The Taxation of ExecutiveCompensation

Brian J HallHarvard Business School and NBER

Jeffrey B LiebmanJohn F Kennedy School of Government Harvard Universityand NBER

EXECUTIVE SUMMARY

Over the past 20 years there has been a dramatic increase in the share ofexecutive compensation paid through stock options We examine theextent to which tax policy has inuenced the composition of executivecompensation and discuss the implications of rising stock-based pay fortax policy We begin by describing the tax rules for executive pay indetail and analyzing how changes in various tax rates affect the taxadvantages of stock options relative to salary and bonus Our empiricalanalysis leads to three conclusions First there is little evidence that taxchanges have played a major role in the dramatic explosion in executivestock-option pay since 1980 Although the tax advantage of options hasapproximately doubled since the early 1980s options currently haveonly a slight tax advantage relative to cashmdashapproximately $4 per $100

We thank James Poterba (the editor) for helpful suggestions and compensation consul-tants Robert Greenberg and Scott Olsen (of Towers Perrin) and Fred Cook (of Frederic WCook and Company) for helpful insights and background information We thank PaulGompers and Andrew Metrick for providing us with the data on the share ownership oflarge institutional investors David Heilman whose Harvard undergraduate thesis rstprovoked us to think about these issues and Austan Goolsbee Gerald Auten and KevinMurphy for comments on an earlier draft Gabriel Leon Humayun Khalid and KatyaRosenblatt provided excellent research assistance

2 Hall amp Liebman

of pretax compensation to the executive A more convincing story for thedramatic explosion in stock options involves changes in corporate gover-nance and the market for corporate control For example there is astrong correlation between the fraction of shares held by large institu-tional investors and the fraction of executive pay in the form of stockoptions a result that holds both longitudinally and cross-sectionallySecond we nd evidence that the million-dollar rule (which limited thecorporate deductibility of non-performance-related executive compensa-tion to $1 million) led rms to adjust the composition of their pay awayfrom salary and toward performance-related pay although our esti-mates suggest that this substitution was minor We nd no evidencethat the regulation decreased the level of total compensation Third weexamine whether there is evidence for signicant shifting of the timingof option exercises in response to changes in tax rates After replicatingGoolsbeersquos (1999) result regarding tax shifting with our data for the 1993tax reform we show that no such shifting occurred in either of the twotax reforms of the 1980s Moreover we nd evidence that much of theunusually large level of option exercises in 1992 was the result of therising stock market rather than the change in marginal tax rates

1 INTRODUCTION

Over the past 15 years there has been a major change in the way thatAmerican executives particularly CEOs are paid For many executivesannual stock-option grants are now greater than cash compensation(salary and bonus) Annual changes in CEO wealth from revaluations ofstock and stock-option holdings completely swamp cash compensationand provide substantial pay-to-performance sensitivity All of this is adramatic change from the early 1980s when the median stock-optiongrant to top executives was zero

In this paper we examine the extent to which tax policy has inu-enced the composition of executive compensation and discuss the impli-cations of rising stock-based pay for tax policy Because top executivesmanage assets worth billions of dollars their compensation arrange-ments and the incentives they face are of substantial importance to theperformance of the US economy Because top executives have veryhigh incomes their responsiveness to taxation has important revenueand efciency implications

We conduct a broad analysis of the taxation of executives We begin bystudying how tax rates affect the degree to which options are favoredrelative to cash and how the tax advantage of options has changed overtime in response to changes in corporate personal and capital gains tax

Taxation of Executive Compensation 3

rates Although the tax advantage of options has approximately doubledsince the early 1980s options currently have only a slight tax advantagerelative to cashmdashapproximately $4 per $100 of pretax compensation

We then analyze what we believe to be the three central policy ques-tions regarding the taxation of executive pay First we examine theextent to which the stock-option explosion has been inuenced by themany changes in tax rates over the past 20 years The evidence suggeststhat changes in taxation have likely had a very modest inuence on theoption explosion Instead changes in corporate governance especiallyin the role of large institutional investors appear to have provided themain impetus for the increase in stock-based pay

Second we examine the effectiveness of tax policies aimed at curbingwhat is deemed by some to be excessive levels of executive compensa-tion Section 162(m) of the Internal Revenue Code (the million-dollarrule) which was enacted in 1993 put a $1-million limit on the deducti-bility (against corporate prots) of non-performance-related executivepay Although we nd evidence that this rule led to a shift in the compo-sition of paymdashaway from salary and toward more performance-relatedbonuses and stock optionsmdashour evidence suggests that the magnitudeof this substitution was small We nd no evidence that the million-dollar rule decreased total executive compensation

Third we analyze the degree to which the tax code is efcient inraising tax revenue from top executives The stock-option explosion hasled to a new and important way for executives to lower their taxes inresponse to changes in tax rates by timing their stock-option gains Forexample in 1993 it was widely reported in the press that well-knownCEOs such as Disneyrsquos Michael Eisner pushed their option gains into1992 in order to avoid paying the higher personal income-tax rates imple-mented in 1993 In an important paper Goolsbee (1999) argues that thetax shifting between 1992 and 1993 was enormous and was the directresult of the increase in marginal tax rates during this period Afterreplicating Goolsbeersquos evidence regarding tax shifting with our data forthe 1993 tax reform we show that no such shifting occurred in responseto either of the two tax reforms of the 1980s Moreover our evidenceindicates that the stock market run-up in 1991 and 1992 was more impor-tant than the change in marginal tax rates in causing the large optiongains observed in 1992

This paper proceeds as follows In the next section we discuss trendsin the level and performance sensitivity of executive compensation Insection 3 we describe the tax and accounting rules concerning executivecompensation In section 4 we analyze how taxes affect the degree towhich options are favored relative to cash In section 5 we examine how

4 Hall amp Liebman

the tax advantage of options has changed over time In section 6 weprovide evidence on the effect of taxation on the composition of execu-tive compensation Section 7 contains empirical analysis of the million-dollar rule Section 8 contains evidence on tax shifting and option gainsSection 9 concludes

2 TRENDS IN TOP EXECUTIVE PAY

In this section we document how top executive pay has changed overtime and discuss how this change has caused the sensitivity of CEOwealth to rm market value to increase substantially There has been alarge increase in the level of CEO pay since 1980 and this growth hasbeen driven by the dramatic increase in stock-option grants during thistime (Hall and Liebman 1998) Although salary and bonuses nearlydoubled over the period in ination-adjusted terms the mean value ofstock-option grants increased by 683 percent The percentage increase inthe median stock-option award cannot be calculated because the me-dian stock-option grant was zero in 1980 The median CEO did notreceive an annual stock-option grant until 1985 Today nearly all topexecutives of large companies receive stock options and the averagestock-option grant is now larger for most top executives than salary andbonus combined

The (ination-adjusted) growth rate of CEO pay since 1980mdashmeasurednarrowly (cash pay grew at an annual rate of 5 percent per year) orbroadly(cash plus option grants grew at almost 9 percent per year) or very broadly(total compensation including stock and stock-option appreciation grewat 115 percent per year)mdashhas been large relative to virtually all othergroups Indeed the growth rate of CEO pay since 1980 has been higheven relative to the pay increases of other high-income earners For ex-ample the cutoff point for being in the top 05 percent of adjusted grossincome (AGI) increased by about 37 percent per year about half the ratefor direct CEO compensation (excluding stock and option appreciation)The only workers who appear to have had faster compensation growththan CEOs are other ldquosuperstarsrdquo The annual pay of professional base-ball players increased by approximately 98 percent per year and that ofprofessional basketball players by 139 percent per year

The increase in stock options has led to a large increase in the equityholdings of top executives and this in turn has led to a dramatic increasein the responsiveness of executive wealth to rm performance Nearlyall of the pay-to-performance sensitivity of executive compensationcomes from equity holdings for a given increase in shareholder valuechanges in the value of an executiversquos stock and stock options are more

Taxation of Executive Compensation 5

than 50 times larger than changes in salary and bonus (Hall and Lieb-man 1998)

As a concrete example the estimates in Hall and Liebman (1998) implythat a 10-percent increase in rm value (of the median company in oursample) leads the company to increase the CEOrsquos salary and bonus byabout $25000 However this same 10-percent increase in shareholdervalue translates into $125-million increase in the value of the CEOrsquosstock and stock-option holdings1

The dramatic rise in the link between CEO wealth and rm perfor-mance can be seen in Figure 1 which shows how two measures of thislink have increased since 19802 The rst measure the JensenndashMurphy(1990) sharing rate (shown on the left scale) is the change in CEO wealthfor a $1000 change in rm value The second measure is the change inCEO wealth for a 10-percent change in rm value (see Baker and Hall1998) Since both measures are strongly affected by rm size (the formerhas a negative correlation and the latter has a positive correlation) thepay-to-performance changes over time are estimated with regression(quantile) analysis that controls for changes in the distribution of rmsizes in the sample over time The gure therefore shows the increasein the pay-to-performance measures over time for a constant-size rmin this case a $1-billion rm (in constant 1998 dollars) The striking fact isthat both measures of the pay-to-performance link have increased bynearly a factor of 10 since 1980 These pay-to-performance increases areeven larger than those we reported in our earlier paper that analyzed theperiod 1980 to 1994 because of the large increase in stock-option grantscombined with the strong stock market performance in 1994 to 1998

3 TAX AND ACCOUNTING RULES

Stock options give an executive the right but not the obligation to buy ashare of the companyrsquos stock at a prespecied pricemdashthe exercise or strikeprice Typically options cannot be exercised immediately That is theyvest (become owned by the executive who can then exercise if he or she

1 Stock-option grants are also very sensitive to changes in rm performance mostly be-cause many grants are multiyear plans that hold the yearly number of options constantand the same number of at-the-money options are worth more when the stock price ishigher and vice versa If stock-option grant sensitivity is also included then about 91percent of pay-to-performance sensitivity comes from stock and stock-option revaluations7 percent comes from stock-option grant changes and less than 2 percent comes fromchanges in salary and bonus (Hall 1999)2 Both measures include only the link created by CEO holdings of stock and stock optionsand ignore the smaller amount of pay-to-performance sensitivity that operates throughchanges in salary bonus and stock-option grants

FIG

UR

E1

Pay

toP

erfo

rman

ceT

heL

ink

betw

een

CE

OW

ealt

han

dSh

areh

olde

rV

alue

Sinc

e19

80

Not

eT

he

anal

ysis

cont

rols

for

the

chan

ging

com

pos

itio

nof

rm

size

over

tim

eT

hu

sth

ese

esti

mat

essh

owth

ep

ay-t

o-p

erfo

rman

celin

kfo

ra

sim

ilar

-siz

edco

mp

any

over

tim

ein

this

case

aco

mp

any

wit

ha

mar

ket

valu

eof

$1bi

llion

Taxation of Executive Compensation 7

wishes) slowly over time Common vesting periods are in the three- tove-year range and options usually vest linearly (eg a four-year op-tion vests at 25 percent at the end of each year) An executive typicallyloses any unvested options upon departure Although options may beexercised as soon as they vest they do not have to be exercised untilthey expire or mature Almost 85 percent of stock-option plans have aterm of exactly ten years with virtually all of the remainder being in theve- to ten-year range About 95 percent of options are granted at themoney or at fair market value which means that the exercise price at grantdate is set equal to the stock price at grant date The remaining 5 percentare either discount options (so-called in-the-money options where theexercise price is below the stock price at grant date) or premium options(so-called out-of-the-money options where the exercise price is abovethe stock price at grant date) The holders of options typically do nothave dividend rights or voting rights even on vested options3

31 Tax Rules and Stock-Option CompensationUnlike salary and bonus stock-option grants are typically an untaxedevent at the time of grant For the most widely used optionsmdashnonquali-ed stock options (NQSOs)mdashexecutives are taxed at the personal incometax rate on option prots (the difference between that stock price and theexercise price times the number of options) when the options are exer-cised The company receives a parallel deduction against corporate in-come at that point If the executive continues to hold the shares afterexercise any subsequent appreciation is taxed at the capital gains rate inthe usual way In 1993 an additional feature was added to the tax code[Internal Revenue Code section 162(M)] that disallowed a corporate de-duction for any executive pay above $1 million that is not performance-based While this rule affects executive salaries most bonuses qualify asperformance-based and standard stock options automatically qualifyTherefore this provision gives companies with highly paid executives anincentive to give more pay in the form of bonuses and stock options asubject we return to in section 7 A summary description of the tax (andaccounting) treatment of cash and option compensation is in Table 1

A far less common type of option which is estimated to account forabout 5 percent of option grants is the incentive stock option (ISO) WhileISOs are similar to NQSOs in their design they are crucially different intwo respects First they have an annual cap of $100000 per executive

3 See Murphy (1999) for details about stock options and Miller and Scholes (1982) on taxincentives

TA

BL

E1

Tax

and

Acc

ount

ing

Tre

atm

ent

ofE

xecu

tive

Com

pens

atio

nC

ash

Pay

men

tsve

rsus

Stan

dard

Exe

cuti

veO

ptio

ns

Tax

atio

nat

Tax

atio

nat

Acc

ount

ing

Com

pens

atio

nD

escr

ipti

onp

erso

nall

evel

corp

orat

ele

vel

trea

tmen

t

Sala

ryan

dbo

nu

sC

ash

pay

men

tsm

ade

Tax

edas

ord

inar

yFu

llyd

edu

ctib

leS

ince

Exp

ense

dag

ains

tea

rn-

toC

EO

in

com

e19

93a

ny

pay

over

$1in

gsin

the

usu

alw

ay

mill

ion

isno

td

educ

t-ib

leu

nles

sit

isp

er-

form

ance

-bas

ed

Non

-qua

lie

dst

ock

Op

tion

sgr

ant

the

ex-

The

gran

ting

ofan

op-

Apa

ralle

ltax

ded

uc-

The

reis

no

exp

ense

rec-

opti

ons

(NQ

SOs)

ecut

ive

the

righ

tbu

tti

onis

typ

ical

lyan

un-

tion

atth

eco

rpor

ate

ogni

tion

(atg

ran

tex

er-

not

the

obli

gati

ont

ota

xed

even

tU

pon

exer

-le

veli

sge

nera

llyal

-ci

seo

rsa

le)

for

op-

buy

shar

esof

stoc

kat

cise

opt

ion

pro

ts

are

low

edu

pon

exer

cise

tion

sif

as

isu

sual

lyan

exer

cise

pric

e(t

ypi-

taxe

dat

the

ord

inar

y(f

orth

eam

ount

ofop

-th

eca

set

he

opti

ons

cally

the

curr

ent

stoc

kra

teI

fth

eex

ecu

tive

tion

pro

ts

)O

pti

ons

are

not

dis

cou

nted

(in

pri

ceat

the

dat

eof

cont

inu

esto

hold

onto

auto

mat

ical

lyqu

alif

yas

the

mon

eyat

gran

tgr

ant)

Th

eop

tion

sth

esh

ares

the

exec

u-

perf

orm

ance

-bas

edp

ayd

ate)

and

the

exer

cise

typ

ical

lyve

stov

era

3ndashti

veis

taxe

din

the

and

ther

efor

ear

eno

tpr

ice

and

num

ber

ofop

-5-

year

per

iod

The

typ

i-us

ualw

ay(a

tth

eca

pi-

subj

ect

toth

e$1

-ti

ons

iskn

own

atgr

ant

calm

atu

rity

is10

year

sta

lgai

nsra

te)

onan

ym

illio

nca

p

dat

eFo

rd

isco

unt

op-

but

the

man

ager

may

furt

her

app

reci

atio

nof

tion

sth

ed

iffe

renc

ebe

-ex

erci

seea

rly

Som

eth

esh

are

pri

ce

twee

nth

est

ock

pri

ceop

tion

sar

edi

scou

ntop

-an

dth

em

arke

tp

rice

isti

ons

(the

exer

cise

pri

ceex

pen

sed

over

the

vest

-is

low

erth

anth

est

ock

ing

per

iod

For

opti

ons

pri

ceat

the

gran

td

ate)

w

ith

vari

able

term

sP

rem

ium

opti

ons

are

(eg

a

vari

able

exer

-th

ere

vers

e(t

heex

er-

cise

pri

cev

esti

ngth

atci

sep

rice

isgr

eate

ris

tied

top

erfo

rman

ce)

than

the

stoc

kp

rice

atth

eop

tion

sar

em

arke

dgr

ant

dat

e)

tom

arke

tan

dex

-p

ense

dd

uri

ngth

eti

me

betw

een

gran

tan

dex

erci

se

Not

eW

eth

ank

com

pen

sati

onco

nsu

ltan

tsSc

ott

Gre

enbe

rgan

dSc

ott

Ols

en(b

oth

ofT

ower

sP

erri

n)

and

Fred

eric

Coo

k(o

fFr

eder

icW

C

ook

and

Ass

ocia

tes)

for

help

fulc

onve

rsat

ion

sin

und

erst

and

ing

the

tax

and

acco

unt

ing

rule

s

10 Hall amp Liebman

Second the tax treatment of ISOs is different ISOs are completely un-taxed at grant or exercise It is only at sale that the executive is taxed andeven then the executive is taxed at the lower capital gains tax rate Thedisadvantage is that the corporation never gets to take a parallel taxdeduction against corporate prots Thus ISOs become more attractiveas the personal tax rate increases and as the corporate tax rate and thecapital gains tax rates fall

Two other related types of compensation restricted stock and stockappreciation rights are worth brief discussion Both however are farless common than standard stock options Restricted stock is payment inthe form of restricted shares the restriction being that the shares vestover time as with options Unlike options the shares typically havevoting and dividend rights With regard to taxation the executive istaxed at the personal rate on the value of the restricted stock as thevesting restrictions lapse However the executive may choose to betaxed at the grant date in which case all subsequent appreciation istaxed at the capital gains rate The dividends paid to the executive aretaxed at the ordinary rate in the usual way The company generallyreceives a parallel deduction equal to the amount of the executiversquosincome when the executive is taxed Unlike stock options restrictedstock is not generally considered to be performance-based and is there-fore subject to the million-dollar rule (unless the vesting of the restrictedstock is performance-based which is sometimes the case)

Stock appreciation rights (SARs) are rights that replicate the payoffs ofstock options with a cash transfer Thus SARs are simpler than options inthat there is no requirement to buy and resell the stock in order to ldquocashoutrdquo SARs generally have the same tax treatment as NQSOs both to theindividual and to the corporation SARs are relatively rare however be-cause they have disadvantageous accounting treatment (described in thenext section) relative to stock options and their main relative advantagevis-a-vis options has been essentially replicated through broker-assistedcashless option exercisesmdashwhereby a broker makes a ldquonanosecondrdquo loanto the executive (to purchase and resell the stock) when the executivewants to ldquocash inrdquo on option prots The tax and accounting treatment ofISOs restricted stock and SARs is summarized in Table 2

32 The Accounting Treatment of OptionsUnlike cash compensation which is expensed against earnings there isgenerally no expense recognition (at grant exercise or sale) for optionswhether they be NQSOs or ISOs As a result compensation consultantsoften point out that stock options are the only form of compensation thatare free in an accounting sense but still deductible for tax purposes

Taxation of Executive Compensation 11

Stock options do however lead to expense recognition if they are dis-counted (in the money at grant date) or if the exercise price and numberof options are not known at grant date For discount options the differ-ence between the stock price and the market price is expensed over thevesting period For options with variable terms (eg a variable exerciseprice vesting that is tied to performance) the options are marked tomarket and expensed during the time between grant and exercise Practi-tioners claim that the accounting treatment of options plays an impor-tant role in the design of option programs Thus plans that have ldquobadaccountingrdquo but are thought by many to have attractive incentive fea-tures are often not even seriously considered by companies Examples ofsuch potentially attractive plans include indexed options (where a CEOprots only if his rmrsquos share price grows relative to some market orindustry benchmark) and option grants that are explicitly performance-related both of which would lead to expenses against earnings

Unlike options restricted stock and SARs do not generally receivefavorable accounting treatment Restricted stock is generally expensedover the period in which the restrictions lapse (usually the vesting pe-riod) The magnitude of the expense is the difference between the cur-rent stock price and the executiversquos cost (if any) SARs are marked tomarket each period and the difference between the stock price and theexercise price is expensed over the outstanding period of SARs

4 THE TAXATION OF EXECUTIVE PAY CASHVERSUS OPTIONS

Stock-based compensation performs two roles in executive compensa-tion arrangements First it helps align the incentives of the executivewith the interests of the rmrsquos shareholders Second it often enables therm to compensate the CEO in a way that is more advantageous from atax standpoint than paying the executive in salary and bonus

41 Agency Theory and Executive Compensation in the Presenceof TaxationIn standard agency-theory models ( Jensen and Meckling 1976) agencycosts are the result of the separation of ownership and control Managersdo not have the same incentives as the owners The optimal incentivecontract for managers balances the benets of high-powered incentives(linking the fortunes of owners and managers through stock and stockoptions for example) with the costs of loading too much risk on risk-averse managers

TA

BL

E2

Tax

and

Acc

ount

ing

Tre

atm

ent

ofE

xecu

tive

Com

pens

atio

nIn

cent

ive

Stoc

kO

ptio

ns(I

SOs)

Res

tric

ted

Stoc

kan

dSt

ock

App

reci

atio

nR

ight

s(S

AR

s)

Tax

atio

nat

Tax

atio

nat

Acc

ount

ing

Com

pen

sati

onD

escr

ipti

onp

erso

nall

evel

corp

orat

ele

vel

trea

tmen

t

Ince

nti

vest

ock

ISO

sar

eid

enti

calt

oIf

the

usu

alco

ndit

ions

Ifth

eu

sual

cond

itio

nsA

sw

ith

NQ

SOs

ther

eop

tion

sN

QSO

sin

thei

rd

esig

n

are

met

th

ere

isno

tax

are

met

the

reis

gene

r-is

gene

rally

noex

pen

seH

owev

ert

hey

have

anat

gran

tor

exer

cise

At

ally

no

corp

orat

eta

xd

e-re

cogn

itio

nfo

rIS

Os

annu

alca

pof

$100

000

sale

all

opti

onp

ro

tsd

uct

ion

atgr

ant

exer

-(e

ver)

on

the

amou

ntth

atan

dsu

bseq

uen

tca

pita

lci

seo

rsa

le

vest

sT

he

tax

trea

t-ap

prec

iati

onar

eta

xed

men

tof

ISO

sal

soat

the

cap

ital

gain

sd

iffe

rs

rate

Res

tric

ted

stoc

kSi

mila

rto

stoc

kop

-T

hein

div

idu

alis

taxe

dT

heco

mp

any

gene

rall

yR

estr

icte

dst

ock

isge

n-ti

ons

but

shar

esat

the

ord

inar

yra

teas

rece

ives

ap

aral

leld

e-er

ally

exp

ense

dov

er(w

hich

vest

slow

ly)

the

rest

rict

ions

lap

se

du

ctio

neq

ual

toth

eth

eve

stin

gp

erio

dT

hera

ther

than

opti

ons

are

How

ever

the

man

ager

amou

ntof

the

exec

u-

amou

nt

exp

ense

dis

gran

ted

toth

eex

ecu

-m

aych

ose

tobe

taxe

dti

versquos

inco

me

whe

nth

eth

ed

iffe

renc

ebe

twee

nti

veU

nlik

eop

tion

sat

gran

td

ate

inw

hich

exec

uti

veis

taxe

dR

e-th

ecu

rren

tst

ock

pric

eth

esh

ares

typi

cally

case

alls

ubs

equ

ent

ap-

stri

cted

stoc

kis

gen

er-

and

the

exec

uti

versquos

have

voti

ngan

dd

ivi-

pre

ciat

ion

ista

xed

atal

lysu

bjec

tto

the

cost

(if

any)

d

end

righ

ts

the

cap

ital

gain

sra

te

mill

ion-

dol

lar

cap

D

ivid

end

sar

eta

xed

atth

eor

din

ary

rate

Stoc

kap

pre

ciat

ion

Rig

hts

that

repl

icat

eG

ener

ally

have

the

Gen

eral

lyha

veth

eU

nlik

eop

tion

sSA

Rs

righ

tsth

epr

ots

ofst

ock

op-

sam

eta

xtr

eatm

ent

assa

me

tax

trea

tmen

tas

are

mar

ked

tom

arke

tti

ons

Rar

ely

use

d

NQ

SOs

NQ

SOs

each

per

iod

an

dth

esi

nce

stoc

kop

tion

sd

iffe

renc

ebe

twee

nth

eha

vem

ore

favo

rabl

eac

-st

ock

pri

cean

dth

eex

-co

unt

ing

trea

tmen

ter

cise

pri

ceis

exp

ense

d(a

ndop

tion

sp

rovi

de

over

the

outs

tand

ing

the

sam

ebe

ne

tof

per

iod

ofSA

Rs

cash

less

exer

cise

thro

ugh

brok

erlo

ans

)

Not

eW

eth

ank

com

pen

sati

onco

nsul

tant

sSc

ott

Gre

enbe

rgan

dSc

ott

Ols

en(b

oth

ofT

ower

sP

erri

n)an

dFr

eder

icC

ook

(of

Fred

eric

WC

ook

and

Ass

ocia

tes)

for

hel

pfu

lcon

vers

atio

nsin

und

erst

and

ing

the

tax

and

acco

unt

ing

rule

s

14 Hall amp Liebman

The effect of taxation on the optimal contract (more precisely theshare of compensation that is performance-related) is ambiguous (evenin the absence of deductibility and deferral) because there are offsettingeffects First by reducing the share of corporate prots received byshareholders taxes diminish the importance to the shareholders of moti-vating the CEO Taken alone this effect would imply that taxes wouldbe expected to reduce the use of performance-based compensation Sec-ond because the government shares in the income received by the CEOthe variance of the CEOrsquos income is reduced raising his utility andlowering the cost to the rm of providing any given set of incentivesThird because taxes will lead executives to provide less effort for anygiven level of incentive-based pay the level of compensation that mustbe provided to compensate the CEO for effort is reduced for a givenamount of incentive These last two factors make it cheaper for the rmto offer contracts to the CEO with large incentive components when taxrates are higher and should therefore increase the use of incentive-based pay Since the net effect of taxation on the level of incentive-basedpay is ambiguous agency theory provides no strong predictions abouthow taxation should affect the optimal composition of option versuscash compensation

42 The Tax Advantages of DeferralWhile agency theory yields ambiguous predictions about the effect oftaxes on the use of options there are direct tax advantages of options inthat options provide executives with a way to defer compensation andthereby lower their taxes However because options also lead to a defer-ral of corporate tax deductibility the tax advantages from a combined(executive and corporate) perspective are not so straightforward In thissection we analyze and measure the tax advantages of stock optionsrelative to salary and bonus compensation We also analyze the taxadvantages of NQSOs relative to ISOs In particular we show how thetax advantages of options change as personal corporate and capitalgains taxes change We then show how the tax advantages of optionshave changed over time in response to changing tax rates

The crucial tax difference between standard options (NQSOs) andcash payment is that option payouts are deferred and the two forms ofcompensation earn different rates of return over the deferral period Anyanalysis of the relative tax advantages of two compensation instrumentsmust consider the tax consequences both to the employer and to theemployeemdashwhat Scholes and Wolfson (1992) call the global contractingperspective Thus in order to make valid comparisons between the twowe compare the tax burden to the executive while holding constant the

Taxation of Executive Compensation 15

posttax cost to the company By keeping posttax employer costs (in NPVterms) constant any package that is preferred by the employee is tax-advantaged in the global-contracting sense

Under this methodology a comparison of the tax advantages of op-tions and cash involves comparing a pretax cash payment of P with anoption payment that has the equivalent posttax NPV to the companyFor NQSOs it is straightforward to show that a pretax payment of P toan executive has exactly the same cost to the company as putting aside Pfor the purposes of paying stock-option payouts later That is if a com-pany pays an executive P today it will have the same amount of moneyin N years as if it had put aside P today let it accumulate at the rate ofreturn earned by the rm and then paid it out (as compensation in theform of option gains4) with any appreciation in year N

Assume that the pretax prot rate is r and the corporate tax rate is TcThen if the company pays P today it receives a deduction today of PTcwhich yields

PTc [1 1 r (1 2 Tc)]N (1)

in N years since the benets of the tax deduction are invested in thecompany and receive the after-tax corporate rate of return Converselyif the company puts P aside today then it grows at the after-tax corpo-rate rate of return in N years to give a tax deduction of

P [1 1 r (1 2 Tc)]N Tc (2)

which is the exact same value Note that the equivalence of these two taxdeductions is analogous to the tax-benet equivalence of front-loadedand back-loaded IRAs

An executiversquos payoff in N years from option prots is equal to Pcompounded at the after-tax corporate rate times 1 2 Tp where Tp is thepersonal tax rate Thus the combined payoffs in N years of the corporatededuction and the executiversquos posttax payoff is

P [1 1 r (1 2 Tc)]N Tc 1 P [1 1 r (1 2 Tc)]

N (1 2 Tp) (3)

where the rst term is the payoff from the corporate deduction and thesecond term is the executiversquos payoff

4 This analysis is not unique to options Any form of deferred compensation that enablesexecutives to invest inside the rm at a higher posttax return or without paying capitalgains taxes would have similar effects

16 Hall amp Liebman

The payoff to salary and bonus in N years is more complicated sincean assumption must be made about how the executive invests the origi-nal cash compensation For example if the executive invests in an instru-ment with taxable interest (eg bonds) then the N-year payoff from aninvestment of P accumulates at the post-corporate-tax post-personal-taxrate of return and is

P (1 2 Tp) [1 1 r (1 2 Tc)(1 2 Tp)]N (4)

However if an executive invests in (non-dividend-paying) equities theinvestment accumulates at the higher post-corporate-tax rate of returnwith the offsetting disadvantage that the capital gains are taxed at thecapital gains rate in year N So the executive receives

P [1 1 r (1 2 Tc)]N (1 2 Tp) 2 Tcg $P (1 2 Tp) [1 1 r (1 2 Tc)]

N 2P (1 2 Tp) (5)

where Tcg is the capital gains tax rate5Thus putting the equations together if an executive is paid cash P

and invests g in bonds and 1 2 g in equities the N-year payoff (includ-ing the payoff to the corporation from the tax deduction) is

P [1 1 r (1 2 Tc)]N Tc 1 g $(P [1 1 r (1 2 Tc)(1 2 Tp)]

N (1 2 Tp)1(1 2 g) (P [1 1 r (1 2 Tc)]

N (1 2 Tp) 2 Tcg $P (1 2 Tp)[1 1 r (1 2 Tc)]

N 2 P (1 2 Tp)) (6)

The tax advantage of options versus salary and bonus therefore is thedifference between the combined (corporate and executive) payoffs inequation (3) and the combined payoffs in equation (6)

43 The Tax Advantage of OptionsWe now turn to analysis of how changes in various tax rates affectthe tax advantages of options relative to cash compensation We focuson the case in which the cash earned by executives is invested entirelyin equities since equities are tax-favored relative to bonds The impor-tant conclusions of this analysis are not substantively different if we

5 In practice executives hold a combination of equities and bondlike instruments withtaxable interest Because equities are tax-favored in the next section we simplify the analy-sis by assuming that executives hold only equities Alternatively we could have assumedthat executives hold only less risky bonds since bonds more closely match the risk proleof a stream of tax savings None of our empirical results are substantively affected by thissimplication

Taxation of Executive Compensation 17

instead assume that a portion of an executiversquos holdings are invested inbonds

431 The Tax Advantage of Options Changing the Corporate TaxRate Stock options are tax-advantaged relative to cash simply becauseoption payouts are deferred allowing the executive to invest at the pre-personal-tax rate of return (with no capital gains tax at the end) ratherthan investing at the after-tax rate of return or paying the capital gainstax Since the advantage of being able to defer taxes is large when thecorporate tax rate is low the tax advantage of options is larger when thecorporate rate is low

To illustrate this point we dene the tax advantage of options to bethe dollar amount by which total after-tax option payoffs (to the corpora-tion and the executive) exceed the total payoffs from salary as denedby the difference of equation (3) and (6) assuming that P (the paymentto the executive) is equal to $100 and g 5 0 (cash compensation is in-vested in equities) We use a ten-year horizon since most options have aten-year term We then calculate how the tax advantage of optionschanges as the corporate tax rate varies from zero to 100 percent holdingthe personal tax rate and the capital gains rate xed at 40 percent and 20percent respectively (which approximates current rates)

The top panel of Figure 2 shows the tax advantage of options relativeto cashmdashprecisely the NPV of the total tax advantage of options for a$100 payment to the CEOmdashas the corporate tax rate changes As ex-pected the tax advantage of options declines as corporate tax rates in-crease At a zero corporate tax rate the tax advantage of options has anNPV of $7 This value is about $350 at a corporate rate of 40 percentWhen the corporate tax rate is 100 percent the tax advantage of optionscompletely disappears since the benets of deferred compensation fallto zero because the after-tax return on equities (and therefore the dis-count rate) falls to zero The tax advantage of options is the NPV of thetax savings from avoiding the capital gains tax on ten years of apprecia-tion which is why it is relatively modest even at the 40-percent corpo-rate rate

432 The Tax Advantage of Options Changing the Personal TaxRate Using the same assumptions as above (but this time holding thecorporate rate at 40 percent and the capital gains rate at 20 percent andvarying the personal rate) the tax advantage of options declines as thepersonal tax rate rises as shown in the middle panel of Figure 2 The taxadvantage of options is higher at low personal tax rates because theposttax base that gives rise to capital gains is higher and the advantage of

Note Holds constant the personal rate at 40 percent and the capital gains rate at 20 percent

Note Holds constant the corporate rate at 40 percent and the capital gains rate at 20 percent

Note Holds constant the personal rate at 40 percent and the corporate rate at 40 percent

FIGURE 2 Tax Advantage of Options NPV of Tax Advantage of Pay-ing $100 in Options (NQSOs) Rather Than in Cash

Taxation of Executive Compensation 19

avoiding the capital gains taxes is greater The tax advantage of optionstherefore declines linearly as the personal tax rate rises At a 100-percentpersonal tax rate the executive receives nothing in either case and byconstruction the companyrsquos tax deduction benet is the same in NPVterms so the tax advantage of options falls to zero also6

433 The Tax Advantage of Options Changing the Capital Gains TaxRate The tax advantage of options for various capital gains rates isdepicted in the bottom panel of Figure 2 under the same assumptions(this time holding both the personal rate and the corporate rate at 40percent while varying the capital gains rate) The tax advantage of avoid-ing the capital gains tax increases as the capital gains tax rate rises

5 HOW HAS THE TAX ADVANTAGE OF OPTIONSCHANGED OVER TIME

The analysis so far has illustrated how the tax advantage of optionsvaries with the changes in tax rates In this section we show how the taxadvantage of options has changed over time as tax rates have changed

51 Top Marginal Tax Rates over TimeThe evolution on the top marginal tax ratesmdashpersonal corporate andcapital gainsmdashfrom 1980 to 1998 is shown in Figure 3 In all cases thetop rate (the rate for taxpayers with the highest incomes) is shownwhich is not always the highest rate since various anomalies (such asclawbacks of exemptions) sometimes lead to marginal tax rates that arehigher than those for the highest income tax payers

The top personal tax rate was 70 percent in 1980 falling to 50 percentin 1982 and 28 percent in 1988 following the 1986 tax act7 The topmarginal tax rate has since risen to 396 percent but is effectively 425percent since there is a 29-percent Medicare surcharge (paid half by theemployer and half by the employee) that has no upper limit The topcorporate tax rate has had only one signicant change since 1980 fallingfrom 46 percent before the 1986 tax act to 34 percent following the act in1988 The top corporate tax rate was increased to 35 percent in 1993 The

6 This is one place where ignoring bond investments has substantive implications Inparticular in the all-bond case the tax advantage of options is nonmonotonicmdashit rises andthen fallsmdashas personal tax rates rise7 During the early 1980s the maximum tax on earned income limited the marginal tax rateon the earnings of high earners so many executives did not face a decline in personal taxrates between 1981 and 1982 The empirical results in this paper are robust to assumingthat executives faced a marginal tax rate of 50 percent in 1980 and 1981

20 Hall amp Liebman

FIGURE 3 Top Marginal Tax Rates since 1980

top capital gains rate has uctuated between 28 percent (approximately)and 20 percent since 1980 and is currently at 20 percent

52 The Evolution of the Tax Advantage of NQSOsThe tax advantage of options relative to cash dened in the same way asin section 4 from 1980 to 1998 is shown in Figure 4 The calculations aredone using the same assumptions as before (ie N 5 10 years and cashis invested in equities) and the statutory top marginal tax rates shown inFigure 3 are used Due to the fall in corporate and personal tax rates thetax advantage of options increased sharply after the Tax Reform Act of1986 This increase in tax advantage has largely been reversed in the1990s as top personal tax rates have crept back up and the capital gainsrate has been reduced

Two key facts emerge from this analysis First there is currently only amoderate tax advantage to standard nonqualied stock optionsmdashon theorder of $4 per $100 of compensation This is because the tax advantagesto the executive of deferring taxes are largely offset by the tax disad-vantages to the company of not being able to deduct option expensesfrom taxable prots until the executive exercises the options Secondalthough the 1986 tax act substantially increased the tax advantage ofoptions more than half of this increase has been reversed in the 1990s

Taxation of Executive Compensation 21

FIGURE 4 Tax Advantage of Options 1980 to 1998

NPV of tax advantage of $100 in options (NQSOs) rather than cash

53 The Tax (Dis)advantage of ISOsSo far the analysis has focused on the tax advantages of NQSOs sincethey are so much more prevalent than ISOs which account for onlyabout 5 percent of option grants The relative scarcity of ISOs can beexplained both by their tax status and by their per-executive cap of$100000 per year

Under what conditions are ISOs tax-preferred to NQSOs For a trans-fer P to an executive the ISO is always tax-preferred by the executive(since the capital gains rate is lower than the personal rate) while theNQSO is always tax-preferred from the companyrsquos perspective (since anISO is not deductible) The key issue however is the relative advan-tages of the two types of options from a global-contracting perspectiveBecause ISOs are not deductible the company is indifferent betweensetting aside P in the form of NQSOs today and setting aside P(1 2 Tc) inthe form of ISOs (Since the NPV of the tax deduction for NQSOs is PTcthe NPV of the payment P is P 2 PTc or P (1 2 Tc))

Equalizing the posttax cost of NQSOs and ISOs we need only look atthe payoff of the executive to determine the condition under which eachis preferred The payoff of NQSOs in N years is

P [1 1 r (1 2 Tc)]N (1 2 Tp) (7)

22 Hall amp Liebman

The same payoff for ISOs is

P (1 2 Tc) [1 1 r (1 2 Tc)]N (1 2 Tcg) (8)

ISOs therefore are tax-advantaged if

Tp 2 Tcg Tc 2 Tc Tcg (9)

Intuitively ISOs are tax-advantaged only if their advantage (the differ-ence between the personal rate and the capital gains rate) is largeenough to offset their cost (the disadvantage of not deducting at thecorporate rate) Note however that the condition is not a simple com-parison between the corporate rate and the personal rate minus thecapital gains rate (Tp 2 Tcg Tc)

Figure 5 shows how the tax advantage of NQSOs relative to ISOs haschanged since 1980 as tax rates have changed Unlike the very modestchanges (around 3 to 4 percent) in the relative tax advantages betweenoptions and cash shown in Figure 4 the relative tax advantage betweenNQSOs and ISOs has seen enormous swings When the top personalrate was 70 percent in 1980 ISOs were tax-favored by a margin greaterthan 6 percent However by 1982 (following the 1981 tax acts that dra-matically lowered the personal tax rate) ISOs became tax-disadvantaged

FIGURE 5 Tax Advantage of NQSOs Relative to ISOs over Time

NPV of tax advantage of $100 payment to executive NQSOs minus ISOs

Taxation of Executive Compensation 23

by about 5 percent By 1988 (following the 1986 tax act) ISOs became tax-disadvantaged by about 18 percent a dramatic swing The raising ofpersonal rates coupled with a decrease in the capital gains rate has sincereduced the tax disadvantage of ISOs to about $4 per $100 of pretaxcompensation Although hard data on ISOs are hard to come by practi-tioners (mostly compensation consultants) claim that ISOs were morepopular prior to 1982 when they were tax-advantaged Although ISOsare less disadvantageous from a tax standpoint relative to NQSOs thanthey were in the late 1980s they are still disadvantageous so it is notsurprising that they have not made a signicant resurgence in recentyears

6 EXPLAINING THE INCREASE IN OPTION-BASEDCOMPENSATION

In this section we analyze whether changes in tax rates affect the compo-sition of executive compensation and consider the relative importanceof tax factors and corporate governance factors in explaining the in-creasing share of compensation paid in the form of stock options Thenumerous changes in tax rates that have occurred since 1980 provide anopportunity to assess whether executive compensation arrangementsrespond to tax incentives in the way that the tax-avoidance model sug-gests We test this model using a panel data set of CEOs in largepublicly-traded US corporations Our identication of the tax effectsrelies on time-series variation in personal corporate and capital gainstax rates along with cross-sectional variation in corporate tax rates

61 DataWe use a 15-year panel data set of CEOs in the largest publicly-tradedUS corporations which is described in Hall and Liebman (1998) Thedata set covers the years from 1980 through 1994 and combines CEOcompensation information from corporate proxies and 10-K lings withstock price and stock return information from CRSP and accounting datafrom Compustat In addition some compensation data from the 1970swere collected in order to construct measures of the value of stock op-tions held by the CEOs in the rst period of the sample8

The feature that distinguishes our data from most other CEO datasets is that with our panel of yearly proxy data on option grants optiongains and total options held we are able to calculate the total value of

8 We are grateful to David Yermack for providing us with some of the data for 1984 to 1991See Yermack (1995) for a discussion of these data

24 Hall amp Liebman

all stock options held by the CEO at a point of time More importantlysince we have the details about the stock options held (number exer-cise price time to maturity etc) we can precisely calculate the changein the value of a CEOrsquos stock option holdings for a given change in rmvalue

62 Identication IssuesThe numerous tax changes over our sample period and the specic waysin which changes in personal corporate and capital gains rates arepredicted to affect the composition of CEO compensation give us un-usually rich sources of identication However because many of thechanges in tax rates would be expected to affect all of the CEOs in oursample in a similar way we will need to pay particular attention toseparating out the tax effects from underlying time trends and otherfactors that changed over time Controlling for underlying trends is par-ticularly important in this study because we are focusing on an out-comemdashthe use of stock optionsmdashthat has increased rapidly over thepast two decades and that many practitioners believe was importantlyaffected by nontax considerations Thus the tax effects we are trying toexplain are deviations from a rising trend

We take two steps to try to separate out the tax effects from theunderlying trend First we include key nontax factors that could poten-tially explain the increasing reliance on performance-based compensa-tion It has been suggested that the dramatic increase in incentive-basedpay is the result of the remarkably poor shareholder returns during the1970s which spurred the LBO and takeover movements of the 1980sAccording to this story (Kaplan 1997) shareholders became much morepowerful via the rise of institutional investors even as the LBO andtakeover movements waned Because the inuence of institutional inves-tors is thought to be one of the most important mechanisms of strongcorporate governance we use the share of each companyrsquos stock ownedby institutional investors (dened as institutions with more than $100million under management) as an explanatory variable9 The percentageof shares owned by large institutional investors increased from about 20percent to almost 50 percent in 1994 an upward trend that closelymatches the sharp rise in the share of compensation in the form ofoptions Indeed in our sample the annual average (over all of the rmsin our sample) share of stock owned by institutional investors has acorrelation of 09 with the annual average share of compensation paid inoptions

9 This variable is described in Gompers and Metrick (1998)

Taxation of Executive Compensation 25

In addition we include two other variables that proxy for strongercorporate governance The rst is the size of the board Evidence suggeststhat boards with smaller numbers of directors reduce the inuence of theCEO and better represent shareholder interests (Yermack 1995) Boardsize fell by approximately 15 percent between 1980 and 1994 The secondvariable is the fraction of the rmrsquos board members who are inside direc-tors or gray directors (noninsiders who have business dealings with thecompany) Firms with fewer outsiders are less likely to act in the share-holdersrsquo interests and we would predict that they would be less likely tohave performance-based compensation for their CEOs The average per-centage of inside and gray directors has decreased from about 45 percentin 1980 to about 35 percent in 1994

Second we allow for cross-sectional variation in corporate tax ratesCorporate marginal tax rates vary because rms may be eligible for tax-loss carrybacks and carryforwards investment tax credits and the alter-native minimum tax (see Auerbach 1986 Auerbach and Altshuler1990 Auerbach and Poterba 1987) Because these cross-sectional differ-ences imply that different rms should respond differently to a givenchange in tax rates they potentially provide us with a way to isolatethe tax effects Firm-specic corporate marginal tax rates are notori-ously difcult to calculate from publicly-available data We use a tricho-tomous variable that equals zero if the rm has tax-loss carryforwardsand negative earnings in a given year equals 05 times the statutorycorporate rate if the rm has only one of those conditions and equalsthe statutory rate if the rm has neither loss carryforwards nor negativeearnings10

The cross-sectional variation in the corporate tax rate may not be ex-ogenous Firms that perform poorly could face low marginal tax ratesand also be particularly likely or unlikely to provide performance-basedpay Because most of our specications use xed effects the correlationwe must be concerned about is between changes in rm performanceand changes in compensation We deal with this by controlling forlagged rm stock market performance in our regressions This shouldeliminate spurious correlation between tax rates and compensation thatis jointly caused by rm performance

10 Graham (1996) presents evidence that the trichotomous variable performs nearly as wellas his recommended simulated marginal tax rate Plesko (1999) compares marginal taxrates calculated from Compustat data with actual marginal tax rates from rm tax returnsHe nds that simple binary variables are more reliable measures of the marginal tax ratethan is either the Graham or the trichotomous variable We obtain substantively similarresults when we switch from the trichotomous variable to the binary variables recommendby Plesko

26 Hall amp Liebman

63 ResultsThe dependent variable in our regressions is the fraction of a CEOrsquos totalannual compensation (measured as the BlackndashScholes value of stock-option grants plus salary and bonus) that comes from stock-optiongrants We choose this reduced-form specication because our interest isin the composition of CEO compensation However the factors thataffect the composition of compensation might also affect the levels andit will be important to keep this in mind in interpreting our results

Optimal-contracting theory generally makes predictions about the in-centives provided by the total amount of rm stock and stock optionsowned by the CEO We choose to model the ow of option grants ratherthan the ultimate performance sensitivity of the CEOrsquos compensationand stock and stock-option holdings because we have in mind anadjustment-cost model in which it takes time for a CEOrsquos holdings ofstock and stock options to reach the optimal level11 In our regressionswe include two measures of the current performance sensitivity of theCEOrsquos wealth to rm performance as explanatory variables to reect thedistance of the CEOrsquos contract from the optimal contract

Table 3 presents results in which the tax effects are identied solely byvariation in tax rates over time The fraction of annual compensation paidin stock options is regressed on tax variables corporate governance vari-ables contract-theory variables and variables reecting stock market re-turns Both of the regressions in this table include rm xed effects Incolumn (1) the fraction in annual compensation paid in stock options isregressed on the log difference in payoff (the combined payoff to both therm and the CEO) from receiving compensation in stock options ratherthan salary and bonus where the payoff calculations are based on statu-tory marginal tax rates This variable is dened as the log difference be-tween equation (3) and equation (6) in section 4 The coefcient on the taxvariable is large and statistically signicant at the 95-percent level Thepoint estimate implies that a one-percent increase in the payoff differencebetween stock options and salary and bonus results in a 24-percentage-point increase in the share of compensation paid in stock options

The corporate governance variables all have the predicted sign Firmswith a higher fraction of their shares owned by institutional investorsuse more performance-based pay (stock options) Firms with large corpo-rate boards or a large share of inside directors are less likely to use stockoptions

The two contract-theory variables are the dollar change in CEO wealth

11 See Core and Guay (1999) for evidence that rms adjust in this way

TABLE 3Compensation Regressions that Rely on Variation in

Tax Rates over Time

(1) (2)

Tax variablesln (difference in payoff)(a) 2393 (0349)ln (12statutory personal 20015 (0034)

rate)ln (12statutory cap gains 0330 (0171)

rate)ln (12statutory corporate 0399 (0148)

rate)

Corporate governance variablesFraction of rmrsquos shares 0043 (029) 0052 (0028)

owned by institutionalinvestors

Size of the rmrsquos board of 20002 (001) 20003 (0001)directors

Fraction of directors who 20082 (033) 20092 (0030)are inside or gray

Contract-theory variablesD (CEO wealth) per $1000 20261 (0095) 20247 (0087)

in FV(b)

D (CEO wealth) per 10 in 2583 (189) 2554 (201)FV(c)

Volatility(d) 521 (424) 261 (369)ln (rm market value) 0019 (0006) 0019 (0006)

Stock market return variablesLagged annual rm stock 0006 (0009) 0003 (0008)

market returnLagged annual SampP 500 re- 20035 (0023) 20029 (0023)

turn

Regression diagnosticsTotal sample size 5179 5179Firms 427 427Adjusted R2 0313 0311

Dependent variable share of compensation paid in options Standard errors in parentheses Bothregressions in this table include rm xed effects

(a) Log difference in payoff is the log difference in combined rm and CEO after-tax share betweenpaying compensation in options and paying salary

(b) This variable is the dollar change in CEO wealth per $1000 change in rm market value (seeJensen and Murphy 1990)

(c) This variable in the dollar charge in CEO wealth per 10-percent change in rm value (see Hall andLiebman 1998)

(d) Standard deviation of rm market value in trillions

28 Hall amp Liebman

per $1000 change in rm value (Jensen and Murphy 1990) and thedollar change in CEO wealth per 10-percent change in rm value (Halland Liebman 1998) As discussed in Baker and Hall (1998) these twowealth sensitivity measures reect two different concepts of how closelyaligned the managersrsquo incentives are with the interests of the sharehold-ers In the regression they are meant to measure how far the CEOrsquosexisting contract and ownership of stock and stock options is from theoptimal level The negative coefcients on the two variables suggeststhat rms that are below their optimal pay-to-performance sensitivityare more likely to give more stock options in the current period

The second column replaces the tax variable motivated by deferraladvantages with the three statutory rates The corporate after-tax sharehas a positive coefcient as we would expect because the tax advantageof stock options increases as the corporate rate falls The coefcient onthe personal tax rate is small and statistically insignicant The sign ofthe coefcient on the capital gains rate is the opposite of that predictedby tax-avoidance theory

Note that these results rely on time-series variation alone This is not avery convincing test Since the log difference in payoffs mostly rises overthe time period covered in our sample as does the share of compensa-tion paid in options it is possible that the tax variable is simply reectingother factors that were changing over this time period that we have notincluded in our regression Therefore it is important to look at specica-tions that rely on cross-sectional variation in the tax variables

Table 4 contains specications that rely on cross-sectional variation incorporate tax rates The rst column repeats the specication from col-umn (1) of Table 3 but uses the trichotomous measures of rm corporatetax rates in place of the statutory rate in calculating each rmrsquos logdifference in payoff from paying compensation in options rather than insalary and bonus As in the previous table we nd the positive coef-cient on the log difference in payoff just as the theory suggest How-ever the coefcient estimate is now much smaller suggesting that aone-percent increase in the log difference in payoff produces a 01-percentage-point increase in the share of compensation paid in options

Introducing cross-sectional variation in the dependent variable doesnot eliminate the potentially spurious correlation over time between thetax variable and the share of compensation paid in options The secondcolumn of Table 4 isolates the pooled cross-sectional variation in corpo-rate tax rates by adding time dummies and dropping the rm xedeffects In this specication the coefcient on the tax variable is indistin-guishable from zero The coefcient on the institutional variable contin-ues to be large and statistically signicant The third column includes

Taxation of Executive Compensation 29

both rm xed effects and time dummies which tests for a relationshipbetween within-rm changes in the dependent variable and within-rmchanges in each explanatory variable In this specication the coefcienton the tax variable is again indistinguishable from zero The coefcientson the corporate governance variables are also statistically insignicant

The third column contains a specication that treats the Tax ReformAct of 1986 (TRA 86) as a natural experiment For rms and CEOs facingthe statutory corporate tax rate TRA 86 increased the relative payoff topaying compensation in options by 17 percent For a rm facing a zeromarginal tax rate it reduced the relative payoff by 45 percent The reduc-tion occurred because TRA 86 reduced the personal tax rate makingsalary and bonus more attractive while the decreased corporate rate hadno effect on zero-tax-rate rms Thus we would expect rms facing zeroor low marginal tax rates to reduce their use of options after TRA 86while rms facing the statutory rate would be expected to increase themFew rms face zero marginal tax rates year after year Therefore we tryto distinguish between rms that often face low tax rates and rms thatnearly always face the statutory rate We dene low-tax rms as oneswhose average marginal tax rate over 1984 to 1986 was below 02412

Then we ran a difference-in-differences regression to see if high-taxrms increased their use of options more than low-tax rms The interac-tion between post-TRA 86 and high corporate tax rates is the key vari-able The estimated coefcient on this variable is positive but small andnot statistically different from zero at the 95-percent level

It is important to note that in this specication as well the main corpo-rate governance variable the share of the rmrsquos shares owned by institu-tional investors performs as theory would predict The variable appearsto be quite robust Table 5 presents regressions of the share of compensa-tion paid in stock options on the fraction of the rmrsquos shares owned byinstitutional investors Since we have already established that these vari-ables have a strong longitudinal relationship the regressions here aredesigned to focus on the cross-sectional relationship between these twovariables The rst and third columns present regressions using data onlyfrom 1994 and nd a coefcient of roughly 02 (column 1 contains noother covariates while column 3 includes all of the other nontax covari-ates from the regressions in Tables 3 and 4) Since the median share of armrsquos stock owned by institutional investors in our sample rises from 19percent in 1980 to 49 percent in 1994 this coefcient would imply that this

12 This cutoff was chosen after inspecting the distribution of tax rates and essentiallyseparates rms that are always at or near the statutory rate from ones that are sometimesbelow it

TA

BL

E4

Com

pens

atio

nR

egre

ssio

nsth

atR

ely

onC

ross

-Sec

tion

alV

aria

tion

inC

orpo

rate

Tax

Rat

es

(1)

(2)

(3)

(4)

Tax

vari

able

s(a)

ln

(dif

fere

nce

inp

ayof

f)0

099

(00

41)

20

038

(00

36)

20

005

(00

44)

Pos

t-T

RA

862

000

5(0

013

)H

igh

corp

orat

eta

xra

te0

018

(00

17)

(Pos

t-T

RA

86)

3hi

gh0

024

(00

16)

corp

orat

eta

xra

te

Cor

por

ate

gove

rnan

ceva

riab

les

Frac

tion

of

rmrsquos

shar

es0

109

(00

27)

011

4(0

015

)2

000

9(0

030

)0

087

(00

34)

owne

dby

inst

itu

tion

alin

vest

ors

Size

ofth

e

rmrsquos

boar

dof

20

004

(00

01)

20

001

(00

01)

20

002

(00

01)

000

1(0

002

)d

irec

tors

Frac

tion

ofd

irec

tors

wh

o2

012

4(0

033

)2

000

3(0

018

)2

004

1(0

034

)0

001

(00

39)

are

insi

de

orgr

ay

Con

trac

t-th

eory

vari

able

sD

(CE

Ow

ealt

h)p

er$1

000

20

238

(00

94)

20

167

(00

53)

20

257

(00

96)

20

021

(01

47)

inFV

(b)

D(C

EO

wea

lth)

per

10

in2

560

(17

8)2

592

(10

5)2

594

(19

6)2

147

(04

86)

FV(c

)

Vol

atili

ty(d

)2

48(4

24)

162

(03

4)1

12(4

28)

106

(06

09)

ln(

rmm

arke

tva

lue)

003

3(

006)

000

8(0

003

)0

009

(00

06)

001

2(0

007

)

Stoc

km

arke

tre

turn

vari

able

sL

agge

dan

nual

rm

stoc

k2

000

2(0

009

)0

011

(00

10)

000

6(0

009

)2

001

2(0

013

)m

arke

tre

turn

Lag

ged

annu

alSamp

P50

02

005

6(0

023

)2

005

3(0

051

)2

003

9(0

049

)2

015

4(0

040

)re

turn

Reg

ress

ion

dia

gnos

tics

T

otal

sam

ple

size

5179

5179

5179

2534

Firm

s42

742

739

6A

dju

sted

R2

031

009

032

005

Firm

xed

effe

cts

Yes

No

Yes

No

Tim

ed

um

mie

sN

oY

esY

esN

o

Dep

end

ent

vari

able

isfr

acti

onof

ann

ualc

omp

ensa

tion

from

stoc

kop

tion

sSt

and

ard

erro

rsin

par

enth

eses

(a)

All

thre

esp

eci

cati

ons

use

tric

hoto

mou

sco

rpor

ate

tax

rate

sT

he

sam

ple

inco

lum

n(3

)u

ses

only

the

year

s19

84to

1988

sin

cew

ear

eu

sing

1986

asa

nat

ural

exp

erim

ent

(b)

Thi

sva

riab

leis

the

dol

lar

chan

gein

CE

Ow

ealt

hp

er$1

000

chan

gein

rm

mar

ketv

alu

e(s

eeJe

nse

nan

dM

urp

hy

1990

)

(c)

Thi

sva

riab

lein

the

dol

lar

char

gein

CE

Ow

ealt

hp

er10

-per

cent

chan

gein

rm

valu

e(s

eeH

alla

nd

Lie

bman

199

8)

(d)

Stan

dar

dd

evia

tion

ofr

mm

arke

tva

lue

intr

illio

ns

32 Hall amp Liebman

TABLE 5Large Institutional Investors and Stock-Option Compensation

The Cross-sectional Relationship between Institutional Investors andOption Grants

(1) (2) (3) (4)

Coefcient on fraction of 217 (067) 145 (014) 177 (069) 120 (015)rmrsquos shares owned byinstitutional investors

Sample period 1994 1980ndash1994 1994 1980ndash1994Contract-theory and stock No No Yes Yes

market covariatesTime dummies No Yes No Yes

The dependent variable is the share of compensation paid in options

factor alone can explain 6 percentage points of the 23-percentage-pointincrease in the median share of compensation paid in options that oc-curred over this time period Of course this variable is only one measureof the strengthening of corporate governance so the total impact of im-proved corporate governance could be greater The second and fourthcolumns of the table use the entire 1980ndash1994 sample and include timedummies to isolate the cross-sectional correlations Again there is evi-dence of a strong cross-sectional relationship between the two variablesalthough the coefcients are slightly smaller perhaps reecting a weakercorporate-governance role in the earlier period In sum there is a strongrelationship in both longitudinal and cross-sectional data between theinstitutional-investor variable and the options-percentage variable13

7 THE MILLION-DOLLAR RULE

In 1993 section 162(m) of the Internal Revenue Code was enactedlimiting the deductibility of executive compensation in excess of onemillion dollars unless the compensation was performance-related Pro-ponents of this legislation argued that it would slow the growth ofexecutive pay and tighten the link between rm performance and execu-tive compensation

The new provision became effective for tax year 1994 and applies tothe CEO and the other four most highly compensated executives in eachcompany Pay that is performance-related such as stock-option grants

13 The one specication in which the fraction owned by institutional investors is notsignicant is in column (2) of Table 4 a regression that includes both xed effects and timeeffects

Taxation of Executive Compensation 33

and bonuses paid for meeting clear performance goals is not affected bythis regulation provided that it has been approved by shareholdersThus the main impact of the provision is to limit the deductibility ofsalary in excess of a million dollars Woodlock and Antenucci (1997)studied the proxy statements of 376 rms and documented that mostrms responded to the new law by qualifying top executive pay asperformance-related Perry and Zenner (1999) identied 25 rms thatreduced salaries from above $1 million to below $1 million and foundthat 23 of the rms cited section 162(m) as their reason for doing so

In this section of the paper we investigate whether this provision hasaffected the level or structure of executive pay using the Execucompdata base which contains information on the highest-paid executives ineach of the rms in the SampP 500 SampP Midcap 400 and SampP SmallCap600 Between 1993 and 1998 the median salary of the Execucomp execu-tives rose from $230000 to $306000 The median value of total compen-sation (salary bonus and option grants) rose from $470000 to $882000Limiting the sample of CEOs indicates that the total compensation of themedian CEO rose from $11 million to $18 million over the period whilethe median salary rose from $441000 to $550000 Thus executive com-pensation clearly continued to rise at a rapid rate after the implementa-tion of the million-dollar rule though perhaps at a slower rate than itotherwise would have

If the regulation affects behavior it should have the largest impact onrms whose executives are receiving salaries that are above the limit orjust below it Thus we might expect to see rms whose executives arereceiving relatively high salaries to rely more heavily on performance-based pay and less heavily on salary in the years following the regulationPerry and Zenner (1999) and Livingston (1999) both nd that this is thecase This trend can be clearly identied in the Execucomp data set Forexample over 1993 to 1998 the median annualized growth rate of salaryfor executives whose 1993 salary was below $500000 in the previous yearwas 85 percent For executives with salary between $500000 and$700000 it was 64 percent for executives between $700000 and $850000it was 50 percent for executives between $750000 and $1 million it was31 percent and for executives above $1 million it was 00 percent14

The observation that executives at lower levels of salary had highergrowth rates is not very strong evidence of an impact of section 162(m)however because it is possible that low-salary executives would havehad higher growth rates even in the absence of the provision To explore

14 Of the 3137 executives for whom we have complete data between 1993 and 1998 2708had 1993 salaries below $500000 while only 35 had 1993 salaries above $1 million

34 Hall amp Liebman

whether this is the case we use Execucomp data from both before andafter the law change Ideally we would want to have data from manyyears before the change to examine whether there is an underlying pat-tern of faster salary growth for low-salary executives Unfortunately theExecucomp data begin in 1992 and therefore provide us with only the1992ndash1993 growth rate in compensation to use as a baseline15

We run regressions pooling data from 1992ndash1993 through 1997ndash1998of the annual percentage change in different forms of compensation on avariable million that is designed to measure the likelihood that anexecutiversquos compensation will be affected by section 162(m) We denemillion as the minimum of 1 and the executiversquos salary divided by $1million in the previous year Thus any executive with salary at or above$1 million would receive a value of 1 while an executive with a salary of$500000 would be coded as a 0516 We include this variable by itself andalso interacted with a dummy variable (after) that equals 1 for years1994 and after Thus the million variable is designed to pick up anyunderlying relationship between the level of salary and the growth rateof salary while the interacted one will identify any differential growthrate of salary by income in the period after section 162(m) took effect Wealso include a full set of year dummies the natural logarithm of thermrsquos market value and the rmrsquos annual rate of return over the twoprevious years as control variables To reduce the sensitivity of our re-sults to outliers we run quantile regressions and robust regressions

Table 6 contains our results The rst four columns are for all topexecutives while the last four columns focus on CEOs Columns (1) and(2) present estimates from robust regression for the percentage change insalary and the percentage change in total compensation (salary bonusstock grants and stock-option grants) In the salary growth regressionsfor all executives the coefcient on million is negative and signicantindicating that there is an underlying relationship between the level ofexecutive compensation and its subsequent growth rate The coefcienton million 3 after is negative as well suggesting that the negativerelationship between the level of salary and its subsequent growth rateintensied slightly after 1994 The relatively small magnitude suggeststhat an executive with a $1 million dollar salary would see his salarygrow at an annual rate that would be about 06 percent slower than anexecutive earning $500000

15 The HallndashLiebman data set contains only a combined salary and bonus variable makingit impossible to use it to analyze this issue16 We also ran the regressions (not reported) allowing for a nonlinear relationship betweenmillion and compensation growth and obtained substantively similar results

TA

BL

E6

The

Eff

ect

ofth

eM

illi

on-D

olla

rR

ule

onSa

lary

and

Tot

alC

ompe

nsat

ion

All

top

exec

uti

ves

CE

Os

only

Rob

ust

regr

essi

onQ

uan

tile

regr

essi

onR

obu

stre

gres

sion

Qu

anti

lere

gres

sion

D

in

Din

D

in

Din

D

in

Din

D

in

Din

sala

ryto

talc

omp

(a)

sala

ryto

talc

omp

sa

lary

tota

lcom

p

sala

ry

tota

lcom

p

mil

lio

n(b

)2

006

02

035

72

006

32

024

12

008

72

023

12

010

32

024

5(0

005

)(0

032

)(0

006

)(0

027

)(0

010

)(0

073

)(0

011

)(0

056

)m

illi

on

32

001

10

026

20

021

20

015

20

016

014

92

003

40

158

aft

er(c

)(0

005

)(0

033

)(0

006

)(0

029

)(0

010

)(0

076

)(0

011

)(0

057

)N

359

5035

950

359

5035

950

757

17

589

757

17

589

Stan

dar

der

rors

inp

aren

thes

esR

egre

ssio

nsal

soin

clu

de

year

du

mm

ies

stoc

km

arke

tre

turn

and

log

mar

ket

valu

e

(a)

Tot

alco

mp

ensa

tion

incl

ud

essa

lary

bon

us

rest

rict

edst

ock

and

stoc

k-op

tion

gran

ts

(b)

mil

lion

isd

ene

das

min

(1s

alar

yin

pre

viou

sye

ars

div

ided

by$1

mil

lion)

(c)

aft

eris

anin

dic

ator

vari

able

equ

alto

zero

befo

reth

ere

gula

tion

took

plac

e(1

993)

and

one

afte

rwar

d

36 Hall amp Liebman

In the second column the coefcient on million 3 after is essentiallyzero suggesting that any decrease in salary brought about by section162(m) was offset by increases in bonus and stock-option grants Theresults are similar for the median regressions There is some evidence ofa very small slowdown in growth of salary for executives likely to beaffected by section 162(m) but no sign of a change in total compensa-tion Taken together this provides evidence of a very small substitutionof performance-related pay for salary

The last four columns run identical tests but limit the sample to CEOsThe interaction of million 3 after continues to be small and negative inthe salary regressions In contrast the impact of 162(m) on total compen-sation appears to be positive and fairly substantial (about 15 percentagepoints per year) in these regressions That is the increase in bonus andstock options following the million-dollar rule more than offset the verysmall decline in salary

All of these conclusions should be interpreted somewhat cautiouslybecause our data provide us with only one pre-1993 control year Never-theless the data are consistent with section 162(m) having led to a veryminor slowdown in salary growth and one that is dramatically smallerthan what one would conclude without controlling for underlying differ-ential salary growth rates at different levels of salary Since total compen-sation for CEOs did not decline and perhaps increased there is evidenceof a minor substitution of performance-related pay for salary in responseto the regulation

8 EXECUTIVE COMPENSATION AND THEELASTICITY OF TAXABLE INCOME

Even if taxes have only a minor effect on executive compensation thestructure of executive compensation is important for tax policy be-cause high-income taxpayers are sources of signicant revenue and arethe focus of important debates about the efciency cost of taxation

Goolsbee (1999) has recently argued that executive decisions to exer-cise options are highly responsive to intertemporal differences in taxrates created by tax reforms He claims that much of the apparent declinein taxable income for high-income taxpayers between 1992 and 1993documented by Feldstein and Feenberg (1996) can be explained by execu-tives shifting option gains into 1992 in order to avoid the higher marginaltax rates that went into effect in 1993 By extension he implies that thehigh elasticities of taxable income with respect to personal tax ratesestimated by comparing the pre- and post-tax-reform taxable income of

Taxation of Executive Compensation 37

high-income individuals reect intertemporal shifting of income and notpermanent effects of taxation If this is true then the deadweight loss oftaxation is lower as is the cost of progressivity

While Goolsbee presents comprehensive evidence that option gainswere unusually high in 1992 his identication of the impact of tax rates onthe timing of option gains depends on correctly specifying a counter-factual level of options that would have been exercised in the absence ofthe tax change He simply assumes a linear time trend The purpose ofGoolsbeersquos time trend is presumably to account for the increasing use ofoptions over time However since options typically require a vestingperiod of at least a couple of years a much more direct measure of ex-pected option gains can be constructed based on the total value of vestedoptions held by the CEO Since options typically have vesting periods thatexceed two years this total can be treated as exogenous since it will not beaffected by the tax changes

More generally the value of options held by the CEO is likely to havea major effect on his exercise decisions If his options are out of themoney he will not have gains to realize If there has been a recent run-up in the market an optimizing CEO may exercise an unusually largeamount of options in order to diversify his overall portfolio and if in-the-money options are about to expire they will be exercised in thecurrent year

Our evidence suggests that much of the apparent tax shifting caninstead be attributed to stock market performance and the timing of pastoption grants Using our 15-year panel we replicate Goolsbeersquos (1999)result for the 1993 period but show that the impact of taxes is notpresent in other time periods Moreover even the result for the 1993period disappears when appropriate controls for past option grants andstock price appreciation are included

The top three rows of Table 7 present data on the taxable incomes ofCEOs in the HallndashLiebman sample For these results our panel has beennarrowed to include only CEOs whose rmrsquos scal years correspondwith tax years In addition we create a balanced panel for each of thethree tax-reform periods by excluding any rm for which there are notcomplete data for the period surrounding the reform

The fourth row of the table shows that the 1993 tax act raised the topmarginal personal income tax rate from 310 to 396 percent effective fortax year 1993 with a further increase occurring in 1994 due to the uncap-ping of the Medicare payroll tax The top row of the table shows thattaxable income was higher in 1992 than for any other year in our sample(all data are in 1994 dollars) Option gains were particularly high in1992mdashmore than double their 1991 level and 68 percent higher than their

TA

BL

E7

Tax

Ref

orm

Sto

ckM

arke

tP

erfo

rman

cea

ndth

eT

imin

gof

Opt

ion

Gai

ns

Ave

rage

stoc

km

arke

tM

ean

CE

OM

ean

CE

Ore

turn

for

Mea

nva

lue

taxa

ble

sala

ryan

dM

ean

CE

OT

opfe

der

al

rms

inof

opti

ons

(Val

ue

ofop

tion

Sam

ple

inco

me

bonu

sop

tion

gain

sp

erso

nal

sam

ple

for

held

atbe

gga

ins)

(va

lue

Ref

orm

size

Yea

r($

mil

)($

mil

)($

mil

)M

TR

pre

vye

ar(

)of

year

ofop

tion

she

ld)

ER

TA

8127

919

800

874

064

80

163

7025

80

460

354

1981

083

10

672

009

869

245

070

61

3919

820

872

067

90

099

5011

90

686

144

1983

099

00

724

025

250

355

116

62

16

TR

A86

295

1985

119

70

837

024

050

76

116

02

0719

861

448

091

70

345

5042

81

776

194

1987

165

50

959

059

338

519

12

229

266

1988

163

21

058

039

028

07

237

71

6419

891

656

105

20

414

2817

92

345

176

1993

272

1990

209

51

028

061

828

232

288

02

15T

axA

ct19

911

875

102

00

547

312

01

268

82

0319

922

678

107

01

250

3146

64

552

275

1993

232

41

209

074

339

620

64

299

173

1994

196

41

262

037

842

515

45

113

074

Sam

ple

isH

allndash

Lie

bman

rm

sw

ith

Dec

embe

r

scal

year

sth

atha

dco

mp

lete

dat

afo

ron

eof

the

thre

eta

x-re

form

per

iod

sst

ud

ied

Taxation of Executive Compensation 39

1994 level These patterns are consistent with Goolsbeersquos claim that tax-payers responded to the anticipated increase in marginal tax rates byshifting gains into 1992

The evidence from the other two tax reforms is less clear suggestingthat there might be factors other than tax rates determining the timing ofoption gains ERTA 1981 reduced the top marginal tax rate from 70percent in 1980 to 50 percent in 1982 However there is no sign thatoption gains were reduced in 1980 and 1981 and then increased in 198217

There was a big increase in option gains in 1983 however Similarly thehighest year of stock-option gains in the period around 1986 was 1987 astrange year to take gains since taxpayers would have known that thepersonal rate would be 10 percentage points lower if they waited untilthe following year18

What all the peak years of option gains have in common is that theyfollow years of strong stock market performance The fth row of Table 7shows the average stock market return during the previous year for therms in our sample The large value of option gains in 1983 followed a355-percent increase in the stock market The 1992 boom in option gainsfollowed a 466-percent increase in the stock market the more modest1987 peak in option gains followed a 19-percent gain (and a 43-percentgain two years before)

Strong stock market performance raises the value of options availableto exercise The sixth row of the table shows the mean value of optionsheld by CEOs at the beginning of the year All three of the peak years ofoption gains correspond with peak years in terms of the value of optionsavailable for exercise Since stock market appreciation during the yearalso increases the value of options available to exercise later in that yearwe need to allow for current-year stock market appreciation as well

Table 8 contains Goolsbee-style regressions for each of the tax re-forms The rst two columns reprint the results from Goolsbee (1999)The third and fourth columns replicate his results in our sample Inparticular with a specication similar to his we are able to replicate hisnding that the current after-tax share has a positive coefcient and thefollowing yearrsquos after-tax share has a coefcient of roughly negative oneColumns (5) through (8) show that this pattern was unique to the 1993tax reform Indeed in the TRA 86 regressions the coefcient on the

17 As we explained above for executives covered by the maximum tax on earned incomethere was not a decline in tax rates between 1981 and 198218 It is possible that the large stock market decline in October 1997 may have promptedsome people to exercise options earlier than they otherwise would have However anincrease in option exercises typically follows stock market increases not declines

TA

BL

E8

Att

empt

sto

Rep

lica

teth

eT

ax-S

hift

ing

Res

ult

inO

ther

Tim

eP

erio

ds

Goo

lsbe

eG

ools

bee

1993

1993

Ent

ire

Ent

ire

regr

essi

onre

gres

sion

refo

rmre

form

TR

A86

TR

A86

ER

TA

81E

RT

A81

1980

ndash199

419

80ndash1

994

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

ln[1

2ta

x(t)

]1

244

108

20

455

032

62

017

62

017

22

016

72

016

70

026

20

068

(01

34)

(01

27)

(05

37)

(05

42)

(02

10)

(02

09)

(00

90)

(00

91)

(00

67)

(00

65)

ln[1

2ta

x(t1

1)]

20

867

20

764

21

214

20

917

047

80

722

20

103

004

30

176

018

3(0

118

)(0

112

)(0

491

)(0

499

)(0

157

)(0

174

)(0

072

)(0

072

)(0

068

)(0

066

)

Tim

e0

135

007

62

000

52

036

40

041

000

80

094

003

50

056

002

7(0

008

)(0

008

)(0

049

)(0

049

)(0

028

)(0

028

)(0

025

)(0

026

)(0

003

)(0

003

)

ln(m

arke

tva

lue)

066

10

489

023

70

334

039

5(0

015

)(0

058

)(0

039

)(0

042

)(0

017

)

Yea

rs19

91ndash1

995

1991

ndash199

519

90ndash1

994

1990

ndash199

419

85ndash1

989

1985

ndash198

919

80ndash1

983

1980

ndash198

319

80ndash1

994

1980

ndash199

4

Ad

just

edR

20

720

770

610

630

690

760

800

810

630

68

Obs

erva

tion

s13

856

132

371

385

136

41

470

143

31

113

106

23

968

385

9

Dep

end

ent

vari

able

isth

elo

gof

taxa

ble

inco

me

All

regr

essi

ons

con

tain

rm

xe

def

fect

sG

ools

bee

regr

essi

on(2

)al

soco

ntai

nstw

oot

her

vari

able

sno

tsh

own

inth

eta

ble

Stan

dar

der

rors

inp

aren

thes

es

Taxation of Executive Compensation 41

TABLE 9Alternative Explanations for the Timing of Option Exercises

1993 Entirereform TRA 86 ERTA 81 1980ndash1994

(1) (2) (3) (4)

ln[12tax(t)] 0108 0111 20051 0201(0324) (0136) (0048) (0057)

ln[12tax(t11)] 20134 0573 0125 0150(0336) (0156) (0053) (0066)

ln(market value) 0418 0177 0338 0418(0061) (0039) (0046) (0163)

Firmrsquos stock market return (t) 0298 0268 0214 0258(0044) (0045) (0031) (0023)

Firmrsquos stock market return 0162 0206 0031 0128(t21) (0041) (0008) (0029) (0021)

ln(value of options held by 0004 0008 0001 0009CEO at start of year) (0003) (0003) (0002) (0001)

Years 1990ndash1994 1985ndash1989 1980ndash1983 1980ndash1994Adjusted R2 064 071 082 068Observations 1349 1427 1036 3812

The dependent variable is the log of taxable income All regressions contain rm xed effects Stan-dard errors in parentheses

following yearrsquos tax rate is positive and signicant Columns (9) and (10)pool all three reforms and fail to replicate the pattern that would reectintertemporal shifting19

It is possible that the ease with which one can shift the timing ofoption gains is asymmetric It may be harder to shift option gains for-ward in time since some options expire and therefore must be exercisednow In addition options were a much smaller portion of compensationin the early 1980s so perhaps the ERTA 81 results are not surprisingNonetheless these results cast doubt on the proposition that this is thechannel through which spurious estimates of the elasticity of taxableincome have occurred for previous tax reforms and they suggest thatwe should look harder for alternative explanations for the timing ofoption gains

Table 9 attempts to predict options gains using the lagged and currentincrease in a rmrsquos market value and the value of the options held by theCEO at the start of the year The two stock market variables stronglypredict option gains while the value of options held has the correct sign

19 The results also fail to suggest large permanent effects of marginal tax rates on taxableincome

42 Hall amp Liebman

in all of the specications and is signicant in the pooled sample eventhough it is highly correlated with the two stock market return variablesIn contrast the tax-timing explanation fails even in the 1993 period

It is important to realize that our results do not invalidate Goolsbeersquoscritique of claims that aggregate revenue data from 1992 and 1993 provethat elasticities of taxable income are large We have simply given analternative explanation for why taxable incomes were unusually high in1992 However our results do cast doubt on the conclusion that largetiming shifts are ubiquitous and capable of explaining away the morecareful estimates of taxable income elasticities that have been performedusing micro data (eg Feldstein 1995 Auten and Carroll 1995)

One further point is in order It is sometimes argued that the combina-tion of increased use of options and the booming stock market explainswhy federal tax revenues have been so much higher than expected dur-ing the 1990s However while large option gains increase revenue fromthe personal income tax they produce offsetting deductions for cor-porations The net impact on revenue is only the difference in the twomarginal tax rates times the option gain While the gap between thepersonal rate and the effective corporate rate is nontrivial the impact oflarge option exercises on total revenue is substantially smaller than theamount by which the exercises inate personal tax revenue and reducecorporate tax revenue

9 CONCLUSION

We have described the tax rules for executive pay in detail and analyzedhow changes in various tax rates affect the tax advantages of stock optionsrelative to salary and bonus We nd that there is a moderate tax advan-tage to optionsmdashcurrently about $4 per $100 in pretax executive paymdashand that changes in the tax advantage of options over time have had atmost a modest impact on the composition of pay Corporate governancefactors particularly the role of large investors appear to be more impor-tant in explaining the dramatic increase in option pay

In addition our evidence suggests that more direct attempts to use taxpolicy to inuence executive compensation have had little effect Wend that the million-dollar rule led companies to substitute perfor-mance-based pay for salary But our evidence suggests that this substitu-tion was quite modest and there is no evidence that the total level of paywas reduced Overall although the stock-option explosion has dramati-cally increased the link between pay and performance this change isdue almost entirely to nontax factors

Taxation of Executive Compensation 43

REFERENCES

Auerbach Alan J (1996) ldquoThe Dynamic Effects of Tax Law Asymmetriesrdquo Re-view of Economic Studies 53451ndash405

mdashmdashmdash and James Poterba (1987) ldquoTax Loss Carryforwards and Corporate TaxIncentivesrdquo In The Effects of Taxation on Capital Accumulation M Feldstein(ed) Chicago University of Chicago Press

Altshuler Rosanne and Alan Auerbach (1990) ldquoThe Signicance of Tax LawAsymmetries An Empirical Investigationrdquo Quarterly Journal of Economics 10561ndash86

Auten Gerald and Robert Carroll (1995) ldquoThe Effect of Income Taxes on House-hold Behaviorrdquo Ofce of Tax Analysis US Department of the Treasury

Baker George P and Brian J Hall (1998) ldquoCEO Incentives and Firm SizerdquoNBER Working Paper no 6868 December

Core John and Wayne Guay (1999) ldquoThe Use of Equity Grants to ManageOptimal Equity Incentive Levelsrdquo Draft

Feldstein M (1995) ldquoThe Effect of Marginal Tax Rates on Taxable Income APanel Study of the 1986 Tax Reform Actrdquo Journal of Political Economy 103 (no3)551ndash72

mdashmdashmdash and D Feenberg (1996) ldquoThe Effect of Increased Tax Rates on TaxableIncome and Economic Efciency A Preliminary Analysis of the 1993 Tax RateIncreasesrdquo In Tax Policy and the Economy vol 10 James M Poterba (ed)Cambridge MA MIT Press

Gompers Paul A and Andrew Metrick (1998) ldquoInstitutional Investors andEquity Pricesrdquo NBER Working Paper no 6723

Goolsbee Austan (1999) ldquoWhat Happens When You Tax the Rich Evidencefrom Executive Compensationrdquo NBER Working Paper no 6333

Graham John R (1996) ldquoProxies for the Corporate Marginal Tax Raterdquo Journal ofFinancial Economics 42187ndash221

Hall Brian J (1999) ldquoThe Design of Multi-year Stock Option Plansrdquo Journal ofApplied Corporate Finance 12(2)97ndash106

mdashmdashmdash and Jeffrey B Liebman (1998) ldquoAre CEOs Really Paid Like Bureau-cratsrdquo Quarterly Journal of Economics CXII(no 3)653ndash691

Jensen Michael C and William M Meckling (1976) ldquoTheory of the Firm Mana-gerial Behavior Agency Costs and Ownership Structurerdquo Journal of FinancialEconomics III305ndash360

mdashmdashmdash and Kevin J Murphy (1990) ldquoPerformance Pay and Top-ManagementIncentivesrdquo Journal of Political Economy XCVIII225ndash264

Kaplan Steven N (1997) ldquoThe Evolution of US Corporate Governance WeAre All Henry Kravis Nowrdquo Draft

Livingstone Jane R (1999) ldquoExecutive Compensation Contracting and Re-sponses to Increased Tax Costsrdquo Louisiana State University June

Miller Merton H and Myron S Scholes (1982) ldquoExecutive CompensationTaxes and Incentivesrdquo In Financial Economics Essays in Honor of Paul CootnerWilliam F Sharpe and Cathryn M Cootner (eds) Princeton NJ Prentice-Hall

Murphy Kevin J (1999) ldquoExecutive Compensationrdquo forthcoming in OrleyAshenfelter and David Card (eds) Handbook of Labor Economics Vol 3 NorthHolland

Perry Todd and Marc Zenner (1999) ldquoPay for Performance Government Regu-

2 Hall amp Liebman

of pretax compensation to the executive A more convincing story for thedramatic explosion in stock options involves changes in corporate gover-nance and the market for corporate control For example there is astrong correlation between the fraction of shares held by large institu-tional investors and the fraction of executive pay in the form of stockoptions a result that holds both longitudinally and cross-sectionallySecond we nd evidence that the million-dollar rule (which limited thecorporate deductibility of non-performance-related executive compensa-tion to $1 million) led rms to adjust the composition of their pay awayfrom salary and toward performance-related pay although our esti-mates suggest that this substitution was minor We nd no evidencethat the regulation decreased the level of total compensation Third weexamine whether there is evidence for signicant shifting of the timingof option exercises in response to changes in tax rates After replicatingGoolsbeersquos (1999) result regarding tax shifting with our data for the 1993tax reform we show that no such shifting occurred in either of the twotax reforms of the 1980s Moreover we nd evidence that much of theunusually large level of option exercises in 1992 was the result of therising stock market rather than the change in marginal tax rates

1 INTRODUCTION

Over the past 15 years there has been a major change in the way thatAmerican executives particularly CEOs are paid For many executivesannual stock-option grants are now greater than cash compensation(salary and bonus) Annual changes in CEO wealth from revaluations ofstock and stock-option holdings completely swamp cash compensationand provide substantial pay-to-performance sensitivity All of this is adramatic change from the early 1980s when the median stock-optiongrant to top executives was zero

In this paper we examine the extent to which tax policy has inu-enced the composition of executive compensation and discuss the impli-cations of rising stock-based pay for tax policy Because top executivesmanage assets worth billions of dollars their compensation arrange-ments and the incentives they face are of substantial importance to theperformance of the US economy Because top executives have veryhigh incomes their responsiveness to taxation has important revenueand efciency implications

We conduct a broad analysis of the taxation of executives We begin bystudying how tax rates affect the degree to which options are favoredrelative to cash and how the tax advantage of options has changed overtime in response to changes in corporate personal and capital gains tax

Taxation of Executive Compensation 3

rates Although the tax advantage of options has approximately doubledsince the early 1980s options currently have only a slight tax advantagerelative to cashmdashapproximately $4 per $100 of pretax compensation

We then analyze what we believe to be the three central policy ques-tions regarding the taxation of executive pay First we examine theextent to which the stock-option explosion has been inuenced by themany changes in tax rates over the past 20 years The evidence suggeststhat changes in taxation have likely had a very modest inuence on theoption explosion Instead changes in corporate governance especiallyin the role of large institutional investors appear to have provided themain impetus for the increase in stock-based pay

Second we examine the effectiveness of tax policies aimed at curbingwhat is deemed by some to be excessive levels of executive compensa-tion Section 162(m) of the Internal Revenue Code (the million-dollarrule) which was enacted in 1993 put a $1-million limit on the deducti-bility (against corporate prots) of non-performance-related executivepay Although we nd evidence that this rule led to a shift in the compo-sition of paymdashaway from salary and toward more performance-relatedbonuses and stock optionsmdashour evidence suggests that the magnitudeof this substitution was small We nd no evidence that the million-dollar rule decreased total executive compensation

Third we analyze the degree to which the tax code is efcient inraising tax revenue from top executives The stock-option explosion hasled to a new and important way for executives to lower their taxes inresponse to changes in tax rates by timing their stock-option gains Forexample in 1993 it was widely reported in the press that well-knownCEOs such as Disneyrsquos Michael Eisner pushed their option gains into1992 in order to avoid paying the higher personal income-tax rates imple-mented in 1993 In an important paper Goolsbee (1999) argues that thetax shifting between 1992 and 1993 was enormous and was the directresult of the increase in marginal tax rates during this period Afterreplicating Goolsbeersquos evidence regarding tax shifting with our data forthe 1993 tax reform we show that no such shifting occurred in responseto either of the two tax reforms of the 1980s Moreover our evidenceindicates that the stock market run-up in 1991 and 1992 was more impor-tant than the change in marginal tax rates in causing the large optiongains observed in 1992

This paper proceeds as follows In the next section we discuss trendsin the level and performance sensitivity of executive compensation Insection 3 we describe the tax and accounting rules concerning executivecompensation In section 4 we analyze how taxes affect the degree towhich options are favored relative to cash In section 5 we examine how

4 Hall amp Liebman

the tax advantage of options has changed over time In section 6 weprovide evidence on the effect of taxation on the composition of execu-tive compensation Section 7 contains empirical analysis of the million-dollar rule Section 8 contains evidence on tax shifting and option gainsSection 9 concludes

2 TRENDS IN TOP EXECUTIVE PAY

In this section we document how top executive pay has changed overtime and discuss how this change has caused the sensitivity of CEOwealth to rm market value to increase substantially There has been alarge increase in the level of CEO pay since 1980 and this growth hasbeen driven by the dramatic increase in stock-option grants during thistime (Hall and Liebman 1998) Although salary and bonuses nearlydoubled over the period in ination-adjusted terms the mean value ofstock-option grants increased by 683 percent The percentage increase inthe median stock-option award cannot be calculated because the me-dian stock-option grant was zero in 1980 The median CEO did notreceive an annual stock-option grant until 1985 Today nearly all topexecutives of large companies receive stock options and the averagestock-option grant is now larger for most top executives than salary andbonus combined

The (ination-adjusted) growth rate of CEO pay since 1980mdashmeasurednarrowly (cash pay grew at an annual rate of 5 percent per year) orbroadly(cash plus option grants grew at almost 9 percent per year) or very broadly(total compensation including stock and stock-option appreciation grewat 115 percent per year)mdashhas been large relative to virtually all othergroups Indeed the growth rate of CEO pay since 1980 has been higheven relative to the pay increases of other high-income earners For ex-ample the cutoff point for being in the top 05 percent of adjusted grossincome (AGI) increased by about 37 percent per year about half the ratefor direct CEO compensation (excluding stock and option appreciation)The only workers who appear to have had faster compensation growththan CEOs are other ldquosuperstarsrdquo The annual pay of professional base-ball players increased by approximately 98 percent per year and that ofprofessional basketball players by 139 percent per year

The increase in stock options has led to a large increase in the equityholdings of top executives and this in turn has led to a dramatic increasein the responsiveness of executive wealth to rm performance Nearlyall of the pay-to-performance sensitivity of executive compensationcomes from equity holdings for a given increase in shareholder valuechanges in the value of an executiversquos stock and stock options are more

Taxation of Executive Compensation 5

than 50 times larger than changes in salary and bonus (Hall and Lieb-man 1998)

As a concrete example the estimates in Hall and Liebman (1998) implythat a 10-percent increase in rm value (of the median company in oursample) leads the company to increase the CEOrsquos salary and bonus byabout $25000 However this same 10-percent increase in shareholdervalue translates into $125-million increase in the value of the CEOrsquosstock and stock-option holdings1

The dramatic rise in the link between CEO wealth and rm perfor-mance can be seen in Figure 1 which shows how two measures of thislink have increased since 19802 The rst measure the JensenndashMurphy(1990) sharing rate (shown on the left scale) is the change in CEO wealthfor a $1000 change in rm value The second measure is the change inCEO wealth for a 10-percent change in rm value (see Baker and Hall1998) Since both measures are strongly affected by rm size (the formerhas a negative correlation and the latter has a positive correlation) thepay-to-performance changes over time are estimated with regression(quantile) analysis that controls for changes in the distribution of rmsizes in the sample over time The gure therefore shows the increasein the pay-to-performance measures over time for a constant-size rmin this case a $1-billion rm (in constant 1998 dollars) The striking fact isthat both measures of the pay-to-performance link have increased bynearly a factor of 10 since 1980 These pay-to-performance increases areeven larger than those we reported in our earlier paper that analyzed theperiod 1980 to 1994 because of the large increase in stock-option grantscombined with the strong stock market performance in 1994 to 1998

3 TAX AND ACCOUNTING RULES

Stock options give an executive the right but not the obligation to buy ashare of the companyrsquos stock at a prespecied pricemdashthe exercise or strikeprice Typically options cannot be exercised immediately That is theyvest (become owned by the executive who can then exercise if he or she

1 Stock-option grants are also very sensitive to changes in rm performance mostly be-cause many grants are multiyear plans that hold the yearly number of options constantand the same number of at-the-money options are worth more when the stock price ishigher and vice versa If stock-option grant sensitivity is also included then about 91percent of pay-to-performance sensitivity comes from stock and stock-option revaluations7 percent comes from stock-option grant changes and less than 2 percent comes fromchanges in salary and bonus (Hall 1999)2 Both measures include only the link created by CEO holdings of stock and stock optionsand ignore the smaller amount of pay-to-performance sensitivity that operates throughchanges in salary bonus and stock-option grants

FIG

UR

E1

Pay

toP

erfo

rman

ceT

heL

ink

betw

een

CE

OW

ealt

han

dSh

areh

olde

rV

alue

Sinc

e19

80

Not

eT

he

anal

ysis

cont

rols

for

the

chan

ging

com

pos

itio

nof

rm

size

over

tim

eT

hu

sth

ese

esti

mat

essh

owth

ep

ay-t

o-p

erfo

rman

celin

kfo

ra

sim

ilar

-siz

edco

mp

any

over

tim

ein

this

case

aco

mp

any

wit

ha

mar

ket

valu

eof

$1bi

llion

Taxation of Executive Compensation 7

wishes) slowly over time Common vesting periods are in the three- tove-year range and options usually vest linearly (eg a four-year op-tion vests at 25 percent at the end of each year) An executive typicallyloses any unvested options upon departure Although options may beexercised as soon as they vest they do not have to be exercised untilthey expire or mature Almost 85 percent of stock-option plans have aterm of exactly ten years with virtually all of the remainder being in theve- to ten-year range About 95 percent of options are granted at themoney or at fair market value which means that the exercise price at grantdate is set equal to the stock price at grant date The remaining 5 percentare either discount options (so-called in-the-money options where theexercise price is below the stock price at grant date) or premium options(so-called out-of-the-money options where the exercise price is abovethe stock price at grant date) The holders of options typically do nothave dividend rights or voting rights even on vested options3

31 Tax Rules and Stock-Option CompensationUnlike salary and bonus stock-option grants are typically an untaxedevent at the time of grant For the most widely used optionsmdashnonquali-ed stock options (NQSOs)mdashexecutives are taxed at the personal incometax rate on option prots (the difference between that stock price and theexercise price times the number of options) when the options are exer-cised The company receives a parallel deduction against corporate in-come at that point If the executive continues to hold the shares afterexercise any subsequent appreciation is taxed at the capital gains rate inthe usual way In 1993 an additional feature was added to the tax code[Internal Revenue Code section 162(M)] that disallowed a corporate de-duction for any executive pay above $1 million that is not performance-based While this rule affects executive salaries most bonuses qualify asperformance-based and standard stock options automatically qualifyTherefore this provision gives companies with highly paid executives anincentive to give more pay in the form of bonuses and stock options asubject we return to in section 7 A summary description of the tax (andaccounting) treatment of cash and option compensation is in Table 1

A far less common type of option which is estimated to account forabout 5 percent of option grants is the incentive stock option (ISO) WhileISOs are similar to NQSOs in their design they are crucially different intwo respects First they have an annual cap of $100000 per executive

3 See Murphy (1999) for details about stock options and Miller and Scholes (1982) on taxincentives

TA

BL

E1

Tax

and

Acc

ount

ing

Tre

atm

ent

ofE

xecu

tive

Com

pens

atio

nC

ash

Pay

men

tsve

rsus

Stan

dard

Exe

cuti

veO

ptio

ns

Tax

atio

nat

Tax

atio

nat

Acc

ount

ing

Com

pens

atio

nD

escr

ipti

onp

erso

nall

evel

corp

orat

ele

vel

trea

tmen

t

Sala

ryan

dbo

nu

sC

ash

pay

men

tsm

ade

Tax

edas

ord

inar

yFu

llyd

edu

ctib

leS

ince

Exp

ense

dag

ains

tea

rn-

toC

EO

in

com

e19

93a

ny

pay

over

$1in

gsin

the

usu

alw

ay

mill

ion

isno

td

educ

t-ib

leu

nles

sit

isp

er-

form

ance

-bas

ed

Non

-qua

lie

dst

ock

Op

tion

sgr

ant

the

ex-

The

gran

ting

ofan

op-

Apa

ralle

ltax

ded

uc-

The

reis

no

exp

ense

rec-

opti

ons

(NQ

SOs)

ecut

ive

the

righ

tbu

tti

onis

typ

ical

lyan

un-

tion

atth

eco

rpor

ate

ogni

tion

(atg

ran

tex

er-

not

the

obli

gati

ont

ota

xed

even

tU

pon

exer

-le

veli

sge

nera

llyal

-ci

seo

rsa

le)

for

op-

buy

shar

esof

stoc

kat

cise

opt

ion

pro

ts

are

low

edu

pon

exer

cise

tion

sif

as

isu

sual

lyan

exer

cise

pric

e(t

ypi-

taxe

dat

the

ord

inar

y(f

orth

eam

ount

ofop

-th

eca

set

he

opti

ons

cally

the

curr

ent

stoc

kra

teI

fth

eex

ecu

tive

tion

pro

ts

)O

pti

ons

are

not

dis

cou

nted

(in

pri

ceat

the

dat

eof

cont

inu

esto

hold

onto

auto

mat

ical

lyqu

alif

yas

the

mon

eyat

gran

tgr

ant)

Th

eop

tion

sth

esh

ares

the

exec

u-

perf

orm

ance

-bas

edp

ayd

ate)

and

the

exer

cise

typ

ical

lyve

stov

era

3ndashti

veis

taxe

din

the

and

ther

efor

ear

eno

tpr

ice

and

num

ber

ofop

-5-

year

per

iod

The

typ

i-us

ualw

ay(a

tth

eca

pi-

subj

ect

toth

e$1

-ti

ons

iskn

own

atgr

ant

calm

atu

rity

is10

year

sta

lgai

nsra

te)

onan

ym

illio

nca

p

dat

eFo

rd

isco

unt

op-

but

the

man

ager

may

furt

her

app

reci

atio

nof

tion

sth

ed

iffe

renc

ebe

-ex

erci

seea

rly

Som

eth

esh

are

pri

ce

twee

nth

est

ock

pri

ceop

tion

sar

edi

scou

ntop

-an

dth

em

arke

tp

rice

isti

ons

(the

exer

cise

pri

ceex

pen

sed

over

the

vest

-is

low

erth

anth

est

ock

ing

per

iod

For

opti

ons

pri

ceat

the

gran

td

ate)

w

ith

vari

able

term

sP

rem

ium

opti

ons

are

(eg

a

vari

able

exer

-th

ere

vers

e(t

heex

er-

cise

pri

cev

esti

ngth

atci

sep

rice

isgr

eate

ris

tied

top

erfo

rman

ce)

than

the

stoc

kp

rice

atth

eop

tion

sar

em

arke

dgr

ant

dat

e)

tom

arke

tan

dex

-p

ense

dd

uri

ngth

eti

me

betw

een

gran

tan

dex

erci

se

Not

eW

eth

ank

com

pen

sati

onco

nsu

ltan

tsSc

ott

Gre

enbe

rgan

dSc

ott

Ols

en(b

oth

ofT

ower

sP

erri

n)

and

Fred

eric

Coo

k(o

fFr

eder

icW

C

ook

and

Ass

ocia

tes)

for

help

fulc

onve

rsat

ion

sin

und

erst

and

ing

the

tax

and

acco

unt

ing

rule

s

10 Hall amp Liebman

Second the tax treatment of ISOs is different ISOs are completely un-taxed at grant or exercise It is only at sale that the executive is taxed andeven then the executive is taxed at the lower capital gains tax rate Thedisadvantage is that the corporation never gets to take a parallel taxdeduction against corporate prots Thus ISOs become more attractiveas the personal tax rate increases and as the corporate tax rate and thecapital gains tax rates fall

Two other related types of compensation restricted stock and stockappreciation rights are worth brief discussion Both however are farless common than standard stock options Restricted stock is payment inthe form of restricted shares the restriction being that the shares vestover time as with options Unlike options the shares typically havevoting and dividend rights With regard to taxation the executive istaxed at the personal rate on the value of the restricted stock as thevesting restrictions lapse However the executive may choose to betaxed at the grant date in which case all subsequent appreciation istaxed at the capital gains rate The dividends paid to the executive aretaxed at the ordinary rate in the usual way The company generallyreceives a parallel deduction equal to the amount of the executiversquosincome when the executive is taxed Unlike stock options restrictedstock is not generally considered to be performance-based and is there-fore subject to the million-dollar rule (unless the vesting of the restrictedstock is performance-based which is sometimes the case)

Stock appreciation rights (SARs) are rights that replicate the payoffs ofstock options with a cash transfer Thus SARs are simpler than options inthat there is no requirement to buy and resell the stock in order to ldquocashoutrdquo SARs generally have the same tax treatment as NQSOs both to theindividual and to the corporation SARs are relatively rare however be-cause they have disadvantageous accounting treatment (described in thenext section) relative to stock options and their main relative advantagevis-a-vis options has been essentially replicated through broker-assistedcashless option exercisesmdashwhereby a broker makes a ldquonanosecondrdquo loanto the executive (to purchase and resell the stock) when the executivewants to ldquocash inrdquo on option prots The tax and accounting treatment ofISOs restricted stock and SARs is summarized in Table 2

32 The Accounting Treatment of OptionsUnlike cash compensation which is expensed against earnings there isgenerally no expense recognition (at grant exercise or sale) for optionswhether they be NQSOs or ISOs As a result compensation consultantsoften point out that stock options are the only form of compensation thatare free in an accounting sense but still deductible for tax purposes

Taxation of Executive Compensation 11

Stock options do however lead to expense recognition if they are dis-counted (in the money at grant date) or if the exercise price and numberof options are not known at grant date For discount options the differ-ence between the stock price and the market price is expensed over thevesting period For options with variable terms (eg a variable exerciseprice vesting that is tied to performance) the options are marked tomarket and expensed during the time between grant and exercise Practi-tioners claim that the accounting treatment of options plays an impor-tant role in the design of option programs Thus plans that have ldquobadaccountingrdquo but are thought by many to have attractive incentive fea-tures are often not even seriously considered by companies Examples ofsuch potentially attractive plans include indexed options (where a CEOprots only if his rmrsquos share price grows relative to some market orindustry benchmark) and option grants that are explicitly performance-related both of which would lead to expenses against earnings

Unlike options restricted stock and SARs do not generally receivefavorable accounting treatment Restricted stock is generally expensedover the period in which the restrictions lapse (usually the vesting pe-riod) The magnitude of the expense is the difference between the cur-rent stock price and the executiversquos cost (if any) SARs are marked tomarket each period and the difference between the stock price and theexercise price is expensed over the outstanding period of SARs

4 THE TAXATION OF EXECUTIVE PAY CASHVERSUS OPTIONS

Stock-based compensation performs two roles in executive compensa-tion arrangements First it helps align the incentives of the executivewith the interests of the rmrsquos shareholders Second it often enables therm to compensate the CEO in a way that is more advantageous from atax standpoint than paying the executive in salary and bonus

41 Agency Theory and Executive Compensation in the Presenceof TaxationIn standard agency-theory models ( Jensen and Meckling 1976) agencycosts are the result of the separation of ownership and control Managersdo not have the same incentives as the owners The optimal incentivecontract for managers balances the benets of high-powered incentives(linking the fortunes of owners and managers through stock and stockoptions for example) with the costs of loading too much risk on risk-averse managers

TA

BL

E2

Tax

and

Acc

ount

ing

Tre

atm

ent

ofE

xecu

tive

Com

pens

atio

nIn

cent

ive

Stoc

kO

ptio

ns(I

SOs)

Res

tric

ted

Stoc

kan

dSt

ock

App

reci

atio

nR

ight

s(S

AR

s)

Tax

atio

nat

Tax

atio

nat

Acc

ount

ing

Com

pen

sati

onD

escr

ipti

onp

erso

nall

evel

corp

orat

ele

vel

trea

tmen

t

Ince

nti

vest

ock

ISO

sar

eid

enti

calt

oIf

the

usu

alco

ndit

ions

Ifth

eu

sual

cond

itio

nsA

sw

ith

NQ

SOs

ther

eop

tion

sN

QSO

sin

thei

rd

esig

n

are

met

th

ere

isno

tax

are

met

the

reis

gene

r-is

gene

rally

noex

pen

seH

owev

ert

hey

have

anat

gran

tor

exer

cise

At

ally

no

corp

orat

eta

xd

e-re

cogn

itio

nfo

rIS

Os

annu

alca

pof

$100

000

sale

all

opti

onp

ro

tsd

uct

ion

atgr

ant

exer

-(e

ver)

on

the

amou

ntth

atan

dsu

bseq

uen

tca

pita

lci

seo

rsa

le

vest

sT

he

tax

trea

t-ap

prec

iati

onar

eta

xed

men

tof

ISO

sal

soat

the

cap

ital

gain

sd

iffe

rs

rate

Res

tric

ted

stoc

kSi

mila

rto

stoc

kop

-T

hein

div

idu

alis

taxe

dT

heco

mp

any

gene

rall

yR

estr

icte

dst

ock

isge

n-ti

ons

but

shar

esat

the

ord

inar

yra

teas

rece

ives

ap

aral

leld

e-er

ally

exp

ense

dov

er(w

hich

vest

slow

ly)

the

rest

rict

ions

lap

se

du

ctio

neq

ual

toth

eth

eve

stin

gp

erio

dT

hera

ther

than

opti

ons

are

How

ever

the

man

ager

amou

ntof

the

exec

u-

amou

nt

exp

ense

dis

gran

ted

toth

eex

ecu

-m

aych

ose

tobe

taxe

dti

versquos

inco

me

whe

nth

eth

ed

iffe

renc

ebe

twee

nti

veU

nlik

eop

tion

sat

gran

td

ate

inw

hich

exec

uti

veis

taxe

dR

e-th

ecu

rren

tst

ock

pric

eth

esh

ares

typi

cally

case

alls

ubs

equ

ent

ap-

stri

cted

stoc

kis

gen

er-

and

the

exec

uti

versquos

have

voti

ngan

dd

ivi-

pre

ciat

ion

ista

xed

atal

lysu

bjec

tto

the

cost

(if

any)

d

end

righ

ts

the

cap

ital

gain

sra

te

mill

ion-

dol

lar

cap

D

ivid

end

sar

eta

xed

atth

eor

din

ary

rate

Stoc

kap

pre

ciat

ion

Rig

hts

that

repl

icat

eG

ener

ally

have

the

Gen

eral

lyha

veth

eU

nlik

eop

tion

sSA

Rs

righ

tsth

epr

ots

ofst

ock

op-

sam

eta

xtr

eatm

ent

assa

me

tax

trea

tmen

tas

are

mar

ked

tom

arke

tti

ons

Rar

ely

use

d

NQ

SOs

NQ

SOs

each

per

iod

an

dth

esi

nce

stoc

kop

tion

sd

iffe

renc

ebe

twee

nth

eha

vem

ore

favo

rabl

eac

-st

ock

pri

cean

dth

eex

-co

unt

ing

trea

tmen

ter

cise

pri

ceis

exp

ense

d(a

ndop

tion

sp

rovi

de

over

the

outs

tand

ing

the

sam

ebe

ne

tof

per

iod

ofSA

Rs

cash

less

exer

cise

thro

ugh

brok

erlo

ans

)

Not

eW

eth

ank

com

pen

sati

onco

nsul

tant

sSc

ott

Gre

enbe

rgan

dSc

ott

Ols

en(b

oth

ofT

ower

sP

erri

n)an

dFr

eder

icC

ook

(of

Fred

eric

WC

ook

and

Ass

ocia

tes)

for

hel

pfu

lcon

vers

atio

nsin

und

erst

and

ing

the

tax

and

acco

unt

ing

rule

s

14 Hall amp Liebman

The effect of taxation on the optimal contract (more precisely theshare of compensation that is performance-related) is ambiguous (evenin the absence of deductibility and deferral) because there are offsettingeffects First by reducing the share of corporate prots received byshareholders taxes diminish the importance to the shareholders of moti-vating the CEO Taken alone this effect would imply that taxes wouldbe expected to reduce the use of performance-based compensation Sec-ond because the government shares in the income received by the CEOthe variance of the CEOrsquos income is reduced raising his utility andlowering the cost to the rm of providing any given set of incentivesThird because taxes will lead executives to provide less effort for anygiven level of incentive-based pay the level of compensation that mustbe provided to compensate the CEO for effort is reduced for a givenamount of incentive These last two factors make it cheaper for the rmto offer contracts to the CEO with large incentive components when taxrates are higher and should therefore increase the use of incentive-based pay Since the net effect of taxation on the level of incentive-basedpay is ambiguous agency theory provides no strong predictions abouthow taxation should affect the optimal composition of option versuscash compensation

42 The Tax Advantages of DeferralWhile agency theory yields ambiguous predictions about the effect oftaxes on the use of options there are direct tax advantages of options inthat options provide executives with a way to defer compensation andthereby lower their taxes However because options also lead to a defer-ral of corporate tax deductibility the tax advantages from a combined(executive and corporate) perspective are not so straightforward In thissection we analyze and measure the tax advantages of stock optionsrelative to salary and bonus compensation We also analyze the taxadvantages of NQSOs relative to ISOs In particular we show how thetax advantages of options change as personal corporate and capitalgains taxes change We then show how the tax advantages of optionshave changed over time in response to changing tax rates

The crucial tax difference between standard options (NQSOs) andcash payment is that option payouts are deferred and the two forms ofcompensation earn different rates of return over the deferral period Anyanalysis of the relative tax advantages of two compensation instrumentsmust consider the tax consequences both to the employer and to theemployeemdashwhat Scholes and Wolfson (1992) call the global contractingperspective Thus in order to make valid comparisons between the twowe compare the tax burden to the executive while holding constant the

Taxation of Executive Compensation 15

posttax cost to the company By keeping posttax employer costs (in NPVterms) constant any package that is preferred by the employee is tax-advantaged in the global-contracting sense

Under this methodology a comparison of the tax advantages of op-tions and cash involves comparing a pretax cash payment of P with anoption payment that has the equivalent posttax NPV to the companyFor NQSOs it is straightforward to show that a pretax payment of P toan executive has exactly the same cost to the company as putting aside Pfor the purposes of paying stock-option payouts later That is if a com-pany pays an executive P today it will have the same amount of moneyin N years as if it had put aside P today let it accumulate at the rate ofreturn earned by the rm and then paid it out (as compensation in theform of option gains4) with any appreciation in year N

Assume that the pretax prot rate is r and the corporate tax rate is TcThen if the company pays P today it receives a deduction today of PTcwhich yields

PTc [1 1 r (1 2 Tc)]N (1)

in N years since the benets of the tax deduction are invested in thecompany and receive the after-tax corporate rate of return Converselyif the company puts P aside today then it grows at the after-tax corpo-rate rate of return in N years to give a tax deduction of

P [1 1 r (1 2 Tc)]N Tc (2)

which is the exact same value Note that the equivalence of these two taxdeductions is analogous to the tax-benet equivalence of front-loadedand back-loaded IRAs

An executiversquos payoff in N years from option prots is equal to Pcompounded at the after-tax corporate rate times 1 2 Tp where Tp is thepersonal tax rate Thus the combined payoffs in N years of the corporatededuction and the executiversquos posttax payoff is

P [1 1 r (1 2 Tc)]N Tc 1 P [1 1 r (1 2 Tc)]

N (1 2 Tp) (3)

where the rst term is the payoff from the corporate deduction and thesecond term is the executiversquos payoff

4 This analysis is not unique to options Any form of deferred compensation that enablesexecutives to invest inside the rm at a higher posttax return or without paying capitalgains taxes would have similar effects

16 Hall amp Liebman

The payoff to salary and bonus in N years is more complicated sincean assumption must be made about how the executive invests the origi-nal cash compensation For example if the executive invests in an instru-ment with taxable interest (eg bonds) then the N-year payoff from aninvestment of P accumulates at the post-corporate-tax post-personal-taxrate of return and is

P (1 2 Tp) [1 1 r (1 2 Tc)(1 2 Tp)]N (4)

However if an executive invests in (non-dividend-paying) equities theinvestment accumulates at the higher post-corporate-tax rate of returnwith the offsetting disadvantage that the capital gains are taxed at thecapital gains rate in year N So the executive receives

P [1 1 r (1 2 Tc)]N (1 2 Tp) 2 Tcg $P (1 2 Tp) [1 1 r (1 2 Tc)]

N 2P (1 2 Tp) (5)

where Tcg is the capital gains tax rate5Thus putting the equations together if an executive is paid cash P

and invests g in bonds and 1 2 g in equities the N-year payoff (includ-ing the payoff to the corporation from the tax deduction) is

P [1 1 r (1 2 Tc)]N Tc 1 g $(P [1 1 r (1 2 Tc)(1 2 Tp)]

N (1 2 Tp)1(1 2 g) (P [1 1 r (1 2 Tc)]

N (1 2 Tp) 2 Tcg $P (1 2 Tp)[1 1 r (1 2 Tc)]

N 2 P (1 2 Tp)) (6)

The tax advantage of options versus salary and bonus therefore is thedifference between the combined (corporate and executive) payoffs inequation (3) and the combined payoffs in equation (6)

43 The Tax Advantage of OptionsWe now turn to analysis of how changes in various tax rates affectthe tax advantages of options relative to cash compensation We focuson the case in which the cash earned by executives is invested entirelyin equities since equities are tax-favored relative to bonds The impor-tant conclusions of this analysis are not substantively different if we

5 In practice executives hold a combination of equities and bondlike instruments withtaxable interest Because equities are tax-favored in the next section we simplify the analy-sis by assuming that executives hold only equities Alternatively we could have assumedthat executives hold only less risky bonds since bonds more closely match the risk proleof a stream of tax savings None of our empirical results are substantively affected by thissimplication

Taxation of Executive Compensation 17

instead assume that a portion of an executiversquos holdings are invested inbonds

431 The Tax Advantage of Options Changing the Corporate TaxRate Stock options are tax-advantaged relative to cash simply becauseoption payouts are deferred allowing the executive to invest at the pre-personal-tax rate of return (with no capital gains tax at the end) ratherthan investing at the after-tax rate of return or paying the capital gainstax Since the advantage of being able to defer taxes is large when thecorporate tax rate is low the tax advantage of options is larger when thecorporate rate is low

To illustrate this point we dene the tax advantage of options to bethe dollar amount by which total after-tax option payoffs (to the corpora-tion and the executive) exceed the total payoffs from salary as denedby the difference of equation (3) and (6) assuming that P (the paymentto the executive) is equal to $100 and g 5 0 (cash compensation is in-vested in equities) We use a ten-year horizon since most options have aten-year term We then calculate how the tax advantage of optionschanges as the corporate tax rate varies from zero to 100 percent holdingthe personal tax rate and the capital gains rate xed at 40 percent and 20percent respectively (which approximates current rates)

The top panel of Figure 2 shows the tax advantage of options relativeto cashmdashprecisely the NPV of the total tax advantage of options for a$100 payment to the CEOmdashas the corporate tax rate changes As ex-pected the tax advantage of options declines as corporate tax rates in-crease At a zero corporate tax rate the tax advantage of options has anNPV of $7 This value is about $350 at a corporate rate of 40 percentWhen the corporate tax rate is 100 percent the tax advantage of optionscompletely disappears since the benets of deferred compensation fallto zero because the after-tax return on equities (and therefore the dis-count rate) falls to zero The tax advantage of options is the NPV of thetax savings from avoiding the capital gains tax on ten years of apprecia-tion which is why it is relatively modest even at the 40-percent corpo-rate rate

432 The Tax Advantage of Options Changing the Personal TaxRate Using the same assumptions as above (but this time holding thecorporate rate at 40 percent and the capital gains rate at 20 percent andvarying the personal rate) the tax advantage of options declines as thepersonal tax rate rises as shown in the middle panel of Figure 2 The taxadvantage of options is higher at low personal tax rates because theposttax base that gives rise to capital gains is higher and the advantage of

Note Holds constant the personal rate at 40 percent and the capital gains rate at 20 percent

Note Holds constant the corporate rate at 40 percent and the capital gains rate at 20 percent

Note Holds constant the personal rate at 40 percent and the corporate rate at 40 percent

FIGURE 2 Tax Advantage of Options NPV of Tax Advantage of Pay-ing $100 in Options (NQSOs) Rather Than in Cash

Taxation of Executive Compensation 19

avoiding the capital gains taxes is greater The tax advantage of optionstherefore declines linearly as the personal tax rate rises At a 100-percentpersonal tax rate the executive receives nothing in either case and byconstruction the companyrsquos tax deduction benet is the same in NPVterms so the tax advantage of options falls to zero also6

433 The Tax Advantage of Options Changing the Capital Gains TaxRate The tax advantage of options for various capital gains rates isdepicted in the bottom panel of Figure 2 under the same assumptions(this time holding both the personal rate and the corporate rate at 40percent while varying the capital gains rate) The tax advantage of avoid-ing the capital gains tax increases as the capital gains tax rate rises

5 HOW HAS THE TAX ADVANTAGE OF OPTIONSCHANGED OVER TIME

The analysis so far has illustrated how the tax advantage of optionsvaries with the changes in tax rates In this section we show how the taxadvantage of options has changed over time as tax rates have changed

51 Top Marginal Tax Rates over TimeThe evolution on the top marginal tax ratesmdashpersonal corporate andcapital gainsmdashfrom 1980 to 1998 is shown in Figure 3 In all cases thetop rate (the rate for taxpayers with the highest incomes) is shownwhich is not always the highest rate since various anomalies (such asclawbacks of exemptions) sometimes lead to marginal tax rates that arehigher than those for the highest income tax payers

The top personal tax rate was 70 percent in 1980 falling to 50 percentin 1982 and 28 percent in 1988 following the 1986 tax act7 The topmarginal tax rate has since risen to 396 percent but is effectively 425percent since there is a 29-percent Medicare surcharge (paid half by theemployer and half by the employee) that has no upper limit The topcorporate tax rate has had only one signicant change since 1980 fallingfrom 46 percent before the 1986 tax act to 34 percent following the act in1988 The top corporate tax rate was increased to 35 percent in 1993 The

6 This is one place where ignoring bond investments has substantive implications Inparticular in the all-bond case the tax advantage of options is nonmonotonicmdashit rises andthen fallsmdashas personal tax rates rise7 During the early 1980s the maximum tax on earned income limited the marginal tax rateon the earnings of high earners so many executives did not face a decline in personal taxrates between 1981 and 1982 The empirical results in this paper are robust to assumingthat executives faced a marginal tax rate of 50 percent in 1980 and 1981

20 Hall amp Liebman

FIGURE 3 Top Marginal Tax Rates since 1980

top capital gains rate has uctuated between 28 percent (approximately)and 20 percent since 1980 and is currently at 20 percent

52 The Evolution of the Tax Advantage of NQSOsThe tax advantage of options relative to cash dened in the same way asin section 4 from 1980 to 1998 is shown in Figure 4 The calculations aredone using the same assumptions as before (ie N 5 10 years and cashis invested in equities) and the statutory top marginal tax rates shown inFigure 3 are used Due to the fall in corporate and personal tax rates thetax advantage of options increased sharply after the Tax Reform Act of1986 This increase in tax advantage has largely been reversed in the1990s as top personal tax rates have crept back up and the capital gainsrate has been reduced

Two key facts emerge from this analysis First there is currently only amoderate tax advantage to standard nonqualied stock optionsmdashon theorder of $4 per $100 of compensation This is because the tax advantagesto the executive of deferring taxes are largely offset by the tax disad-vantages to the company of not being able to deduct option expensesfrom taxable prots until the executive exercises the options Secondalthough the 1986 tax act substantially increased the tax advantage ofoptions more than half of this increase has been reversed in the 1990s

Taxation of Executive Compensation 21

FIGURE 4 Tax Advantage of Options 1980 to 1998

NPV of tax advantage of $100 in options (NQSOs) rather than cash

53 The Tax (Dis)advantage of ISOsSo far the analysis has focused on the tax advantages of NQSOs sincethey are so much more prevalent than ISOs which account for onlyabout 5 percent of option grants The relative scarcity of ISOs can beexplained both by their tax status and by their per-executive cap of$100000 per year

Under what conditions are ISOs tax-preferred to NQSOs For a trans-fer P to an executive the ISO is always tax-preferred by the executive(since the capital gains rate is lower than the personal rate) while theNQSO is always tax-preferred from the companyrsquos perspective (since anISO is not deductible) The key issue however is the relative advan-tages of the two types of options from a global-contracting perspectiveBecause ISOs are not deductible the company is indifferent betweensetting aside P in the form of NQSOs today and setting aside P(1 2 Tc) inthe form of ISOs (Since the NPV of the tax deduction for NQSOs is PTcthe NPV of the payment P is P 2 PTc or P (1 2 Tc))

Equalizing the posttax cost of NQSOs and ISOs we need only look atthe payoff of the executive to determine the condition under which eachis preferred The payoff of NQSOs in N years is

P [1 1 r (1 2 Tc)]N (1 2 Tp) (7)

22 Hall amp Liebman

The same payoff for ISOs is

P (1 2 Tc) [1 1 r (1 2 Tc)]N (1 2 Tcg) (8)

ISOs therefore are tax-advantaged if

Tp 2 Tcg Tc 2 Tc Tcg (9)

Intuitively ISOs are tax-advantaged only if their advantage (the differ-ence between the personal rate and the capital gains rate) is largeenough to offset their cost (the disadvantage of not deducting at thecorporate rate) Note however that the condition is not a simple com-parison between the corporate rate and the personal rate minus thecapital gains rate (Tp 2 Tcg Tc)

Figure 5 shows how the tax advantage of NQSOs relative to ISOs haschanged since 1980 as tax rates have changed Unlike the very modestchanges (around 3 to 4 percent) in the relative tax advantages betweenoptions and cash shown in Figure 4 the relative tax advantage betweenNQSOs and ISOs has seen enormous swings When the top personalrate was 70 percent in 1980 ISOs were tax-favored by a margin greaterthan 6 percent However by 1982 (following the 1981 tax acts that dra-matically lowered the personal tax rate) ISOs became tax-disadvantaged

FIGURE 5 Tax Advantage of NQSOs Relative to ISOs over Time

NPV of tax advantage of $100 payment to executive NQSOs minus ISOs

Taxation of Executive Compensation 23

by about 5 percent By 1988 (following the 1986 tax act) ISOs became tax-disadvantaged by about 18 percent a dramatic swing The raising ofpersonal rates coupled with a decrease in the capital gains rate has sincereduced the tax disadvantage of ISOs to about $4 per $100 of pretaxcompensation Although hard data on ISOs are hard to come by practi-tioners (mostly compensation consultants) claim that ISOs were morepopular prior to 1982 when they were tax-advantaged Although ISOsare less disadvantageous from a tax standpoint relative to NQSOs thanthey were in the late 1980s they are still disadvantageous so it is notsurprising that they have not made a signicant resurgence in recentyears

6 EXPLAINING THE INCREASE IN OPTION-BASEDCOMPENSATION

In this section we analyze whether changes in tax rates affect the compo-sition of executive compensation and consider the relative importanceof tax factors and corporate governance factors in explaining the in-creasing share of compensation paid in the form of stock options Thenumerous changes in tax rates that have occurred since 1980 provide anopportunity to assess whether executive compensation arrangementsrespond to tax incentives in the way that the tax-avoidance model sug-gests We test this model using a panel data set of CEOs in largepublicly-traded US corporations Our identication of the tax effectsrelies on time-series variation in personal corporate and capital gainstax rates along with cross-sectional variation in corporate tax rates

61 DataWe use a 15-year panel data set of CEOs in the largest publicly-tradedUS corporations which is described in Hall and Liebman (1998) Thedata set covers the years from 1980 through 1994 and combines CEOcompensation information from corporate proxies and 10-K lings withstock price and stock return information from CRSP and accounting datafrom Compustat In addition some compensation data from the 1970swere collected in order to construct measures of the value of stock op-tions held by the CEOs in the rst period of the sample8

The feature that distinguishes our data from most other CEO datasets is that with our panel of yearly proxy data on option grants optiongains and total options held we are able to calculate the total value of

8 We are grateful to David Yermack for providing us with some of the data for 1984 to 1991See Yermack (1995) for a discussion of these data

24 Hall amp Liebman

all stock options held by the CEO at a point of time More importantlysince we have the details about the stock options held (number exer-cise price time to maturity etc) we can precisely calculate the changein the value of a CEOrsquos stock option holdings for a given change in rmvalue

62 Identication IssuesThe numerous tax changes over our sample period and the specic waysin which changes in personal corporate and capital gains rates arepredicted to affect the composition of CEO compensation give us un-usually rich sources of identication However because many of thechanges in tax rates would be expected to affect all of the CEOs in oursample in a similar way we will need to pay particular attention toseparating out the tax effects from underlying time trends and otherfactors that changed over time Controlling for underlying trends is par-ticularly important in this study because we are focusing on an out-comemdashthe use of stock optionsmdashthat has increased rapidly over thepast two decades and that many practitioners believe was importantlyaffected by nontax considerations Thus the tax effects we are trying toexplain are deviations from a rising trend

We take two steps to try to separate out the tax effects from theunderlying trend First we include key nontax factors that could poten-tially explain the increasing reliance on performance-based compensa-tion It has been suggested that the dramatic increase in incentive-basedpay is the result of the remarkably poor shareholder returns during the1970s which spurred the LBO and takeover movements of the 1980sAccording to this story (Kaplan 1997) shareholders became much morepowerful via the rise of institutional investors even as the LBO andtakeover movements waned Because the inuence of institutional inves-tors is thought to be one of the most important mechanisms of strongcorporate governance we use the share of each companyrsquos stock ownedby institutional investors (dened as institutions with more than $100million under management) as an explanatory variable9 The percentageof shares owned by large institutional investors increased from about 20percent to almost 50 percent in 1994 an upward trend that closelymatches the sharp rise in the share of compensation in the form ofoptions Indeed in our sample the annual average (over all of the rmsin our sample) share of stock owned by institutional investors has acorrelation of 09 with the annual average share of compensation paid inoptions

9 This variable is described in Gompers and Metrick (1998)

Taxation of Executive Compensation 25

In addition we include two other variables that proxy for strongercorporate governance The rst is the size of the board Evidence suggeststhat boards with smaller numbers of directors reduce the inuence of theCEO and better represent shareholder interests (Yermack 1995) Boardsize fell by approximately 15 percent between 1980 and 1994 The secondvariable is the fraction of the rmrsquos board members who are inside direc-tors or gray directors (noninsiders who have business dealings with thecompany) Firms with fewer outsiders are less likely to act in the share-holdersrsquo interests and we would predict that they would be less likely tohave performance-based compensation for their CEOs The average per-centage of inside and gray directors has decreased from about 45 percentin 1980 to about 35 percent in 1994

Second we allow for cross-sectional variation in corporate tax ratesCorporate marginal tax rates vary because rms may be eligible for tax-loss carrybacks and carryforwards investment tax credits and the alter-native minimum tax (see Auerbach 1986 Auerbach and Altshuler1990 Auerbach and Poterba 1987) Because these cross-sectional differ-ences imply that different rms should respond differently to a givenchange in tax rates they potentially provide us with a way to isolatethe tax effects Firm-specic corporate marginal tax rates are notori-ously difcult to calculate from publicly-available data We use a tricho-tomous variable that equals zero if the rm has tax-loss carryforwardsand negative earnings in a given year equals 05 times the statutorycorporate rate if the rm has only one of those conditions and equalsthe statutory rate if the rm has neither loss carryforwards nor negativeearnings10

The cross-sectional variation in the corporate tax rate may not be ex-ogenous Firms that perform poorly could face low marginal tax ratesand also be particularly likely or unlikely to provide performance-basedpay Because most of our specications use xed effects the correlationwe must be concerned about is between changes in rm performanceand changes in compensation We deal with this by controlling forlagged rm stock market performance in our regressions This shouldeliminate spurious correlation between tax rates and compensation thatis jointly caused by rm performance

10 Graham (1996) presents evidence that the trichotomous variable performs nearly as wellas his recommended simulated marginal tax rate Plesko (1999) compares marginal taxrates calculated from Compustat data with actual marginal tax rates from rm tax returnsHe nds that simple binary variables are more reliable measures of the marginal tax ratethan is either the Graham or the trichotomous variable We obtain substantively similarresults when we switch from the trichotomous variable to the binary variables recommendby Plesko

26 Hall amp Liebman

63 ResultsThe dependent variable in our regressions is the fraction of a CEOrsquos totalannual compensation (measured as the BlackndashScholes value of stock-option grants plus salary and bonus) that comes from stock-optiongrants We choose this reduced-form specication because our interest isin the composition of CEO compensation However the factors thataffect the composition of compensation might also affect the levels andit will be important to keep this in mind in interpreting our results

Optimal-contracting theory generally makes predictions about the in-centives provided by the total amount of rm stock and stock optionsowned by the CEO We choose to model the ow of option grants ratherthan the ultimate performance sensitivity of the CEOrsquos compensationand stock and stock-option holdings because we have in mind anadjustment-cost model in which it takes time for a CEOrsquos holdings ofstock and stock options to reach the optimal level11 In our regressionswe include two measures of the current performance sensitivity of theCEOrsquos wealth to rm performance as explanatory variables to reect thedistance of the CEOrsquos contract from the optimal contract

Table 3 presents results in which the tax effects are identied solely byvariation in tax rates over time The fraction of annual compensation paidin stock options is regressed on tax variables corporate governance vari-ables contract-theory variables and variables reecting stock market re-turns Both of the regressions in this table include rm xed effects Incolumn (1) the fraction in annual compensation paid in stock options isregressed on the log difference in payoff (the combined payoff to both therm and the CEO) from receiving compensation in stock options ratherthan salary and bonus where the payoff calculations are based on statu-tory marginal tax rates This variable is dened as the log difference be-tween equation (3) and equation (6) in section 4 The coefcient on the taxvariable is large and statistically signicant at the 95-percent level Thepoint estimate implies that a one-percent increase in the payoff differencebetween stock options and salary and bonus results in a 24-percentage-point increase in the share of compensation paid in stock options

The corporate governance variables all have the predicted sign Firmswith a higher fraction of their shares owned by institutional investorsuse more performance-based pay (stock options) Firms with large corpo-rate boards or a large share of inside directors are less likely to use stockoptions

The two contract-theory variables are the dollar change in CEO wealth

11 See Core and Guay (1999) for evidence that rms adjust in this way

TABLE 3Compensation Regressions that Rely on Variation in

Tax Rates over Time

(1) (2)

Tax variablesln (difference in payoff)(a) 2393 (0349)ln (12statutory personal 20015 (0034)

rate)ln (12statutory cap gains 0330 (0171)

rate)ln (12statutory corporate 0399 (0148)

rate)

Corporate governance variablesFraction of rmrsquos shares 0043 (029) 0052 (0028)

owned by institutionalinvestors

Size of the rmrsquos board of 20002 (001) 20003 (0001)directors

Fraction of directors who 20082 (033) 20092 (0030)are inside or gray

Contract-theory variablesD (CEO wealth) per $1000 20261 (0095) 20247 (0087)

in FV(b)

D (CEO wealth) per 10 in 2583 (189) 2554 (201)FV(c)

Volatility(d) 521 (424) 261 (369)ln (rm market value) 0019 (0006) 0019 (0006)

Stock market return variablesLagged annual rm stock 0006 (0009) 0003 (0008)

market returnLagged annual SampP 500 re- 20035 (0023) 20029 (0023)

turn

Regression diagnosticsTotal sample size 5179 5179Firms 427 427Adjusted R2 0313 0311

Dependent variable share of compensation paid in options Standard errors in parentheses Bothregressions in this table include rm xed effects

(a) Log difference in payoff is the log difference in combined rm and CEO after-tax share betweenpaying compensation in options and paying salary

(b) This variable is the dollar change in CEO wealth per $1000 change in rm market value (seeJensen and Murphy 1990)

(c) This variable in the dollar charge in CEO wealth per 10-percent change in rm value (see Hall andLiebman 1998)

(d) Standard deviation of rm market value in trillions

28 Hall amp Liebman

per $1000 change in rm value (Jensen and Murphy 1990) and thedollar change in CEO wealth per 10-percent change in rm value (Halland Liebman 1998) As discussed in Baker and Hall (1998) these twowealth sensitivity measures reect two different concepts of how closelyaligned the managersrsquo incentives are with the interests of the sharehold-ers In the regression they are meant to measure how far the CEOrsquosexisting contract and ownership of stock and stock options is from theoptimal level The negative coefcients on the two variables suggeststhat rms that are below their optimal pay-to-performance sensitivityare more likely to give more stock options in the current period

The second column replaces the tax variable motivated by deferraladvantages with the three statutory rates The corporate after-tax sharehas a positive coefcient as we would expect because the tax advantageof stock options increases as the corporate rate falls The coefcient onthe personal tax rate is small and statistically insignicant The sign ofthe coefcient on the capital gains rate is the opposite of that predictedby tax-avoidance theory

Note that these results rely on time-series variation alone This is not avery convincing test Since the log difference in payoffs mostly rises overthe time period covered in our sample as does the share of compensa-tion paid in options it is possible that the tax variable is simply reectingother factors that were changing over this time period that we have notincluded in our regression Therefore it is important to look at specica-tions that rely on cross-sectional variation in the tax variables

Table 4 contains specications that rely on cross-sectional variation incorporate tax rates The rst column repeats the specication from col-umn (1) of Table 3 but uses the trichotomous measures of rm corporatetax rates in place of the statutory rate in calculating each rmrsquos logdifference in payoff from paying compensation in options rather than insalary and bonus As in the previous table we nd the positive coef-cient on the log difference in payoff just as the theory suggest How-ever the coefcient estimate is now much smaller suggesting that aone-percent increase in the log difference in payoff produces a 01-percentage-point increase in the share of compensation paid in options

Introducing cross-sectional variation in the dependent variable doesnot eliminate the potentially spurious correlation over time between thetax variable and the share of compensation paid in options The secondcolumn of Table 4 isolates the pooled cross-sectional variation in corpo-rate tax rates by adding time dummies and dropping the rm xedeffects In this specication the coefcient on the tax variable is indistin-guishable from zero The coefcient on the institutional variable contin-ues to be large and statistically signicant The third column includes

Taxation of Executive Compensation 29

both rm xed effects and time dummies which tests for a relationshipbetween within-rm changes in the dependent variable and within-rmchanges in each explanatory variable In this specication the coefcienton the tax variable is again indistinguishable from zero The coefcientson the corporate governance variables are also statistically insignicant

The third column contains a specication that treats the Tax ReformAct of 1986 (TRA 86) as a natural experiment For rms and CEOs facingthe statutory corporate tax rate TRA 86 increased the relative payoff topaying compensation in options by 17 percent For a rm facing a zeromarginal tax rate it reduced the relative payoff by 45 percent The reduc-tion occurred because TRA 86 reduced the personal tax rate makingsalary and bonus more attractive while the decreased corporate rate hadno effect on zero-tax-rate rms Thus we would expect rms facing zeroor low marginal tax rates to reduce their use of options after TRA 86while rms facing the statutory rate would be expected to increase themFew rms face zero marginal tax rates year after year Therefore we tryto distinguish between rms that often face low tax rates and rms thatnearly always face the statutory rate We dene low-tax rms as oneswhose average marginal tax rate over 1984 to 1986 was below 02412

Then we ran a difference-in-differences regression to see if high-taxrms increased their use of options more than low-tax rms The interac-tion between post-TRA 86 and high corporate tax rates is the key vari-able The estimated coefcient on this variable is positive but small andnot statistically different from zero at the 95-percent level

It is important to note that in this specication as well the main corpo-rate governance variable the share of the rmrsquos shares owned by institu-tional investors performs as theory would predict The variable appearsto be quite robust Table 5 presents regressions of the share of compensa-tion paid in stock options on the fraction of the rmrsquos shares owned byinstitutional investors Since we have already established that these vari-ables have a strong longitudinal relationship the regressions here aredesigned to focus on the cross-sectional relationship between these twovariables The rst and third columns present regressions using data onlyfrom 1994 and nd a coefcient of roughly 02 (column 1 contains noother covariates while column 3 includes all of the other nontax covari-ates from the regressions in Tables 3 and 4) Since the median share of armrsquos stock owned by institutional investors in our sample rises from 19percent in 1980 to 49 percent in 1994 this coefcient would imply that this

12 This cutoff was chosen after inspecting the distribution of tax rates and essentiallyseparates rms that are always at or near the statutory rate from ones that are sometimesbelow it

TA

BL

E4

Com

pens

atio

nR

egre

ssio

nsth

atR

ely

onC

ross

-Sec

tion

alV

aria

tion

inC

orpo

rate

Tax

Rat

es

(1)

(2)

(3)

(4)

Tax

vari

able

s(a)

ln

(dif

fere

nce

inp

ayof

f)0

099

(00

41)

20

038

(00

36)

20

005

(00

44)

Pos

t-T

RA

862

000

5(0

013

)H

igh

corp

orat

eta

xra

te0

018

(00

17)

(Pos

t-T

RA

86)

3hi

gh0

024

(00

16)

corp

orat

eta

xra

te

Cor

por

ate

gove

rnan

ceva

riab

les

Frac

tion

of

rmrsquos

shar

es0

109

(00

27)

011

4(0

015

)2

000

9(0

030

)0

087

(00

34)

owne

dby

inst

itu

tion

alin

vest

ors

Size

ofth

e

rmrsquos

boar

dof

20

004

(00

01)

20

001

(00

01)

20

002

(00

01)

000

1(0

002

)d

irec

tors

Frac

tion

ofd

irec

tors

wh

o2

012

4(0

033

)2

000

3(0

018

)2

004

1(0

034

)0

001

(00

39)

are

insi

de

orgr

ay

Con

trac

t-th

eory

vari

able

sD

(CE

Ow

ealt

h)p

er$1

000

20

238

(00

94)

20

167

(00

53)

20

257

(00

96)

20

021

(01

47)

inFV

(b)

D(C

EO

wea

lth)

per

10

in2

560

(17

8)2

592

(10

5)2

594

(19

6)2

147

(04

86)

FV(c

)

Vol

atili

ty(d

)2

48(4

24)

162

(03

4)1

12(4

28)

106

(06

09)

ln(

rmm

arke

tva

lue)

003

3(

006)

000

8(0

003

)0

009

(00

06)

001

2(0

007

)

Stoc

km

arke

tre

turn

vari

able

sL

agge

dan

nual

rm

stoc

k2

000

2(0

009

)0

011

(00

10)

000

6(0

009

)2

001

2(0

013

)m

arke

tre

turn

Lag

ged

annu

alSamp

P50

02

005

6(0

023

)2

005

3(0

051

)2

003

9(0

049

)2

015

4(0

040

)re

turn

Reg

ress

ion

dia

gnos

tics

T

otal

sam

ple

size

5179

5179

5179

2534

Firm

s42

742

739

6A

dju

sted

R2

031

009

032

005

Firm

xed

effe

cts

Yes

No

Yes

No

Tim

ed

um

mie

sN

oY

esY

esN

o

Dep

end

ent

vari

able

isfr

acti

onof

ann

ualc

omp

ensa

tion

from

stoc

kop

tion

sSt

and

ard

erro

rsin

par

enth

eses

(a)

All

thre

esp

eci

cati

ons

use

tric

hoto

mou

sco

rpor

ate

tax

rate

sT

he

sam

ple

inco

lum

n(3

)u

ses

only

the

year

s19

84to

1988

sin

cew

ear

eu

sing

1986

asa

nat

ural

exp

erim

ent

(b)

Thi

sva

riab

leis

the

dol

lar

chan

gein

CE

Ow

ealt

hp

er$1

000

chan

gein

rm

mar

ketv

alu

e(s

eeJe

nse

nan

dM

urp

hy

1990

)

(c)

Thi

sva

riab

lein

the

dol

lar

char

gein

CE

Ow

ealt

hp

er10

-per

cent

chan

gein

rm

valu

e(s

eeH

alla

nd

Lie

bman

199

8)

(d)

Stan

dar

dd

evia

tion

ofr

mm

arke

tva

lue

intr

illio

ns

32 Hall amp Liebman

TABLE 5Large Institutional Investors and Stock-Option Compensation

The Cross-sectional Relationship between Institutional Investors andOption Grants

(1) (2) (3) (4)

Coefcient on fraction of 217 (067) 145 (014) 177 (069) 120 (015)rmrsquos shares owned byinstitutional investors

Sample period 1994 1980ndash1994 1994 1980ndash1994Contract-theory and stock No No Yes Yes

market covariatesTime dummies No Yes No Yes

The dependent variable is the share of compensation paid in options

factor alone can explain 6 percentage points of the 23-percentage-pointincrease in the median share of compensation paid in options that oc-curred over this time period Of course this variable is only one measureof the strengthening of corporate governance so the total impact of im-proved corporate governance could be greater The second and fourthcolumns of the table use the entire 1980ndash1994 sample and include timedummies to isolate the cross-sectional correlations Again there is evi-dence of a strong cross-sectional relationship between the two variablesalthough the coefcients are slightly smaller perhaps reecting a weakercorporate-governance role in the earlier period In sum there is a strongrelationship in both longitudinal and cross-sectional data between theinstitutional-investor variable and the options-percentage variable13

7 THE MILLION-DOLLAR RULE

In 1993 section 162(m) of the Internal Revenue Code was enactedlimiting the deductibility of executive compensation in excess of onemillion dollars unless the compensation was performance-related Pro-ponents of this legislation argued that it would slow the growth ofexecutive pay and tighten the link between rm performance and execu-tive compensation

The new provision became effective for tax year 1994 and applies tothe CEO and the other four most highly compensated executives in eachcompany Pay that is performance-related such as stock-option grants

13 The one specication in which the fraction owned by institutional investors is notsignicant is in column (2) of Table 4 a regression that includes both xed effects and timeeffects

Taxation of Executive Compensation 33

and bonuses paid for meeting clear performance goals is not affected bythis regulation provided that it has been approved by shareholdersThus the main impact of the provision is to limit the deductibility ofsalary in excess of a million dollars Woodlock and Antenucci (1997)studied the proxy statements of 376 rms and documented that mostrms responded to the new law by qualifying top executive pay asperformance-related Perry and Zenner (1999) identied 25 rms thatreduced salaries from above $1 million to below $1 million and foundthat 23 of the rms cited section 162(m) as their reason for doing so

In this section of the paper we investigate whether this provision hasaffected the level or structure of executive pay using the Execucompdata base which contains information on the highest-paid executives ineach of the rms in the SampP 500 SampP Midcap 400 and SampP SmallCap600 Between 1993 and 1998 the median salary of the Execucomp execu-tives rose from $230000 to $306000 The median value of total compen-sation (salary bonus and option grants) rose from $470000 to $882000Limiting the sample of CEOs indicates that the total compensation of themedian CEO rose from $11 million to $18 million over the period whilethe median salary rose from $441000 to $550000 Thus executive com-pensation clearly continued to rise at a rapid rate after the implementa-tion of the million-dollar rule though perhaps at a slower rate than itotherwise would have

If the regulation affects behavior it should have the largest impact onrms whose executives are receiving salaries that are above the limit orjust below it Thus we might expect to see rms whose executives arereceiving relatively high salaries to rely more heavily on performance-based pay and less heavily on salary in the years following the regulationPerry and Zenner (1999) and Livingston (1999) both nd that this is thecase This trend can be clearly identied in the Execucomp data set Forexample over 1993 to 1998 the median annualized growth rate of salaryfor executives whose 1993 salary was below $500000 in the previous yearwas 85 percent For executives with salary between $500000 and$700000 it was 64 percent for executives between $700000 and $850000it was 50 percent for executives between $750000 and $1 million it was31 percent and for executives above $1 million it was 00 percent14

The observation that executives at lower levels of salary had highergrowth rates is not very strong evidence of an impact of section 162(m)however because it is possible that low-salary executives would havehad higher growth rates even in the absence of the provision To explore

14 Of the 3137 executives for whom we have complete data between 1993 and 1998 2708had 1993 salaries below $500000 while only 35 had 1993 salaries above $1 million

34 Hall amp Liebman

whether this is the case we use Execucomp data from both before andafter the law change Ideally we would want to have data from manyyears before the change to examine whether there is an underlying pat-tern of faster salary growth for low-salary executives Unfortunately theExecucomp data begin in 1992 and therefore provide us with only the1992ndash1993 growth rate in compensation to use as a baseline15

We run regressions pooling data from 1992ndash1993 through 1997ndash1998of the annual percentage change in different forms of compensation on avariable million that is designed to measure the likelihood that anexecutiversquos compensation will be affected by section 162(m) We denemillion as the minimum of 1 and the executiversquos salary divided by $1million in the previous year Thus any executive with salary at or above$1 million would receive a value of 1 while an executive with a salary of$500000 would be coded as a 0516 We include this variable by itself andalso interacted with a dummy variable (after) that equals 1 for years1994 and after Thus the million variable is designed to pick up anyunderlying relationship between the level of salary and the growth rateof salary while the interacted one will identify any differential growthrate of salary by income in the period after section 162(m) took effect Wealso include a full set of year dummies the natural logarithm of thermrsquos market value and the rmrsquos annual rate of return over the twoprevious years as control variables To reduce the sensitivity of our re-sults to outliers we run quantile regressions and robust regressions

Table 6 contains our results The rst four columns are for all topexecutives while the last four columns focus on CEOs Columns (1) and(2) present estimates from robust regression for the percentage change insalary and the percentage change in total compensation (salary bonusstock grants and stock-option grants) In the salary growth regressionsfor all executives the coefcient on million is negative and signicantindicating that there is an underlying relationship between the level ofexecutive compensation and its subsequent growth rate The coefcienton million 3 after is negative as well suggesting that the negativerelationship between the level of salary and its subsequent growth rateintensied slightly after 1994 The relatively small magnitude suggeststhat an executive with a $1 million dollar salary would see his salarygrow at an annual rate that would be about 06 percent slower than anexecutive earning $500000

15 The HallndashLiebman data set contains only a combined salary and bonus variable makingit impossible to use it to analyze this issue16 We also ran the regressions (not reported) allowing for a nonlinear relationship betweenmillion and compensation growth and obtained substantively similar results

TA

BL

E6

The

Eff

ect

ofth

eM

illi

on-D

olla

rR

ule

onSa

lary

and

Tot

alC

ompe

nsat

ion

All

top

exec

uti

ves

CE

Os

only

Rob

ust

regr

essi

onQ

uan

tile

regr

essi

onR

obu

stre

gres

sion

Qu

anti

lere

gres

sion

D

in

Din

D

in

Din

D

in

Din

D

in

Din

sala

ryto

talc

omp

(a)

sala

ryto

talc

omp

sa

lary

tota

lcom

p

sala

ry

tota

lcom

p

mil

lio

n(b

)2

006

02

035

72

006

32

024

12

008

72

023

12

010

32

024

5(0

005

)(0

032

)(0

006

)(0

027

)(0

010

)(0

073

)(0

011

)(0

056

)m

illi

on

32

001

10

026

20

021

20

015

20

016

014

92

003

40

158

aft

er(c

)(0

005

)(0

033

)(0

006

)(0

029

)(0

010

)(0

076

)(0

011

)(0

057

)N

359

5035

950

359

5035

950

757

17

589

757

17

589

Stan

dar

der

rors

inp

aren

thes

esR

egre

ssio

nsal

soin

clu

de

year

du

mm

ies

stoc

km

arke

tre

turn

and

log

mar

ket

valu

e

(a)

Tot

alco

mp

ensa

tion

incl

ud

essa

lary

bon

us

rest

rict

edst

ock

and

stoc

k-op

tion

gran

ts

(b)

mil

lion

isd

ene

das

min

(1s

alar

yin

pre

viou

sye

ars

div

ided

by$1

mil

lion)

(c)

aft

eris

anin

dic

ator

vari

able

equ

alto

zero

befo

reth

ere

gula

tion

took

plac

e(1

993)

and

one

afte

rwar

d

36 Hall amp Liebman

In the second column the coefcient on million 3 after is essentiallyzero suggesting that any decrease in salary brought about by section162(m) was offset by increases in bonus and stock-option grants Theresults are similar for the median regressions There is some evidence ofa very small slowdown in growth of salary for executives likely to beaffected by section 162(m) but no sign of a change in total compensa-tion Taken together this provides evidence of a very small substitutionof performance-related pay for salary

The last four columns run identical tests but limit the sample to CEOsThe interaction of million 3 after continues to be small and negative inthe salary regressions In contrast the impact of 162(m) on total compen-sation appears to be positive and fairly substantial (about 15 percentagepoints per year) in these regressions That is the increase in bonus andstock options following the million-dollar rule more than offset the verysmall decline in salary

All of these conclusions should be interpreted somewhat cautiouslybecause our data provide us with only one pre-1993 control year Never-theless the data are consistent with section 162(m) having led to a veryminor slowdown in salary growth and one that is dramatically smallerthan what one would conclude without controlling for underlying differ-ential salary growth rates at different levels of salary Since total compen-sation for CEOs did not decline and perhaps increased there is evidenceof a minor substitution of performance-related pay for salary in responseto the regulation

8 EXECUTIVE COMPENSATION AND THEELASTICITY OF TAXABLE INCOME

Even if taxes have only a minor effect on executive compensation thestructure of executive compensation is important for tax policy be-cause high-income taxpayers are sources of signicant revenue and arethe focus of important debates about the efciency cost of taxation

Goolsbee (1999) has recently argued that executive decisions to exer-cise options are highly responsive to intertemporal differences in taxrates created by tax reforms He claims that much of the apparent declinein taxable income for high-income taxpayers between 1992 and 1993documented by Feldstein and Feenberg (1996) can be explained by execu-tives shifting option gains into 1992 in order to avoid the higher marginaltax rates that went into effect in 1993 By extension he implies that thehigh elasticities of taxable income with respect to personal tax ratesestimated by comparing the pre- and post-tax-reform taxable income of

Taxation of Executive Compensation 37

high-income individuals reect intertemporal shifting of income and notpermanent effects of taxation If this is true then the deadweight loss oftaxation is lower as is the cost of progressivity

While Goolsbee presents comprehensive evidence that option gainswere unusually high in 1992 his identication of the impact of tax rates onthe timing of option gains depends on correctly specifying a counter-factual level of options that would have been exercised in the absence ofthe tax change He simply assumes a linear time trend The purpose ofGoolsbeersquos time trend is presumably to account for the increasing use ofoptions over time However since options typically require a vestingperiod of at least a couple of years a much more direct measure of ex-pected option gains can be constructed based on the total value of vestedoptions held by the CEO Since options typically have vesting periods thatexceed two years this total can be treated as exogenous since it will not beaffected by the tax changes

More generally the value of options held by the CEO is likely to havea major effect on his exercise decisions If his options are out of themoney he will not have gains to realize If there has been a recent run-up in the market an optimizing CEO may exercise an unusually largeamount of options in order to diversify his overall portfolio and if in-the-money options are about to expire they will be exercised in thecurrent year

Our evidence suggests that much of the apparent tax shifting caninstead be attributed to stock market performance and the timing of pastoption grants Using our 15-year panel we replicate Goolsbeersquos (1999)result for the 1993 period but show that the impact of taxes is notpresent in other time periods Moreover even the result for the 1993period disappears when appropriate controls for past option grants andstock price appreciation are included

The top three rows of Table 7 present data on the taxable incomes ofCEOs in the HallndashLiebman sample For these results our panel has beennarrowed to include only CEOs whose rmrsquos scal years correspondwith tax years In addition we create a balanced panel for each of thethree tax-reform periods by excluding any rm for which there are notcomplete data for the period surrounding the reform

The fourth row of the table shows that the 1993 tax act raised the topmarginal personal income tax rate from 310 to 396 percent effective fortax year 1993 with a further increase occurring in 1994 due to the uncap-ping of the Medicare payroll tax The top row of the table shows thattaxable income was higher in 1992 than for any other year in our sample(all data are in 1994 dollars) Option gains were particularly high in1992mdashmore than double their 1991 level and 68 percent higher than their

TA

BL

E7

Tax

Ref

orm

Sto

ckM

arke

tP

erfo

rman

cea

ndth

eT

imin

gof

Opt

ion

Gai

ns

Ave

rage

stoc

km

arke

tM

ean

CE

OM

ean

CE

Ore

turn

for

Mea

nva

lue

taxa

ble

sala

ryan

dM

ean

CE

OT

opfe

der

al

rms

inof

opti

ons

(Val

ue

ofop

tion

Sam

ple

inco

me

bonu

sop

tion

gain

sp

erso

nal

sam

ple

for

held

atbe

gga

ins)

(va

lue

Ref

orm

size

Yea

r($

mil

)($

mil

)($

mil

)M

TR

pre

vye

ar(

)of

year

ofop

tion

she

ld)

ER

TA

8127

919

800

874

064

80

163

7025

80

460

354

1981

083

10

672

009

869

245

070

61

3919

820

872

067

90

099

5011

90

686

144

1983

099

00

724

025

250

355

116

62

16

TR

A86

295

1985

119

70

837

024

050

76

116

02

0719

861

448

091

70

345

5042

81

776

194

1987

165

50

959

059

338

519

12

229

266

1988

163

21

058

039

028

07

237

71

6419

891

656

105

20

414

2817

92

345

176

1993

272

1990

209

51

028

061

828

232

288

02

15T

axA

ct19

911

875

102

00

547

312

01

268

82

0319

922

678

107

01

250

3146

64

552

275

1993

232

41

209

074

339

620

64

299

173

1994

196

41

262

037

842

515

45

113

074

Sam

ple

isH

allndash

Lie

bman

rm

sw

ith

Dec

embe

r

scal

year

sth

atha

dco

mp

lete

dat

afo

ron

eof

the

thre

eta

x-re

form

per

iod

sst

ud

ied

Taxation of Executive Compensation 39

1994 level These patterns are consistent with Goolsbeersquos claim that tax-payers responded to the anticipated increase in marginal tax rates byshifting gains into 1992

The evidence from the other two tax reforms is less clear suggestingthat there might be factors other than tax rates determining the timing ofoption gains ERTA 1981 reduced the top marginal tax rate from 70percent in 1980 to 50 percent in 1982 However there is no sign thatoption gains were reduced in 1980 and 1981 and then increased in 198217

There was a big increase in option gains in 1983 however Similarly thehighest year of stock-option gains in the period around 1986 was 1987 astrange year to take gains since taxpayers would have known that thepersonal rate would be 10 percentage points lower if they waited untilthe following year18

What all the peak years of option gains have in common is that theyfollow years of strong stock market performance The fth row of Table 7shows the average stock market return during the previous year for therms in our sample The large value of option gains in 1983 followed a355-percent increase in the stock market The 1992 boom in option gainsfollowed a 466-percent increase in the stock market the more modest1987 peak in option gains followed a 19-percent gain (and a 43-percentgain two years before)

Strong stock market performance raises the value of options availableto exercise The sixth row of the table shows the mean value of optionsheld by CEOs at the beginning of the year All three of the peak years ofoption gains correspond with peak years in terms of the value of optionsavailable for exercise Since stock market appreciation during the yearalso increases the value of options available to exercise later in that yearwe need to allow for current-year stock market appreciation as well

Table 8 contains Goolsbee-style regressions for each of the tax re-forms The rst two columns reprint the results from Goolsbee (1999)The third and fourth columns replicate his results in our sample Inparticular with a specication similar to his we are able to replicate hisnding that the current after-tax share has a positive coefcient and thefollowing yearrsquos after-tax share has a coefcient of roughly negative oneColumns (5) through (8) show that this pattern was unique to the 1993tax reform Indeed in the TRA 86 regressions the coefcient on the

17 As we explained above for executives covered by the maximum tax on earned incomethere was not a decline in tax rates between 1981 and 198218 It is possible that the large stock market decline in October 1997 may have promptedsome people to exercise options earlier than they otherwise would have However anincrease in option exercises typically follows stock market increases not declines

TA

BL

E8

Att

empt

sto

Rep

lica

teth

eT

ax-S

hift

ing

Res

ult

inO

ther

Tim

eP

erio

ds

Goo

lsbe

eG

ools

bee

1993

1993

Ent

ire

Ent

ire

regr

essi

onre

gres

sion

refo

rmre

form

TR

A86

TR

A86

ER

TA

81E

RT

A81

1980

ndash199

419

80ndash1

994

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

ln[1

2ta

x(t)

]1

244

108

20

455

032

62

017

62

017

22

016

72

016

70

026

20

068

(01

34)

(01

27)

(05

37)

(05

42)

(02

10)

(02

09)

(00

90)

(00

91)

(00

67)

(00

65)

ln[1

2ta

x(t1

1)]

20

867

20

764

21

214

20

917

047

80

722

20

103

004

30

176

018

3(0

118

)(0

112

)(0

491

)(0

499

)(0

157

)(0

174

)(0

072

)(0

072

)(0

068

)(0

066

)

Tim

e0

135

007

62

000

52

036

40

041

000

80

094

003

50

056

002

7(0

008

)(0

008

)(0

049

)(0

049

)(0

028

)(0

028

)(0

025

)(0

026

)(0

003

)(0

003

)

ln(m

arke

tva

lue)

066

10

489

023

70

334

039

5(0

015

)(0

058

)(0

039

)(0

042

)(0

017

)

Yea

rs19

91ndash1

995

1991

ndash199

519

90ndash1

994

1990

ndash199

419

85ndash1

989

1985

ndash198

919

80ndash1

983

1980

ndash198

319

80ndash1

994

1980

ndash199

4

Ad

just

edR

20

720

770

610

630

690

760

800

810

630

68

Obs

erva

tion

s13

856

132

371

385

136

41

470

143

31

113

106

23

968

385

9

Dep

end

ent

vari

able

isth

elo

gof

taxa

ble

inco

me

All

regr

essi

ons

con

tain

rm

xe

def

fect

sG

ools

bee

regr

essi

on(2

)al

soco

ntai

nstw

oot

her

vari

able

sno

tsh

own

inth

eta

ble

Stan

dar

der

rors

inp

aren

thes

es

Taxation of Executive Compensation 41

TABLE 9Alternative Explanations for the Timing of Option Exercises

1993 Entirereform TRA 86 ERTA 81 1980ndash1994

(1) (2) (3) (4)

ln[12tax(t)] 0108 0111 20051 0201(0324) (0136) (0048) (0057)

ln[12tax(t11)] 20134 0573 0125 0150(0336) (0156) (0053) (0066)

ln(market value) 0418 0177 0338 0418(0061) (0039) (0046) (0163)

Firmrsquos stock market return (t) 0298 0268 0214 0258(0044) (0045) (0031) (0023)

Firmrsquos stock market return 0162 0206 0031 0128(t21) (0041) (0008) (0029) (0021)

ln(value of options held by 0004 0008 0001 0009CEO at start of year) (0003) (0003) (0002) (0001)

Years 1990ndash1994 1985ndash1989 1980ndash1983 1980ndash1994Adjusted R2 064 071 082 068Observations 1349 1427 1036 3812

The dependent variable is the log of taxable income All regressions contain rm xed effects Stan-dard errors in parentheses

following yearrsquos tax rate is positive and signicant Columns (9) and (10)pool all three reforms and fail to replicate the pattern that would reectintertemporal shifting19

It is possible that the ease with which one can shift the timing ofoption gains is asymmetric It may be harder to shift option gains for-ward in time since some options expire and therefore must be exercisednow In addition options were a much smaller portion of compensationin the early 1980s so perhaps the ERTA 81 results are not surprisingNonetheless these results cast doubt on the proposition that this is thechannel through which spurious estimates of the elasticity of taxableincome have occurred for previous tax reforms and they suggest thatwe should look harder for alternative explanations for the timing ofoption gains

Table 9 attempts to predict options gains using the lagged and currentincrease in a rmrsquos market value and the value of the options held by theCEO at the start of the year The two stock market variables stronglypredict option gains while the value of options held has the correct sign

19 The results also fail to suggest large permanent effects of marginal tax rates on taxableincome

42 Hall amp Liebman

in all of the specications and is signicant in the pooled sample eventhough it is highly correlated with the two stock market return variablesIn contrast the tax-timing explanation fails even in the 1993 period

It is important to realize that our results do not invalidate Goolsbeersquoscritique of claims that aggregate revenue data from 1992 and 1993 provethat elasticities of taxable income are large We have simply given analternative explanation for why taxable incomes were unusually high in1992 However our results do cast doubt on the conclusion that largetiming shifts are ubiquitous and capable of explaining away the morecareful estimates of taxable income elasticities that have been performedusing micro data (eg Feldstein 1995 Auten and Carroll 1995)

One further point is in order It is sometimes argued that the combina-tion of increased use of options and the booming stock market explainswhy federal tax revenues have been so much higher than expected dur-ing the 1990s However while large option gains increase revenue fromthe personal income tax they produce offsetting deductions for cor-porations The net impact on revenue is only the difference in the twomarginal tax rates times the option gain While the gap between thepersonal rate and the effective corporate rate is nontrivial the impact oflarge option exercises on total revenue is substantially smaller than theamount by which the exercises inate personal tax revenue and reducecorporate tax revenue

9 CONCLUSION

We have described the tax rules for executive pay in detail and analyzedhow changes in various tax rates affect the tax advantages of stock optionsrelative to salary and bonus We nd that there is a moderate tax advan-tage to optionsmdashcurrently about $4 per $100 in pretax executive paymdashand that changes in the tax advantage of options over time have had atmost a modest impact on the composition of pay Corporate governancefactors particularly the role of large investors appear to be more impor-tant in explaining the dramatic increase in option pay

In addition our evidence suggests that more direct attempts to use taxpolicy to inuence executive compensation have had little effect Wend that the million-dollar rule led companies to substitute perfor-mance-based pay for salary But our evidence suggests that this substitu-tion was quite modest and there is no evidence that the total level of paywas reduced Overall although the stock-option explosion has dramati-cally increased the link between pay and performance this change isdue almost entirely to nontax factors

Taxation of Executive Compensation 43

REFERENCES

Auerbach Alan J (1996) ldquoThe Dynamic Effects of Tax Law Asymmetriesrdquo Re-view of Economic Studies 53451ndash405

mdashmdashmdash and James Poterba (1987) ldquoTax Loss Carryforwards and Corporate TaxIncentivesrdquo In The Effects of Taxation on Capital Accumulation M Feldstein(ed) Chicago University of Chicago Press

Altshuler Rosanne and Alan Auerbach (1990) ldquoThe Signicance of Tax LawAsymmetries An Empirical Investigationrdquo Quarterly Journal of Economics 10561ndash86

Auten Gerald and Robert Carroll (1995) ldquoThe Effect of Income Taxes on House-hold Behaviorrdquo Ofce of Tax Analysis US Department of the Treasury

Baker George P and Brian J Hall (1998) ldquoCEO Incentives and Firm SizerdquoNBER Working Paper no 6868 December

Core John and Wayne Guay (1999) ldquoThe Use of Equity Grants to ManageOptimal Equity Incentive Levelsrdquo Draft

Feldstein M (1995) ldquoThe Effect of Marginal Tax Rates on Taxable Income APanel Study of the 1986 Tax Reform Actrdquo Journal of Political Economy 103 (no3)551ndash72

mdashmdashmdash and D Feenberg (1996) ldquoThe Effect of Increased Tax Rates on TaxableIncome and Economic Efciency A Preliminary Analysis of the 1993 Tax RateIncreasesrdquo In Tax Policy and the Economy vol 10 James M Poterba (ed)Cambridge MA MIT Press

Gompers Paul A and Andrew Metrick (1998) ldquoInstitutional Investors andEquity Pricesrdquo NBER Working Paper no 6723

Goolsbee Austan (1999) ldquoWhat Happens When You Tax the Rich Evidencefrom Executive Compensationrdquo NBER Working Paper no 6333

Graham John R (1996) ldquoProxies for the Corporate Marginal Tax Raterdquo Journal ofFinancial Economics 42187ndash221

Hall Brian J (1999) ldquoThe Design of Multi-year Stock Option Plansrdquo Journal ofApplied Corporate Finance 12(2)97ndash106

mdashmdashmdash and Jeffrey B Liebman (1998) ldquoAre CEOs Really Paid Like Bureau-cratsrdquo Quarterly Journal of Economics CXII(no 3)653ndash691

Jensen Michael C and William M Meckling (1976) ldquoTheory of the Firm Mana-gerial Behavior Agency Costs and Ownership Structurerdquo Journal of FinancialEconomics III305ndash360

mdashmdashmdash and Kevin J Murphy (1990) ldquoPerformance Pay and Top-ManagementIncentivesrdquo Journal of Political Economy XCVIII225ndash264

Kaplan Steven N (1997) ldquoThe Evolution of US Corporate Governance WeAre All Henry Kravis Nowrdquo Draft

Livingstone Jane R (1999) ldquoExecutive Compensation Contracting and Re-sponses to Increased Tax Costsrdquo Louisiana State University June

Miller Merton H and Myron S Scholes (1982) ldquoExecutive CompensationTaxes and Incentivesrdquo In Financial Economics Essays in Honor of Paul CootnerWilliam F Sharpe and Cathryn M Cootner (eds) Princeton NJ Prentice-Hall

Murphy Kevin J (1999) ldquoExecutive Compensationrdquo forthcoming in OrleyAshenfelter and David Card (eds) Handbook of Labor Economics Vol 3 NorthHolland

Perry Todd and Marc Zenner (1999) ldquoPay for Performance Government Regu-

Taxation of Executive Compensation 3

rates Although the tax advantage of options has approximately doubledsince the early 1980s options currently have only a slight tax advantagerelative to cashmdashapproximately $4 per $100 of pretax compensation

We then analyze what we believe to be the three central policy ques-tions regarding the taxation of executive pay First we examine theextent to which the stock-option explosion has been inuenced by themany changes in tax rates over the past 20 years The evidence suggeststhat changes in taxation have likely had a very modest inuence on theoption explosion Instead changes in corporate governance especiallyin the role of large institutional investors appear to have provided themain impetus for the increase in stock-based pay

Second we examine the effectiveness of tax policies aimed at curbingwhat is deemed by some to be excessive levels of executive compensa-tion Section 162(m) of the Internal Revenue Code (the million-dollarrule) which was enacted in 1993 put a $1-million limit on the deducti-bility (against corporate prots) of non-performance-related executivepay Although we nd evidence that this rule led to a shift in the compo-sition of paymdashaway from salary and toward more performance-relatedbonuses and stock optionsmdashour evidence suggests that the magnitudeof this substitution was small We nd no evidence that the million-dollar rule decreased total executive compensation

Third we analyze the degree to which the tax code is efcient inraising tax revenue from top executives The stock-option explosion hasled to a new and important way for executives to lower their taxes inresponse to changes in tax rates by timing their stock-option gains Forexample in 1993 it was widely reported in the press that well-knownCEOs such as Disneyrsquos Michael Eisner pushed their option gains into1992 in order to avoid paying the higher personal income-tax rates imple-mented in 1993 In an important paper Goolsbee (1999) argues that thetax shifting between 1992 and 1993 was enormous and was the directresult of the increase in marginal tax rates during this period Afterreplicating Goolsbeersquos evidence regarding tax shifting with our data forthe 1993 tax reform we show that no such shifting occurred in responseto either of the two tax reforms of the 1980s Moreover our evidenceindicates that the stock market run-up in 1991 and 1992 was more impor-tant than the change in marginal tax rates in causing the large optiongains observed in 1992

This paper proceeds as follows In the next section we discuss trendsin the level and performance sensitivity of executive compensation Insection 3 we describe the tax and accounting rules concerning executivecompensation In section 4 we analyze how taxes affect the degree towhich options are favored relative to cash In section 5 we examine how

4 Hall amp Liebman

the tax advantage of options has changed over time In section 6 weprovide evidence on the effect of taxation on the composition of execu-tive compensation Section 7 contains empirical analysis of the million-dollar rule Section 8 contains evidence on tax shifting and option gainsSection 9 concludes

2 TRENDS IN TOP EXECUTIVE PAY

In this section we document how top executive pay has changed overtime and discuss how this change has caused the sensitivity of CEOwealth to rm market value to increase substantially There has been alarge increase in the level of CEO pay since 1980 and this growth hasbeen driven by the dramatic increase in stock-option grants during thistime (Hall and Liebman 1998) Although salary and bonuses nearlydoubled over the period in ination-adjusted terms the mean value ofstock-option grants increased by 683 percent The percentage increase inthe median stock-option award cannot be calculated because the me-dian stock-option grant was zero in 1980 The median CEO did notreceive an annual stock-option grant until 1985 Today nearly all topexecutives of large companies receive stock options and the averagestock-option grant is now larger for most top executives than salary andbonus combined

The (ination-adjusted) growth rate of CEO pay since 1980mdashmeasurednarrowly (cash pay grew at an annual rate of 5 percent per year) orbroadly(cash plus option grants grew at almost 9 percent per year) or very broadly(total compensation including stock and stock-option appreciation grewat 115 percent per year)mdashhas been large relative to virtually all othergroups Indeed the growth rate of CEO pay since 1980 has been higheven relative to the pay increases of other high-income earners For ex-ample the cutoff point for being in the top 05 percent of adjusted grossincome (AGI) increased by about 37 percent per year about half the ratefor direct CEO compensation (excluding stock and option appreciation)The only workers who appear to have had faster compensation growththan CEOs are other ldquosuperstarsrdquo The annual pay of professional base-ball players increased by approximately 98 percent per year and that ofprofessional basketball players by 139 percent per year

The increase in stock options has led to a large increase in the equityholdings of top executives and this in turn has led to a dramatic increasein the responsiveness of executive wealth to rm performance Nearlyall of the pay-to-performance sensitivity of executive compensationcomes from equity holdings for a given increase in shareholder valuechanges in the value of an executiversquos stock and stock options are more

Taxation of Executive Compensation 5

than 50 times larger than changes in salary and bonus (Hall and Lieb-man 1998)

As a concrete example the estimates in Hall and Liebman (1998) implythat a 10-percent increase in rm value (of the median company in oursample) leads the company to increase the CEOrsquos salary and bonus byabout $25000 However this same 10-percent increase in shareholdervalue translates into $125-million increase in the value of the CEOrsquosstock and stock-option holdings1

The dramatic rise in the link between CEO wealth and rm perfor-mance can be seen in Figure 1 which shows how two measures of thislink have increased since 19802 The rst measure the JensenndashMurphy(1990) sharing rate (shown on the left scale) is the change in CEO wealthfor a $1000 change in rm value The second measure is the change inCEO wealth for a 10-percent change in rm value (see Baker and Hall1998) Since both measures are strongly affected by rm size (the formerhas a negative correlation and the latter has a positive correlation) thepay-to-performance changes over time are estimated with regression(quantile) analysis that controls for changes in the distribution of rmsizes in the sample over time The gure therefore shows the increasein the pay-to-performance measures over time for a constant-size rmin this case a $1-billion rm (in constant 1998 dollars) The striking fact isthat both measures of the pay-to-performance link have increased bynearly a factor of 10 since 1980 These pay-to-performance increases areeven larger than those we reported in our earlier paper that analyzed theperiod 1980 to 1994 because of the large increase in stock-option grantscombined with the strong stock market performance in 1994 to 1998

3 TAX AND ACCOUNTING RULES

Stock options give an executive the right but not the obligation to buy ashare of the companyrsquos stock at a prespecied pricemdashthe exercise or strikeprice Typically options cannot be exercised immediately That is theyvest (become owned by the executive who can then exercise if he or she

1 Stock-option grants are also very sensitive to changes in rm performance mostly be-cause many grants are multiyear plans that hold the yearly number of options constantand the same number of at-the-money options are worth more when the stock price ishigher and vice versa If stock-option grant sensitivity is also included then about 91percent of pay-to-performance sensitivity comes from stock and stock-option revaluations7 percent comes from stock-option grant changes and less than 2 percent comes fromchanges in salary and bonus (Hall 1999)2 Both measures include only the link created by CEO holdings of stock and stock optionsand ignore the smaller amount of pay-to-performance sensitivity that operates throughchanges in salary bonus and stock-option grants

FIG

UR

E1

Pay

toP

erfo

rman

ceT

heL

ink

betw

een

CE

OW

ealt

han

dSh

areh

olde

rV

alue

Sinc

e19

80

Not

eT

he

anal

ysis

cont

rols

for

the

chan

ging

com

pos

itio

nof

rm

size

over

tim

eT

hu

sth

ese

esti

mat

essh

owth

ep

ay-t

o-p

erfo

rman

celin

kfo

ra

sim

ilar

-siz

edco

mp

any

over

tim

ein

this

case

aco

mp

any

wit

ha

mar

ket

valu

eof

$1bi

llion

Taxation of Executive Compensation 7

wishes) slowly over time Common vesting periods are in the three- tove-year range and options usually vest linearly (eg a four-year op-tion vests at 25 percent at the end of each year) An executive typicallyloses any unvested options upon departure Although options may beexercised as soon as they vest they do not have to be exercised untilthey expire or mature Almost 85 percent of stock-option plans have aterm of exactly ten years with virtually all of the remainder being in theve- to ten-year range About 95 percent of options are granted at themoney or at fair market value which means that the exercise price at grantdate is set equal to the stock price at grant date The remaining 5 percentare either discount options (so-called in-the-money options where theexercise price is below the stock price at grant date) or premium options(so-called out-of-the-money options where the exercise price is abovethe stock price at grant date) The holders of options typically do nothave dividend rights or voting rights even on vested options3

31 Tax Rules and Stock-Option CompensationUnlike salary and bonus stock-option grants are typically an untaxedevent at the time of grant For the most widely used optionsmdashnonquali-ed stock options (NQSOs)mdashexecutives are taxed at the personal incometax rate on option prots (the difference between that stock price and theexercise price times the number of options) when the options are exer-cised The company receives a parallel deduction against corporate in-come at that point If the executive continues to hold the shares afterexercise any subsequent appreciation is taxed at the capital gains rate inthe usual way In 1993 an additional feature was added to the tax code[Internal Revenue Code section 162(M)] that disallowed a corporate de-duction for any executive pay above $1 million that is not performance-based While this rule affects executive salaries most bonuses qualify asperformance-based and standard stock options automatically qualifyTherefore this provision gives companies with highly paid executives anincentive to give more pay in the form of bonuses and stock options asubject we return to in section 7 A summary description of the tax (andaccounting) treatment of cash and option compensation is in Table 1

A far less common type of option which is estimated to account forabout 5 percent of option grants is the incentive stock option (ISO) WhileISOs are similar to NQSOs in their design they are crucially different intwo respects First they have an annual cap of $100000 per executive

3 See Murphy (1999) for details about stock options and Miller and Scholes (1982) on taxincentives

TA

BL

E1

Tax

and

Acc

ount

ing

Tre

atm

ent

ofE

xecu

tive

Com

pens

atio

nC

ash

Pay

men

tsve

rsus

Stan

dard

Exe

cuti

veO

ptio

ns

Tax

atio

nat

Tax

atio

nat

Acc

ount

ing

Com

pens

atio

nD

escr

ipti

onp

erso

nall

evel

corp

orat

ele

vel

trea

tmen

t

Sala

ryan

dbo

nu

sC

ash

pay

men

tsm

ade

Tax

edas

ord

inar

yFu

llyd

edu

ctib

leS

ince

Exp

ense

dag

ains

tea

rn-

toC

EO

in

com

e19

93a

ny

pay

over

$1in

gsin

the

usu

alw

ay

mill

ion

isno

td

educ

t-ib

leu

nles

sit

isp

er-

form

ance

-bas

ed

Non

-qua

lie

dst

ock

Op

tion

sgr

ant

the

ex-

The

gran

ting

ofan

op-

Apa

ralle

ltax

ded

uc-

The

reis

no

exp

ense

rec-

opti

ons

(NQ

SOs)

ecut

ive

the

righ

tbu

tti

onis

typ

ical

lyan

un-

tion

atth

eco

rpor

ate

ogni

tion

(atg

ran

tex

er-

not

the

obli

gati

ont

ota

xed

even

tU

pon

exer

-le

veli

sge

nera

llyal

-ci

seo

rsa

le)

for

op-

buy

shar

esof

stoc

kat

cise

opt

ion

pro

ts

are

low

edu

pon

exer

cise

tion

sif

as

isu

sual

lyan

exer

cise

pric

e(t

ypi-

taxe

dat

the

ord

inar

y(f

orth

eam

ount

ofop

-th

eca

set

he

opti

ons

cally

the

curr

ent

stoc

kra

teI

fth

eex

ecu

tive

tion

pro

ts

)O

pti

ons

are

not

dis

cou

nted

(in

pri

ceat

the

dat

eof

cont

inu

esto

hold

onto

auto

mat

ical

lyqu

alif

yas

the

mon

eyat

gran

tgr

ant)

Th

eop

tion

sth

esh

ares

the

exec

u-

perf

orm

ance

-bas

edp

ayd

ate)

and

the

exer

cise

typ

ical

lyve

stov

era

3ndashti

veis

taxe

din

the

and

ther

efor

ear

eno

tpr

ice

and

num

ber

ofop

-5-

year

per

iod

The

typ

i-us

ualw

ay(a

tth

eca

pi-

subj

ect

toth

e$1

-ti

ons

iskn

own

atgr

ant

calm

atu

rity

is10

year

sta

lgai

nsra

te)

onan

ym

illio

nca

p

dat

eFo

rd

isco

unt

op-

but

the

man

ager

may

furt

her

app

reci

atio

nof

tion

sth

ed

iffe

renc

ebe

-ex

erci

seea

rly

Som

eth

esh

are

pri

ce

twee

nth

est

ock

pri

ceop

tion

sar

edi

scou

ntop

-an

dth

em

arke

tp

rice

isti

ons

(the

exer

cise

pri

ceex

pen

sed

over

the

vest

-is

low

erth

anth

est

ock

ing

per

iod

For

opti

ons

pri

ceat

the

gran

td

ate)

w

ith

vari

able

term

sP

rem

ium

opti

ons

are

(eg

a

vari

able

exer

-th

ere

vers

e(t

heex

er-

cise

pri

cev

esti

ngth

atci

sep

rice

isgr

eate

ris

tied

top

erfo

rman

ce)

than

the

stoc

kp

rice

atth

eop

tion

sar

em

arke

dgr

ant

dat

e)

tom

arke

tan

dex

-p

ense

dd

uri

ngth

eti

me

betw

een

gran

tan

dex

erci

se

Not

eW

eth

ank

com

pen

sati

onco

nsu

ltan

tsSc

ott

Gre

enbe

rgan

dSc

ott

Ols

en(b

oth

ofT

ower

sP

erri

n)

and

Fred

eric

Coo

k(o

fFr

eder

icW

C

ook

and

Ass

ocia

tes)

for

help

fulc

onve

rsat

ion

sin

und

erst

and

ing

the

tax

and

acco

unt

ing

rule

s

10 Hall amp Liebman

Second the tax treatment of ISOs is different ISOs are completely un-taxed at grant or exercise It is only at sale that the executive is taxed andeven then the executive is taxed at the lower capital gains tax rate Thedisadvantage is that the corporation never gets to take a parallel taxdeduction against corporate prots Thus ISOs become more attractiveas the personal tax rate increases and as the corporate tax rate and thecapital gains tax rates fall

Two other related types of compensation restricted stock and stockappreciation rights are worth brief discussion Both however are farless common than standard stock options Restricted stock is payment inthe form of restricted shares the restriction being that the shares vestover time as with options Unlike options the shares typically havevoting and dividend rights With regard to taxation the executive istaxed at the personal rate on the value of the restricted stock as thevesting restrictions lapse However the executive may choose to betaxed at the grant date in which case all subsequent appreciation istaxed at the capital gains rate The dividends paid to the executive aretaxed at the ordinary rate in the usual way The company generallyreceives a parallel deduction equal to the amount of the executiversquosincome when the executive is taxed Unlike stock options restrictedstock is not generally considered to be performance-based and is there-fore subject to the million-dollar rule (unless the vesting of the restrictedstock is performance-based which is sometimes the case)

Stock appreciation rights (SARs) are rights that replicate the payoffs ofstock options with a cash transfer Thus SARs are simpler than options inthat there is no requirement to buy and resell the stock in order to ldquocashoutrdquo SARs generally have the same tax treatment as NQSOs both to theindividual and to the corporation SARs are relatively rare however be-cause they have disadvantageous accounting treatment (described in thenext section) relative to stock options and their main relative advantagevis-a-vis options has been essentially replicated through broker-assistedcashless option exercisesmdashwhereby a broker makes a ldquonanosecondrdquo loanto the executive (to purchase and resell the stock) when the executivewants to ldquocash inrdquo on option prots The tax and accounting treatment ofISOs restricted stock and SARs is summarized in Table 2

32 The Accounting Treatment of OptionsUnlike cash compensation which is expensed against earnings there isgenerally no expense recognition (at grant exercise or sale) for optionswhether they be NQSOs or ISOs As a result compensation consultantsoften point out that stock options are the only form of compensation thatare free in an accounting sense but still deductible for tax purposes

Taxation of Executive Compensation 11

Stock options do however lead to expense recognition if they are dis-counted (in the money at grant date) or if the exercise price and numberof options are not known at grant date For discount options the differ-ence between the stock price and the market price is expensed over thevesting period For options with variable terms (eg a variable exerciseprice vesting that is tied to performance) the options are marked tomarket and expensed during the time between grant and exercise Practi-tioners claim that the accounting treatment of options plays an impor-tant role in the design of option programs Thus plans that have ldquobadaccountingrdquo but are thought by many to have attractive incentive fea-tures are often not even seriously considered by companies Examples ofsuch potentially attractive plans include indexed options (where a CEOprots only if his rmrsquos share price grows relative to some market orindustry benchmark) and option grants that are explicitly performance-related both of which would lead to expenses against earnings

Unlike options restricted stock and SARs do not generally receivefavorable accounting treatment Restricted stock is generally expensedover the period in which the restrictions lapse (usually the vesting pe-riod) The magnitude of the expense is the difference between the cur-rent stock price and the executiversquos cost (if any) SARs are marked tomarket each period and the difference between the stock price and theexercise price is expensed over the outstanding period of SARs

4 THE TAXATION OF EXECUTIVE PAY CASHVERSUS OPTIONS

Stock-based compensation performs two roles in executive compensa-tion arrangements First it helps align the incentives of the executivewith the interests of the rmrsquos shareholders Second it often enables therm to compensate the CEO in a way that is more advantageous from atax standpoint than paying the executive in salary and bonus

41 Agency Theory and Executive Compensation in the Presenceof TaxationIn standard agency-theory models ( Jensen and Meckling 1976) agencycosts are the result of the separation of ownership and control Managersdo not have the same incentives as the owners The optimal incentivecontract for managers balances the benets of high-powered incentives(linking the fortunes of owners and managers through stock and stockoptions for example) with the costs of loading too much risk on risk-averse managers

TA

BL

E2

Tax

and

Acc

ount

ing

Tre

atm

ent

ofE

xecu

tive

Com

pens

atio

nIn

cent

ive

Stoc

kO

ptio

ns(I

SOs)

Res

tric

ted

Stoc

kan

dSt

ock

App

reci

atio

nR

ight

s(S

AR

s)

Tax

atio

nat

Tax

atio

nat

Acc

ount

ing

Com

pen

sati

onD

escr

ipti

onp

erso

nall

evel

corp

orat

ele

vel

trea

tmen

t

Ince

nti

vest

ock

ISO

sar

eid

enti

calt

oIf

the

usu

alco

ndit

ions

Ifth

eu

sual

cond

itio

nsA

sw

ith

NQ

SOs

ther

eop

tion

sN

QSO

sin

thei

rd

esig

n

are

met

th

ere

isno

tax

are

met

the

reis

gene

r-is

gene

rally

noex

pen

seH

owev

ert

hey

have

anat

gran

tor

exer

cise

At

ally

no

corp

orat

eta

xd

e-re

cogn

itio

nfo

rIS

Os

annu

alca

pof

$100

000

sale

all

opti

onp

ro

tsd

uct

ion

atgr

ant

exer

-(e

ver)

on

the

amou

ntth

atan

dsu

bseq

uen

tca

pita

lci

seo

rsa

le

vest

sT

he

tax

trea

t-ap

prec

iati

onar

eta

xed

men

tof

ISO

sal

soat

the

cap

ital

gain

sd

iffe

rs

rate

Res

tric

ted

stoc

kSi

mila

rto

stoc

kop

-T

hein

div

idu

alis

taxe

dT

heco

mp

any

gene

rall

yR

estr

icte

dst

ock

isge

n-ti

ons

but

shar

esat

the

ord

inar

yra

teas

rece

ives

ap

aral

leld

e-er

ally

exp

ense

dov

er(w

hich

vest

slow

ly)

the

rest

rict

ions

lap

se

du

ctio

neq

ual

toth

eth

eve

stin

gp

erio

dT

hera

ther

than

opti

ons

are

How

ever

the

man

ager

amou

ntof

the

exec

u-

amou

nt

exp

ense

dis

gran

ted

toth

eex

ecu

-m

aych

ose

tobe

taxe

dti

versquos

inco

me

whe

nth

eth

ed

iffe

renc

ebe

twee

nti

veU

nlik

eop

tion

sat

gran

td

ate

inw

hich

exec

uti

veis

taxe

dR

e-th

ecu

rren

tst

ock

pric

eth

esh

ares

typi

cally

case

alls

ubs

equ

ent

ap-

stri

cted

stoc

kis

gen

er-

and

the

exec

uti

versquos

have

voti

ngan

dd

ivi-

pre

ciat

ion

ista

xed

atal

lysu

bjec

tto

the

cost

(if

any)

d

end

righ

ts

the

cap

ital

gain

sra

te

mill

ion-

dol

lar

cap

D

ivid

end

sar

eta

xed

atth

eor

din

ary

rate

Stoc

kap

pre

ciat

ion

Rig

hts

that

repl

icat

eG

ener

ally

have

the

Gen

eral

lyha

veth

eU

nlik

eop

tion

sSA

Rs

righ

tsth

epr

ots

ofst

ock

op-

sam

eta

xtr

eatm

ent

assa

me

tax

trea

tmen

tas

are

mar

ked

tom

arke

tti

ons

Rar

ely

use

d

NQ

SOs

NQ

SOs

each

per

iod

an

dth

esi

nce

stoc

kop

tion

sd

iffe

renc

ebe

twee

nth

eha

vem

ore

favo

rabl

eac

-st

ock

pri

cean

dth

eex

-co

unt

ing

trea

tmen

ter

cise

pri

ceis

exp

ense

d(a

ndop

tion

sp

rovi

de

over

the

outs

tand

ing

the

sam

ebe

ne

tof

per

iod

ofSA

Rs

cash

less

exer

cise

thro

ugh

brok

erlo

ans

)

Not

eW

eth

ank

com

pen

sati

onco

nsul

tant

sSc

ott

Gre

enbe

rgan

dSc

ott

Ols

en(b

oth

ofT

ower

sP

erri

n)an

dFr

eder

icC

ook

(of

Fred

eric

WC

ook

and

Ass

ocia

tes)

for

hel

pfu

lcon

vers

atio

nsin

und

erst

and

ing

the

tax

and

acco

unt

ing

rule

s

14 Hall amp Liebman

The effect of taxation on the optimal contract (more precisely theshare of compensation that is performance-related) is ambiguous (evenin the absence of deductibility and deferral) because there are offsettingeffects First by reducing the share of corporate prots received byshareholders taxes diminish the importance to the shareholders of moti-vating the CEO Taken alone this effect would imply that taxes wouldbe expected to reduce the use of performance-based compensation Sec-ond because the government shares in the income received by the CEOthe variance of the CEOrsquos income is reduced raising his utility andlowering the cost to the rm of providing any given set of incentivesThird because taxes will lead executives to provide less effort for anygiven level of incentive-based pay the level of compensation that mustbe provided to compensate the CEO for effort is reduced for a givenamount of incentive These last two factors make it cheaper for the rmto offer contracts to the CEO with large incentive components when taxrates are higher and should therefore increase the use of incentive-based pay Since the net effect of taxation on the level of incentive-basedpay is ambiguous agency theory provides no strong predictions abouthow taxation should affect the optimal composition of option versuscash compensation

42 The Tax Advantages of DeferralWhile agency theory yields ambiguous predictions about the effect oftaxes on the use of options there are direct tax advantages of options inthat options provide executives with a way to defer compensation andthereby lower their taxes However because options also lead to a defer-ral of corporate tax deductibility the tax advantages from a combined(executive and corporate) perspective are not so straightforward In thissection we analyze and measure the tax advantages of stock optionsrelative to salary and bonus compensation We also analyze the taxadvantages of NQSOs relative to ISOs In particular we show how thetax advantages of options change as personal corporate and capitalgains taxes change We then show how the tax advantages of optionshave changed over time in response to changing tax rates

The crucial tax difference between standard options (NQSOs) andcash payment is that option payouts are deferred and the two forms ofcompensation earn different rates of return over the deferral period Anyanalysis of the relative tax advantages of two compensation instrumentsmust consider the tax consequences both to the employer and to theemployeemdashwhat Scholes and Wolfson (1992) call the global contractingperspective Thus in order to make valid comparisons between the twowe compare the tax burden to the executive while holding constant the

Taxation of Executive Compensation 15

posttax cost to the company By keeping posttax employer costs (in NPVterms) constant any package that is preferred by the employee is tax-advantaged in the global-contracting sense

Under this methodology a comparison of the tax advantages of op-tions and cash involves comparing a pretax cash payment of P with anoption payment that has the equivalent posttax NPV to the companyFor NQSOs it is straightforward to show that a pretax payment of P toan executive has exactly the same cost to the company as putting aside Pfor the purposes of paying stock-option payouts later That is if a com-pany pays an executive P today it will have the same amount of moneyin N years as if it had put aside P today let it accumulate at the rate ofreturn earned by the rm and then paid it out (as compensation in theform of option gains4) with any appreciation in year N

Assume that the pretax prot rate is r and the corporate tax rate is TcThen if the company pays P today it receives a deduction today of PTcwhich yields

PTc [1 1 r (1 2 Tc)]N (1)

in N years since the benets of the tax deduction are invested in thecompany and receive the after-tax corporate rate of return Converselyif the company puts P aside today then it grows at the after-tax corpo-rate rate of return in N years to give a tax deduction of

P [1 1 r (1 2 Tc)]N Tc (2)

which is the exact same value Note that the equivalence of these two taxdeductions is analogous to the tax-benet equivalence of front-loadedand back-loaded IRAs

An executiversquos payoff in N years from option prots is equal to Pcompounded at the after-tax corporate rate times 1 2 Tp where Tp is thepersonal tax rate Thus the combined payoffs in N years of the corporatededuction and the executiversquos posttax payoff is

P [1 1 r (1 2 Tc)]N Tc 1 P [1 1 r (1 2 Tc)]

N (1 2 Tp) (3)

where the rst term is the payoff from the corporate deduction and thesecond term is the executiversquos payoff

4 This analysis is not unique to options Any form of deferred compensation that enablesexecutives to invest inside the rm at a higher posttax return or without paying capitalgains taxes would have similar effects

16 Hall amp Liebman

The payoff to salary and bonus in N years is more complicated sincean assumption must be made about how the executive invests the origi-nal cash compensation For example if the executive invests in an instru-ment with taxable interest (eg bonds) then the N-year payoff from aninvestment of P accumulates at the post-corporate-tax post-personal-taxrate of return and is

P (1 2 Tp) [1 1 r (1 2 Tc)(1 2 Tp)]N (4)

However if an executive invests in (non-dividend-paying) equities theinvestment accumulates at the higher post-corporate-tax rate of returnwith the offsetting disadvantage that the capital gains are taxed at thecapital gains rate in year N So the executive receives

P [1 1 r (1 2 Tc)]N (1 2 Tp) 2 Tcg $P (1 2 Tp) [1 1 r (1 2 Tc)]

N 2P (1 2 Tp) (5)

where Tcg is the capital gains tax rate5Thus putting the equations together if an executive is paid cash P

and invests g in bonds and 1 2 g in equities the N-year payoff (includ-ing the payoff to the corporation from the tax deduction) is

P [1 1 r (1 2 Tc)]N Tc 1 g $(P [1 1 r (1 2 Tc)(1 2 Tp)]

N (1 2 Tp)1(1 2 g) (P [1 1 r (1 2 Tc)]

N (1 2 Tp) 2 Tcg $P (1 2 Tp)[1 1 r (1 2 Tc)]

N 2 P (1 2 Tp)) (6)

The tax advantage of options versus salary and bonus therefore is thedifference between the combined (corporate and executive) payoffs inequation (3) and the combined payoffs in equation (6)

43 The Tax Advantage of OptionsWe now turn to analysis of how changes in various tax rates affectthe tax advantages of options relative to cash compensation We focuson the case in which the cash earned by executives is invested entirelyin equities since equities are tax-favored relative to bonds The impor-tant conclusions of this analysis are not substantively different if we

5 In practice executives hold a combination of equities and bondlike instruments withtaxable interest Because equities are tax-favored in the next section we simplify the analy-sis by assuming that executives hold only equities Alternatively we could have assumedthat executives hold only less risky bonds since bonds more closely match the risk proleof a stream of tax savings None of our empirical results are substantively affected by thissimplication

Taxation of Executive Compensation 17

instead assume that a portion of an executiversquos holdings are invested inbonds

431 The Tax Advantage of Options Changing the Corporate TaxRate Stock options are tax-advantaged relative to cash simply becauseoption payouts are deferred allowing the executive to invest at the pre-personal-tax rate of return (with no capital gains tax at the end) ratherthan investing at the after-tax rate of return or paying the capital gainstax Since the advantage of being able to defer taxes is large when thecorporate tax rate is low the tax advantage of options is larger when thecorporate rate is low

To illustrate this point we dene the tax advantage of options to bethe dollar amount by which total after-tax option payoffs (to the corpora-tion and the executive) exceed the total payoffs from salary as denedby the difference of equation (3) and (6) assuming that P (the paymentto the executive) is equal to $100 and g 5 0 (cash compensation is in-vested in equities) We use a ten-year horizon since most options have aten-year term We then calculate how the tax advantage of optionschanges as the corporate tax rate varies from zero to 100 percent holdingthe personal tax rate and the capital gains rate xed at 40 percent and 20percent respectively (which approximates current rates)

The top panel of Figure 2 shows the tax advantage of options relativeto cashmdashprecisely the NPV of the total tax advantage of options for a$100 payment to the CEOmdashas the corporate tax rate changes As ex-pected the tax advantage of options declines as corporate tax rates in-crease At a zero corporate tax rate the tax advantage of options has anNPV of $7 This value is about $350 at a corporate rate of 40 percentWhen the corporate tax rate is 100 percent the tax advantage of optionscompletely disappears since the benets of deferred compensation fallto zero because the after-tax return on equities (and therefore the dis-count rate) falls to zero The tax advantage of options is the NPV of thetax savings from avoiding the capital gains tax on ten years of apprecia-tion which is why it is relatively modest even at the 40-percent corpo-rate rate

432 The Tax Advantage of Options Changing the Personal TaxRate Using the same assumptions as above (but this time holding thecorporate rate at 40 percent and the capital gains rate at 20 percent andvarying the personal rate) the tax advantage of options declines as thepersonal tax rate rises as shown in the middle panel of Figure 2 The taxadvantage of options is higher at low personal tax rates because theposttax base that gives rise to capital gains is higher and the advantage of

Note Holds constant the personal rate at 40 percent and the capital gains rate at 20 percent

Note Holds constant the corporate rate at 40 percent and the capital gains rate at 20 percent

Note Holds constant the personal rate at 40 percent and the corporate rate at 40 percent

FIGURE 2 Tax Advantage of Options NPV of Tax Advantage of Pay-ing $100 in Options (NQSOs) Rather Than in Cash

Taxation of Executive Compensation 19

avoiding the capital gains taxes is greater The tax advantage of optionstherefore declines linearly as the personal tax rate rises At a 100-percentpersonal tax rate the executive receives nothing in either case and byconstruction the companyrsquos tax deduction benet is the same in NPVterms so the tax advantage of options falls to zero also6

433 The Tax Advantage of Options Changing the Capital Gains TaxRate The tax advantage of options for various capital gains rates isdepicted in the bottom panel of Figure 2 under the same assumptions(this time holding both the personal rate and the corporate rate at 40percent while varying the capital gains rate) The tax advantage of avoid-ing the capital gains tax increases as the capital gains tax rate rises

5 HOW HAS THE TAX ADVANTAGE OF OPTIONSCHANGED OVER TIME

The analysis so far has illustrated how the tax advantage of optionsvaries with the changes in tax rates In this section we show how the taxadvantage of options has changed over time as tax rates have changed

51 Top Marginal Tax Rates over TimeThe evolution on the top marginal tax ratesmdashpersonal corporate andcapital gainsmdashfrom 1980 to 1998 is shown in Figure 3 In all cases thetop rate (the rate for taxpayers with the highest incomes) is shownwhich is not always the highest rate since various anomalies (such asclawbacks of exemptions) sometimes lead to marginal tax rates that arehigher than those for the highest income tax payers

The top personal tax rate was 70 percent in 1980 falling to 50 percentin 1982 and 28 percent in 1988 following the 1986 tax act7 The topmarginal tax rate has since risen to 396 percent but is effectively 425percent since there is a 29-percent Medicare surcharge (paid half by theemployer and half by the employee) that has no upper limit The topcorporate tax rate has had only one signicant change since 1980 fallingfrom 46 percent before the 1986 tax act to 34 percent following the act in1988 The top corporate tax rate was increased to 35 percent in 1993 The

6 This is one place where ignoring bond investments has substantive implications Inparticular in the all-bond case the tax advantage of options is nonmonotonicmdashit rises andthen fallsmdashas personal tax rates rise7 During the early 1980s the maximum tax on earned income limited the marginal tax rateon the earnings of high earners so many executives did not face a decline in personal taxrates between 1981 and 1982 The empirical results in this paper are robust to assumingthat executives faced a marginal tax rate of 50 percent in 1980 and 1981

20 Hall amp Liebman

FIGURE 3 Top Marginal Tax Rates since 1980

top capital gains rate has uctuated between 28 percent (approximately)and 20 percent since 1980 and is currently at 20 percent

52 The Evolution of the Tax Advantage of NQSOsThe tax advantage of options relative to cash dened in the same way asin section 4 from 1980 to 1998 is shown in Figure 4 The calculations aredone using the same assumptions as before (ie N 5 10 years and cashis invested in equities) and the statutory top marginal tax rates shown inFigure 3 are used Due to the fall in corporate and personal tax rates thetax advantage of options increased sharply after the Tax Reform Act of1986 This increase in tax advantage has largely been reversed in the1990s as top personal tax rates have crept back up and the capital gainsrate has been reduced

Two key facts emerge from this analysis First there is currently only amoderate tax advantage to standard nonqualied stock optionsmdashon theorder of $4 per $100 of compensation This is because the tax advantagesto the executive of deferring taxes are largely offset by the tax disad-vantages to the company of not being able to deduct option expensesfrom taxable prots until the executive exercises the options Secondalthough the 1986 tax act substantially increased the tax advantage ofoptions more than half of this increase has been reversed in the 1990s

Taxation of Executive Compensation 21

FIGURE 4 Tax Advantage of Options 1980 to 1998

NPV of tax advantage of $100 in options (NQSOs) rather than cash

53 The Tax (Dis)advantage of ISOsSo far the analysis has focused on the tax advantages of NQSOs sincethey are so much more prevalent than ISOs which account for onlyabout 5 percent of option grants The relative scarcity of ISOs can beexplained both by their tax status and by their per-executive cap of$100000 per year

Under what conditions are ISOs tax-preferred to NQSOs For a trans-fer P to an executive the ISO is always tax-preferred by the executive(since the capital gains rate is lower than the personal rate) while theNQSO is always tax-preferred from the companyrsquos perspective (since anISO is not deductible) The key issue however is the relative advan-tages of the two types of options from a global-contracting perspectiveBecause ISOs are not deductible the company is indifferent betweensetting aside P in the form of NQSOs today and setting aside P(1 2 Tc) inthe form of ISOs (Since the NPV of the tax deduction for NQSOs is PTcthe NPV of the payment P is P 2 PTc or P (1 2 Tc))

Equalizing the posttax cost of NQSOs and ISOs we need only look atthe payoff of the executive to determine the condition under which eachis preferred The payoff of NQSOs in N years is

P [1 1 r (1 2 Tc)]N (1 2 Tp) (7)

22 Hall amp Liebman

The same payoff for ISOs is

P (1 2 Tc) [1 1 r (1 2 Tc)]N (1 2 Tcg) (8)

ISOs therefore are tax-advantaged if

Tp 2 Tcg Tc 2 Tc Tcg (9)

Intuitively ISOs are tax-advantaged only if their advantage (the differ-ence between the personal rate and the capital gains rate) is largeenough to offset their cost (the disadvantage of not deducting at thecorporate rate) Note however that the condition is not a simple com-parison between the corporate rate and the personal rate minus thecapital gains rate (Tp 2 Tcg Tc)

Figure 5 shows how the tax advantage of NQSOs relative to ISOs haschanged since 1980 as tax rates have changed Unlike the very modestchanges (around 3 to 4 percent) in the relative tax advantages betweenoptions and cash shown in Figure 4 the relative tax advantage betweenNQSOs and ISOs has seen enormous swings When the top personalrate was 70 percent in 1980 ISOs were tax-favored by a margin greaterthan 6 percent However by 1982 (following the 1981 tax acts that dra-matically lowered the personal tax rate) ISOs became tax-disadvantaged

FIGURE 5 Tax Advantage of NQSOs Relative to ISOs over Time

NPV of tax advantage of $100 payment to executive NQSOs minus ISOs

Taxation of Executive Compensation 23

by about 5 percent By 1988 (following the 1986 tax act) ISOs became tax-disadvantaged by about 18 percent a dramatic swing The raising ofpersonal rates coupled with a decrease in the capital gains rate has sincereduced the tax disadvantage of ISOs to about $4 per $100 of pretaxcompensation Although hard data on ISOs are hard to come by practi-tioners (mostly compensation consultants) claim that ISOs were morepopular prior to 1982 when they were tax-advantaged Although ISOsare less disadvantageous from a tax standpoint relative to NQSOs thanthey were in the late 1980s they are still disadvantageous so it is notsurprising that they have not made a signicant resurgence in recentyears

6 EXPLAINING THE INCREASE IN OPTION-BASEDCOMPENSATION

In this section we analyze whether changes in tax rates affect the compo-sition of executive compensation and consider the relative importanceof tax factors and corporate governance factors in explaining the in-creasing share of compensation paid in the form of stock options Thenumerous changes in tax rates that have occurred since 1980 provide anopportunity to assess whether executive compensation arrangementsrespond to tax incentives in the way that the tax-avoidance model sug-gests We test this model using a panel data set of CEOs in largepublicly-traded US corporations Our identication of the tax effectsrelies on time-series variation in personal corporate and capital gainstax rates along with cross-sectional variation in corporate tax rates

61 DataWe use a 15-year panel data set of CEOs in the largest publicly-tradedUS corporations which is described in Hall and Liebman (1998) Thedata set covers the years from 1980 through 1994 and combines CEOcompensation information from corporate proxies and 10-K lings withstock price and stock return information from CRSP and accounting datafrom Compustat In addition some compensation data from the 1970swere collected in order to construct measures of the value of stock op-tions held by the CEOs in the rst period of the sample8

The feature that distinguishes our data from most other CEO datasets is that with our panel of yearly proxy data on option grants optiongains and total options held we are able to calculate the total value of

8 We are grateful to David Yermack for providing us with some of the data for 1984 to 1991See Yermack (1995) for a discussion of these data

24 Hall amp Liebman

all stock options held by the CEO at a point of time More importantlysince we have the details about the stock options held (number exer-cise price time to maturity etc) we can precisely calculate the changein the value of a CEOrsquos stock option holdings for a given change in rmvalue

62 Identication IssuesThe numerous tax changes over our sample period and the specic waysin which changes in personal corporate and capital gains rates arepredicted to affect the composition of CEO compensation give us un-usually rich sources of identication However because many of thechanges in tax rates would be expected to affect all of the CEOs in oursample in a similar way we will need to pay particular attention toseparating out the tax effects from underlying time trends and otherfactors that changed over time Controlling for underlying trends is par-ticularly important in this study because we are focusing on an out-comemdashthe use of stock optionsmdashthat has increased rapidly over thepast two decades and that many practitioners believe was importantlyaffected by nontax considerations Thus the tax effects we are trying toexplain are deviations from a rising trend

We take two steps to try to separate out the tax effects from theunderlying trend First we include key nontax factors that could poten-tially explain the increasing reliance on performance-based compensa-tion It has been suggested that the dramatic increase in incentive-basedpay is the result of the remarkably poor shareholder returns during the1970s which spurred the LBO and takeover movements of the 1980sAccording to this story (Kaplan 1997) shareholders became much morepowerful via the rise of institutional investors even as the LBO andtakeover movements waned Because the inuence of institutional inves-tors is thought to be one of the most important mechanisms of strongcorporate governance we use the share of each companyrsquos stock ownedby institutional investors (dened as institutions with more than $100million under management) as an explanatory variable9 The percentageof shares owned by large institutional investors increased from about 20percent to almost 50 percent in 1994 an upward trend that closelymatches the sharp rise in the share of compensation in the form ofoptions Indeed in our sample the annual average (over all of the rmsin our sample) share of stock owned by institutional investors has acorrelation of 09 with the annual average share of compensation paid inoptions

9 This variable is described in Gompers and Metrick (1998)

Taxation of Executive Compensation 25

In addition we include two other variables that proxy for strongercorporate governance The rst is the size of the board Evidence suggeststhat boards with smaller numbers of directors reduce the inuence of theCEO and better represent shareholder interests (Yermack 1995) Boardsize fell by approximately 15 percent between 1980 and 1994 The secondvariable is the fraction of the rmrsquos board members who are inside direc-tors or gray directors (noninsiders who have business dealings with thecompany) Firms with fewer outsiders are less likely to act in the share-holdersrsquo interests and we would predict that they would be less likely tohave performance-based compensation for their CEOs The average per-centage of inside and gray directors has decreased from about 45 percentin 1980 to about 35 percent in 1994

Second we allow for cross-sectional variation in corporate tax ratesCorporate marginal tax rates vary because rms may be eligible for tax-loss carrybacks and carryforwards investment tax credits and the alter-native minimum tax (see Auerbach 1986 Auerbach and Altshuler1990 Auerbach and Poterba 1987) Because these cross-sectional differ-ences imply that different rms should respond differently to a givenchange in tax rates they potentially provide us with a way to isolatethe tax effects Firm-specic corporate marginal tax rates are notori-ously difcult to calculate from publicly-available data We use a tricho-tomous variable that equals zero if the rm has tax-loss carryforwardsand negative earnings in a given year equals 05 times the statutorycorporate rate if the rm has only one of those conditions and equalsthe statutory rate if the rm has neither loss carryforwards nor negativeearnings10

The cross-sectional variation in the corporate tax rate may not be ex-ogenous Firms that perform poorly could face low marginal tax ratesand also be particularly likely or unlikely to provide performance-basedpay Because most of our specications use xed effects the correlationwe must be concerned about is between changes in rm performanceand changes in compensation We deal with this by controlling forlagged rm stock market performance in our regressions This shouldeliminate spurious correlation between tax rates and compensation thatis jointly caused by rm performance

10 Graham (1996) presents evidence that the trichotomous variable performs nearly as wellas his recommended simulated marginal tax rate Plesko (1999) compares marginal taxrates calculated from Compustat data with actual marginal tax rates from rm tax returnsHe nds that simple binary variables are more reliable measures of the marginal tax ratethan is either the Graham or the trichotomous variable We obtain substantively similarresults when we switch from the trichotomous variable to the binary variables recommendby Plesko

26 Hall amp Liebman

63 ResultsThe dependent variable in our regressions is the fraction of a CEOrsquos totalannual compensation (measured as the BlackndashScholes value of stock-option grants plus salary and bonus) that comes from stock-optiongrants We choose this reduced-form specication because our interest isin the composition of CEO compensation However the factors thataffect the composition of compensation might also affect the levels andit will be important to keep this in mind in interpreting our results

Optimal-contracting theory generally makes predictions about the in-centives provided by the total amount of rm stock and stock optionsowned by the CEO We choose to model the ow of option grants ratherthan the ultimate performance sensitivity of the CEOrsquos compensationand stock and stock-option holdings because we have in mind anadjustment-cost model in which it takes time for a CEOrsquos holdings ofstock and stock options to reach the optimal level11 In our regressionswe include two measures of the current performance sensitivity of theCEOrsquos wealth to rm performance as explanatory variables to reect thedistance of the CEOrsquos contract from the optimal contract

Table 3 presents results in which the tax effects are identied solely byvariation in tax rates over time The fraction of annual compensation paidin stock options is regressed on tax variables corporate governance vari-ables contract-theory variables and variables reecting stock market re-turns Both of the regressions in this table include rm xed effects Incolumn (1) the fraction in annual compensation paid in stock options isregressed on the log difference in payoff (the combined payoff to both therm and the CEO) from receiving compensation in stock options ratherthan salary and bonus where the payoff calculations are based on statu-tory marginal tax rates This variable is dened as the log difference be-tween equation (3) and equation (6) in section 4 The coefcient on the taxvariable is large and statistically signicant at the 95-percent level Thepoint estimate implies that a one-percent increase in the payoff differencebetween stock options and salary and bonus results in a 24-percentage-point increase in the share of compensation paid in stock options

The corporate governance variables all have the predicted sign Firmswith a higher fraction of their shares owned by institutional investorsuse more performance-based pay (stock options) Firms with large corpo-rate boards or a large share of inside directors are less likely to use stockoptions

The two contract-theory variables are the dollar change in CEO wealth

11 See Core and Guay (1999) for evidence that rms adjust in this way

TABLE 3Compensation Regressions that Rely on Variation in

Tax Rates over Time

(1) (2)

Tax variablesln (difference in payoff)(a) 2393 (0349)ln (12statutory personal 20015 (0034)

rate)ln (12statutory cap gains 0330 (0171)

rate)ln (12statutory corporate 0399 (0148)

rate)

Corporate governance variablesFraction of rmrsquos shares 0043 (029) 0052 (0028)

owned by institutionalinvestors

Size of the rmrsquos board of 20002 (001) 20003 (0001)directors

Fraction of directors who 20082 (033) 20092 (0030)are inside or gray

Contract-theory variablesD (CEO wealth) per $1000 20261 (0095) 20247 (0087)

in FV(b)

D (CEO wealth) per 10 in 2583 (189) 2554 (201)FV(c)

Volatility(d) 521 (424) 261 (369)ln (rm market value) 0019 (0006) 0019 (0006)

Stock market return variablesLagged annual rm stock 0006 (0009) 0003 (0008)

market returnLagged annual SampP 500 re- 20035 (0023) 20029 (0023)

turn

Regression diagnosticsTotal sample size 5179 5179Firms 427 427Adjusted R2 0313 0311

Dependent variable share of compensation paid in options Standard errors in parentheses Bothregressions in this table include rm xed effects

(a) Log difference in payoff is the log difference in combined rm and CEO after-tax share betweenpaying compensation in options and paying salary

(b) This variable is the dollar change in CEO wealth per $1000 change in rm market value (seeJensen and Murphy 1990)

(c) This variable in the dollar charge in CEO wealth per 10-percent change in rm value (see Hall andLiebman 1998)

(d) Standard deviation of rm market value in trillions

28 Hall amp Liebman

per $1000 change in rm value (Jensen and Murphy 1990) and thedollar change in CEO wealth per 10-percent change in rm value (Halland Liebman 1998) As discussed in Baker and Hall (1998) these twowealth sensitivity measures reect two different concepts of how closelyaligned the managersrsquo incentives are with the interests of the sharehold-ers In the regression they are meant to measure how far the CEOrsquosexisting contract and ownership of stock and stock options is from theoptimal level The negative coefcients on the two variables suggeststhat rms that are below their optimal pay-to-performance sensitivityare more likely to give more stock options in the current period

The second column replaces the tax variable motivated by deferraladvantages with the three statutory rates The corporate after-tax sharehas a positive coefcient as we would expect because the tax advantageof stock options increases as the corporate rate falls The coefcient onthe personal tax rate is small and statistically insignicant The sign ofthe coefcient on the capital gains rate is the opposite of that predictedby tax-avoidance theory

Note that these results rely on time-series variation alone This is not avery convincing test Since the log difference in payoffs mostly rises overthe time period covered in our sample as does the share of compensa-tion paid in options it is possible that the tax variable is simply reectingother factors that were changing over this time period that we have notincluded in our regression Therefore it is important to look at specica-tions that rely on cross-sectional variation in the tax variables

Table 4 contains specications that rely on cross-sectional variation incorporate tax rates The rst column repeats the specication from col-umn (1) of Table 3 but uses the trichotomous measures of rm corporatetax rates in place of the statutory rate in calculating each rmrsquos logdifference in payoff from paying compensation in options rather than insalary and bonus As in the previous table we nd the positive coef-cient on the log difference in payoff just as the theory suggest How-ever the coefcient estimate is now much smaller suggesting that aone-percent increase in the log difference in payoff produces a 01-percentage-point increase in the share of compensation paid in options

Introducing cross-sectional variation in the dependent variable doesnot eliminate the potentially spurious correlation over time between thetax variable and the share of compensation paid in options The secondcolumn of Table 4 isolates the pooled cross-sectional variation in corpo-rate tax rates by adding time dummies and dropping the rm xedeffects In this specication the coefcient on the tax variable is indistin-guishable from zero The coefcient on the institutional variable contin-ues to be large and statistically signicant The third column includes

Taxation of Executive Compensation 29

both rm xed effects and time dummies which tests for a relationshipbetween within-rm changes in the dependent variable and within-rmchanges in each explanatory variable In this specication the coefcienton the tax variable is again indistinguishable from zero The coefcientson the corporate governance variables are also statistically insignicant

The third column contains a specication that treats the Tax ReformAct of 1986 (TRA 86) as a natural experiment For rms and CEOs facingthe statutory corporate tax rate TRA 86 increased the relative payoff topaying compensation in options by 17 percent For a rm facing a zeromarginal tax rate it reduced the relative payoff by 45 percent The reduc-tion occurred because TRA 86 reduced the personal tax rate makingsalary and bonus more attractive while the decreased corporate rate hadno effect on zero-tax-rate rms Thus we would expect rms facing zeroor low marginal tax rates to reduce their use of options after TRA 86while rms facing the statutory rate would be expected to increase themFew rms face zero marginal tax rates year after year Therefore we tryto distinguish between rms that often face low tax rates and rms thatnearly always face the statutory rate We dene low-tax rms as oneswhose average marginal tax rate over 1984 to 1986 was below 02412

Then we ran a difference-in-differences regression to see if high-taxrms increased their use of options more than low-tax rms The interac-tion between post-TRA 86 and high corporate tax rates is the key vari-able The estimated coefcient on this variable is positive but small andnot statistically different from zero at the 95-percent level

It is important to note that in this specication as well the main corpo-rate governance variable the share of the rmrsquos shares owned by institu-tional investors performs as theory would predict The variable appearsto be quite robust Table 5 presents regressions of the share of compensa-tion paid in stock options on the fraction of the rmrsquos shares owned byinstitutional investors Since we have already established that these vari-ables have a strong longitudinal relationship the regressions here aredesigned to focus on the cross-sectional relationship between these twovariables The rst and third columns present regressions using data onlyfrom 1994 and nd a coefcient of roughly 02 (column 1 contains noother covariates while column 3 includes all of the other nontax covari-ates from the regressions in Tables 3 and 4) Since the median share of armrsquos stock owned by institutional investors in our sample rises from 19percent in 1980 to 49 percent in 1994 this coefcient would imply that this

12 This cutoff was chosen after inspecting the distribution of tax rates and essentiallyseparates rms that are always at or near the statutory rate from ones that are sometimesbelow it

TA

BL

E4

Com

pens

atio

nR

egre

ssio

nsth

atR

ely

onC

ross

-Sec

tion

alV

aria

tion

inC

orpo

rate

Tax

Rat

es

(1)

(2)

(3)

(4)

Tax

vari

able

s(a)

ln

(dif

fere

nce

inp

ayof

f)0

099

(00

41)

20

038

(00

36)

20

005

(00

44)

Pos

t-T

RA

862

000

5(0

013

)H

igh

corp

orat

eta

xra

te0

018

(00

17)

(Pos

t-T

RA

86)

3hi

gh0

024

(00

16)

corp

orat

eta

xra

te

Cor

por

ate

gove

rnan

ceva

riab

les

Frac

tion

of

rmrsquos

shar

es0

109

(00

27)

011

4(0

015

)2

000

9(0

030

)0

087

(00

34)

owne

dby

inst

itu

tion

alin

vest

ors

Size

ofth

e

rmrsquos

boar

dof

20

004

(00

01)

20

001

(00

01)

20

002

(00

01)

000

1(0

002

)d

irec

tors

Frac

tion

ofd

irec

tors

wh

o2

012

4(0

033

)2

000

3(0

018

)2

004

1(0

034

)0

001

(00

39)

are

insi

de

orgr

ay

Con

trac

t-th

eory

vari

able

sD

(CE

Ow

ealt

h)p

er$1

000

20

238

(00

94)

20

167

(00

53)

20

257

(00

96)

20

021

(01

47)

inFV

(b)

D(C

EO

wea

lth)

per

10

in2

560

(17

8)2

592

(10

5)2

594

(19

6)2

147

(04

86)

FV(c

)

Vol

atili

ty(d

)2

48(4

24)

162

(03

4)1

12(4

28)

106

(06

09)

ln(

rmm

arke

tva

lue)

003

3(

006)

000

8(0

003

)0

009

(00

06)

001

2(0

007

)

Stoc

km

arke

tre

turn

vari

able

sL

agge

dan

nual

rm

stoc

k2

000

2(0

009

)0

011

(00

10)

000

6(0

009

)2

001

2(0

013

)m

arke

tre

turn

Lag

ged

annu

alSamp

P50

02

005

6(0

023

)2

005

3(0

051

)2

003

9(0

049

)2

015

4(0

040

)re

turn

Reg

ress

ion

dia

gnos

tics

T

otal

sam

ple

size

5179

5179

5179

2534

Firm

s42

742

739

6A

dju

sted

R2

031

009

032

005

Firm

xed

effe

cts

Yes

No

Yes

No

Tim

ed

um

mie

sN

oY

esY

esN

o

Dep

end

ent

vari

able

isfr

acti

onof

ann

ualc

omp

ensa

tion

from

stoc

kop

tion

sSt

and

ard

erro

rsin

par

enth

eses

(a)

All

thre

esp

eci

cati

ons

use

tric

hoto

mou

sco

rpor

ate

tax

rate

sT

he

sam

ple

inco

lum

n(3

)u

ses

only

the

year

s19

84to

1988

sin

cew

ear

eu

sing

1986

asa

nat

ural

exp

erim

ent

(b)

Thi

sva

riab

leis

the

dol

lar

chan

gein

CE

Ow

ealt

hp

er$1

000

chan

gein

rm

mar

ketv

alu

e(s

eeJe

nse

nan

dM

urp

hy

1990

)

(c)

Thi

sva

riab

lein

the

dol

lar

char

gein

CE

Ow

ealt

hp

er10

-per

cent

chan

gein

rm

valu

e(s

eeH

alla

nd

Lie

bman

199

8)

(d)

Stan

dar

dd

evia

tion

ofr

mm

arke

tva

lue

intr

illio

ns

32 Hall amp Liebman

TABLE 5Large Institutional Investors and Stock-Option Compensation

The Cross-sectional Relationship between Institutional Investors andOption Grants

(1) (2) (3) (4)

Coefcient on fraction of 217 (067) 145 (014) 177 (069) 120 (015)rmrsquos shares owned byinstitutional investors

Sample period 1994 1980ndash1994 1994 1980ndash1994Contract-theory and stock No No Yes Yes

market covariatesTime dummies No Yes No Yes

The dependent variable is the share of compensation paid in options

factor alone can explain 6 percentage points of the 23-percentage-pointincrease in the median share of compensation paid in options that oc-curred over this time period Of course this variable is only one measureof the strengthening of corporate governance so the total impact of im-proved corporate governance could be greater The second and fourthcolumns of the table use the entire 1980ndash1994 sample and include timedummies to isolate the cross-sectional correlations Again there is evi-dence of a strong cross-sectional relationship between the two variablesalthough the coefcients are slightly smaller perhaps reecting a weakercorporate-governance role in the earlier period In sum there is a strongrelationship in both longitudinal and cross-sectional data between theinstitutional-investor variable and the options-percentage variable13

7 THE MILLION-DOLLAR RULE

In 1993 section 162(m) of the Internal Revenue Code was enactedlimiting the deductibility of executive compensation in excess of onemillion dollars unless the compensation was performance-related Pro-ponents of this legislation argued that it would slow the growth ofexecutive pay and tighten the link between rm performance and execu-tive compensation

The new provision became effective for tax year 1994 and applies tothe CEO and the other four most highly compensated executives in eachcompany Pay that is performance-related such as stock-option grants

13 The one specication in which the fraction owned by institutional investors is notsignicant is in column (2) of Table 4 a regression that includes both xed effects and timeeffects

Taxation of Executive Compensation 33

and bonuses paid for meeting clear performance goals is not affected bythis regulation provided that it has been approved by shareholdersThus the main impact of the provision is to limit the deductibility ofsalary in excess of a million dollars Woodlock and Antenucci (1997)studied the proxy statements of 376 rms and documented that mostrms responded to the new law by qualifying top executive pay asperformance-related Perry and Zenner (1999) identied 25 rms thatreduced salaries from above $1 million to below $1 million and foundthat 23 of the rms cited section 162(m) as their reason for doing so

In this section of the paper we investigate whether this provision hasaffected the level or structure of executive pay using the Execucompdata base which contains information on the highest-paid executives ineach of the rms in the SampP 500 SampP Midcap 400 and SampP SmallCap600 Between 1993 and 1998 the median salary of the Execucomp execu-tives rose from $230000 to $306000 The median value of total compen-sation (salary bonus and option grants) rose from $470000 to $882000Limiting the sample of CEOs indicates that the total compensation of themedian CEO rose from $11 million to $18 million over the period whilethe median salary rose from $441000 to $550000 Thus executive com-pensation clearly continued to rise at a rapid rate after the implementa-tion of the million-dollar rule though perhaps at a slower rate than itotherwise would have

If the regulation affects behavior it should have the largest impact onrms whose executives are receiving salaries that are above the limit orjust below it Thus we might expect to see rms whose executives arereceiving relatively high salaries to rely more heavily on performance-based pay and less heavily on salary in the years following the regulationPerry and Zenner (1999) and Livingston (1999) both nd that this is thecase This trend can be clearly identied in the Execucomp data set Forexample over 1993 to 1998 the median annualized growth rate of salaryfor executives whose 1993 salary was below $500000 in the previous yearwas 85 percent For executives with salary between $500000 and$700000 it was 64 percent for executives between $700000 and $850000it was 50 percent for executives between $750000 and $1 million it was31 percent and for executives above $1 million it was 00 percent14

The observation that executives at lower levels of salary had highergrowth rates is not very strong evidence of an impact of section 162(m)however because it is possible that low-salary executives would havehad higher growth rates even in the absence of the provision To explore

14 Of the 3137 executives for whom we have complete data between 1993 and 1998 2708had 1993 salaries below $500000 while only 35 had 1993 salaries above $1 million

34 Hall amp Liebman

whether this is the case we use Execucomp data from both before andafter the law change Ideally we would want to have data from manyyears before the change to examine whether there is an underlying pat-tern of faster salary growth for low-salary executives Unfortunately theExecucomp data begin in 1992 and therefore provide us with only the1992ndash1993 growth rate in compensation to use as a baseline15

We run regressions pooling data from 1992ndash1993 through 1997ndash1998of the annual percentage change in different forms of compensation on avariable million that is designed to measure the likelihood that anexecutiversquos compensation will be affected by section 162(m) We denemillion as the minimum of 1 and the executiversquos salary divided by $1million in the previous year Thus any executive with salary at or above$1 million would receive a value of 1 while an executive with a salary of$500000 would be coded as a 0516 We include this variable by itself andalso interacted with a dummy variable (after) that equals 1 for years1994 and after Thus the million variable is designed to pick up anyunderlying relationship between the level of salary and the growth rateof salary while the interacted one will identify any differential growthrate of salary by income in the period after section 162(m) took effect Wealso include a full set of year dummies the natural logarithm of thermrsquos market value and the rmrsquos annual rate of return over the twoprevious years as control variables To reduce the sensitivity of our re-sults to outliers we run quantile regressions and robust regressions

Table 6 contains our results The rst four columns are for all topexecutives while the last four columns focus on CEOs Columns (1) and(2) present estimates from robust regression for the percentage change insalary and the percentage change in total compensation (salary bonusstock grants and stock-option grants) In the salary growth regressionsfor all executives the coefcient on million is negative and signicantindicating that there is an underlying relationship between the level ofexecutive compensation and its subsequent growth rate The coefcienton million 3 after is negative as well suggesting that the negativerelationship between the level of salary and its subsequent growth rateintensied slightly after 1994 The relatively small magnitude suggeststhat an executive with a $1 million dollar salary would see his salarygrow at an annual rate that would be about 06 percent slower than anexecutive earning $500000

15 The HallndashLiebman data set contains only a combined salary and bonus variable makingit impossible to use it to analyze this issue16 We also ran the regressions (not reported) allowing for a nonlinear relationship betweenmillion and compensation growth and obtained substantively similar results

TA

BL

E6

The

Eff

ect

ofth

eM

illi

on-D

olla

rR

ule

onSa

lary

and

Tot

alC

ompe

nsat

ion

All

top

exec

uti

ves

CE

Os

only

Rob

ust

regr

essi

onQ

uan

tile

regr

essi

onR

obu

stre

gres

sion

Qu

anti

lere

gres

sion

D

in

Din

D

in

Din

D

in

Din

D

in

Din

sala

ryto

talc

omp

(a)

sala

ryto

talc

omp

sa

lary

tota

lcom

p

sala

ry

tota

lcom

p

mil

lio

n(b

)2

006

02

035

72

006

32

024

12

008

72

023

12

010

32

024

5(0

005

)(0

032

)(0

006

)(0

027

)(0

010

)(0

073

)(0

011

)(0

056

)m

illi

on

32

001

10

026

20

021

20

015

20

016

014

92

003

40

158

aft

er(c

)(0

005

)(0

033

)(0

006

)(0

029

)(0

010

)(0

076

)(0

011

)(0

057

)N

359

5035

950

359

5035

950

757

17

589

757

17

589

Stan

dar

der

rors

inp

aren

thes

esR

egre

ssio

nsal

soin

clu

de

year

du

mm

ies

stoc

km

arke

tre

turn

and

log

mar

ket

valu

e

(a)

Tot

alco

mp

ensa

tion

incl

ud

essa

lary

bon

us

rest

rict

edst

ock

and

stoc

k-op

tion

gran

ts

(b)

mil

lion

isd

ene

das

min

(1s

alar

yin

pre

viou

sye

ars

div

ided

by$1

mil

lion)

(c)

aft

eris

anin

dic

ator

vari

able

equ

alto

zero

befo

reth

ere

gula

tion

took

plac

e(1

993)

and

one

afte

rwar

d

36 Hall amp Liebman

In the second column the coefcient on million 3 after is essentiallyzero suggesting that any decrease in salary brought about by section162(m) was offset by increases in bonus and stock-option grants Theresults are similar for the median regressions There is some evidence ofa very small slowdown in growth of salary for executives likely to beaffected by section 162(m) but no sign of a change in total compensa-tion Taken together this provides evidence of a very small substitutionof performance-related pay for salary

The last four columns run identical tests but limit the sample to CEOsThe interaction of million 3 after continues to be small and negative inthe salary regressions In contrast the impact of 162(m) on total compen-sation appears to be positive and fairly substantial (about 15 percentagepoints per year) in these regressions That is the increase in bonus andstock options following the million-dollar rule more than offset the verysmall decline in salary

All of these conclusions should be interpreted somewhat cautiouslybecause our data provide us with only one pre-1993 control year Never-theless the data are consistent with section 162(m) having led to a veryminor slowdown in salary growth and one that is dramatically smallerthan what one would conclude without controlling for underlying differ-ential salary growth rates at different levels of salary Since total compen-sation for CEOs did not decline and perhaps increased there is evidenceof a minor substitution of performance-related pay for salary in responseto the regulation

8 EXECUTIVE COMPENSATION AND THEELASTICITY OF TAXABLE INCOME

Even if taxes have only a minor effect on executive compensation thestructure of executive compensation is important for tax policy be-cause high-income taxpayers are sources of signicant revenue and arethe focus of important debates about the efciency cost of taxation

Goolsbee (1999) has recently argued that executive decisions to exer-cise options are highly responsive to intertemporal differences in taxrates created by tax reforms He claims that much of the apparent declinein taxable income for high-income taxpayers between 1992 and 1993documented by Feldstein and Feenberg (1996) can be explained by execu-tives shifting option gains into 1992 in order to avoid the higher marginaltax rates that went into effect in 1993 By extension he implies that thehigh elasticities of taxable income with respect to personal tax ratesestimated by comparing the pre- and post-tax-reform taxable income of

Taxation of Executive Compensation 37

high-income individuals reect intertemporal shifting of income and notpermanent effects of taxation If this is true then the deadweight loss oftaxation is lower as is the cost of progressivity

While Goolsbee presents comprehensive evidence that option gainswere unusually high in 1992 his identication of the impact of tax rates onthe timing of option gains depends on correctly specifying a counter-factual level of options that would have been exercised in the absence ofthe tax change He simply assumes a linear time trend The purpose ofGoolsbeersquos time trend is presumably to account for the increasing use ofoptions over time However since options typically require a vestingperiod of at least a couple of years a much more direct measure of ex-pected option gains can be constructed based on the total value of vestedoptions held by the CEO Since options typically have vesting periods thatexceed two years this total can be treated as exogenous since it will not beaffected by the tax changes

More generally the value of options held by the CEO is likely to havea major effect on his exercise decisions If his options are out of themoney he will not have gains to realize If there has been a recent run-up in the market an optimizing CEO may exercise an unusually largeamount of options in order to diversify his overall portfolio and if in-the-money options are about to expire they will be exercised in thecurrent year

Our evidence suggests that much of the apparent tax shifting caninstead be attributed to stock market performance and the timing of pastoption grants Using our 15-year panel we replicate Goolsbeersquos (1999)result for the 1993 period but show that the impact of taxes is notpresent in other time periods Moreover even the result for the 1993period disappears when appropriate controls for past option grants andstock price appreciation are included

The top three rows of Table 7 present data on the taxable incomes ofCEOs in the HallndashLiebman sample For these results our panel has beennarrowed to include only CEOs whose rmrsquos scal years correspondwith tax years In addition we create a balanced panel for each of thethree tax-reform periods by excluding any rm for which there are notcomplete data for the period surrounding the reform

The fourth row of the table shows that the 1993 tax act raised the topmarginal personal income tax rate from 310 to 396 percent effective fortax year 1993 with a further increase occurring in 1994 due to the uncap-ping of the Medicare payroll tax The top row of the table shows thattaxable income was higher in 1992 than for any other year in our sample(all data are in 1994 dollars) Option gains were particularly high in1992mdashmore than double their 1991 level and 68 percent higher than their

TA

BL

E7

Tax

Ref

orm

Sto

ckM

arke

tP

erfo

rman

cea

ndth

eT

imin

gof

Opt

ion

Gai

ns

Ave

rage

stoc

km

arke

tM

ean

CE

OM

ean

CE

Ore

turn

for

Mea

nva

lue

taxa

ble

sala

ryan

dM

ean

CE

OT

opfe

der

al

rms

inof

opti

ons

(Val

ue

ofop

tion

Sam

ple

inco

me

bonu

sop

tion

gain

sp

erso

nal

sam

ple

for

held

atbe

gga

ins)

(va

lue

Ref

orm

size

Yea

r($

mil

)($

mil

)($

mil

)M

TR

pre

vye

ar(

)of

year

ofop

tion

she

ld)

ER

TA

8127

919

800

874

064

80

163

7025

80

460

354

1981

083

10

672

009

869

245

070

61

3919

820

872

067

90

099

5011

90

686

144

1983

099

00

724

025

250

355

116

62

16

TR

A86

295

1985

119

70

837

024

050

76

116

02

0719

861

448

091

70

345

5042

81

776

194

1987

165

50

959

059

338

519

12

229

266

1988

163

21

058

039

028

07

237

71

6419

891

656

105

20

414

2817

92

345

176

1993

272

1990

209

51

028

061

828

232

288

02

15T

axA

ct19

911

875

102

00

547

312

01

268

82

0319

922

678

107

01

250

3146

64

552

275

1993

232

41

209

074

339

620

64

299

173

1994

196

41

262

037

842

515

45

113

074

Sam

ple

isH

allndash

Lie

bman

rm

sw

ith

Dec

embe

r

scal

year

sth

atha

dco

mp

lete

dat

afo

ron

eof

the

thre

eta

x-re

form

per

iod

sst

ud

ied

Taxation of Executive Compensation 39

1994 level These patterns are consistent with Goolsbeersquos claim that tax-payers responded to the anticipated increase in marginal tax rates byshifting gains into 1992

The evidence from the other two tax reforms is less clear suggestingthat there might be factors other than tax rates determining the timing ofoption gains ERTA 1981 reduced the top marginal tax rate from 70percent in 1980 to 50 percent in 1982 However there is no sign thatoption gains were reduced in 1980 and 1981 and then increased in 198217

There was a big increase in option gains in 1983 however Similarly thehighest year of stock-option gains in the period around 1986 was 1987 astrange year to take gains since taxpayers would have known that thepersonal rate would be 10 percentage points lower if they waited untilthe following year18

What all the peak years of option gains have in common is that theyfollow years of strong stock market performance The fth row of Table 7shows the average stock market return during the previous year for therms in our sample The large value of option gains in 1983 followed a355-percent increase in the stock market The 1992 boom in option gainsfollowed a 466-percent increase in the stock market the more modest1987 peak in option gains followed a 19-percent gain (and a 43-percentgain two years before)

Strong stock market performance raises the value of options availableto exercise The sixth row of the table shows the mean value of optionsheld by CEOs at the beginning of the year All three of the peak years ofoption gains correspond with peak years in terms of the value of optionsavailable for exercise Since stock market appreciation during the yearalso increases the value of options available to exercise later in that yearwe need to allow for current-year stock market appreciation as well

Table 8 contains Goolsbee-style regressions for each of the tax re-forms The rst two columns reprint the results from Goolsbee (1999)The third and fourth columns replicate his results in our sample Inparticular with a specication similar to his we are able to replicate hisnding that the current after-tax share has a positive coefcient and thefollowing yearrsquos after-tax share has a coefcient of roughly negative oneColumns (5) through (8) show that this pattern was unique to the 1993tax reform Indeed in the TRA 86 regressions the coefcient on the

17 As we explained above for executives covered by the maximum tax on earned incomethere was not a decline in tax rates between 1981 and 198218 It is possible that the large stock market decline in October 1997 may have promptedsome people to exercise options earlier than they otherwise would have However anincrease in option exercises typically follows stock market increases not declines

TA

BL

E8

Att

empt

sto

Rep

lica

teth

eT

ax-S

hift

ing

Res

ult

inO

ther

Tim

eP

erio

ds

Goo

lsbe

eG

ools

bee

1993

1993

Ent

ire

Ent

ire

regr

essi

onre

gres

sion

refo

rmre

form

TR

A86

TR

A86

ER

TA

81E

RT

A81

1980

ndash199

419

80ndash1

994

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

ln[1

2ta

x(t)

]1

244

108

20

455

032

62

017

62

017

22

016

72

016

70

026

20

068

(01

34)

(01

27)

(05

37)

(05

42)

(02

10)

(02

09)

(00

90)

(00

91)

(00

67)

(00

65)

ln[1

2ta

x(t1

1)]

20

867

20

764

21

214

20

917

047

80

722

20

103

004

30

176

018

3(0

118

)(0

112

)(0

491

)(0

499

)(0

157

)(0

174

)(0

072

)(0

072

)(0

068

)(0

066

)

Tim

e0

135

007

62

000

52

036

40

041

000

80

094

003

50

056

002

7(0

008

)(0

008

)(0

049

)(0

049

)(0

028

)(0

028

)(0

025

)(0

026

)(0

003

)(0

003

)

ln(m

arke

tva

lue)

066

10

489

023

70

334

039

5(0

015

)(0

058

)(0

039

)(0

042

)(0

017

)

Yea

rs19

91ndash1

995

1991

ndash199

519

90ndash1

994

1990

ndash199

419

85ndash1

989

1985

ndash198

919

80ndash1

983

1980

ndash198

319

80ndash1

994

1980

ndash199

4

Ad

just

edR

20

720

770

610

630

690

760

800

810

630

68

Obs

erva

tion

s13

856

132

371

385

136

41

470

143

31

113

106

23

968

385

9

Dep

end

ent

vari

able

isth

elo

gof

taxa

ble

inco

me

All

regr

essi

ons

con

tain

rm

xe

def

fect

sG

ools

bee

regr

essi

on(2

)al

soco

ntai

nstw

oot

her

vari

able

sno

tsh

own

inth

eta

ble

Stan

dar

der

rors

inp

aren

thes

es

Taxation of Executive Compensation 41

TABLE 9Alternative Explanations for the Timing of Option Exercises

1993 Entirereform TRA 86 ERTA 81 1980ndash1994

(1) (2) (3) (4)

ln[12tax(t)] 0108 0111 20051 0201(0324) (0136) (0048) (0057)

ln[12tax(t11)] 20134 0573 0125 0150(0336) (0156) (0053) (0066)

ln(market value) 0418 0177 0338 0418(0061) (0039) (0046) (0163)

Firmrsquos stock market return (t) 0298 0268 0214 0258(0044) (0045) (0031) (0023)

Firmrsquos stock market return 0162 0206 0031 0128(t21) (0041) (0008) (0029) (0021)

ln(value of options held by 0004 0008 0001 0009CEO at start of year) (0003) (0003) (0002) (0001)

Years 1990ndash1994 1985ndash1989 1980ndash1983 1980ndash1994Adjusted R2 064 071 082 068Observations 1349 1427 1036 3812

The dependent variable is the log of taxable income All regressions contain rm xed effects Stan-dard errors in parentheses

following yearrsquos tax rate is positive and signicant Columns (9) and (10)pool all three reforms and fail to replicate the pattern that would reectintertemporal shifting19

It is possible that the ease with which one can shift the timing ofoption gains is asymmetric It may be harder to shift option gains for-ward in time since some options expire and therefore must be exercisednow In addition options were a much smaller portion of compensationin the early 1980s so perhaps the ERTA 81 results are not surprisingNonetheless these results cast doubt on the proposition that this is thechannel through which spurious estimates of the elasticity of taxableincome have occurred for previous tax reforms and they suggest thatwe should look harder for alternative explanations for the timing ofoption gains

Table 9 attempts to predict options gains using the lagged and currentincrease in a rmrsquos market value and the value of the options held by theCEO at the start of the year The two stock market variables stronglypredict option gains while the value of options held has the correct sign

19 The results also fail to suggest large permanent effects of marginal tax rates on taxableincome

42 Hall amp Liebman

in all of the specications and is signicant in the pooled sample eventhough it is highly correlated with the two stock market return variablesIn contrast the tax-timing explanation fails even in the 1993 period

It is important to realize that our results do not invalidate Goolsbeersquoscritique of claims that aggregate revenue data from 1992 and 1993 provethat elasticities of taxable income are large We have simply given analternative explanation for why taxable incomes were unusually high in1992 However our results do cast doubt on the conclusion that largetiming shifts are ubiquitous and capable of explaining away the morecareful estimates of taxable income elasticities that have been performedusing micro data (eg Feldstein 1995 Auten and Carroll 1995)

One further point is in order It is sometimes argued that the combina-tion of increased use of options and the booming stock market explainswhy federal tax revenues have been so much higher than expected dur-ing the 1990s However while large option gains increase revenue fromthe personal income tax they produce offsetting deductions for cor-porations The net impact on revenue is only the difference in the twomarginal tax rates times the option gain While the gap between thepersonal rate and the effective corporate rate is nontrivial the impact oflarge option exercises on total revenue is substantially smaller than theamount by which the exercises inate personal tax revenue and reducecorporate tax revenue

9 CONCLUSION

We have described the tax rules for executive pay in detail and analyzedhow changes in various tax rates affect the tax advantages of stock optionsrelative to salary and bonus We nd that there is a moderate tax advan-tage to optionsmdashcurrently about $4 per $100 in pretax executive paymdashand that changes in the tax advantage of options over time have had atmost a modest impact on the composition of pay Corporate governancefactors particularly the role of large investors appear to be more impor-tant in explaining the dramatic increase in option pay

In addition our evidence suggests that more direct attempts to use taxpolicy to inuence executive compensation have had little effect Wend that the million-dollar rule led companies to substitute perfor-mance-based pay for salary But our evidence suggests that this substitu-tion was quite modest and there is no evidence that the total level of paywas reduced Overall although the stock-option explosion has dramati-cally increased the link between pay and performance this change isdue almost entirely to nontax factors

Taxation of Executive Compensation 43

REFERENCES

Auerbach Alan J (1996) ldquoThe Dynamic Effects of Tax Law Asymmetriesrdquo Re-view of Economic Studies 53451ndash405

mdashmdashmdash and James Poterba (1987) ldquoTax Loss Carryforwards and Corporate TaxIncentivesrdquo In The Effects of Taxation on Capital Accumulation M Feldstein(ed) Chicago University of Chicago Press

Altshuler Rosanne and Alan Auerbach (1990) ldquoThe Signicance of Tax LawAsymmetries An Empirical Investigationrdquo Quarterly Journal of Economics 10561ndash86

Auten Gerald and Robert Carroll (1995) ldquoThe Effect of Income Taxes on House-hold Behaviorrdquo Ofce of Tax Analysis US Department of the Treasury

Baker George P and Brian J Hall (1998) ldquoCEO Incentives and Firm SizerdquoNBER Working Paper no 6868 December

Core John and Wayne Guay (1999) ldquoThe Use of Equity Grants to ManageOptimal Equity Incentive Levelsrdquo Draft

Feldstein M (1995) ldquoThe Effect of Marginal Tax Rates on Taxable Income APanel Study of the 1986 Tax Reform Actrdquo Journal of Political Economy 103 (no3)551ndash72

mdashmdashmdash and D Feenberg (1996) ldquoThe Effect of Increased Tax Rates on TaxableIncome and Economic Efciency A Preliminary Analysis of the 1993 Tax RateIncreasesrdquo In Tax Policy and the Economy vol 10 James M Poterba (ed)Cambridge MA MIT Press

Gompers Paul A and Andrew Metrick (1998) ldquoInstitutional Investors andEquity Pricesrdquo NBER Working Paper no 6723

Goolsbee Austan (1999) ldquoWhat Happens When You Tax the Rich Evidencefrom Executive Compensationrdquo NBER Working Paper no 6333

Graham John R (1996) ldquoProxies for the Corporate Marginal Tax Raterdquo Journal ofFinancial Economics 42187ndash221

Hall Brian J (1999) ldquoThe Design of Multi-year Stock Option Plansrdquo Journal ofApplied Corporate Finance 12(2)97ndash106

mdashmdashmdash and Jeffrey B Liebman (1998) ldquoAre CEOs Really Paid Like Bureau-cratsrdquo Quarterly Journal of Economics CXII(no 3)653ndash691

Jensen Michael C and William M Meckling (1976) ldquoTheory of the Firm Mana-gerial Behavior Agency Costs and Ownership Structurerdquo Journal of FinancialEconomics III305ndash360

mdashmdashmdash and Kevin J Murphy (1990) ldquoPerformance Pay and Top-ManagementIncentivesrdquo Journal of Political Economy XCVIII225ndash264

Kaplan Steven N (1997) ldquoThe Evolution of US Corporate Governance WeAre All Henry Kravis Nowrdquo Draft

Livingstone Jane R (1999) ldquoExecutive Compensation Contracting and Re-sponses to Increased Tax Costsrdquo Louisiana State University June

Miller Merton H and Myron S Scholes (1982) ldquoExecutive CompensationTaxes and Incentivesrdquo In Financial Economics Essays in Honor of Paul CootnerWilliam F Sharpe and Cathryn M Cootner (eds) Princeton NJ Prentice-Hall

Murphy Kevin J (1999) ldquoExecutive Compensationrdquo forthcoming in OrleyAshenfelter and David Card (eds) Handbook of Labor Economics Vol 3 NorthHolland

Perry Todd and Marc Zenner (1999) ldquoPay for Performance Government Regu-

4 Hall amp Liebman

the tax advantage of options has changed over time In section 6 weprovide evidence on the effect of taxation on the composition of execu-tive compensation Section 7 contains empirical analysis of the million-dollar rule Section 8 contains evidence on tax shifting and option gainsSection 9 concludes

2 TRENDS IN TOP EXECUTIVE PAY

In this section we document how top executive pay has changed overtime and discuss how this change has caused the sensitivity of CEOwealth to rm market value to increase substantially There has been alarge increase in the level of CEO pay since 1980 and this growth hasbeen driven by the dramatic increase in stock-option grants during thistime (Hall and Liebman 1998) Although salary and bonuses nearlydoubled over the period in ination-adjusted terms the mean value ofstock-option grants increased by 683 percent The percentage increase inthe median stock-option award cannot be calculated because the me-dian stock-option grant was zero in 1980 The median CEO did notreceive an annual stock-option grant until 1985 Today nearly all topexecutives of large companies receive stock options and the averagestock-option grant is now larger for most top executives than salary andbonus combined

The (ination-adjusted) growth rate of CEO pay since 1980mdashmeasurednarrowly (cash pay grew at an annual rate of 5 percent per year) orbroadly(cash plus option grants grew at almost 9 percent per year) or very broadly(total compensation including stock and stock-option appreciation grewat 115 percent per year)mdashhas been large relative to virtually all othergroups Indeed the growth rate of CEO pay since 1980 has been higheven relative to the pay increases of other high-income earners For ex-ample the cutoff point for being in the top 05 percent of adjusted grossincome (AGI) increased by about 37 percent per year about half the ratefor direct CEO compensation (excluding stock and option appreciation)The only workers who appear to have had faster compensation growththan CEOs are other ldquosuperstarsrdquo The annual pay of professional base-ball players increased by approximately 98 percent per year and that ofprofessional basketball players by 139 percent per year

The increase in stock options has led to a large increase in the equityholdings of top executives and this in turn has led to a dramatic increasein the responsiveness of executive wealth to rm performance Nearlyall of the pay-to-performance sensitivity of executive compensationcomes from equity holdings for a given increase in shareholder valuechanges in the value of an executiversquos stock and stock options are more

Taxation of Executive Compensation 5

than 50 times larger than changes in salary and bonus (Hall and Lieb-man 1998)

As a concrete example the estimates in Hall and Liebman (1998) implythat a 10-percent increase in rm value (of the median company in oursample) leads the company to increase the CEOrsquos salary and bonus byabout $25000 However this same 10-percent increase in shareholdervalue translates into $125-million increase in the value of the CEOrsquosstock and stock-option holdings1

The dramatic rise in the link between CEO wealth and rm perfor-mance can be seen in Figure 1 which shows how two measures of thislink have increased since 19802 The rst measure the JensenndashMurphy(1990) sharing rate (shown on the left scale) is the change in CEO wealthfor a $1000 change in rm value The second measure is the change inCEO wealth for a 10-percent change in rm value (see Baker and Hall1998) Since both measures are strongly affected by rm size (the formerhas a negative correlation and the latter has a positive correlation) thepay-to-performance changes over time are estimated with regression(quantile) analysis that controls for changes in the distribution of rmsizes in the sample over time The gure therefore shows the increasein the pay-to-performance measures over time for a constant-size rmin this case a $1-billion rm (in constant 1998 dollars) The striking fact isthat both measures of the pay-to-performance link have increased bynearly a factor of 10 since 1980 These pay-to-performance increases areeven larger than those we reported in our earlier paper that analyzed theperiod 1980 to 1994 because of the large increase in stock-option grantscombined with the strong stock market performance in 1994 to 1998

3 TAX AND ACCOUNTING RULES

Stock options give an executive the right but not the obligation to buy ashare of the companyrsquos stock at a prespecied pricemdashthe exercise or strikeprice Typically options cannot be exercised immediately That is theyvest (become owned by the executive who can then exercise if he or she

1 Stock-option grants are also very sensitive to changes in rm performance mostly be-cause many grants are multiyear plans that hold the yearly number of options constantand the same number of at-the-money options are worth more when the stock price ishigher and vice versa If stock-option grant sensitivity is also included then about 91percent of pay-to-performance sensitivity comes from stock and stock-option revaluations7 percent comes from stock-option grant changes and less than 2 percent comes fromchanges in salary and bonus (Hall 1999)2 Both measures include only the link created by CEO holdings of stock and stock optionsand ignore the smaller amount of pay-to-performance sensitivity that operates throughchanges in salary bonus and stock-option grants

FIG

UR

E1

Pay

toP

erfo

rman

ceT

heL

ink

betw

een

CE

OW

ealt

han

dSh

areh

olde

rV

alue

Sinc

e19

80

Not

eT

he

anal

ysis

cont

rols

for

the

chan

ging

com

pos

itio

nof

rm

size

over

tim

eT

hu

sth

ese

esti

mat

essh

owth

ep

ay-t

o-p

erfo

rman

celin

kfo

ra

sim

ilar

-siz

edco

mp

any

over

tim

ein

this

case

aco

mp

any

wit

ha

mar

ket

valu

eof

$1bi

llion

Taxation of Executive Compensation 7

wishes) slowly over time Common vesting periods are in the three- tove-year range and options usually vest linearly (eg a four-year op-tion vests at 25 percent at the end of each year) An executive typicallyloses any unvested options upon departure Although options may beexercised as soon as they vest they do not have to be exercised untilthey expire or mature Almost 85 percent of stock-option plans have aterm of exactly ten years with virtually all of the remainder being in theve- to ten-year range About 95 percent of options are granted at themoney or at fair market value which means that the exercise price at grantdate is set equal to the stock price at grant date The remaining 5 percentare either discount options (so-called in-the-money options where theexercise price is below the stock price at grant date) or premium options(so-called out-of-the-money options where the exercise price is abovethe stock price at grant date) The holders of options typically do nothave dividend rights or voting rights even on vested options3

31 Tax Rules and Stock-Option CompensationUnlike salary and bonus stock-option grants are typically an untaxedevent at the time of grant For the most widely used optionsmdashnonquali-ed stock options (NQSOs)mdashexecutives are taxed at the personal incometax rate on option prots (the difference between that stock price and theexercise price times the number of options) when the options are exer-cised The company receives a parallel deduction against corporate in-come at that point If the executive continues to hold the shares afterexercise any subsequent appreciation is taxed at the capital gains rate inthe usual way In 1993 an additional feature was added to the tax code[Internal Revenue Code section 162(M)] that disallowed a corporate de-duction for any executive pay above $1 million that is not performance-based While this rule affects executive salaries most bonuses qualify asperformance-based and standard stock options automatically qualifyTherefore this provision gives companies with highly paid executives anincentive to give more pay in the form of bonuses and stock options asubject we return to in section 7 A summary description of the tax (andaccounting) treatment of cash and option compensation is in Table 1

A far less common type of option which is estimated to account forabout 5 percent of option grants is the incentive stock option (ISO) WhileISOs are similar to NQSOs in their design they are crucially different intwo respects First they have an annual cap of $100000 per executive

3 See Murphy (1999) for details about stock options and Miller and Scholes (1982) on taxincentives

TA

BL

E1

Tax

and

Acc

ount

ing

Tre

atm

ent

ofE

xecu

tive

Com

pens

atio

nC

ash

Pay

men

tsve

rsus

Stan

dard

Exe

cuti

veO

ptio

ns

Tax

atio

nat

Tax

atio

nat

Acc

ount

ing

Com

pens

atio

nD

escr

ipti

onp

erso

nall

evel

corp

orat

ele

vel

trea

tmen

t

Sala

ryan

dbo

nu

sC

ash

pay

men

tsm

ade

Tax

edas

ord

inar

yFu

llyd

edu

ctib

leS

ince

Exp

ense

dag

ains

tea

rn-

toC

EO

in

com

e19

93a

ny

pay

over

$1in

gsin

the

usu

alw

ay

mill

ion

isno

td

educ

t-ib

leu

nles

sit

isp

er-

form

ance

-bas

ed

Non

-qua

lie

dst

ock

Op

tion

sgr

ant

the

ex-

The

gran

ting

ofan

op-

Apa

ralle

ltax

ded

uc-

The

reis

no

exp

ense

rec-

opti

ons

(NQ

SOs)

ecut

ive

the

righ

tbu

tti

onis

typ

ical

lyan

un-

tion

atth

eco

rpor

ate

ogni

tion

(atg

ran

tex

er-

not

the

obli

gati

ont

ota

xed

even

tU

pon

exer

-le

veli

sge

nera

llyal

-ci

seo

rsa

le)

for

op-

buy

shar

esof

stoc

kat

cise

opt

ion

pro

ts

are

low

edu

pon

exer

cise

tion

sif

as

isu

sual

lyan

exer

cise

pric

e(t

ypi-

taxe

dat

the

ord

inar

y(f

orth

eam

ount

ofop

-th

eca

set

he

opti

ons

cally

the

curr

ent

stoc

kra

teI

fth

eex

ecu

tive

tion

pro

ts

)O

pti

ons

are

not

dis

cou

nted

(in

pri

ceat

the

dat

eof

cont

inu

esto

hold

onto

auto

mat

ical

lyqu

alif

yas

the

mon

eyat

gran

tgr

ant)

Th

eop

tion

sth

esh

ares

the

exec

u-

perf

orm

ance

-bas

edp

ayd

ate)

and

the

exer

cise

typ

ical

lyve

stov

era

3ndashti

veis

taxe

din

the

and

ther

efor

ear

eno

tpr

ice

and

num

ber

ofop

-5-

year

per

iod

The

typ

i-us

ualw

ay(a

tth

eca

pi-

subj

ect

toth

e$1

-ti

ons

iskn

own

atgr

ant

calm

atu

rity

is10

year

sta

lgai

nsra

te)

onan

ym

illio

nca

p

dat

eFo

rd

isco

unt

op-

but

the

man

ager

may

furt

her

app

reci

atio

nof

tion

sth

ed

iffe

renc

ebe

-ex

erci

seea

rly

Som

eth

esh

are

pri

ce

twee

nth

est

ock

pri

ceop

tion

sar

edi

scou

ntop

-an

dth

em

arke

tp

rice

isti

ons

(the

exer

cise

pri

ceex

pen

sed

over

the

vest

-is

low

erth

anth

est

ock

ing

per

iod

For

opti

ons

pri

ceat

the

gran

td

ate)

w

ith

vari

able

term

sP

rem

ium

opti

ons

are

(eg

a

vari

able

exer

-th

ere

vers

e(t

heex

er-

cise

pri

cev

esti

ngth

atci

sep

rice

isgr

eate

ris

tied

top

erfo

rman

ce)

than

the

stoc

kp

rice

atth

eop

tion

sar

em

arke

dgr

ant

dat

e)

tom

arke

tan

dex

-p

ense

dd

uri

ngth

eti

me

betw

een

gran

tan

dex

erci

se

Not

eW

eth

ank

com

pen

sati

onco

nsu

ltan

tsSc

ott

Gre

enbe

rgan

dSc

ott

Ols

en(b

oth

ofT

ower

sP

erri

n)

and

Fred

eric

Coo

k(o

fFr

eder

icW

C

ook

and

Ass

ocia

tes)

for

help

fulc

onve

rsat

ion

sin

und

erst

and

ing

the

tax

and

acco

unt

ing

rule

s

10 Hall amp Liebman

Second the tax treatment of ISOs is different ISOs are completely un-taxed at grant or exercise It is only at sale that the executive is taxed andeven then the executive is taxed at the lower capital gains tax rate Thedisadvantage is that the corporation never gets to take a parallel taxdeduction against corporate prots Thus ISOs become more attractiveas the personal tax rate increases and as the corporate tax rate and thecapital gains tax rates fall

Two other related types of compensation restricted stock and stockappreciation rights are worth brief discussion Both however are farless common than standard stock options Restricted stock is payment inthe form of restricted shares the restriction being that the shares vestover time as with options Unlike options the shares typically havevoting and dividend rights With regard to taxation the executive istaxed at the personal rate on the value of the restricted stock as thevesting restrictions lapse However the executive may choose to betaxed at the grant date in which case all subsequent appreciation istaxed at the capital gains rate The dividends paid to the executive aretaxed at the ordinary rate in the usual way The company generallyreceives a parallel deduction equal to the amount of the executiversquosincome when the executive is taxed Unlike stock options restrictedstock is not generally considered to be performance-based and is there-fore subject to the million-dollar rule (unless the vesting of the restrictedstock is performance-based which is sometimes the case)

Stock appreciation rights (SARs) are rights that replicate the payoffs ofstock options with a cash transfer Thus SARs are simpler than options inthat there is no requirement to buy and resell the stock in order to ldquocashoutrdquo SARs generally have the same tax treatment as NQSOs both to theindividual and to the corporation SARs are relatively rare however be-cause they have disadvantageous accounting treatment (described in thenext section) relative to stock options and their main relative advantagevis-a-vis options has been essentially replicated through broker-assistedcashless option exercisesmdashwhereby a broker makes a ldquonanosecondrdquo loanto the executive (to purchase and resell the stock) when the executivewants to ldquocash inrdquo on option prots The tax and accounting treatment ofISOs restricted stock and SARs is summarized in Table 2

32 The Accounting Treatment of OptionsUnlike cash compensation which is expensed against earnings there isgenerally no expense recognition (at grant exercise or sale) for optionswhether they be NQSOs or ISOs As a result compensation consultantsoften point out that stock options are the only form of compensation thatare free in an accounting sense but still deductible for tax purposes

Taxation of Executive Compensation 11

Stock options do however lead to expense recognition if they are dis-counted (in the money at grant date) or if the exercise price and numberof options are not known at grant date For discount options the differ-ence between the stock price and the market price is expensed over thevesting period For options with variable terms (eg a variable exerciseprice vesting that is tied to performance) the options are marked tomarket and expensed during the time between grant and exercise Practi-tioners claim that the accounting treatment of options plays an impor-tant role in the design of option programs Thus plans that have ldquobadaccountingrdquo but are thought by many to have attractive incentive fea-tures are often not even seriously considered by companies Examples ofsuch potentially attractive plans include indexed options (where a CEOprots only if his rmrsquos share price grows relative to some market orindustry benchmark) and option grants that are explicitly performance-related both of which would lead to expenses against earnings

Unlike options restricted stock and SARs do not generally receivefavorable accounting treatment Restricted stock is generally expensedover the period in which the restrictions lapse (usually the vesting pe-riod) The magnitude of the expense is the difference between the cur-rent stock price and the executiversquos cost (if any) SARs are marked tomarket each period and the difference between the stock price and theexercise price is expensed over the outstanding period of SARs

4 THE TAXATION OF EXECUTIVE PAY CASHVERSUS OPTIONS

Stock-based compensation performs two roles in executive compensa-tion arrangements First it helps align the incentives of the executivewith the interests of the rmrsquos shareholders Second it often enables therm to compensate the CEO in a way that is more advantageous from atax standpoint than paying the executive in salary and bonus

41 Agency Theory and Executive Compensation in the Presenceof TaxationIn standard agency-theory models ( Jensen and Meckling 1976) agencycosts are the result of the separation of ownership and control Managersdo not have the same incentives as the owners The optimal incentivecontract for managers balances the benets of high-powered incentives(linking the fortunes of owners and managers through stock and stockoptions for example) with the costs of loading too much risk on risk-averse managers

TA

BL

E2

Tax

and

Acc

ount

ing

Tre

atm

ent

ofE

xecu

tive

Com

pens

atio

nIn

cent

ive

Stoc

kO

ptio

ns(I

SOs)

Res

tric

ted

Stoc

kan

dSt

ock

App

reci

atio

nR

ight

s(S

AR

s)

Tax

atio

nat

Tax

atio

nat

Acc

ount

ing

Com

pen

sati

onD

escr

ipti

onp

erso

nall

evel

corp

orat

ele

vel

trea

tmen

t

Ince

nti

vest

ock

ISO

sar

eid

enti

calt

oIf

the

usu

alco

ndit

ions

Ifth

eu

sual

cond

itio

nsA

sw

ith

NQ

SOs

ther

eop

tion

sN

QSO

sin

thei

rd

esig

n

are

met

th

ere

isno

tax

are

met

the

reis

gene

r-is

gene

rally

noex

pen

seH

owev

ert

hey

have

anat

gran

tor

exer

cise

At

ally

no

corp

orat

eta

xd

e-re

cogn

itio

nfo

rIS

Os

annu

alca

pof

$100

000

sale

all

opti

onp

ro

tsd

uct

ion

atgr

ant

exer

-(e

ver)

on

the

amou

ntth

atan

dsu

bseq

uen

tca

pita

lci

seo

rsa

le

vest

sT

he

tax

trea

t-ap

prec

iati

onar

eta

xed

men

tof

ISO

sal

soat

the

cap

ital

gain

sd

iffe

rs

rate

Res

tric

ted

stoc

kSi

mila

rto

stoc

kop

-T

hein

div

idu

alis

taxe

dT

heco

mp

any

gene

rall

yR

estr

icte

dst

ock

isge

n-ti

ons

but

shar

esat

the

ord

inar

yra

teas

rece

ives

ap

aral

leld

e-er

ally

exp

ense

dov

er(w

hich

vest

slow

ly)

the

rest

rict

ions

lap

se

du

ctio

neq

ual

toth

eth

eve

stin

gp

erio

dT

hera

ther

than

opti

ons

are

How

ever

the

man

ager

amou

ntof

the

exec

u-

amou

nt

exp

ense

dis

gran

ted

toth

eex

ecu

-m

aych

ose

tobe

taxe

dti

versquos

inco

me

whe

nth

eth

ed

iffe

renc

ebe

twee

nti

veU

nlik

eop

tion

sat

gran

td

ate

inw

hich

exec

uti

veis

taxe

dR

e-th

ecu

rren

tst

ock

pric

eth

esh

ares

typi

cally

case

alls

ubs

equ

ent

ap-

stri

cted

stoc

kis

gen

er-

and

the

exec

uti

versquos

have

voti

ngan

dd

ivi-

pre

ciat

ion

ista

xed

atal

lysu

bjec

tto

the

cost

(if

any)

d

end

righ

ts

the

cap

ital

gain

sra

te

mill

ion-

dol

lar

cap

D

ivid

end

sar

eta

xed

atth

eor

din

ary

rate

Stoc

kap

pre

ciat

ion

Rig

hts

that

repl

icat

eG

ener

ally

have

the

Gen

eral

lyha

veth

eU

nlik

eop

tion

sSA

Rs

righ

tsth

epr

ots

ofst

ock

op-

sam

eta

xtr

eatm

ent

assa

me

tax

trea

tmen

tas

are

mar

ked

tom

arke

tti

ons

Rar

ely

use

d

NQ

SOs

NQ

SOs

each

per

iod

an

dth

esi

nce

stoc

kop

tion

sd

iffe

renc

ebe

twee

nth

eha

vem

ore

favo

rabl

eac

-st

ock

pri

cean

dth

eex

-co

unt

ing

trea

tmen

ter

cise

pri

ceis

exp

ense

d(a

ndop

tion

sp

rovi

de

over

the

outs

tand

ing

the

sam

ebe

ne

tof

per

iod

ofSA

Rs

cash

less

exer

cise

thro

ugh

brok

erlo

ans

)

Not

eW

eth

ank

com

pen

sati

onco

nsul

tant

sSc

ott

Gre

enbe

rgan

dSc

ott

Ols

en(b

oth

ofT

ower

sP

erri

n)an

dFr

eder

icC

ook

(of

Fred

eric

WC

ook

and

Ass

ocia

tes)

for

hel

pfu

lcon

vers

atio

nsin

und

erst

and

ing

the

tax

and

acco

unt

ing

rule

s

14 Hall amp Liebman

The effect of taxation on the optimal contract (more precisely theshare of compensation that is performance-related) is ambiguous (evenin the absence of deductibility and deferral) because there are offsettingeffects First by reducing the share of corporate prots received byshareholders taxes diminish the importance to the shareholders of moti-vating the CEO Taken alone this effect would imply that taxes wouldbe expected to reduce the use of performance-based compensation Sec-ond because the government shares in the income received by the CEOthe variance of the CEOrsquos income is reduced raising his utility andlowering the cost to the rm of providing any given set of incentivesThird because taxes will lead executives to provide less effort for anygiven level of incentive-based pay the level of compensation that mustbe provided to compensate the CEO for effort is reduced for a givenamount of incentive These last two factors make it cheaper for the rmto offer contracts to the CEO with large incentive components when taxrates are higher and should therefore increase the use of incentive-based pay Since the net effect of taxation on the level of incentive-basedpay is ambiguous agency theory provides no strong predictions abouthow taxation should affect the optimal composition of option versuscash compensation

42 The Tax Advantages of DeferralWhile agency theory yields ambiguous predictions about the effect oftaxes on the use of options there are direct tax advantages of options inthat options provide executives with a way to defer compensation andthereby lower their taxes However because options also lead to a defer-ral of corporate tax deductibility the tax advantages from a combined(executive and corporate) perspective are not so straightforward In thissection we analyze and measure the tax advantages of stock optionsrelative to salary and bonus compensation We also analyze the taxadvantages of NQSOs relative to ISOs In particular we show how thetax advantages of options change as personal corporate and capitalgains taxes change We then show how the tax advantages of optionshave changed over time in response to changing tax rates

The crucial tax difference between standard options (NQSOs) andcash payment is that option payouts are deferred and the two forms ofcompensation earn different rates of return over the deferral period Anyanalysis of the relative tax advantages of two compensation instrumentsmust consider the tax consequences both to the employer and to theemployeemdashwhat Scholes and Wolfson (1992) call the global contractingperspective Thus in order to make valid comparisons between the twowe compare the tax burden to the executive while holding constant the

Taxation of Executive Compensation 15

posttax cost to the company By keeping posttax employer costs (in NPVterms) constant any package that is preferred by the employee is tax-advantaged in the global-contracting sense

Under this methodology a comparison of the tax advantages of op-tions and cash involves comparing a pretax cash payment of P with anoption payment that has the equivalent posttax NPV to the companyFor NQSOs it is straightforward to show that a pretax payment of P toan executive has exactly the same cost to the company as putting aside Pfor the purposes of paying stock-option payouts later That is if a com-pany pays an executive P today it will have the same amount of moneyin N years as if it had put aside P today let it accumulate at the rate ofreturn earned by the rm and then paid it out (as compensation in theform of option gains4) with any appreciation in year N

Assume that the pretax prot rate is r and the corporate tax rate is TcThen if the company pays P today it receives a deduction today of PTcwhich yields

PTc [1 1 r (1 2 Tc)]N (1)

in N years since the benets of the tax deduction are invested in thecompany and receive the after-tax corporate rate of return Converselyif the company puts P aside today then it grows at the after-tax corpo-rate rate of return in N years to give a tax deduction of

P [1 1 r (1 2 Tc)]N Tc (2)

which is the exact same value Note that the equivalence of these two taxdeductions is analogous to the tax-benet equivalence of front-loadedand back-loaded IRAs

An executiversquos payoff in N years from option prots is equal to Pcompounded at the after-tax corporate rate times 1 2 Tp where Tp is thepersonal tax rate Thus the combined payoffs in N years of the corporatededuction and the executiversquos posttax payoff is

P [1 1 r (1 2 Tc)]N Tc 1 P [1 1 r (1 2 Tc)]

N (1 2 Tp) (3)

where the rst term is the payoff from the corporate deduction and thesecond term is the executiversquos payoff

4 This analysis is not unique to options Any form of deferred compensation that enablesexecutives to invest inside the rm at a higher posttax return or without paying capitalgains taxes would have similar effects

16 Hall amp Liebman

The payoff to salary and bonus in N years is more complicated sincean assumption must be made about how the executive invests the origi-nal cash compensation For example if the executive invests in an instru-ment with taxable interest (eg bonds) then the N-year payoff from aninvestment of P accumulates at the post-corporate-tax post-personal-taxrate of return and is

P (1 2 Tp) [1 1 r (1 2 Tc)(1 2 Tp)]N (4)

However if an executive invests in (non-dividend-paying) equities theinvestment accumulates at the higher post-corporate-tax rate of returnwith the offsetting disadvantage that the capital gains are taxed at thecapital gains rate in year N So the executive receives

P [1 1 r (1 2 Tc)]N (1 2 Tp) 2 Tcg $P (1 2 Tp) [1 1 r (1 2 Tc)]

N 2P (1 2 Tp) (5)

where Tcg is the capital gains tax rate5Thus putting the equations together if an executive is paid cash P

and invests g in bonds and 1 2 g in equities the N-year payoff (includ-ing the payoff to the corporation from the tax deduction) is

P [1 1 r (1 2 Tc)]N Tc 1 g $(P [1 1 r (1 2 Tc)(1 2 Tp)]

N (1 2 Tp)1(1 2 g) (P [1 1 r (1 2 Tc)]

N (1 2 Tp) 2 Tcg $P (1 2 Tp)[1 1 r (1 2 Tc)]

N 2 P (1 2 Tp)) (6)

The tax advantage of options versus salary and bonus therefore is thedifference between the combined (corporate and executive) payoffs inequation (3) and the combined payoffs in equation (6)

43 The Tax Advantage of OptionsWe now turn to analysis of how changes in various tax rates affectthe tax advantages of options relative to cash compensation We focuson the case in which the cash earned by executives is invested entirelyin equities since equities are tax-favored relative to bonds The impor-tant conclusions of this analysis are not substantively different if we

5 In practice executives hold a combination of equities and bondlike instruments withtaxable interest Because equities are tax-favored in the next section we simplify the analy-sis by assuming that executives hold only equities Alternatively we could have assumedthat executives hold only less risky bonds since bonds more closely match the risk proleof a stream of tax savings None of our empirical results are substantively affected by thissimplication

Taxation of Executive Compensation 17

instead assume that a portion of an executiversquos holdings are invested inbonds

431 The Tax Advantage of Options Changing the Corporate TaxRate Stock options are tax-advantaged relative to cash simply becauseoption payouts are deferred allowing the executive to invest at the pre-personal-tax rate of return (with no capital gains tax at the end) ratherthan investing at the after-tax rate of return or paying the capital gainstax Since the advantage of being able to defer taxes is large when thecorporate tax rate is low the tax advantage of options is larger when thecorporate rate is low

To illustrate this point we dene the tax advantage of options to bethe dollar amount by which total after-tax option payoffs (to the corpora-tion and the executive) exceed the total payoffs from salary as denedby the difference of equation (3) and (6) assuming that P (the paymentto the executive) is equal to $100 and g 5 0 (cash compensation is in-vested in equities) We use a ten-year horizon since most options have aten-year term We then calculate how the tax advantage of optionschanges as the corporate tax rate varies from zero to 100 percent holdingthe personal tax rate and the capital gains rate xed at 40 percent and 20percent respectively (which approximates current rates)

The top panel of Figure 2 shows the tax advantage of options relativeto cashmdashprecisely the NPV of the total tax advantage of options for a$100 payment to the CEOmdashas the corporate tax rate changes As ex-pected the tax advantage of options declines as corporate tax rates in-crease At a zero corporate tax rate the tax advantage of options has anNPV of $7 This value is about $350 at a corporate rate of 40 percentWhen the corporate tax rate is 100 percent the tax advantage of optionscompletely disappears since the benets of deferred compensation fallto zero because the after-tax return on equities (and therefore the dis-count rate) falls to zero The tax advantage of options is the NPV of thetax savings from avoiding the capital gains tax on ten years of apprecia-tion which is why it is relatively modest even at the 40-percent corpo-rate rate

432 The Tax Advantage of Options Changing the Personal TaxRate Using the same assumptions as above (but this time holding thecorporate rate at 40 percent and the capital gains rate at 20 percent andvarying the personal rate) the tax advantage of options declines as thepersonal tax rate rises as shown in the middle panel of Figure 2 The taxadvantage of options is higher at low personal tax rates because theposttax base that gives rise to capital gains is higher and the advantage of

Note Holds constant the personal rate at 40 percent and the capital gains rate at 20 percent

Note Holds constant the corporate rate at 40 percent and the capital gains rate at 20 percent

Note Holds constant the personal rate at 40 percent and the corporate rate at 40 percent

FIGURE 2 Tax Advantage of Options NPV of Tax Advantage of Pay-ing $100 in Options (NQSOs) Rather Than in Cash

Taxation of Executive Compensation 19

avoiding the capital gains taxes is greater The tax advantage of optionstherefore declines linearly as the personal tax rate rises At a 100-percentpersonal tax rate the executive receives nothing in either case and byconstruction the companyrsquos tax deduction benet is the same in NPVterms so the tax advantage of options falls to zero also6

433 The Tax Advantage of Options Changing the Capital Gains TaxRate The tax advantage of options for various capital gains rates isdepicted in the bottom panel of Figure 2 under the same assumptions(this time holding both the personal rate and the corporate rate at 40percent while varying the capital gains rate) The tax advantage of avoid-ing the capital gains tax increases as the capital gains tax rate rises

5 HOW HAS THE TAX ADVANTAGE OF OPTIONSCHANGED OVER TIME

The analysis so far has illustrated how the tax advantage of optionsvaries with the changes in tax rates In this section we show how the taxadvantage of options has changed over time as tax rates have changed

51 Top Marginal Tax Rates over TimeThe evolution on the top marginal tax ratesmdashpersonal corporate andcapital gainsmdashfrom 1980 to 1998 is shown in Figure 3 In all cases thetop rate (the rate for taxpayers with the highest incomes) is shownwhich is not always the highest rate since various anomalies (such asclawbacks of exemptions) sometimes lead to marginal tax rates that arehigher than those for the highest income tax payers

The top personal tax rate was 70 percent in 1980 falling to 50 percentin 1982 and 28 percent in 1988 following the 1986 tax act7 The topmarginal tax rate has since risen to 396 percent but is effectively 425percent since there is a 29-percent Medicare surcharge (paid half by theemployer and half by the employee) that has no upper limit The topcorporate tax rate has had only one signicant change since 1980 fallingfrom 46 percent before the 1986 tax act to 34 percent following the act in1988 The top corporate tax rate was increased to 35 percent in 1993 The

6 This is one place where ignoring bond investments has substantive implications Inparticular in the all-bond case the tax advantage of options is nonmonotonicmdashit rises andthen fallsmdashas personal tax rates rise7 During the early 1980s the maximum tax on earned income limited the marginal tax rateon the earnings of high earners so many executives did not face a decline in personal taxrates between 1981 and 1982 The empirical results in this paper are robust to assumingthat executives faced a marginal tax rate of 50 percent in 1980 and 1981

20 Hall amp Liebman

FIGURE 3 Top Marginal Tax Rates since 1980

top capital gains rate has uctuated between 28 percent (approximately)and 20 percent since 1980 and is currently at 20 percent

52 The Evolution of the Tax Advantage of NQSOsThe tax advantage of options relative to cash dened in the same way asin section 4 from 1980 to 1998 is shown in Figure 4 The calculations aredone using the same assumptions as before (ie N 5 10 years and cashis invested in equities) and the statutory top marginal tax rates shown inFigure 3 are used Due to the fall in corporate and personal tax rates thetax advantage of options increased sharply after the Tax Reform Act of1986 This increase in tax advantage has largely been reversed in the1990s as top personal tax rates have crept back up and the capital gainsrate has been reduced

Two key facts emerge from this analysis First there is currently only amoderate tax advantage to standard nonqualied stock optionsmdashon theorder of $4 per $100 of compensation This is because the tax advantagesto the executive of deferring taxes are largely offset by the tax disad-vantages to the company of not being able to deduct option expensesfrom taxable prots until the executive exercises the options Secondalthough the 1986 tax act substantially increased the tax advantage ofoptions more than half of this increase has been reversed in the 1990s

Taxation of Executive Compensation 21

FIGURE 4 Tax Advantage of Options 1980 to 1998

NPV of tax advantage of $100 in options (NQSOs) rather than cash

53 The Tax (Dis)advantage of ISOsSo far the analysis has focused on the tax advantages of NQSOs sincethey are so much more prevalent than ISOs which account for onlyabout 5 percent of option grants The relative scarcity of ISOs can beexplained both by their tax status and by their per-executive cap of$100000 per year

Under what conditions are ISOs tax-preferred to NQSOs For a trans-fer P to an executive the ISO is always tax-preferred by the executive(since the capital gains rate is lower than the personal rate) while theNQSO is always tax-preferred from the companyrsquos perspective (since anISO is not deductible) The key issue however is the relative advan-tages of the two types of options from a global-contracting perspectiveBecause ISOs are not deductible the company is indifferent betweensetting aside P in the form of NQSOs today and setting aside P(1 2 Tc) inthe form of ISOs (Since the NPV of the tax deduction for NQSOs is PTcthe NPV of the payment P is P 2 PTc or P (1 2 Tc))

Equalizing the posttax cost of NQSOs and ISOs we need only look atthe payoff of the executive to determine the condition under which eachis preferred The payoff of NQSOs in N years is

P [1 1 r (1 2 Tc)]N (1 2 Tp) (7)

22 Hall amp Liebman

The same payoff for ISOs is

P (1 2 Tc) [1 1 r (1 2 Tc)]N (1 2 Tcg) (8)

ISOs therefore are tax-advantaged if

Tp 2 Tcg Tc 2 Tc Tcg (9)

Intuitively ISOs are tax-advantaged only if their advantage (the differ-ence between the personal rate and the capital gains rate) is largeenough to offset their cost (the disadvantage of not deducting at thecorporate rate) Note however that the condition is not a simple com-parison between the corporate rate and the personal rate minus thecapital gains rate (Tp 2 Tcg Tc)

Figure 5 shows how the tax advantage of NQSOs relative to ISOs haschanged since 1980 as tax rates have changed Unlike the very modestchanges (around 3 to 4 percent) in the relative tax advantages betweenoptions and cash shown in Figure 4 the relative tax advantage betweenNQSOs and ISOs has seen enormous swings When the top personalrate was 70 percent in 1980 ISOs were tax-favored by a margin greaterthan 6 percent However by 1982 (following the 1981 tax acts that dra-matically lowered the personal tax rate) ISOs became tax-disadvantaged

FIGURE 5 Tax Advantage of NQSOs Relative to ISOs over Time

NPV of tax advantage of $100 payment to executive NQSOs minus ISOs

Taxation of Executive Compensation 23

by about 5 percent By 1988 (following the 1986 tax act) ISOs became tax-disadvantaged by about 18 percent a dramatic swing The raising ofpersonal rates coupled with a decrease in the capital gains rate has sincereduced the tax disadvantage of ISOs to about $4 per $100 of pretaxcompensation Although hard data on ISOs are hard to come by practi-tioners (mostly compensation consultants) claim that ISOs were morepopular prior to 1982 when they were tax-advantaged Although ISOsare less disadvantageous from a tax standpoint relative to NQSOs thanthey were in the late 1980s they are still disadvantageous so it is notsurprising that they have not made a signicant resurgence in recentyears

6 EXPLAINING THE INCREASE IN OPTION-BASEDCOMPENSATION

In this section we analyze whether changes in tax rates affect the compo-sition of executive compensation and consider the relative importanceof tax factors and corporate governance factors in explaining the in-creasing share of compensation paid in the form of stock options Thenumerous changes in tax rates that have occurred since 1980 provide anopportunity to assess whether executive compensation arrangementsrespond to tax incentives in the way that the tax-avoidance model sug-gests We test this model using a panel data set of CEOs in largepublicly-traded US corporations Our identication of the tax effectsrelies on time-series variation in personal corporate and capital gainstax rates along with cross-sectional variation in corporate tax rates

61 DataWe use a 15-year panel data set of CEOs in the largest publicly-tradedUS corporations which is described in Hall and Liebman (1998) Thedata set covers the years from 1980 through 1994 and combines CEOcompensation information from corporate proxies and 10-K lings withstock price and stock return information from CRSP and accounting datafrom Compustat In addition some compensation data from the 1970swere collected in order to construct measures of the value of stock op-tions held by the CEOs in the rst period of the sample8

The feature that distinguishes our data from most other CEO datasets is that with our panel of yearly proxy data on option grants optiongains and total options held we are able to calculate the total value of

8 We are grateful to David Yermack for providing us with some of the data for 1984 to 1991See Yermack (1995) for a discussion of these data

24 Hall amp Liebman

all stock options held by the CEO at a point of time More importantlysince we have the details about the stock options held (number exer-cise price time to maturity etc) we can precisely calculate the changein the value of a CEOrsquos stock option holdings for a given change in rmvalue

62 Identication IssuesThe numerous tax changes over our sample period and the specic waysin which changes in personal corporate and capital gains rates arepredicted to affect the composition of CEO compensation give us un-usually rich sources of identication However because many of thechanges in tax rates would be expected to affect all of the CEOs in oursample in a similar way we will need to pay particular attention toseparating out the tax effects from underlying time trends and otherfactors that changed over time Controlling for underlying trends is par-ticularly important in this study because we are focusing on an out-comemdashthe use of stock optionsmdashthat has increased rapidly over thepast two decades and that many practitioners believe was importantlyaffected by nontax considerations Thus the tax effects we are trying toexplain are deviations from a rising trend

We take two steps to try to separate out the tax effects from theunderlying trend First we include key nontax factors that could poten-tially explain the increasing reliance on performance-based compensa-tion It has been suggested that the dramatic increase in incentive-basedpay is the result of the remarkably poor shareholder returns during the1970s which spurred the LBO and takeover movements of the 1980sAccording to this story (Kaplan 1997) shareholders became much morepowerful via the rise of institutional investors even as the LBO andtakeover movements waned Because the inuence of institutional inves-tors is thought to be one of the most important mechanisms of strongcorporate governance we use the share of each companyrsquos stock ownedby institutional investors (dened as institutions with more than $100million under management) as an explanatory variable9 The percentageof shares owned by large institutional investors increased from about 20percent to almost 50 percent in 1994 an upward trend that closelymatches the sharp rise in the share of compensation in the form ofoptions Indeed in our sample the annual average (over all of the rmsin our sample) share of stock owned by institutional investors has acorrelation of 09 with the annual average share of compensation paid inoptions

9 This variable is described in Gompers and Metrick (1998)

Taxation of Executive Compensation 25

In addition we include two other variables that proxy for strongercorporate governance The rst is the size of the board Evidence suggeststhat boards with smaller numbers of directors reduce the inuence of theCEO and better represent shareholder interests (Yermack 1995) Boardsize fell by approximately 15 percent between 1980 and 1994 The secondvariable is the fraction of the rmrsquos board members who are inside direc-tors or gray directors (noninsiders who have business dealings with thecompany) Firms with fewer outsiders are less likely to act in the share-holdersrsquo interests and we would predict that they would be less likely tohave performance-based compensation for their CEOs The average per-centage of inside and gray directors has decreased from about 45 percentin 1980 to about 35 percent in 1994

Second we allow for cross-sectional variation in corporate tax ratesCorporate marginal tax rates vary because rms may be eligible for tax-loss carrybacks and carryforwards investment tax credits and the alter-native minimum tax (see Auerbach 1986 Auerbach and Altshuler1990 Auerbach and Poterba 1987) Because these cross-sectional differ-ences imply that different rms should respond differently to a givenchange in tax rates they potentially provide us with a way to isolatethe tax effects Firm-specic corporate marginal tax rates are notori-ously difcult to calculate from publicly-available data We use a tricho-tomous variable that equals zero if the rm has tax-loss carryforwardsand negative earnings in a given year equals 05 times the statutorycorporate rate if the rm has only one of those conditions and equalsthe statutory rate if the rm has neither loss carryforwards nor negativeearnings10

The cross-sectional variation in the corporate tax rate may not be ex-ogenous Firms that perform poorly could face low marginal tax ratesand also be particularly likely or unlikely to provide performance-basedpay Because most of our specications use xed effects the correlationwe must be concerned about is between changes in rm performanceand changes in compensation We deal with this by controlling forlagged rm stock market performance in our regressions This shouldeliminate spurious correlation between tax rates and compensation thatis jointly caused by rm performance

10 Graham (1996) presents evidence that the trichotomous variable performs nearly as wellas his recommended simulated marginal tax rate Plesko (1999) compares marginal taxrates calculated from Compustat data with actual marginal tax rates from rm tax returnsHe nds that simple binary variables are more reliable measures of the marginal tax ratethan is either the Graham or the trichotomous variable We obtain substantively similarresults when we switch from the trichotomous variable to the binary variables recommendby Plesko

26 Hall amp Liebman

63 ResultsThe dependent variable in our regressions is the fraction of a CEOrsquos totalannual compensation (measured as the BlackndashScholes value of stock-option grants plus salary and bonus) that comes from stock-optiongrants We choose this reduced-form specication because our interest isin the composition of CEO compensation However the factors thataffect the composition of compensation might also affect the levels andit will be important to keep this in mind in interpreting our results

Optimal-contracting theory generally makes predictions about the in-centives provided by the total amount of rm stock and stock optionsowned by the CEO We choose to model the ow of option grants ratherthan the ultimate performance sensitivity of the CEOrsquos compensationand stock and stock-option holdings because we have in mind anadjustment-cost model in which it takes time for a CEOrsquos holdings ofstock and stock options to reach the optimal level11 In our regressionswe include two measures of the current performance sensitivity of theCEOrsquos wealth to rm performance as explanatory variables to reect thedistance of the CEOrsquos contract from the optimal contract

Table 3 presents results in which the tax effects are identied solely byvariation in tax rates over time The fraction of annual compensation paidin stock options is regressed on tax variables corporate governance vari-ables contract-theory variables and variables reecting stock market re-turns Both of the regressions in this table include rm xed effects Incolumn (1) the fraction in annual compensation paid in stock options isregressed on the log difference in payoff (the combined payoff to both therm and the CEO) from receiving compensation in stock options ratherthan salary and bonus where the payoff calculations are based on statu-tory marginal tax rates This variable is dened as the log difference be-tween equation (3) and equation (6) in section 4 The coefcient on the taxvariable is large and statistically signicant at the 95-percent level Thepoint estimate implies that a one-percent increase in the payoff differencebetween stock options and salary and bonus results in a 24-percentage-point increase in the share of compensation paid in stock options

The corporate governance variables all have the predicted sign Firmswith a higher fraction of their shares owned by institutional investorsuse more performance-based pay (stock options) Firms with large corpo-rate boards or a large share of inside directors are less likely to use stockoptions

The two contract-theory variables are the dollar change in CEO wealth

11 See Core and Guay (1999) for evidence that rms adjust in this way

TABLE 3Compensation Regressions that Rely on Variation in

Tax Rates over Time

(1) (2)

Tax variablesln (difference in payoff)(a) 2393 (0349)ln (12statutory personal 20015 (0034)

rate)ln (12statutory cap gains 0330 (0171)

rate)ln (12statutory corporate 0399 (0148)

rate)

Corporate governance variablesFraction of rmrsquos shares 0043 (029) 0052 (0028)

owned by institutionalinvestors

Size of the rmrsquos board of 20002 (001) 20003 (0001)directors

Fraction of directors who 20082 (033) 20092 (0030)are inside or gray

Contract-theory variablesD (CEO wealth) per $1000 20261 (0095) 20247 (0087)

in FV(b)

D (CEO wealth) per 10 in 2583 (189) 2554 (201)FV(c)

Volatility(d) 521 (424) 261 (369)ln (rm market value) 0019 (0006) 0019 (0006)

Stock market return variablesLagged annual rm stock 0006 (0009) 0003 (0008)

market returnLagged annual SampP 500 re- 20035 (0023) 20029 (0023)

turn

Regression diagnosticsTotal sample size 5179 5179Firms 427 427Adjusted R2 0313 0311

Dependent variable share of compensation paid in options Standard errors in parentheses Bothregressions in this table include rm xed effects

(a) Log difference in payoff is the log difference in combined rm and CEO after-tax share betweenpaying compensation in options and paying salary

(b) This variable is the dollar change in CEO wealth per $1000 change in rm market value (seeJensen and Murphy 1990)

(c) This variable in the dollar charge in CEO wealth per 10-percent change in rm value (see Hall andLiebman 1998)

(d) Standard deviation of rm market value in trillions

28 Hall amp Liebman

per $1000 change in rm value (Jensen and Murphy 1990) and thedollar change in CEO wealth per 10-percent change in rm value (Halland Liebman 1998) As discussed in Baker and Hall (1998) these twowealth sensitivity measures reect two different concepts of how closelyaligned the managersrsquo incentives are with the interests of the sharehold-ers In the regression they are meant to measure how far the CEOrsquosexisting contract and ownership of stock and stock options is from theoptimal level The negative coefcients on the two variables suggeststhat rms that are below their optimal pay-to-performance sensitivityare more likely to give more stock options in the current period

The second column replaces the tax variable motivated by deferraladvantages with the three statutory rates The corporate after-tax sharehas a positive coefcient as we would expect because the tax advantageof stock options increases as the corporate rate falls The coefcient onthe personal tax rate is small and statistically insignicant The sign ofthe coefcient on the capital gains rate is the opposite of that predictedby tax-avoidance theory

Note that these results rely on time-series variation alone This is not avery convincing test Since the log difference in payoffs mostly rises overthe time period covered in our sample as does the share of compensa-tion paid in options it is possible that the tax variable is simply reectingother factors that were changing over this time period that we have notincluded in our regression Therefore it is important to look at specica-tions that rely on cross-sectional variation in the tax variables

Table 4 contains specications that rely on cross-sectional variation incorporate tax rates The rst column repeats the specication from col-umn (1) of Table 3 but uses the trichotomous measures of rm corporatetax rates in place of the statutory rate in calculating each rmrsquos logdifference in payoff from paying compensation in options rather than insalary and bonus As in the previous table we nd the positive coef-cient on the log difference in payoff just as the theory suggest How-ever the coefcient estimate is now much smaller suggesting that aone-percent increase in the log difference in payoff produces a 01-percentage-point increase in the share of compensation paid in options

Introducing cross-sectional variation in the dependent variable doesnot eliminate the potentially spurious correlation over time between thetax variable and the share of compensation paid in options The secondcolumn of Table 4 isolates the pooled cross-sectional variation in corpo-rate tax rates by adding time dummies and dropping the rm xedeffects In this specication the coefcient on the tax variable is indistin-guishable from zero The coefcient on the institutional variable contin-ues to be large and statistically signicant The third column includes

Taxation of Executive Compensation 29

both rm xed effects and time dummies which tests for a relationshipbetween within-rm changes in the dependent variable and within-rmchanges in each explanatory variable In this specication the coefcienton the tax variable is again indistinguishable from zero The coefcientson the corporate governance variables are also statistically insignicant

The third column contains a specication that treats the Tax ReformAct of 1986 (TRA 86) as a natural experiment For rms and CEOs facingthe statutory corporate tax rate TRA 86 increased the relative payoff topaying compensation in options by 17 percent For a rm facing a zeromarginal tax rate it reduced the relative payoff by 45 percent The reduc-tion occurred because TRA 86 reduced the personal tax rate makingsalary and bonus more attractive while the decreased corporate rate hadno effect on zero-tax-rate rms Thus we would expect rms facing zeroor low marginal tax rates to reduce their use of options after TRA 86while rms facing the statutory rate would be expected to increase themFew rms face zero marginal tax rates year after year Therefore we tryto distinguish between rms that often face low tax rates and rms thatnearly always face the statutory rate We dene low-tax rms as oneswhose average marginal tax rate over 1984 to 1986 was below 02412

Then we ran a difference-in-differences regression to see if high-taxrms increased their use of options more than low-tax rms The interac-tion between post-TRA 86 and high corporate tax rates is the key vari-able The estimated coefcient on this variable is positive but small andnot statistically different from zero at the 95-percent level

It is important to note that in this specication as well the main corpo-rate governance variable the share of the rmrsquos shares owned by institu-tional investors performs as theory would predict The variable appearsto be quite robust Table 5 presents regressions of the share of compensa-tion paid in stock options on the fraction of the rmrsquos shares owned byinstitutional investors Since we have already established that these vari-ables have a strong longitudinal relationship the regressions here aredesigned to focus on the cross-sectional relationship between these twovariables The rst and third columns present regressions using data onlyfrom 1994 and nd a coefcient of roughly 02 (column 1 contains noother covariates while column 3 includes all of the other nontax covari-ates from the regressions in Tables 3 and 4) Since the median share of armrsquos stock owned by institutional investors in our sample rises from 19percent in 1980 to 49 percent in 1994 this coefcient would imply that this

12 This cutoff was chosen after inspecting the distribution of tax rates and essentiallyseparates rms that are always at or near the statutory rate from ones that are sometimesbelow it

TA

BL

E4

Com

pens

atio

nR

egre

ssio

nsth

atR

ely

onC

ross

-Sec

tion

alV

aria

tion

inC

orpo

rate

Tax

Rat

es

(1)

(2)

(3)

(4)

Tax

vari

able

s(a)

ln

(dif

fere

nce

inp

ayof

f)0

099

(00

41)

20

038

(00

36)

20

005

(00

44)

Pos

t-T

RA

862

000

5(0

013

)H

igh

corp

orat

eta

xra

te0

018

(00

17)

(Pos

t-T

RA

86)

3hi

gh0

024

(00

16)

corp

orat

eta

xra

te

Cor

por

ate

gove

rnan

ceva

riab

les

Frac

tion

of

rmrsquos

shar

es0

109

(00

27)

011

4(0

015

)2

000

9(0

030

)0

087

(00

34)

owne

dby

inst

itu

tion

alin

vest

ors

Size

ofth

e

rmrsquos

boar

dof

20

004

(00

01)

20

001

(00

01)

20

002

(00

01)

000

1(0

002

)d

irec

tors

Frac

tion

ofd

irec

tors

wh

o2

012

4(0

033

)2

000

3(0

018

)2

004

1(0

034

)0

001

(00

39)

are

insi

de

orgr

ay

Con

trac

t-th

eory

vari

able

sD

(CE

Ow

ealt

h)p

er$1

000

20

238

(00

94)

20

167

(00

53)

20

257

(00

96)

20

021

(01

47)

inFV

(b)

D(C

EO

wea

lth)

per

10

in2

560

(17

8)2

592

(10

5)2

594

(19

6)2

147

(04

86)

FV(c

)

Vol

atili

ty(d

)2

48(4

24)

162

(03

4)1

12(4

28)

106

(06

09)

ln(

rmm

arke

tva

lue)

003

3(

006)

000

8(0

003

)0

009

(00

06)

001

2(0

007

)

Stoc

km

arke

tre

turn

vari

able

sL

agge

dan

nual

rm

stoc

k2

000

2(0

009

)0

011

(00

10)

000

6(0

009

)2

001

2(0

013

)m

arke

tre

turn

Lag

ged

annu

alSamp

P50

02

005

6(0

023

)2

005

3(0

051

)2

003

9(0

049

)2

015

4(0

040

)re

turn

Reg

ress

ion

dia

gnos

tics

T

otal

sam

ple

size

5179

5179

5179

2534

Firm

s42

742

739

6A

dju

sted

R2

031

009

032

005

Firm

xed

effe

cts

Yes

No

Yes

No

Tim

ed

um

mie

sN

oY

esY

esN

o

Dep

end

ent

vari

able

isfr

acti

onof

ann

ualc

omp

ensa

tion

from

stoc

kop

tion

sSt

and

ard

erro

rsin

par

enth

eses

(a)

All

thre

esp

eci

cati

ons

use

tric

hoto

mou

sco

rpor

ate

tax

rate

sT

he

sam

ple

inco

lum

n(3

)u

ses

only

the

year

s19

84to

1988

sin

cew

ear

eu

sing

1986

asa

nat

ural

exp

erim

ent

(b)

Thi

sva

riab

leis

the

dol

lar

chan

gein

CE

Ow

ealt

hp

er$1

000

chan

gein

rm

mar

ketv

alu

e(s

eeJe

nse

nan

dM

urp

hy

1990

)

(c)

Thi

sva

riab

lein

the

dol

lar

char

gein

CE

Ow

ealt

hp

er10

-per

cent

chan

gein

rm

valu

e(s

eeH

alla

nd

Lie

bman

199

8)

(d)

Stan

dar

dd

evia

tion

ofr

mm

arke

tva

lue

intr

illio

ns

32 Hall amp Liebman

TABLE 5Large Institutional Investors and Stock-Option Compensation

The Cross-sectional Relationship between Institutional Investors andOption Grants

(1) (2) (3) (4)

Coefcient on fraction of 217 (067) 145 (014) 177 (069) 120 (015)rmrsquos shares owned byinstitutional investors

Sample period 1994 1980ndash1994 1994 1980ndash1994Contract-theory and stock No No Yes Yes

market covariatesTime dummies No Yes No Yes

The dependent variable is the share of compensation paid in options

factor alone can explain 6 percentage points of the 23-percentage-pointincrease in the median share of compensation paid in options that oc-curred over this time period Of course this variable is only one measureof the strengthening of corporate governance so the total impact of im-proved corporate governance could be greater The second and fourthcolumns of the table use the entire 1980ndash1994 sample and include timedummies to isolate the cross-sectional correlations Again there is evi-dence of a strong cross-sectional relationship between the two variablesalthough the coefcients are slightly smaller perhaps reecting a weakercorporate-governance role in the earlier period In sum there is a strongrelationship in both longitudinal and cross-sectional data between theinstitutional-investor variable and the options-percentage variable13

7 THE MILLION-DOLLAR RULE

In 1993 section 162(m) of the Internal Revenue Code was enactedlimiting the deductibility of executive compensation in excess of onemillion dollars unless the compensation was performance-related Pro-ponents of this legislation argued that it would slow the growth ofexecutive pay and tighten the link between rm performance and execu-tive compensation

The new provision became effective for tax year 1994 and applies tothe CEO and the other four most highly compensated executives in eachcompany Pay that is performance-related such as stock-option grants

13 The one specication in which the fraction owned by institutional investors is notsignicant is in column (2) of Table 4 a regression that includes both xed effects and timeeffects

Taxation of Executive Compensation 33

and bonuses paid for meeting clear performance goals is not affected bythis regulation provided that it has been approved by shareholdersThus the main impact of the provision is to limit the deductibility ofsalary in excess of a million dollars Woodlock and Antenucci (1997)studied the proxy statements of 376 rms and documented that mostrms responded to the new law by qualifying top executive pay asperformance-related Perry and Zenner (1999) identied 25 rms thatreduced salaries from above $1 million to below $1 million and foundthat 23 of the rms cited section 162(m) as their reason for doing so

In this section of the paper we investigate whether this provision hasaffected the level or structure of executive pay using the Execucompdata base which contains information on the highest-paid executives ineach of the rms in the SampP 500 SampP Midcap 400 and SampP SmallCap600 Between 1993 and 1998 the median salary of the Execucomp execu-tives rose from $230000 to $306000 The median value of total compen-sation (salary bonus and option grants) rose from $470000 to $882000Limiting the sample of CEOs indicates that the total compensation of themedian CEO rose from $11 million to $18 million over the period whilethe median salary rose from $441000 to $550000 Thus executive com-pensation clearly continued to rise at a rapid rate after the implementa-tion of the million-dollar rule though perhaps at a slower rate than itotherwise would have

If the regulation affects behavior it should have the largest impact onrms whose executives are receiving salaries that are above the limit orjust below it Thus we might expect to see rms whose executives arereceiving relatively high salaries to rely more heavily on performance-based pay and less heavily on salary in the years following the regulationPerry and Zenner (1999) and Livingston (1999) both nd that this is thecase This trend can be clearly identied in the Execucomp data set Forexample over 1993 to 1998 the median annualized growth rate of salaryfor executives whose 1993 salary was below $500000 in the previous yearwas 85 percent For executives with salary between $500000 and$700000 it was 64 percent for executives between $700000 and $850000it was 50 percent for executives between $750000 and $1 million it was31 percent and for executives above $1 million it was 00 percent14

The observation that executives at lower levels of salary had highergrowth rates is not very strong evidence of an impact of section 162(m)however because it is possible that low-salary executives would havehad higher growth rates even in the absence of the provision To explore

14 Of the 3137 executives for whom we have complete data between 1993 and 1998 2708had 1993 salaries below $500000 while only 35 had 1993 salaries above $1 million

34 Hall amp Liebman

whether this is the case we use Execucomp data from both before andafter the law change Ideally we would want to have data from manyyears before the change to examine whether there is an underlying pat-tern of faster salary growth for low-salary executives Unfortunately theExecucomp data begin in 1992 and therefore provide us with only the1992ndash1993 growth rate in compensation to use as a baseline15

We run regressions pooling data from 1992ndash1993 through 1997ndash1998of the annual percentage change in different forms of compensation on avariable million that is designed to measure the likelihood that anexecutiversquos compensation will be affected by section 162(m) We denemillion as the minimum of 1 and the executiversquos salary divided by $1million in the previous year Thus any executive with salary at or above$1 million would receive a value of 1 while an executive with a salary of$500000 would be coded as a 0516 We include this variable by itself andalso interacted with a dummy variable (after) that equals 1 for years1994 and after Thus the million variable is designed to pick up anyunderlying relationship between the level of salary and the growth rateof salary while the interacted one will identify any differential growthrate of salary by income in the period after section 162(m) took effect Wealso include a full set of year dummies the natural logarithm of thermrsquos market value and the rmrsquos annual rate of return over the twoprevious years as control variables To reduce the sensitivity of our re-sults to outliers we run quantile regressions and robust regressions

Table 6 contains our results The rst four columns are for all topexecutives while the last four columns focus on CEOs Columns (1) and(2) present estimates from robust regression for the percentage change insalary and the percentage change in total compensation (salary bonusstock grants and stock-option grants) In the salary growth regressionsfor all executives the coefcient on million is negative and signicantindicating that there is an underlying relationship between the level ofexecutive compensation and its subsequent growth rate The coefcienton million 3 after is negative as well suggesting that the negativerelationship between the level of salary and its subsequent growth rateintensied slightly after 1994 The relatively small magnitude suggeststhat an executive with a $1 million dollar salary would see his salarygrow at an annual rate that would be about 06 percent slower than anexecutive earning $500000

15 The HallndashLiebman data set contains only a combined salary and bonus variable makingit impossible to use it to analyze this issue16 We also ran the regressions (not reported) allowing for a nonlinear relationship betweenmillion and compensation growth and obtained substantively similar results

TA

BL

E6

The

Eff

ect

ofth

eM

illi

on-D

olla

rR

ule

onSa

lary

and

Tot

alC

ompe

nsat

ion

All

top

exec

uti

ves

CE

Os

only

Rob

ust

regr

essi

onQ

uan

tile

regr

essi

onR

obu

stre

gres

sion

Qu

anti

lere

gres

sion

D

in

Din

D

in

Din

D

in

Din

D

in

Din

sala

ryto

talc

omp

(a)

sala

ryto

talc

omp

sa

lary

tota

lcom

p

sala

ry

tota

lcom

p

mil

lio

n(b

)2

006

02

035

72

006

32

024

12

008

72

023

12

010

32

024

5(0

005

)(0

032

)(0

006

)(0

027

)(0

010

)(0

073

)(0

011

)(0

056

)m

illi

on

32

001

10

026

20

021

20

015

20

016

014

92

003

40

158

aft

er(c

)(0

005

)(0

033

)(0

006

)(0

029

)(0

010

)(0

076

)(0

011

)(0

057

)N

359

5035

950

359

5035

950

757

17

589

757

17

589

Stan

dar

der

rors

inp

aren

thes

esR

egre

ssio

nsal

soin

clu

de

year

du

mm

ies

stoc

km

arke

tre

turn

and

log

mar

ket

valu

e

(a)

Tot

alco

mp

ensa

tion

incl

ud

essa

lary

bon

us

rest

rict

edst

ock

and

stoc

k-op

tion

gran

ts

(b)

mil

lion

isd

ene

das

min

(1s

alar

yin

pre

viou

sye

ars

div

ided

by$1

mil

lion)

(c)

aft

eris

anin

dic

ator

vari

able

equ

alto

zero

befo

reth

ere

gula

tion

took

plac

e(1

993)

and

one

afte

rwar

d

36 Hall amp Liebman

In the second column the coefcient on million 3 after is essentiallyzero suggesting that any decrease in salary brought about by section162(m) was offset by increases in bonus and stock-option grants Theresults are similar for the median regressions There is some evidence ofa very small slowdown in growth of salary for executives likely to beaffected by section 162(m) but no sign of a change in total compensa-tion Taken together this provides evidence of a very small substitutionof performance-related pay for salary

The last four columns run identical tests but limit the sample to CEOsThe interaction of million 3 after continues to be small and negative inthe salary regressions In contrast the impact of 162(m) on total compen-sation appears to be positive and fairly substantial (about 15 percentagepoints per year) in these regressions That is the increase in bonus andstock options following the million-dollar rule more than offset the verysmall decline in salary

All of these conclusions should be interpreted somewhat cautiouslybecause our data provide us with only one pre-1993 control year Never-theless the data are consistent with section 162(m) having led to a veryminor slowdown in salary growth and one that is dramatically smallerthan what one would conclude without controlling for underlying differ-ential salary growth rates at different levels of salary Since total compen-sation for CEOs did not decline and perhaps increased there is evidenceof a minor substitution of performance-related pay for salary in responseto the regulation

8 EXECUTIVE COMPENSATION AND THEELASTICITY OF TAXABLE INCOME

Even if taxes have only a minor effect on executive compensation thestructure of executive compensation is important for tax policy be-cause high-income taxpayers are sources of signicant revenue and arethe focus of important debates about the efciency cost of taxation

Goolsbee (1999) has recently argued that executive decisions to exer-cise options are highly responsive to intertemporal differences in taxrates created by tax reforms He claims that much of the apparent declinein taxable income for high-income taxpayers between 1992 and 1993documented by Feldstein and Feenberg (1996) can be explained by execu-tives shifting option gains into 1992 in order to avoid the higher marginaltax rates that went into effect in 1993 By extension he implies that thehigh elasticities of taxable income with respect to personal tax ratesestimated by comparing the pre- and post-tax-reform taxable income of

Taxation of Executive Compensation 37

high-income individuals reect intertemporal shifting of income and notpermanent effects of taxation If this is true then the deadweight loss oftaxation is lower as is the cost of progressivity

While Goolsbee presents comprehensive evidence that option gainswere unusually high in 1992 his identication of the impact of tax rates onthe timing of option gains depends on correctly specifying a counter-factual level of options that would have been exercised in the absence ofthe tax change He simply assumes a linear time trend The purpose ofGoolsbeersquos time trend is presumably to account for the increasing use ofoptions over time However since options typically require a vestingperiod of at least a couple of years a much more direct measure of ex-pected option gains can be constructed based on the total value of vestedoptions held by the CEO Since options typically have vesting periods thatexceed two years this total can be treated as exogenous since it will not beaffected by the tax changes

More generally the value of options held by the CEO is likely to havea major effect on his exercise decisions If his options are out of themoney he will not have gains to realize If there has been a recent run-up in the market an optimizing CEO may exercise an unusually largeamount of options in order to diversify his overall portfolio and if in-the-money options are about to expire they will be exercised in thecurrent year

Our evidence suggests that much of the apparent tax shifting caninstead be attributed to stock market performance and the timing of pastoption grants Using our 15-year panel we replicate Goolsbeersquos (1999)result for the 1993 period but show that the impact of taxes is notpresent in other time periods Moreover even the result for the 1993period disappears when appropriate controls for past option grants andstock price appreciation are included

The top three rows of Table 7 present data on the taxable incomes ofCEOs in the HallndashLiebman sample For these results our panel has beennarrowed to include only CEOs whose rmrsquos scal years correspondwith tax years In addition we create a balanced panel for each of thethree tax-reform periods by excluding any rm for which there are notcomplete data for the period surrounding the reform

The fourth row of the table shows that the 1993 tax act raised the topmarginal personal income tax rate from 310 to 396 percent effective fortax year 1993 with a further increase occurring in 1994 due to the uncap-ping of the Medicare payroll tax The top row of the table shows thattaxable income was higher in 1992 than for any other year in our sample(all data are in 1994 dollars) Option gains were particularly high in1992mdashmore than double their 1991 level and 68 percent higher than their

TA

BL

E7

Tax

Ref

orm

Sto

ckM

arke

tP

erfo

rman

cea

ndth

eT

imin

gof

Opt

ion

Gai

ns

Ave

rage

stoc

km

arke

tM

ean

CE

OM

ean

CE

Ore

turn

for

Mea

nva

lue

taxa

ble

sala

ryan

dM

ean

CE

OT

opfe

der

al

rms

inof

opti

ons

(Val

ue

ofop

tion

Sam

ple

inco

me

bonu

sop

tion

gain

sp

erso

nal

sam

ple

for

held

atbe

gga

ins)

(va

lue

Ref

orm

size

Yea

r($

mil

)($

mil

)($

mil

)M

TR

pre

vye

ar(

)of

year

ofop

tion

she

ld)

ER

TA

8127

919

800

874

064

80

163

7025

80

460

354

1981

083

10

672

009

869

245

070

61

3919

820

872

067

90

099

5011

90

686

144

1983

099

00

724

025

250

355

116

62

16

TR

A86

295

1985

119

70

837

024

050

76

116

02

0719

861

448

091

70

345

5042

81

776

194

1987

165

50

959

059

338

519

12

229

266

1988

163

21

058

039

028

07

237

71

6419

891

656

105

20

414

2817

92

345

176

1993

272

1990

209

51

028

061

828

232

288

02

15T

axA

ct19

911

875

102

00

547

312

01

268

82

0319

922

678

107

01

250

3146

64

552

275

1993

232

41

209

074

339

620

64

299

173

1994

196

41

262

037

842

515

45

113

074

Sam

ple

isH

allndash

Lie

bman

rm

sw

ith

Dec

embe

r

scal

year

sth

atha

dco

mp

lete

dat

afo

ron

eof

the

thre

eta

x-re

form

per

iod

sst

ud

ied

Taxation of Executive Compensation 39

1994 level These patterns are consistent with Goolsbeersquos claim that tax-payers responded to the anticipated increase in marginal tax rates byshifting gains into 1992

The evidence from the other two tax reforms is less clear suggestingthat there might be factors other than tax rates determining the timing ofoption gains ERTA 1981 reduced the top marginal tax rate from 70percent in 1980 to 50 percent in 1982 However there is no sign thatoption gains were reduced in 1980 and 1981 and then increased in 198217

There was a big increase in option gains in 1983 however Similarly thehighest year of stock-option gains in the period around 1986 was 1987 astrange year to take gains since taxpayers would have known that thepersonal rate would be 10 percentage points lower if they waited untilthe following year18

What all the peak years of option gains have in common is that theyfollow years of strong stock market performance The fth row of Table 7shows the average stock market return during the previous year for therms in our sample The large value of option gains in 1983 followed a355-percent increase in the stock market The 1992 boom in option gainsfollowed a 466-percent increase in the stock market the more modest1987 peak in option gains followed a 19-percent gain (and a 43-percentgain two years before)

Strong stock market performance raises the value of options availableto exercise The sixth row of the table shows the mean value of optionsheld by CEOs at the beginning of the year All three of the peak years ofoption gains correspond with peak years in terms of the value of optionsavailable for exercise Since stock market appreciation during the yearalso increases the value of options available to exercise later in that yearwe need to allow for current-year stock market appreciation as well

Table 8 contains Goolsbee-style regressions for each of the tax re-forms The rst two columns reprint the results from Goolsbee (1999)The third and fourth columns replicate his results in our sample Inparticular with a specication similar to his we are able to replicate hisnding that the current after-tax share has a positive coefcient and thefollowing yearrsquos after-tax share has a coefcient of roughly negative oneColumns (5) through (8) show that this pattern was unique to the 1993tax reform Indeed in the TRA 86 regressions the coefcient on the

17 As we explained above for executives covered by the maximum tax on earned incomethere was not a decline in tax rates between 1981 and 198218 It is possible that the large stock market decline in October 1997 may have promptedsome people to exercise options earlier than they otherwise would have However anincrease in option exercises typically follows stock market increases not declines

TA

BL

E8

Att

empt

sto

Rep

lica

teth

eT

ax-S

hift

ing

Res

ult

inO

ther

Tim

eP

erio

ds

Goo

lsbe

eG

ools

bee

1993

1993

Ent

ire

Ent

ire

regr

essi

onre

gres

sion

refo

rmre

form

TR

A86

TR

A86

ER

TA

81E

RT

A81

1980

ndash199

419

80ndash1

994

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

ln[1

2ta

x(t)

]1

244

108

20

455

032

62

017

62

017

22

016

72

016

70

026

20

068

(01

34)

(01

27)

(05

37)

(05

42)

(02

10)

(02

09)

(00

90)

(00

91)

(00

67)

(00

65)

ln[1

2ta

x(t1

1)]

20

867

20

764

21

214

20

917

047

80

722

20

103

004

30

176

018

3(0

118

)(0

112

)(0

491

)(0

499

)(0

157

)(0

174

)(0

072

)(0

072

)(0

068

)(0

066

)

Tim

e0

135

007

62

000

52

036

40

041

000

80

094

003

50

056

002

7(0

008

)(0

008

)(0

049

)(0

049

)(0

028

)(0

028

)(0

025

)(0

026

)(0

003

)(0

003

)

ln(m

arke

tva

lue)

066

10

489

023

70

334

039

5(0

015

)(0

058

)(0

039

)(0

042

)(0

017

)

Yea

rs19

91ndash1

995

1991

ndash199

519

90ndash1

994

1990

ndash199

419

85ndash1

989

1985

ndash198

919

80ndash1

983

1980

ndash198

319

80ndash1

994

1980

ndash199

4

Ad

just

edR

20

720

770

610

630

690

760

800

810

630

68

Obs

erva

tion

s13

856

132

371

385

136

41

470

143

31

113

106

23

968

385

9

Dep

end

ent

vari

able

isth

elo

gof

taxa

ble

inco

me

All

regr

essi

ons

con

tain

rm

xe

def

fect

sG

ools

bee

regr

essi

on(2

)al

soco

ntai

nstw

oot

her

vari

able

sno

tsh

own

inth

eta

ble

Stan

dar

der

rors

inp

aren

thes

es

Taxation of Executive Compensation 41

TABLE 9Alternative Explanations for the Timing of Option Exercises

1993 Entirereform TRA 86 ERTA 81 1980ndash1994

(1) (2) (3) (4)

ln[12tax(t)] 0108 0111 20051 0201(0324) (0136) (0048) (0057)

ln[12tax(t11)] 20134 0573 0125 0150(0336) (0156) (0053) (0066)

ln(market value) 0418 0177 0338 0418(0061) (0039) (0046) (0163)

Firmrsquos stock market return (t) 0298 0268 0214 0258(0044) (0045) (0031) (0023)

Firmrsquos stock market return 0162 0206 0031 0128(t21) (0041) (0008) (0029) (0021)

ln(value of options held by 0004 0008 0001 0009CEO at start of year) (0003) (0003) (0002) (0001)

Years 1990ndash1994 1985ndash1989 1980ndash1983 1980ndash1994Adjusted R2 064 071 082 068Observations 1349 1427 1036 3812

The dependent variable is the log of taxable income All regressions contain rm xed effects Stan-dard errors in parentheses

following yearrsquos tax rate is positive and signicant Columns (9) and (10)pool all three reforms and fail to replicate the pattern that would reectintertemporal shifting19

It is possible that the ease with which one can shift the timing ofoption gains is asymmetric It may be harder to shift option gains for-ward in time since some options expire and therefore must be exercisednow In addition options were a much smaller portion of compensationin the early 1980s so perhaps the ERTA 81 results are not surprisingNonetheless these results cast doubt on the proposition that this is thechannel through which spurious estimates of the elasticity of taxableincome have occurred for previous tax reforms and they suggest thatwe should look harder for alternative explanations for the timing ofoption gains

Table 9 attempts to predict options gains using the lagged and currentincrease in a rmrsquos market value and the value of the options held by theCEO at the start of the year The two stock market variables stronglypredict option gains while the value of options held has the correct sign

19 The results also fail to suggest large permanent effects of marginal tax rates on taxableincome

42 Hall amp Liebman

in all of the specications and is signicant in the pooled sample eventhough it is highly correlated with the two stock market return variablesIn contrast the tax-timing explanation fails even in the 1993 period

It is important to realize that our results do not invalidate Goolsbeersquoscritique of claims that aggregate revenue data from 1992 and 1993 provethat elasticities of taxable income are large We have simply given analternative explanation for why taxable incomes were unusually high in1992 However our results do cast doubt on the conclusion that largetiming shifts are ubiquitous and capable of explaining away the morecareful estimates of taxable income elasticities that have been performedusing micro data (eg Feldstein 1995 Auten and Carroll 1995)

One further point is in order It is sometimes argued that the combina-tion of increased use of options and the booming stock market explainswhy federal tax revenues have been so much higher than expected dur-ing the 1990s However while large option gains increase revenue fromthe personal income tax they produce offsetting deductions for cor-porations The net impact on revenue is only the difference in the twomarginal tax rates times the option gain While the gap between thepersonal rate and the effective corporate rate is nontrivial the impact oflarge option exercises on total revenue is substantially smaller than theamount by which the exercises inate personal tax revenue and reducecorporate tax revenue

9 CONCLUSION

We have described the tax rules for executive pay in detail and analyzedhow changes in various tax rates affect the tax advantages of stock optionsrelative to salary and bonus We nd that there is a moderate tax advan-tage to optionsmdashcurrently about $4 per $100 in pretax executive paymdashand that changes in the tax advantage of options over time have had atmost a modest impact on the composition of pay Corporate governancefactors particularly the role of large investors appear to be more impor-tant in explaining the dramatic increase in option pay

In addition our evidence suggests that more direct attempts to use taxpolicy to inuence executive compensation have had little effect Wend that the million-dollar rule led companies to substitute perfor-mance-based pay for salary But our evidence suggests that this substitu-tion was quite modest and there is no evidence that the total level of paywas reduced Overall although the stock-option explosion has dramati-cally increased the link between pay and performance this change isdue almost entirely to nontax factors

Taxation of Executive Compensation 43

REFERENCES

Auerbach Alan J (1996) ldquoThe Dynamic Effects of Tax Law Asymmetriesrdquo Re-view of Economic Studies 53451ndash405

mdashmdashmdash and James Poterba (1987) ldquoTax Loss Carryforwards and Corporate TaxIncentivesrdquo In The Effects of Taxation on Capital Accumulation M Feldstein(ed) Chicago University of Chicago Press

Altshuler Rosanne and Alan Auerbach (1990) ldquoThe Signicance of Tax LawAsymmetries An Empirical Investigationrdquo Quarterly Journal of Economics 10561ndash86

Auten Gerald and Robert Carroll (1995) ldquoThe Effect of Income Taxes on House-hold Behaviorrdquo Ofce of Tax Analysis US Department of the Treasury

Baker George P and Brian J Hall (1998) ldquoCEO Incentives and Firm SizerdquoNBER Working Paper no 6868 December

Core John and Wayne Guay (1999) ldquoThe Use of Equity Grants to ManageOptimal Equity Incentive Levelsrdquo Draft

Feldstein M (1995) ldquoThe Effect of Marginal Tax Rates on Taxable Income APanel Study of the 1986 Tax Reform Actrdquo Journal of Political Economy 103 (no3)551ndash72

mdashmdashmdash and D Feenberg (1996) ldquoThe Effect of Increased Tax Rates on TaxableIncome and Economic Efciency A Preliminary Analysis of the 1993 Tax RateIncreasesrdquo In Tax Policy and the Economy vol 10 James M Poterba (ed)Cambridge MA MIT Press

Gompers Paul A and Andrew Metrick (1998) ldquoInstitutional Investors andEquity Pricesrdquo NBER Working Paper no 6723

Goolsbee Austan (1999) ldquoWhat Happens When You Tax the Rich Evidencefrom Executive Compensationrdquo NBER Working Paper no 6333

Graham John R (1996) ldquoProxies for the Corporate Marginal Tax Raterdquo Journal ofFinancial Economics 42187ndash221

Hall Brian J (1999) ldquoThe Design of Multi-year Stock Option Plansrdquo Journal ofApplied Corporate Finance 12(2)97ndash106

mdashmdashmdash and Jeffrey B Liebman (1998) ldquoAre CEOs Really Paid Like Bureau-cratsrdquo Quarterly Journal of Economics CXII(no 3)653ndash691

Jensen Michael C and William M Meckling (1976) ldquoTheory of the Firm Mana-gerial Behavior Agency Costs and Ownership Structurerdquo Journal of FinancialEconomics III305ndash360

mdashmdashmdash and Kevin J Murphy (1990) ldquoPerformance Pay and Top-ManagementIncentivesrdquo Journal of Political Economy XCVIII225ndash264

Kaplan Steven N (1997) ldquoThe Evolution of US Corporate Governance WeAre All Henry Kravis Nowrdquo Draft

Livingstone Jane R (1999) ldquoExecutive Compensation Contracting and Re-sponses to Increased Tax Costsrdquo Louisiana State University June

Miller Merton H and Myron S Scholes (1982) ldquoExecutive CompensationTaxes and Incentivesrdquo In Financial Economics Essays in Honor of Paul CootnerWilliam F Sharpe and Cathryn M Cootner (eds) Princeton NJ Prentice-Hall

Murphy Kevin J (1999) ldquoExecutive Compensationrdquo forthcoming in OrleyAshenfelter and David Card (eds) Handbook of Labor Economics Vol 3 NorthHolland

Perry Todd and Marc Zenner (1999) ldquoPay for Performance Government Regu-

Taxation of Executive Compensation 5

than 50 times larger than changes in salary and bonus (Hall and Lieb-man 1998)

As a concrete example the estimates in Hall and Liebman (1998) implythat a 10-percent increase in rm value (of the median company in oursample) leads the company to increase the CEOrsquos salary and bonus byabout $25000 However this same 10-percent increase in shareholdervalue translates into $125-million increase in the value of the CEOrsquosstock and stock-option holdings1

The dramatic rise in the link between CEO wealth and rm perfor-mance can be seen in Figure 1 which shows how two measures of thislink have increased since 19802 The rst measure the JensenndashMurphy(1990) sharing rate (shown on the left scale) is the change in CEO wealthfor a $1000 change in rm value The second measure is the change inCEO wealth for a 10-percent change in rm value (see Baker and Hall1998) Since both measures are strongly affected by rm size (the formerhas a negative correlation and the latter has a positive correlation) thepay-to-performance changes over time are estimated with regression(quantile) analysis that controls for changes in the distribution of rmsizes in the sample over time The gure therefore shows the increasein the pay-to-performance measures over time for a constant-size rmin this case a $1-billion rm (in constant 1998 dollars) The striking fact isthat both measures of the pay-to-performance link have increased bynearly a factor of 10 since 1980 These pay-to-performance increases areeven larger than those we reported in our earlier paper that analyzed theperiod 1980 to 1994 because of the large increase in stock-option grantscombined with the strong stock market performance in 1994 to 1998

3 TAX AND ACCOUNTING RULES

Stock options give an executive the right but not the obligation to buy ashare of the companyrsquos stock at a prespecied pricemdashthe exercise or strikeprice Typically options cannot be exercised immediately That is theyvest (become owned by the executive who can then exercise if he or she

1 Stock-option grants are also very sensitive to changes in rm performance mostly be-cause many grants are multiyear plans that hold the yearly number of options constantand the same number of at-the-money options are worth more when the stock price ishigher and vice versa If stock-option grant sensitivity is also included then about 91percent of pay-to-performance sensitivity comes from stock and stock-option revaluations7 percent comes from stock-option grant changes and less than 2 percent comes fromchanges in salary and bonus (Hall 1999)2 Both measures include only the link created by CEO holdings of stock and stock optionsand ignore the smaller amount of pay-to-performance sensitivity that operates throughchanges in salary bonus and stock-option grants

FIG

UR

E1

Pay

toP

erfo

rman

ceT

heL

ink

betw

een

CE

OW

ealt

han

dSh

areh

olde

rV

alue

Sinc

e19

80

Not

eT

he

anal

ysis

cont

rols

for

the

chan

ging

com

pos

itio

nof

rm

size

over

tim

eT

hu

sth

ese

esti

mat

essh

owth

ep

ay-t

o-p

erfo

rman

celin

kfo

ra

sim

ilar

-siz

edco

mp

any

over

tim

ein

this

case

aco

mp

any

wit

ha

mar

ket

valu

eof

$1bi

llion

Taxation of Executive Compensation 7

wishes) slowly over time Common vesting periods are in the three- tove-year range and options usually vest linearly (eg a four-year op-tion vests at 25 percent at the end of each year) An executive typicallyloses any unvested options upon departure Although options may beexercised as soon as they vest they do not have to be exercised untilthey expire or mature Almost 85 percent of stock-option plans have aterm of exactly ten years with virtually all of the remainder being in theve- to ten-year range About 95 percent of options are granted at themoney or at fair market value which means that the exercise price at grantdate is set equal to the stock price at grant date The remaining 5 percentare either discount options (so-called in-the-money options where theexercise price is below the stock price at grant date) or premium options(so-called out-of-the-money options where the exercise price is abovethe stock price at grant date) The holders of options typically do nothave dividend rights or voting rights even on vested options3

31 Tax Rules and Stock-Option CompensationUnlike salary and bonus stock-option grants are typically an untaxedevent at the time of grant For the most widely used optionsmdashnonquali-ed stock options (NQSOs)mdashexecutives are taxed at the personal incometax rate on option prots (the difference between that stock price and theexercise price times the number of options) when the options are exer-cised The company receives a parallel deduction against corporate in-come at that point If the executive continues to hold the shares afterexercise any subsequent appreciation is taxed at the capital gains rate inthe usual way In 1993 an additional feature was added to the tax code[Internal Revenue Code section 162(M)] that disallowed a corporate de-duction for any executive pay above $1 million that is not performance-based While this rule affects executive salaries most bonuses qualify asperformance-based and standard stock options automatically qualifyTherefore this provision gives companies with highly paid executives anincentive to give more pay in the form of bonuses and stock options asubject we return to in section 7 A summary description of the tax (andaccounting) treatment of cash and option compensation is in Table 1

A far less common type of option which is estimated to account forabout 5 percent of option grants is the incentive stock option (ISO) WhileISOs are similar to NQSOs in their design they are crucially different intwo respects First they have an annual cap of $100000 per executive

3 See Murphy (1999) for details about stock options and Miller and Scholes (1982) on taxincentives

TA

BL

E1

Tax

and

Acc

ount

ing

Tre

atm

ent

ofE

xecu

tive

Com

pens

atio

nC

ash

Pay

men

tsve

rsus

Stan

dard

Exe

cuti

veO

ptio

ns

Tax

atio

nat

Tax

atio

nat

Acc

ount

ing

Com

pens

atio

nD

escr

ipti

onp

erso

nall

evel

corp

orat

ele

vel

trea

tmen

t

Sala

ryan

dbo

nu

sC

ash

pay

men

tsm

ade

Tax

edas

ord

inar

yFu

llyd

edu

ctib

leS

ince

Exp

ense

dag

ains

tea

rn-

toC

EO

in

com

e19

93a

ny

pay

over

$1in

gsin

the

usu

alw

ay

mill

ion

isno

td

educ

t-ib

leu

nles

sit

isp

er-

form

ance

-bas

ed

Non

-qua

lie

dst

ock

Op

tion

sgr

ant

the

ex-

The

gran

ting

ofan

op-

Apa

ralle

ltax

ded

uc-

The

reis

no

exp

ense

rec-

opti

ons

(NQ

SOs)

ecut

ive

the

righ

tbu

tti

onis

typ

ical

lyan

un-

tion

atth

eco

rpor

ate

ogni

tion

(atg

ran

tex

er-

not

the

obli

gati

ont

ota

xed

even

tU

pon

exer

-le

veli

sge

nera

llyal

-ci

seo

rsa

le)

for

op-

buy

shar

esof

stoc

kat

cise

opt

ion

pro

ts

are

low

edu

pon

exer

cise

tion

sif

as

isu

sual

lyan

exer

cise

pric

e(t

ypi-

taxe

dat

the

ord

inar

y(f

orth

eam

ount

ofop

-th

eca

set

he

opti

ons

cally

the

curr

ent

stoc

kra

teI

fth

eex

ecu

tive

tion

pro

ts

)O

pti

ons

are

not

dis

cou

nted

(in

pri

ceat

the

dat

eof

cont

inu

esto

hold

onto

auto

mat

ical

lyqu

alif

yas

the

mon

eyat

gran

tgr

ant)

Th

eop

tion

sth

esh

ares

the

exec

u-

perf

orm

ance

-bas

edp

ayd

ate)

and

the

exer

cise

typ

ical

lyve

stov

era

3ndashti

veis

taxe

din

the

and

ther

efor

ear

eno

tpr

ice

and

num

ber

ofop

-5-

year

per

iod

The

typ

i-us

ualw

ay(a

tth

eca

pi-

subj

ect

toth

e$1

-ti

ons

iskn

own

atgr

ant

calm

atu

rity

is10

year

sta

lgai

nsra

te)

onan

ym

illio

nca

p

dat

eFo

rd

isco

unt

op-

but

the

man

ager

may

furt

her

app

reci

atio

nof

tion

sth

ed

iffe

renc

ebe

-ex

erci

seea

rly

Som

eth

esh

are

pri

ce

twee

nth

est

ock

pri

ceop

tion

sar

edi

scou

ntop

-an

dth

em

arke

tp

rice

isti

ons

(the

exer

cise

pri

ceex

pen

sed

over

the

vest

-is

low

erth

anth

est

ock

ing

per

iod

For

opti

ons

pri

ceat

the

gran

td

ate)

w

ith

vari

able

term

sP

rem

ium

opti

ons

are

(eg

a

vari

able

exer

-th

ere

vers

e(t

heex

er-

cise

pri

cev

esti

ngth

atci

sep

rice

isgr

eate

ris

tied

top

erfo

rman

ce)

than

the

stoc

kp

rice

atth

eop

tion

sar

em

arke

dgr

ant

dat

e)

tom

arke

tan

dex

-p

ense

dd

uri

ngth

eti

me

betw

een

gran

tan

dex

erci

se

Not

eW

eth

ank

com

pen

sati

onco

nsu

ltan

tsSc

ott

Gre

enbe

rgan

dSc

ott

Ols

en(b

oth

ofT

ower

sP

erri

n)

and

Fred

eric

Coo

k(o

fFr

eder

icW

C

ook

and

Ass

ocia

tes)

for

help

fulc

onve

rsat

ion

sin

und

erst

and

ing

the

tax

and

acco

unt

ing

rule

s

10 Hall amp Liebman

Second the tax treatment of ISOs is different ISOs are completely un-taxed at grant or exercise It is only at sale that the executive is taxed andeven then the executive is taxed at the lower capital gains tax rate Thedisadvantage is that the corporation never gets to take a parallel taxdeduction against corporate prots Thus ISOs become more attractiveas the personal tax rate increases and as the corporate tax rate and thecapital gains tax rates fall

Two other related types of compensation restricted stock and stockappreciation rights are worth brief discussion Both however are farless common than standard stock options Restricted stock is payment inthe form of restricted shares the restriction being that the shares vestover time as with options Unlike options the shares typically havevoting and dividend rights With regard to taxation the executive istaxed at the personal rate on the value of the restricted stock as thevesting restrictions lapse However the executive may choose to betaxed at the grant date in which case all subsequent appreciation istaxed at the capital gains rate The dividends paid to the executive aretaxed at the ordinary rate in the usual way The company generallyreceives a parallel deduction equal to the amount of the executiversquosincome when the executive is taxed Unlike stock options restrictedstock is not generally considered to be performance-based and is there-fore subject to the million-dollar rule (unless the vesting of the restrictedstock is performance-based which is sometimes the case)

Stock appreciation rights (SARs) are rights that replicate the payoffs ofstock options with a cash transfer Thus SARs are simpler than options inthat there is no requirement to buy and resell the stock in order to ldquocashoutrdquo SARs generally have the same tax treatment as NQSOs both to theindividual and to the corporation SARs are relatively rare however be-cause they have disadvantageous accounting treatment (described in thenext section) relative to stock options and their main relative advantagevis-a-vis options has been essentially replicated through broker-assistedcashless option exercisesmdashwhereby a broker makes a ldquonanosecondrdquo loanto the executive (to purchase and resell the stock) when the executivewants to ldquocash inrdquo on option prots The tax and accounting treatment ofISOs restricted stock and SARs is summarized in Table 2

32 The Accounting Treatment of OptionsUnlike cash compensation which is expensed against earnings there isgenerally no expense recognition (at grant exercise or sale) for optionswhether they be NQSOs or ISOs As a result compensation consultantsoften point out that stock options are the only form of compensation thatare free in an accounting sense but still deductible for tax purposes

Taxation of Executive Compensation 11

Stock options do however lead to expense recognition if they are dis-counted (in the money at grant date) or if the exercise price and numberof options are not known at grant date For discount options the differ-ence between the stock price and the market price is expensed over thevesting period For options with variable terms (eg a variable exerciseprice vesting that is tied to performance) the options are marked tomarket and expensed during the time between grant and exercise Practi-tioners claim that the accounting treatment of options plays an impor-tant role in the design of option programs Thus plans that have ldquobadaccountingrdquo but are thought by many to have attractive incentive fea-tures are often not even seriously considered by companies Examples ofsuch potentially attractive plans include indexed options (where a CEOprots only if his rmrsquos share price grows relative to some market orindustry benchmark) and option grants that are explicitly performance-related both of which would lead to expenses against earnings

Unlike options restricted stock and SARs do not generally receivefavorable accounting treatment Restricted stock is generally expensedover the period in which the restrictions lapse (usually the vesting pe-riod) The magnitude of the expense is the difference between the cur-rent stock price and the executiversquos cost (if any) SARs are marked tomarket each period and the difference between the stock price and theexercise price is expensed over the outstanding period of SARs

4 THE TAXATION OF EXECUTIVE PAY CASHVERSUS OPTIONS

Stock-based compensation performs two roles in executive compensa-tion arrangements First it helps align the incentives of the executivewith the interests of the rmrsquos shareholders Second it often enables therm to compensate the CEO in a way that is more advantageous from atax standpoint than paying the executive in salary and bonus

41 Agency Theory and Executive Compensation in the Presenceof TaxationIn standard agency-theory models ( Jensen and Meckling 1976) agencycosts are the result of the separation of ownership and control Managersdo not have the same incentives as the owners The optimal incentivecontract for managers balances the benets of high-powered incentives(linking the fortunes of owners and managers through stock and stockoptions for example) with the costs of loading too much risk on risk-averse managers

TA

BL

E2

Tax

and

Acc

ount

ing

Tre

atm

ent

ofE

xecu

tive

Com

pens

atio

nIn

cent

ive

Stoc

kO

ptio

ns(I

SOs)

Res

tric

ted

Stoc

kan

dSt

ock

App

reci

atio

nR

ight

s(S

AR

s)

Tax

atio

nat

Tax

atio

nat

Acc

ount

ing

Com

pen

sati

onD

escr

ipti

onp

erso

nall

evel

corp

orat

ele

vel

trea

tmen

t

Ince

nti

vest

ock

ISO

sar

eid

enti

calt

oIf

the

usu

alco

ndit

ions

Ifth

eu

sual

cond

itio

nsA

sw

ith

NQ

SOs

ther

eop

tion

sN

QSO

sin

thei

rd

esig

n

are

met

th

ere

isno

tax

are

met

the

reis

gene

r-is

gene

rally

noex

pen

seH

owev

ert

hey

have

anat

gran

tor

exer

cise

At

ally

no

corp

orat

eta

xd

e-re

cogn

itio

nfo

rIS

Os

annu

alca

pof

$100

000

sale

all

opti

onp

ro

tsd

uct

ion

atgr

ant

exer

-(e

ver)

on

the

amou

ntth

atan

dsu

bseq

uen

tca

pita

lci

seo

rsa

le

vest

sT

he

tax

trea

t-ap

prec

iati

onar

eta

xed

men

tof

ISO

sal

soat

the

cap

ital

gain

sd

iffe

rs

rate

Res

tric

ted

stoc

kSi

mila

rto

stoc

kop

-T

hein

div

idu

alis

taxe

dT

heco

mp

any

gene

rall

yR

estr

icte

dst

ock

isge

n-ti

ons

but

shar

esat

the

ord

inar

yra

teas

rece

ives

ap

aral

leld

e-er

ally

exp

ense

dov

er(w

hich

vest

slow

ly)

the

rest

rict

ions

lap

se

du

ctio

neq

ual

toth

eth

eve

stin

gp

erio

dT

hera

ther

than

opti

ons

are

How

ever

the

man

ager

amou

ntof

the

exec

u-

amou

nt

exp

ense

dis

gran

ted

toth

eex

ecu

-m

aych

ose

tobe

taxe

dti

versquos

inco

me

whe

nth

eth

ed

iffe

renc

ebe

twee

nti

veU

nlik

eop

tion

sat

gran

td

ate

inw

hich

exec

uti

veis

taxe

dR

e-th

ecu

rren

tst

ock

pric

eth

esh

ares

typi

cally

case

alls

ubs

equ

ent

ap-

stri

cted

stoc

kis

gen

er-

and

the

exec

uti

versquos

have

voti

ngan

dd

ivi-

pre

ciat

ion

ista

xed

atal

lysu

bjec

tto

the

cost

(if

any)

d

end

righ

ts

the

cap

ital

gain

sra

te

mill

ion-

dol

lar

cap

D

ivid

end

sar

eta

xed

atth

eor

din

ary

rate

Stoc

kap

pre

ciat

ion

Rig

hts

that

repl

icat

eG

ener

ally

have

the

Gen

eral

lyha

veth

eU

nlik

eop

tion

sSA

Rs

righ

tsth

epr

ots

ofst

ock

op-

sam

eta

xtr

eatm

ent

assa

me

tax

trea

tmen

tas

are

mar

ked

tom

arke

tti

ons

Rar

ely

use

d

NQ

SOs

NQ

SOs

each

per

iod

an

dth

esi

nce

stoc

kop

tion

sd

iffe

renc

ebe

twee

nth

eha

vem

ore

favo

rabl

eac

-st

ock

pri

cean

dth

eex

-co

unt

ing

trea

tmen

ter

cise

pri

ceis

exp

ense

d(a

ndop

tion

sp

rovi

de

over

the

outs

tand

ing

the

sam

ebe

ne

tof

per

iod

ofSA

Rs

cash

less

exer

cise

thro

ugh

brok

erlo

ans

)

Not

eW

eth

ank

com

pen

sati

onco

nsul

tant

sSc

ott

Gre

enbe

rgan

dSc

ott

Ols

en(b

oth

ofT

ower

sP

erri

n)an

dFr

eder

icC

ook

(of

Fred

eric

WC

ook

and

Ass

ocia

tes)

for

hel

pfu

lcon

vers

atio

nsin

und

erst

and

ing

the

tax

and

acco

unt

ing

rule

s

14 Hall amp Liebman

The effect of taxation on the optimal contract (more precisely theshare of compensation that is performance-related) is ambiguous (evenin the absence of deductibility and deferral) because there are offsettingeffects First by reducing the share of corporate prots received byshareholders taxes diminish the importance to the shareholders of moti-vating the CEO Taken alone this effect would imply that taxes wouldbe expected to reduce the use of performance-based compensation Sec-ond because the government shares in the income received by the CEOthe variance of the CEOrsquos income is reduced raising his utility andlowering the cost to the rm of providing any given set of incentivesThird because taxes will lead executives to provide less effort for anygiven level of incentive-based pay the level of compensation that mustbe provided to compensate the CEO for effort is reduced for a givenamount of incentive These last two factors make it cheaper for the rmto offer contracts to the CEO with large incentive components when taxrates are higher and should therefore increase the use of incentive-based pay Since the net effect of taxation on the level of incentive-basedpay is ambiguous agency theory provides no strong predictions abouthow taxation should affect the optimal composition of option versuscash compensation

42 The Tax Advantages of DeferralWhile agency theory yields ambiguous predictions about the effect oftaxes on the use of options there are direct tax advantages of options inthat options provide executives with a way to defer compensation andthereby lower their taxes However because options also lead to a defer-ral of corporate tax deductibility the tax advantages from a combined(executive and corporate) perspective are not so straightforward In thissection we analyze and measure the tax advantages of stock optionsrelative to salary and bonus compensation We also analyze the taxadvantages of NQSOs relative to ISOs In particular we show how thetax advantages of options change as personal corporate and capitalgains taxes change We then show how the tax advantages of optionshave changed over time in response to changing tax rates

The crucial tax difference between standard options (NQSOs) andcash payment is that option payouts are deferred and the two forms ofcompensation earn different rates of return over the deferral period Anyanalysis of the relative tax advantages of two compensation instrumentsmust consider the tax consequences both to the employer and to theemployeemdashwhat Scholes and Wolfson (1992) call the global contractingperspective Thus in order to make valid comparisons between the twowe compare the tax burden to the executive while holding constant the

Taxation of Executive Compensation 15

posttax cost to the company By keeping posttax employer costs (in NPVterms) constant any package that is preferred by the employee is tax-advantaged in the global-contracting sense

Under this methodology a comparison of the tax advantages of op-tions and cash involves comparing a pretax cash payment of P with anoption payment that has the equivalent posttax NPV to the companyFor NQSOs it is straightforward to show that a pretax payment of P toan executive has exactly the same cost to the company as putting aside Pfor the purposes of paying stock-option payouts later That is if a com-pany pays an executive P today it will have the same amount of moneyin N years as if it had put aside P today let it accumulate at the rate ofreturn earned by the rm and then paid it out (as compensation in theform of option gains4) with any appreciation in year N

Assume that the pretax prot rate is r and the corporate tax rate is TcThen if the company pays P today it receives a deduction today of PTcwhich yields

PTc [1 1 r (1 2 Tc)]N (1)

in N years since the benets of the tax deduction are invested in thecompany and receive the after-tax corporate rate of return Converselyif the company puts P aside today then it grows at the after-tax corpo-rate rate of return in N years to give a tax deduction of

P [1 1 r (1 2 Tc)]N Tc (2)

which is the exact same value Note that the equivalence of these two taxdeductions is analogous to the tax-benet equivalence of front-loadedand back-loaded IRAs

An executiversquos payoff in N years from option prots is equal to Pcompounded at the after-tax corporate rate times 1 2 Tp where Tp is thepersonal tax rate Thus the combined payoffs in N years of the corporatededuction and the executiversquos posttax payoff is

P [1 1 r (1 2 Tc)]N Tc 1 P [1 1 r (1 2 Tc)]

N (1 2 Tp) (3)

where the rst term is the payoff from the corporate deduction and thesecond term is the executiversquos payoff

4 This analysis is not unique to options Any form of deferred compensation that enablesexecutives to invest inside the rm at a higher posttax return or without paying capitalgains taxes would have similar effects

16 Hall amp Liebman

The payoff to salary and bonus in N years is more complicated sincean assumption must be made about how the executive invests the origi-nal cash compensation For example if the executive invests in an instru-ment with taxable interest (eg bonds) then the N-year payoff from aninvestment of P accumulates at the post-corporate-tax post-personal-taxrate of return and is

P (1 2 Tp) [1 1 r (1 2 Tc)(1 2 Tp)]N (4)

However if an executive invests in (non-dividend-paying) equities theinvestment accumulates at the higher post-corporate-tax rate of returnwith the offsetting disadvantage that the capital gains are taxed at thecapital gains rate in year N So the executive receives

P [1 1 r (1 2 Tc)]N (1 2 Tp) 2 Tcg $P (1 2 Tp) [1 1 r (1 2 Tc)]

N 2P (1 2 Tp) (5)

where Tcg is the capital gains tax rate5Thus putting the equations together if an executive is paid cash P

and invests g in bonds and 1 2 g in equities the N-year payoff (includ-ing the payoff to the corporation from the tax deduction) is

P [1 1 r (1 2 Tc)]N Tc 1 g $(P [1 1 r (1 2 Tc)(1 2 Tp)]

N (1 2 Tp)1(1 2 g) (P [1 1 r (1 2 Tc)]

N (1 2 Tp) 2 Tcg $P (1 2 Tp)[1 1 r (1 2 Tc)]

N 2 P (1 2 Tp)) (6)

The tax advantage of options versus salary and bonus therefore is thedifference between the combined (corporate and executive) payoffs inequation (3) and the combined payoffs in equation (6)

43 The Tax Advantage of OptionsWe now turn to analysis of how changes in various tax rates affectthe tax advantages of options relative to cash compensation We focuson the case in which the cash earned by executives is invested entirelyin equities since equities are tax-favored relative to bonds The impor-tant conclusions of this analysis are not substantively different if we

5 In practice executives hold a combination of equities and bondlike instruments withtaxable interest Because equities are tax-favored in the next section we simplify the analy-sis by assuming that executives hold only equities Alternatively we could have assumedthat executives hold only less risky bonds since bonds more closely match the risk proleof a stream of tax savings None of our empirical results are substantively affected by thissimplication

Taxation of Executive Compensation 17

instead assume that a portion of an executiversquos holdings are invested inbonds

431 The Tax Advantage of Options Changing the Corporate TaxRate Stock options are tax-advantaged relative to cash simply becauseoption payouts are deferred allowing the executive to invest at the pre-personal-tax rate of return (with no capital gains tax at the end) ratherthan investing at the after-tax rate of return or paying the capital gainstax Since the advantage of being able to defer taxes is large when thecorporate tax rate is low the tax advantage of options is larger when thecorporate rate is low

To illustrate this point we dene the tax advantage of options to bethe dollar amount by which total after-tax option payoffs (to the corpora-tion and the executive) exceed the total payoffs from salary as denedby the difference of equation (3) and (6) assuming that P (the paymentto the executive) is equal to $100 and g 5 0 (cash compensation is in-vested in equities) We use a ten-year horizon since most options have aten-year term We then calculate how the tax advantage of optionschanges as the corporate tax rate varies from zero to 100 percent holdingthe personal tax rate and the capital gains rate xed at 40 percent and 20percent respectively (which approximates current rates)

The top panel of Figure 2 shows the tax advantage of options relativeto cashmdashprecisely the NPV of the total tax advantage of options for a$100 payment to the CEOmdashas the corporate tax rate changes As ex-pected the tax advantage of options declines as corporate tax rates in-crease At a zero corporate tax rate the tax advantage of options has anNPV of $7 This value is about $350 at a corporate rate of 40 percentWhen the corporate tax rate is 100 percent the tax advantage of optionscompletely disappears since the benets of deferred compensation fallto zero because the after-tax return on equities (and therefore the dis-count rate) falls to zero The tax advantage of options is the NPV of thetax savings from avoiding the capital gains tax on ten years of apprecia-tion which is why it is relatively modest even at the 40-percent corpo-rate rate

432 The Tax Advantage of Options Changing the Personal TaxRate Using the same assumptions as above (but this time holding thecorporate rate at 40 percent and the capital gains rate at 20 percent andvarying the personal rate) the tax advantage of options declines as thepersonal tax rate rises as shown in the middle panel of Figure 2 The taxadvantage of options is higher at low personal tax rates because theposttax base that gives rise to capital gains is higher and the advantage of

Note Holds constant the personal rate at 40 percent and the capital gains rate at 20 percent

Note Holds constant the corporate rate at 40 percent and the capital gains rate at 20 percent

Note Holds constant the personal rate at 40 percent and the corporate rate at 40 percent

FIGURE 2 Tax Advantage of Options NPV of Tax Advantage of Pay-ing $100 in Options (NQSOs) Rather Than in Cash

Taxation of Executive Compensation 19

avoiding the capital gains taxes is greater The tax advantage of optionstherefore declines linearly as the personal tax rate rises At a 100-percentpersonal tax rate the executive receives nothing in either case and byconstruction the companyrsquos tax deduction benet is the same in NPVterms so the tax advantage of options falls to zero also6

433 The Tax Advantage of Options Changing the Capital Gains TaxRate The tax advantage of options for various capital gains rates isdepicted in the bottom panel of Figure 2 under the same assumptions(this time holding both the personal rate and the corporate rate at 40percent while varying the capital gains rate) The tax advantage of avoid-ing the capital gains tax increases as the capital gains tax rate rises

5 HOW HAS THE TAX ADVANTAGE OF OPTIONSCHANGED OVER TIME

The analysis so far has illustrated how the tax advantage of optionsvaries with the changes in tax rates In this section we show how the taxadvantage of options has changed over time as tax rates have changed

51 Top Marginal Tax Rates over TimeThe evolution on the top marginal tax ratesmdashpersonal corporate andcapital gainsmdashfrom 1980 to 1998 is shown in Figure 3 In all cases thetop rate (the rate for taxpayers with the highest incomes) is shownwhich is not always the highest rate since various anomalies (such asclawbacks of exemptions) sometimes lead to marginal tax rates that arehigher than those for the highest income tax payers

The top personal tax rate was 70 percent in 1980 falling to 50 percentin 1982 and 28 percent in 1988 following the 1986 tax act7 The topmarginal tax rate has since risen to 396 percent but is effectively 425percent since there is a 29-percent Medicare surcharge (paid half by theemployer and half by the employee) that has no upper limit The topcorporate tax rate has had only one signicant change since 1980 fallingfrom 46 percent before the 1986 tax act to 34 percent following the act in1988 The top corporate tax rate was increased to 35 percent in 1993 The

6 This is one place where ignoring bond investments has substantive implications Inparticular in the all-bond case the tax advantage of options is nonmonotonicmdashit rises andthen fallsmdashas personal tax rates rise7 During the early 1980s the maximum tax on earned income limited the marginal tax rateon the earnings of high earners so many executives did not face a decline in personal taxrates between 1981 and 1982 The empirical results in this paper are robust to assumingthat executives faced a marginal tax rate of 50 percent in 1980 and 1981

20 Hall amp Liebman

FIGURE 3 Top Marginal Tax Rates since 1980

top capital gains rate has uctuated between 28 percent (approximately)and 20 percent since 1980 and is currently at 20 percent

52 The Evolution of the Tax Advantage of NQSOsThe tax advantage of options relative to cash dened in the same way asin section 4 from 1980 to 1998 is shown in Figure 4 The calculations aredone using the same assumptions as before (ie N 5 10 years and cashis invested in equities) and the statutory top marginal tax rates shown inFigure 3 are used Due to the fall in corporate and personal tax rates thetax advantage of options increased sharply after the Tax Reform Act of1986 This increase in tax advantage has largely been reversed in the1990s as top personal tax rates have crept back up and the capital gainsrate has been reduced

Two key facts emerge from this analysis First there is currently only amoderate tax advantage to standard nonqualied stock optionsmdashon theorder of $4 per $100 of compensation This is because the tax advantagesto the executive of deferring taxes are largely offset by the tax disad-vantages to the company of not being able to deduct option expensesfrom taxable prots until the executive exercises the options Secondalthough the 1986 tax act substantially increased the tax advantage ofoptions more than half of this increase has been reversed in the 1990s

Taxation of Executive Compensation 21

FIGURE 4 Tax Advantage of Options 1980 to 1998

NPV of tax advantage of $100 in options (NQSOs) rather than cash

53 The Tax (Dis)advantage of ISOsSo far the analysis has focused on the tax advantages of NQSOs sincethey are so much more prevalent than ISOs which account for onlyabout 5 percent of option grants The relative scarcity of ISOs can beexplained both by their tax status and by their per-executive cap of$100000 per year

Under what conditions are ISOs tax-preferred to NQSOs For a trans-fer P to an executive the ISO is always tax-preferred by the executive(since the capital gains rate is lower than the personal rate) while theNQSO is always tax-preferred from the companyrsquos perspective (since anISO is not deductible) The key issue however is the relative advan-tages of the two types of options from a global-contracting perspectiveBecause ISOs are not deductible the company is indifferent betweensetting aside P in the form of NQSOs today and setting aside P(1 2 Tc) inthe form of ISOs (Since the NPV of the tax deduction for NQSOs is PTcthe NPV of the payment P is P 2 PTc or P (1 2 Tc))

Equalizing the posttax cost of NQSOs and ISOs we need only look atthe payoff of the executive to determine the condition under which eachis preferred The payoff of NQSOs in N years is

P [1 1 r (1 2 Tc)]N (1 2 Tp) (7)

22 Hall amp Liebman

The same payoff for ISOs is

P (1 2 Tc) [1 1 r (1 2 Tc)]N (1 2 Tcg) (8)

ISOs therefore are tax-advantaged if

Tp 2 Tcg Tc 2 Tc Tcg (9)

Intuitively ISOs are tax-advantaged only if their advantage (the differ-ence between the personal rate and the capital gains rate) is largeenough to offset their cost (the disadvantage of not deducting at thecorporate rate) Note however that the condition is not a simple com-parison between the corporate rate and the personal rate minus thecapital gains rate (Tp 2 Tcg Tc)

Figure 5 shows how the tax advantage of NQSOs relative to ISOs haschanged since 1980 as tax rates have changed Unlike the very modestchanges (around 3 to 4 percent) in the relative tax advantages betweenoptions and cash shown in Figure 4 the relative tax advantage betweenNQSOs and ISOs has seen enormous swings When the top personalrate was 70 percent in 1980 ISOs were tax-favored by a margin greaterthan 6 percent However by 1982 (following the 1981 tax acts that dra-matically lowered the personal tax rate) ISOs became tax-disadvantaged

FIGURE 5 Tax Advantage of NQSOs Relative to ISOs over Time

NPV of tax advantage of $100 payment to executive NQSOs minus ISOs

Taxation of Executive Compensation 23

by about 5 percent By 1988 (following the 1986 tax act) ISOs became tax-disadvantaged by about 18 percent a dramatic swing The raising ofpersonal rates coupled with a decrease in the capital gains rate has sincereduced the tax disadvantage of ISOs to about $4 per $100 of pretaxcompensation Although hard data on ISOs are hard to come by practi-tioners (mostly compensation consultants) claim that ISOs were morepopular prior to 1982 when they were tax-advantaged Although ISOsare less disadvantageous from a tax standpoint relative to NQSOs thanthey were in the late 1980s they are still disadvantageous so it is notsurprising that they have not made a signicant resurgence in recentyears

6 EXPLAINING THE INCREASE IN OPTION-BASEDCOMPENSATION

In this section we analyze whether changes in tax rates affect the compo-sition of executive compensation and consider the relative importanceof tax factors and corporate governance factors in explaining the in-creasing share of compensation paid in the form of stock options Thenumerous changes in tax rates that have occurred since 1980 provide anopportunity to assess whether executive compensation arrangementsrespond to tax incentives in the way that the tax-avoidance model sug-gests We test this model using a panel data set of CEOs in largepublicly-traded US corporations Our identication of the tax effectsrelies on time-series variation in personal corporate and capital gainstax rates along with cross-sectional variation in corporate tax rates

61 DataWe use a 15-year panel data set of CEOs in the largest publicly-tradedUS corporations which is described in Hall and Liebman (1998) Thedata set covers the years from 1980 through 1994 and combines CEOcompensation information from corporate proxies and 10-K lings withstock price and stock return information from CRSP and accounting datafrom Compustat In addition some compensation data from the 1970swere collected in order to construct measures of the value of stock op-tions held by the CEOs in the rst period of the sample8

The feature that distinguishes our data from most other CEO datasets is that with our panel of yearly proxy data on option grants optiongains and total options held we are able to calculate the total value of

8 We are grateful to David Yermack for providing us with some of the data for 1984 to 1991See Yermack (1995) for a discussion of these data

24 Hall amp Liebman

all stock options held by the CEO at a point of time More importantlysince we have the details about the stock options held (number exer-cise price time to maturity etc) we can precisely calculate the changein the value of a CEOrsquos stock option holdings for a given change in rmvalue

62 Identication IssuesThe numerous tax changes over our sample period and the specic waysin which changes in personal corporate and capital gains rates arepredicted to affect the composition of CEO compensation give us un-usually rich sources of identication However because many of thechanges in tax rates would be expected to affect all of the CEOs in oursample in a similar way we will need to pay particular attention toseparating out the tax effects from underlying time trends and otherfactors that changed over time Controlling for underlying trends is par-ticularly important in this study because we are focusing on an out-comemdashthe use of stock optionsmdashthat has increased rapidly over thepast two decades and that many practitioners believe was importantlyaffected by nontax considerations Thus the tax effects we are trying toexplain are deviations from a rising trend

We take two steps to try to separate out the tax effects from theunderlying trend First we include key nontax factors that could poten-tially explain the increasing reliance on performance-based compensa-tion It has been suggested that the dramatic increase in incentive-basedpay is the result of the remarkably poor shareholder returns during the1970s which spurred the LBO and takeover movements of the 1980sAccording to this story (Kaplan 1997) shareholders became much morepowerful via the rise of institutional investors even as the LBO andtakeover movements waned Because the inuence of institutional inves-tors is thought to be one of the most important mechanisms of strongcorporate governance we use the share of each companyrsquos stock ownedby institutional investors (dened as institutions with more than $100million under management) as an explanatory variable9 The percentageof shares owned by large institutional investors increased from about 20percent to almost 50 percent in 1994 an upward trend that closelymatches the sharp rise in the share of compensation in the form ofoptions Indeed in our sample the annual average (over all of the rmsin our sample) share of stock owned by institutional investors has acorrelation of 09 with the annual average share of compensation paid inoptions

9 This variable is described in Gompers and Metrick (1998)

Taxation of Executive Compensation 25

In addition we include two other variables that proxy for strongercorporate governance The rst is the size of the board Evidence suggeststhat boards with smaller numbers of directors reduce the inuence of theCEO and better represent shareholder interests (Yermack 1995) Boardsize fell by approximately 15 percent between 1980 and 1994 The secondvariable is the fraction of the rmrsquos board members who are inside direc-tors or gray directors (noninsiders who have business dealings with thecompany) Firms with fewer outsiders are less likely to act in the share-holdersrsquo interests and we would predict that they would be less likely tohave performance-based compensation for their CEOs The average per-centage of inside and gray directors has decreased from about 45 percentin 1980 to about 35 percent in 1994

Second we allow for cross-sectional variation in corporate tax ratesCorporate marginal tax rates vary because rms may be eligible for tax-loss carrybacks and carryforwards investment tax credits and the alter-native minimum tax (see Auerbach 1986 Auerbach and Altshuler1990 Auerbach and Poterba 1987) Because these cross-sectional differ-ences imply that different rms should respond differently to a givenchange in tax rates they potentially provide us with a way to isolatethe tax effects Firm-specic corporate marginal tax rates are notori-ously difcult to calculate from publicly-available data We use a tricho-tomous variable that equals zero if the rm has tax-loss carryforwardsand negative earnings in a given year equals 05 times the statutorycorporate rate if the rm has only one of those conditions and equalsthe statutory rate if the rm has neither loss carryforwards nor negativeearnings10

The cross-sectional variation in the corporate tax rate may not be ex-ogenous Firms that perform poorly could face low marginal tax ratesand also be particularly likely or unlikely to provide performance-basedpay Because most of our specications use xed effects the correlationwe must be concerned about is between changes in rm performanceand changes in compensation We deal with this by controlling forlagged rm stock market performance in our regressions This shouldeliminate spurious correlation between tax rates and compensation thatis jointly caused by rm performance

10 Graham (1996) presents evidence that the trichotomous variable performs nearly as wellas his recommended simulated marginal tax rate Plesko (1999) compares marginal taxrates calculated from Compustat data with actual marginal tax rates from rm tax returnsHe nds that simple binary variables are more reliable measures of the marginal tax ratethan is either the Graham or the trichotomous variable We obtain substantively similarresults when we switch from the trichotomous variable to the binary variables recommendby Plesko

26 Hall amp Liebman

63 ResultsThe dependent variable in our regressions is the fraction of a CEOrsquos totalannual compensation (measured as the BlackndashScholes value of stock-option grants plus salary and bonus) that comes from stock-optiongrants We choose this reduced-form specication because our interest isin the composition of CEO compensation However the factors thataffect the composition of compensation might also affect the levels andit will be important to keep this in mind in interpreting our results

Optimal-contracting theory generally makes predictions about the in-centives provided by the total amount of rm stock and stock optionsowned by the CEO We choose to model the ow of option grants ratherthan the ultimate performance sensitivity of the CEOrsquos compensationand stock and stock-option holdings because we have in mind anadjustment-cost model in which it takes time for a CEOrsquos holdings ofstock and stock options to reach the optimal level11 In our regressionswe include two measures of the current performance sensitivity of theCEOrsquos wealth to rm performance as explanatory variables to reect thedistance of the CEOrsquos contract from the optimal contract

Table 3 presents results in which the tax effects are identied solely byvariation in tax rates over time The fraction of annual compensation paidin stock options is regressed on tax variables corporate governance vari-ables contract-theory variables and variables reecting stock market re-turns Both of the regressions in this table include rm xed effects Incolumn (1) the fraction in annual compensation paid in stock options isregressed on the log difference in payoff (the combined payoff to both therm and the CEO) from receiving compensation in stock options ratherthan salary and bonus where the payoff calculations are based on statu-tory marginal tax rates This variable is dened as the log difference be-tween equation (3) and equation (6) in section 4 The coefcient on the taxvariable is large and statistically signicant at the 95-percent level Thepoint estimate implies that a one-percent increase in the payoff differencebetween stock options and salary and bonus results in a 24-percentage-point increase in the share of compensation paid in stock options

The corporate governance variables all have the predicted sign Firmswith a higher fraction of their shares owned by institutional investorsuse more performance-based pay (stock options) Firms with large corpo-rate boards or a large share of inside directors are less likely to use stockoptions

The two contract-theory variables are the dollar change in CEO wealth

11 See Core and Guay (1999) for evidence that rms adjust in this way

TABLE 3Compensation Regressions that Rely on Variation in

Tax Rates over Time

(1) (2)

Tax variablesln (difference in payoff)(a) 2393 (0349)ln (12statutory personal 20015 (0034)

rate)ln (12statutory cap gains 0330 (0171)

rate)ln (12statutory corporate 0399 (0148)

rate)

Corporate governance variablesFraction of rmrsquos shares 0043 (029) 0052 (0028)

owned by institutionalinvestors

Size of the rmrsquos board of 20002 (001) 20003 (0001)directors

Fraction of directors who 20082 (033) 20092 (0030)are inside or gray

Contract-theory variablesD (CEO wealth) per $1000 20261 (0095) 20247 (0087)

in FV(b)

D (CEO wealth) per 10 in 2583 (189) 2554 (201)FV(c)

Volatility(d) 521 (424) 261 (369)ln (rm market value) 0019 (0006) 0019 (0006)

Stock market return variablesLagged annual rm stock 0006 (0009) 0003 (0008)

market returnLagged annual SampP 500 re- 20035 (0023) 20029 (0023)

turn

Regression diagnosticsTotal sample size 5179 5179Firms 427 427Adjusted R2 0313 0311

Dependent variable share of compensation paid in options Standard errors in parentheses Bothregressions in this table include rm xed effects

(a) Log difference in payoff is the log difference in combined rm and CEO after-tax share betweenpaying compensation in options and paying salary

(b) This variable is the dollar change in CEO wealth per $1000 change in rm market value (seeJensen and Murphy 1990)

(c) This variable in the dollar charge in CEO wealth per 10-percent change in rm value (see Hall andLiebman 1998)

(d) Standard deviation of rm market value in trillions

28 Hall amp Liebman

per $1000 change in rm value (Jensen and Murphy 1990) and thedollar change in CEO wealth per 10-percent change in rm value (Halland Liebman 1998) As discussed in Baker and Hall (1998) these twowealth sensitivity measures reect two different concepts of how closelyaligned the managersrsquo incentives are with the interests of the sharehold-ers In the regression they are meant to measure how far the CEOrsquosexisting contract and ownership of stock and stock options is from theoptimal level The negative coefcients on the two variables suggeststhat rms that are below their optimal pay-to-performance sensitivityare more likely to give more stock options in the current period

The second column replaces the tax variable motivated by deferraladvantages with the three statutory rates The corporate after-tax sharehas a positive coefcient as we would expect because the tax advantageof stock options increases as the corporate rate falls The coefcient onthe personal tax rate is small and statistically insignicant The sign ofthe coefcient on the capital gains rate is the opposite of that predictedby tax-avoidance theory

Note that these results rely on time-series variation alone This is not avery convincing test Since the log difference in payoffs mostly rises overthe time period covered in our sample as does the share of compensa-tion paid in options it is possible that the tax variable is simply reectingother factors that were changing over this time period that we have notincluded in our regression Therefore it is important to look at specica-tions that rely on cross-sectional variation in the tax variables

Table 4 contains specications that rely on cross-sectional variation incorporate tax rates The rst column repeats the specication from col-umn (1) of Table 3 but uses the trichotomous measures of rm corporatetax rates in place of the statutory rate in calculating each rmrsquos logdifference in payoff from paying compensation in options rather than insalary and bonus As in the previous table we nd the positive coef-cient on the log difference in payoff just as the theory suggest How-ever the coefcient estimate is now much smaller suggesting that aone-percent increase in the log difference in payoff produces a 01-percentage-point increase in the share of compensation paid in options

Introducing cross-sectional variation in the dependent variable doesnot eliminate the potentially spurious correlation over time between thetax variable and the share of compensation paid in options The secondcolumn of Table 4 isolates the pooled cross-sectional variation in corpo-rate tax rates by adding time dummies and dropping the rm xedeffects In this specication the coefcient on the tax variable is indistin-guishable from zero The coefcient on the institutional variable contin-ues to be large and statistically signicant The third column includes

Taxation of Executive Compensation 29

both rm xed effects and time dummies which tests for a relationshipbetween within-rm changes in the dependent variable and within-rmchanges in each explanatory variable In this specication the coefcienton the tax variable is again indistinguishable from zero The coefcientson the corporate governance variables are also statistically insignicant

The third column contains a specication that treats the Tax ReformAct of 1986 (TRA 86) as a natural experiment For rms and CEOs facingthe statutory corporate tax rate TRA 86 increased the relative payoff topaying compensation in options by 17 percent For a rm facing a zeromarginal tax rate it reduced the relative payoff by 45 percent The reduc-tion occurred because TRA 86 reduced the personal tax rate makingsalary and bonus more attractive while the decreased corporate rate hadno effect on zero-tax-rate rms Thus we would expect rms facing zeroor low marginal tax rates to reduce their use of options after TRA 86while rms facing the statutory rate would be expected to increase themFew rms face zero marginal tax rates year after year Therefore we tryto distinguish between rms that often face low tax rates and rms thatnearly always face the statutory rate We dene low-tax rms as oneswhose average marginal tax rate over 1984 to 1986 was below 02412

Then we ran a difference-in-differences regression to see if high-taxrms increased their use of options more than low-tax rms The interac-tion between post-TRA 86 and high corporate tax rates is the key vari-able The estimated coefcient on this variable is positive but small andnot statistically different from zero at the 95-percent level

It is important to note that in this specication as well the main corpo-rate governance variable the share of the rmrsquos shares owned by institu-tional investors performs as theory would predict The variable appearsto be quite robust Table 5 presents regressions of the share of compensa-tion paid in stock options on the fraction of the rmrsquos shares owned byinstitutional investors Since we have already established that these vari-ables have a strong longitudinal relationship the regressions here aredesigned to focus on the cross-sectional relationship between these twovariables The rst and third columns present regressions using data onlyfrom 1994 and nd a coefcient of roughly 02 (column 1 contains noother covariates while column 3 includes all of the other nontax covari-ates from the regressions in Tables 3 and 4) Since the median share of armrsquos stock owned by institutional investors in our sample rises from 19percent in 1980 to 49 percent in 1994 this coefcient would imply that this

12 This cutoff was chosen after inspecting the distribution of tax rates and essentiallyseparates rms that are always at or near the statutory rate from ones that are sometimesbelow it

TA

BL

E4

Com

pens

atio

nR

egre

ssio

nsth

atR

ely

onC

ross

-Sec

tion

alV

aria

tion

inC

orpo

rate

Tax

Rat

es

(1)

(2)

(3)

(4)

Tax

vari

able

s(a)

ln

(dif

fere

nce

inp

ayof

f)0

099

(00

41)

20

038

(00

36)

20

005

(00

44)

Pos

t-T

RA

862

000

5(0

013

)H

igh

corp

orat

eta

xra

te0

018

(00

17)

(Pos

t-T

RA

86)

3hi

gh0

024

(00

16)

corp

orat

eta

xra

te

Cor

por

ate

gove

rnan

ceva

riab

les

Frac

tion

of

rmrsquos

shar

es0

109

(00

27)

011

4(0

015

)2

000

9(0

030

)0

087

(00

34)

owne

dby

inst

itu

tion

alin

vest

ors

Size

ofth

e

rmrsquos

boar

dof

20

004

(00

01)

20

001

(00

01)

20

002

(00

01)

000

1(0

002

)d

irec

tors

Frac

tion

ofd

irec

tors

wh

o2

012

4(0

033

)2

000

3(0

018

)2

004

1(0

034

)0

001

(00

39)

are

insi

de

orgr

ay

Con

trac

t-th

eory

vari

able

sD

(CE

Ow

ealt

h)p

er$1

000

20

238

(00

94)

20

167

(00

53)

20

257

(00

96)

20

021

(01

47)

inFV

(b)

D(C

EO

wea

lth)

per

10

in2

560

(17

8)2

592

(10

5)2

594

(19

6)2

147

(04

86)

FV(c

)

Vol

atili

ty(d

)2

48(4

24)

162

(03

4)1

12(4

28)

106

(06

09)

ln(

rmm

arke

tva

lue)

003

3(

006)

000

8(0

003

)0

009

(00

06)

001

2(0

007

)

Stoc

km

arke

tre

turn

vari

able

sL

agge

dan

nual

rm

stoc

k2

000

2(0

009

)0

011

(00

10)

000

6(0

009

)2

001

2(0

013

)m

arke

tre

turn

Lag

ged

annu

alSamp

P50

02

005

6(0

023

)2

005

3(0

051

)2

003

9(0

049

)2

015

4(0

040

)re

turn

Reg

ress

ion

dia

gnos

tics

T

otal

sam

ple

size

5179

5179

5179

2534

Firm

s42

742

739

6A

dju

sted

R2

031

009

032

005

Firm

xed

effe

cts

Yes

No

Yes

No

Tim

ed

um

mie

sN

oY

esY

esN

o

Dep

end

ent

vari

able

isfr

acti

onof

ann

ualc

omp

ensa

tion

from

stoc

kop

tion

sSt

and

ard

erro

rsin

par

enth

eses

(a)

All

thre

esp

eci

cati

ons

use

tric

hoto

mou

sco

rpor

ate

tax

rate

sT

he

sam

ple

inco

lum

n(3

)u

ses

only

the

year

s19

84to

1988

sin

cew

ear

eu

sing

1986

asa

nat

ural

exp

erim

ent

(b)

Thi

sva

riab

leis

the

dol

lar

chan

gein

CE

Ow

ealt

hp

er$1

000

chan

gein

rm

mar

ketv

alu

e(s

eeJe

nse

nan

dM

urp

hy

1990

)

(c)

Thi

sva

riab

lein

the

dol

lar

char

gein

CE

Ow

ealt

hp

er10

-per

cent

chan

gein

rm

valu

e(s

eeH

alla

nd

Lie

bman

199

8)

(d)

Stan

dar

dd

evia

tion

ofr

mm

arke

tva

lue

intr

illio

ns

32 Hall amp Liebman

TABLE 5Large Institutional Investors and Stock-Option Compensation

The Cross-sectional Relationship between Institutional Investors andOption Grants

(1) (2) (3) (4)

Coefcient on fraction of 217 (067) 145 (014) 177 (069) 120 (015)rmrsquos shares owned byinstitutional investors

Sample period 1994 1980ndash1994 1994 1980ndash1994Contract-theory and stock No No Yes Yes

market covariatesTime dummies No Yes No Yes

The dependent variable is the share of compensation paid in options

factor alone can explain 6 percentage points of the 23-percentage-pointincrease in the median share of compensation paid in options that oc-curred over this time period Of course this variable is only one measureof the strengthening of corporate governance so the total impact of im-proved corporate governance could be greater The second and fourthcolumns of the table use the entire 1980ndash1994 sample and include timedummies to isolate the cross-sectional correlations Again there is evi-dence of a strong cross-sectional relationship between the two variablesalthough the coefcients are slightly smaller perhaps reecting a weakercorporate-governance role in the earlier period In sum there is a strongrelationship in both longitudinal and cross-sectional data between theinstitutional-investor variable and the options-percentage variable13

7 THE MILLION-DOLLAR RULE

In 1993 section 162(m) of the Internal Revenue Code was enactedlimiting the deductibility of executive compensation in excess of onemillion dollars unless the compensation was performance-related Pro-ponents of this legislation argued that it would slow the growth ofexecutive pay and tighten the link between rm performance and execu-tive compensation

The new provision became effective for tax year 1994 and applies tothe CEO and the other four most highly compensated executives in eachcompany Pay that is performance-related such as stock-option grants

13 The one specication in which the fraction owned by institutional investors is notsignicant is in column (2) of Table 4 a regression that includes both xed effects and timeeffects

Taxation of Executive Compensation 33

and bonuses paid for meeting clear performance goals is not affected bythis regulation provided that it has been approved by shareholdersThus the main impact of the provision is to limit the deductibility ofsalary in excess of a million dollars Woodlock and Antenucci (1997)studied the proxy statements of 376 rms and documented that mostrms responded to the new law by qualifying top executive pay asperformance-related Perry and Zenner (1999) identied 25 rms thatreduced salaries from above $1 million to below $1 million and foundthat 23 of the rms cited section 162(m) as their reason for doing so

In this section of the paper we investigate whether this provision hasaffected the level or structure of executive pay using the Execucompdata base which contains information on the highest-paid executives ineach of the rms in the SampP 500 SampP Midcap 400 and SampP SmallCap600 Between 1993 and 1998 the median salary of the Execucomp execu-tives rose from $230000 to $306000 The median value of total compen-sation (salary bonus and option grants) rose from $470000 to $882000Limiting the sample of CEOs indicates that the total compensation of themedian CEO rose from $11 million to $18 million over the period whilethe median salary rose from $441000 to $550000 Thus executive com-pensation clearly continued to rise at a rapid rate after the implementa-tion of the million-dollar rule though perhaps at a slower rate than itotherwise would have

If the regulation affects behavior it should have the largest impact onrms whose executives are receiving salaries that are above the limit orjust below it Thus we might expect to see rms whose executives arereceiving relatively high salaries to rely more heavily on performance-based pay and less heavily on salary in the years following the regulationPerry and Zenner (1999) and Livingston (1999) both nd that this is thecase This trend can be clearly identied in the Execucomp data set Forexample over 1993 to 1998 the median annualized growth rate of salaryfor executives whose 1993 salary was below $500000 in the previous yearwas 85 percent For executives with salary between $500000 and$700000 it was 64 percent for executives between $700000 and $850000it was 50 percent for executives between $750000 and $1 million it was31 percent and for executives above $1 million it was 00 percent14

The observation that executives at lower levels of salary had highergrowth rates is not very strong evidence of an impact of section 162(m)however because it is possible that low-salary executives would havehad higher growth rates even in the absence of the provision To explore

14 Of the 3137 executives for whom we have complete data between 1993 and 1998 2708had 1993 salaries below $500000 while only 35 had 1993 salaries above $1 million

34 Hall amp Liebman

whether this is the case we use Execucomp data from both before andafter the law change Ideally we would want to have data from manyyears before the change to examine whether there is an underlying pat-tern of faster salary growth for low-salary executives Unfortunately theExecucomp data begin in 1992 and therefore provide us with only the1992ndash1993 growth rate in compensation to use as a baseline15

We run regressions pooling data from 1992ndash1993 through 1997ndash1998of the annual percentage change in different forms of compensation on avariable million that is designed to measure the likelihood that anexecutiversquos compensation will be affected by section 162(m) We denemillion as the minimum of 1 and the executiversquos salary divided by $1million in the previous year Thus any executive with salary at or above$1 million would receive a value of 1 while an executive with a salary of$500000 would be coded as a 0516 We include this variable by itself andalso interacted with a dummy variable (after) that equals 1 for years1994 and after Thus the million variable is designed to pick up anyunderlying relationship between the level of salary and the growth rateof salary while the interacted one will identify any differential growthrate of salary by income in the period after section 162(m) took effect Wealso include a full set of year dummies the natural logarithm of thermrsquos market value and the rmrsquos annual rate of return over the twoprevious years as control variables To reduce the sensitivity of our re-sults to outliers we run quantile regressions and robust regressions

Table 6 contains our results The rst four columns are for all topexecutives while the last four columns focus on CEOs Columns (1) and(2) present estimates from robust regression for the percentage change insalary and the percentage change in total compensation (salary bonusstock grants and stock-option grants) In the salary growth regressionsfor all executives the coefcient on million is negative and signicantindicating that there is an underlying relationship between the level ofexecutive compensation and its subsequent growth rate The coefcienton million 3 after is negative as well suggesting that the negativerelationship between the level of salary and its subsequent growth rateintensied slightly after 1994 The relatively small magnitude suggeststhat an executive with a $1 million dollar salary would see his salarygrow at an annual rate that would be about 06 percent slower than anexecutive earning $500000

15 The HallndashLiebman data set contains only a combined salary and bonus variable makingit impossible to use it to analyze this issue16 We also ran the regressions (not reported) allowing for a nonlinear relationship betweenmillion and compensation growth and obtained substantively similar results

TA

BL

E6

The

Eff

ect

ofth

eM

illi

on-D

olla

rR

ule

onSa

lary

and

Tot

alC

ompe

nsat

ion

All

top

exec

uti

ves

CE

Os

only

Rob

ust

regr

essi

onQ

uan

tile

regr

essi

onR

obu

stre

gres

sion

Qu

anti

lere

gres

sion

D

in

Din

D

in

Din

D

in

Din

D

in

Din

sala

ryto

talc

omp

(a)

sala

ryto

talc

omp

sa

lary

tota

lcom

p

sala

ry

tota

lcom

p

mil

lio

n(b

)2

006

02

035

72

006

32

024

12

008

72

023

12

010

32

024

5(0

005

)(0

032

)(0

006

)(0

027

)(0

010

)(0

073

)(0

011

)(0

056

)m

illi

on

32

001

10

026

20

021

20

015

20

016

014

92

003

40

158

aft

er(c

)(0

005

)(0

033

)(0

006

)(0

029

)(0

010

)(0

076

)(0

011

)(0

057

)N

359

5035

950

359

5035

950

757

17

589

757

17

589

Stan

dar

der

rors

inp

aren

thes

esR

egre

ssio

nsal

soin

clu

de

year

du

mm

ies

stoc

km

arke

tre

turn

and

log

mar

ket

valu

e

(a)

Tot

alco

mp

ensa

tion

incl

ud

essa

lary

bon

us

rest

rict

edst

ock

and

stoc

k-op

tion

gran

ts

(b)

mil

lion

isd

ene

das

min

(1s

alar

yin

pre

viou

sye

ars

div

ided

by$1

mil

lion)

(c)

aft

eris

anin

dic

ator

vari

able

equ

alto

zero

befo

reth

ere

gula

tion

took

plac

e(1

993)

and

one

afte

rwar

d

36 Hall amp Liebman

In the second column the coefcient on million 3 after is essentiallyzero suggesting that any decrease in salary brought about by section162(m) was offset by increases in bonus and stock-option grants Theresults are similar for the median regressions There is some evidence ofa very small slowdown in growth of salary for executives likely to beaffected by section 162(m) but no sign of a change in total compensa-tion Taken together this provides evidence of a very small substitutionof performance-related pay for salary

The last four columns run identical tests but limit the sample to CEOsThe interaction of million 3 after continues to be small and negative inthe salary regressions In contrast the impact of 162(m) on total compen-sation appears to be positive and fairly substantial (about 15 percentagepoints per year) in these regressions That is the increase in bonus andstock options following the million-dollar rule more than offset the verysmall decline in salary

All of these conclusions should be interpreted somewhat cautiouslybecause our data provide us with only one pre-1993 control year Never-theless the data are consistent with section 162(m) having led to a veryminor slowdown in salary growth and one that is dramatically smallerthan what one would conclude without controlling for underlying differ-ential salary growth rates at different levels of salary Since total compen-sation for CEOs did not decline and perhaps increased there is evidenceof a minor substitution of performance-related pay for salary in responseto the regulation

8 EXECUTIVE COMPENSATION AND THEELASTICITY OF TAXABLE INCOME

Even if taxes have only a minor effect on executive compensation thestructure of executive compensation is important for tax policy be-cause high-income taxpayers are sources of signicant revenue and arethe focus of important debates about the efciency cost of taxation

Goolsbee (1999) has recently argued that executive decisions to exer-cise options are highly responsive to intertemporal differences in taxrates created by tax reforms He claims that much of the apparent declinein taxable income for high-income taxpayers between 1992 and 1993documented by Feldstein and Feenberg (1996) can be explained by execu-tives shifting option gains into 1992 in order to avoid the higher marginaltax rates that went into effect in 1993 By extension he implies that thehigh elasticities of taxable income with respect to personal tax ratesestimated by comparing the pre- and post-tax-reform taxable income of

Taxation of Executive Compensation 37

high-income individuals reect intertemporal shifting of income and notpermanent effects of taxation If this is true then the deadweight loss oftaxation is lower as is the cost of progressivity

While Goolsbee presents comprehensive evidence that option gainswere unusually high in 1992 his identication of the impact of tax rates onthe timing of option gains depends on correctly specifying a counter-factual level of options that would have been exercised in the absence ofthe tax change He simply assumes a linear time trend The purpose ofGoolsbeersquos time trend is presumably to account for the increasing use ofoptions over time However since options typically require a vestingperiod of at least a couple of years a much more direct measure of ex-pected option gains can be constructed based on the total value of vestedoptions held by the CEO Since options typically have vesting periods thatexceed two years this total can be treated as exogenous since it will not beaffected by the tax changes

More generally the value of options held by the CEO is likely to havea major effect on his exercise decisions If his options are out of themoney he will not have gains to realize If there has been a recent run-up in the market an optimizing CEO may exercise an unusually largeamount of options in order to diversify his overall portfolio and if in-the-money options are about to expire they will be exercised in thecurrent year

Our evidence suggests that much of the apparent tax shifting caninstead be attributed to stock market performance and the timing of pastoption grants Using our 15-year panel we replicate Goolsbeersquos (1999)result for the 1993 period but show that the impact of taxes is notpresent in other time periods Moreover even the result for the 1993period disappears when appropriate controls for past option grants andstock price appreciation are included

The top three rows of Table 7 present data on the taxable incomes ofCEOs in the HallndashLiebman sample For these results our panel has beennarrowed to include only CEOs whose rmrsquos scal years correspondwith tax years In addition we create a balanced panel for each of thethree tax-reform periods by excluding any rm for which there are notcomplete data for the period surrounding the reform

The fourth row of the table shows that the 1993 tax act raised the topmarginal personal income tax rate from 310 to 396 percent effective fortax year 1993 with a further increase occurring in 1994 due to the uncap-ping of the Medicare payroll tax The top row of the table shows thattaxable income was higher in 1992 than for any other year in our sample(all data are in 1994 dollars) Option gains were particularly high in1992mdashmore than double their 1991 level and 68 percent higher than their

TA

BL

E7

Tax

Ref

orm

Sto

ckM

arke

tP

erfo

rman

cea

ndth

eT

imin

gof

Opt

ion

Gai

ns

Ave

rage

stoc

km

arke

tM

ean

CE

OM

ean

CE

Ore

turn

for

Mea

nva

lue

taxa

ble

sala

ryan

dM

ean

CE

OT

opfe

der

al

rms

inof

opti

ons

(Val

ue

ofop

tion

Sam

ple

inco

me

bonu

sop

tion

gain

sp

erso

nal

sam

ple

for

held

atbe

gga

ins)

(va

lue

Ref

orm

size

Yea

r($

mil

)($

mil

)($

mil

)M

TR

pre

vye

ar(

)of

year

ofop

tion

she

ld)

ER

TA

8127

919

800

874

064

80

163

7025

80

460

354

1981

083

10

672

009

869

245

070

61

3919

820

872

067

90

099

5011

90

686

144

1983

099

00

724

025

250

355

116

62

16

TR

A86

295

1985

119

70

837

024

050

76

116

02

0719

861

448

091

70

345

5042

81

776

194

1987

165

50

959

059

338

519

12

229

266

1988

163

21

058

039

028

07

237

71

6419

891

656

105

20

414

2817

92

345

176

1993

272

1990

209

51

028

061

828

232

288

02

15T

axA

ct19

911

875

102

00

547

312

01

268

82

0319

922

678

107

01

250

3146

64

552

275

1993

232

41

209

074

339

620

64

299

173

1994

196

41

262

037

842

515

45

113

074

Sam

ple

isH

allndash

Lie

bman

rm

sw

ith

Dec

embe

r

scal

year

sth

atha

dco

mp

lete

dat

afo

ron

eof

the

thre

eta

x-re

form

per

iod

sst

ud

ied

Taxation of Executive Compensation 39

1994 level These patterns are consistent with Goolsbeersquos claim that tax-payers responded to the anticipated increase in marginal tax rates byshifting gains into 1992

The evidence from the other two tax reforms is less clear suggestingthat there might be factors other than tax rates determining the timing ofoption gains ERTA 1981 reduced the top marginal tax rate from 70percent in 1980 to 50 percent in 1982 However there is no sign thatoption gains were reduced in 1980 and 1981 and then increased in 198217

There was a big increase in option gains in 1983 however Similarly thehighest year of stock-option gains in the period around 1986 was 1987 astrange year to take gains since taxpayers would have known that thepersonal rate would be 10 percentage points lower if they waited untilthe following year18

What all the peak years of option gains have in common is that theyfollow years of strong stock market performance The fth row of Table 7shows the average stock market return during the previous year for therms in our sample The large value of option gains in 1983 followed a355-percent increase in the stock market The 1992 boom in option gainsfollowed a 466-percent increase in the stock market the more modest1987 peak in option gains followed a 19-percent gain (and a 43-percentgain two years before)

Strong stock market performance raises the value of options availableto exercise The sixth row of the table shows the mean value of optionsheld by CEOs at the beginning of the year All three of the peak years ofoption gains correspond with peak years in terms of the value of optionsavailable for exercise Since stock market appreciation during the yearalso increases the value of options available to exercise later in that yearwe need to allow for current-year stock market appreciation as well

Table 8 contains Goolsbee-style regressions for each of the tax re-forms The rst two columns reprint the results from Goolsbee (1999)The third and fourth columns replicate his results in our sample Inparticular with a specication similar to his we are able to replicate hisnding that the current after-tax share has a positive coefcient and thefollowing yearrsquos after-tax share has a coefcient of roughly negative oneColumns (5) through (8) show that this pattern was unique to the 1993tax reform Indeed in the TRA 86 regressions the coefcient on the

17 As we explained above for executives covered by the maximum tax on earned incomethere was not a decline in tax rates between 1981 and 198218 It is possible that the large stock market decline in October 1997 may have promptedsome people to exercise options earlier than they otherwise would have However anincrease in option exercises typically follows stock market increases not declines

TA

BL

E8

Att

empt

sto

Rep

lica

teth

eT

ax-S

hift

ing

Res

ult

inO

ther

Tim

eP

erio

ds

Goo

lsbe

eG

ools

bee

1993

1993

Ent

ire

Ent

ire

regr

essi

onre

gres

sion

refo

rmre

form

TR

A86

TR

A86

ER

TA

81E

RT

A81

1980

ndash199

419

80ndash1

994

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

ln[1

2ta

x(t)

]1

244

108

20

455

032

62

017

62

017

22

016

72

016

70

026

20

068

(01

34)

(01

27)

(05

37)

(05

42)

(02

10)

(02

09)

(00

90)

(00

91)

(00

67)

(00

65)

ln[1

2ta

x(t1

1)]

20

867

20

764

21

214

20

917

047

80

722

20

103

004

30

176

018

3(0

118

)(0

112

)(0

491

)(0

499

)(0

157

)(0

174

)(0

072

)(0

072

)(0

068

)(0

066

)

Tim

e0

135

007

62

000

52

036

40

041

000

80

094

003

50

056

002

7(0

008

)(0

008

)(0

049

)(0

049

)(0

028

)(0

028

)(0

025

)(0

026

)(0

003

)(0

003

)

ln(m

arke

tva

lue)

066

10

489

023

70

334

039

5(0

015

)(0

058

)(0

039

)(0

042

)(0

017

)

Yea

rs19

91ndash1

995

1991

ndash199

519

90ndash1

994

1990

ndash199

419

85ndash1

989

1985

ndash198

919

80ndash1

983

1980

ndash198

319

80ndash1

994

1980

ndash199

4

Ad

just

edR

20

720

770

610

630

690

760

800

810

630

68

Obs

erva

tion

s13

856

132

371

385

136

41

470

143

31

113

106

23

968

385

9

Dep

end

ent

vari

able

isth

elo

gof

taxa

ble

inco

me

All

regr

essi

ons

con

tain

rm

xe

def

fect

sG

ools

bee

regr

essi

on(2

)al

soco

ntai

nstw

oot

her

vari

able

sno

tsh

own

inth

eta

ble

Stan

dar

der

rors

inp

aren

thes

es

Taxation of Executive Compensation 41

TABLE 9Alternative Explanations for the Timing of Option Exercises

1993 Entirereform TRA 86 ERTA 81 1980ndash1994

(1) (2) (3) (4)

ln[12tax(t)] 0108 0111 20051 0201(0324) (0136) (0048) (0057)

ln[12tax(t11)] 20134 0573 0125 0150(0336) (0156) (0053) (0066)

ln(market value) 0418 0177 0338 0418(0061) (0039) (0046) (0163)

Firmrsquos stock market return (t) 0298 0268 0214 0258(0044) (0045) (0031) (0023)

Firmrsquos stock market return 0162 0206 0031 0128(t21) (0041) (0008) (0029) (0021)

ln(value of options held by 0004 0008 0001 0009CEO at start of year) (0003) (0003) (0002) (0001)

Years 1990ndash1994 1985ndash1989 1980ndash1983 1980ndash1994Adjusted R2 064 071 082 068Observations 1349 1427 1036 3812

The dependent variable is the log of taxable income All regressions contain rm xed effects Stan-dard errors in parentheses

following yearrsquos tax rate is positive and signicant Columns (9) and (10)pool all three reforms and fail to replicate the pattern that would reectintertemporal shifting19

It is possible that the ease with which one can shift the timing ofoption gains is asymmetric It may be harder to shift option gains for-ward in time since some options expire and therefore must be exercisednow In addition options were a much smaller portion of compensationin the early 1980s so perhaps the ERTA 81 results are not surprisingNonetheless these results cast doubt on the proposition that this is thechannel through which spurious estimates of the elasticity of taxableincome have occurred for previous tax reforms and they suggest thatwe should look harder for alternative explanations for the timing ofoption gains

Table 9 attempts to predict options gains using the lagged and currentincrease in a rmrsquos market value and the value of the options held by theCEO at the start of the year The two stock market variables stronglypredict option gains while the value of options held has the correct sign

19 The results also fail to suggest large permanent effects of marginal tax rates on taxableincome

42 Hall amp Liebman

in all of the specications and is signicant in the pooled sample eventhough it is highly correlated with the two stock market return variablesIn contrast the tax-timing explanation fails even in the 1993 period

It is important to realize that our results do not invalidate Goolsbeersquoscritique of claims that aggregate revenue data from 1992 and 1993 provethat elasticities of taxable income are large We have simply given analternative explanation for why taxable incomes were unusually high in1992 However our results do cast doubt on the conclusion that largetiming shifts are ubiquitous and capable of explaining away the morecareful estimates of taxable income elasticities that have been performedusing micro data (eg Feldstein 1995 Auten and Carroll 1995)

One further point is in order It is sometimes argued that the combina-tion of increased use of options and the booming stock market explainswhy federal tax revenues have been so much higher than expected dur-ing the 1990s However while large option gains increase revenue fromthe personal income tax they produce offsetting deductions for cor-porations The net impact on revenue is only the difference in the twomarginal tax rates times the option gain While the gap between thepersonal rate and the effective corporate rate is nontrivial the impact oflarge option exercises on total revenue is substantially smaller than theamount by which the exercises inate personal tax revenue and reducecorporate tax revenue

9 CONCLUSION

We have described the tax rules for executive pay in detail and analyzedhow changes in various tax rates affect the tax advantages of stock optionsrelative to salary and bonus We nd that there is a moderate tax advan-tage to optionsmdashcurrently about $4 per $100 in pretax executive paymdashand that changes in the tax advantage of options over time have had atmost a modest impact on the composition of pay Corporate governancefactors particularly the role of large investors appear to be more impor-tant in explaining the dramatic increase in option pay

In addition our evidence suggests that more direct attempts to use taxpolicy to inuence executive compensation have had little effect Wend that the million-dollar rule led companies to substitute perfor-mance-based pay for salary But our evidence suggests that this substitu-tion was quite modest and there is no evidence that the total level of paywas reduced Overall although the stock-option explosion has dramati-cally increased the link between pay and performance this change isdue almost entirely to nontax factors

Taxation of Executive Compensation 43

REFERENCES

Auerbach Alan J (1996) ldquoThe Dynamic Effects of Tax Law Asymmetriesrdquo Re-view of Economic Studies 53451ndash405

mdashmdashmdash and James Poterba (1987) ldquoTax Loss Carryforwards and Corporate TaxIncentivesrdquo In The Effects of Taxation on Capital Accumulation M Feldstein(ed) Chicago University of Chicago Press

Altshuler Rosanne and Alan Auerbach (1990) ldquoThe Signicance of Tax LawAsymmetries An Empirical Investigationrdquo Quarterly Journal of Economics 10561ndash86

Auten Gerald and Robert Carroll (1995) ldquoThe Effect of Income Taxes on House-hold Behaviorrdquo Ofce of Tax Analysis US Department of the Treasury

Baker George P and Brian J Hall (1998) ldquoCEO Incentives and Firm SizerdquoNBER Working Paper no 6868 December

Core John and Wayne Guay (1999) ldquoThe Use of Equity Grants to ManageOptimal Equity Incentive Levelsrdquo Draft

Feldstein M (1995) ldquoThe Effect of Marginal Tax Rates on Taxable Income APanel Study of the 1986 Tax Reform Actrdquo Journal of Political Economy 103 (no3)551ndash72

mdashmdashmdash and D Feenberg (1996) ldquoThe Effect of Increased Tax Rates on TaxableIncome and Economic Efciency A Preliminary Analysis of the 1993 Tax RateIncreasesrdquo In Tax Policy and the Economy vol 10 James M Poterba (ed)Cambridge MA MIT Press

Gompers Paul A and Andrew Metrick (1998) ldquoInstitutional Investors andEquity Pricesrdquo NBER Working Paper no 6723

Goolsbee Austan (1999) ldquoWhat Happens When You Tax the Rich Evidencefrom Executive Compensationrdquo NBER Working Paper no 6333

Graham John R (1996) ldquoProxies for the Corporate Marginal Tax Raterdquo Journal ofFinancial Economics 42187ndash221

Hall Brian J (1999) ldquoThe Design of Multi-year Stock Option Plansrdquo Journal ofApplied Corporate Finance 12(2)97ndash106

mdashmdashmdash and Jeffrey B Liebman (1998) ldquoAre CEOs Really Paid Like Bureau-cratsrdquo Quarterly Journal of Economics CXII(no 3)653ndash691

Jensen Michael C and William M Meckling (1976) ldquoTheory of the Firm Mana-gerial Behavior Agency Costs and Ownership Structurerdquo Journal of FinancialEconomics III305ndash360

mdashmdashmdash and Kevin J Murphy (1990) ldquoPerformance Pay and Top-ManagementIncentivesrdquo Journal of Political Economy XCVIII225ndash264

Kaplan Steven N (1997) ldquoThe Evolution of US Corporate Governance WeAre All Henry Kravis Nowrdquo Draft

Livingstone Jane R (1999) ldquoExecutive Compensation Contracting and Re-sponses to Increased Tax Costsrdquo Louisiana State University June

Miller Merton H and Myron S Scholes (1982) ldquoExecutive CompensationTaxes and Incentivesrdquo In Financial Economics Essays in Honor of Paul CootnerWilliam F Sharpe and Cathryn M Cootner (eds) Princeton NJ Prentice-Hall

Murphy Kevin J (1999) ldquoExecutive Compensationrdquo forthcoming in OrleyAshenfelter and David Card (eds) Handbook of Labor Economics Vol 3 NorthHolland

Perry Todd and Marc Zenner (1999) ldquoPay for Performance Government Regu-

FIG

UR

E1

Pay

toP

erfo

rman

ceT

heL

ink

betw

een

CE

OW

ealt

han

dSh

areh

olde

rV

alue

Sinc

e19

80

Not

eT

he

anal

ysis

cont

rols

for

the

chan

ging

com

pos

itio

nof

rm

size

over

tim

eT

hu

sth

ese

esti

mat

essh

owth

ep

ay-t

o-p

erfo

rman

celin

kfo

ra

sim

ilar

-siz

edco

mp

any

over

tim

ein

this

case

aco

mp

any

wit

ha

mar

ket

valu

eof

$1bi

llion

Taxation of Executive Compensation 7

wishes) slowly over time Common vesting periods are in the three- tove-year range and options usually vest linearly (eg a four-year op-tion vests at 25 percent at the end of each year) An executive typicallyloses any unvested options upon departure Although options may beexercised as soon as they vest they do not have to be exercised untilthey expire or mature Almost 85 percent of stock-option plans have aterm of exactly ten years with virtually all of the remainder being in theve- to ten-year range About 95 percent of options are granted at themoney or at fair market value which means that the exercise price at grantdate is set equal to the stock price at grant date The remaining 5 percentare either discount options (so-called in-the-money options where theexercise price is below the stock price at grant date) or premium options(so-called out-of-the-money options where the exercise price is abovethe stock price at grant date) The holders of options typically do nothave dividend rights or voting rights even on vested options3

31 Tax Rules and Stock-Option CompensationUnlike salary and bonus stock-option grants are typically an untaxedevent at the time of grant For the most widely used optionsmdashnonquali-ed stock options (NQSOs)mdashexecutives are taxed at the personal incometax rate on option prots (the difference between that stock price and theexercise price times the number of options) when the options are exer-cised The company receives a parallel deduction against corporate in-come at that point If the executive continues to hold the shares afterexercise any subsequent appreciation is taxed at the capital gains rate inthe usual way In 1993 an additional feature was added to the tax code[Internal Revenue Code section 162(M)] that disallowed a corporate de-duction for any executive pay above $1 million that is not performance-based While this rule affects executive salaries most bonuses qualify asperformance-based and standard stock options automatically qualifyTherefore this provision gives companies with highly paid executives anincentive to give more pay in the form of bonuses and stock options asubject we return to in section 7 A summary description of the tax (andaccounting) treatment of cash and option compensation is in Table 1

A far less common type of option which is estimated to account forabout 5 percent of option grants is the incentive stock option (ISO) WhileISOs are similar to NQSOs in their design they are crucially different intwo respects First they have an annual cap of $100000 per executive

3 See Murphy (1999) for details about stock options and Miller and Scholes (1982) on taxincentives

TA

BL

E1

Tax

and

Acc

ount

ing

Tre

atm

ent

ofE

xecu

tive

Com

pens

atio

nC

ash

Pay

men

tsve

rsus

Stan

dard

Exe

cuti

veO

ptio

ns

Tax

atio

nat

Tax

atio

nat

Acc

ount

ing

Com

pens

atio

nD

escr

ipti

onp

erso

nall

evel

corp

orat

ele

vel

trea

tmen

t

Sala

ryan

dbo

nu

sC

ash

pay

men

tsm

ade

Tax

edas

ord

inar

yFu

llyd

edu

ctib

leS

ince

Exp

ense

dag

ains

tea

rn-

toC

EO

in

com

e19

93a

ny

pay

over

$1in

gsin

the

usu

alw

ay

mill

ion

isno

td

educ

t-ib

leu

nles

sit

isp

er-

form

ance

-bas

ed

Non

-qua

lie

dst

ock

Op

tion

sgr

ant

the

ex-

The

gran

ting

ofan

op-

Apa

ralle

ltax

ded

uc-

The

reis

no

exp

ense

rec-

opti

ons

(NQ

SOs)

ecut

ive

the

righ

tbu

tti

onis

typ

ical

lyan

un-

tion

atth

eco

rpor

ate

ogni

tion

(atg

ran

tex

er-

not

the

obli

gati

ont

ota

xed

even

tU

pon

exer

-le

veli

sge

nera

llyal

-ci

seo

rsa

le)

for

op-

buy

shar

esof

stoc

kat

cise

opt

ion

pro

ts

are

low

edu

pon

exer

cise

tion

sif

as

isu

sual

lyan

exer

cise

pric

e(t

ypi-

taxe

dat

the

ord

inar

y(f

orth

eam

ount

ofop

-th

eca

set

he

opti

ons

cally

the

curr

ent

stoc

kra

teI

fth

eex

ecu

tive

tion

pro

ts

)O

pti

ons

are

not

dis

cou

nted

(in

pri

ceat

the

dat

eof

cont

inu

esto

hold

onto

auto

mat

ical

lyqu

alif

yas

the

mon

eyat

gran

tgr

ant)

Th

eop

tion

sth

esh

ares

the

exec

u-

perf

orm

ance

-bas

edp

ayd

ate)

and

the

exer

cise

typ

ical

lyve

stov

era

3ndashti

veis

taxe

din

the

and

ther

efor

ear

eno

tpr

ice

and

num

ber

ofop

-5-

year

per

iod

The

typ

i-us

ualw

ay(a

tth

eca

pi-

subj

ect

toth

e$1

-ti

ons

iskn

own

atgr

ant

calm

atu

rity

is10

year

sta

lgai

nsra

te)

onan

ym

illio

nca

p

dat

eFo

rd

isco

unt

op-

but

the

man

ager

may

furt

her

app

reci

atio

nof

tion

sth

ed

iffe

renc

ebe

-ex

erci

seea

rly

Som

eth

esh

are

pri

ce

twee

nth

est

ock

pri

ceop

tion

sar

edi

scou

ntop

-an

dth

em

arke

tp

rice

isti

ons

(the

exer

cise

pri

ceex

pen

sed

over

the

vest

-is

low

erth

anth

est

ock

ing

per

iod

For

opti

ons

pri

ceat

the

gran

td

ate)

w

ith

vari

able

term

sP

rem

ium

opti

ons

are

(eg

a

vari

able

exer

-th

ere

vers

e(t

heex

er-

cise

pri

cev

esti

ngth

atci

sep

rice

isgr

eate

ris

tied

top

erfo

rman

ce)

than

the

stoc

kp

rice

atth

eop

tion

sar

em

arke

dgr

ant

dat

e)

tom

arke

tan

dex

-p

ense

dd

uri

ngth

eti

me

betw

een

gran

tan

dex

erci

se

Not

eW

eth

ank

com

pen

sati

onco

nsu

ltan

tsSc

ott

Gre

enbe

rgan

dSc

ott

Ols

en(b

oth

ofT

ower

sP

erri

n)

and

Fred

eric

Coo

k(o

fFr

eder

icW

C

ook

and

Ass

ocia

tes)

for

help

fulc

onve

rsat

ion

sin

und

erst

and

ing

the

tax

and

acco

unt

ing

rule

s

10 Hall amp Liebman

Second the tax treatment of ISOs is different ISOs are completely un-taxed at grant or exercise It is only at sale that the executive is taxed andeven then the executive is taxed at the lower capital gains tax rate Thedisadvantage is that the corporation never gets to take a parallel taxdeduction against corporate prots Thus ISOs become more attractiveas the personal tax rate increases and as the corporate tax rate and thecapital gains tax rates fall

Two other related types of compensation restricted stock and stockappreciation rights are worth brief discussion Both however are farless common than standard stock options Restricted stock is payment inthe form of restricted shares the restriction being that the shares vestover time as with options Unlike options the shares typically havevoting and dividend rights With regard to taxation the executive istaxed at the personal rate on the value of the restricted stock as thevesting restrictions lapse However the executive may choose to betaxed at the grant date in which case all subsequent appreciation istaxed at the capital gains rate The dividends paid to the executive aretaxed at the ordinary rate in the usual way The company generallyreceives a parallel deduction equal to the amount of the executiversquosincome when the executive is taxed Unlike stock options restrictedstock is not generally considered to be performance-based and is there-fore subject to the million-dollar rule (unless the vesting of the restrictedstock is performance-based which is sometimes the case)

Stock appreciation rights (SARs) are rights that replicate the payoffs ofstock options with a cash transfer Thus SARs are simpler than options inthat there is no requirement to buy and resell the stock in order to ldquocashoutrdquo SARs generally have the same tax treatment as NQSOs both to theindividual and to the corporation SARs are relatively rare however be-cause they have disadvantageous accounting treatment (described in thenext section) relative to stock options and their main relative advantagevis-a-vis options has been essentially replicated through broker-assistedcashless option exercisesmdashwhereby a broker makes a ldquonanosecondrdquo loanto the executive (to purchase and resell the stock) when the executivewants to ldquocash inrdquo on option prots The tax and accounting treatment ofISOs restricted stock and SARs is summarized in Table 2

32 The Accounting Treatment of OptionsUnlike cash compensation which is expensed against earnings there isgenerally no expense recognition (at grant exercise or sale) for optionswhether they be NQSOs or ISOs As a result compensation consultantsoften point out that stock options are the only form of compensation thatare free in an accounting sense but still deductible for tax purposes

Taxation of Executive Compensation 11

Stock options do however lead to expense recognition if they are dis-counted (in the money at grant date) or if the exercise price and numberof options are not known at grant date For discount options the differ-ence between the stock price and the market price is expensed over thevesting period For options with variable terms (eg a variable exerciseprice vesting that is tied to performance) the options are marked tomarket and expensed during the time between grant and exercise Practi-tioners claim that the accounting treatment of options plays an impor-tant role in the design of option programs Thus plans that have ldquobadaccountingrdquo but are thought by many to have attractive incentive fea-tures are often not even seriously considered by companies Examples ofsuch potentially attractive plans include indexed options (where a CEOprots only if his rmrsquos share price grows relative to some market orindustry benchmark) and option grants that are explicitly performance-related both of which would lead to expenses against earnings

Unlike options restricted stock and SARs do not generally receivefavorable accounting treatment Restricted stock is generally expensedover the period in which the restrictions lapse (usually the vesting pe-riod) The magnitude of the expense is the difference between the cur-rent stock price and the executiversquos cost (if any) SARs are marked tomarket each period and the difference between the stock price and theexercise price is expensed over the outstanding period of SARs

4 THE TAXATION OF EXECUTIVE PAY CASHVERSUS OPTIONS

Stock-based compensation performs two roles in executive compensa-tion arrangements First it helps align the incentives of the executivewith the interests of the rmrsquos shareholders Second it often enables therm to compensate the CEO in a way that is more advantageous from atax standpoint than paying the executive in salary and bonus

41 Agency Theory and Executive Compensation in the Presenceof TaxationIn standard agency-theory models ( Jensen and Meckling 1976) agencycosts are the result of the separation of ownership and control Managersdo not have the same incentives as the owners The optimal incentivecontract for managers balances the benets of high-powered incentives(linking the fortunes of owners and managers through stock and stockoptions for example) with the costs of loading too much risk on risk-averse managers

TA

BL

E2

Tax

and

Acc

ount

ing

Tre

atm

ent

ofE

xecu

tive

Com

pens

atio

nIn

cent

ive

Stoc

kO

ptio

ns(I

SOs)

Res

tric

ted

Stoc

kan

dSt

ock

App

reci

atio

nR

ight

s(S

AR

s)

Tax

atio

nat

Tax

atio

nat

Acc

ount

ing

Com

pen

sati

onD

escr

ipti

onp

erso

nall

evel

corp

orat

ele

vel

trea

tmen

t

Ince

nti

vest

ock

ISO

sar

eid

enti

calt

oIf

the

usu

alco

ndit

ions

Ifth

eu

sual

cond

itio

nsA

sw

ith

NQ

SOs

ther

eop

tion

sN

QSO

sin

thei

rd

esig

n

are

met

th

ere

isno

tax

are

met

the

reis

gene

r-is

gene

rally

noex

pen

seH

owev

ert

hey

have

anat

gran

tor

exer

cise

At

ally

no

corp

orat

eta

xd

e-re

cogn

itio

nfo

rIS

Os

annu

alca

pof

$100

000

sale

all

opti

onp

ro

tsd

uct

ion

atgr

ant

exer

-(e

ver)

on

the

amou

ntth

atan

dsu

bseq

uen

tca

pita

lci

seo

rsa

le

vest

sT

he

tax

trea

t-ap

prec

iati

onar

eta

xed

men

tof

ISO

sal

soat

the

cap

ital

gain

sd

iffe

rs

rate

Res

tric

ted

stoc

kSi

mila

rto

stoc

kop

-T

hein

div

idu

alis

taxe

dT

heco

mp

any

gene

rall

yR

estr

icte

dst

ock

isge

n-ti

ons

but

shar

esat

the

ord

inar

yra

teas

rece

ives

ap

aral

leld

e-er

ally

exp

ense

dov

er(w

hich

vest

slow

ly)

the

rest

rict

ions

lap

se

du

ctio

neq

ual

toth

eth

eve

stin

gp

erio

dT

hera

ther

than

opti

ons

are

How

ever

the

man

ager

amou

ntof

the

exec

u-

amou

nt

exp

ense

dis

gran

ted

toth

eex

ecu

-m

aych

ose

tobe

taxe

dti

versquos

inco

me

whe

nth

eth

ed

iffe

renc

ebe

twee

nti

veU

nlik

eop

tion

sat

gran

td

ate

inw

hich

exec

uti

veis

taxe

dR

e-th

ecu

rren

tst

ock

pric

eth

esh

ares

typi

cally

case

alls

ubs

equ

ent

ap-

stri

cted

stoc

kis

gen

er-

and

the

exec

uti

versquos

have

voti

ngan

dd

ivi-

pre

ciat

ion

ista

xed

atal

lysu

bjec

tto

the

cost

(if

any)

d

end

righ

ts

the

cap

ital

gain

sra

te

mill

ion-

dol

lar

cap

D

ivid

end

sar

eta

xed

atth

eor

din

ary

rate

Stoc

kap

pre

ciat

ion

Rig

hts

that

repl

icat

eG

ener

ally

have

the

Gen

eral

lyha

veth

eU

nlik

eop

tion

sSA

Rs

righ

tsth

epr

ots

ofst

ock

op-

sam

eta

xtr

eatm

ent

assa

me

tax

trea

tmen

tas

are

mar

ked

tom

arke

tti

ons

Rar

ely

use

d

NQ

SOs

NQ

SOs

each

per

iod

an

dth

esi

nce

stoc

kop

tion

sd

iffe

renc

ebe

twee

nth

eha

vem

ore

favo

rabl

eac

-st

ock

pri

cean

dth

eex

-co

unt

ing

trea

tmen

ter

cise

pri

ceis

exp

ense

d(a

ndop

tion

sp

rovi

de

over

the

outs

tand

ing

the

sam

ebe

ne

tof

per

iod

ofSA

Rs

cash

less

exer

cise

thro

ugh

brok

erlo

ans

)

Not

eW

eth

ank

com

pen

sati

onco

nsul

tant

sSc

ott

Gre

enbe

rgan

dSc

ott

Ols

en(b

oth

ofT

ower

sP

erri

n)an

dFr

eder

icC

ook

(of

Fred

eric

WC

ook

and

Ass

ocia

tes)

for

hel

pfu

lcon

vers

atio

nsin

und

erst

and

ing

the

tax

and

acco

unt

ing

rule

s

14 Hall amp Liebman

The effect of taxation on the optimal contract (more precisely theshare of compensation that is performance-related) is ambiguous (evenin the absence of deductibility and deferral) because there are offsettingeffects First by reducing the share of corporate prots received byshareholders taxes diminish the importance to the shareholders of moti-vating the CEO Taken alone this effect would imply that taxes wouldbe expected to reduce the use of performance-based compensation Sec-ond because the government shares in the income received by the CEOthe variance of the CEOrsquos income is reduced raising his utility andlowering the cost to the rm of providing any given set of incentivesThird because taxes will lead executives to provide less effort for anygiven level of incentive-based pay the level of compensation that mustbe provided to compensate the CEO for effort is reduced for a givenamount of incentive These last two factors make it cheaper for the rmto offer contracts to the CEO with large incentive components when taxrates are higher and should therefore increase the use of incentive-based pay Since the net effect of taxation on the level of incentive-basedpay is ambiguous agency theory provides no strong predictions abouthow taxation should affect the optimal composition of option versuscash compensation

42 The Tax Advantages of DeferralWhile agency theory yields ambiguous predictions about the effect oftaxes on the use of options there are direct tax advantages of options inthat options provide executives with a way to defer compensation andthereby lower their taxes However because options also lead to a defer-ral of corporate tax deductibility the tax advantages from a combined(executive and corporate) perspective are not so straightforward In thissection we analyze and measure the tax advantages of stock optionsrelative to salary and bonus compensation We also analyze the taxadvantages of NQSOs relative to ISOs In particular we show how thetax advantages of options change as personal corporate and capitalgains taxes change We then show how the tax advantages of optionshave changed over time in response to changing tax rates

The crucial tax difference between standard options (NQSOs) andcash payment is that option payouts are deferred and the two forms ofcompensation earn different rates of return over the deferral period Anyanalysis of the relative tax advantages of two compensation instrumentsmust consider the tax consequences both to the employer and to theemployeemdashwhat Scholes and Wolfson (1992) call the global contractingperspective Thus in order to make valid comparisons between the twowe compare the tax burden to the executive while holding constant the

Taxation of Executive Compensation 15

posttax cost to the company By keeping posttax employer costs (in NPVterms) constant any package that is preferred by the employee is tax-advantaged in the global-contracting sense

Under this methodology a comparison of the tax advantages of op-tions and cash involves comparing a pretax cash payment of P with anoption payment that has the equivalent posttax NPV to the companyFor NQSOs it is straightforward to show that a pretax payment of P toan executive has exactly the same cost to the company as putting aside Pfor the purposes of paying stock-option payouts later That is if a com-pany pays an executive P today it will have the same amount of moneyin N years as if it had put aside P today let it accumulate at the rate ofreturn earned by the rm and then paid it out (as compensation in theform of option gains4) with any appreciation in year N

Assume that the pretax prot rate is r and the corporate tax rate is TcThen if the company pays P today it receives a deduction today of PTcwhich yields

PTc [1 1 r (1 2 Tc)]N (1)

in N years since the benets of the tax deduction are invested in thecompany and receive the after-tax corporate rate of return Converselyif the company puts P aside today then it grows at the after-tax corpo-rate rate of return in N years to give a tax deduction of

P [1 1 r (1 2 Tc)]N Tc (2)

which is the exact same value Note that the equivalence of these two taxdeductions is analogous to the tax-benet equivalence of front-loadedand back-loaded IRAs

An executiversquos payoff in N years from option prots is equal to Pcompounded at the after-tax corporate rate times 1 2 Tp where Tp is thepersonal tax rate Thus the combined payoffs in N years of the corporatededuction and the executiversquos posttax payoff is

P [1 1 r (1 2 Tc)]N Tc 1 P [1 1 r (1 2 Tc)]

N (1 2 Tp) (3)

where the rst term is the payoff from the corporate deduction and thesecond term is the executiversquos payoff

4 This analysis is not unique to options Any form of deferred compensation that enablesexecutives to invest inside the rm at a higher posttax return or without paying capitalgains taxes would have similar effects

16 Hall amp Liebman

The payoff to salary and bonus in N years is more complicated sincean assumption must be made about how the executive invests the origi-nal cash compensation For example if the executive invests in an instru-ment with taxable interest (eg bonds) then the N-year payoff from aninvestment of P accumulates at the post-corporate-tax post-personal-taxrate of return and is

P (1 2 Tp) [1 1 r (1 2 Tc)(1 2 Tp)]N (4)

However if an executive invests in (non-dividend-paying) equities theinvestment accumulates at the higher post-corporate-tax rate of returnwith the offsetting disadvantage that the capital gains are taxed at thecapital gains rate in year N So the executive receives

P [1 1 r (1 2 Tc)]N (1 2 Tp) 2 Tcg $P (1 2 Tp) [1 1 r (1 2 Tc)]

N 2P (1 2 Tp) (5)

where Tcg is the capital gains tax rate5Thus putting the equations together if an executive is paid cash P

and invests g in bonds and 1 2 g in equities the N-year payoff (includ-ing the payoff to the corporation from the tax deduction) is

P [1 1 r (1 2 Tc)]N Tc 1 g $(P [1 1 r (1 2 Tc)(1 2 Tp)]

N (1 2 Tp)1(1 2 g) (P [1 1 r (1 2 Tc)]

N (1 2 Tp) 2 Tcg $P (1 2 Tp)[1 1 r (1 2 Tc)]

N 2 P (1 2 Tp)) (6)

The tax advantage of options versus salary and bonus therefore is thedifference between the combined (corporate and executive) payoffs inequation (3) and the combined payoffs in equation (6)

43 The Tax Advantage of OptionsWe now turn to analysis of how changes in various tax rates affectthe tax advantages of options relative to cash compensation We focuson the case in which the cash earned by executives is invested entirelyin equities since equities are tax-favored relative to bonds The impor-tant conclusions of this analysis are not substantively different if we

5 In practice executives hold a combination of equities and bondlike instruments withtaxable interest Because equities are tax-favored in the next section we simplify the analy-sis by assuming that executives hold only equities Alternatively we could have assumedthat executives hold only less risky bonds since bonds more closely match the risk proleof a stream of tax savings None of our empirical results are substantively affected by thissimplication

Taxation of Executive Compensation 17

instead assume that a portion of an executiversquos holdings are invested inbonds

431 The Tax Advantage of Options Changing the Corporate TaxRate Stock options are tax-advantaged relative to cash simply becauseoption payouts are deferred allowing the executive to invest at the pre-personal-tax rate of return (with no capital gains tax at the end) ratherthan investing at the after-tax rate of return or paying the capital gainstax Since the advantage of being able to defer taxes is large when thecorporate tax rate is low the tax advantage of options is larger when thecorporate rate is low

To illustrate this point we dene the tax advantage of options to bethe dollar amount by which total after-tax option payoffs (to the corpora-tion and the executive) exceed the total payoffs from salary as denedby the difference of equation (3) and (6) assuming that P (the paymentto the executive) is equal to $100 and g 5 0 (cash compensation is in-vested in equities) We use a ten-year horizon since most options have aten-year term We then calculate how the tax advantage of optionschanges as the corporate tax rate varies from zero to 100 percent holdingthe personal tax rate and the capital gains rate xed at 40 percent and 20percent respectively (which approximates current rates)

The top panel of Figure 2 shows the tax advantage of options relativeto cashmdashprecisely the NPV of the total tax advantage of options for a$100 payment to the CEOmdashas the corporate tax rate changes As ex-pected the tax advantage of options declines as corporate tax rates in-crease At a zero corporate tax rate the tax advantage of options has anNPV of $7 This value is about $350 at a corporate rate of 40 percentWhen the corporate tax rate is 100 percent the tax advantage of optionscompletely disappears since the benets of deferred compensation fallto zero because the after-tax return on equities (and therefore the dis-count rate) falls to zero The tax advantage of options is the NPV of thetax savings from avoiding the capital gains tax on ten years of apprecia-tion which is why it is relatively modest even at the 40-percent corpo-rate rate

432 The Tax Advantage of Options Changing the Personal TaxRate Using the same assumptions as above (but this time holding thecorporate rate at 40 percent and the capital gains rate at 20 percent andvarying the personal rate) the tax advantage of options declines as thepersonal tax rate rises as shown in the middle panel of Figure 2 The taxadvantage of options is higher at low personal tax rates because theposttax base that gives rise to capital gains is higher and the advantage of

Note Holds constant the personal rate at 40 percent and the capital gains rate at 20 percent

Note Holds constant the corporate rate at 40 percent and the capital gains rate at 20 percent

Note Holds constant the personal rate at 40 percent and the corporate rate at 40 percent

FIGURE 2 Tax Advantage of Options NPV of Tax Advantage of Pay-ing $100 in Options (NQSOs) Rather Than in Cash

Taxation of Executive Compensation 19

avoiding the capital gains taxes is greater The tax advantage of optionstherefore declines linearly as the personal tax rate rises At a 100-percentpersonal tax rate the executive receives nothing in either case and byconstruction the companyrsquos tax deduction benet is the same in NPVterms so the tax advantage of options falls to zero also6

433 The Tax Advantage of Options Changing the Capital Gains TaxRate The tax advantage of options for various capital gains rates isdepicted in the bottom panel of Figure 2 under the same assumptions(this time holding both the personal rate and the corporate rate at 40percent while varying the capital gains rate) The tax advantage of avoid-ing the capital gains tax increases as the capital gains tax rate rises

5 HOW HAS THE TAX ADVANTAGE OF OPTIONSCHANGED OVER TIME

The analysis so far has illustrated how the tax advantage of optionsvaries with the changes in tax rates In this section we show how the taxadvantage of options has changed over time as tax rates have changed

51 Top Marginal Tax Rates over TimeThe evolution on the top marginal tax ratesmdashpersonal corporate andcapital gainsmdashfrom 1980 to 1998 is shown in Figure 3 In all cases thetop rate (the rate for taxpayers with the highest incomes) is shownwhich is not always the highest rate since various anomalies (such asclawbacks of exemptions) sometimes lead to marginal tax rates that arehigher than those for the highest income tax payers

The top personal tax rate was 70 percent in 1980 falling to 50 percentin 1982 and 28 percent in 1988 following the 1986 tax act7 The topmarginal tax rate has since risen to 396 percent but is effectively 425percent since there is a 29-percent Medicare surcharge (paid half by theemployer and half by the employee) that has no upper limit The topcorporate tax rate has had only one signicant change since 1980 fallingfrom 46 percent before the 1986 tax act to 34 percent following the act in1988 The top corporate tax rate was increased to 35 percent in 1993 The

6 This is one place where ignoring bond investments has substantive implications Inparticular in the all-bond case the tax advantage of options is nonmonotonicmdashit rises andthen fallsmdashas personal tax rates rise7 During the early 1980s the maximum tax on earned income limited the marginal tax rateon the earnings of high earners so many executives did not face a decline in personal taxrates between 1981 and 1982 The empirical results in this paper are robust to assumingthat executives faced a marginal tax rate of 50 percent in 1980 and 1981

20 Hall amp Liebman

FIGURE 3 Top Marginal Tax Rates since 1980

top capital gains rate has uctuated between 28 percent (approximately)and 20 percent since 1980 and is currently at 20 percent

52 The Evolution of the Tax Advantage of NQSOsThe tax advantage of options relative to cash dened in the same way asin section 4 from 1980 to 1998 is shown in Figure 4 The calculations aredone using the same assumptions as before (ie N 5 10 years and cashis invested in equities) and the statutory top marginal tax rates shown inFigure 3 are used Due to the fall in corporate and personal tax rates thetax advantage of options increased sharply after the Tax Reform Act of1986 This increase in tax advantage has largely been reversed in the1990s as top personal tax rates have crept back up and the capital gainsrate has been reduced

Two key facts emerge from this analysis First there is currently only amoderate tax advantage to standard nonqualied stock optionsmdashon theorder of $4 per $100 of compensation This is because the tax advantagesto the executive of deferring taxes are largely offset by the tax disad-vantages to the company of not being able to deduct option expensesfrom taxable prots until the executive exercises the options Secondalthough the 1986 tax act substantially increased the tax advantage ofoptions more than half of this increase has been reversed in the 1990s

Taxation of Executive Compensation 21

FIGURE 4 Tax Advantage of Options 1980 to 1998

NPV of tax advantage of $100 in options (NQSOs) rather than cash

53 The Tax (Dis)advantage of ISOsSo far the analysis has focused on the tax advantages of NQSOs sincethey are so much more prevalent than ISOs which account for onlyabout 5 percent of option grants The relative scarcity of ISOs can beexplained both by their tax status and by their per-executive cap of$100000 per year

Under what conditions are ISOs tax-preferred to NQSOs For a trans-fer P to an executive the ISO is always tax-preferred by the executive(since the capital gains rate is lower than the personal rate) while theNQSO is always tax-preferred from the companyrsquos perspective (since anISO is not deductible) The key issue however is the relative advan-tages of the two types of options from a global-contracting perspectiveBecause ISOs are not deductible the company is indifferent betweensetting aside P in the form of NQSOs today and setting aside P(1 2 Tc) inthe form of ISOs (Since the NPV of the tax deduction for NQSOs is PTcthe NPV of the payment P is P 2 PTc or P (1 2 Tc))

Equalizing the posttax cost of NQSOs and ISOs we need only look atthe payoff of the executive to determine the condition under which eachis preferred The payoff of NQSOs in N years is

P [1 1 r (1 2 Tc)]N (1 2 Tp) (7)

22 Hall amp Liebman

The same payoff for ISOs is

P (1 2 Tc) [1 1 r (1 2 Tc)]N (1 2 Tcg) (8)

ISOs therefore are tax-advantaged if

Tp 2 Tcg Tc 2 Tc Tcg (9)

Intuitively ISOs are tax-advantaged only if their advantage (the differ-ence between the personal rate and the capital gains rate) is largeenough to offset their cost (the disadvantage of not deducting at thecorporate rate) Note however that the condition is not a simple com-parison between the corporate rate and the personal rate minus thecapital gains rate (Tp 2 Tcg Tc)

Figure 5 shows how the tax advantage of NQSOs relative to ISOs haschanged since 1980 as tax rates have changed Unlike the very modestchanges (around 3 to 4 percent) in the relative tax advantages betweenoptions and cash shown in Figure 4 the relative tax advantage betweenNQSOs and ISOs has seen enormous swings When the top personalrate was 70 percent in 1980 ISOs were tax-favored by a margin greaterthan 6 percent However by 1982 (following the 1981 tax acts that dra-matically lowered the personal tax rate) ISOs became tax-disadvantaged

FIGURE 5 Tax Advantage of NQSOs Relative to ISOs over Time

NPV of tax advantage of $100 payment to executive NQSOs minus ISOs

Taxation of Executive Compensation 23

by about 5 percent By 1988 (following the 1986 tax act) ISOs became tax-disadvantaged by about 18 percent a dramatic swing The raising ofpersonal rates coupled with a decrease in the capital gains rate has sincereduced the tax disadvantage of ISOs to about $4 per $100 of pretaxcompensation Although hard data on ISOs are hard to come by practi-tioners (mostly compensation consultants) claim that ISOs were morepopular prior to 1982 when they were tax-advantaged Although ISOsare less disadvantageous from a tax standpoint relative to NQSOs thanthey were in the late 1980s they are still disadvantageous so it is notsurprising that they have not made a signicant resurgence in recentyears

6 EXPLAINING THE INCREASE IN OPTION-BASEDCOMPENSATION

In this section we analyze whether changes in tax rates affect the compo-sition of executive compensation and consider the relative importanceof tax factors and corporate governance factors in explaining the in-creasing share of compensation paid in the form of stock options Thenumerous changes in tax rates that have occurred since 1980 provide anopportunity to assess whether executive compensation arrangementsrespond to tax incentives in the way that the tax-avoidance model sug-gests We test this model using a panel data set of CEOs in largepublicly-traded US corporations Our identication of the tax effectsrelies on time-series variation in personal corporate and capital gainstax rates along with cross-sectional variation in corporate tax rates

61 DataWe use a 15-year panel data set of CEOs in the largest publicly-tradedUS corporations which is described in Hall and Liebman (1998) Thedata set covers the years from 1980 through 1994 and combines CEOcompensation information from corporate proxies and 10-K lings withstock price and stock return information from CRSP and accounting datafrom Compustat In addition some compensation data from the 1970swere collected in order to construct measures of the value of stock op-tions held by the CEOs in the rst period of the sample8

The feature that distinguishes our data from most other CEO datasets is that with our panel of yearly proxy data on option grants optiongains and total options held we are able to calculate the total value of

8 We are grateful to David Yermack for providing us with some of the data for 1984 to 1991See Yermack (1995) for a discussion of these data

24 Hall amp Liebman

all stock options held by the CEO at a point of time More importantlysince we have the details about the stock options held (number exer-cise price time to maturity etc) we can precisely calculate the changein the value of a CEOrsquos stock option holdings for a given change in rmvalue

62 Identication IssuesThe numerous tax changes over our sample period and the specic waysin which changes in personal corporate and capital gains rates arepredicted to affect the composition of CEO compensation give us un-usually rich sources of identication However because many of thechanges in tax rates would be expected to affect all of the CEOs in oursample in a similar way we will need to pay particular attention toseparating out the tax effects from underlying time trends and otherfactors that changed over time Controlling for underlying trends is par-ticularly important in this study because we are focusing on an out-comemdashthe use of stock optionsmdashthat has increased rapidly over thepast two decades and that many practitioners believe was importantlyaffected by nontax considerations Thus the tax effects we are trying toexplain are deviations from a rising trend

We take two steps to try to separate out the tax effects from theunderlying trend First we include key nontax factors that could poten-tially explain the increasing reliance on performance-based compensa-tion It has been suggested that the dramatic increase in incentive-basedpay is the result of the remarkably poor shareholder returns during the1970s which spurred the LBO and takeover movements of the 1980sAccording to this story (Kaplan 1997) shareholders became much morepowerful via the rise of institutional investors even as the LBO andtakeover movements waned Because the inuence of institutional inves-tors is thought to be one of the most important mechanisms of strongcorporate governance we use the share of each companyrsquos stock ownedby institutional investors (dened as institutions with more than $100million under management) as an explanatory variable9 The percentageof shares owned by large institutional investors increased from about 20percent to almost 50 percent in 1994 an upward trend that closelymatches the sharp rise in the share of compensation in the form ofoptions Indeed in our sample the annual average (over all of the rmsin our sample) share of stock owned by institutional investors has acorrelation of 09 with the annual average share of compensation paid inoptions

9 This variable is described in Gompers and Metrick (1998)

Taxation of Executive Compensation 25

In addition we include two other variables that proxy for strongercorporate governance The rst is the size of the board Evidence suggeststhat boards with smaller numbers of directors reduce the inuence of theCEO and better represent shareholder interests (Yermack 1995) Boardsize fell by approximately 15 percent between 1980 and 1994 The secondvariable is the fraction of the rmrsquos board members who are inside direc-tors or gray directors (noninsiders who have business dealings with thecompany) Firms with fewer outsiders are less likely to act in the share-holdersrsquo interests and we would predict that they would be less likely tohave performance-based compensation for their CEOs The average per-centage of inside and gray directors has decreased from about 45 percentin 1980 to about 35 percent in 1994

Second we allow for cross-sectional variation in corporate tax ratesCorporate marginal tax rates vary because rms may be eligible for tax-loss carrybacks and carryforwards investment tax credits and the alter-native minimum tax (see Auerbach 1986 Auerbach and Altshuler1990 Auerbach and Poterba 1987) Because these cross-sectional differ-ences imply that different rms should respond differently to a givenchange in tax rates they potentially provide us with a way to isolatethe tax effects Firm-specic corporate marginal tax rates are notori-ously difcult to calculate from publicly-available data We use a tricho-tomous variable that equals zero if the rm has tax-loss carryforwardsand negative earnings in a given year equals 05 times the statutorycorporate rate if the rm has only one of those conditions and equalsthe statutory rate if the rm has neither loss carryforwards nor negativeearnings10

The cross-sectional variation in the corporate tax rate may not be ex-ogenous Firms that perform poorly could face low marginal tax ratesand also be particularly likely or unlikely to provide performance-basedpay Because most of our specications use xed effects the correlationwe must be concerned about is between changes in rm performanceand changes in compensation We deal with this by controlling forlagged rm stock market performance in our regressions This shouldeliminate spurious correlation between tax rates and compensation thatis jointly caused by rm performance

10 Graham (1996) presents evidence that the trichotomous variable performs nearly as wellas his recommended simulated marginal tax rate Plesko (1999) compares marginal taxrates calculated from Compustat data with actual marginal tax rates from rm tax returnsHe nds that simple binary variables are more reliable measures of the marginal tax ratethan is either the Graham or the trichotomous variable We obtain substantively similarresults when we switch from the trichotomous variable to the binary variables recommendby Plesko

26 Hall amp Liebman

63 ResultsThe dependent variable in our regressions is the fraction of a CEOrsquos totalannual compensation (measured as the BlackndashScholes value of stock-option grants plus salary and bonus) that comes from stock-optiongrants We choose this reduced-form specication because our interest isin the composition of CEO compensation However the factors thataffect the composition of compensation might also affect the levels andit will be important to keep this in mind in interpreting our results

Optimal-contracting theory generally makes predictions about the in-centives provided by the total amount of rm stock and stock optionsowned by the CEO We choose to model the ow of option grants ratherthan the ultimate performance sensitivity of the CEOrsquos compensationand stock and stock-option holdings because we have in mind anadjustment-cost model in which it takes time for a CEOrsquos holdings ofstock and stock options to reach the optimal level11 In our regressionswe include two measures of the current performance sensitivity of theCEOrsquos wealth to rm performance as explanatory variables to reect thedistance of the CEOrsquos contract from the optimal contract

Table 3 presents results in which the tax effects are identied solely byvariation in tax rates over time The fraction of annual compensation paidin stock options is regressed on tax variables corporate governance vari-ables contract-theory variables and variables reecting stock market re-turns Both of the regressions in this table include rm xed effects Incolumn (1) the fraction in annual compensation paid in stock options isregressed on the log difference in payoff (the combined payoff to both therm and the CEO) from receiving compensation in stock options ratherthan salary and bonus where the payoff calculations are based on statu-tory marginal tax rates This variable is dened as the log difference be-tween equation (3) and equation (6) in section 4 The coefcient on the taxvariable is large and statistically signicant at the 95-percent level Thepoint estimate implies that a one-percent increase in the payoff differencebetween stock options and salary and bonus results in a 24-percentage-point increase in the share of compensation paid in stock options

The corporate governance variables all have the predicted sign Firmswith a higher fraction of their shares owned by institutional investorsuse more performance-based pay (stock options) Firms with large corpo-rate boards or a large share of inside directors are less likely to use stockoptions

The two contract-theory variables are the dollar change in CEO wealth

11 See Core and Guay (1999) for evidence that rms adjust in this way

TABLE 3Compensation Regressions that Rely on Variation in

Tax Rates over Time

(1) (2)

Tax variablesln (difference in payoff)(a) 2393 (0349)ln (12statutory personal 20015 (0034)

rate)ln (12statutory cap gains 0330 (0171)

rate)ln (12statutory corporate 0399 (0148)

rate)

Corporate governance variablesFraction of rmrsquos shares 0043 (029) 0052 (0028)

owned by institutionalinvestors

Size of the rmrsquos board of 20002 (001) 20003 (0001)directors

Fraction of directors who 20082 (033) 20092 (0030)are inside or gray

Contract-theory variablesD (CEO wealth) per $1000 20261 (0095) 20247 (0087)

in FV(b)

D (CEO wealth) per 10 in 2583 (189) 2554 (201)FV(c)

Volatility(d) 521 (424) 261 (369)ln (rm market value) 0019 (0006) 0019 (0006)

Stock market return variablesLagged annual rm stock 0006 (0009) 0003 (0008)

market returnLagged annual SampP 500 re- 20035 (0023) 20029 (0023)

turn

Regression diagnosticsTotal sample size 5179 5179Firms 427 427Adjusted R2 0313 0311

Dependent variable share of compensation paid in options Standard errors in parentheses Bothregressions in this table include rm xed effects

(a) Log difference in payoff is the log difference in combined rm and CEO after-tax share betweenpaying compensation in options and paying salary

(b) This variable is the dollar change in CEO wealth per $1000 change in rm market value (seeJensen and Murphy 1990)

(c) This variable in the dollar charge in CEO wealth per 10-percent change in rm value (see Hall andLiebman 1998)

(d) Standard deviation of rm market value in trillions

28 Hall amp Liebman

per $1000 change in rm value (Jensen and Murphy 1990) and thedollar change in CEO wealth per 10-percent change in rm value (Halland Liebman 1998) As discussed in Baker and Hall (1998) these twowealth sensitivity measures reect two different concepts of how closelyaligned the managersrsquo incentives are with the interests of the sharehold-ers In the regression they are meant to measure how far the CEOrsquosexisting contract and ownership of stock and stock options is from theoptimal level The negative coefcients on the two variables suggeststhat rms that are below their optimal pay-to-performance sensitivityare more likely to give more stock options in the current period

The second column replaces the tax variable motivated by deferraladvantages with the three statutory rates The corporate after-tax sharehas a positive coefcient as we would expect because the tax advantageof stock options increases as the corporate rate falls The coefcient onthe personal tax rate is small and statistically insignicant The sign ofthe coefcient on the capital gains rate is the opposite of that predictedby tax-avoidance theory

Note that these results rely on time-series variation alone This is not avery convincing test Since the log difference in payoffs mostly rises overthe time period covered in our sample as does the share of compensa-tion paid in options it is possible that the tax variable is simply reectingother factors that were changing over this time period that we have notincluded in our regression Therefore it is important to look at specica-tions that rely on cross-sectional variation in the tax variables

Table 4 contains specications that rely on cross-sectional variation incorporate tax rates The rst column repeats the specication from col-umn (1) of Table 3 but uses the trichotomous measures of rm corporatetax rates in place of the statutory rate in calculating each rmrsquos logdifference in payoff from paying compensation in options rather than insalary and bonus As in the previous table we nd the positive coef-cient on the log difference in payoff just as the theory suggest How-ever the coefcient estimate is now much smaller suggesting that aone-percent increase in the log difference in payoff produces a 01-percentage-point increase in the share of compensation paid in options

Introducing cross-sectional variation in the dependent variable doesnot eliminate the potentially spurious correlation over time between thetax variable and the share of compensation paid in options The secondcolumn of Table 4 isolates the pooled cross-sectional variation in corpo-rate tax rates by adding time dummies and dropping the rm xedeffects In this specication the coefcient on the tax variable is indistin-guishable from zero The coefcient on the institutional variable contin-ues to be large and statistically signicant The third column includes

Taxation of Executive Compensation 29

both rm xed effects and time dummies which tests for a relationshipbetween within-rm changes in the dependent variable and within-rmchanges in each explanatory variable In this specication the coefcienton the tax variable is again indistinguishable from zero The coefcientson the corporate governance variables are also statistically insignicant

The third column contains a specication that treats the Tax ReformAct of 1986 (TRA 86) as a natural experiment For rms and CEOs facingthe statutory corporate tax rate TRA 86 increased the relative payoff topaying compensation in options by 17 percent For a rm facing a zeromarginal tax rate it reduced the relative payoff by 45 percent The reduc-tion occurred because TRA 86 reduced the personal tax rate makingsalary and bonus more attractive while the decreased corporate rate hadno effect on zero-tax-rate rms Thus we would expect rms facing zeroor low marginal tax rates to reduce their use of options after TRA 86while rms facing the statutory rate would be expected to increase themFew rms face zero marginal tax rates year after year Therefore we tryto distinguish between rms that often face low tax rates and rms thatnearly always face the statutory rate We dene low-tax rms as oneswhose average marginal tax rate over 1984 to 1986 was below 02412

Then we ran a difference-in-differences regression to see if high-taxrms increased their use of options more than low-tax rms The interac-tion between post-TRA 86 and high corporate tax rates is the key vari-able The estimated coefcient on this variable is positive but small andnot statistically different from zero at the 95-percent level

It is important to note that in this specication as well the main corpo-rate governance variable the share of the rmrsquos shares owned by institu-tional investors performs as theory would predict The variable appearsto be quite robust Table 5 presents regressions of the share of compensa-tion paid in stock options on the fraction of the rmrsquos shares owned byinstitutional investors Since we have already established that these vari-ables have a strong longitudinal relationship the regressions here aredesigned to focus on the cross-sectional relationship between these twovariables The rst and third columns present regressions using data onlyfrom 1994 and nd a coefcient of roughly 02 (column 1 contains noother covariates while column 3 includes all of the other nontax covari-ates from the regressions in Tables 3 and 4) Since the median share of armrsquos stock owned by institutional investors in our sample rises from 19percent in 1980 to 49 percent in 1994 this coefcient would imply that this

12 This cutoff was chosen after inspecting the distribution of tax rates and essentiallyseparates rms that are always at or near the statutory rate from ones that are sometimesbelow it

TA

BL

E4

Com

pens

atio

nR

egre

ssio

nsth

atR

ely

onC

ross

-Sec

tion

alV

aria

tion

inC

orpo

rate

Tax

Rat

es

(1)

(2)

(3)

(4)

Tax

vari

able

s(a)

ln

(dif

fere

nce

inp

ayof

f)0

099

(00

41)

20

038

(00

36)

20

005

(00

44)

Pos

t-T

RA

862

000

5(0

013

)H

igh

corp

orat

eta

xra

te0

018

(00

17)

(Pos

t-T

RA

86)

3hi

gh0

024

(00

16)

corp

orat

eta

xra

te

Cor

por

ate

gove

rnan

ceva

riab

les

Frac

tion

of

rmrsquos

shar

es0

109

(00

27)

011

4(0

015

)2

000

9(0

030

)0

087

(00

34)

owne

dby

inst

itu

tion

alin

vest

ors

Size

ofth

e

rmrsquos

boar

dof

20

004

(00

01)

20

001

(00

01)

20

002

(00

01)

000

1(0

002

)d

irec

tors

Frac

tion

ofd

irec

tors

wh

o2

012

4(0

033

)2

000

3(0

018

)2

004

1(0

034

)0

001

(00

39)

are

insi

de

orgr

ay

Con

trac

t-th

eory

vari

able

sD

(CE

Ow

ealt

h)p

er$1

000

20

238

(00

94)

20

167

(00

53)

20

257

(00

96)

20

021

(01

47)

inFV

(b)

D(C

EO

wea

lth)

per

10

in2

560

(17

8)2

592

(10

5)2

594

(19

6)2

147

(04

86)

FV(c

)

Vol

atili

ty(d

)2

48(4

24)

162

(03

4)1

12(4

28)

106

(06

09)

ln(

rmm

arke

tva

lue)

003

3(

006)

000

8(0

003

)0

009

(00

06)

001

2(0

007

)

Stoc

km

arke

tre

turn

vari

able

sL

agge

dan

nual

rm

stoc

k2

000

2(0

009

)0

011

(00

10)

000

6(0

009

)2

001

2(0

013

)m

arke

tre

turn

Lag

ged

annu

alSamp

P50

02

005

6(0

023

)2

005

3(0

051

)2

003

9(0

049

)2

015

4(0

040

)re

turn

Reg

ress

ion

dia

gnos

tics

T

otal

sam

ple

size

5179

5179

5179

2534

Firm

s42

742

739

6A

dju

sted

R2

031

009

032

005

Firm

xed

effe

cts

Yes

No

Yes

No

Tim

ed

um

mie

sN

oY

esY

esN

o

Dep

end

ent

vari

able

isfr

acti

onof

ann

ualc

omp

ensa

tion

from

stoc

kop

tion

sSt

and

ard

erro

rsin

par

enth

eses

(a)

All

thre

esp

eci

cati

ons

use

tric

hoto

mou

sco

rpor

ate

tax

rate

sT

he

sam

ple

inco

lum

n(3

)u

ses

only

the

year

s19

84to

1988

sin

cew

ear

eu

sing

1986

asa

nat

ural

exp

erim

ent

(b)

Thi

sva

riab

leis

the

dol

lar

chan

gein

CE

Ow

ealt

hp

er$1

000

chan

gein

rm

mar

ketv

alu

e(s

eeJe

nse

nan

dM

urp

hy

1990

)

(c)

Thi

sva

riab

lein

the

dol

lar

char

gein

CE

Ow

ealt

hp

er10

-per

cent

chan

gein

rm

valu

e(s

eeH

alla

nd

Lie

bman

199

8)

(d)

Stan

dar

dd

evia

tion

ofr

mm

arke

tva

lue

intr

illio

ns

32 Hall amp Liebman

TABLE 5Large Institutional Investors and Stock-Option Compensation

The Cross-sectional Relationship between Institutional Investors andOption Grants

(1) (2) (3) (4)

Coefcient on fraction of 217 (067) 145 (014) 177 (069) 120 (015)rmrsquos shares owned byinstitutional investors

Sample period 1994 1980ndash1994 1994 1980ndash1994Contract-theory and stock No No Yes Yes

market covariatesTime dummies No Yes No Yes

The dependent variable is the share of compensation paid in options

factor alone can explain 6 percentage points of the 23-percentage-pointincrease in the median share of compensation paid in options that oc-curred over this time period Of course this variable is only one measureof the strengthening of corporate governance so the total impact of im-proved corporate governance could be greater The second and fourthcolumns of the table use the entire 1980ndash1994 sample and include timedummies to isolate the cross-sectional correlations Again there is evi-dence of a strong cross-sectional relationship between the two variablesalthough the coefcients are slightly smaller perhaps reecting a weakercorporate-governance role in the earlier period In sum there is a strongrelationship in both longitudinal and cross-sectional data between theinstitutional-investor variable and the options-percentage variable13

7 THE MILLION-DOLLAR RULE

In 1993 section 162(m) of the Internal Revenue Code was enactedlimiting the deductibility of executive compensation in excess of onemillion dollars unless the compensation was performance-related Pro-ponents of this legislation argued that it would slow the growth ofexecutive pay and tighten the link between rm performance and execu-tive compensation

The new provision became effective for tax year 1994 and applies tothe CEO and the other four most highly compensated executives in eachcompany Pay that is performance-related such as stock-option grants

13 The one specication in which the fraction owned by institutional investors is notsignicant is in column (2) of Table 4 a regression that includes both xed effects and timeeffects

Taxation of Executive Compensation 33

and bonuses paid for meeting clear performance goals is not affected bythis regulation provided that it has been approved by shareholdersThus the main impact of the provision is to limit the deductibility ofsalary in excess of a million dollars Woodlock and Antenucci (1997)studied the proxy statements of 376 rms and documented that mostrms responded to the new law by qualifying top executive pay asperformance-related Perry and Zenner (1999) identied 25 rms thatreduced salaries from above $1 million to below $1 million and foundthat 23 of the rms cited section 162(m) as their reason for doing so

In this section of the paper we investigate whether this provision hasaffected the level or structure of executive pay using the Execucompdata base which contains information on the highest-paid executives ineach of the rms in the SampP 500 SampP Midcap 400 and SampP SmallCap600 Between 1993 and 1998 the median salary of the Execucomp execu-tives rose from $230000 to $306000 The median value of total compen-sation (salary bonus and option grants) rose from $470000 to $882000Limiting the sample of CEOs indicates that the total compensation of themedian CEO rose from $11 million to $18 million over the period whilethe median salary rose from $441000 to $550000 Thus executive com-pensation clearly continued to rise at a rapid rate after the implementa-tion of the million-dollar rule though perhaps at a slower rate than itotherwise would have

If the regulation affects behavior it should have the largest impact onrms whose executives are receiving salaries that are above the limit orjust below it Thus we might expect to see rms whose executives arereceiving relatively high salaries to rely more heavily on performance-based pay and less heavily on salary in the years following the regulationPerry and Zenner (1999) and Livingston (1999) both nd that this is thecase This trend can be clearly identied in the Execucomp data set Forexample over 1993 to 1998 the median annualized growth rate of salaryfor executives whose 1993 salary was below $500000 in the previous yearwas 85 percent For executives with salary between $500000 and$700000 it was 64 percent for executives between $700000 and $850000it was 50 percent for executives between $750000 and $1 million it was31 percent and for executives above $1 million it was 00 percent14

The observation that executives at lower levels of salary had highergrowth rates is not very strong evidence of an impact of section 162(m)however because it is possible that low-salary executives would havehad higher growth rates even in the absence of the provision To explore

14 Of the 3137 executives for whom we have complete data between 1993 and 1998 2708had 1993 salaries below $500000 while only 35 had 1993 salaries above $1 million

34 Hall amp Liebman

whether this is the case we use Execucomp data from both before andafter the law change Ideally we would want to have data from manyyears before the change to examine whether there is an underlying pat-tern of faster salary growth for low-salary executives Unfortunately theExecucomp data begin in 1992 and therefore provide us with only the1992ndash1993 growth rate in compensation to use as a baseline15

We run regressions pooling data from 1992ndash1993 through 1997ndash1998of the annual percentage change in different forms of compensation on avariable million that is designed to measure the likelihood that anexecutiversquos compensation will be affected by section 162(m) We denemillion as the minimum of 1 and the executiversquos salary divided by $1million in the previous year Thus any executive with salary at or above$1 million would receive a value of 1 while an executive with a salary of$500000 would be coded as a 0516 We include this variable by itself andalso interacted with a dummy variable (after) that equals 1 for years1994 and after Thus the million variable is designed to pick up anyunderlying relationship between the level of salary and the growth rateof salary while the interacted one will identify any differential growthrate of salary by income in the period after section 162(m) took effect Wealso include a full set of year dummies the natural logarithm of thermrsquos market value and the rmrsquos annual rate of return over the twoprevious years as control variables To reduce the sensitivity of our re-sults to outliers we run quantile regressions and robust regressions

Table 6 contains our results The rst four columns are for all topexecutives while the last four columns focus on CEOs Columns (1) and(2) present estimates from robust regression for the percentage change insalary and the percentage change in total compensation (salary bonusstock grants and stock-option grants) In the salary growth regressionsfor all executives the coefcient on million is negative and signicantindicating that there is an underlying relationship between the level ofexecutive compensation and its subsequent growth rate The coefcienton million 3 after is negative as well suggesting that the negativerelationship between the level of salary and its subsequent growth rateintensied slightly after 1994 The relatively small magnitude suggeststhat an executive with a $1 million dollar salary would see his salarygrow at an annual rate that would be about 06 percent slower than anexecutive earning $500000

15 The HallndashLiebman data set contains only a combined salary and bonus variable makingit impossible to use it to analyze this issue16 We also ran the regressions (not reported) allowing for a nonlinear relationship betweenmillion and compensation growth and obtained substantively similar results

TA

BL

E6

The

Eff

ect

ofth

eM

illi

on-D

olla

rR

ule

onSa

lary

and

Tot

alC

ompe

nsat

ion

All

top

exec

uti

ves

CE

Os

only

Rob

ust

regr

essi

onQ

uan

tile

regr

essi

onR

obu

stre

gres

sion

Qu

anti

lere

gres

sion

D

in

Din

D

in

Din

D

in

Din

D

in

Din

sala

ryto

talc

omp

(a)

sala

ryto

talc

omp

sa

lary

tota

lcom

p

sala

ry

tota

lcom

p

mil

lio

n(b

)2

006

02

035

72

006

32

024

12

008

72

023

12

010

32

024

5(0

005

)(0

032

)(0

006

)(0

027

)(0

010

)(0

073

)(0

011

)(0

056

)m

illi

on

32

001

10

026

20

021

20

015

20

016

014

92

003

40

158

aft

er(c

)(0

005

)(0

033

)(0

006

)(0

029

)(0

010

)(0

076

)(0

011

)(0

057

)N

359

5035

950

359

5035

950

757

17

589

757

17

589

Stan

dar

der

rors

inp

aren

thes

esR

egre

ssio

nsal

soin

clu

de

year

du

mm

ies

stoc

km

arke

tre

turn

and

log

mar

ket

valu

e

(a)

Tot

alco

mp

ensa

tion

incl

ud

essa

lary

bon

us

rest

rict

edst

ock

and

stoc

k-op

tion

gran

ts

(b)

mil

lion

isd

ene

das

min

(1s

alar

yin

pre

viou

sye

ars

div

ided

by$1

mil

lion)

(c)

aft

eris

anin

dic

ator

vari

able

equ

alto

zero

befo

reth

ere

gula

tion

took

plac

e(1

993)

and

one

afte

rwar

d

36 Hall amp Liebman

In the second column the coefcient on million 3 after is essentiallyzero suggesting that any decrease in salary brought about by section162(m) was offset by increases in bonus and stock-option grants Theresults are similar for the median regressions There is some evidence ofa very small slowdown in growth of salary for executives likely to beaffected by section 162(m) but no sign of a change in total compensa-tion Taken together this provides evidence of a very small substitutionof performance-related pay for salary

The last four columns run identical tests but limit the sample to CEOsThe interaction of million 3 after continues to be small and negative inthe salary regressions In contrast the impact of 162(m) on total compen-sation appears to be positive and fairly substantial (about 15 percentagepoints per year) in these regressions That is the increase in bonus andstock options following the million-dollar rule more than offset the verysmall decline in salary

All of these conclusions should be interpreted somewhat cautiouslybecause our data provide us with only one pre-1993 control year Never-theless the data are consistent with section 162(m) having led to a veryminor slowdown in salary growth and one that is dramatically smallerthan what one would conclude without controlling for underlying differ-ential salary growth rates at different levels of salary Since total compen-sation for CEOs did not decline and perhaps increased there is evidenceof a minor substitution of performance-related pay for salary in responseto the regulation

8 EXECUTIVE COMPENSATION AND THEELASTICITY OF TAXABLE INCOME

Even if taxes have only a minor effect on executive compensation thestructure of executive compensation is important for tax policy be-cause high-income taxpayers are sources of signicant revenue and arethe focus of important debates about the efciency cost of taxation

Goolsbee (1999) has recently argued that executive decisions to exer-cise options are highly responsive to intertemporal differences in taxrates created by tax reforms He claims that much of the apparent declinein taxable income for high-income taxpayers between 1992 and 1993documented by Feldstein and Feenberg (1996) can be explained by execu-tives shifting option gains into 1992 in order to avoid the higher marginaltax rates that went into effect in 1993 By extension he implies that thehigh elasticities of taxable income with respect to personal tax ratesestimated by comparing the pre- and post-tax-reform taxable income of

Taxation of Executive Compensation 37

high-income individuals reect intertemporal shifting of income and notpermanent effects of taxation If this is true then the deadweight loss oftaxation is lower as is the cost of progressivity

While Goolsbee presents comprehensive evidence that option gainswere unusually high in 1992 his identication of the impact of tax rates onthe timing of option gains depends on correctly specifying a counter-factual level of options that would have been exercised in the absence ofthe tax change He simply assumes a linear time trend The purpose ofGoolsbeersquos time trend is presumably to account for the increasing use ofoptions over time However since options typically require a vestingperiod of at least a couple of years a much more direct measure of ex-pected option gains can be constructed based on the total value of vestedoptions held by the CEO Since options typically have vesting periods thatexceed two years this total can be treated as exogenous since it will not beaffected by the tax changes

More generally the value of options held by the CEO is likely to havea major effect on his exercise decisions If his options are out of themoney he will not have gains to realize If there has been a recent run-up in the market an optimizing CEO may exercise an unusually largeamount of options in order to diversify his overall portfolio and if in-the-money options are about to expire they will be exercised in thecurrent year

Our evidence suggests that much of the apparent tax shifting caninstead be attributed to stock market performance and the timing of pastoption grants Using our 15-year panel we replicate Goolsbeersquos (1999)result for the 1993 period but show that the impact of taxes is notpresent in other time periods Moreover even the result for the 1993period disappears when appropriate controls for past option grants andstock price appreciation are included

The top three rows of Table 7 present data on the taxable incomes ofCEOs in the HallndashLiebman sample For these results our panel has beennarrowed to include only CEOs whose rmrsquos scal years correspondwith tax years In addition we create a balanced panel for each of thethree tax-reform periods by excluding any rm for which there are notcomplete data for the period surrounding the reform

The fourth row of the table shows that the 1993 tax act raised the topmarginal personal income tax rate from 310 to 396 percent effective fortax year 1993 with a further increase occurring in 1994 due to the uncap-ping of the Medicare payroll tax The top row of the table shows thattaxable income was higher in 1992 than for any other year in our sample(all data are in 1994 dollars) Option gains were particularly high in1992mdashmore than double their 1991 level and 68 percent higher than their

TA

BL

E7

Tax

Ref

orm

Sto

ckM

arke

tP

erfo

rman

cea

ndth

eT

imin

gof

Opt

ion

Gai

ns

Ave

rage

stoc

km

arke

tM

ean

CE

OM

ean

CE

Ore

turn

for

Mea

nva

lue

taxa

ble

sala

ryan

dM

ean

CE

OT

opfe

der

al

rms

inof

opti

ons

(Val

ue

ofop

tion

Sam

ple

inco

me

bonu

sop

tion

gain

sp

erso

nal

sam

ple

for

held

atbe

gga

ins)

(va

lue

Ref

orm

size

Yea

r($

mil

)($

mil

)($

mil

)M

TR

pre

vye

ar(

)of

year

ofop

tion

she

ld)

ER

TA

8127

919

800

874

064

80

163

7025

80

460

354

1981

083

10

672

009

869

245

070

61

3919

820

872

067

90

099

5011

90

686

144

1983

099

00

724

025

250

355

116

62

16

TR

A86

295

1985

119

70

837

024

050

76

116

02

0719

861

448

091

70

345

5042

81

776

194

1987

165

50

959

059

338

519

12

229

266

1988

163

21

058

039

028

07

237

71

6419

891

656

105

20

414

2817

92

345

176

1993

272

1990

209

51

028

061

828

232

288

02

15T

axA

ct19

911

875

102

00

547

312

01

268

82

0319

922

678

107

01

250

3146

64

552

275

1993

232

41

209

074

339

620

64

299

173

1994

196

41

262

037

842

515

45

113

074

Sam

ple

isH

allndash

Lie

bman

rm

sw

ith

Dec

embe

r

scal

year

sth

atha

dco

mp

lete

dat

afo

ron

eof

the

thre

eta

x-re

form

per

iod

sst

ud

ied

Taxation of Executive Compensation 39

1994 level These patterns are consistent with Goolsbeersquos claim that tax-payers responded to the anticipated increase in marginal tax rates byshifting gains into 1992

The evidence from the other two tax reforms is less clear suggestingthat there might be factors other than tax rates determining the timing ofoption gains ERTA 1981 reduced the top marginal tax rate from 70percent in 1980 to 50 percent in 1982 However there is no sign thatoption gains were reduced in 1980 and 1981 and then increased in 198217

There was a big increase in option gains in 1983 however Similarly thehighest year of stock-option gains in the period around 1986 was 1987 astrange year to take gains since taxpayers would have known that thepersonal rate would be 10 percentage points lower if they waited untilthe following year18

What all the peak years of option gains have in common is that theyfollow years of strong stock market performance The fth row of Table 7shows the average stock market return during the previous year for therms in our sample The large value of option gains in 1983 followed a355-percent increase in the stock market The 1992 boom in option gainsfollowed a 466-percent increase in the stock market the more modest1987 peak in option gains followed a 19-percent gain (and a 43-percentgain two years before)

Strong stock market performance raises the value of options availableto exercise The sixth row of the table shows the mean value of optionsheld by CEOs at the beginning of the year All three of the peak years ofoption gains correspond with peak years in terms of the value of optionsavailable for exercise Since stock market appreciation during the yearalso increases the value of options available to exercise later in that yearwe need to allow for current-year stock market appreciation as well

Table 8 contains Goolsbee-style regressions for each of the tax re-forms The rst two columns reprint the results from Goolsbee (1999)The third and fourth columns replicate his results in our sample Inparticular with a specication similar to his we are able to replicate hisnding that the current after-tax share has a positive coefcient and thefollowing yearrsquos after-tax share has a coefcient of roughly negative oneColumns (5) through (8) show that this pattern was unique to the 1993tax reform Indeed in the TRA 86 regressions the coefcient on the

17 As we explained above for executives covered by the maximum tax on earned incomethere was not a decline in tax rates between 1981 and 198218 It is possible that the large stock market decline in October 1997 may have promptedsome people to exercise options earlier than they otherwise would have However anincrease in option exercises typically follows stock market increases not declines

TA

BL

E8

Att

empt

sto

Rep

lica

teth

eT

ax-S

hift

ing

Res

ult

inO

ther

Tim

eP

erio

ds

Goo

lsbe

eG

ools

bee

1993

1993

Ent

ire

Ent

ire

regr

essi

onre

gres

sion

refo

rmre

form

TR

A86

TR

A86

ER

TA

81E

RT

A81

1980

ndash199

419

80ndash1

994

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

ln[1

2ta

x(t)

]1

244

108

20

455

032

62

017

62

017

22

016

72

016

70

026

20

068

(01

34)

(01

27)

(05

37)

(05

42)

(02

10)

(02

09)

(00

90)

(00

91)

(00

67)

(00

65)

ln[1

2ta

x(t1

1)]

20

867

20

764

21

214

20

917

047

80

722

20

103

004

30

176

018

3(0

118

)(0

112

)(0

491

)(0

499

)(0

157

)(0

174

)(0

072

)(0

072

)(0

068

)(0

066

)

Tim

e0

135

007

62

000

52

036

40

041

000

80

094

003

50

056

002

7(0

008

)(0

008

)(0

049

)(0

049

)(0

028

)(0

028

)(0

025

)(0

026

)(0

003

)(0

003

)

ln(m

arke

tva

lue)

066

10

489

023

70

334

039

5(0

015

)(0

058

)(0

039

)(0

042

)(0

017

)

Yea

rs19

91ndash1

995

1991

ndash199

519

90ndash1

994

1990

ndash199

419

85ndash1

989

1985

ndash198

919

80ndash1

983

1980

ndash198

319

80ndash1

994

1980

ndash199

4

Ad

just

edR

20

720

770

610

630

690

760

800

810

630

68

Obs

erva

tion

s13

856

132

371

385

136

41

470

143

31

113

106

23

968

385

9

Dep

end

ent

vari

able

isth

elo

gof

taxa

ble

inco

me

All

regr

essi

ons

con

tain

rm

xe

def

fect

sG

ools

bee

regr

essi

on(2

)al

soco

ntai

nstw

oot

her

vari

able

sno

tsh

own

inth

eta

ble

Stan

dar

der

rors

inp

aren

thes

es

Taxation of Executive Compensation 41

TABLE 9Alternative Explanations for the Timing of Option Exercises

1993 Entirereform TRA 86 ERTA 81 1980ndash1994

(1) (2) (3) (4)

ln[12tax(t)] 0108 0111 20051 0201(0324) (0136) (0048) (0057)

ln[12tax(t11)] 20134 0573 0125 0150(0336) (0156) (0053) (0066)

ln(market value) 0418 0177 0338 0418(0061) (0039) (0046) (0163)

Firmrsquos stock market return (t) 0298 0268 0214 0258(0044) (0045) (0031) (0023)

Firmrsquos stock market return 0162 0206 0031 0128(t21) (0041) (0008) (0029) (0021)

ln(value of options held by 0004 0008 0001 0009CEO at start of year) (0003) (0003) (0002) (0001)

Years 1990ndash1994 1985ndash1989 1980ndash1983 1980ndash1994Adjusted R2 064 071 082 068Observations 1349 1427 1036 3812

The dependent variable is the log of taxable income All regressions contain rm xed effects Stan-dard errors in parentheses

following yearrsquos tax rate is positive and signicant Columns (9) and (10)pool all three reforms and fail to replicate the pattern that would reectintertemporal shifting19

It is possible that the ease with which one can shift the timing ofoption gains is asymmetric It may be harder to shift option gains for-ward in time since some options expire and therefore must be exercisednow In addition options were a much smaller portion of compensationin the early 1980s so perhaps the ERTA 81 results are not surprisingNonetheless these results cast doubt on the proposition that this is thechannel through which spurious estimates of the elasticity of taxableincome have occurred for previous tax reforms and they suggest thatwe should look harder for alternative explanations for the timing ofoption gains

Table 9 attempts to predict options gains using the lagged and currentincrease in a rmrsquos market value and the value of the options held by theCEO at the start of the year The two stock market variables stronglypredict option gains while the value of options held has the correct sign

19 The results also fail to suggest large permanent effects of marginal tax rates on taxableincome

42 Hall amp Liebman

in all of the specications and is signicant in the pooled sample eventhough it is highly correlated with the two stock market return variablesIn contrast the tax-timing explanation fails even in the 1993 period

It is important to realize that our results do not invalidate Goolsbeersquoscritique of claims that aggregate revenue data from 1992 and 1993 provethat elasticities of taxable income are large We have simply given analternative explanation for why taxable incomes were unusually high in1992 However our results do cast doubt on the conclusion that largetiming shifts are ubiquitous and capable of explaining away the morecareful estimates of taxable income elasticities that have been performedusing micro data (eg Feldstein 1995 Auten and Carroll 1995)

One further point is in order It is sometimes argued that the combina-tion of increased use of options and the booming stock market explainswhy federal tax revenues have been so much higher than expected dur-ing the 1990s However while large option gains increase revenue fromthe personal income tax they produce offsetting deductions for cor-porations The net impact on revenue is only the difference in the twomarginal tax rates times the option gain While the gap between thepersonal rate and the effective corporate rate is nontrivial the impact oflarge option exercises on total revenue is substantially smaller than theamount by which the exercises inate personal tax revenue and reducecorporate tax revenue

9 CONCLUSION

We have described the tax rules for executive pay in detail and analyzedhow changes in various tax rates affect the tax advantages of stock optionsrelative to salary and bonus We nd that there is a moderate tax advan-tage to optionsmdashcurrently about $4 per $100 in pretax executive paymdashand that changes in the tax advantage of options over time have had atmost a modest impact on the composition of pay Corporate governancefactors particularly the role of large investors appear to be more impor-tant in explaining the dramatic increase in option pay

In addition our evidence suggests that more direct attempts to use taxpolicy to inuence executive compensation have had little effect Wend that the million-dollar rule led companies to substitute perfor-mance-based pay for salary But our evidence suggests that this substitu-tion was quite modest and there is no evidence that the total level of paywas reduced Overall although the stock-option explosion has dramati-cally increased the link between pay and performance this change isdue almost entirely to nontax factors

Taxation of Executive Compensation 43

REFERENCES

Auerbach Alan J (1996) ldquoThe Dynamic Effects of Tax Law Asymmetriesrdquo Re-view of Economic Studies 53451ndash405

mdashmdashmdash and James Poterba (1987) ldquoTax Loss Carryforwards and Corporate TaxIncentivesrdquo In The Effects of Taxation on Capital Accumulation M Feldstein(ed) Chicago University of Chicago Press

Altshuler Rosanne and Alan Auerbach (1990) ldquoThe Signicance of Tax LawAsymmetries An Empirical Investigationrdquo Quarterly Journal of Economics 10561ndash86

Auten Gerald and Robert Carroll (1995) ldquoThe Effect of Income Taxes on House-hold Behaviorrdquo Ofce of Tax Analysis US Department of the Treasury

Baker George P and Brian J Hall (1998) ldquoCEO Incentives and Firm SizerdquoNBER Working Paper no 6868 December

Core John and Wayne Guay (1999) ldquoThe Use of Equity Grants to ManageOptimal Equity Incentive Levelsrdquo Draft

Feldstein M (1995) ldquoThe Effect of Marginal Tax Rates on Taxable Income APanel Study of the 1986 Tax Reform Actrdquo Journal of Political Economy 103 (no3)551ndash72

mdashmdashmdash and D Feenberg (1996) ldquoThe Effect of Increased Tax Rates on TaxableIncome and Economic Efciency A Preliminary Analysis of the 1993 Tax RateIncreasesrdquo In Tax Policy and the Economy vol 10 James M Poterba (ed)Cambridge MA MIT Press

Gompers Paul A and Andrew Metrick (1998) ldquoInstitutional Investors andEquity Pricesrdquo NBER Working Paper no 6723

Goolsbee Austan (1999) ldquoWhat Happens When You Tax the Rich Evidencefrom Executive Compensationrdquo NBER Working Paper no 6333

Graham John R (1996) ldquoProxies for the Corporate Marginal Tax Raterdquo Journal ofFinancial Economics 42187ndash221

Hall Brian J (1999) ldquoThe Design of Multi-year Stock Option Plansrdquo Journal ofApplied Corporate Finance 12(2)97ndash106

mdashmdashmdash and Jeffrey B Liebman (1998) ldquoAre CEOs Really Paid Like Bureau-cratsrdquo Quarterly Journal of Economics CXII(no 3)653ndash691

Jensen Michael C and William M Meckling (1976) ldquoTheory of the Firm Mana-gerial Behavior Agency Costs and Ownership Structurerdquo Journal of FinancialEconomics III305ndash360

mdashmdashmdash and Kevin J Murphy (1990) ldquoPerformance Pay and Top-ManagementIncentivesrdquo Journal of Political Economy XCVIII225ndash264

Kaplan Steven N (1997) ldquoThe Evolution of US Corporate Governance WeAre All Henry Kravis Nowrdquo Draft

Livingstone Jane R (1999) ldquoExecutive Compensation Contracting and Re-sponses to Increased Tax Costsrdquo Louisiana State University June

Miller Merton H and Myron S Scholes (1982) ldquoExecutive CompensationTaxes and Incentivesrdquo In Financial Economics Essays in Honor of Paul CootnerWilliam F Sharpe and Cathryn M Cootner (eds) Princeton NJ Prentice-Hall

Murphy Kevin J (1999) ldquoExecutive Compensationrdquo forthcoming in OrleyAshenfelter and David Card (eds) Handbook of Labor Economics Vol 3 NorthHolland

Perry Todd and Marc Zenner (1999) ldquoPay for Performance Government Regu-

Taxation of Executive Compensation 7

wishes) slowly over time Common vesting periods are in the three- tove-year range and options usually vest linearly (eg a four-year op-tion vests at 25 percent at the end of each year) An executive typicallyloses any unvested options upon departure Although options may beexercised as soon as they vest they do not have to be exercised untilthey expire or mature Almost 85 percent of stock-option plans have aterm of exactly ten years with virtually all of the remainder being in theve- to ten-year range About 95 percent of options are granted at themoney or at fair market value which means that the exercise price at grantdate is set equal to the stock price at grant date The remaining 5 percentare either discount options (so-called in-the-money options where theexercise price is below the stock price at grant date) or premium options(so-called out-of-the-money options where the exercise price is abovethe stock price at grant date) The holders of options typically do nothave dividend rights or voting rights even on vested options3

31 Tax Rules and Stock-Option CompensationUnlike salary and bonus stock-option grants are typically an untaxedevent at the time of grant For the most widely used optionsmdashnonquali-ed stock options (NQSOs)mdashexecutives are taxed at the personal incometax rate on option prots (the difference between that stock price and theexercise price times the number of options) when the options are exer-cised The company receives a parallel deduction against corporate in-come at that point If the executive continues to hold the shares afterexercise any subsequent appreciation is taxed at the capital gains rate inthe usual way In 1993 an additional feature was added to the tax code[Internal Revenue Code section 162(M)] that disallowed a corporate de-duction for any executive pay above $1 million that is not performance-based While this rule affects executive salaries most bonuses qualify asperformance-based and standard stock options automatically qualifyTherefore this provision gives companies with highly paid executives anincentive to give more pay in the form of bonuses and stock options asubject we return to in section 7 A summary description of the tax (andaccounting) treatment of cash and option compensation is in Table 1

A far less common type of option which is estimated to account forabout 5 percent of option grants is the incentive stock option (ISO) WhileISOs are similar to NQSOs in their design they are crucially different intwo respects First they have an annual cap of $100000 per executive

3 See Murphy (1999) for details about stock options and Miller and Scholes (1982) on taxincentives

TA

BL

E1

Tax

and

Acc

ount

ing

Tre

atm

ent

ofE

xecu

tive

Com

pens

atio

nC

ash

Pay

men

tsve

rsus

Stan

dard

Exe

cuti

veO

ptio

ns

Tax

atio

nat

Tax

atio

nat

Acc

ount

ing

Com

pens

atio

nD

escr

ipti

onp

erso

nall

evel

corp

orat

ele

vel

trea

tmen

t

Sala

ryan

dbo

nu

sC

ash

pay

men

tsm

ade

Tax

edas

ord

inar

yFu

llyd

edu

ctib

leS

ince

Exp

ense

dag

ains

tea

rn-

toC

EO

in

com

e19

93a

ny

pay

over

$1in

gsin

the

usu

alw

ay

mill

ion

isno

td

educ

t-ib

leu

nles

sit

isp

er-

form

ance

-bas

ed

Non

-qua

lie

dst

ock

Op

tion

sgr

ant

the

ex-

The

gran

ting

ofan

op-

Apa

ralle

ltax

ded

uc-

The

reis

no

exp

ense

rec-

opti

ons

(NQ

SOs)

ecut

ive

the

righ

tbu

tti

onis

typ

ical

lyan

un-

tion

atth

eco

rpor

ate

ogni

tion

(atg

ran

tex

er-

not

the

obli

gati

ont

ota

xed

even

tU

pon

exer

-le

veli

sge

nera

llyal

-ci

seo

rsa

le)

for

op-

buy

shar

esof

stoc

kat

cise

opt

ion

pro

ts

are

low

edu

pon

exer

cise

tion

sif

as

isu

sual

lyan

exer

cise

pric

e(t

ypi-

taxe

dat

the

ord

inar

y(f

orth

eam

ount

ofop

-th

eca

set

he

opti

ons

cally

the

curr

ent

stoc

kra

teI

fth

eex

ecu

tive

tion

pro

ts

)O

pti

ons

are

not

dis

cou

nted

(in

pri

ceat

the

dat

eof

cont

inu

esto

hold

onto

auto

mat

ical

lyqu

alif

yas

the

mon

eyat

gran

tgr

ant)

Th

eop

tion

sth

esh

ares

the

exec

u-

perf

orm

ance

-bas

edp

ayd

ate)

and

the

exer

cise

typ

ical

lyve

stov

era

3ndashti

veis

taxe

din

the

and

ther

efor

ear

eno

tpr

ice

and

num

ber

ofop

-5-

year

per

iod

The

typ

i-us

ualw

ay(a

tth

eca

pi-

subj

ect

toth

e$1

-ti

ons

iskn

own

atgr

ant

calm

atu

rity

is10

year

sta

lgai

nsra

te)

onan

ym

illio

nca

p

dat

eFo

rd

isco

unt

op-

but

the

man

ager

may

furt

her

app

reci

atio

nof

tion

sth

ed

iffe

renc

ebe

-ex

erci

seea

rly

Som

eth

esh

are

pri

ce

twee

nth

est

ock

pri

ceop

tion

sar

edi

scou

ntop

-an

dth

em

arke

tp

rice

isti

ons

(the

exer

cise

pri

ceex

pen

sed

over

the

vest

-is

low

erth

anth

est

ock

ing

per

iod

For

opti

ons

pri

ceat

the

gran

td

ate)

w

ith

vari

able

term

sP

rem

ium

opti

ons

are

(eg

a

vari

able

exer

-th

ere

vers

e(t

heex

er-

cise

pri

cev

esti

ngth

atci

sep

rice

isgr

eate

ris

tied

top

erfo

rman

ce)

than

the

stoc

kp

rice

atth

eop

tion

sar

em

arke

dgr

ant

dat

e)

tom

arke

tan

dex

-p

ense

dd

uri

ngth

eti

me

betw

een

gran

tan

dex

erci

se

Not

eW

eth

ank

com

pen

sati

onco

nsu

ltan

tsSc

ott

Gre

enbe

rgan

dSc

ott

Ols

en(b

oth

ofT

ower

sP

erri

n)

and

Fred

eric

Coo

k(o

fFr

eder

icW

C

ook

and

Ass

ocia

tes)

for

help

fulc

onve

rsat

ion

sin

und

erst

and

ing

the

tax

and

acco

unt

ing

rule

s

10 Hall amp Liebman

Second the tax treatment of ISOs is different ISOs are completely un-taxed at grant or exercise It is only at sale that the executive is taxed andeven then the executive is taxed at the lower capital gains tax rate Thedisadvantage is that the corporation never gets to take a parallel taxdeduction against corporate prots Thus ISOs become more attractiveas the personal tax rate increases and as the corporate tax rate and thecapital gains tax rates fall

Two other related types of compensation restricted stock and stockappreciation rights are worth brief discussion Both however are farless common than standard stock options Restricted stock is payment inthe form of restricted shares the restriction being that the shares vestover time as with options Unlike options the shares typically havevoting and dividend rights With regard to taxation the executive istaxed at the personal rate on the value of the restricted stock as thevesting restrictions lapse However the executive may choose to betaxed at the grant date in which case all subsequent appreciation istaxed at the capital gains rate The dividends paid to the executive aretaxed at the ordinary rate in the usual way The company generallyreceives a parallel deduction equal to the amount of the executiversquosincome when the executive is taxed Unlike stock options restrictedstock is not generally considered to be performance-based and is there-fore subject to the million-dollar rule (unless the vesting of the restrictedstock is performance-based which is sometimes the case)

Stock appreciation rights (SARs) are rights that replicate the payoffs ofstock options with a cash transfer Thus SARs are simpler than options inthat there is no requirement to buy and resell the stock in order to ldquocashoutrdquo SARs generally have the same tax treatment as NQSOs both to theindividual and to the corporation SARs are relatively rare however be-cause they have disadvantageous accounting treatment (described in thenext section) relative to stock options and their main relative advantagevis-a-vis options has been essentially replicated through broker-assistedcashless option exercisesmdashwhereby a broker makes a ldquonanosecondrdquo loanto the executive (to purchase and resell the stock) when the executivewants to ldquocash inrdquo on option prots The tax and accounting treatment ofISOs restricted stock and SARs is summarized in Table 2

32 The Accounting Treatment of OptionsUnlike cash compensation which is expensed against earnings there isgenerally no expense recognition (at grant exercise or sale) for optionswhether they be NQSOs or ISOs As a result compensation consultantsoften point out that stock options are the only form of compensation thatare free in an accounting sense but still deductible for tax purposes

Taxation of Executive Compensation 11

Stock options do however lead to expense recognition if they are dis-counted (in the money at grant date) or if the exercise price and numberof options are not known at grant date For discount options the differ-ence between the stock price and the market price is expensed over thevesting period For options with variable terms (eg a variable exerciseprice vesting that is tied to performance) the options are marked tomarket and expensed during the time between grant and exercise Practi-tioners claim that the accounting treatment of options plays an impor-tant role in the design of option programs Thus plans that have ldquobadaccountingrdquo but are thought by many to have attractive incentive fea-tures are often not even seriously considered by companies Examples ofsuch potentially attractive plans include indexed options (where a CEOprots only if his rmrsquos share price grows relative to some market orindustry benchmark) and option grants that are explicitly performance-related both of which would lead to expenses against earnings

Unlike options restricted stock and SARs do not generally receivefavorable accounting treatment Restricted stock is generally expensedover the period in which the restrictions lapse (usually the vesting pe-riod) The magnitude of the expense is the difference between the cur-rent stock price and the executiversquos cost (if any) SARs are marked tomarket each period and the difference between the stock price and theexercise price is expensed over the outstanding period of SARs

4 THE TAXATION OF EXECUTIVE PAY CASHVERSUS OPTIONS

Stock-based compensation performs two roles in executive compensa-tion arrangements First it helps align the incentives of the executivewith the interests of the rmrsquos shareholders Second it often enables therm to compensate the CEO in a way that is more advantageous from atax standpoint than paying the executive in salary and bonus

41 Agency Theory and Executive Compensation in the Presenceof TaxationIn standard agency-theory models ( Jensen and Meckling 1976) agencycosts are the result of the separation of ownership and control Managersdo not have the same incentives as the owners The optimal incentivecontract for managers balances the benets of high-powered incentives(linking the fortunes of owners and managers through stock and stockoptions for example) with the costs of loading too much risk on risk-averse managers

TA

BL

E2

Tax

and

Acc

ount

ing

Tre

atm

ent

ofE

xecu

tive

Com

pens

atio

nIn

cent

ive

Stoc

kO

ptio

ns(I

SOs)

Res

tric

ted

Stoc

kan

dSt

ock

App

reci

atio

nR

ight

s(S

AR

s)

Tax

atio

nat

Tax

atio

nat

Acc

ount

ing

Com

pen

sati

onD

escr

ipti

onp

erso

nall

evel

corp

orat

ele

vel

trea

tmen

t

Ince

nti

vest

ock

ISO

sar

eid

enti

calt

oIf

the

usu

alco

ndit

ions

Ifth

eu

sual

cond

itio

nsA

sw

ith

NQ

SOs

ther

eop

tion

sN

QSO

sin

thei

rd

esig

n

are

met

th

ere

isno

tax

are

met

the

reis

gene

r-is

gene

rally

noex

pen

seH

owev

ert

hey

have

anat

gran

tor

exer

cise

At

ally

no

corp

orat

eta

xd

e-re

cogn

itio

nfo

rIS

Os

annu

alca

pof

$100

000

sale

all

opti

onp

ro

tsd

uct

ion

atgr

ant

exer

-(e

ver)

on

the

amou

ntth

atan

dsu

bseq

uen

tca

pita

lci

seo

rsa

le

vest

sT

he

tax

trea

t-ap

prec

iati

onar

eta

xed

men

tof

ISO

sal

soat

the

cap

ital

gain

sd

iffe

rs

rate

Res

tric

ted

stoc

kSi

mila

rto

stoc

kop

-T

hein

div

idu

alis

taxe

dT

heco

mp

any

gene

rall

yR

estr

icte

dst

ock

isge

n-ti

ons

but

shar

esat

the

ord

inar

yra

teas

rece

ives

ap

aral

leld

e-er

ally

exp

ense

dov

er(w

hich

vest

slow

ly)

the

rest

rict

ions

lap

se

du

ctio

neq

ual

toth

eth

eve

stin

gp

erio

dT

hera

ther

than

opti

ons

are

How

ever

the

man

ager

amou

ntof

the

exec

u-

amou

nt

exp

ense

dis

gran

ted

toth

eex

ecu

-m

aych

ose

tobe

taxe

dti

versquos

inco

me

whe

nth

eth

ed

iffe

renc

ebe

twee

nti

veU

nlik

eop

tion

sat

gran

td

ate

inw

hich

exec

uti

veis

taxe

dR

e-th

ecu

rren

tst

ock

pric

eth

esh

ares

typi

cally

case

alls

ubs

equ

ent

ap-

stri

cted

stoc

kis

gen

er-

and

the

exec

uti

versquos

have

voti

ngan

dd

ivi-

pre

ciat

ion

ista

xed

atal

lysu

bjec

tto

the

cost

(if

any)

d

end

righ

ts

the

cap

ital

gain

sra

te

mill

ion-

dol

lar

cap

D

ivid

end

sar

eta

xed

atth

eor

din

ary

rate

Stoc

kap

pre

ciat

ion

Rig

hts

that

repl

icat

eG

ener

ally

have

the

Gen

eral

lyha

veth

eU

nlik

eop

tion

sSA

Rs

righ

tsth

epr

ots

ofst

ock

op-

sam

eta

xtr

eatm

ent

assa

me

tax

trea

tmen

tas

are

mar

ked

tom

arke

tti

ons

Rar

ely

use

d

NQ

SOs

NQ

SOs

each

per

iod

an

dth

esi

nce

stoc

kop

tion

sd

iffe

renc

ebe

twee

nth

eha

vem

ore

favo

rabl

eac

-st

ock

pri

cean

dth

eex

-co

unt

ing

trea

tmen

ter

cise

pri

ceis

exp

ense

d(a

ndop

tion

sp

rovi

de

over

the

outs

tand

ing

the

sam

ebe

ne

tof

per

iod

ofSA

Rs

cash

less

exer

cise

thro

ugh

brok

erlo

ans

)

Not

eW

eth

ank

com

pen

sati

onco

nsul

tant

sSc

ott

Gre

enbe

rgan

dSc

ott

Ols

en(b

oth

ofT

ower

sP

erri

n)an

dFr

eder

icC

ook

(of

Fred

eric

WC

ook

and

Ass

ocia

tes)

for

hel

pfu

lcon

vers

atio

nsin

und

erst

and

ing

the

tax

and

acco

unt

ing

rule

s

14 Hall amp Liebman

The effect of taxation on the optimal contract (more precisely theshare of compensation that is performance-related) is ambiguous (evenin the absence of deductibility and deferral) because there are offsettingeffects First by reducing the share of corporate prots received byshareholders taxes diminish the importance to the shareholders of moti-vating the CEO Taken alone this effect would imply that taxes wouldbe expected to reduce the use of performance-based compensation Sec-ond because the government shares in the income received by the CEOthe variance of the CEOrsquos income is reduced raising his utility andlowering the cost to the rm of providing any given set of incentivesThird because taxes will lead executives to provide less effort for anygiven level of incentive-based pay the level of compensation that mustbe provided to compensate the CEO for effort is reduced for a givenamount of incentive These last two factors make it cheaper for the rmto offer contracts to the CEO with large incentive components when taxrates are higher and should therefore increase the use of incentive-based pay Since the net effect of taxation on the level of incentive-basedpay is ambiguous agency theory provides no strong predictions abouthow taxation should affect the optimal composition of option versuscash compensation

42 The Tax Advantages of DeferralWhile agency theory yields ambiguous predictions about the effect oftaxes on the use of options there are direct tax advantages of options inthat options provide executives with a way to defer compensation andthereby lower their taxes However because options also lead to a defer-ral of corporate tax deductibility the tax advantages from a combined(executive and corporate) perspective are not so straightforward In thissection we analyze and measure the tax advantages of stock optionsrelative to salary and bonus compensation We also analyze the taxadvantages of NQSOs relative to ISOs In particular we show how thetax advantages of options change as personal corporate and capitalgains taxes change We then show how the tax advantages of optionshave changed over time in response to changing tax rates

The crucial tax difference between standard options (NQSOs) andcash payment is that option payouts are deferred and the two forms ofcompensation earn different rates of return over the deferral period Anyanalysis of the relative tax advantages of two compensation instrumentsmust consider the tax consequences both to the employer and to theemployeemdashwhat Scholes and Wolfson (1992) call the global contractingperspective Thus in order to make valid comparisons between the twowe compare the tax burden to the executive while holding constant the

Taxation of Executive Compensation 15

posttax cost to the company By keeping posttax employer costs (in NPVterms) constant any package that is preferred by the employee is tax-advantaged in the global-contracting sense

Under this methodology a comparison of the tax advantages of op-tions and cash involves comparing a pretax cash payment of P with anoption payment that has the equivalent posttax NPV to the companyFor NQSOs it is straightforward to show that a pretax payment of P toan executive has exactly the same cost to the company as putting aside Pfor the purposes of paying stock-option payouts later That is if a com-pany pays an executive P today it will have the same amount of moneyin N years as if it had put aside P today let it accumulate at the rate ofreturn earned by the rm and then paid it out (as compensation in theform of option gains4) with any appreciation in year N

Assume that the pretax prot rate is r and the corporate tax rate is TcThen if the company pays P today it receives a deduction today of PTcwhich yields

PTc [1 1 r (1 2 Tc)]N (1)

in N years since the benets of the tax deduction are invested in thecompany and receive the after-tax corporate rate of return Converselyif the company puts P aside today then it grows at the after-tax corpo-rate rate of return in N years to give a tax deduction of

P [1 1 r (1 2 Tc)]N Tc (2)

which is the exact same value Note that the equivalence of these two taxdeductions is analogous to the tax-benet equivalence of front-loadedand back-loaded IRAs

An executiversquos payoff in N years from option prots is equal to Pcompounded at the after-tax corporate rate times 1 2 Tp where Tp is thepersonal tax rate Thus the combined payoffs in N years of the corporatededuction and the executiversquos posttax payoff is

P [1 1 r (1 2 Tc)]N Tc 1 P [1 1 r (1 2 Tc)]

N (1 2 Tp) (3)

where the rst term is the payoff from the corporate deduction and thesecond term is the executiversquos payoff

4 This analysis is not unique to options Any form of deferred compensation that enablesexecutives to invest inside the rm at a higher posttax return or without paying capitalgains taxes would have similar effects

16 Hall amp Liebman

The payoff to salary and bonus in N years is more complicated sincean assumption must be made about how the executive invests the origi-nal cash compensation For example if the executive invests in an instru-ment with taxable interest (eg bonds) then the N-year payoff from aninvestment of P accumulates at the post-corporate-tax post-personal-taxrate of return and is

P (1 2 Tp) [1 1 r (1 2 Tc)(1 2 Tp)]N (4)

However if an executive invests in (non-dividend-paying) equities theinvestment accumulates at the higher post-corporate-tax rate of returnwith the offsetting disadvantage that the capital gains are taxed at thecapital gains rate in year N So the executive receives

P [1 1 r (1 2 Tc)]N (1 2 Tp) 2 Tcg $P (1 2 Tp) [1 1 r (1 2 Tc)]

N 2P (1 2 Tp) (5)

where Tcg is the capital gains tax rate5Thus putting the equations together if an executive is paid cash P

and invests g in bonds and 1 2 g in equities the N-year payoff (includ-ing the payoff to the corporation from the tax deduction) is

P [1 1 r (1 2 Tc)]N Tc 1 g $(P [1 1 r (1 2 Tc)(1 2 Tp)]

N (1 2 Tp)1(1 2 g) (P [1 1 r (1 2 Tc)]

N (1 2 Tp) 2 Tcg $P (1 2 Tp)[1 1 r (1 2 Tc)]

N 2 P (1 2 Tp)) (6)

The tax advantage of options versus salary and bonus therefore is thedifference between the combined (corporate and executive) payoffs inequation (3) and the combined payoffs in equation (6)

43 The Tax Advantage of OptionsWe now turn to analysis of how changes in various tax rates affectthe tax advantages of options relative to cash compensation We focuson the case in which the cash earned by executives is invested entirelyin equities since equities are tax-favored relative to bonds The impor-tant conclusions of this analysis are not substantively different if we

5 In practice executives hold a combination of equities and bondlike instruments withtaxable interest Because equities are tax-favored in the next section we simplify the analy-sis by assuming that executives hold only equities Alternatively we could have assumedthat executives hold only less risky bonds since bonds more closely match the risk proleof a stream of tax savings None of our empirical results are substantively affected by thissimplication

Taxation of Executive Compensation 17

instead assume that a portion of an executiversquos holdings are invested inbonds

431 The Tax Advantage of Options Changing the Corporate TaxRate Stock options are tax-advantaged relative to cash simply becauseoption payouts are deferred allowing the executive to invest at the pre-personal-tax rate of return (with no capital gains tax at the end) ratherthan investing at the after-tax rate of return or paying the capital gainstax Since the advantage of being able to defer taxes is large when thecorporate tax rate is low the tax advantage of options is larger when thecorporate rate is low

To illustrate this point we dene the tax advantage of options to bethe dollar amount by which total after-tax option payoffs (to the corpora-tion and the executive) exceed the total payoffs from salary as denedby the difference of equation (3) and (6) assuming that P (the paymentto the executive) is equal to $100 and g 5 0 (cash compensation is in-vested in equities) We use a ten-year horizon since most options have aten-year term We then calculate how the tax advantage of optionschanges as the corporate tax rate varies from zero to 100 percent holdingthe personal tax rate and the capital gains rate xed at 40 percent and 20percent respectively (which approximates current rates)

The top panel of Figure 2 shows the tax advantage of options relativeto cashmdashprecisely the NPV of the total tax advantage of options for a$100 payment to the CEOmdashas the corporate tax rate changes As ex-pected the tax advantage of options declines as corporate tax rates in-crease At a zero corporate tax rate the tax advantage of options has anNPV of $7 This value is about $350 at a corporate rate of 40 percentWhen the corporate tax rate is 100 percent the tax advantage of optionscompletely disappears since the benets of deferred compensation fallto zero because the after-tax return on equities (and therefore the dis-count rate) falls to zero The tax advantage of options is the NPV of thetax savings from avoiding the capital gains tax on ten years of apprecia-tion which is why it is relatively modest even at the 40-percent corpo-rate rate

432 The Tax Advantage of Options Changing the Personal TaxRate Using the same assumptions as above (but this time holding thecorporate rate at 40 percent and the capital gains rate at 20 percent andvarying the personal rate) the tax advantage of options declines as thepersonal tax rate rises as shown in the middle panel of Figure 2 The taxadvantage of options is higher at low personal tax rates because theposttax base that gives rise to capital gains is higher and the advantage of

Note Holds constant the personal rate at 40 percent and the capital gains rate at 20 percent

Note Holds constant the corporate rate at 40 percent and the capital gains rate at 20 percent

Note Holds constant the personal rate at 40 percent and the corporate rate at 40 percent

FIGURE 2 Tax Advantage of Options NPV of Tax Advantage of Pay-ing $100 in Options (NQSOs) Rather Than in Cash

Taxation of Executive Compensation 19

avoiding the capital gains taxes is greater The tax advantage of optionstherefore declines linearly as the personal tax rate rises At a 100-percentpersonal tax rate the executive receives nothing in either case and byconstruction the companyrsquos tax deduction benet is the same in NPVterms so the tax advantage of options falls to zero also6

433 The Tax Advantage of Options Changing the Capital Gains TaxRate The tax advantage of options for various capital gains rates isdepicted in the bottom panel of Figure 2 under the same assumptions(this time holding both the personal rate and the corporate rate at 40percent while varying the capital gains rate) The tax advantage of avoid-ing the capital gains tax increases as the capital gains tax rate rises

5 HOW HAS THE TAX ADVANTAGE OF OPTIONSCHANGED OVER TIME

The analysis so far has illustrated how the tax advantage of optionsvaries with the changes in tax rates In this section we show how the taxadvantage of options has changed over time as tax rates have changed

51 Top Marginal Tax Rates over TimeThe evolution on the top marginal tax ratesmdashpersonal corporate andcapital gainsmdashfrom 1980 to 1998 is shown in Figure 3 In all cases thetop rate (the rate for taxpayers with the highest incomes) is shownwhich is not always the highest rate since various anomalies (such asclawbacks of exemptions) sometimes lead to marginal tax rates that arehigher than those for the highest income tax payers

The top personal tax rate was 70 percent in 1980 falling to 50 percentin 1982 and 28 percent in 1988 following the 1986 tax act7 The topmarginal tax rate has since risen to 396 percent but is effectively 425percent since there is a 29-percent Medicare surcharge (paid half by theemployer and half by the employee) that has no upper limit The topcorporate tax rate has had only one signicant change since 1980 fallingfrom 46 percent before the 1986 tax act to 34 percent following the act in1988 The top corporate tax rate was increased to 35 percent in 1993 The

6 This is one place where ignoring bond investments has substantive implications Inparticular in the all-bond case the tax advantage of options is nonmonotonicmdashit rises andthen fallsmdashas personal tax rates rise7 During the early 1980s the maximum tax on earned income limited the marginal tax rateon the earnings of high earners so many executives did not face a decline in personal taxrates between 1981 and 1982 The empirical results in this paper are robust to assumingthat executives faced a marginal tax rate of 50 percent in 1980 and 1981

20 Hall amp Liebman

FIGURE 3 Top Marginal Tax Rates since 1980

top capital gains rate has uctuated between 28 percent (approximately)and 20 percent since 1980 and is currently at 20 percent

52 The Evolution of the Tax Advantage of NQSOsThe tax advantage of options relative to cash dened in the same way asin section 4 from 1980 to 1998 is shown in Figure 4 The calculations aredone using the same assumptions as before (ie N 5 10 years and cashis invested in equities) and the statutory top marginal tax rates shown inFigure 3 are used Due to the fall in corporate and personal tax rates thetax advantage of options increased sharply after the Tax Reform Act of1986 This increase in tax advantage has largely been reversed in the1990s as top personal tax rates have crept back up and the capital gainsrate has been reduced

Two key facts emerge from this analysis First there is currently only amoderate tax advantage to standard nonqualied stock optionsmdashon theorder of $4 per $100 of compensation This is because the tax advantagesto the executive of deferring taxes are largely offset by the tax disad-vantages to the company of not being able to deduct option expensesfrom taxable prots until the executive exercises the options Secondalthough the 1986 tax act substantially increased the tax advantage ofoptions more than half of this increase has been reversed in the 1990s

Taxation of Executive Compensation 21

FIGURE 4 Tax Advantage of Options 1980 to 1998

NPV of tax advantage of $100 in options (NQSOs) rather than cash

53 The Tax (Dis)advantage of ISOsSo far the analysis has focused on the tax advantages of NQSOs sincethey are so much more prevalent than ISOs which account for onlyabout 5 percent of option grants The relative scarcity of ISOs can beexplained both by their tax status and by their per-executive cap of$100000 per year

Under what conditions are ISOs tax-preferred to NQSOs For a trans-fer P to an executive the ISO is always tax-preferred by the executive(since the capital gains rate is lower than the personal rate) while theNQSO is always tax-preferred from the companyrsquos perspective (since anISO is not deductible) The key issue however is the relative advan-tages of the two types of options from a global-contracting perspectiveBecause ISOs are not deductible the company is indifferent betweensetting aside P in the form of NQSOs today and setting aside P(1 2 Tc) inthe form of ISOs (Since the NPV of the tax deduction for NQSOs is PTcthe NPV of the payment P is P 2 PTc or P (1 2 Tc))

Equalizing the posttax cost of NQSOs and ISOs we need only look atthe payoff of the executive to determine the condition under which eachis preferred The payoff of NQSOs in N years is

P [1 1 r (1 2 Tc)]N (1 2 Tp) (7)

22 Hall amp Liebman

The same payoff for ISOs is

P (1 2 Tc) [1 1 r (1 2 Tc)]N (1 2 Tcg) (8)

ISOs therefore are tax-advantaged if

Tp 2 Tcg Tc 2 Tc Tcg (9)

Intuitively ISOs are tax-advantaged only if their advantage (the differ-ence between the personal rate and the capital gains rate) is largeenough to offset their cost (the disadvantage of not deducting at thecorporate rate) Note however that the condition is not a simple com-parison between the corporate rate and the personal rate minus thecapital gains rate (Tp 2 Tcg Tc)

Figure 5 shows how the tax advantage of NQSOs relative to ISOs haschanged since 1980 as tax rates have changed Unlike the very modestchanges (around 3 to 4 percent) in the relative tax advantages betweenoptions and cash shown in Figure 4 the relative tax advantage betweenNQSOs and ISOs has seen enormous swings When the top personalrate was 70 percent in 1980 ISOs were tax-favored by a margin greaterthan 6 percent However by 1982 (following the 1981 tax acts that dra-matically lowered the personal tax rate) ISOs became tax-disadvantaged

FIGURE 5 Tax Advantage of NQSOs Relative to ISOs over Time

NPV of tax advantage of $100 payment to executive NQSOs minus ISOs

Taxation of Executive Compensation 23

by about 5 percent By 1988 (following the 1986 tax act) ISOs became tax-disadvantaged by about 18 percent a dramatic swing The raising ofpersonal rates coupled with a decrease in the capital gains rate has sincereduced the tax disadvantage of ISOs to about $4 per $100 of pretaxcompensation Although hard data on ISOs are hard to come by practi-tioners (mostly compensation consultants) claim that ISOs were morepopular prior to 1982 when they were tax-advantaged Although ISOsare less disadvantageous from a tax standpoint relative to NQSOs thanthey were in the late 1980s they are still disadvantageous so it is notsurprising that they have not made a signicant resurgence in recentyears

6 EXPLAINING THE INCREASE IN OPTION-BASEDCOMPENSATION

In this section we analyze whether changes in tax rates affect the compo-sition of executive compensation and consider the relative importanceof tax factors and corporate governance factors in explaining the in-creasing share of compensation paid in the form of stock options Thenumerous changes in tax rates that have occurred since 1980 provide anopportunity to assess whether executive compensation arrangementsrespond to tax incentives in the way that the tax-avoidance model sug-gests We test this model using a panel data set of CEOs in largepublicly-traded US corporations Our identication of the tax effectsrelies on time-series variation in personal corporate and capital gainstax rates along with cross-sectional variation in corporate tax rates

61 DataWe use a 15-year panel data set of CEOs in the largest publicly-tradedUS corporations which is described in Hall and Liebman (1998) Thedata set covers the years from 1980 through 1994 and combines CEOcompensation information from corporate proxies and 10-K lings withstock price and stock return information from CRSP and accounting datafrom Compustat In addition some compensation data from the 1970swere collected in order to construct measures of the value of stock op-tions held by the CEOs in the rst period of the sample8

The feature that distinguishes our data from most other CEO datasets is that with our panel of yearly proxy data on option grants optiongains and total options held we are able to calculate the total value of

8 We are grateful to David Yermack for providing us with some of the data for 1984 to 1991See Yermack (1995) for a discussion of these data

24 Hall amp Liebman

all stock options held by the CEO at a point of time More importantlysince we have the details about the stock options held (number exer-cise price time to maturity etc) we can precisely calculate the changein the value of a CEOrsquos stock option holdings for a given change in rmvalue

62 Identication IssuesThe numerous tax changes over our sample period and the specic waysin which changes in personal corporate and capital gains rates arepredicted to affect the composition of CEO compensation give us un-usually rich sources of identication However because many of thechanges in tax rates would be expected to affect all of the CEOs in oursample in a similar way we will need to pay particular attention toseparating out the tax effects from underlying time trends and otherfactors that changed over time Controlling for underlying trends is par-ticularly important in this study because we are focusing on an out-comemdashthe use of stock optionsmdashthat has increased rapidly over thepast two decades and that many practitioners believe was importantlyaffected by nontax considerations Thus the tax effects we are trying toexplain are deviations from a rising trend

We take two steps to try to separate out the tax effects from theunderlying trend First we include key nontax factors that could poten-tially explain the increasing reliance on performance-based compensa-tion It has been suggested that the dramatic increase in incentive-basedpay is the result of the remarkably poor shareholder returns during the1970s which spurred the LBO and takeover movements of the 1980sAccording to this story (Kaplan 1997) shareholders became much morepowerful via the rise of institutional investors even as the LBO andtakeover movements waned Because the inuence of institutional inves-tors is thought to be one of the most important mechanisms of strongcorporate governance we use the share of each companyrsquos stock ownedby institutional investors (dened as institutions with more than $100million under management) as an explanatory variable9 The percentageof shares owned by large institutional investors increased from about 20percent to almost 50 percent in 1994 an upward trend that closelymatches the sharp rise in the share of compensation in the form ofoptions Indeed in our sample the annual average (over all of the rmsin our sample) share of stock owned by institutional investors has acorrelation of 09 with the annual average share of compensation paid inoptions

9 This variable is described in Gompers and Metrick (1998)

Taxation of Executive Compensation 25

In addition we include two other variables that proxy for strongercorporate governance The rst is the size of the board Evidence suggeststhat boards with smaller numbers of directors reduce the inuence of theCEO and better represent shareholder interests (Yermack 1995) Boardsize fell by approximately 15 percent between 1980 and 1994 The secondvariable is the fraction of the rmrsquos board members who are inside direc-tors or gray directors (noninsiders who have business dealings with thecompany) Firms with fewer outsiders are less likely to act in the share-holdersrsquo interests and we would predict that they would be less likely tohave performance-based compensation for their CEOs The average per-centage of inside and gray directors has decreased from about 45 percentin 1980 to about 35 percent in 1994

Second we allow for cross-sectional variation in corporate tax ratesCorporate marginal tax rates vary because rms may be eligible for tax-loss carrybacks and carryforwards investment tax credits and the alter-native minimum tax (see Auerbach 1986 Auerbach and Altshuler1990 Auerbach and Poterba 1987) Because these cross-sectional differ-ences imply that different rms should respond differently to a givenchange in tax rates they potentially provide us with a way to isolatethe tax effects Firm-specic corporate marginal tax rates are notori-ously difcult to calculate from publicly-available data We use a tricho-tomous variable that equals zero if the rm has tax-loss carryforwardsand negative earnings in a given year equals 05 times the statutorycorporate rate if the rm has only one of those conditions and equalsthe statutory rate if the rm has neither loss carryforwards nor negativeearnings10

The cross-sectional variation in the corporate tax rate may not be ex-ogenous Firms that perform poorly could face low marginal tax ratesand also be particularly likely or unlikely to provide performance-basedpay Because most of our specications use xed effects the correlationwe must be concerned about is between changes in rm performanceand changes in compensation We deal with this by controlling forlagged rm stock market performance in our regressions This shouldeliminate spurious correlation between tax rates and compensation thatis jointly caused by rm performance

10 Graham (1996) presents evidence that the trichotomous variable performs nearly as wellas his recommended simulated marginal tax rate Plesko (1999) compares marginal taxrates calculated from Compustat data with actual marginal tax rates from rm tax returnsHe nds that simple binary variables are more reliable measures of the marginal tax ratethan is either the Graham or the trichotomous variable We obtain substantively similarresults when we switch from the trichotomous variable to the binary variables recommendby Plesko

26 Hall amp Liebman

63 ResultsThe dependent variable in our regressions is the fraction of a CEOrsquos totalannual compensation (measured as the BlackndashScholes value of stock-option grants plus salary and bonus) that comes from stock-optiongrants We choose this reduced-form specication because our interest isin the composition of CEO compensation However the factors thataffect the composition of compensation might also affect the levels andit will be important to keep this in mind in interpreting our results

Optimal-contracting theory generally makes predictions about the in-centives provided by the total amount of rm stock and stock optionsowned by the CEO We choose to model the ow of option grants ratherthan the ultimate performance sensitivity of the CEOrsquos compensationand stock and stock-option holdings because we have in mind anadjustment-cost model in which it takes time for a CEOrsquos holdings ofstock and stock options to reach the optimal level11 In our regressionswe include two measures of the current performance sensitivity of theCEOrsquos wealth to rm performance as explanatory variables to reect thedistance of the CEOrsquos contract from the optimal contract

Table 3 presents results in which the tax effects are identied solely byvariation in tax rates over time The fraction of annual compensation paidin stock options is regressed on tax variables corporate governance vari-ables contract-theory variables and variables reecting stock market re-turns Both of the regressions in this table include rm xed effects Incolumn (1) the fraction in annual compensation paid in stock options isregressed on the log difference in payoff (the combined payoff to both therm and the CEO) from receiving compensation in stock options ratherthan salary and bonus where the payoff calculations are based on statu-tory marginal tax rates This variable is dened as the log difference be-tween equation (3) and equation (6) in section 4 The coefcient on the taxvariable is large and statistically signicant at the 95-percent level Thepoint estimate implies that a one-percent increase in the payoff differencebetween stock options and salary and bonus results in a 24-percentage-point increase in the share of compensation paid in stock options

The corporate governance variables all have the predicted sign Firmswith a higher fraction of their shares owned by institutional investorsuse more performance-based pay (stock options) Firms with large corpo-rate boards or a large share of inside directors are less likely to use stockoptions

The two contract-theory variables are the dollar change in CEO wealth

11 See Core and Guay (1999) for evidence that rms adjust in this way

TABLE 3Compensation Regressions that Rely on Variation in

Tax Rates over Time

(1) (2)

Tax variablesln (difference in payoff)(a) 2393 (0349)ln (12statutory personal 20015 (0034)

rate)ln (12statutory cap gains 0330 (0171)

rate)ln (12statutory corporate 0399 (0148)

rate)

Corporate governance variablesFraction of rmrsquos shares 0043 (029) 0052 (0028)

owned by institutionalinvestors

Size of the rmrsquos board of 20002 (001) 20003 (0001)directors

Fraction of directors who 20082 (033) 20092 (0030)are inside or gray

Contract-theory variablesD (CEO wealth) per $1000 20261 (0095) 20247 (0087)

in FV(b)

D (CEO wealth) per 10 in 2583 (189) 2554 (201)FV(c)

Volatility(d) 521 (424) 261 (369)ln (rm market value) 0019 (0006) 0019 (0006)

Stock market return variablesLagged annual rm stock 0006 (0009) 0003 (0008)

market returnLagged annual SampP 500 re- 20035 (0023) 20029 (0023)

turn

Regression diagnosticsTotal sample size 5179 5179Firms 427 427Adjusted R2 0313 0311

Dependent variable share of compensation paid in options Standard errors in parentheses Bothregressions in this table include rm xed effects

(a) Log difference in payoff is the log difference in combined rm and CEO after-tax share betweenpaying compensation in options and paying salary

(b) This variable is the dollar change in CEO wealth per $1000 change in rm market value (seeJensen and Murphy 1990)

(c) This variable in the dollar charge in CEO wealth per 10-percent change in rm value (see Hall andLiebman 1998)

(d) Standard deviation of rm market value in trillions

28 Hall amp Liebman

per $1000 change in rm value (Jensen and Murphy 1990) and thedollar change in CEO wealth per 10-percent change in rm value (Halland Liebman 1998) As discussed in Baker and Hall (1998) these twowealth sensitivity measures reect two different concepts of how closelyaligned the managersrsquo incentives are with the interests of the sharehold-ers In the regression they are meant to measure how far the CEOrsquosexisting contract and ownership of stock and stock options is from theoptimal level The negative coefcients on the two variables suggeststhat rms that are below their optimal pay-to-performance sensitivityare more likely to give more stock options in the current period

The second column replaces the tax variable motivated by deferraladvantages with the three statutory rates The corporate after-tax sharehas a positive coefcient as we would expect because the tax advantageof stock options increases as the corporate rate falls The coefcient onthe personal tax rate is small and statistically insignicant The sign ofthe coefcient on the capital gains rate is the opposite of that predictedby tax-avoidance theory

Note that these results rely on time-series variation alone This is not avery convincing test Since the log difference in payoffs mostly rises overthe time period covered in our sample as does the share of compensa-tion paid in options it is possible that the tax variable is simply reectingother factors that were changing over this time period that we have notincluded in our regression Therefore it is important to look at specica-tions that rely on cross-sectional variation in the tax variables

Table 4 contains specications that rely on cross-sectional variation incorporate tax rates The rst column repeats the specication from col-umn (1) of Table 3 but uses the trichotomous measures of rm corporatetax rates in place of the statutory rate in calculating each rmrsquos logdifference in payoff from paying compensation in options rather than insalary and bonus As in the previous table we nd the positive coef-cient on the log difference in payoff just as the theory suggest How-ever the coefcient estimate is now much smaller suggesting that aone-percent increase in the log difference in payoff produces a 01-percentage-point increase in the share of compensation paid in options

Introducing cross-sectional variation in the dependent variable doesnot eliminate the potentially spurious correlation over time between thetax variable and the share of compensation paid in options The secondcolumn of Table 4 isolates the pooled cross-sectional variation in corpo-rate tax rates by adding time dummies and dropping the rm xedeffects In this specication the coefcient on the tax variable is indistin-guishable from zero The coefcient on the institutional variable contin-ues to be large and statistically signicant The third column includes

Taxation of Executive Compensation 29

both rm xed effects and time dummies which tests for a relationshipbetween within-rm changes in the dependent variable and within-rmchanges in each explanatory variable In this specication the coefcienton the tax variable is again indistinguishable from zero The coefcientson the corporate governance variables are also statistically insignicant

The third column contains a specication that treats the Tax ReformAct of 1986 (TRA 86) as a natural experiment For rms and CEOs facingthe statutory corporate tax rate TRA 86 increased the relative payoff topaying compensation in options by 17 percent For a rm facing a zeromarginal tax rate it reduced the relative payoff by 45 percent The reduc-tion occurred because TRA 86 reduced the personal tax rate makingsalary and bonus more attractive while the decreased corporate rate hadno effect on zero-tax-rate rms Thus we would expect rms facing zeroor low marginal tax rates to reduce their use of options after TRA 86while rms facing the statutory rate would be expected to increase themFew rms face zero marginal tax rates year after year Therefore we tryto distinguish between rms that often face low tax rates and rms thatnearly always face the statutory rate We dene low-tax rms as oneswhose average marginal tax rate over 1984 to 1986 was below 02412

Then we ran a difference-in-differences regression to see if high-taxrms increased their use of options more than low-tax rms The interac-tion between post-TRA 86 and high corporate tax rates is the key vari-able The estimated coefcient on this variable is positive but small andnot statistically different from zero at the 95-percent level

It is important to note that in this specication as well the main corpo-rate governance variable the share of the rmrsquos shares owned by institu-tional investors performs as theory would predict The variable appearsto be quite robust Table 5 presents regressions of the share of compensa-tion paid in stock options on the fraction of the rmrsquos shares owned byinstitutional investors Since we have already established that these vari-ables have a strong longitudinal relationship the regressions here aredesigned to focus on the cross-sectional relationship between these twovariables The rst and third columns present regressions using data onlyfrom 1994 and nd a coefcient of roughly 02 (column 1 contains noother covariates while column 3 includes all of the other nontax covari-ates from the regressions in Tables 3 and 4) Since the median share of armrsquos stock owned by institutional investors in our sample rises from 19percent in 1980 to 49 percent in 1994 this coefcient would imply that this

12 This cutoff was chosen after inspecting the distribution of tax rates and essentiallyseparates rms that are always at or near the statutory rate from ones that are sometimesbelow it

TA

BL

E4

Com

pens

atio

nR

egre

ssio

nsth

atR

ely

onC

ross

-Sec

tion

alV

aria

tion

inC

orpo

rate

Tax

Rat

es

(1)

(2)

(3)

(4)

Tax

vari

able

s(a)

ln

(dif

fere

nce

inp

ayof

f)0

099

(00

41)

20

038

(00

36)

20

005

(00

44)

Pos

t-T

RA

862

000

5(0

013

)H

igh

corp

orat

eta

xra

te0

018

(00

17)

(Pos

t-T

RA

86)

3hi

gh0

024

(00

16)

corp

orat

eta

xra

te

Cor

por

ate

gove

rnan

ceva

riab

les

Frac

tion

of

rmrsquos

shar

es0

109

(00

27)

011

4(0

015

)2

000

9(0

030

)0

087

(00

34)

owne

dby

inst

itu

tion

alin

vest

ors

Size

ofth

e

rmrsquos

boar

dof

20

004

(00

01)

20

001

(00

01)

20

002

(00

01)

000

1(0

002

)d

irec

tors

Frac

tion

ofd

irec

tors

wh

o2

012

4(0

033

)2

000

3(0

018

)2

004

1(0

034

)0

001

(00

39)

are

insi

de

orgr

ay

Con

trac

t-th

eory

vari

able

sD

(CE

Ow

ealt

h)p

er$1

000

20

238

(00

94)

20

167

(00

53)

20

257

(00

96)

20

021

(01

47)

inFV

(b)

D(C

EO

wea

lth)

per

10

in2

560

(17

8)2

592

(10

5)2

594

(19

6)2

147

(04

86)

FV(c

)

Vol

atili

ty(d

)2

48(4

24)

162

(03

4)1

12(4

28)

106

(06

09)

ln(

rmm

arke

tva

lue)

003

3(

006)

000

8(0

003

)0

009

(00

06)

001

2(0

007

)

Stoc

km

arke

tre

turn

vari

able

sL

agge

dan

nual

rm

stoc

k2

000

2(0

009

)0

011

(00

10)

000

6(0

009

)2

001

2(0

013

)m

arke

tre

turn

Lag

ged

annu

alSamp

P50

02

005

6(0

023

)2

005

3(0

051

)2

003

9(0

049

)2

015

4(0

040

)re

turn

Reg

ress

ion

dia

gnos

tics

T

otal

sam

ple

size

5179

5179

5179

2534

Firm

s42

742

739

6A

dju

sted

R2

031

009

032

005

Firm

xed

effe

cts

Yes

No

Yes

No

Tim

ed

um

mie

sN

oY

esY

esN

o

Dep

end

ent

vari

able

isfr

acti

onof

ann

ualc

omp

ensa

tion

from

stoc

kop

tion

sSt

and

ard

erro

rsin

par

enth

eses

(a)

All

thre

esp

eci

cati

ons

use

tric

hoto

mou

sco

rpor

ate

tax

rate

sT

he

sam

ple

inco

lum

n(3

)u

ses

only

the

year

s19

84to

1988

sin

cew

ear

eu

sing

1986

asa

nat

ural

exp

erim

ent

(b)

Thi

sva

riab

leis

the

dol

lar

chan

gein

CE

Ow

ealt

hp

er$1

000

chan

gein

rm

mar

ketv

alu

e(s

eeJe

nse

nan

dM

urp

hy

1990

)

(c)

Thi

sva

riab

lein

the

dol

lar

char

gein

CE

Ow

ealt

hp

er10

-per

cent

chan

gein

rm

valu

e(s

eeH

alla

nd

Lie

bman

199

8)

(d)

Stan

dar

dd

evia

tion

ofr

mm

arke

tva

lue

intr

illio

ns

32 Hall amp Liebman

TABLE 5Large Institutional Investors and Stock-Option Compensation

The Cross-sectional Relationship between Institutional Investors andOption Grants

(1) (2) (3) (4)

Coefcient on fraction of 217 (067) 145 (014) 177 (069) 120 (015)rmrsquos shares owned byinstitutional investors

Sample period 1994 1980ndash1994 1994 1980ndash1994Contract-theory and stock No No Yes Yes

market covariatesTime dummies No Yes No Yes

The dependent variable is the share of compensation paid in options

factor alone can explain 6 percentage points of the 23-percentage-pointincrease in the median share of compensation paid in options that oc-curred over this time period Of course this variable is only one measureof the strengthening of corporate governance so the total impact of im-proved corporate governance could be greater The second and fourthcolumns of the table use the entire 1980ndash1994 sample and include timedummies to isolate the cross-sectional correlations Again there is evi-dence of a strong cross-sectional relationship between the two variablesalthough the coefcients are slightly smaller perhaps reecting a weakercorporate-governance role in the earlier period In sum there is a strongrelationship in both longitudinal and cross-sectional data between theinstitutional-investor variable and the options-percentage variable13

7 THE MILLION-DOLLAR RULE

In 1993 section 162(m) of the Internal Revenue Code was enactedlimiting the deductibility of executive compensation in excess of onemillion dollars unless the compensation was performance-related Pro-ponents of this legislation argued that it would slow the growth ofexecutive pay and tighten the link between rm performance and execu-tive compensation

The new provision became effective for tax year 1994 and applies tothe CEO and the other four most highly compensated executives in eachcompany Pay that is performance-related such as stock-option grants

13 The one specication in which the fraction owned by institutional investors is notsignicant is in column (2) of Table 4 a regression that includes both xed effects and timeeffects

Taxation of Executive Compensation 33

and bonuses paid for meeting clear performance goals is not affected bythis regulation provided that it has been approved by shareholdersThus the main impact of the provision is to limit the deductibility ofsalary in excess of a million dollars Woodlock and Antenucci (1997)studied the proxy statements of 376 rms and documented that mostrms responded to the new law by qualifying top executive pay asperformance-related Perry and Zenner (1999) identied 25 rms thatreduced salaries from above $1 million to below $1 million and foundthat 23 of the rms cited section 162(m) as their reason for doing so

In this section of the paper we investigate whether this provision hasaffected the level or structure of executive pay using the Execucompdata base which contains information on the highest-paid executives ineach of the rms in the SampP 500 SampP Midcap 400 and SampP SmallCap600 Between 1993 and 1998 the median salary of the Execucomp execu-tives rose from $230000 to $306000 The median value of total compen-sation (salary bonus and option grants) rose from $470000 to $882000Limiting the sample of CEOs indicates that the total compensation of themedian CEO rose from $11 million to $18 million over the period whilethe median salary rose from $441000 to $550000 Thus executive com-pensation clearly continued to rise at a rapid rate after the implementa-tion of the million-dollar rule though perhaps at a slower rate than itotherwise would have

If the regulation affects behavior it should have the largest impact onrms whose executives are receiving salaries that are above the limit orjust below it Thus we might expect to see rms whose executives arereceiving relatively high salaries to rely more heavily on performance-based pay and less heavily on salary in the years following the regulationPerry and Zenner (1999) and Livingston (1999) both nd that this is thecase This trend can be clearly identied in the Execucomp data set Forexample over 1993 to 1998 the median annualized growth rate of salaryfor executives whose 1993 salary was below $500000 in the previous yearwas 85 percent For executives with salary between $500000 and$700000 it was 64 percent for executives between $700000 and $850000it was 50 percent for executives between $750000 and $1 million it was31 percent and for executives above $1 million it was 00 percent14

The observation that executives at lower levels of salary had highergrowth rates is not very strong evidence of an impact of section 162(m)however because it is possible that low-salary executives would havehad higher growth rates even in the absence of the provision To explore

14 Of the 3137 executives for whom we have complete data between 1993 and 1998 2708had 1993 salaries below $500000 while only 35 had 1993 salaries above $1 million

34 Hall amp Liebman

whether this is the case we use Execucomp data from both before andafter the law change Ideally we would want to have data from manyyears before the change to examine whether there is an underlying pat-tern of faster salary growth for low-salary executives Unfortunately theExecucomp data begin in 1992 and therefore provide us with only the1992ndash1993 growth rate in compensation to use as a baseline15

We run regressions pooling data from 1992ndash1993 through 1997ndash1998of the annual percentage change in different forms of compensation on avariable million that is designed to measure the likelihood that anexecutiversquos compensation will be affected by section 162(m) We denemillion as the minimum of 1 and the executiversquos salary divided by $1million in the previous year Thus any executive with salary at or above$1 million would receive a value of 1 while an executive with a salary of$500000 would be coded as a 0516 We include this variable by itself andalso interacted with a dummy variable (after) that equals 1 for years1994 and after Thus the million variable is designed to pick up anyunderlying relationship between the level of salary and the growth rateof salary while the interacted one will identify any differential growthrate of salary by income in the period after section 162(m) took effect Wealso include a full set of year dummies the natural logarithm of thermrsquos market value and the rmrsquos annual rate of return over the twoprevious years as control variables To reduce the sensitivity of our re-sults to outliers we run quantile regressions and robust regressions

Table 6 contains our results The rst four columns are for all topexecutives while the last four columns focus on CEOs Columns (1) and(2) present estimates from robust regression for the percentage change insalary and the percentage change in total compensation (salary bonusstock grants and stock-option grants) In the salary growth regressionsfor all executives the coefcient on million is negative and signicantindicating that there is an underlying relationship between the level ofexecutive compensation and its subsequent growth rate The coefcienton million 3 after is negative as well suggesting that the negativerelationship between the level of salary and its subsequent growth rateintensied slightly after 1994 The relatively small magnitude suggeststhat an executive with a $1 million dollar salary would see his salarygrow at an annual rate that would be about 06 percent slower than anexecutive earning $500000

15 The HallndashLiebman data set contains only a combined salary and bonus variable makingit impossible to use it to analyze this issue16 We also ran the regressions (not reported) allowing for a nonlinear relationship betweenmillion and compensation growth and obtained substantively similar results

TA

BL

E6

The

Eff

ect

ofth

eM

illi

on-D

olla

rR

ule

onSa

lary

and

Tot

alC

ompe

nsat

ion

All

top

exec

uti

ves

CE

Os

only

Rob

ust

regr

essi

onQ

uan

tile

regr

essi

onR

obu

stre

gres

sion

Qu

anti

lere

gres

sion

D

in

Din

D

in

Din

D

in

Din

D

in

Din

sala

ryto

talc

omp

(a)

sala

ryto

talc

omp

sa

lary

tota

lcom

p

sala

ry

tota

lcom

p

mil

lio

n(b

)2

006

02

035

72

006

32

024

12

008

72

023

12

010

32

024

5(0

005

)(0

032

)(0

006

)(0

027

)(0

010

)(0

073

)(0

011

)(0

056

)m

illi

on

32

001

10

026

20

021

20

015

20

016

014

92

003

40

158

aft

er(c

)(0

005

)(0

033

)(0

006

)(0

029

)(0

010

)(0

076

)(0

011

)(0

057

)N

359

5035

950

359

5035

950

757

17

589

757

17

589

Stan

dar

der

rors

inp

aren

thes

esR

egre

ssio

nsal

soin

clu

de

year

du

mm

ies

stoc

km

arke

tre

turn

and

log

mar

ket

valu

e

(a)

Tot

alco

mp

ensa

tion

incl

ud

essa

lary

bon

us

rest

rict

edst

ock

and

stoc

k-op

tion

gran

ts

(b)

mil

lion

isd

ene

das

min

(1s

alar

yin

pre

viou

sye

ars

div

ided

by$1

mil

lion)

(c)

aft

eris

anin

dic

ator

vari

able

equ

alto

zero

befo

reth

ere

gula

tion

took

plac

e(1

993)

and

one

afte

rwar

d

36 Hall amp Liebman

In the second column the coefcient on million 3 after is essentiallyzero suggesting that any decrease in salary brought about by section162(m) was offset by increases in bonus and stock-option grants Theresults are similar for the median regressions There is some evidence ofa very small slowdown in growth of salary for executives likely to beaffected by section 162(m) but no sign of a change in total compensa-tion Taken together this provides evidence of a very small substitutionof performance-related pay for salary

The last four columns run identical tests but limit the sample to CEOsThe interaction of million 3 after continues to be small and negative inthe salary regressions In contrast the impact of 162(m) on total compen-sation appears to be positive and fairly substantial (about 15 percentagepoints per year) in these regressions That is the increase in bonus andstock options following the million-dollar rule more than offset the verysmall decline in salary

All of these conclusions should be interpreted somewhat cautiouslybecause our data provide us with only one pre-1993 control year Never-theless the data are consistent with section 162(m) having led to a veryminor slowdown in salary growth and one that is dramatically smallerthan what one would conclude without controlling for underlying differ-ential salary growth rates at different levels of salary Since total compen-sation for CEOs did not decline and perhaps increased there is evidenceof a minor substitution of performance-related pay for salary in responseto the regulation

8 EXECUTIVE COMPENSATION AND THEELASTICITY OF TAXABLE INCOME

Even if taxes have only a minor effect on executive compensation thestructure of executive compensation is important for tax policy be-cause high-income taxpayers are sources of signicant revenue and arethe focus of important debates about the efciency cost of taxation

Goolsbee (1999) has recently argued that executive decisions to exer-cise options are highly responsive to intertemporal differences in taxrates created by tax reforms He claims that much of the apparent declinein taxable income for high-income taxpayers between 1992 and 1993documented by Feldstein and Feenberg (1996) can be explained by execu-tives shifting option gains into 1992 in order to avoid the higher marginaltax rates that went into effect in 1993 By extension he implies that thehigh elasticities of taxable income with respect to personal tax ratesestimated by comparing the pre- and post-tax-reform taxable income of

Taxation of Executive Compensation 37

high-income individuals reect intertemporal shifting of income and notpermanent effects of taxation If this is true then the deadweight loss oftaxation is lower as is the cost of progressivity

While Goolsbee presents comprehensive evidence that option gainswere unusually high in 1992 his identication of the impact of tax rates onthe timing of option gains depends on correctly specifying a counter-factual level of options that would have been exercised in the absence ofthe tax change He simply assumes a linear time trend The purpose ofGoolsbeersquos time trend is presumably to account for the increasing use ofoptions over time However since options typically require a vestingperiod of at least a couple of years a much more direct measure of ex-pected option gains can be constructed based on the total value of vestedoptions held by the CEO Since options typically have vesting periods thatexceed two years this total can be treated as exogenous since it will not beaffected by the tax changes

More generally the value of options held by the CEO is likely to havea major effect on his exercise decisions If his options are out of themoney he will not have gains to realize If there has been a recent run-up in the market an optimizing CEO may exercise an unusually largeamount of options in order to diversify his overall portfolio and if in-the-money options are about to expire they will be exercised in thecurrent year

Our evidence suggests that much of the apparent tax shifting caninstead be attributed to stock market performance and the timing of pastoption grants Using our 15-year panel we replicate Goolsbeersquos (1999)result for the 1993 period but show that the impact of taxes is notpresent in other time periods Moreover even the result for the 1993period disappears when appropriate controls for past option grants andstock price appreciation are included

The top three rows of Table 7 present data on the taxable incomes ofCEOs in the HallndashLiebman sample For these results our panel has beennarrowed to include only CEOs whose rmrsquos scal years correspondwith tax years In addition we create a balanced panel for each of thethree tax-reform periods by excluding any rm for which there are notcomplete data for the period surrounding the reform

The fourth row of the table shows that the 1993 tax act raised the topmarginal personal income tax rate from 310 to 396 percent effective fortax year 1993 with a further increase occurring in 1994 due to the uncap-ping of the Medicare payroll tax The top row of the table shows thattaxable income was higher in 1992 than for any other year in our sample(all data are in 1994 dollars) Option gains were particularly high in1992mdashmore than double their 1991 level and 68 percent higher than their

TA

BL

E7

Tax

Ref

orm

Sto

ckM

arke

tP

erfo

rman

cea

ndth

eT

imin

gof

Opt

ion

Gai

ns

Ave

rage

stoc

km

arke

tM

ean

CE

OM

ean

CE

Ore

turn

for

Mea

nva

lue

taxa

ble

sala

ryan

dM

ean

CE

OT

opfe

der

al

rms

inof

opti

ons

(Val

ue

ofop

tion

Sam

ple

inco

me

bonu

sop

tion

gain

sp

erso

nal

sam

ple

for

held

atbe

gga

ins)

(va

lue

Ref

orm

size

Yea

r($

mil

)($

mil

)($

mil

)M

TR

pre

vye

ar(

)of

year

ofop

tion

she

ld)

ER

TA

8127

919

800

874

064

80

163

7025

80

460

354

1981

083

10

672

009

869

245

070

61

3919

820

872

067

90

099

5011

90

686

144

1983

099

00

724

025

250

355

116

62

16

TR

A86

295

1985

119

70

837

024

050

76

116

02

0719

861

448

091

70

345

5042

81

776

194

1987

165

50

959

059

338

519

12

229

266

1988

163

21

058

039

028

07

237

71

6419

891

656

105

20

414

2817

92

345

176

1993

272

1990

209

51

028

061

828

232

288

02

15T

axA

ct19

911

875

102

00

547

312

01

268

82

0319

922

678

107

01

250

3146

64

552

275

1993

232

41

209

074

339

620

64

299

173

1994

196

41

262

037

842

515

45

113

074

Sam

ple

isH

allndash

Lie

bman

rm

sw

ith

Dec

embe

r

scal

year

sth

atha

dco

mp

lete

dat

afo

ron

eof

the

thre

eta

x-re

form

per

iod

sst

ud

ied

Taxation of Executive Compensation 39

1994 level These patterns are consistent with Goolsbeersquos claim that tax-payers responded to the anticipated increase in marginal tax rates byshifting gains into 1992

The evidence from the other two tax reforms is less clear suggestingthat there might be factors other than tax rates determining the timing ofoption gains ERTA 1981 reduced the top marginal tax rate from 70percent in 1980 to 50 percent in 1982 However there is no sign thatoption gains were reduced in 1980 and 1981 and then increased in 198217

There was a big increase in option gains in 1983 however Similarly thehighest year of stock-option gains in the period around 1986 was 1987 astrange year to take gains since taxpayers would have known that thepersonal rate would be 10 percentage points lower if they waited untilthe following year18

What all the peak years of option gains have in common is that theyfollow years of strong stock market performance The fth row of Table 7shows the average stock market return during the previous year for therms in our sample The large value of option gains in 1983 followed a355-percent increase in the stock market The 1992 boom in option gainsfollowed a 466-percent increase in the stock market the more modest1987 peak in option gains followed a 19-percent gain (and a 43-percentgain two years before)

Strong stock market performance raises the value of options availableto exercise The sixth row of the table shows the mean value of optionsheld by CEOs at the beginning of the year All three of the peak years ofoption gains correspond with peak years in terms of the value of optionsavailable for exercise Since stock market appreciation during the yearalso increases the value of options available to exercise later in that yearwe need to allow for current-year stock market appreciation as well

Table 8 contains Goolsbee-style regressions for each of the tax re-forms The rst two columns reprint the results from Goolsbee (1999)The third and fourth columns replicate his results in our sample Inparticular with a specication similar to his we are able to replicate hisnding that the current after-tax share has a positive coefcient and thefollowing yearrsquos after-tax share has a coefcient of roughly negative oneColumns (5) through (8) show that this pattern was unique to the 1993tax reform Indeed in the TRA 86 regressions the coefcient on the

17 As we explained above for executives covered by the maximum tax on earned incomethere was not a decline in tax rates between 1981 and 198218 It is possible that the large stock market decline in October 1997 may have promptedsome people to exercise options earlier than they otherwise would have However anincrease in option exercises typically follows stock market increases not declines

TA

BL

E8

Att

empt

sto

Rep

lica

teth

eT

ax-S

hift

ing

Res

ult

inO

ther

Tim

eP

erio

ds

Goo

lsbe

eG

ools

bee

1993

1993

Ent

ire

Ent

ire

regr

essi

onre

gres

sion

refo

rmre

form

TR

A86

TR

A86

ER

TA

81E

RT

A81

1980

ndash199

419

80ndash1

994

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

ln[1

2ta

x(t)

]1

244

108

20

455

032

62

017

62

017

22

016

72

016

70

026

20

068

(01

34)

(01

27)

(05

37)

(05

42)

(02

10)

(02

09)

(00

90)

(00

91)

(00

67)

(00

65)

ln[1

2ta

x(t1

1)]

20

867

20

764

21

214

20

917

047

80

722

20

103

004

30

176

018

3(0

118

)(0

112

)(0

491

)(0

499

)(0

157

)(0

174

)(0

072

)(0

072

)(0

068

)(0

066

)

Tim

e0

135

007

62

000

52

036

40

041

000

80

094

003

50

056

002

7(0

008

)(0

008

)(0

049

)(0

049

)(0

028

)(0

028

)(0

025

)(0

026

)(0

003

)(0

003

)

ln(m

arke

tva

lue)

066

10

489

023

70

334

039

5(0

015

)(0

058

)(0

039

)(0

042

)(0

017

)

Yea

rs19

91ndash1

995

1991

ndash199

519

90ndash1

994

1990

ndash199

419

85ndash1

989

1985

ndash198

919

80ndash1

983

1980

ndash198

319

80ndash1

994

1980

ndash199

4

Ad

just

edR

20

720

770

610

630

690

760

800

810

630

68

Obs

erva

tion

s13

856

132

371

385

136

41

470

143

31

113

106

23

968

385

9

Dep

end

ent

vari

able

isth

elo

gof

taxa

ble

inco

me

All

regr

essi

ons

con

tain

rm

xe

def

fect

sG

ools

bee

regr

essi

on(2

)al

soco

ntai

nstw

oot

her

vari

able

sno

tsh

own

inth

eta

ble

Stan

dar

der

rors

inp

aren

thes

es

Taxation of Executive Compensation 41

TABLE 9Alternative Explanations for the Timing of Option Exercises

1993 Entirereform TRA 86 ERTA 81 1980ndash1994

(1) (2) (3) (4)

ln[12tax(t)] 0108 0111 20051 0201(0324) (0136) (0048) (0057)

ln[12tax(t11)] 20134 0573 0125 0150(0336) (0156) (0053) (0066)

ln(market value) 0418 0177 0338 0418(0061) (0039) (0046) (0163)

Firmrsquos stock market return (t) 0298 0268 0214 0258(0044) (0045) (0031) (0023)

Firmrsquos stock market return 0162 0206 0031 0128(t21) (0041) (0008) (0029) (0021)

ln(value of options held by 0004 0008 0001 0009CEO at start of year) (0003) (0003) (0002) (0001)

Years 1990ndash1994 1985ndash1989 1980ndash1983 1980ndash1994Adjusted R2 064 071 082 068Observations 1349 1427 1036 3812

The dependent variable is the log of taxable income All regressions contain rm xed effects Stan-dard errors in parentheses

following yearrsquos tax rate is positive and signicant Columns (9) and (10)pool all three reforms and fail to replicate the pattern that would reectintertemporal shifting19

It is possible that the ease with which one can shift the timing ofoption gains is asymmetric It may be harder to shift option gains for-ward in time since some options expire and therefore must be exercisednow In addition options were a much smaller portion of compensationin the early 1980s so perhaps the ERTA 81 results are not surprisingNonetheless these results cast doubt on the proposition that this is thechannel through which spurious estimates of the elasticity of taxableincome have occurred for previous tax reforms and they suggest thatwe should look harder for alternative explanations for the timing ofoption gains

Table 9 attempts to predict options gains using the lagged and currentincrease in a rmrsquos market value and the value of the options held by theCEO at the start of the year The two stock market variables stronglypredict option gains while the value of options held has the correct sign

19 The results also fail to suggest large permanent effects of marginal tax rates on taxableincome

42 Hall amp Liebman

in all of the specications and is signicant in the pooled sample eventhough it is highly correlated with the two stock market return variablesIn contrast the tax-timing explanation fails even in the 1993 period

It is important to realize that our results do not invalidate Goolsbeersquoscritique of claims that aggregate revenue data from 1992 and 1993 provethat elasticities of taxable income are large We have simply given analternative explanation for why taxable incomes were unusually high in1992 However our results do cast doubt on the conclusion that largetiming shifts are ubiquitous and capable of explaining away the morecareful estimates of taxable income elasticities that have been performedusing micro data (eg Feldstein 1995 Auten and Carroll 1995)

One further point is in order It is sometimes argued that the combina-tion of increased use of options and the booming stock market explainswhy federal tax revenues have been so much higher than expected dur-ing the 1990s However while large option gains increase revenue fromthe personal income tax they produce offsetting deductions for cor-porations The net impact on revenue is only the difference in the twomarginal tax rates times the option gain While the gap between thepersonal rate and the effective corporate rate is nontrivial the impact oflarge option exercises on total revenue is substantially smaller than theamount by which the exercises inate personal tax revenue and reducecorporate tax revenue

9 CONCLUSION

We have described the tax rules for executive pay in detail and analyzedhow changes in various tax rates affect the tax advantages of stock optionsrelative to salary and bonus We nd that there is a moderate tax advan-tage to optionsmdashcurrently about $4 per $100 in pretax executive paymdashand that changes in the tax advantage of options over time have had atmost a modest impact on the composition of pay Corporate governancefactors particularly the role of large investors appear to be more impor-tant in explaining the dramatic increase in option pay

In addition our evidence suggests that more direct attempts to use taxpolicy to inuence executive compensation have had little effect Wend that the million-dollar rule led companies to substitute perfor-mance-based pay for salary But our evidence suggests that this substitu-tion was quite modest and there is no evidence that the total level of paywas reduced Overall although the stock-option explosion has dramati-cally increased the link between pay and performance this change isdue almost entirely to nontax factors

Taxation of Executive Compensation 43

REFERENCES

Auerbach Alan J (1996) ldquoThe Dynamic Effects of Tax Law Asymmetriesrdquo Re-view of Economic Studies 53451ndash405

mdashmdashmdash and James Poterba (1987) ldquoTax Loss Carryforwards and Corporate TaxIncentivesrdquo In The Effects of Taxation on Capital Accumulation M Feldstein(ed) Chicago University of Chicago Press

Altshuler Rosanne and Alan Auerbach (1990) ldquoThe Signicance of Tax LawAsymmetries An Empirical Investigationrdquo Quarterly Journal of Economics 10561ndash86

Auten Gerald and Robert Carroll (1995) ldquoThe Effect of Income Taxes on House-hold Behaviorrdquo Ofce of Tax Analysis US Department of the Treasury

Baker George P and Brian J Hall (1998) ldquoCEO Incentives and Firm SizerdquoNBER Working Paper no 6868 December

Core John and Wayne Guay (1999) ldquoThe Use of Equity Grants to ManageOptimal Equity Incentive Levelsrdquo Draft

Feldstein M (1995) ldquoThe Effect of Marginal Tax Rates on Taxable Income APanel Study of the 1986 Tax Reform Actrdquo Journal of Political Economy 103 (no3)551ndash72

mdashmdashmdash and D Feenberg (1996) ldquoThe Effect of Increased Tax Rates on TaxableIncome and Economic Efciency A Preliminary Analysis of the 1993 Tax RateIncreasesrdquo In Tax Policy and the Economy vol 10 James M Poterba (ed)Cambridge MA MIT Press

Gompers Paul A and Andrew Metrick (1998) ldquoInstitutional Investors andEquity Pricesrdquo NBER Working Paper no 6723

Goolsbee Austan (1999) ldquoWhat Happens When You Tax the Rich Evidencefrom Executive Compensationrdquo NBER Working Paper no 6333

Graham John R (1996) ldquoProxies for the Corporate Marginal Tax Raterdquo Journal ofFinancial Economics 42187ndash221

Hall Brian J (1999) ldquoThe Design of Multi-year Stock Option Plansrdquo Journal ofApplied Corporate Finance 12(2)97ndash106

mdashmdashmdash and Jeffrey B Liebman (1998) ldquoAre CEOs Really Paid Like Bureau-cratsrdquo Quarterly Journal of Economics CXII(no 3)653ndash691

Jensen Michael C and William M Meckling (1976) ldquoTheory of the Firm Mana-gerial Behavior Agency Costs and Ownership Structurerdquo Journal of FinancialEconomics III305ndash360

mdashmdashmdash and Kevin J Murphy (1990) ldquoPerformance Pay and Top-ManagementIncentivesrdquo Journal of Political Economy XCVIII225ndash264

Kaplan Steven N (1997) ldquoThe Evolution of US Corporate Governance WeAre All Henry Kravis Nowrdquo Draft

Livingstone Jane R (1999) ldquoExecutive Compensation Contracting and Re-sponses to Increased Tax Costsrdquo Louisiana State University June

Miller Merton H and Myron S Scholes (1982) ldquoExecutive CompensationTaxes and Incentivesrdquo In Financial Economics Essays in Honor of Paul CootnerWilliam F Sharpe and Cathryn M Cootner (eds) Princeton NJ Prentice-Hall

Murphy Kevin J (1999) ldquoExecutive Compensationrdquo forthcoming in OrleyAshenfelter and David Card (eds) Handbook of Labor Economics Vol 3 NorthHolland

Perry Todd and Marc Zenner (1999) ldquoPay for Performance Government Regu-

TA

BL

E1

Tax

and

Acc

ount

ing

Tre

atm

ent

ofE

xecu

tive

Com

pens

atio

nC

ash

Pay

men

tsve

rsus

Stan

dard

Exe

cuti

veO

ptio

ns

Tax

atio

nat

Tax

atio

nat

Acc

ount

ing

Com

pens

atio

nD

escr

ipti

onp

erso

nall

evel

corp

orat

ele

vel

trea

tmen

t

Sala

ryan

dbo

nu

sC

ash

pay

men

tsm

ade

Tax

edas

ord

inar

yFu

llyd

edu

ctib

leS

ince

Exp

ense

dag

ains

tea

rn-

toC

EO

in

com

e19

93a

ny

pay

over

$1in

gsin

the

usu

alw

ay

mill

ion

isno

td

educ

t-ib

leu

nles

sit

isp

er-

form

ance

-bas

ed

Non

-qua

lie

dst

ock

Op

tion

sgr

ant

the

ex-

The

gran

ting

ofan

op-

Apa

ralle

ltax

ded

uc-

The

reis

no

exp

ense

rec-

opti

ons

(NQ

SOs)

ecut

ive

the

righ

tbu

tti

onis

typ

ical

lyan

un-

tion

atth

eco

rpor

ate

ogni

tion

(atg

ran

tex

er-

not

the

obli

gati

ont

ota

xed

even

tU

pon

exer

-le

veli

sge

nera

llyal

-ci

seo

rsa

le)

for

op-

buy

shar

esof

stoc

kat

cise

opt

ion

pro

ts

are

low

edu

pon

exer

cise

tion

sif

as

isu

sual

lyan

exer

cise

pric

e(t

ypi-

taxe

dat

the

ord

inar

y(f

orth

eam

ount

ofop

-th

eca

set

he

opti

ons

cally

the

curr

ent

stoc

kra

teI

fth

eex

ecu

tive

tion

pro

ts

)O

pti

ons

are

not

dis

cou

nted

(in

pri

ceat

the

dat

eof

cont

inu

esto

hold

onto

auto

mat

ical

lyqu

alif

yas

the

mon

eyat

gran

tgr

ant)

Th

eop

tion

sth

esh

ares

the

exec

u-

perf

orm

ance

-bas

edp

ayd

ate)

and

the

exer

cise

typ

ical

lyve

stov

era

3ndashti

veis

taxe

din

the

and

ther

efor

ear

eno

tpr

ice

and

num

ber

ofop

-5-

year

per

iod

The

typ

i-us

ualw

ay(a

tth

eca

pi-

subj

ect

toth

e$1

-ti

ons

iskn

own

atgr

ant

calm

atu

rity

is10

year

sta

lgai

nsra

te)

onan

ym

illio

nca

p

dat

eFo

rd

isco

unt

op-

but

the

man

ager

may

furt

her

app

reci

atio

nof

tion

sth

ed

iffe

renc

ebe

-ex

erci

seea

rly

Som

eth

esh

are

pri

ce

twee

nth

est

ock

pri

ceop

tion

sar

edi

scou

ntop

-an

dth

em

arke

tp

rice

isti

ons

(the

exer

cise

pri

ceex

pen

sed

over

the

vest

-is

low

erth

anth

est

ock

ing

per

iod

For

opti

ons

pri

ceat

the

gran

td

ate)

w

ith

vari

able

term

sP

rem

ium

opti

ons

are

(eg

a

vari

able

exer

-th

ere

vers

e(t

heex

er-

cise

pri

cev

esti

ngth

atci

sep

rice

isgr

eate

ris

tied

top

erfo

rman

ce)

than

the

stoc

kp

rice

atth

eop

tion

sar

em

arke

dgr

ant

dat

e)

tom

arke

tan

dex

-p

ense

dd

uri

ngth

eti

me

betw

een

gran

tan

dex

erci

se

Not

eW

eth

ank

com

pen

sati

onco

nsu

ltan

tsSc

ott

Gre

enbe

rgan

dSc

ott

Ols

en(b

oth

ofT

ower

sP

erri

n)

and

Fred

eric

Coo

k(o

fFr

eder

icW

C

ook

and

Ass

ocia

tes)

for

help

fulc

onve

rsat

ion

sin

und

erst

and

ing

the

tax

and

acco

unt

ing

rule

s

10 Hall amp Liebman

Second the tax treatment of ISOs is different ISOs are completely un-taxed at grant or exercise It is only at sale that the executive is taxed andeven then the executive is taxed at the lower capital gains tax rate Thedisadvantage is that the corporation never gets to take a parallel taxdeduction against corporate prots Thus ISOs become more attractiveas the personal tax rate increases and as the corporate tax rate and thecapital gains tax rates fall

Two other related types of compensation restricted stock and stockappreciation rights are worth brief discussion Both however are farless common than standard stock options Restricted stock is payment inthe form of restricted shares the restriction being that the shares vestover time as with options Unlike options the shares typically havevoting and dividend rights With regard to taxation the executive istaxed at the personal rate on the value of the restricted stock as thevesting restrictions lapse However the executive may choose to betaxed at the grant date in which case all subsequent appreciation istaxed at the capital gains rate The dividends paid to the executive aretaxed at the ordinary rate in the usual way The company generallyreceives a parallel deduction equal to the amount of the executiversquosincome when the executive is taxed Unlike stock options restrictedstock is not generally considered to be performance-based and is there-fore subject to the million-dollar rule (unless the vesting of the restrictedstock is performance-based which is sometimes the case)

Stock appreciation rights (SARs) are rights that replicate the payoffs ofstock options with a cash transfer Thus SARs are simpler than options inthat there is no requirement to buy and resell the stock in order to ldquocashoutrdquo SARs generally have the same tax treatment as NQSOs both to theindividual and to the corporation SARs are relatively rare however be-cause they have disadvantageous accounting treatment (described in thenext section) relative to stock options and their main relative advantagevis-a-vis options has been essentially replicated through broker-assistedcashless option exercisesmdashwhereby a broker makes a ldquonanosecondrdquo loanto the executive (to purchase and resell the stock) when the executivewants to ldquocash inrdquo on option prots The tax and accounting treatment ofISOs restricted stock and SARs is summarized in Table 2

32 The Accounting Treatment of OptionsUnlike cash compensation which is expensed against earnings there isgenerally no expense recognition (at grant exercise or sale) for optionswhether they be NQSOs or ISOs As a result compensation consultantsoften point out that stock options are the only form of compensation thatare free in an accounting sense but still deductible for tax purposes

Taxation of Executive Compensation 11

Stock options do however lead to expense recognition if they are dis-counted (in the money at grant date) or if the exercise price and numberof options are not known at grant date For discount options the differ-ence between the stock price and the market price is expensed over thevesting period For options with variable terms (eg a variable exerciseprice vesting that is tied to performance) the options are marked tomarket and expensed during the time between grant and exercise Practi-tioners claim that the accounting treatment of options plays an impor-tant role in the design of option programs Thus plans that have ldquobadaccountingrdquo but are thought by many to have attractive incentive fea-tures are often not even seriously considered by companies Examples ofsuch potentially attractive plans include indexed options (where a CEOprots only if his rmrsquos share price grows relative to some market orindustry benchmark) and option grants that are explicitly performance-related both of which would lead to expenses against earnings

Unlike options restricted stock and SARs do not generally receivefavorable accounting treatment Restricted stock is generally expensedover the period in which the restrictions lapse (usually the vesting pe-riod) The magnitude of the expense is the difference between the cur-rent stock price and the executiversquos cost (if any) SARs are marked tomarket each period and the difference between the stock price and theexercise price is expensed over the outstanding period of SARs

4 THE TAXATION OF EXECUTIVE PAY CASHVERSUS OPTIONS

Stock-based compensation performs two roles in executive compensa-tion arrangements First it helps align the incentives of the executivewith the interests of the rmrsquos shareholders Second it often enables therm to compensate the CEO in a way that is more advantageous from atax standpoint than paying the executive in salary and bonus

41 Agency Theory and Executive Compensation in the Presenceof TaxationIn standard agency-theory models ( Jensen and Meckling 1976) agencycosts are the result of the separation of ownership and control Managersdo not have the same incentives as the owners The optimal incentivecontract for managers balances the benets of high-powered incentives(linking the fortunes of owners and managers through stock and stockoptions for example) with the costs of loading too much risk on risk-averse managers

TA

BL

E2

Tax

and

Acc

ount

ing

Tre

atm

ent

ofE

xecu

tive

Com

pens

atio

nIn

cent

ive

Stoc

kO

ptio

ns(I

SOs)

Res

tric

ted

Stoc

kan

dSt

ock

App

reci

atio

nR

ight

s(S

AR

s)

Tax

atio

nat

Tax

atio

nat

Acc

ount

ing

Com

pen

sati

onD

escr

ipti

onp

erso

nall

evel

corp

orat

ele

vel

trea

tmen

t

Ince

nti

vest

ock

ISO

sar

eid

enti

calt

oIf

the

usu

alco

ndit

ions

Ifth

eu

sual

cond

itio

nsA

sw

ith

NQ

SOs

ther

eop

tion

sN

QSO

sin

thei

rd

esig

n

are

met

th

ere

isno

tax

are

met

the

reis

gene

r-is

gene

rally

noex

pen

seH

owev

ert

hey

have

anat

gran

tor

exer

cise

At

ally

no

corp

orat

eta

xd

e-re

cogn

itio

nfo

rIS

Os

annu

alca

pof

$100

000

sale

all

opti

onp

ro

tsd

uct

ion

atgr

ant

exer

-(e

ver)

on

the

amou

ntth

atan

dsu

bseq

uen

tca

pita

lci

seo

rsa

le

vest

sT

he

tax

trea

t-ap

prec

iati

onar

eta

xed

men

tof

ISO

sal

soat

the

cap

ital

gain

sd

iffe

rs

rate

Res

tric

ted

stoc

kSi

mila

rto

stoc

kop

-T

hein

div

idu

alis

taxe

dT

heco

mp

any

gene

rall

yR

estr

icte

dst

ock

isge

n-ti

ons

but

shar

esat

the

ord

inar

yra

teas

rece

ives

ap

aral

leld

e-er

ally

exp

ense

dov

er(w

hich

vest

slow

ly)

the

rest

rict

ions

lap

se

du

ctio

neq

ual

toth

eth

eve

stin

gp

erio

dT

hera

ther

than

opti

ons

are

How

ever

the

man

ager

amou

ntof

the

exec

u-

amou

nt

exp

ense

dis

gran

ted

toth

eex

ecu

-m

aych

ose

tobe

taxe

dti

versquos

inco

me

whe

nth

eth

ed

iffe

renc

ebe

twee

nti

veU

nlik

eop

tion

sat

gran

td

ate

inw

hich

exec

uti

veis

taxe

dR

e-th

ecu

rren

tst

ock

pric

eth

esh

ares

typi

cally

case

alls

ubs

equ

ent

ap-

stri

cted

stoc

kis

gen

er-

and

the

exec

uti

versquos

have

voti

ngan

dd

ivi-

pre

ciat

ion

ista

xed

atal

lysu

bjec

tto

the

cost

(if

any)

d

end

righ

ts

the

cap

ital

gain

sra

te

mill

ion-

dol

lar

cap

D

ivid

end

sar

eta

xed

atth

eor

din

ary

rate

Stoc

kap

pre

ciat

ion

Rig

hts

that

repl

icat

eG

ener

ally

have

the

Gen

eral

lyha

veth

eU

nlik

eop

tion

sSA

Rs

righ

tsth

epr

ots

ofst

ock

op-

sam

eta

xtr

eatm

ent

assa

me

tax

trea

tmen

tas

are

mar

ked

tom

arke

tti

ons

Rar

ely

use

d

NQ

SOs

NQ

SOs

each

per

iod

an

dth

esi

nce

stoc

kop

tion

sd

iffe

renc

ebe

twee

nth

eha

vem

ore

favo

rabl

eac

-st

ock

pri

cean

dth

eex

-co

unt

ing

trea

tmen

ter

cise

pri

ceis

exp

ense

d(a

ndop

tion

sp

rovi

de

over

the

outs

tand

ing

the

sam

ebe

ne

tof

per

iod

ofSA

Rs

cash

less

exer

cise

thro

ugh

brok

erlo

ans

)

Not

eW

eth

ank

com

pen

sati

onco

nsul

tant

sSc

ott

Gre

enbe

rgan

dSc

ott

Ols

en(b

oth

ofT

ower

sP

erri

n)an

dFr

eder

icC

ook

(of

Fred

eric

WC

ook

and

Ass

ocia

tes)

for

hel

pfu

lcon

vers

atio

nsin

und

erst

and

ing

the

tax

and

acco

unt

ing

rule

s

14 Hall amp Liebman

The effect of taxation on the optimal contract (more precisely theshare of compensation that is performance-related) is ambiguous (evenin the absence of deductibility and deferral) because there are offsettingeffects First by reducing the share of corporate prots received byshareholders taxes diminish the importance to the shareholders of moti-vating the CEO Taken alone this effect would imply that taxes wouldbe expected to reduce the use of performance-based compensation Sec-ond because the government shares in the income received by the CEOthe variance of the CEOrsquos income is reduced raising his utility andlowering the cost to the rm of providing any given set of incentivesThird because taxes will lead executives to provide less effort for anygiven level of incentive-based pay the level of compensation that mustbe provided to compensate the CEO for effort is reduced for a givenamount of incentive These last two factors make it cheaper for the rmto offer contracts to the CEO with large incentive components when taxrates are higher and should therefore increase the use of incentive-based pay Since the net effect of taxation on the level of incentive-basedpay is ambiguous agency theory provides no strong predictions abouthow taxation should affect the optimal composition of option versuscash compensation

42 The Tax Advantages of DeferralWhile agency theory yields ambiguous predictions about the effect oftaxes on the use of options there are direct tax advantages of options inthat options provide executives with a way to defer compensation andthereby lower their taxes However because options also lead to a defer-ral of corporate tax deductibility the tax advantages from a combined(executive and corporate) perspective are not so straightforward In thissection we analyze and measure the tax advantages of stock optionsrelative to salary and bonus compensation We also analyze the taxadvantages of NQSOs relative to ISOs In particular we show how thetax advantages of options change as personal corporate and capitalgains taxes change We then show how the tax advantages of optionshave changed over time in response to changing tax rates

The crucial tax difference between standard options (NQSOs) andcash payment is that option payouts are deferred and the two forms ofcompensation earn different rates of return over the deferral period Anyanalysis of the relative tax advantages of two compensation instrumentsmust consider the tax consequences both to the employer and to theemployeemdashwhat Scholes and Wolfson (1992) call the global contractingperspective Thus in order to make valid comparisons between the twowe compare the tax burden to the executive while holding constant the

Taxation of Executive Compensation 15

posttax cost to the company By keeping posttax employer costs (in NPVterms) constant any package that is preferred by the employee is tax-advantaged in the global-contracting sense

Under this methodology a comparison of the tax advantages of op-tions and cash involves comparing a pretax cash payment of P with anoption payment that has the equivalent posttax NPV to the companyFor NQSOs it is straightforward to show that a pretax payment of P toan executive has exactly the same cost to the company as putting aside Pfor the purposes of paying stock-option payouts later That is if a com-pany pays an executive P today it will have the same amount of moneyin N years as if it had put aside P today let it accumulate at the rate ofreturn earned by the rm and then paid it out (as compensation in theform of option gains4) with any appreciation in year N

Assume that the pretax prot rate is r and the corporate tax rate is TcThen if the company pays P today it receives a deduction today of PTcwhich yields

PTc [1 1 r (1 2 Tc)]N (1)

in N years since the benets of the tax deduction are invested in thecompany and receive the after-tax corporate rate of return Converselyif the company puts P aside today then it grows at the after-tax corpo-rate rate of return in N years to give a tax deduction of

P [1 1 r (1 2 Tc)]N Tc (2)

which is the exact same value Note that the equivalence of these two taxdeductions is analogous to the tax-benet equivalence of front-loadedand back-loaded IRAs

An executiversquos payoff in N years from option prots is equal to Pcompounded at the after-tax corporate rate times 1 2 Tp where Tp is thepersonal tax rate Thus the combined payoffs in N years of the corporatededuction and the executiversquos posttax payoff is

P [1 1 r (1 2 Tc)]N Tc 1 P [1 1 r (1 2 Tc)]

N (1 2 Tp) (3)

where the rst term is the payoff from the corporate deduction and thesecond term is the executiversquos payoff

4 This analysis is not unique to options Any form of deferred compensation that enablesexecutives to invest inside the rm at a higher posttax return or without paying capitalgains taxes would have similar effects

16 Hall amp Liebman

The payoff to salary and bonus in N years is more complicated sincean assumption must be made about how the executive invests the origi-nal cash compensation For example if the executive invests in an instru-ment with taxable interest (eg bonds) then the N-year payoff from aninvestment of P accumulates at the post-corporate-tax post-personal-taxrate of return and is

P (1 2 Tp) [1 1 r (1 2 Tc)(1 2 Tp)]N (4)

However if an executive invests in (non-dividend-paying) equities theinvestment accumulates at the higher post-corporate-tax rate of returnwith the offsetting disadvantage that the capital gains are taxed at thecapital gains rate in year N So the executive receives

P [1 1 r (1 2 Tc)]N (1 2 Tp) 2 Tcg $P (1 2 Tp) [1 1 r (1 2 Tc)]

N 2P (1 2 Tp) (5)

where Tcg is the capital gains tax rate5Thus putting the equations together if an executive is paid cash P

and invests g in bonds and 1 2 g in equities the N-year payoff (includ-ing the payoff to the corporation from the tax deduction) is

P [1 1 r (1 2 Tc)]N Tc 1 g $(P [1 1 r (1 2 Tc)(1 2 Tp)]

N (1 2 Tp)1(1 2 g) (P [1 1 r (1 2 Tc)]

N (1 2 Tp) 2 Tcg $P (1 2 Tp)[1 1 r (1 2 Tc)]

N 2 P (1 2 Tp)) (6)

The tax advantage of options versus salary and bonus therefore is thedifference between the combined (corporate and executive) payoffs inequation (3) and the combined payoffs in equation (6)

43 The Tax Advantage of OptionsWe now turn to analysis of how changes in various tax rates affectthe tax advantages of options relative to cash compensation We focuson the case in which the cash earned by executives is invested entirelyin equities since equities are tax-favored relative to bonds The impor-tant conclusions of this analysis are not substantively different if we

5 In practice executives hold a combination of equities and bondlike instruments withtaxable interest Because equities are tax-favored in the next section we simplify the analy-sis by assuming that executives hold only equities Alternatively we could have assumedthat executives hold only less risky bonds since bonds more closely match the risk proleof a stream of tax savings None of our empirical results are substantively affected by thissimplication

Taxation of Executive Compensation 17

instead assume that a portion of an executiversquos holdings are invested inbonds

431 The Tax Advantage of Options Changing the Corporate TaxRate Stock options are tax-advantaged relative to cash simply becauseoption payouts are deferred allowing the executive to invest at the pre-personal-tax rate of return (with no capital gains tax at the end) ratherthan investing at the after-tax rate of return or paying the capital gainstax Since the advantage of being able to defer taxes is large when thecorporate tax rate is low the tax advantage of options is larger when thecorporate rate is low

To illustrate this point we dene the tax advantage of options to bethe dollar amount by which total after-tax option payoffs (to the corpora-tion and the executive) exceed the total payoffs from salary as denedby the difference of equation (3) and (6) assuming that P (the paymentto the executive) is equal to $100 and g 5 0 (cash compensation is in-vested in equities) We use a ten-year horizon since most options have aten-year term We then calculate how the tax advantage of optionschanges as the corporate tax rate varies from zero to 100 percent holdingthe personal tax rate and the capital gains rate xed at 40 percent and 20percent respectively (which approximates current rates)

The top panel of Figure 2 shows the tax advantage of options relativeto cashmdashprecisely the NPV of the total tax advantage of options for a$100 payment to the CEOmdashas the corporate tax rate changes As ex-pected the tax advantage of options declines as corporate tax rates in-crease At a zero corporate tax rate the tax advantage of options has anNPV of $7 This value is about $350 at a corporate rate of 40 percentWhen the corporate tax rate is 100 percent the tax advantage of optionscompletely disappears since the benets of deferred compensation fallto zero because the after-tax return on equities (and therefore the dis-count rate) falls to zero The tax advantage of options is the NPV of thetax savings from avoiding the capital gains tax on ten years of apprecia-tion which is why it is relatively modest even at the 40-percent corpo-rate rate

432 The Tax Advantage of Options Changing the Personal TaxRate Using the same assumptions as above (but this time holding thecorporate rate at 40 percent and the capital gains rate at 20 percent andvarying the personal rate) the tax advantage of options declines as thepersonal tax rate rises as shown in the middle panel of Figure 2 The taxadvantage of options is higher at low personal tax rates because theposttax base that gives rise to capital gains is higher and the advantage of

Note Holds constant the personal rate at 40 percent and the capital gains rate at 20 percent

Note Holds constant the corporate rate at 40 percent and the capital gains rate at 20 percent

Note Holds constant the personal rate at 40 percent and the corporate rate at 40 percent

FIGURE 2 Tax Advantage of Options NPV of Tax Advantage of Pay-ing $100 in Options (NQSOs) Rather Than in Cash

Taxation of Executive Compensation 19

avoiding the capital gains taxes is greater The tax advantage of optionstherefore declines linearly as the personal tax rate rises At a 100-percentpersonal tax rate the executive receives nothing in either case and byconstruction the companyrsquos tax deduction benet is the same in NPVterms so the tax advantage of options falls to zero also6

433 The Tax Advantage of Options Changing the Capital Gains TaxRate The tax advantage of options for various capital gains rates isdepicted in the bottom panel of Figure 2 under the same assumptions(this time holding both the personal rate and the corporate rate at 40percent while varying the capital gains rate) The tax advantage of avoid-ing the capital gains tax increases as the capital gains tax rate rises

5 HOW HAS THE TAX ADVANTAGE OF OPTIONSCHANGED OVER TIME

The analysis so far has illustrated how the tax advantage of optionsvaries with the changes in tax rates In this section we show how the taxadvantage of options has changed over time as tax rates have changed

51 Top Marginal Tax Rates over TimeThe evolution on the top marginal tax ratesmdashpersonal corporate andcapital gainsmdashfrom 1980 to 1998 is shown in Figure 3 In all cases thetop rate (the rate for taxpayers with the highest incomes) is shownwhich is not always the highest rate since various anomalies (such asclawbacks of exemptions) sometimes lead to marginal tax rates that arehigher than those for the highest income tax payers

The top personal tax rate was 70 percent in 1980 falling to 50 percentin 1982 and 28 percent in 1988 following the 1986 tax act7 The topmarginal tax rate has since risen to 396 percent but is effectively 425percent since there is a 29-percent Medicare surcharge (paid half by theemployer and half by the employee) that has no upper limit The topcorporate tax rate has had only one signicant change since 1980 fallingfrom 46 percent before the 1986 tax act to 34 percent following the act in1988 The top corporate tax rate was increased to 35 percent in 1993 The

6 This is one place where ignoring bond investments has substantive implications Inparticular in the all-bond case the tax advantage of options is nonmonotonicmdashit rises andthen fallsmdashas personal tax rates rise7 During the early 1980s the maximum tax on earned income limited the marginal tax rateon the earnings of high earners so many executives did not face a decline in personal taxrates between 1981 and 1982 The empirical results in this paper are robust to assumingthat executives faced a marginal tax rate of 50 percent in 1980 and 1981

20 Hall amp Liebman

FIGURE 3 Top Marginal Tax Rates since 1980

top capital gains rate has uctuated between 28 percent (approximately)and 20 percent since 1980 and is currently at 20 percent

52 The Evolution of the Tax Advantage of NQSOsThe tax advantage of options relative to cash dened in the same way asin section 4 from 1980 to 1998 is shown in Figure 4 The calculations aredone using the same assumptions as before (ie N 5 10 years and cashis invested in equities) and the statutory top marginal tax rates shown inFigure 3 are used Due to the fall in corporate and personal tax rates thetax advantage of options increased sharply after the Tax Reform Act of1986 This increase in tax advantage has largely been reversed in the1990s as top personal tax rates have crept back up and the capital gainsrate has been reduced

Two key facts emerge from this analysis First there is currently only amoderate tax advantage to standard nonqualied stock optionsmdashon theorder of $4 per $100 of compensation This is because the tax advantagesto the executive of deferring taxes are largely offset by the tax disad-vantages to the company of not being able to deduct option expensesfrom taxable prots until the executive exercises the options Secondalthough the 1986 tax act substantially increased the tax advantage ofoptions more than half of this increase has been reversed in the 1990s

Taxation of Executive Compensation 21

FIGURE 4 Tax Advantage of Options 1980 to 1998

NPV of tax advantage of $100 in options (NQSOs) rather than cash

53 The Tax (Dis)advantage of ISOsSo far the analysis has focused on the tax advantages of NQSOs sincethey are so much more prevalent than ISOs which account for onlyabout 5 percent of option grants The relative scarcity of ISOs can beexplained both by their tax status and by their per-executive cap of$100000 per year

Under what conditions are ISOs tax-preferred to NQSOs For a trans-fer P to an executive the ISO is always tax-preferred by the executive(since the capital gains rate is lower than the personal rate) while theNQSO is always tax-preferred from the companyrsquos perspective (since anISO is not deductible) The key issue however is the relative advan-tages of the two types of options from a global-contracting perspectiveBecause ISOs are not deductible the company is indifferent betweensetting aside P in the form of NQSOs today and setting aside P(1 2 Tc) inthe form of ISOs (Since the NPV of the tax deduction for NQSOs is PTcthe NPV of the payment P is P 2 PTc or P (1 2 Tc))

Equalizing the posttax cost of NQSOs and ISOs we need only look atthe payoff of the executive to determine the condition under which eachis preferred The payoff of NQSOs in N years is

P [1 1 r (1 2 Tc)]N (1 2 Tp) (7)

22 Hall amp Liebman

The same payoff for ISOs is

P (1 2 Tc) [1 1 r (1 2 Tc)]N (1 2 Tcg) (8)

ISOs therefore are tax-advantaged if

Tp 2 Tcg Tc 2 Tc Tcg (9)

Intuitively ISOs are tax-advantaged only if their advantage (the differ-ence between the personal rate and the capital gains rate) is largeenough to offset their cost (the disadvantage of not deducting at thecorporate rate) Note however that the condition is not a simple com-parison between the corporate rate and the personal rate minus thecapital gains rate (Tp 2 Tcg Tc)

Figure 5 shows how the tax advantage of NQSOs relative to ISOs haschanged since 1980 as tax rates have changed Unlike the very modestchanges (around 3 to 4 percent) in the relative tax advantages betweenoptions and cash shown in Figure 4 the relative tax advantage betweenNQSOs and ISOs has seen enormous swings When the top personalrate was 70 percent in 1980 ISOs were tax-favored by a margin greaterthan 6 percent However by 1982 (following the 1981 tax acts that dra-matically lowered the personal tax rate) ISOs became tax-disadvantaged

FIGURE 5 Tax Advantage of NQSOs Relative to ISOs over Time

NPV of tax advantage of $100 payment to executive NQSOs minus ISOs

Taxation of Executive Compensation 23

by about 5 percent By 1988 (following the 1986 tax act) ISOs became tax-disadvantaged by about 18 percent a dramatic swing The raising ofpersonal rates coupled with a decrease in the capital gains rate has sincereduced the tax disadvantage of ISOs to about $4 per $100 of pretaxcompensation Although hard data on ISOs are hard to come by practi-tioners (mostly compensation consultants) claim that ISOs were morepopular prior to 1982 when they were tax-advantaged Although ISOsare less disadvantageous from a tax standpoint relative to NQSOs thanthey were in the late 1980s they are still disadvantageous so it is notsurprising that they have not made a signicant resurgence in recentyears

6 EXPLAINING THE INCREASE IN OPTION-BASEDCOMPENSATION

In this section we analyze whether changes in tax rates affect the compo-sition of executive compensation and consider the relative importanceof tax factors and corporate governance factors in explaining the in-creasing share of compensation paid in the form of stock options Thenumerous changes in tax rates that have occurred since 1980 provide anopportunity to assess whether executive compensation arrangementsrespond to tax incentives in the way that the tax-avoidance model sug-gests We test this model using a panel data set of CEOs in largepublicly-traded US corporations Our identication of the tax effectsrelies on time-series variation in personal corporate and capital gainstax rates along with cross-sectional variation in corporate tax rates

61 DataWe use a 15-year panel data set of CEOs in the largest publicly-tradedUS corporations which is described in Hall and Liebman (1998) Thedata set covers the years from 1980 through 1994 and combines CEOcompensation information from corporate proxies and 10-K lings withstock price and stock return information from CRSP and accounting datafrom Compustat In addition some compensation data from the 1970swere collected in order to construct measures of the value of stock op-tions held by the CEOs in the rst period of the sample8

The feature that distinguishes our data from most other CEO datasets is that with our panel of yearly proxy data on option grants optiongains and total options held we are able to calculate the total value of

8 We are grateful to David Yermack for providing us with some of the data for 1984 to 1991See Yermack (1995) for a discussion of these data

24 Hall amp Liebman

all stock options held by the CEO at a point of time More importantlysince we have the details about the stock options held (number exer-cise price time to maturity etc) we can precisely calculate the changein the value of a CEOrsquos stock option holdings for a given change in rmvalue

62 Identication IssuesThe numerous tax changes over our sample period and the specic waysin which changes in personal corporate and capital gains rates arepredicted to affect the composition of CEO compensation give us un-usually rich sources of identication However because many of thechanges in tax rates would be expected to affect all of the CEOs in oursample in a similar way we will need to pay particular attention toseparating out the tax effects from underlying time trends and otherfactors that changed over time Controlling for underlying trends is par-ticularly important in this study because we are focusing on an out-comemdashthe use of stock optionsmdashthat has increased rapidly over thepast two decades and that many practitioners believe was importantlyaffected by nontax considerations Thus the tax effects we are trying toexplain are deviations from a rising trend

We take two steps to try to separate out the tax effects from theunderlying trend First we include key nontax factors that could poten-tially explain the increasing reliance on performance-based compensa-tion It has been suggested that the dramatic increase in incentive-basedpay is the result of the remarkably poor shareholder returns during the1970s which spurred the LBO and takeover movements of the 1980sAccording to this story (Kaplan 1997) shareholders became much morepowerful via the rise of institutional investors even as the LBO andtakeover movements waned Because the inuence of institutional inves-tors is thought to be one of the most important mechanisms of strongcorporate governance we use the share of each companyrsquos stock ownedby institutional investors (dened as institutions with more than $100million under management) as an explanatory variable9 The percentageof shares owned by large institutional investors increased from about 20percent to almost 50 percent in 1994 an upward trend that closelymatches the sharp rise in the share of compensation in the form ofoptions Indeed in our sample the annual average (over all of the rmsin our sample) share of stock owned by institutional investors has acorrelation of 09 with the annual average share of compensation paid inoptions

9 This variable is described in Gompers and Metrick (1998)

Taxation of Executive Compensation 25

In addition we include two other variables that proxy for strongercorporate governance The rst is the size of the board Evidence suggeststhat boards with smaller numbers of directors reduce the inuence of theCEO and better represent shareholder interests (Yermack 1995) Boardsize fell by approximately 15 percent between 1980 and 1994 The secondvariable is the fraction of the rmrsquos board members who are inside direc-tors or gray directors (noninsiders who have business dealings with thecompany) Firms with fewer outsiders are less likely to act in the share-holdersrsquo interests and we would predict that they would be less likely tohave performance-based compensation for their CEOs The average per-centage of inside and gray directors has decreased from about 45 percentin 1980 to about 35 percent in 1994

Second we allow for cross-sectional variation in corporate tax ratesCorporate marginal tax rates vary because rms may be eligible for tax-loss carrybacks and carryforwards investment tax credits and the alter-native minimum tax (see Auerbach 1986 Auerbach and Altshuler1990 Auerbach and Poterba 1987) Because these cross-sectional differ-ences imply that different rms should respond differently to a givenchange in tax rates they potentially provide us with a way to isolatethe tax effects Firm-specic corporate marginal tax rates are notori-ously difcult to calculate from publicly-available data We use a tricho-tomous variable that equals zero if the rm has tax-loss carryforwardsand negative earnings in a given year equals 05 times the statutorycorporate rate if the rm has only one of those conditions and equalsthe statutory rate if the rm has neither loss carryforwards nor negativeearnings10

The cross-sectional variation in the corporate tax rate may not be ex-ogenous Firms that perform poorly could face low marginal tax ratesand also be particularly likely or unlikely to provide performance-basedpay Because most of our specications use xed effects the correlationwe must be concerned about is between changes in rm performanceand changes in compensation We deal with this by controlling forlagged rm stock market performance in our regressions This shouldeliminate spurious correlation between tax rates and compensation thatis jointly caused by rm performance

10 Graham (1996) presents evidence that the trichotomous variable performs nearly as wellas his recommended simulated marginal tax rate Plesko (1999) compares marginal taxrates calculated from Compustat data with actual marginal tax rates from rm tax returnsHe nds that simple binary variables are more reliable measures of the marginal tax ratethan is either the Graham or the trichotomous variable We obtain substantively similarresults when we switch from the trichotomous variable to the binary variables recommendby Plesko

26 Hall amp Liebman

63 ResultsThe dependent variable in our regressions is the fraction of a CEOrsquos totalannual compensation (measured as the BlackndashScholes value of stock-option grants plus salary and bonus) that comes from stock-optiongrants We choose this reduced-form specication because our interest isin the composition of CEO compensation However the factors thataffect the composition of compensation might also affect the levels andit will be important to keep this in mind in interpreting our results

Optimal-contracting theory generally makes predictions about the in-centives provided by the total amount of rm stock and stock optionsowned by the CEO We choose to model the ow of option grants ratherthan the ultimate performance sensitivity of the CEOrsquos compensationand stock and stock-option holdings because we have in mind anadjustment-cost model in which it takes time for a CEOrsquos holdings ofstock and stock options to reach the optimal level11 In our regressionswe include two measures of the current performance sensitivity of theCEOrsquos wealth to rm performance as explanatory variables to reect thedistance of the CEOrsquos contract from the optimal contract

Table 3 presents results in which the tax effects are identied solely byvariation in tax rates over time The fraction of annual compensation paidin stock options is regressed on tax variables corporate governance vari-ables contract-theory variables and variables reecting stock market re-turns Both of the regressions in this table include rm xed effects Incolumn (1) the fraction in annual compensation paid in stock options isregressed on the log difference in payoff (the combined payoff to both therm and the CEO) from receiving compensation in stock options ratherthan salary and bonus where the payoff calculations are based on statu-tory marginal tax rates This variable is dened as the log difference be-tween equation (3) and equation (6) in section 4 The coefcient on the taxvariable is large and statistically signicant at the 95-percent level Thepoint estimate implies that a one-percent increase in the payoff differencebetween stock options and salary and bonus results in a 24-percentage-point increase in the share of compensation paid in stock options

The corporate governance variables all have the predicted sign Firmswith a higher fraction of their shares owned by institutional investorsuse more performance-based pay (stock options) Firms with large corpo-rate boards or a large share of inside directors are less likely to use stockoptions

The two contract-theory variables are the dollar change in CEO wealth

11 See Core and Guay (1999) for evidence that rms adjust in this way

TABLE 3Compensation Regressions that Rely on Variation in

Tax Rates over Time

(1) (2)

Tax variablesln (difference in payoff)(a) 2393 (0349)ln (12statutory personal 20015 (0034)

rate)ln (12statutory cap gains 0330 (0171)

rate)ln (12statutory corporate 0399 (0148)

rate)

Corporate governance variablesFraction of rmrsquos shares 0043 (029) 0052 (0028)

owned by institutionalinvestors

Size of the rmrsquos board of 20002 (001) 20003 (0001)directors

Fraction of directors who 20082 (033) 20092 (0030)are inside or gray

Contract-theory variablesD (CEO wealth) per $1000 20261 (0095) 20247 (0087)

in FV(b)

D (CEO wealth) per 10 in 2583 (189) 2554 (201)FV(c)

Volatility(d) 521 (424) 261 (369)ln (rm market value) 0019 (0006) 0019 (0006)

Stock market return variablesLagged annual rm stock 0006 (0009) 0003 (0008)

market returnLagged annual SampP 500 re- 20035 (0023) 20029 (0023)

turn

Regression diagnosticsTotal sample size 5179 5179Firms 427 427Adjusted R2 0313 0311

Dependent variable share of compensation paid in options Standard errors in parentheses Bothregressions in this table include rm xed effects

(a) Log difference in payoff is the log difference in combined rm and CEO after-tax share betweenpaying compensation in options and paying salary

(b) This variable is the dollar change in CEO wealth per $1000 change in rm market value (seeJensen and Murphy 1990)

(c) This variable in the dollar charge in CEO wealth per 10-percent change in rm value (see Hall andLiebman 1998)

(d) Standard deviation of rm market value in trillions

28 Hall amp Liebman

per $1000 change in rm value (Jensen and Murphy 1990) and thedollar change in CEO wealth per 10-percent change in rm value (Halland Liebman 1998) As discussed in Baker and Hall (1998) these twowealth sensitivity measures reect two different concepts of how closelyaligned the managersrsquo incentives are with the interests of the sharehold-ers In the regression they are meant to measure how far the CEOrsquosexisting contract and ownership of stock and stock options is from theoptimal level The negative coefcients on the two variables suggeststhat rms that are below their optimal pay-to-performance sensitivityare more likely to give more stock options in the current period

The second column replaces the tax variable motivated by deferraladvantages with the three statutory rates The corporate after-tax sharehas a positive coefcient as we would expect because the tax advantageof stock options increases as the corporate rate falls The coefcient onthe personal tax rate is small and statistically insignicant The sign ofthe coefcient on the capital gains rate is the opposite of that predictedby tax-avoidance theory

Note that these results rely on time-series variation alone This is not avery convincing test Since the log difference in payoffs mostly rises overthe time period covered in our sample as does the share of compensa-tion paid in options it is possible that the tax variable is simply reectingother factors that were changing over this time period that we have notincluded in our regression Therefore it is important to look at specica-tions that rely on cross-sectional variation in the tax variables

Table 4 contains specications that rely on cross-sectional variation incorporate tax rates The rst column repeats the specication from col-umn (1) of Table 3 but uses the trichotomous measures of rm corporatetax rates in place of the statutory rate in calculating each rmrsquos logdifference in payoff from paying compensation in options rather than insalary and bonus As in the previous table we nd the positive coef-cient on the log difference in payoff just as the theory suggest How-ever the coefcient estimate is now much smaller suggesting that aone-percent increase in the log difference in payoff produces a 01-percentage-point increase in the share of compensation paid in options

Introducing cross-sectional variation in the dependent variable doesnot eliminate the potentially spurious correlation over time between thetax variable and the share of compensation paid in options The secondcolumn of Table 4 isolates the pooled cross-sectional variation in corpo-rate tax rates by adding time dummies and dropping the rm xedeffects In this specication the coefcient on the tax variable is indistin-guishable from zero The coefcient on the institutional variable contin-ues to be large and statistically signicant The third column includes

Taxation of Executive Compensation 29

both rm xed effects and time dummies which tests for a relationshipbetween within-rm changes in the dependent variable and within-rmchanges in each explanatory variable In this specication the coefcienton the tax variable is again indistinguishable from zero The coefcientson the corporate governance variables are also statistically insignicant

The third column contains a specication that treats the Tax ReformAct of 1986 (TRA 86) as a natural experiment For rms and CEOs facingthe statutory corporate tax rate TRA 86 increased the relative payoff topaying compensation in options by 17 percent For a rm facing a zeromarginal tax rate it reduced the relative payoff by 45 percent The reduc-tion occurred because TRA 86 reduced the personal tax rate makingsalary and bonus more attractive while the decreased corporate rate hadno effect on zero-tax-rate rms Thus we would expect rms facing zeroor low marginal tax rates to reduce their use of options after TRA 86while rms facing the statutory rate would be expected to increase themFew rms face zero marginal tax rates year after year Therefore we tryto distinguish between rms that often face low tax rates and rms thatnearly always face the statutory rate We dene low-tax rms as oneswhose average marginal tax rate over 1984 to 1986 was below 02412

Then we ran a difference-in-differences regression to see if high-taxrms increased their use of options more than low-tax rms The interac-tion between post-TRA 86 and high corporate tax rates is the key vari-able The estimated coefcient on this variable is positive but small andnot statistically different from zero at the 95-percent level

It is important to note that in this specication as well the main corpo-rate governance variable the share of the rmrsquos shares owned by institu-tional investors performs as theory would predict The variable appearsto be quite robust Table 5 presents regressions of the share of compensa-tion paid in stock options on the fraction of the rmrsquos shares owned byinstitutional investors Since we have already established that these vari-ables have a strong longitudinal relationship the regressions here aredesigned to focus on the cross-sectional relationship between these twovariables The rst and third columns present regressions using data onlyfrom 1994 and nd a coefcient of roughly 02 (column 1 contains noother covariates while column 3 includes all of the other nontax covari-ates from the regressions in Tables 3 and 4) Since the median share of armrsquos stock owned by institutional investors in our sample rises from 19percent in 1980 to 49 percent in 1994 this coefcient would imply that this

12 This cutoff was chosen after inspecting the distribution of tax rates and essentiallyseparates rms that are always at or near the statutory rate from ones that are sometimesbelow it

TA

BL

E4

Com

pens

atio

nR

egre

ssio

nsth

atR

ely

onC

ross

-Sec

tion

alV

aria

tion

inC

orpo

rate

Tax

Rat

es

(1)

(2)

(3)

(4)

Tax

vari

able

s(a)

ln

(dif

fere

nce

inp

ayof

f)0

099

(00

41)

20

038

(00

36)

20

005

(00

44)

Pos

t-T

RA

862

000

5(0

013

)H

igh

corp

orat

eta

xra

te0

018

(00

17)

(Pos

t-T

RA

86)

3hi

gh0

024

(00

16)

corp

orat

eta

xra

te

Cor

por

ate

gove

rnan

ceva

riab

les

Frac

tion

of

rmrsquos

shar

es0

109

(00

27)

011

4(0

015

)2

000

9(0

030

)0

087

(00

34)

owne

dby

inst

itu

tion

alin

vest

ors

Size

ofth

e

rmrsquos

boar

dof

20

004

(00

01)

20

001

(00

01)

20

002

(00

01)

000

1(0

002

)d

irec

tors

Frac

tion

ofd

irec

tors

wh

o2

012

4(0

033

)2

000

3(0

018

)2

004

1(0

034

)0

001

(00

39)

are

insi

de

orgr

ay

Con

trac

t-th

eory

vari

able

sD

(CE

Ow

ealt

h)p

er$1

000

20

238

(00

94)

20

167

(00

53)

20

257

(00

96)

20

021

(01

47)

inFV

(b)

D(C

EO

wea

lth)

per

10

in2

560

(17

8)2

592

(10

5)2

594

(19

6)2

147

(04

86)

FV(c

)

Vol

atili

ty(d

)2

48(4

24)

162

(03

4)1

12(4

28)

106

(06

09)

ln(

rmm

arke

tva

lue)

003

3(

006)

000

8(0

003

)0

009

(00

06)

001

2(0

007

)

Stoc

km

arke

tre

turn

vari

able

sL

agge

dan

nual

rm

stoc

k2

000

2(0

009

)0

011

(00

10)

000

6(0

009

)2

001

2(0

013

)m

arke

tre

turn

Lag

ged

annu

alSamp

P50

02

005

6(0

023

)2

005

3(0

051

)2

003

9(0

049

)2

015

4(0

040

)re

turn

Reg

ress

ion

dia

gnos

tics

T

otal

sam

ple

size

5179

5179

5179

2534

Firm

s42

742

739

6A

dju

sted

R2

031

009

032

005

Firm

xed

effe

cts

Yes

No

Yes

No

Tim

ed

um

mie

sN

oY

esY

esN

o

Dep

end

ent

vari

able

isfr

acti

onof

ann

ualc

omp

ensa

tion

from

stoc

kop

tion

sSt

and

ard

erro

rsin

par

enth

eses

(a)

All

thre

esp

eci

cati

ons

use

tric

hoto

mou

sco

rpor

ate

tax

rate

sT

he

sam

ple

inco

lum

n(3

)u

ses

only

the

year

s19

84to

1988

sin

cew

ear

eu

sing

1986

asa

nat

ural

exp

erim

ent

(b)

Thi

sva

riab

leis

the

dol

lar

chan

gein

CE

Ow

ealt

hp

er$1

000

chan

gein

rm

mar

ketv

alu

e(s

eeJe

nse

nan

dM

urp

hy

1990

)

(c)

Thi

sva

riab

lein

the

dol

lar

char

gein

CE

Ow

ealt

hp

er10

-per

cent

chan

gein

rm

valu

e(s

eeH

alla

nd

Lie

bman

199

8)

(d)

Stan

dar

dd

evia

tion

ofr

mm

arke

tva

lue

intr

illio

ns

32 Hall amp Liebman

TABLE 5Large Institutional Investors and Stock-Option Compensation

The Cross-sectional Relationship between Institutional Investors andOption Grants

(1) (2) (3) (4)

Coefcient on fraction of 217 (067) 145 (014) 177 (069) 120 (015)rmrsquos shares owned byinstitutional investors

Sample period 1994 1980ndash1994 1994 1980ndash1994Contract-theory and stock No No Yes Yes

market covariatesTime dummies No Yes No Yes

The dependent variable is the share of compensation paid in options

factor alone can explain 6 percentage points of the 23-percentage-pointincrease in the median share of compensation paid in options that oc-curred over this time period Of course this variable is only one measureof the strengthening of corporate governance so the total impact of im-proved corporate governance could be greater The second and fourthcolumns of the table use the entire 1980ndash1994 sample and include timedummies to isolate the cross-sectional correlations Again there is evi-dence of a strong cross-sectional relationship between the two variablesalthough the coefcients are slightly smaller perhaps reecting a weakercorporate-governance role in the earlier period In sum there is a strongrelationship in both longitudinal and cross-sectional data between theinstitutional-investor variable and the options-percentage variable13

7 THE MILLION-DOLLAR RULE

In 1993 section 162(m) of the Internal Revenue Code was enactedlimiting the deductibility of executive compensation in excess of onemillion dollars unless the compensation was performance-related Pro-ponents of this legislation argued that it would slow the growth ofexecutive pay and tighten the link between rm performance and execu-tive compensation

The new provision became effective for tax year 1994 and applies tothe CEO and the other four most highly compensated executives in eachcompany Pay that is performance-related such as stock-option grants

13 The one specication in which the fraction owned by institutional investors is notsignicant is in column (2) of Table 4 a regression that includes both xed effects and timeeffects

Taxation of Executive Compensation 33

and bonuses paid for meeting clear performance goals is not affected bythis regulation provided that it has been approved by shareholdersThus the main impact of the provision is to limit the deductibility ofsalary in excess of a million dollars Woodlock and Antenucci (1997)studied the proxy statements of 376 rms and documented that mostrms responded to the new law by qualifying top executive pay asperformance-related Perry and Zenner (1999) identied 25 rms thatreduced salaries from above $1 million to below $1 million and foundthat 23 of the rms cited section 162(m) as their reason for doing so

In this section of the paper we investigate whether this provision hasaffected the level or structure of executive pay using the Execucompdata base which contains information on the highest-paid executives ineach of the rms in the SampP 500 SampP Midcap 400 and SampP SmallCap600 Between 1993 and 1998 the median salary of the Execucomp execu-tives rose from $230000 to $306000 The median value of total compen-sation (salary bonus and option grants) rose from $470000 to $882000Limiting the sample of CEOs indicates that the total compensation of themedian CEO rose from $11 million to $18 million over the period whilethe median salary rose from $441000 to $550000 Thus executive com-pensation clearly continued to rise at a rapid rate after the implementa-tion of the million-dollar rule though perhaps at a slower rate than itotherwise would have

If the regulation affects behavior it should have the largest impact onrms whose executives are receiving salaries that are above the limit orjust below it Thus we might expect to see rms whose executives arereceiving relatively high salaries to rely more heavily on performance-based pay and less heavily on salary in the years following the regulationPerry and Zenner (1999) and Livingston (1999) both nd that this is thecase This trend can be clearly identied in the Execucomp data set Forexample over 1993 to 1998 the median annualized growth rate of salaryfor executives whose 1993 salary was below $500000 in the previous yearwas 85 percent For executives with salary between $500000 and$700000 it was 64 percent for executives between $700000 and $850000it was 50 percent for executives between $750000 and $1 million it was31 percent and for executives above $1 million it was 00 percent14

The observation that executives at lower levels of salary had highergrowth rates is not very strong evidence of an impact of section 162(m)however because it is possible that low-salary executives would havehad higher growth rates even in the absence of the provision To explore

14 Of the 3137 executives for whom we have complete data between 1993 and 1998 2708had 1993 salaries below $500000 while only 35 had 1993 salaries above $1 million

34 Hall amp Liebman

whether this is the case we use Execucomp data from both before andafter the law change Ideally we would want to have data from manyyears before the change to examine whether there is an underlying pat-tern of faster salary growth for low-salary executives Unfortunately theExecucomp data begin in 1992 and therefore provide us with only the1992ndash1993 growth rate in compensation to use as a baseline15

We run regressions pooling data from 1992ndash1993 through 1997ndash1998of the annual percentage change in different forms of compensation on avariable million that is designed to measure the likelihood that anexecutiversquos compensation will be affected by section 162(m) We denemillion as the minimum of 1 and the executiversquos salary divided by $1million in the previous year Thus any executive with salary at or above$1 million would receive a value of 1 while an executive with a salary of$500000 would be coded as a 0516 We include this variable by itself andalso interacted with a dummy variable (after) that equals 1 for years1994 and after Thus the million variable is designed to pick up anyunderlying relationship between the level of salary and the growth rateof salary while the interacted one will identify any differential growthrate of salary by income in the period after section 162(m) took effect Wealso include a full set of year dummies the natural logarithm of thermrsquos market value and the rmrsquos annual rate of return over the twoprevious years as control variables To reduce the sensitivity of our re-sults to outliers we run quantile regressions and robust regressions

Table 6 contains our results The rst four columns are for all topexecutives while the last four columns focus on CEOs Columns (1) and(2) present estimates from robust regression for the percentage change insalary and the percentage change in total compensation (salary bonusstock grants and stock-option grants) In the salary growth regressionsfor all executives the coefcient on million is negative and signicantindicating that there is an underlying relationship between the level ofexecutive compensation and its subsequent growth rate The coefcienton million 3 after is negative as well suggesting that the negativerelationship between the level of salary and its subsequent growth rateintensied slightly after 1994 The relatively small magnitude suggeststhat an executive with a $1 million dollar salary would see his salarygrow at an annual rate that would be about 06 percent slower than anexecutive earning $500000

15 The HallndashLiebman data set contains only a combined salary and bonus variable makingit impossible to use it to analyze this issue16 We also ran the regressions (not reported) allowing for a nonlinear relationship betweenmillion and compensation growth and obtained substantively similar results

TA

BL

E6

The

Eff

ect

ofth

eM

illi

on-D

olla

rR

ule

onSa

lary

and

Tot

alC

ompe

nsat

ion

All

top

exec

uti

ves

CE

Os

only

Rob

ust

regr

essi

onQ

uan

tile

regr

essi

onR

obu

stre

gres

sion

Qu

anti

lere

gres

sion

D

in

Din

D

in

Din

D

in

Din

D

in

Din

sala

ryto

talc

omp

(a)

sala

ryto

talc

omp

sa

lary

tota

lcom

p

sala

ry

tota

lcom

p

mil

lio

n(b

)2

006

02

035

72

006

32

024

12

008

72

023

12

010

32

024

5(0

005

)(0

032

)(0

006

)(0

027

)(0

010

)(0

073

)(0

011

)(0

056

)m

illi

on

32

001

10

026

20

021

20

015

20

016

014

92

003

40

158

aft

er(c

)(0

005

)(0

033

)(0

006

)(0

029

)(0

010

)(0

076

)(0

011

)(0

057

)N

359

5035

950

359

5035

950

757

17

589

757

17

589

Stan

dar

der

rors

inp

aren

thes

esR

egre

ssio

nsal

soin

clu

de

year

du

mm

ies

stoc

km

arke

tre

turn

and

log

mar

ket

valu

e

(a)

Tot

alco

mp

ensa

tion

incl

ud

essa

lary

bon

us

rest

rict

edst

ock

and

stoc

k-op

tion

gran

ts

(b)

mil

lion

isd

ene

das

min

(1s

alar

yin

pre

viou

sye

ars

div

ided

by$1

mil

lion)

(c)

aft

eris

anin

dic

ator

vari

able

equ

alto

zero

befo

reth

ere

gula

tion

took

plac

e(1

993)

and

one

afte

rwar

d

36 Hall amp Liebman

In the second column the coefcient on million 3 after is essentiallyzero suggesting that any decrease in salary brought about by section162(m) was offset by increases in bonus and stock-option grants Theresults are similar for the median regressions There is some evidence ofa very small slowdown in growth of salary for executives likely to beaffected by section 162(m) but no sign of a change in total compensa-tion Taken together this provides evidence of a very small substitutionof performance-related pay for salary

The last four columns run identical tests but limit the sample to CEOsThe interaction of million 3 after continues to be small and negative inthe salary regressions In contrast the impact of 162(m) on total compen-sation appears to be positive and fairly substantial (about 15 percentagepoints per year) in these regressions That is the increase in bonus andstock options following the million-dollar rule more than offset the verysmall decline in salary

All of these conclusions should be interpreted somewhat cautiouslybecause our data provide us with only one pre-1993 control year Never-theless the data are consistent with section 162(m) having led to a veryminor slowdown in salary growth and one that is dramatically smallerthan what one would conclude without controlling for underlying differ-ential salary growth rates at different levels of salary Since total compen-sation for CEOs did not decline and perhaps increased there is evidenceof a minor substitution of performance-related pay for salary in responseto the regulation

8 EXECUTIVE COMPENSATION AND THEELASTICITY OF TAXABLE INCOME

Even if taxes have only a minor effect on executive compensation thestructure of executive compensation is important for tax policy be-cause high-income taxpayers are sources of signicant revenue and arethe focus of important debates about the efciency cost of taxation

Goolsbee (1999) has recently argued that executive decisions to exer-cise options are highly responsive to intertemporal differences in taxrates created by tax reforms He claims that much of the apparent declinein taxable income for high-income taxpayers between 1992 and 1993documented by Feldstein and Feenberg (1996) can be explained by execu-tives shifting option gains into 1992 in order to avoid the higher marginaltax rates that went into effect in 1993 By extension he implies that thehigh elasticities of taxable income with respect to personal tax ratesestimated by comparing the pre- and post-tax-reform taxable income of

Taxation of Executive Compensation 37

high-income individuals reect intertemporal shifting of income and notpermanent effects of taxation If this is true then the deadweight loss oftaxation is lower as is the cost of progressivity

While Goolsbee presents comprehensive evidence that option gainswere unusually high in 1992 his identication of the impact of tax rates onthe timing of option gains depends on correctly specifying a counter-factual level of options that would have been exercised in the absence ofthe tax change He simply assumes a linear time trend The purpose ofGoolsbeersquos time trend is presumably to account for the increasing use ofoptions over time However since options typically require a vestingperiod of at least a couple of years a much more direct measure of ex-pected option gains can be constructed based on the total value of vestedoptions held by the CEO Since options typically have vesting periods thatexceed two years this total can be treated as exogenous since it will not beaffected by the tax changes

More generally the value of options held by the CEO is likely to havea major effect on his exercise decisions If his options are out of themoney he will not have gains to realize If there has been a recent run-up in the market an optimizing CEO may exercise an unusually largeamount of options in order to diversify his overall portfolio and if in-the-money options are about to expire they will be exercised in thecurrent year

Our evidence suggests that much of the apparent tax shifting caninstead be attributed to stock market performance and the timing of pastoption grants Using our 15-year panel we replicate Goolsbeersquos (1999)result for the 1993 period but show that the impact of taxes is notpresent in other time periods Moreover even the result for the 1993period disappears when appropriate controls for past option grants andstock price appreciation are included

The top three rows of Table 7 present data on the taxable incomes ofCEOs in the HallndashLiebman sample For these results our panel has beennarrowed to include only CEOs whose rmrsquos scal years correspondwith tax years In addition we create a balanced panel for each of thethree tax-reform periods by excluding any rm for which there are notcomplete data for the period surrounding the reform

The fourth row of the table shows that the 1993 tax act raised the topmarginal personal income tax rate from 310 to 396 percent effective fortax year 1993 with a further increase occurring in 1994 due to the uncap-ping of the Medicare payroll tax The top row of the table shows thattaxable income was higher in 1992 than for any other year in our sample(all data are in 1994 dollars) Option gains were particularly high in1992mdashmore than double their 1991 level and 68 percent higher than their

TA

BL

E7

Tax

Ref

orm

Sto

ckM

arke

tP

erfo

rman

cea

ndth

eT

imin

gof

Opt

ion

Gai

ns

Ave

rage

stoc

km

arke

tM

ean

CE

OM

ean

CE

Ore

turn

for

Mea

nva

lue

taxa

ble

sala

ryan

dM

ean

CE

OT

opfe

der

al

rms

inof

opti

ons

(Val

ue

ofop

tion

Sam

ple

inco

me

bonu

sop

tion

gain

sp

erso

nal

sam

ple

for

held

atbe

gga

ins)

(va

lue

Ref

orm

size

Yea

r($

mil

)($

mil

)($

mil

)M

TR

pre

vye

ar(

)of

year

ofop

tion

she

ld)

ER

TA

8127

919

800

874

064

80

163

7025

80

460

354

1981

083

10

672

009

869

245

070

61

3919

820

872

067

90

099

5011

90

686

144

1983

099

00

724

025

250

355

116

62

16

TR

A86

295

1985

119

70

837

024

050

76

116

02

0719

861

448

091

70

345

5042

81

776

194

1987

165

50

959

059

338

519

12

229

266

1988

163

21

058

039

028

07

237

71

6419

891

656

105

20

414

2817

92

345

176

1993

272

1990

209

51

028

061

828

232

288

02

15T

axA

ct19

911

875

102

00

547

312

01

268

82

0319

922

678

107

01

250

3146

64

552

275

1993

232

41

209

074

339

620

64

299

173

1994

196

41

262

037

842

515

45

113

074

Sam

ple

isH

allndash

Lie

bman

rm

sw

ith

Dec

embe

r

scal

year

sth

atha

dco

mp

lete

dat

afo

ron

eof

the

thre

eta

x-re

form

per

iod

sst

ud

ied

Taxation of Executive Compensation 39

1994 level These patterns are consistent with Goolsbeersquos claim that tax-payers responded to the anticipated increase in marginal tax rates byshifting gains into 1992

The evidence from the other two tax reforms is less clear suggestingthat there might be factors other than tax rates determining the timing ofoption gains ERTA 1981 reduced the top marginal tax rate from 70percent in 1980 to 50 percent in 1982 However there is no sign thatoption gains were reduced in 1980 and 1981 and then increased in 198217

There was a big increase in option gains in 1983 however Similarly thehighest year of stock-option gains in the period around 1986 was 1987 astrange year to take gains since taxpayers would have known that thepersonal rate would be 10 percentage points lower if they waited untilthe following year18

What all the peak years of option gains have in common is that theyfollow years of strong stock market performance The fth row of Table 7shows the average stock market return during the previous year for therms in our sample The large value of option gains in 1983 followed a355-percent increase in the stock market The 1992 boom in option gainsfollowed a 466-percent increase in the stock market the more modest1987 peak in option gains followed a 19-percent gain (and a 43-percentgain two years before)

Strong stock market performance raises the value of options availableto exercise The sixth row of the table shows the mean value of optionsheld by CEOs at the beginning of the year All three of the peak years ofoption gains correspond with peak years in terms of the value of optionsavailable for exercise Since stock market appreciation during the yearalso increases the value of options available to exercise later in that yearwe need to allow for current-year stock market appreciation as well

Table 8 contains Goolsbee-style regressions for each of the tax re-forms The rst two columns reprint the results from Goolsbee (1999)The third and fourth columns replicate his results in our sample Inparticular with a specication similar to his we are able to replicate hisnding that the current after-tax share has a positive coefcient and thefollowing yearrsquos after-tax share has a coefcient of roughly negative oneColumns (5) through (8) show that this pattern was unique to the 1993tax reform Indeed in the TRA 86 regressions the coefcient on the

17 As we explained above for executives covered by the maximum tax on earned incomethere was not a decline in tax rates between 1981 and 198218 It is possible that the large stock market decline in October 1997 may have promptedsome people to exercise options earlier than they otherwise would have However anincrease in option exercises typically follows stock market increases not declines

TA

BL

E8

Att

empt

sto

Rep

lica

teth

eT

ax-S

hift

ing

Res

ult

inO

ther

Tim

eP

erio

ds

Goo

lsbe

eG

ools

bee

1993

1993

Ent

ire

Ent

ire

regr

essi

onre

gres

sion

refo

rmre

form

TR

A86

TR

A86

ER

TA

81E

RT

A81

1980

ndash199

419

80ndash1

994

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

ln[1

2ta

x(t)

]1

244

108

20

455

032

62

017

62

017

22

016

72

016

70

026

20

068

(01

34)

(01

27)

(05

37)

(05

42)

(02

10)

(02

09)

(00

90)

(00

91)

(00

67)

(00

65)

ln[1

2ta

x(t1

1)]

20

867

20

764

21

214

20

917

047

80

722

20

103

004

30

176

018

3(0

118

)(0

112

)(0

491

)(0

499

)(0

157

)(0

174

)(0

072

)(0

072

)(0

068

)(0

066

)

Tim

e0

135

007

62

000

52

036

40

041

000

80

094

003

50

056

002

7(0

008

)(0

008

)(0

049

)(0

049

)(0

028

)(0

028

)(0

025

)(0

026

)(0

003

)(0

003

)

ln(m

arke

tva

lue)

066

10

489

023

70

334

039

5(0

015

)(0

058

)(0

039

)(0

042

)(0

017

)

Yea

rs19

91ndash1

995

1991

ndash199

519

90ndash1

994

1990

ndash199

419

85ndash1

989

1985

ndash198

919

80ndash1

983

1980

ndash198

319

80ndash1

994

1980

ndash199

4

Ad

just

edR

20

720

770

610

630

690

760

800

810

630

68

Obs

erva

tion

s13

856

132

371

385

136

41

470

143

31

113

106

23

968

385

9

Dep

end

ent

vari

able

isth

elo

gof

taxa

ble

inco

me

All

regr

essi

ons

con

tain

rm

xe

def

fect

sG

ools

bee

regr

essi

on(2

)al

soco

ntai

nstw

oot

her

vari

able

sno

tsh

own

inth

eta

ble

Stan

dar

der

rors

inp

aren

thes

es

Taxation of Executive Compensation 41

TABLE 9Alternative Explanations for the Timing of Option Exercises

1993 Entirereform TRA 86 ERTA 81 1980ndash1994

(1) (2) (3) (4)

ln[12tax(t)] 0108 0111 20051 0201(0324) (0136) (0048) (0057)

ln[12tax(t11)] 20134 0573 0125 0150(0336) (0156) (0053) (0066)

ln(market value) 0418 0177 0338 0418(0061) (0039) (0046) (0163)

Firmrsquos stock market return (t) 0298 0268 0214 0258(0044) (0045) (0031) (0023)

Firmrsquos stock market return 0162 0206 0031 0128(t21) (0041) (0008) (0029) (0021)

ln(value of options held by 0004 0008 0001 0009CEO at start of year) (0003) (0003) (0002) (0001)

Years 1990ndash1994 1985ndash1989 1980ndash1983 1980ndash1994Adjusted R2 064 071 082 068Observations 1349 1427 1036 3812

The dependent variable is the log of taxable income All regressions contain rm xed effects Stan-dard errors in parentheses

following yearrsquos tax rate is positive and signicant Columns (9) and (10)pool all three reforms and fail to replicate the pattern that would reectintertemporal shifting19

It is possible that the ease with which one can shift the timing ofoption gains is asymmetric It may be harder to shift option gains for-ward in time since some options expire and therefore must be exercisednow In addition options were a much smaller portion of compensationin the early 1980s so perhaps the ERTA 81 results are not surprisingNonetheless these results cast doubt on the proposition that this is thechannel through which spurious estimates of the elasticity of taxableincome have occurred for previous tax reforms and they suggest thatwe should look harder for alternative explanations for the timing ofoption gains

Table 9 attempts to predict options gains using the lagged and currentincrease in a rmrsquos market value and the value of the options held by theCEO at the start of the year The two stock market variables stronglypredict option gains while the value of options held has the correct sign

19 The results also fail to suggest large permanent effects of marginal tax rates on taxableincome

42 Hall amp Liebman

in all of the specications and is signicant in the pooled sample eventhough it is highly correlated with the two stock market return variablesIn contrast the tax-timing explanation fails even in the 1993 period

It is important to realize that our results do not invalidate Goolsbeersquoscritique of claims that aggregate revenue data from 1992 and 1993 provethat elasticities of taxable income are large We have simply given analternative explanation for why taxable incomes were unusually high in1992 However our results do cast doubt on the conclusion that largetiming shifts are ubiquitous and capable of explaining away the morecareful estimates of taxable income elasticities that have been performedusing micro data (eg Feldstein 1995 Auten and Carroll 1995)

One further point is in order It is sometimes argued that the combina-tion of increased use of options and the booming stock market explainswhy federal tax revenues have been so much higher than expected dur-ing the 1990s However while large option gains increase revenue fromthe personal income tax they produce offsetting deductions for cor-porations The net impact on revenue is only the difference in the twomarginal tax rates times the option gain While the gap between thepersonal rate and the effective corporate rate is nontrivial the impact oflarge option exercises on total revenue is substantially smaller than theamount by which the exercises inate personal tax revenue and reducecorporate tax revenue

9 CONCLUSION

We have described the tax rules for executive pay in detail and analyzedhow changes in various tax rates affect the tax advantages of stock optionsrelative to salary and bonus We nd that there is a moderate tax advan-tage to optionsmdashcurrently about $4 per $100 in pretax executive paymdashand that changes in the tax advantage of options over time have had atmost a modest impact on the composition of pay Corporate governancefactors particularly the role of large investors appear to be more impor-tant in explaining the dramatic increase in option pay

In addition our evidence suggests that more direct attempts to use taxpolicy to inuence executive compensation have had little effect Wend that the million-dollar rule led companies to substitute perfor-mance-based pay for salary But our evidence suggests that this substitu-tion was quite modest and there is no evidence that the total level of paywas reduced Overall although the stock-option explosion has dramati-cally increased the link between pay and performance this change isdue almost entirely to nontax factors

Taxation of Executive Compensation 43

REFERENCES

Auerbach Alan J (1996) ldquoThe Dynamic Effects of Tax Law Asymmetriesrdquo Re-view of Economic Studies 53451ndash405

mdashmdashmdash and James Poterba (1987) ldquoTax Loss Carryforwards and Corporate TaxIncentivesrdquo In The Effects of Taxation on Capital Accumulation M Feldstein(ed) Chicago University of Chicago Press

Altshuler Rosanne and Alan Auerbach (1990) ldquoThe Signicance of Tax LawAsymmetries An Empirical Investigationrdquo Quarterly Journal of Economics 10561ndash86

Auten Gerald and Robert Carroll (1995) ldquoThe Effect of Income Taxes on House-hold Behaviorrdquo Ofce of Tax Analysis US Department of the Treasury

Baker George P and Brian J Hall (1998) ldquoCEO Incentives and Firm SizerdquoNBER Working Paper no 6868 December

Core John and Wayne Guay (1999) ldquoThe Use of Equity Grants to ManageOptimal Equity Incentive Levelsrdquo Draft

Feldstein M (1995) ldquoThe Effect of Marginal Tax Rates on Taxable Income APanel Study of the 1986 Tax Reform Actrdquo Journal of Political Economy 103 (no3)551ndash72

mdashmdashmdash and D Feenberg (1996) ldquoThe Effect of Increased Tax Rates on TaxableIncome and Economic Efciency A Preliminary Analysis of the 1993 Tax RateIncreasesrdquo In Tax Policy and the Economy vol 10 James M Poterba (ed)Cambridge MA MIT Press

Gompers Paul A and Andrew Metrick (1998) ldquoInstitutional Investors andEquity Pricesrdquo NBER Working Paper no 6723

Goolsbee Austan (1999) ldquoWhat Happens When You Tax the Rich Evidencefrom Executive Compensationrdquo NBER Working Paper no 6333

Graham John R (1996) ldquoProxies for the Corporate Marginal Tax Raterdquo Journal ofFinancial Economics 42187ndash221

Hall Brian J (1999) ldquoThe Design of Multi-year Stock Option Plansrdquo Journal ofApplied Corporate Finance 12(2)97ndash106

mdashmdashmdash and Jeffrey B Liebman (1998) ldquoAre CEOs Really Paid Like Bureau-cratsrdquo Quarterly Journal of Economics CXII(no 3)653ndash691

Jensen Michael C and William M Meckling (1976) ldquoTheory of the Firm Mana-gerial Behavior Agency Costs and Ownership Structurerdquo Journal of FinancialEconomics III305ndash360

mdashmdashmdash and Kevin J Murphy (1990) ldquoPerformance Pay and Top-ManagementIncentivesrdquo Journal of Political Economy XCVIII225ndash264

Kaplan Steven N (1997) ldquoThe Evolution of US Corporate Governance WeAre All Henry Kravis Nowrdquo Draft

Livingstone Jane R (1999) ldquoExecutive Compensation Contracting and Re-sponses to Increased Tax Costsrdquo Louisiana State University June

Miller Merton H and Myron S Scholes (1982) ldquoExecutive CompensationTaxes and Incentivesrdquo In Financial Economics Essays in Honor of Paul CootnerWilliam F Sharpe and Cathryn M Cootner (eds) Princeton NJ Prentice-Hall

Murphy Kevin J (1999) ldquoExecutive Compensationrdquo forthcoming in OrleyAshenfelter and David Card (eds) Handbook of Labor Economics Vol 3 NorthHolland

Perry Todd and Marc Zenner (1999) ldquoPay for Performance Government Regu-

typ

ical

lyve

stov

era

3ndashti

veis

taxe

din

the

and

ther

efor

ear

eno

tpr

ice

and

num

ber

ofop

-5-

year

per

iod

The

typ

i-us

ualw

ay(a

tth

eca

pi-

subj

ect

toth

e$1

-ti

ons

iskn

own

atgr

ant

calm

atu

rity

is10

year

sta

lgai

nsra

te)

onan

ym

illio

nca

p

dat

eFo

rd

isco

unt

op-

but

the

man

ager

may

furt

her

app

reci

atio

nof

tion

sth

ed

iffe

renc

ebe

-ex

erci

seea

rly

Som

eth

esh

are

pri

ce

twee

nth

est

ock

pri

ceop

tion

sar

edi

scou

ntop

-an

dth

em

arke

tp

rice

isti

ons

(the

exer

cise

pri

ceex

pen

sed

over

the

vest

-is

low

erth

anth

est

ock

ing

per

iod

For

opti

ons

pri

ceat

the

gran

td

ate)

w

ith

vari

able

term

sP

rem

ium

opti

ons

are

(eg

a

vari

able

exer

-th

ere

vers

e(t

heex

er-

cise

pri

cev

esti

ngth

atci

sep

rice

isgr

eate

ris

tied

top

erfo

rman

ce)

than

the

stoc

kp

rice

atth

eop

tion

sar

em

arke

dgr

ant

dat

e)

tom

arke

tan

dex

-p

ense

dd

uri

ngth

eti

me

betw

een

gran

tan

dex

erci

se

Not

eW

eth

ank

com

pen

sati

onco

nsu

ltan

tsSc

ott

Gre

enbe

rgan

dSc

ott

Ols

en(b

oth

ofT

ower

sP

erri

n)

and

Fred

eric

Coo

k(o

fFr

eder

icW

C

ook

and

Ass

ocia

tes)

for

help

fulc

onve

rsat

ion

sin

und

erst

and

ing

the

tax

and

acco

unt

ing

rule

s

10 Hall amp Liebman

Second the tax treatment of ISOs is different ISOs are completely un-taxed at grant or exercise It is only at sale that the executive is taxed andeven then the executive is taxed at the lower capital gains tax rate Thedisadvantage is that the corporation never gets to take a parallel taxdeduction against corporate prots Thus ISOs become more attractiveas the personal tax rate increases and as the corporate tax rate and thecapital gains tax rates fall

Two other related types of compensation restricted stock and stockappreciation rights are worth brief discussion Both however are farless common than standard stock options Restricted stock is payment inthe form of restricted shares the restriction being that the shares vestover time as with options Unlike options the shares typically havevoting and dividend rights With regard to taxation the executive istaxed at the personal rate on the value of the restricted stock as thevesting restrictions lapse However the executive may choose to betaxed at the grant date in which case all subsequent appreciation istaxed at the capital gains rate The dividends paid to the executive aretaxed at the ordinary rate in the usual way The company generallyreceives a parallel deduction equal to the amount of the executiversquosincome when the executive is taxed Unlike stock options restrictedstock is not generally considered to be performance-based and is there-fore subject to the million-dollar rule (unless the vesting of the restrictedstock is performance-based which is sometimes the case)

Stock appreciation rights (SARs) are rights that replicate the payoffs ofstock options with a cash transfer Thus SARs are simpler than options inthat there is no requirement to buy and resell the stock in order to ldquocashoutrdquo SARs generally have the same tax treatment as NQSOs both to theindividual and to the corporation SARs are relatively rare however be-cause they have disadvantageous accounting treatment (described in thenext section) relative to stock options and their main relative advantagevis-a-vis options has been essentially replicated through broker-assistedcashless option exercisesmdashwhereby a broker makes a ldquonanosecondrdquo loanto the executive (to purchase and resell the stock) when the executivewants to ldquocash inrdquo on option prots The tax and accounting treatment ofISOs restricted stock and SARs is summarized in Table 2

32 The Accounting Treatment of OptionsUnlike cash compensation which is expensed against earnings there isgenerally no expense recognition (at grant exercise or sale) for optionswhether they be NQSOs or ISOs As a result compensation consultantsoften point out that stock options are the only form of compensation thatare free in an accounting sense but still deductible for tax purposes

Taxation of Executive Compensation 11

Stock options do however lead to expense recognition if they are dis-counted (in the money at grant date) or if the exercise price and numberof options are not known at grant date For discount options the differ-ence between the stock price and the market price is expensed over thevesting period For options with variable terms (eg a variable exerciseprice vesting that is tied to performance) the options are marked tomarket and expensed during the time between grant and exercise Practi-tioners claim that the accounting treatment of options plays an impor-tant role in the design of option programs Thus plans that have ldquobadaccountingrdquo but are thought by many to have attractive incentive fea-tures are often not even seriously considered by companies Examples ofsuch potentially attractive plans include indexed options (where a CEOprots only if his rmrsquos share price grows relative to some market orindustry benchmark) and option grants that are explicitly performance-related both of which would lead to expenses against earnings

Unlike options restricted stock and SARs do not generally receivefavorable accounting treatment Restricted stock is generally expensedover the period in which the restrictions lapse (usually the vesting pe-riod) The magnitude of the expense is the difference between the cur-rent stock price and the executiversquos cost (if any) SARs are marked tomarket each period and the difference between the stock price and theexercise price is expensed over the outstanding period of SARs

4 THE TAXATION OF EXECUTIVE PAY CASHVERSUS OPTIONS

Stock-based compensation performs two roles in executive compensa-tion arrangements First it helps align the incentives of the executivewith the interests of the rmrsquos shareholders Second it often enables therm to compensate the CEO in a way that is more advantageous from atax standpoint than paying the executive in salary and bonus

41 Agency Theory and Executive Compensation in the Presenceof TaxationIn standard agency-theory models ( Jensen and Meckling 1976) agencycosts are the result of the separation of ownership and control Managersdo not have the same incentives as the owners The optimal incentivecontract for managers balances the benets of high-powered incentives(linking the fortunes of owners and managers through stock and stockoptions for example) with the costs of loading too much risk on risk-averse managers

TA

BL

E2

Tax

and

Acc

ount

ing

Tre

atm

ent

ofE

xecu

tive

Com

pens

atio

nIn

cent

ive

Stoc

kO

ptio

ns(I

SOs)

Res

tric

ted

Stoc

kan

dSt

ock

App

reci

atio

nR

ight

s(S

AR

s)

Tax

atio

nat

Tax

atio

nat

Acc

ount

ing

Com

pen

sati

onD

escr

ipti

onp

erso

nall

evel

corp

orat

ele

vel

trea

tmen

t

Ince

nti

vest

ock

ISO

sar

eid

enti

calt

oIf

the

usu

alco

ndit

ions

Ifth

eu

sual

cond

itio

nsA

sw

ith

NQ

SOs

ther

eop

tion

sN

QSO

sin

thei

rd

esig

n

are

met

th

ere

isno

tax

are

met

the

reis

gene

r-is

gene

rally

noex

pen

seH

owev

ert

hey

have

anat

gran

tor

exer

cise

At

ally

no

corp

orat

eta

xd

e-re

cogn

itio

nfo

rIS

Os

annu

alca

pof

$100

000

sale

all

opti

onp

ro

tsd

uct

ion

atgr

ant

exer

-(e

ver)

on

the

amou

ntth

atan

dsu

bseq

uen

tca

pita

lci

seo

rsa

le

vest

sT

he

tax

trea

t-ap

prec

iati

onar

eta

xed

men

tof

ISO

sal

soat

the

cap

ital

gain

sd

iffe

rs

rate

Res

tric

ted

stoc

kSi

mila

rto

stoc

kop

-T

hein

div

idu

alis

taxe

dT

heco

mp

any

gene

rall

yR

estr

icte

dst

ock

isge

n-ti

ons

but

shar

esat

the

ord

inar

yra

teas

rece

ives

ap

aral

leld

e-er

ally

exp

ense

dov

er(w

hich

vest

slow

ly)

the

rest

rict

ions

lap

se

du

ctio

neq

ual

toth

eth

eve

stin

gp

erio

dT

hera

ther

than

opti

ons

are

How

ever

the

man

ager

amou

ntof

the

exec

u-

amou

nt

exp

ense

dis

gran

ted

toth

eex

ecu

-m

aych

ose

tobe

taxe

dti

versquos

inco

me

whe

nth

eth

ed

iffe

renc

ebe

twee

nti

veU

nlik

eop

tion

sat

gran

td

ate

inw

hich

exec

uti

veis

taxe

dR

e-th

ecu

rren

tst

ock

pric

eth

esh

ares

typi

cally

case

alls

ubs

equ

ent

ap-

stri

cted

stoc

kis

gen

er-

and

the

exec

uti

versquos

have

voti

ngan

dd

ivi-

pre

ciat

ion

ista

xed

atal

lysu

bjec

tto

the

cost

(if

any)

d

end

righ

ts

the

cap

ital

gain

sra

te

mill

ion-

dol

lar

cap

D

ivid

end

sar

eta

xed

atth

eor

din

ary

rate

Stoc

kap

pre

ciat

ion

Rig

hts

that

repl

icat

eG

ener

ally

have

the

Gen

eral

lyha

veth

eU

nlik

eop

tion

sSA

Rs

righ

tsth

epr

ots

ofst

ock

op-

sam

eta

xtr

eatm

ent

assa

me

tax

trea

tmen

tas

are

mar

ked

tom

arke

tti

ons

Rar

ely

use

d

NQ

SOs

NQ

SOs

each

per

iod

an

dth

esi

nce

stoc

kop

tion

sd

iffe

renc

ebe

twee

nth

eha

vem

ore

favo

rabl

eac

-st

ock

pri

cean

dth

eex

-co

unt

ing

trea

tmen

ter

cise

pri

ceis

exp

ense

d(a

ndop

tion

sp

rovi

de

over

the

outs

tand

ing

the

sam

ebe

ne

tof

per

iod

ofSA

Rs

cash

less

exer

cise

thro

ugh

brok

erlo

ans

)

Not

eW

eth

ank

com

pen

sati

onco

nsul

tant

sSc

ott

Gre

enbe

rgan

dSc

ott

Ols

en(b

oth

ofT

ower

sP

erri

n)an

dFr

eder

icC

ook

(of

Fred

eric

WC

ook

and

Ass

ocia

tes)

for

hel

pfu

lcon

vers

atio

nsin

und

erst

and

ing

the

tax

and

acco

unt

ing

rule

s

14 Hall amp Liebman

The effect of taxation on the optimal contract (more precisely theshare of compensation that is performance-related) is ambiguous (evenin the absence of deductibility and deferral) because there are offsettingeffects First by reducing the share of corporate prots received byshareholders taxes diminish the importance to the shareholders of moti-vating the CEO Taken alone this effect would imply that taxes wouldbe expected to reduce the use of performance-based compensation Sec-ond because the government shares in the income received by the CEOthe variance of the CEOrsquos income is reduced raising his utility andlowering the cost to the rm of providing any given set of incentivesThird because taxes will lead executives to provide less effort for anygiven level of incentive-based pay the level of compensation that mustbe provided to compensate the CEO for effort is reduced for a givenamount of incentive These last two factors make it cheaper for the rmto offer contracts to the CEO with large incentive components when taxrates are higher and should therefore increase the use of incentive-based pay Since the net effect of taxation on the level of incentive-basedpay is ambiguous agency theory provides no strong predictions abouthow taxation should affect the optimal composition of option versuscash compensation

42 The Tax Advantages of DeferralWhile agency theory yields ambiguous predictions about the effect oftaxes on the use of options there are direct tax advantages of options inthat options provide executives with a way to defer compensation andthereby lower their taxes However because options also lead to a defer-ral of corporate tax deductibility the tax advantages from a combined(executive and corporate) perspective are not so straightforward In thissection we analyze and measure the tax advantages of stock optionsrelative to salary and bonus compensation We also analyze the taxadvantages of NQSOs relative to ISOs In particular we show how thetax advantages of options change as personal corporate and capitalgains taxes change We then show how the tax advantages of optionshave changed over time in response to changing tax rates

The crucial tax difference between standard options (NQSOs) andcash payment is that option payouts are deferred and the two forms ofcompensation earn different rates of return over the deferral period Anyanalysis of the relative tax advantages of two compensation instrumentsmust consider the tax consequences both to the employer and to theemployeemdashwhat Scholes and Wolfson (1992) call the global contractingperspective Thus in order to make valid comparisons between the twowe compare the tax burden to the executive while holding constant the

Taxation of Executive Compensation 15

posttax cost to the company By keeping posttax employer costs (in NPVterms) constant any package that is preferred by the employee is tax-advantaged in the global-contracting sense

Under this methodology a comparison of the tax advantages of op-tions and cash involves comparing a pretax cash payment of P with anoption payment that has the equivalent posttax NPV to the companyFor NQSOs it is straightforward to show that a pretax payment of P toan executive has exactly the same cost to the company as putting aside Pfor the purposes of paying stock-option payouts later That is if a com-pany pays an executive P today it will have the same amount of moneyin N years as if it had put aside P today let it accumulate at the rate ofreturn earned by the rm and then paid it out (as compensation in theform of option gains4) with any appreciation in year N

Assume that the pretax prot rate is r and the corporate tax rate is TcThen if the company pays P today it receives a deduction today of PTcwhich yields

PTc [1 1 r (1 2 Tc)]N (1)

in N years since the benets of the tax deduction are invested in thecompany and receive the after-tax corporate rate of return Converselyif the company puts P aside today then it grows at the after-tax corpo-rate rate of return in N years to give a tax deduction of

P [1 1 r (1 2 Tc)]N Tc (2)

which is the exact same value Note that the equivalence of these two taxdeductions is analogous to the tax-benet equivalence of front-loadedand back-loaded IRAs

An executiversquos payoff in N years from option prots is equal to Pcompounded at the after-tax corporate rate times 1 2 Tp where Tp is thepersonal tax rate Thus the combined payoffs in N years of the corporatededuction and the executiversquos posttax payoff is

P [1 1 r (1 2 Tc)]N Tc 1 P [1 1 r (1 2 Tc)]

N (1 2 Tp) (3)

where the rst term is the payoff from the corporate deduction and thesecond term is the executiversquos payoff

4 This analysis is not unique to options Any form of deferred compensation that enablesexecutives to invest inside the rm at a higher posttax return or without paying capitalgains taxes would have similar effects

16 Hall amp Liebman

The payoff to salary and bonus in N years is more complicated sincean assumption must be made about how the executive invests the origi-nal cash compensation For example if the executive invests in an instru-ment with taxable interest (eg bonds) then the N-year payoff from aninvestment of P accumulates at the post-corporate-tax post-personal-taxrate of return and is

P (1 2 Tp) [1 1 r (1 2 Tc)(1 2 Tp)]N (4)

However if an executive invests in (non-dividend-paying) equities theinvestment accumulates at the higher post-corporate-tax rate of returnwith the offsetting disadvantage that the capital gains are taxed at thecapital gains rate in year N So the executive receives

P [1 1 r (1 2 Tc)]N (1 2 Tp) 2 Tcg $P (1 2 Tp) [1 1 r (1 2 Tc)]

N 2P (1 2 Tp) (5)

where Tcg is the capital gains tax rate5Thus putting the equations together if an executive is paid cash P

and invests g in bonds and 1 2 g in equities the N-year payoff (includ-ing the payoff to the corporation from the tax deduction) is

P [1 1 r (1 2 Tc)]N Tc 1 g $(P [1 1 r (1 2 Tc)(1 2 Tp)]

N (1 2 Tp)1(1 2 g) (P [1 1 r (1 2 Tc)]

N (1 2 Tp) 2 Tcg $P (1 2 Tp)[1 1 r (1 2 Tc)]

N 2 P (1 2 Tp)) (6)

The tax advantage of options versus salary and bonus therefore is thedifference between the combined (corporate and executive) payoffs inequation (3) and the combined payoffs in equation (6)

43 The Tax Advantage of OptionsWe now turn to analysis of how changes in various tax rates affectthe tax advantages of options relative to cash compensation We focuson the case in which the cash earned by executives is invested entirelyin equities since equities are tax-favored relative to bonds The impor-tant conclusions of this analysis are not substantively different if we

5 In practice executives hold a combination of equities and bondlike instruments withtaxable interest Because equities are tax-favored in the next section we simplify the analy-sis by assuming that executives hold only equities Alternatively we could have assumedthat executives hold only less risky bonds since bonds more closely match the risk proleof a stream of tax savings None of our empirical results are substantively affected by thissimplication

Taxation of Executive Compensation 17

instead assume that a portion of an executiversquos holdings are invested inbonds

431 The Tax Advantage of Options Changing the Corporate TaxRate Stock options are tax-advantaged relative to cash simply becauseoption payouts are deferred allowing the executive to invest at the pre-personal-tax rate of return (with no capital gains tax at the end) ratherthan investing at the after-tax rate of return or paying the capital gainstax Since the advantage of being able to defer taxes is large when thecorporate tax rate is low the tax advantage of options is larger when thecorporate rate is low

To illustrate this point we dene the tax advantage of options to bethe dollar amount by which total after-tax option payoffs (to the corpora-tion and the executive) exceed the total payoffs from salary as denedby the difference of equation (3) and (6) assuming that P (the paymentto the executive) is equal to $100 and g 5 0 (cash compensation is in-vested in equities) We use a ten-year horizon since most options have aten-year term We then calculate how the tax advantage of optionschanges as the corporate tax rate varies from zero to 100 percent holdingthe personal tax rate and the capital gains rate xed at 40 percent and 20percent respectively (which approximates current rates)

The top panel of Figure 2 shows the tax advantage of options relativeto cashmdashprecisely the NPV of the total tax advantage of options for a$100 payment to the CEOmdashas the corporate tax rate changes As ex-pected the tax advantage of options declines as corporate tax rates in-crease At a zero corporate tax rate the tax advantage of options has anNPV of $7 This value is about $350 at a corporate rate of 40 percentWhen the corporate tax rate is 100 percent the tax advantage of optionscompletely disappears since the benets of deferred compensation fallto zero because the after-tax return on equities (and therefore the dis-count rate) falls to zero The tax advantage of options is the NPV of thetax savings from avoiding the capital gains tax on ten years of apprecia-tion which is why it is relatively modest even at the 40-percent corpo-rate rate

432 The Tax Advantage of Options Changing the Personal TaxRate Using the same assumptions as above (but this time holding thecorporate rate at 40 percent and the capital gains rate at 20 percent andvarying the personal rate) the tax advantage of options declines as thepersonal tax rate rises as shown in the middle panel of Figure 2 The taxadvantage of options is higher at low personal tax rates because theposttax base that gives rise to capital gains is higher and the advantage of

Note Holds constant the personal rate at 40 percent and the capital gains rate at 20 percent

Note Holds constant the corporate rate at 40 percent and the capital gains rate at 20 percent

Note Holds constant the personal rate at 40 percent and the corporate rate at 40 percent

FIGURE 2 Tax Advantage of Options NPV of Tax Advantage of Pay-ing $100 in Options (NQSOs) Rather Than in Cash

Taxation of Executive Compensation 19

avoiding the capital gains taxes is greater The tax advantage of optionstherefore declines linearly as the personal tax rate rises At a 100-percentpersonal tax rate the executive receives nothing in either case and byconstruction the companyrsquos tax deduction benet is the same in NPVterms so the tax advantage of options falls to zero also6

433 The Tax Advantage of Options Changing the Capital Gains TaxRate The tax advantage of options for various capital gains rates isdepicted in the bottom panel of Figure 2 under the same assumptions(this time holding both the personal rate and the corporate rate at 40percent while varying the capital gains rate) The tax advantage of avoid-ing the capital gains tax increases as the capital gains tax rate rises

5 HOW HAS THE TAX ADVANTAGE OF OPTIONSCHANGED OVER TIME

The analysis so far has illustrated how the tax advantage of optionsvaries with the changes in tax rates In this section we show how the taxadvantage of options has changed over time as tax rates have changed

51 Top Marginal Tax Rates over TimeThe evolution on the top marginal tax ratesmdashpersonal corporate andcapital gainsmdashfrom 1980 to 1998 is shown in Figure 3 In all cases thetop rate (the rate for taxpayers with the highest incomes) is shownwhich is not always the highest rate since various anomalies (such asclawbacks of exemptions) sometimes lead to marginal tax rates that arehigher than those for the highest income tax payers

The top personal tax rate was 70 percent in 1980 falling to 50 percentin 1982 and 28 percent in 1988 following the 1986 tax act7 The topmarginal tax rate has since risen to 396 percent but is effectively 425percent since there is a 29-percent Medicare surcharge (paid half by theemployer and half by the employee) that has no upper limit The topcorporate tax rate has had only one signicant change since 1980 fallingfrom 46 percent before the 1986 tax act to 34 percent following the act in1988 The top corporate tax rate was increased to 35 percent in 1993 The

6 This is one place where ignoring bond investments has substantive implications Inparticular in the all-bond case the tax advantage of options is nonmonotonicmdashit rises andthen fallsmdashas personal tax rates rise7 During the early 1980s the maximum tax on earned income limited the marginal tax rateon the earnings of high earners so many executives did not face a decline in personal taxrates between 1981 and 1982 The empirical results in this paper are robust to assumingthat executives faced a marginal tax rate of 50 percent in 1980 and 1981

20 Hall amp Liebman

FIGURE 3 Top Marginal Tax Rates since 1980

top capital gains rate has uctuated between 28 percent (approximately)and 20 percent since 1980 and is currently at 20 percent

52 The Evolution of the Tax Advantage of NQSOsThe tax advantage of options relative to cash dened in the same way asin section 4 from 1980 to 1998 is shown in Figure 4 The calculations aredone using the same assumptions as before (ie N 5 10 years and cashis invested in equities) and the statutory top marginal tax rates shown inFigure 3 are used Due to the fall in corporate and personal tax rates thetax advantage of options increased sharply after the Tax Reform Act of1986 This increase in tax advantage has largely been reversed in the1990s as top personal tax rates have crept back up and the capital gainsrate has been reduced

Two key facts emerge from this analysis First there is currently only amoderate tax advantage to standard nonqualied stock optionsmdashon theorder of $4 per $100 of compensation This is because the tax advantagesto the executive of deferring taxes are largely offset by the tax disad-vantages to the company of not being able to deduct option expensesfrom taxable prots until the executive exercises the options Secondalthough the 1986 tax act substantially increased the tax advantage ofoptions more than half of this increase has been reversed in the 1990s

Taxation of Executive Compensation 21

FIGURE 4 Tax Advantage of Options 1980 to 1998

NPV of tax advantage of $100 in options (NQSOs) rather than cash

53 The Tax (Dis)advantage of ISOsSo far the analysis has focused on the tax advantages of NQSOs sincethey are so much more prevalent than ISOs which account for onlyabout 5 percent of option grants The relative scarcity of ISOs can beexplained both by their tax status and by their per-executive cap of$100000 per year

Under what conditions are ISOs tax-preferred to NQSOs For a trans-fer P to an executive the ISO is always tax-preferred by the executive(since the capital gains rate is lower than the personal rate) while theNQSO is always tax-preferred from the companyrsquos perspective (since anISO is not deductible) The key issue however is the relative advan-tages of the two types of options from a global-contracting perspectiveBecause ISOs are not deductible the company is indifferent betweensetting aside P in the form of NQSOs today and setting aside P(1 2 Tc) inthe form of ISOs (Since the NPV of the tax deduction for NQSOs is PTcthe NPV of the payment P is P 2 PTc or P (1 2 Tc))

Equalizing the posttax cost of NQSOs and ISOs we need only look atthe payoff of the executive to determine the condition under which eachis preferred The payoff of NQSOs in N years is

P [1 1 r (1 2 Tc)]N (1 2 Tp) (7)

22 Hall amp Liebman

The same payoff for ISOs is

P (1 2 Tc) [1 1 r (1 2 Tc)]N (1 2 Tcg) (8)

ISOs therefore are tax-advantaged if

Tp 2 Tcg Tc 2 Tc Tcg (9)

Intuitively ISOs are tax-advantaged only if their advantage (the differ-ence between the personal rate and the capital gains rate) is largeenough to offset their cost (the disadvantage of not deducting at thecorporate rate) Note however that the condition is not a simple com-parison between the corporate rate and the personal rate minus thecapital gains rate (Tp 2 Tcg Tc)

Figure 5 shows how the tax advantage of NQSOs relative to ISOs haschanged since 1980 as tax rates have changed Unlike the very modestchanges (around 3 to 4 percent) in the relative tax advantages betweenoptions and cash shown in Figure 4 the relative tax advantage betweenNQSOs and ISOs has seen enormous swings When the top personalrate was 70 percent in 1980 ISOs were tax-favored by a margin greaterthan 6 percent However by 1982 (following the 1981 tax acts that dra-matically lowered the personal tax rate) ISOs became tax-disadvantaged

FIGURE 5 Tax Advantage of NQSOs Relative to ISOs over Time

NPV of tax advantage of $100 payment to executive NQSOs minus ISOs

Taxation of Executive Compensation 23

by about 5 percent By 1988 (following the 1986 tax act) ISOs became tax-disadvantaged by about 18 percent a dramatic swing The raising ofpersonal rates coupled with a decrease in the capital gains rate has sincereduced the tax disadvantage of ISOs to about $4 per $100 of pretaxcompensation Although hard data on ISOs are hard to come by practi-tioners (mostly compensation consultants) claim that ISOs were morepopular prior to 1982 when they were tax-advantaged Although ISOsare less disadvantageous from a tax standpoint relative to NQSOs thanthey were in the late 1980s they are still disadvantageous so it is notsurprising that they have not made a signicant resurgence in recentyears

6 EXPLAINING THE INCREASE IN OPTION-BASEDCOMPENSATION

In this section we analyze whether changes in tax rates affect the compo-sition of executive compensation and consider the relative importanceof tax factors and corporate governance factors in explaining the in-creasing share of compensation paid in the form of stock options Thenumerous changes in tax rates that have occurred since 1980 provide anopportunity to assess whether executive compensation arrangementsrespond to tax incentives in the way that the tax-avoidance model sug-gests We test this model using a panel data set of CEOs in largepublicly-traded US corporations Our identication of the tax effectsrelies on time-series variation in personal corporate and capital gainstax rates along with cross-sectional variation in corporate tax rates

61 DataWe use a 15-year panel data set of CEOs in the largest publicly-tradedUS corporations which is described in Hall and Liebman (1998) Thedata set covers the years from 1980 through 1994 and combines CEOcompensation information from corporate proxies and 10-K lings withstock price and stock return information from CRSP and accounting datafrom Compustat In addition some compensation data from the 1970swere collected in order to construct measures of the value of stock op-tions held by the CEOs in the rst period of the sample8

The feature that distinguishes our data from most other CEO datasets is that with our panel of yearly proxy data on option grants optiongains and total options held we are able to calculate the total value of

8 We are grateful to David Yermack for providing us with some of the data for 1984 to 1991See Yermack (1995) for a discussion of these data

24 Hall amp Liebman

all stock options held by the CEO at a point of time More importantlysince we have the details about the stock options held (number exer-cise price time to maturity etc) we can precisely calculate the changein the value of a CEOrsquos stock option holdings for a given change in rmvalue

62 Identication IssuesThe numerous tax changes over our sample period and the specic waysin which changes in personal corporate and capital gains rates arepredicted to affect the composition of CEO compensation give us un-usually rich sources of identication However because many of thechanges in tax rates would be expected to affect all of the CEOs in oursample in a similar way we will need to pay particular attention toseparating out the tax effects from underlying time trends and otherfactors that changed over time Controlling for underlying trends is par-ticularly important in this study because we are focusing on an out-comemdashthe use of stock optionsmdashthat has increased rapidly over thepast two decades and that many practitioners believe was importantlyaffected by nontax considerations Thus the tax effects we are trying toexplain are deviations from a rising trend

We take two steps to try to separate out the tax effects from theunderlying trend First we include key nontax factors that could poten-tially explain the increasing reliance on performance-based compensa-tion It has been suggested that the dramatic increase in incentive-basedpay is the result of the remarkably poor shareholder returns during the1970s which spurred the LBO and takeover movements of the 1980sAccording to this story (Kaplan 1997) shareholders became much morepowerful via the rise of institutional investors even as the LBO andtakeover movements waned Because the inuence of institutional inves-tors is thought to be one of the most important mechanisms of strongcorporate governance we use the share of each companyrsquos stock ownedby institutional investors (dened as institutions with more than $100million under management) as an explanatory variable9 The percentageof shares owned by large institutional investors increased from about 20percent to almost 50 percent in 1994 an upward trend that closelymatches the sharp rise in the share of compensation in the form ofoptions Indeed in our sample the annual average (over all of the rmsin our sample) share of stock owned by institutional investors has acorrelation of 09 with the annual average share of compensation paid inoptions

9 This variable is described in Gompers and Metrick (1998)

Taxation of Executive Compensation 25

In addition we include two other variables that proxy for strongercorporate governance The rst is the size of the board Evidence suggeststhat boards with smaller numbers of directors reduce the inuence of theCEO and better represent shareholder interests (Yermack 1995) Boardsize fell by approximately 15 percent between 1980 and 1994 The secondvariable is the fraction of the rmrsquos board members who are inside direc-tors or gray directors (noninsiders who have business dealings with thecompany) Firms with fewer outsiders are less likely to act in the share-holdersrsquo interests and we would predict that they would be less likely tohave performance-based compensation for their CEOs The average per-centage of inside and gray directors has decreased from about 45 percentin 1980 to about 35 percent in 1994

Second we allow for cross-sectional variation in corporate tax ratesCorporate marginal tax rates vary because rms may be eligible for tax-loss carrybacks and carryforwards investment tax credits and the alter-native minimum tax (see Auerbach 1986 Auerbach and Altshuler1990 Auerbach and Poterba 1987) Because these cross-sectional differ-ences imply that different rms should respond differently to a givenchange in tax rates they potentially provide us with a way to isolatethe tax effects Firm-specic corporate marginal tax rates are notori-ously difcult to calculate from publicly-available data We use a tricho-tomous variable that equals zero if the rm has tax-loss carryforwardsand negative earnings in a given year equals 05 times the statutorycorporate rate if the rm has only one of those conditions and equalsthe statutory rate if the rm has neither loss carryforwards nor negativeearnings10

The cross-sectional variation in the corporate tax rate may not be ex-ogenous Firms that perform poorly could face low marginal tax ratesand also be particularly likely or unlikely to provide performance-basedpay Because most of our specications use xed effects the correlationwe must be concerned about is between changes in rm performanceand changes in compensation We deal with this by controlling forlagged rm stock market performance in our regressions This shouldeliminate spurious correlation between tax rates and compensation thatis jointly caused by rm performance

10 Graham (1996) presents evidence that the trichotomous variable performs nearly as wellas his recommended simulated marginal tax rate Plesko (1999) compares marginal taxrates calculated from Compustat data with actual marginal tax rates from rm tax returnsHe nds that simple binary variables are more reliable measures of the marginal tax ratethan is either the Graham or the trichotomous variable We obtain substantively similarresults when we switch from the trichotomous variable to the binary variables recommendby Plesko

26 Hall amp Liebman

63 ResultsThe dependent variable in our regressions is the fraction of a CEOrsquos totalannual compensation (measured as the BlackndashScholes value of stock-option grants plus salary and bonus) that comes from stock-optiongrants We choose this reduced-form specication because our interest isin the composition of CEO compensation However the factors thataffect the composition of compensation might also affect the levels andit will be important to keep this in mind in interpreting our results

Optimal-contracting theory generally makes predictions about the in-centives provided by the total amount of rm stock and stock optionsowned by the CEO We choose to model the ow of option grants ratherthan the ultimate performance sensitivity of the CEOrsquos compensationand stock and stock-option holdings because we have in mind anadjustment-cost model in which it takes time for a CEOrsquos holdings ofstock and stock options to reach the optimal level11 In our regressionswe include two measures of the current performance sensitivity of theCEOrsquos wealth to rm performance as explanatory variables to reect thedistance of the CEOrsquos contract from the optimal contract

Table 3 presents results in which the tax effects are identied solely byvariation in tax rates over time The fraction of annual compensation paidin stock options is regressed on tax variables corporate governance vari-ables contract-theory variables and variables reecting stock market re-turns Both of the regressions in this table include rm xed effects Incolumn (1) the fraction in annual compensation paid in stock options isregressed on the log difference in payoff (the combined payoff to both therm and the CEO) from receiving compensation in stock options ratherthan salary and bonus where the payoff calculations are based on statu-tory marginal tax rates This variable is dened as the log difference be-tween equation (3) and equation (6) in section 4 The coefcient on the taxvariable is large and statistically signicant at the 95-percent level Thepoint estimate implies that a one-percent increase in the payoff differencebetween stock options and salary and bonus results in a 24-percentage-point increase in the share of compensation paid in stock options

The corporate governance variables all have the predicted sign Firmswith a higher fraction of their shares owned by institutional investorsuse more performance-based pay (stock options) Firms with large corpo-rate boards or a large share of inside directors are less likely to use stockoptions

The two contract-theory variables are the dollar change in CEO wealth

11 See Core and Guay (1999) for evidence that rms adjust in this way

TABLE 3Compensation Regressions that Rely on Variation in

Tax Rates over Time

(1) (2)

Tax variablesln (difference in payoff)(a) 2393 (0349)ln (12statutory personal 20015 (0034)

rate)ln (12statutory cap gains 0330 (0171)

rate)ln (12statutory corporate 0399 (0148)

rate)

Corporate governance variablesFraction of rmrsquos shares 0043 (029) 0052 (0028)

owned by institutionalinvestors

Size of the rmrsquos board of 20002 (001) 20003 (0001)directors

Fraction of directors who 20082 (033) 20092 (0030)are inside or gray

Contract-theory variablesD (CEO wealth) per $1000 20261 (0095) 20247 (0087)

in FV(b)

D (CEO wealth) per 10 in 2583 (189) 2554 (201)FV(c)

Volatility(d) 521 (424) 261 (369)ln (rm market value) 0019 (0006) 0019 (0006)

Stock market return variablesLagged annual rm stock 0006 (0009) 0003 (0008)

market returnLagged annual SampP 500 re- 20035 (0023) 20029 (0023)

turn

Regression diagnosticsTotal sample size 5179 5179Firms 427 427Adjusted R2 0313 0311

Dependent variable share of compensation paid in options Standard errors in parentheses Bothregressions in this table include rm xed effects

(a) Log difference in payoff is the log difference in combined rm and CEO after-tax share betweenpaying compensation in options and paying salary

(b) This variable is the dollar change in CEO wealth per $1000 change in rm market value (seeJensen and Murphy 1990)

(c) This variable in the dollar charge in CEO wealth per 10-percent change in rm value (see Hall andLiebman 1998)

(d) Standard deviation of rm market value in trillions

28 Hall amp Liebman

per $1000 change in rm value (Jensen and Murphy 1990) and thedollar change in CEO wealth per 10-percent change in rm value (Halland Liebman 1998) As discussed in Baker and Hall (1998) these twowealth sensitivity measures reect two different concepts of how closelyaligned the managersrsquo incentives are with the interests of the sharehold-ers In the regression they are meant to measure how far the CEOrsquosexisting contract and ownership of stock and stock options is from theoptimal level The negative coefcients on the two variables suggeststhat rms that are below their optimal pay-to-performance sensitivityare more likely to give more stock options in the current period

The second column replaces the tax variable motivated by deferraladvantages with the three statutory rates The corporate after-tax sharehas a positive coefcient as we would expect because the tax advantageof stock options increases as the corporate rate falls The coefcient onthe personal tax rate is small and statistically insignicant The sign ofthe coefcient on the capital gains rate is the opposite of that predictedby tax-avoidance theory

Note that these results rely on time-series variation alone This is not avery convincing test Since the log difference in payoffs mostly rises overthe time period covered in our sample as does the share of compensa-tion paid in options it is possible that the tax variable is simply reectingother factors that were changing over this time period that we have notincluded in our regression Therefore it is important to look at specica-tions that rely on cross-sectional variation in the tax variables

Table 4 contains specications that rely on cross-sectional variation incorporate tax rates The rst column repeats the specication from col-umn (1) of Table 3 but uses the trichotomous measures of rm corporatetax rates in place of the statutory rate in calculating each rmrsquos logdifference in payoff from paying compensation in options rather than insalary and bonus As in the previous table we nd the positive coef-cient on the log difference in payoff just as the theory suggest How-ever the coefcient estimate is now much smaller suggesting that aone-percent increase in the log difference in payoff produces a 01-percentage-point increase in the share of compensation paid in options

Introducing cross-sectional variation in the dependent variable doesnot eliminate the potentially spurious correlation over time between thetax variable and the share of compensation paid in options The secondcolumn of Table 4 isolates the pooled cross-sectional variation in corpo-rate tax rates by adding time dummies and dropping the rm xedeffects In this specication the coefcient on the tax variable is indistin-guishable from zero The coefcient on the institutional variable contin-ues to be large and statistically signicant The third column includes

Taxation of Executive Compensation 29

both rm xed effects and time dummies which tests for a relationshipbetween within-rm changes in the dependent variable and within-rmchanges in each explanatory variable In this specication the coefcienton the tax variable is again indistinguishable from zero The coefcientson the corporate governance variables are also statistically insignicant

The third column contains a specication that treats the Tax ReformAct of 1986 (TRA 86) as a natural experiment For rms and CEOs facingthe statutory corporate tax rate TRA 86 increased the relative payoff topaying compensation in options by 17 percent For a rm facing a zeromarginal tax rate it reduced the relative payoff by 45 percent The reduc-tion occurred because TRA 86 reduced the personal tax rate makingsalary and bonus more attractive while the decreased corporate rate hadno effect on zero-tax-rate rms Thus we would expect rms facing zeroor low marginal tax rates to reduce their use of options after TRA 86while rms facing the statutory rate would be expected to increase themFew rms face zero marginal tax rates year after year Therefore we tryto distinguish between rms that often face low tax rates and rms thatnearly always face the statutory rate We dene low-tax rms as oneswhose average marginal tax rate over 1984 to 1986 was below 02412

Then we ran a difference-in-differences regression to see if high-taxrms increased their use of options more than low-tax rms The interac-tion between post-TRA 86 and high corporate tax rates is the key vari-able The estimated coefcient on this variable is positive but small andnot statistically different from zero at the 95-percent level

It is important to note that in this specication as well the main corpo-rate governance variable the share of the rmrsquos shares owned by institu-tional investors performs as theory would predict The variable appearsto be quite robust Table 5 presents regressions of the share of compensa-tion paid in stock options on the fraction of the rmrsquos shares owned byinstitutional investors Since we have already established that these vari-ables have a strong longitudinal relationship the regressions here aredesigned to focus on the cross-sectional relationship between these twovariables The rst and third columns present regressions using data onlyfrom 1994 and nd a coefcient of roughly 02 (column 1 contains noother covariates while column 3 includes all of the other nontax covari-ates from the regressions in Tables 3 and 4) Since the median share of armrsquos stock owned by institutional investors in our sample rises from 19percent in 1980 to 49 percent in 1994 this coefcient would imply that this

12 This cutoff was chosen after inspecting the distribution of tax rates and essentiallyseparates rms that are always at or near the statutory rate from ones that are sometimesbelow it

TA

BL

E4

Com

pens

atio

nR

egre

ssio

nsth

atR

ely

onC

ross

-Sec

tion

alV

aria

tion

inC

orpo

rate

Tax

Rat

es

(1)

(2)

(3)

(4)

Tax

vari

able

s(a)

ln

(dif

fere

nce

inp

ayof

f)0

099

(00

41)

20

038

(00

36)

20

005

(00

44)

Pos

t-T

RA

862

000

5(0

013

)H

igh

corp

orat

eta

xra

te0

018

(00

17)

(Pos

t-T

RA

86)

3hi

gh0

024

(00

16)

corp

orat

eta

xra

te

Cor

por

ate

gove

rnan

ceva

riab

les

Frac

tion

of

rmrsquos

shar

es0

109

(00

27)

011

4(0

015

)2

000

9(0

030

)0

087

(00

34)

owne

dby

inst

itu

tion

alin

vest

ors

Size

ofth

e

rmrsquos

boar

dof

20

004

(00

01)

20

001

(00

01)

20

002

(00

01)

000

1(0

002

)d

irec

tors

Frac

tion

ofd

irec

tors

wh

o2

012

4(0

033

)2

000

3(0

018

)2

004

1(0

034

)0

001

(00

39)

are

insi

de

orgr

ay

Con

trac

t-th

eory

vari

able

sD

(CE

Ow

ealt

h)p

er$1

000

20

238

(00

94)

20

167

(00

53)

20

257

(00

96)

20

021

(01

47)

inFV

(b)

D(C

EO

wea

lth)

per

10

in2

560

(17

8)2

592

(10

5)2

594

(19

6)2

147

(04

86)

FV(c

)

Vol

atili

ty(d

)2

48(4

24)

162

(03

4)1

12(4

28)

106

(06

09)

ln(

rmm

arke

tva

lue)

003

3(

006)

000

8(0

003

)0

009

(00

06)

001

2(0

007

)

Stoc

km

arke

tre

turn

vari

able

sL

agge

dan

nual

rm

stoc

k2

000

2(0

009

)0

011

(00

10)

000

6(0

009

)2

001

2(0

013

)m

arke

tre

turn

Lag

ged

annu

alSamp

P50

02

005

6(0

023

)2

005

3(0

051

)2

003

9(0

049

)2

015

4(0

040

)re

turn

Reg

ress

ion

dia

gnos

tics

T

otal

sam

ple

size

5179

5179

5179

2534

Firm

s42

742

739

6A

dju

sted

R2

031

009

032

005

Firm

xed

effe

cts

Yes

No

Yes

No

Tim

ed

um

mie

sN

oY

esY

esN

o

Dep

end

ent

vari

able

isfr

acti

onof

ann

ualc

omp

ensa

tion

from

stoc

kop

tion

sSt

and

ard

erro

rsin

par

enth

eses

(a)

All

thre

esp

eci

cati

ons

use

tric

hoto

mou

sco

rpor

ate

tax

rate

sT

he

sam

ple

inco

lum

n(3

)u

ses

only

the

year

s19

84to

1988

sin

cew

ear

eu

sing

1986

asa

nat

ural

exp

erim

ent

(b)

Thi

sva

riab

leis

the

dol

lar

chan

gein

CE

Ow

ealt

hp

er$1

000

chan

gein

rm

mar

ketv

alu

e(s

eeJe

nse

nan

dM

urp

hy

1990

)

(c)

Thi

sva

riab

lein

the

dol

lar

char

gein

CE

Ow

ealt

hp

er10

-per

cent

chan

gein

rm

valu

e(s

eeH

alla

nd

Lie

bman

199

8)

(d)

Stan

dar

dd

evia

tion

ofr

mm

arke

tva

lue

intr

illio

ns

32 Hall amp Liebman

TABLE 5Large Institutional Investors and Stock-Option Compensation

The Cross-sectional Relationship between Institutional Investors andOption Grants

(1) (2) (3) (4)

Coefcient on fraction of 217 (067) 145 (014) 177 (069) 120 (015)rmrsquos shares owned byinstitutional investors

Sample period 1994 1980ndash1994 1994 1980ndash1994Contract-theory and stock No No Yes Yes

market covariatesTime dummies No Yes No Yes

The dependent variable is the share of compensation paid in options

factor alone can explain 6 percentage points of the 23-percentage-pointincrease in the median share of compensation paid in options that oc-curred over this time period Of course this variable is only one measureof the strengthening of corporate governance so the total impact of im-proved corporate governance could be greater The second and fourthcolumns of the table use the entire 1980ndash1994 sample and include timedummies to isolate the cross-sectional correlations Again there is evi-dence of a strong cross-sectional relationship between the two variablesalthough the coefcients are slightly smaller perhaps reecting a weakercorporate-governance role in the earlier period In sum there is a strongrelationship in both longitudinal and cross-sectional data between theinstitutional-investor variable and the options-percentage variable13

7 THE MILLION-DOLLAR RULE

In 1993 section 162(m) of the Internal Revenue Code was enactedlimiting the deductibility of executive compensation in excess of onemillion dollars unless the compensation was performance-related Pro-ponents of this legislation argued that it would slow the growth ofexecutive pay and tighten the link between rm performance and execu-tive compensation

The new provision became effective for tax year 1994 and applies tothe CEO and the other four most highly compensated executives in eachcompany Pay that is performance-related such as stock-option grants

13 The one specication in which the fraction owned by institutional investors is notsignicant is in column (2) of Table 4 a regression that includes both xed effects and timeeffects

Taxation of Executive Compensation 33

and bonuses paid for meeting clear performance goals is not affected bythis regulation provided that it has been approved by shareholdersThus the main impact of the provision is to limit the deductibility ofsalary in excess of a million dollars Woodlock and Antenucci (1997)studied the proxy statements of 376 rms and documented that mostrms responded to the new law by qualifying top executive pay asperformance-related Perry and Zenner (1999) identied 25 rms thatreduced salaries from above $1 million to below $1 million and foundthat 23 of the rms cited section 162(m) as their reason for doing so

In this section of the paper we investigate whether this provision hasaffected the level or structure of executive pay using the Execucompdata base which contains information on the highest-paid executives ineach of the rms in the SampP 500 SampP Midcap 400 and SampP SmallCap600 Between 1993 and 1998 the median salary of the Execucomp execu-tives rose from $230000 to $306000 The median value of total compen-sation (salary bonus and option grants) rose from $470000 to $882000Limiting the sample of CEOs indicates that the total compensation of themedian CEO rose from $11 million to $18 million over the period whilethe median salary rose from $441000 to $550000 Thus executive com-pensation clearly continued to rise at a rapid rate after the implementa-tion of the million-dollar rule though perhaps at a slower rate than itotherwise would have

If the regulation affects behavior it should have the largest impact onrms whose executives are receiving salaries that are above the limit orjust below it Thus we might expect to see rms whose executives arereceiving relatively high salaries to rely more heavily on performance-based pay and less heavily on salary in the years following the regulationPerry and Zenner (1999) and Livingston (1999) both nd that this is thecase This trend can be clearly identied in the Execucomp data set Forexample over 1993 to 1998 the median annualized growth rate of salaryfor executives whose 1993 salary was below $500000 in the previous yearwas 85 percent For executives with salary between $500000 and$700000 it was 64 percent for executives between $700000 and $850000it was 50 percent for executives between $750000 and $1 million it was31 percent and for executives above $1 million it was 00 percent14

The observation that executives at lower levels of salary had highergrowth rates is not very strong evidence of an impact of section 162(m)however because it is possible that low-salary executives would havehad higher growth rates even in the absence of the provision To explore

14 Of the 3137 executives for whom we have complete data between 1993 and 1998 2708had 1993 salaries below $500000 while only 35 had 1993 salaries above $1 million

34 Hall amp Liebman

whether this is the case we use Execucomp data from both before andafter the law change Ideally we would want to have data from manyyears before the change to examine whether there is an underlying pat-tern of faster salary growth for low-salary executives Unfortunately theExecucomp data begin in 1992 and therefore provide us with only the1992ndash1993 growth rate in compensation to use as a baseline15

We run regressions pooling data from 1992ndash1993 through 1997ndash1998of the annual percentage change in different forms of compensation on avariable million that is designed to measure the likelihood that anexecutiversquos compensation will be affected by section 162(m) We denemillion as the minimum of 1 and the executiversquos salary divided by $1million in the previous year Thus any executive with salary at or above$1 million would receive a value of 1 while an executive with a salary of$500000 would be coded as a 0516 We include this variable by itself andalso interacted with a dummy variable (after) that equals 1 for years1994 and after Thus the million variable is designed to pick up anyunderlying relationship between the level of salary and the growth rateof salary while the interacted one will identify any differential growthrate of salary by income in the period after section 162(m) took effect Wealso include a full set of year dummies the natural logarithm of thermrsquos market value and the rmrsquos annual rate of return over the twoprevious years as control variables To reduce the sensitivity of our re-sults to outliers we run quantile regressions and robust regressions

Table 6 contains our results The rst four columns are for all topexecutives while the last four columns focus on CEOs Columns (1) and(2) present estimates from robust regression for the percentage change insalary and the percentage change in total compensation (salary bonusstock grants and stock-option grants) In the salary growth regressionsfor all executives the coefcient on million is negative and signicantindicating that there is an underlying relationship between the level ofexecutive compensation and its subsequent growth rate The coefcienton million 3 after is negative as well suggesting that the negativerelationship between the level of salary and its subsequent growth rateintensied slightly after 1994 The relatively small magnitude suggeststhat an executive with a $1 million dollar salary would see his salarygrow at an annual rate that would be about 06 percent slower than anexecutive earning $500000

15 The HallndashLiebman data set contains only a combined salary and bonus variable makingit impossible to use it to analyze this issue16 We also ran the regressions (not reported) allowing for a nonlinear relationship betweenmillion and compensation growth and obtained substantively similar results

TA

BL

E6

The

Eff

ect

ofth

eM

illi

on-D

olla

rR

ule

onSa

lary

and

Tot

alC

ompe

nsat

ion

All

top

exec

uti

ves

CE

Os

only

Rob

ust

regr

essi

onQ

uan

tile

regr

essi

onR

obu

stre

gres

sion

Qu

anti

lere

gres

sion

D

in

Din

D

in

Din

D

in

Din

D

in

Din

sala

ryto

talc

omp

(a)

sala

ryto

talc

omp

sa

lary

tota

lcom

p

sala

ry

tota

lcom

p

mil

lio

n(b

)2

006

02

035

72

006

32

024

12

008

72

023

12

010

32

024

5(0

005

)(0

032

)(0

006

)(0

027

)(0

010

)(0

073

)(0

011

)(0

056

)m

illi

on

32

001

10

026

20

021

20

015

20

016

014

92

003

40

158

aft

er(c

)(0

005

)(0

033

)(0

006

)(0

029

)(0

010

)(0

076

)(0

011

)(0

057

)N

359

5035

950

359

5035

950

757

17

589

757

17

589

Stan

dar

der

rors

inp

aren

thes

esR

egre

ssio

nsal

soin

clu

de

year

du

mm

ies

stoc

km

arke

tre

turn

and

log

mar

ket

valu

e

(a)

Tot

alco

mp

ensa

tion

incl

ud

essa

lary

bon

us

rest

rict

edst

ock

and

stoc

k-op

tion

gran

ts

(b)

mil

lion

isd

ene

das

min

(1s

alar

yin

pre

viou

sye

ars

div

ided

by$1

mil

lion)

(c)

aft

eris

anin

dic

ator

vari

able

equ

alto

zero

befo

reth

ere

gula

tion

took

plac

e(1

993)

and

one

afte

rwar

d

36 Hall amp Liebman

In the second column the coefcient on million 3 after is essentiallyzero suggesting that any decrease in salary brought about by section162(m) was offset by increases in bonus and stock-option grants Theresults are similar for the median regressions There is some evidence ofa very small slowdown in growth of salary for executives likely to beaffected by section 162(m) but no sign of a change in total compensa-tion Taken together this provides evidence of a very small substitutionof performance-related pay for salary

The last four columns run identical tests but limit the sample to CEOsThe interaction of million 3 after continues to be small and negative inthe salary regressions In contrast the impact of 162(m) on total compen-sation appears to be positive and fairly substantial (about 15 percentagepoints per year) in these regressions That is the increase in bonus andstock options following the million-dollar rule more than offset the verysmall decline in salary

All of these conclusions should be interpreted somewhat cautiouslybecause our data provide us with only one pre-1993 control year Never-theless the data are consistent with section 162(m) having led to a veryminor slowdown in salary growth and one that is dramatically smallerthan what one would conclude without controlling for underlying differ-ential salary growth rates at different levels of salary Since total compen-sation for CEOs did not decline and perhaps increased there is evidenceof a minor substitution of performance-related pay for salary in responseto the regulation

8 EXECUTIVE COMPENSATION AND THEELASTICITY OF TAXABLE INCOME

Even if taxes have only a minor effect on executive compensation thestructure of executive compensation is important for tax policy be-cause high-income taxpayers are sources of signicant revenue and arethe focus of important debates about the efciency cost of taxation

Goolsbee (1999) has recently argued that executive decisions to exer-cise options are highly responsive to intertemporal differences in taxrates created by tax reforms He claims that much of the apparent declinein taxable income for high-income taxpayers between 1992 and 1993documented by Feldstein and Feenberg (1996) can be explained by execu-tives shifting option gains into 1992 in order to avoid the higher marginaltax rates that went into effect in 1993 By extension he implies that thehigh elasticities of taxable income with respect to personal tax ratesestimated by comparing the pre- and post-tax-reform taxable income of

Taxation of Executive Compensation 37

high-income individuals reect intertemporal shifting of income and notpermanent effects of taxation If this is true then the deadweight loss oftaxation is lower as is the cost of progressivity

While Goolsbee presents comprehensive evidence that option gainswere unusually high in 1992 his identication of the impact of tax rates onthe timing of option gains depends on correctly specifying a counter-factual level of options that would have been exercised in the absence ofthe tax change He simply assumes a linear time trend The purpose ofGoolsbeersquos time trend is presumably to account for the increasing use ofoptions over time However since options typically require a vestingperiod of at least a couple of years a much more direct measure of ex-pected option gains can be constructed based on the total value of vestedoptions held by the CEO Since options typically have vesting periods thatexceed two years this total can be treated as exogenous since it will not beaffected by the tax changes

More generally the value of options held by the CEO is likely to havea major effect on his exercise decisions If his options are out of themoney he will not have gains to realize If there has been a recent run-up in the market an optimizing CEO may exercise an unusually largeamount of options in order to diversify his overall portfolio and if in-the-money options are about to expire they will be exercised in thecurrent year

Our evidence suggests that much of the apparent tax shifting caninstead be attributed to stock market performance and the timing of pastoption grants Using our 15-year panel we replicate Goolsbeersquos (1999)result for the 1993 period but show that the impact of taxes is notpresent in other time periods Moreover even the result for the 1993period disappears when appropriate controls for past option grants andstock price appreciation are included

The top three rows of Table 7 present data on the taxable incomes ofCEOs in the HallndashLiebman sample For these results our panel has beennarrowed to include only CEOs whose rmrsquos scal years correspondwith tax years In addition we create a balanced panel for each of thethree tax-reform periods by excluding any rm for which there are notcomplete data for the period surrounding the reform

The fourth row of the table shows that the 1993 tax act raised the topmarginal personal income tax rate from 310 to 396 percent effective fortax year 1993 with a further increase occurring in 1994 due to the uncap-ping of the Medicare payroll tax The top row of the table shows thattaxable income was higher in 1992 than for any other year in our sample(all data are in 1994 dollars) Option gains were particularly high in1992mdashmore than double their 1991 level and 68 percent higher than their

TA

BL

E7

Tax

Ref

orm

Sto

ckM

arke

tP

erfo

rman

cea

ndth

eT

imin

gof

Opt

ion

Gai

ns

Ave

rage

stoc

km

arke

tM

ean

CE

OM

ean

CE

Ore

turn

for

Mea

nva

lue

taxa

ble

sala

ryan

dM

ean

CE

OT

opfe

der

al

rms

inof

opti

ons

(Val

ue

ofop

tion

Sam

ple

inco

me

bonu

sop

tion

gain

sp

erso

nal

sam

ple

for

held

atbe

gga

ins)

(va

lue

Ref

orm

size

Yea

r($

mil

)($

mil

)($

mil

)M

TR

pre

vye

ar(

)of

year

ofop

tion

she

ld)

ER

TA

8127

919

800

874

064

80

163

7025

80

460

354

1981

083

10

672

009

869

245

070

61

3919

820

872

067

90

099

5011

90

686

144

1983

099

00

724

025

250

355

116

62

16

TR

A86

295

1985

119

70

837

024

050

76

116

02

0719

861

448

091

70

345

5042

81

776

194

1987

165

50

959

059

338

519

12

229

266

1988

163

21

058

039

028

07

237

71

6419

891

656

105

20

414

2817

92

345

176

1993

272

1990

209

51

028

061

828

232

288

02

15T

axA

ct19

911

875

102

00

547

312

01

268

82

0319

922

678

107

01

250

3146

64

552

275

1993

232

41

209

074

339

620

64

299

173

1994

196

41

262

037

842

515

45

113

074

Sam

ple

isH

allndash

Lie

bman

rm

sw

ith

Dec

embe

r

scal

year

sth

atha

dco

mp

lete

dat

afo

ron

eof

the

thre

eta

x-re

form

per

iod

sst

ud

ied

Taxation of Executive Compensation 39

1994 level These patterns are consistent with Goolsbeersquos claim that tax-payers responded to the anticipated increase in marginal tax rates byshifting gains into 1992

The evidence from the other two tax reforms is less clear suggestingthat there might be factors other than tax rates determining the timing ofoption gains ERTA 1981 reduced the top marginal tax rate from 70percent in 1980 to 50 percent in 1982 However there is no sign thatoption gains were reduced in 1980 and 1981 and then increased in 198217

There was a big increase in option gains in 1983 however Similarly thehighest year of stock-option gains in the period around 1986 was 1987 astrange year to take gains since taxpayers would have known that thepersonal rate would be 10 percentage points lower if they waited untilthe following year18

What all the peak years of option gains have in common is that theyfollow years of strong stock market performance The fth row of Table 7shows the average stock market return during the previous year for therms in our sample The large value of option gains in 1983 followed a355-percent increase in the stock market The 1992 boom in option gainsfollowed a 466-percent increase in the stock market the more modest1987 peak in option gains followed a 19-percent gain (and a 43-percentgain two years before)

Strong stock market performance raises the value of options availableto exercise The sixth row of the table shows the mean value of optionsheld by CEOs at the beginning of the year All three of the peak years ofoption gains correspond with peak years in terms of the value of optionsavailable for exercise Since stock market appreciation during the yearalso increases the value of options available to exercise later in that yearwe need to allow for current-year stock market appreciation as well

Table 8 contains Goolsbee-style regressions for each of the tax re-forms The rst two columns reprint the results from Goolsbee (1999)The third and fourth columns replicate his results in our sample Inparticular with a specication similar to his we are able to replicate hisnding that the current after-tax share has a positive coefcient and thefollowing yearrsquos after-tax share has a coefcient of roughly negative oneColumns (5) through (8) show that this pattern was unique to the 1993tax reform Indeed in the TRA 86 regressions the coefcient on the

17 As we explained above for executives covered by the maximum tax on earned incomethere was not a decline in tax rates between 1981 and 198218 It is possible that the large stock market decline in October 1997 may have promptedsome people to exercise options earlier than they otherwise would have However anincrease in option exercises typically follows stock market increases not declines

TA

BL

E8

Att

empt

sto

Rep

lica

teth

eT

ax-S

hift

ing

Res

ult

inO

ther

Tim

eP

erio

ds

Goo

lsbe

eG

ools

bee

1993

1993

Ent

ire

Ent

ire

regr

essi

onre

gres

sion

refo

rmre

form

TR

A86

TR

A86

ER

TA

81E

RT

A81

1980

ndash199

419

80ndash1

994

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

ln[1

2ta

x(t)

]1

244

108

20

455

032

62

017

62

017

22

016

72

016

70

026

20

068

(01

34)

(01

27)

(05

37)

(05

42)

(02

10)

(02

09)

(00

90)

(00

91)

(00

67)

(00

65)

ln[1

2ta

x(t1

1)]

20

867

20

764

21

214

20

917

047

80

722

20

103

004

30

176

018

3(0

118

)(0

112

)(0

491

)(0

499

)(0

157

)(0

174

)(0

072

)(0

072

)(0

068

)(0

066

)

Tim

e0

135

007

62

000

52

036

40

041

000

80

094

003

50

056

002

7(0

008

)(0

008

)(0

049

)(0

049

)(0

028

)(0

028

)(0

025

)(0

026

)(0

003

)(0

003

)

ln(m

arke

tva

lue)

066

10

489

023

70

334

039

5(0

015

)(0

058

)(0

039

)(0

042

)(0

017

)

Yea

rs19

91ndash1

995

1991

ndash199

519

90ndash1

994

1990

ndash199

419

85ndash1

989

1985

ndash198

919

80ndash1

983

1980

ndash198

319

80ndash1

994

1980

ndash199

4

Ad

just

edR

20

720

770

610

630

690

760

800

810

630

68

Obs

erva

tion

s13

856

132

371

385

136

41

470

143

31

113

106

23

968

385

9

Dep

end

ent

vari

able

isth

elo

gof

taxa

ble

inco

me

All

regr

essi

ons

con

tain

rm

xe

def

fect

sG

ools

bee

regr

essi

on(2

)al

soco

ntai

nstw

oot

her

vari

able

sno

tsh

own

inth

eta

ble

Stan

dar

der

rors

inp

aren

thes

es

Taxation of Executive Compensation 41

TABLE 9Alternative Explanations for the Timing of Option Exercises

1993 Entirereform TRA 86 ERTA 81 1980ndash1994

(1) (2) (3) (4)

ln[12tax(t)] 0108 0111 20051 0201(0324) (0136) (0048) (0057)

ln[12tax(t11)] 20134 0573 0125 0150(0336) (0156) (0053) (0066)

ln(market value) 0418 0177 0338 0418(0061) (0039) (0046) (0163)

Firmrsquos stock market return (t) 0298 0268 0214 0258(0044) (0045) (0031) (0023)

Firmrsquos stock market return 0162 0206 0031 0128(t21) (0041) (0008) (0029) (0021)

ln(value of options held by 0004 0008 0001 0009CEO at start of year) (0003) (0003) (0002) (0001)

Years 1990ndash1994 1985ndash1989 1980ndash1983 1980ndash1994Adjusted R2 064 071 082 068Observations 1349 1427 1036 3812

The dependent variable is the log of taxable income All regressions contain rm xed effects Stan-dard errors in parentheses

following yearrsquos tax rate is positive and signicant Columns (9) and (10)pool all three reforms and fail to replicate the pattern that would reectintertemporal shifting19

It is possible that the ease with which one can shift the timing ofoption gains is asymmetric It may be harder to shift option gains for-ward in time since some options expire and therefore must be exercisednow In addition options were a much smaller portion of compensationin the early 1980s so perhaps the ERTA 81 results are not surprisingNonetheless these results cast doubt on the proposition that this is thechannel through which spurious estimates of the elasticity of taxableincome have occurred for previous tax reforms and they suggest thatwe should look harder for alternative explanations for the timing ofoption gains

Table 9 attempts to predict options gains using the lagged and currentincrease in a rmrsquos market value and the value of the options held by theCEO at the start of the year The two stock market variables stronglypredict option gains while the value of options held has the correct sign

19 The results also fail to suggest large permanent effects of marginal tax rates on taxableincome

42 Hall amp Liebman

in all of the specications and is signicant in the pooled sample eventhough it is highly correlated with the two stock market return variablesIn contrast the tax-timing explanation fails even in the 1993 period

It is important to realize that our results do not invalidate Goolsbeersquoscritique of claims that aggregate revenue data from 1992 and 1993 provethat elasticities of taxable income are large We have simply given analternative explanation for why taxable incomes were unusually high in1992 However our results do cast doubt on the conclusion that largetiming shifts are ubiquitous and capable of explaining away the morecareful estimates of taxable income elasticities that have been performedusing micro data (eg Feldstein 1995 Auten and Carroll 1995)

One further point is in order It is sometimes argued that the combina-tion of increased use of options and the booming stock market explainswhy federal tax revenues have been so much higher than expected dur-ing the 1990s However while large option gains increase revenue fromthe personal income tax they produce offsetting deductions for cor-porations The net impact on revenue is only the difference in the twomarginal tax rates times the option gain While the gap between thepersonal rate and the effective corporate rate is nontrivial the impact oflarge option exercises on total revenue is substantially smaller than theamount by which the exercises inate personal tax revenue and reducecorporate tax revenue

9 CONCLUSION

We have described the tax rules for executive pay in detail and analyzedhow changes in various tax rates affect the tax advantages of stock optionsrelative to salary and bonus We nd that there is a moderate tax advan-tage to optionsmdashcurrently about $4 per $100 in pretax executive paymdashand that changes in the tax advantage of options over time have had atmost a modest impact on the composition of pay Corporate governancefactors particularly the role of large investors appear to be more impor-tant in explaining the dramatic increase in option pay

In addition our evidence suggests that more direct attempts to use taxpolicy to inuence executive compensation have had little effect Wend that the million-dollar rule led companies to substitute perfor-mance-based pay for salary But our evidence suggests that this substitu-tion was quite modest and there is no evidence that the total level of paywas reduced Overall although the stock-option explosion has dramati-cally increased the link between pay and performance this change isdue almost entirely to nontax factors

Taxation of Executive Compensation 43

REFERENCES

Auerbach Alan J (1996) ldquoThe Dynamic Effects of Tax Law Asymmetriesrdquo Re-view of Economic Studies 53451ndash405

mdashmdashmdash and James Poterba (1987) ldquoTax Loss Carryforwards and Corporate TaxIncentivesrdquo In The Effects of Taxation on Capital Accumulation M Feldstein(ed) Chicago University of Chicago Press

Altshuler Rosanne and Alan Auerbach (1990) ldquoThe Signicance of Tax LawAsymmetries An Empirical Investigationrdquo Quarterly Journal of Economics 10561ndash86

Auten Gerald and Robert Carroll (1995) ldquoThe Effect of Income Taxes on House-hold Behaviorrdquo Ofce of Tax Analysis US Department of the Treasury

Baker George P and Brian J Hall (1998) ldquoCEO Incentives and Firm SizerdquoNBER Working Paper no 6868 December

Core John and Wayne Guay (1999) ldquoThe Use of Equity Grants to ManageOptimal Equity Incentive Levelsrdquo Draft

Feldstein M (1995) ldquoThe Effect of Marginal Tax Rates on Taxable Income APanel Study of the 1986 Tax Reform Actrdquo Journal of Political Economy 103 (no3)551ndash72

mdashmdashmdash and D Feenberg (1996) ldquoThe Effect of Increased Tax Rates on TaxableIncome and Economic Efciency A Preliminary Analysis of the 1993 Tax RateIncreasesrdquo In Tax Policy and the Economy vol 10 James M Poterba (ed)Cambridge MA MIT Press

Gompers Paul A and Andrew Metrick (1998) ldquoInstitutional Investors andEquity Pricesrdquo NBER Working Paper no 6723

Goolsbee Austan (1999) ldquoWhat Happens When You Tax the Rich Evidencefrom Executive Compensationrdquo NBER Working Paper no 6333

Graham John R (1996) ldquoProxies for the Corporate Marginal Tax Raterdquo Journal ofFinancial Economics 42187ndash221

Hall Brian J (1999) ldquoThe Design of Multi-year Stock Option Plansrdquo Journal ofApplied Corporate Finance 12(2)97ndash106

mdashmdashmdash and Jeffrey B Liebman (1998) ldquoAre CEOs Really Paid Like Bureau-cratsrdquo Quarterly Journal of Economics CXII(no 3)653ndash691

Jensen Michael C and William M Meckling (1976) ldquoTheory of the Firm Mana-gerial Behavior Agency Costs and Ownership Structurerdquo Journal of FinancialEconomics III305ndash360

mdashmdashmdash and Kevin J Murphy (1990) ldquoPerformance Pay and Top-ManagementIncentivesrdquo Journal of Political Economy XCVIII225ndash264

Kaplan Steven N (1997) ldquoThe Evolution of US Corporate Governance WeAre All Henry Kravis Nowrdquo Draft

Livingstone Jane R (1999) ldquoExecutive Compensation Contracting and Re-sponses to Increased Tax Costsrdquo Louisiana State University June

Miller Merton H and Myron S Scholes (1982) ldquoExecutive CompensationTaxes and Incentivesrdquo In Financial Economics Essays in Honor of Paul CootnerWilliam F Sharpe and Cathryn M Cootner (eds) Princeton NJ Prentice-Hall

Murphy Kevin J (1999) ldquoExecutive Compensationrdquo forthcoming in OrleyAshenfelter and David Card (eds) Handbook of Labor Economics Vol 3 NorthHolland

Perry Todd and Marc Zenner (1999) ldquoPay for Performance Government Regu-

10 Hall amp Liebman

Second the tax treatment of ISOs is different ISOs are completely un-taxed at grant or exercise It is only at sale that the executive is taxed andeven then the executive is taxed at the lower capital gains tax rate Thedisadvantage is that the corporation never gets to take a parallel taxdeduction against corporate prots Thus ISOs become more attractiveas the personal tax rate increases and as the corporate tax rate and thecapital gains tax rates fall

Two other related types of compensation restricted stock and stockappreciation rights are worth brief discussion Both however are farless common than standard stock options Restricted stock is payment inthe form of restricted shares the restriction being that the shares vestover time as with options Unlike options the shares typically havevoting and dividend rights With regard to taxation the executive istaxed at the personal rate on the value of the restricted stock as thevesting restrictions lapse However the executive may choose to betaxed at the grant date in which case all subsequent appreciation istaxed at the capital gains rate The dividends paid to the executive aretaxed at the ordinary rate in the usual way The company generallyreceives a parallel deduction equal to the amount of the executiversquosincome when the executive is taxed Unlike stock options restrictedstock is not generally considered to be performance-based and is there-fore subject to the million-dollar rule (unless the vesting of the restrictedstock is performance-based which is sometimes the case)

Stock appreciation rights (SARs) are rights that replicate the payoffs ofstock options with a cash transfer Thus SARs are simpler than options inthat there is no requirement to buy and resell the stock in order to ldquocashoutrdquo SARs generally have the same tax treatment as NQSOs both to theindividual and to the corporation SARs are relatively rare however be-cause they have disadvantageous accounting treatment (described in thenext section) relative to stock options and their main relative advantagevis-a-vis options has been essentially replicated through broker-assistedcashless option exercisesmdashwhereby a broker makes a ldquonanosecondrdquo loanto the executive (to purchase and resell the stock) when the executivewants to ldquocash inrdquo on option prots The tax and accounting treatment ofISOs restricted stock and SARs is summarized in Table 2

32 The Accounting Treatment of OptionsUnlike cash compensation which is expensed against earnings there isgenerally no expense recognition (at grant exercise or sale) for optionswhether they be NQSOs or ISOs As a result compensation consultantsoften point out that stock options are the only form of compensation thatare free in an accounting sense but still deductible for tax purposes

Taxation of Executive Compensation 11

Stock options do however lead to expense recognition if they are dis-counted (in the money at grant date) or if the exercise price and numberof options are not known at grant date For discount options the differ-ence between the stock price and the market price is expensed over thevesting period For options with variable terms (eg a variable exerciseprice vesting that is tied to performance) the options are marked tomarket and expensed during the time between grant and exercise Practi-tioners claim that the accounting treatment of options plays an impor-tant role in the design of option programs Thus plans that have ldquobadaccountingrdquo but are thought by many to have attractive incentive fea-tures are often not even seriously considered by companies Examples ofsuch potentially attractive plans include indexed options (where a CEOprots only if his rmrsquos share price grows relative to some market orindustry benchmark) and option grants that are explicitly performance-related both of which would lead to expenses against earnings

Unlike options restricted stock and SARs do not generally receivefavorable accounting treatment Restricted stock is generally expensedover the period in which the restrictions lapse (usually the vesting pe-riod) The magnitude of the expense is the difference between the cur-rent stock price and the executiversquos cost (if any) SARs are marked tomarket each period and the difference between the stock price and theexercise price is expensed over the outstanding period of SARs

4 THE TAXATION OF EXECUTIVE PAY CASHVERSUS OPTIONS

Stock-based compensation performs two roles in executive compensa-tion arrangements First it helps align the incentives of the executivewith the interests of the rmrsquos shareholders Second it often enables therm to compensate the CEO in a way that is more advantageous from atax standpoint than paying the executive in salary and bonus

41 Agency Theory and Executive Compensation in the Presenceof TaxationIn standard agency-theory models ( Jensen and Meckling 1976) agencycosts are the result of the separation of ownership and control Managersdo not have the same incentives as the owners The optimal incentivecontract for managers balances the benets of high-powered incentives(linking the fortunes of owners and managers through stock and stockoptions for example) with the costs of loading too much risk on risk-averse managers

TA

BL

E2

Tax

and

Acc

ount

ing

Tre

atm

ent

ofE

xecu

tive

Com

pens

atio

nIn

cent

ive

Stoc

kO

ptio

ns(I

SOs)

Res

tric

ted

Stoc

kan

dSt

ock

App

reci

atio

nR

ight

s(S

AR

s)

Tax

atio

nat

Tax

atio

nat

Acc

ount

ing

Com

pen

sati

onD

escr

ipti

onp

erso

nall

evel

corp

orat

ele

vel

trea

tmen

t

Ince

nti

vest

ock

ISO

sar

eid

enti

calt

oIf

the

usu

alco

ndit

ions

Ifth

eu

sual

cond

itio

nsA

sw

ith

NQ

SOs

ther

eop

tion

sN

QSO

sin

thei

rd

esig

n

are

met

th

ere

isno

tax

are

met

the

reis

gene

r-is

gene

rally

noex

pen

seH

owev

ert

hey

have

anat

gran

tor

exer

cise

At

ally

no

corp

orat

eta

xd

e-re

cogn

itio

nfo

rIS

Os

annu

alca

pof

$100

000

sale

all

opti

onp

ro

tsd

uct

ion

atgr

ant

exer

-(e

ver)

on

the

amou

ntth

atan

dsu

bseq

uen

tca

pita

lci

seo

rsa

le

vest

sT

he

tax

trea

t-ap

prec

iati

onar

eta

xed

men

tof

ISO

sal

soat

the

cap

ital

gain

sd

iffe

rs

rate

Res

tric

ted

stoc

kSi

mila

rto

stoc

kop

-T

hein

div

idu

alis

taxe

dT

heco

mp

any

gene

rall

yR

estr

icte

dst

ock

isge

n-ti

ons

but

shar

esat

the

ord

inar

yra

teas

rece

ives

ap

aral

leld

e-er

ally

exp

ense

dov

er(w

hich

vest

slow

ly)

the

rest

rict

ions

lap

se

du

ctio

neq

ual

toth

eth

eve

stin

gp

erio

dT

hera

ther

than

opti

ons

are

How

ever

the

man

ager

amou

ntof

the

exec

u-

amou

nt

exp

ense

dis

gran

ted

toth

eex

ecu

-m

aych

ose

tobe

taxe

dti

versquos

inco

me

whe

nth

eth

ed

iffe

renc

ebe

twee

nti

veU

nlik

eop

tion

sat

gran

td

ate

inw

hich

exec

uti

veis

taxe

dR

e-th

ecu

rren

tst

ock

pric

eth

esh

ares

typi

cally

case

alls

ubs

equ

ent

ap-

stri

cted

stoc

kis

gen

er-

and

the

exec

uti

versquos

have

voti

ngan

dd

ivi-

pre

ciat

ion

ista

xed

atal

lysu

bjec

tto

the

cost

(if

any)

d

end

righ

ts

the

cap

ital

gain

sra

te

mill

ion-

dol

lar

cap

D

ivid

end

sar

eta

xed

atth

eor

din

ary

rate

Stoc

kap

pre

ciat

ion

Rig

hts

that

repl

icat

eG

ener

ally

have

the

Gen

eral

lyha

veth

eU

nlik

eop

tion

sSA

Rs

righ

tsth

epr

ots

ofst

ock

op-

sam

eta

xtr

eatm

ent

assa

me

tax

trea

tmen

tas

are

mar

ked

tom

arke

tti

ons

Rar

ely

use

d

NQ

SOs

NQ

SOs

each

per

iod

an

dth

esi

nce

stoc

kop

tion

sd

iffe

renc

ebe

twee

nth

eha

vem

ore

favo

rabl

eac

-st

ock

pri

cean

dth

eex

-co

unt

ing

trea

tmen

ter

cise

pri

ceis

exp

ense

d(a

ndop

tion

sp

rovi

de

over

the

outs

tand

ing

the

sam

ebe

ne

tof

per

iod

ofSA

Rs

cash

less

exer

cise

thro

ugh

brok

erlo

ans

)

Not

eW

eth

ank

com

pen

sati

onco

nsul

tant

sSc

ott

Gre

enbe

rgan

dSc

ott

Ols

en(b

oth

ofT

ower

sP

erri

n)an

dFr

eder

icC

ook

(of

Fred

eric

WC

ook

and

Ass

ocia

tes)

for

hel

pfu

lcon

vers

atio

nsin

und

erst

and

ing

the

tax

and

acco

unt

ing

rule

s

14 Hall amp Liebman

The effect of taxation on the optimal contract (more precisely theshare of compensation that is performance-related) is ambiguous (evenin the absence of deductibility and deferral) because there are offsettingeffects First by reducing the share of corporate prots received byshareholders taxes diminish the importance to the shareholders of moti-vating the CEO Taken alone this effect would imply that taxes wouldbe expected to reduce the use of performance-based compensation Sec-ond because the government shares in the income received by the CEOthe variance of the CEOrsquos income is reduced raising his utility andlowering the cost to the rm of providing any given set of incentivesThird because taxes will lead executives to provide less effort for anygiven level of incentive-based pay the level of compensation that mustbe provided to compensate the CEO for effort is reduced for a givenamount of incentive These last two factors make it cheaper for the rmto offer contracts to the CEO with large incentive components when taxrates are higher and should therefore increase the use of incentive-based pay Since the net effect of taxation on the level of incentive-basedpay is ambiguous agency theory provides no strong predictions abouthow taxation should affect the optimal composition of option versuscash compensation

42 The Tax Advantages of DeferralWhile agency theory yields ambiguous predictions about the effect oftaxes on the use of options there are direct tax advantages of options inthat options provide executives with a way to defer compensation andthereby lower their taxes However because options also lead to a defer-ral of corporate tax deductibility the tax advantages from a combined(executive and corporate) perspective are not so straightforward In thissection we analyze and measure the tax advantages of stock optionsrelative to salary and bonus compensation We also analyze the taxadvantages of NQSOs relative to ISOs In particular we show how thetax advantages of options change as personal corporate and capitalgains taxes change We then show how the tax advantages of optionshave changed over time in response to changing tax rates

The crucial tax difference between standard options (NQSOs) andcash payment is that option payouts are deferred and the two forms ofcompensation earn different rates of return over the deferral period Anyanalysis of the relative tax advantages of two compensation instrumentsmust consider the tax consequences both to the employer and to theemployeemdashwhat Scholes and Wolfson (1992) call the global contractingperspective Thus in order to make valid comparisons between the twowe compare the tax burden to the executive while holding constant the

Taxation of Executive Compensation 15

posttax cost to the company By keeping posttax employer costs (in NPVterms) constant any package that is preferred by the employee is tax-advantaged in the global-contracting sense

Under this methodology a comparison of the tax advantages of op-tions and cash involves comparing a pretax cash payment of P with anoption payment that has the equivalent posttax NPV to the companyFor NQSOs it is straightforward to show that a pretax payment of P toan executive has exactly the same cost to the company as putting aside Pfor the purposes of paying stock-option payouts later That is if a com-pany pays an executive P today it will have the same amount of moneyin N years as if it had put aside P today let it accumulate at the rate ofreturn earned by the rm and then paid it out (as compensation in theform of option gains4) with any appreciation in year N

Assume that the pretax prot rate is r and the corporate tax rate is TcThen if the company pays P today it receives a deduction today of PTcwhich yields

PTc [1 1 r (1 2 Tc)]N (1)

in N years since the benets of the tax deduction are invested in thecompany and receive the after-tax corporate rate of return Converselyif the company puts P aside today then it grows at the after-tax corpo-rate rate of return in N years to give a tax deduction of

P [1 1 r (1 2 Tc)]N Tc (2)

which is the exact same value Note that the equivalence of these two taxdeductions is analogous to the tax-benet equivalence of front-loadedand back-loaded IRAs

An executiversquos payoff in N years from option prots is equal to Pcompounded at the after-tax corporate rate times 1 2 Tp where Tp is thepersonal tax rate Thus the combined payoffs in N years of the corporatededuction and the executiversquos posttax payoff is

P [1 1 r (1 2 Tc)]N Tc 1 P [1 1 r (1 2 Tc)]

N (1 2 Tp) (3)

where the rst term is the payoff from the corporate deduction and thesecond term is the executiversquos payoff

4 This analysis is not unique to options Any form of deferred compensation that enablesexecutives to invest inside the rm at a higher posttax return or without paying capitalgains taxes would have similar effects

16 Hall amp Liebman

The payoff to salary and bonus in N years is more complicated sincean assumption must be made about how the executive invests the origi-nal cash compensation For example if the executive invests in an instru-ment with taxable interest (eg bonds) then the N-year payoff from aninvestment of P accumulates at the post-corporate-tax post-personal-taxrate of return and is

P (1 2 Tp) [1 1 r (1 2 Tc)(1 2 Tp)]N (4)

However if an executive invests in (non-dividend-paying) equities theinvestment accumulates at the higher post-corporate-tax rate of returnwith the offsetting disadvantage that the capital gains are taxed at thecapital gains rate in year N So the executive receives

P [1 1 r (1 2 Tc)]N (1 2 Tp) 2 Tcg $P (1 2 Tp) [1 1 r (1 2 Tc)]

N 2P (1 2 Tp) (5)

where Tcg is the capital gains tax rate5Thus putting the equations together if an executive is paid cash P

and invests g in bonds and 1 2 g in equities the N-year payoff (includ-ing the payoff to the corporation from the tax deduction) is

P [1 1 r (1 2 Tc)]N Tc 1 g $(P [1 1 r (1 2 Tc)(1 2 Tp)]

N (1 2 Tp)1(1 2 g) (P [1 1 r (1 2 Tc)]

N (1 2 Tp) 2 Tcg $P (1 2 Tp)[1 1 r (1 2 Tc)]

N 2 P (1 2 Tp)) (6)

The tax advantage of options versus salary and bonus therefore is thedifference between the combined (corporate and executive) payoffs inequation (3) and the combined payoffs in equation (6)

43 The Tax Advantage of OptionsWe now turn to analysis of how changes in various tax rates affectthe tax advantages of options relative to cash compensation We focuson the case in which the cash earned by executives is invested entirelyin equities since equities are tax-favored relative to bonds The impor-tant conclusions of this analysis are not substantively different if we

5 In practice executives hold a combination of equities and bondlike instruments withtaxable interest Because equities are tax-favored in the next section we simplify the analy-sis by assuming that executives hold only equities Alternatively we could have assumedthat executives hold only less risky bonds since bonds more closely match the risk proleof a stream of tax savings None of our empirical results are substantively affected by thissimplication

Taxation of Executive Compensation 17

instead assume that a portion of an executiversquos holdings are invested inbonds

431 The Tax Advantage of Options Changing the Corporate TaxRate Stock options are tax-advantaged relative to cash simply becauseoption payouts are deferred allowing the executive to invest at the pre-personal-tax rate of return (with no capital gains tax at the end) ratherthan investing at the after-tax rate of return or paying the capital gainstax Since the advantage of being able to defer taxes is large when thecorporate tax rate is low the tax advantage of options is larger when thecorporate rate is low

To illustrate this point we dene the tax advantage of options to bethe dollar amount by which total after-tax option payoffs (to the corpora-tion and the executive) exceed the total payoffs from salary as denedby the difference of equation (3) and (6) assuming that P (the paymentto the executive) is equal to $100 and g 5 0 (cash compensation is in-vested in equities) We use a ten-year horizon since most options have aten-year term We then calculate how the tax advantage of optionschanges as the corporate tax rate varies from zero to 100 percent holdingthe personal tax rate and the capital gains rate xed at 40 percent and 20percent respectively (which approximates current rates)

The top panel of Figure 2 shows the tax advantage of options relativeto cashmdashprecisely the NPV of the total tax advantage of options for a$100 payment to the CEOmdashas the corporate tax rate changes As ex-pected the tax advantage of options declines as corporate tax rates in-crease At a zero corporate tax rate the tax advantage of options has anNPV of $7 This value is about $350 at a corporate rate of 40 percentWhen the corporate tax rate is 100 percent the tax advantage of optionscompletely disappears since the benets of deferred compensation fallto zero because the after-tax return on equities (and therefore the dis-count rate) falls to zero The tax advantage of options is the NPV of thetax savings from avoiding the capital gains tax on ten years of apprecia-tion which is why it is relatively modest even at the 40-percent corpo-rate rate

432 The Tax Advantage of Options Changing the Personal TaxRate Using the same assumptions as above (but this time holding thecorporate rate at 40 percent and the capital gains rate at 20 percent andvarying the personal rate) the tax advantage of options declines as thepersonal tax rate rises as shown in the middle panel of Figure 2 The taxadvantage of options is higher at low personal tax rates because theposttax base that gives rise to capital gains is higher and the advantage of

Note Holds constant the personal rate at 40 percent and the capital gains rate at 20 percent

Note Holds constant the corporate rate at 40 percent and the capital gains rate at 20 percent

Note Holds constant the personal rate at 40 percent and the corporate rate at 40 percent

FIGURE 2 Tax Advantage of Options NPV of Tax Advantage of Pay-ing $100 in Options (NQSOs) Rather Than in Cash

Taxation of Executive Compensation 19

avoiding the capital gains taxes is greater The tax advantage of optionstherefore declines linearly as the personal tax rate rises At a 100-percentpersonal tax rate the executive receives nothing in either case and byconstruction the companyrsquos tax deduction benet is the same in NPVterms so the tax advantage of options falls to zero also6

433 The Tax Advantage of Options Changing the Capital Gains TaxRate The tax advantage of options for various capital gains rates isdepicted in the bottom panel of Figure 2 under the same assumptions(this time holding both the personal rate and the corporate rate at 40percent while varying the capital gains rate) The tax advantage of avoid-ing the capital gains tax increases as the capital gains tax rate rises

5 HOW HAS THE TAX ADVANTAGE OF OPTIONSCHANGED OVER TIME

The analysis so far has illustrated how the tax advantage of optionsvaries with the changes in tax rates In this section we show how the taxadvantage of options has changed over time as tax rates have changed

51 Top Marginal Tax Rates over TimeThe evolution on the top marginal tax ratesmdashpersonal corporate andcapital gainsmdashfrom 1980 to 1998 is shown in Figure 3 In all cases thetop rate (the rate for taxpayers with the highest incomes) is shownwhich is not always the highest rate since various anomalies (such asclawbacks of exemptions) sometimes lead to marginal tax rates that arehigher than those for the highest income tax payers

The top personal tax rate was 70 percent in 1980 falling to 50 percentin 1982 and 28 percent in 1988 following the 1986 tax act7 The topmarginal tax rate has since risen to 396 percent but is effectively 425percent since there is a 29-percent Medicare surcharge (paid half by theemployer and half by the employee) that has no upper limit The topcorporate tax rate has had only one signicant change since 1980 fallingfrom 46 percent before the 1986 tax act to 34 percent following the act in1988 The top corporate tax rate was increased to 35 percent in 1993 The

6 This is one place where ignoring bond investments has substantive implications Inparticular in the all-bond case the tax advantage of options is nonmonotonicmdashit rises andthen fallsmdashas personal tax rates rise7 During the early 1980s the maximum tax on earned income limited the marginal tax rateon the earnings of high earners so many executives did not face a decline in personal taxrates between 1981 and 1982 The empirical results in this paper are robust to assumingthat executives faced a marginal tax rate of 50 percent in 1980 and 1981

20 Hall amp Liebman

FIGURE 3 Top Marginal Tax Rates since 1980

top capital gains rate has uctuated between 28 percent (approximately)and 20 percent since 1980 and is currently at 20 percent

52 The Evolution of the Tax Advantage of NQSOsThe tax advantage of options relative to cash dened in the same way asin section 4 from 1980 to 1998 is shown in Figure 4 The calculations aredone using the same assumptions as before (ie N 5 10 years and cashis invested in equities) and the statutory top marginal tax rates shown inFigure 3 are used Due to the fall in corporate and personal tax rates thetax advantage of options increased sharply after the Tax Reform Act of1986 This increase in tax advantage has largely been reversed in the1990s as top personal tax rates have crept back up and the capital gainsrate has been reduced

Two key facts emerge from this analysis First there is currently only amoderate tax advantage to standard nonqualied stock optionsmdashon theorder of $4 per $100 of compensation This is because the tax advantagesto the executive of deferring taxes are largely offset by the tax disad-vantages to the company of not being able to deduct option expensesfrom taxable prots until the executive exercises the options Secondalthough the 1986 tax act substantially increased the tax advantage ofoptions more than half of this increase has been reversed in the 1990s

Taxation of Executive Compensation 21

FIGURE 4 Tax Advantage of Options 1980 to 1998

NPV of tax advantage of $100 in options (NQSOs) rather than cash

53 The Tax (Dis)advantage of ISOsSo far the analysis has focused on the tax advantages of NQSOs sincethey are so much more prevalent than ISOs which account for onlyabout 5 percent of option grants The relative scarcity of ISOs can beexplained both by their tax status and by their per-executive cap of$100000 per year

Under what conditions are ISOs tax-preferred to NQSOs For a trans-fer P to an executive the ISO is always tax-preferred by the executive(since the capital gains rate is lower than the personal rate) while theNQSO is always tax-preferred from the companyrsquos perspective (since anISO is not deductible) The key issue however is the relative advan-tages of the two types of options from a global-contracting perspectiveBecause ISOs are not deductible the company is indifferent betweensetting aside P in the form of NQSOs today and setting aside P(1 2 Tc) inthe form of ISOs (Since the NPV of the tax deduction for NQSOs is PTcthe NPV of the payment P is P 2 PTc or P (1 2 Tc))

Equalizing the posttax cost of NQSOs and ISOs we need only look atthe payoff of the executive to determine the condition under which eachis preferred The payoff of NQSOs in N years is

P [1 1 r (1 2 Tc)]N (1 2 Tp) (7)

22 Hall amp Liebman

The same payoff for ISOs is

P (1 2 Tc) [1 1 r (1 2 Tc)]N (1 2 Tcg) (8)

ISOs therefore are tax-advantaged if

Tp 2 Tcg Tc 2 Tc Tcg (9)

Intuitively ISOs are tax-advantaged only if their advantage (the differ-ence between the personal rate and the capital gains rate) is largeenough to offset their cost (the disadvantage of not deducting at thecorporate rate) Note however that the condition is not a simple com-parison between the corporate rate and the personal rate minus thecapital gains rate (Tp 2 Tcg Tc)

Figure 5 shows how the tax advantage of NQSOs relative to ISOs haschanged since 1980 as tax rates have changed Unlike the very modestchanges (around 3 to 4 percent) in the relative tax advantages betweenoptions and cash shown in Figure 4 the relative tax advantage betweenNQSOs and ISOs has seen enormous swings When the top personalrate was 70 percent in 1980 ISOs were tax-favored by a margin greaterthan 6 percent However by 1982 (following the 1981 tax acts that dra-matically lowered the personal tax rate) ISOs became tax-disadvantaged

FIGURE 5 Tax Advantage of NQSOs Relative to ISOs over Time

NPV of tax advantage of $100 payment to executive NQSOs minus ISOs

Taxation of Executive Compensation 23

by about 5 percent By 1988 (following the 1986 tax act) ISOs became tax-disadvantaged by about 18 percent a dramatic swing The raising ofpersonal rates coupled with a decrease in the capital gains rate has sincereduced the tax disadvantage of ISOs to about $4 per $100 of pretaxcompensation Although hard data on ISOs are hard to come by practi-tioners (mostly compensation consultants) claim that ISOs were morepopular prior to 1982 when they were tax-advantaged Although ISOsare less disadvantageous from a tax standpoint relative to NQSOs thanthey were in the late 1980s they are still disadvantageous so it is notsurprising that they have not made a signicant resurgence in recentyears

6 EXPLAINING THE INCREASE IN OPTION-BASEDCOMPENSATION

In this section we analyze whether changes in tax rates affect the compo-sition of executive compensation and consider the relative importanceof tax factors and corporate governance factors in explaining the in-creasing share of compensation paid in the form of stock options Thenumerous changes in tax rates that have occurred since 1980 provide anopportunity to assess whether executive compensation arrangementsrespond to tax incentives in the way that the tax-avoidance model sug-gests We test this model using a panel data set of CEOs in largepublicly-traded US corporations Our identication of the tax effectsrelies on time-series variation in personal corporate and capital gainstax rates along with cross-sectional variation in corporate tax rates

61 DataWe use a 15-year panel data set of CEOs in the largest publicly-tradedUS corporations which is described in Hall and Liebman (1998) Thedata set covers the years from 1980 through 1994 and combines CEOcompensation information from corporate proxies and 10-K lings withstock price and stock return information from CRSP and accounting datafrom Compustat In addition some compensation data from the 1970swere collected in order to construct measures of the value of stock op-tions held by the CEOs in the rst period of the sample8

The feature that distinguishes our data from most other CEO datasets is that with our panel of yearly proxy data on option grants optiongains and total options held we are able to calculate the total value of

8 We are grateful to David Yermack for providing us with some of the data for 1984 to 1991See Yermack (1995) for a discussion of these data

24 Hall amp Liebman

all stock options held by the CEO at a point of time More importantlysince we have the details about the stock options held (number exer-cise price time to maturity etc) we can precisely calculate the changein the value of a CEOrsquos stock option holdings for a given change in rmvalue

62 Identication IssuesThe numerous tax changes over our sample period and the specic waysin which changes in personal corporate and capital gains rates arepredicted to affect the composition of CEO compensation give us un-usually rich sources of identication However because many of thechanges in tax rates would be expected to affect all of the CEOs in oursample in a similar way we will need to pay particular attention toseparating out the tax effects from underlying time trends and otherfactors that changed over time Controlling for underlying trends is par-ticularly important in this study because we are focusing on an out-comemdashthe use of stock optionsmdashthat has increased rapidly over thepast two decades and that many practitioners believe was importantlyaffected by nontax considerations Thus the tax effects we are trying toexplain are deviations from a rising trend

We take two steps to try to separate out the tax effects from theunderlying trend First we include key nontax factors that could poten-tially explain the increasing reliance on performance-based compensa-tion It has been suggested that the dramatic increase in incentive-basedpay is the result of the remarkably poor shareholder returns during the1970s which spurred the LBO and takeover movements of the 1980sAccording to this story (Kaplan 1997) shareholders became much morepowerful via the rise of institutional investors even as the LBO andtakeover movements waned Because the inuence of institutional inves-tors is thought to be one of the most important mechanisms of strongcorporate governance we use the share of each companyrsquos stock ownedby institutional investors (dened as institutions with more than $100million under management) as an explanatory variable9 The percentageof shares owned by large institutional investors increased from about 20percent to almost 50 percent in 1994 an upward trend that closelymatches the sharp rise in the share of compensation in the form ofoptions Indeed in our sample the annual average (over all of the rmsin our sample) share of stock owned by institutional investors has acorrelation of 09 with the annual average share of compensation paid inoptions

9 This variable is described in Gompers and Metrick (1998)

Taxation of Executive Compensation 25

In addition we include two other variables that proxy for strongercorporate governance The rst is the size of the board Evidence suggeststhat boards with smaller numbers of directors reduce the inuence of theCEO and better represent shareholder interests (Yermack 1995) Boardsize fell by approximately 15 percent between 1980 and 1994 The secondvariable is the fraction of the rmrsquos board members who are inside direc-tors or gray directors (noninsiders who have business dealings with thecompany) Firms with fewer outsiders are less likely to act in the share-holdersrsquo interests and we would predict that they would be less likely tohave performance-based compensation for their CEOs The average per-centage of inside and gray directors has decreased from about 45 percentin 1980 to about 35 percent in 1994

Second we allow for cross-sectional variation in corporate tax ratesCorporate marginal tax rates vary because rms may be eligible for tax-loss carrybacks and carryforwards investment tax credits and the alter-native minimum tax (see Auerbach 1986 Auerbach and Altshuler1990 Auerbach and Poterba 1987) Because these cross-sectional differ-ences imply that different rms should respond differently to a givenchange in tax rates they potentially provide us with a way to isolatethe tax effects Firm-specic corporate marginal tax rates are notori-ously difcult to calculate from publicly-available data We use a tricho-tomous variable that equals zero if the rm has tax-loss carryforwardsand negative earnings in a given year equals 05 times the statutorycorporate rate if the rm has only one of those conditions and equalsthe statutory rate if the rm has neither loss carryforwards nor negativeearnings10

The cross-sectional variation in the corporate tax rate may not be ex-ogenous Firms that perform poorly could face low marginal tax ratesand also be particularly likely or unlikely to provide performance-basedpay Because most of our specications use xed effects the correlationwe must be concerned about is between changes in rm performanceand changes in compensation We deal with this by controlling forlagged rm stock market performance in our regressions This shouldeliminate spurious correlation between tax rates and compensation thatis jointly caused by rm performance

10 Graham (1996) presents evidence that the trichotomous variable performs nearly as wellas his recommended simulated marginal tax rate Plesko (1999) compares marginal taxrates calculated from Compustat data with actual marginal tax rates from rm tax returnsHe nds that simple binary variables are more reliable measures of the marginal tax ratethan is either the Graham or the trichotomous variable We obtain substantively similarresults when we switch from the trichotomous variable to the binary variables recommendby Plesko

26 Hall amp Liebman

63 ResultsThe dependent variable in our regressions is the fraction of a CEOrsquos totalannual compensation (measured as the BlackndashScholes value of stock-option grants plus salary and bonus) that comes from stock-optiongrants We choose this reduced-form specication because our interest isin the composition of CEO compensation However the factors thataffect the composition of compensation might also affect the levels andit will be important to keep this in mind in interpreting our results

Optimal-contracting theory generally makes predictions about the in-centives provided by the total amount of rm stock and stock optionsowned by the CEO We choose to model the ow of option grants ratherthan the ultimate performance sensitivity of the CEOrsquos compensationand stock and stock-option holdings because we have in mind anadjustment-cost model in which it takes time for a CEOrsquos holdings ofstock and stock options to reach the optimal level11 In our regressionswe include two measures of the current performance sensitivity of theCEOrsquos wealth to rm performance as explanatory variables to reect thedistance of the CEOrsquos contract from the optimal contract

Table 3 presents results in which the tax effects are identied solely byvariation in tax rates over time The fraction of annual compensation paidin stock options is regressed on tax variables corporate governance vari-ables contract-theory variables and variables reecting stock market re-turns Both of the regressions in this table include rm xed effects Incolumn (1) the fraction in annual compensation paid in stock options isregressed on the log difference in payoff (the combined payoff to both therm and the CEO) from receiving compensation in stock options ratherthan salary and bonus where the payoff calculations are based on statu-tory marginal tax rates This variable is dened as the log difference be-tween equation (3) and equation (6) in section 4 The coefcient on the taxvariable is large and statistically signicant at the 95-percent level Thepoint estimate implies that a one-percent increase in the payoff differencebetween stock options and salary and bonus results in a 24-percentage-point increase in the share of compensation paid in stock options

The corporate governance variables all have the predicted sign Firmswith a higher fraction of their shares owned by institutional investorsuse more performance-based pay (stock options) Firms with large corpo-rate boards or a large share of inside directors are less likely to use stockoptions

The two contract-theory variables are the dollar change in CEO wealth

11 See Core and Guay (1999) for evidence that rms adjust in this way

TABLE 3Compensation Regressions that Rely on Variation in

Tax Rates over Time

(1) (2)

Tax variablesln (difference in payoff)(a) 2393 (0349)ln (12statutory personal 20015 (0034)

rate)ln (12statutory cap gains 0330 (0171)

rate)ln (12statutory corporate 0399 (0148)

rate)

Corporate governance variablesFraction of rmrsquos shares 0043 (029) 0052 (0028)

owned by institutionalinvestors

Size of the rmrsquos board of 20002 (001) 20003 (0001)directors

Fraction of directors who 20082 (033) 20092 (0030)are inside or gray

Contract-theory variablesD (CEO wealth) per $1000 20261 (0095) 20247 (0087)

in FV(b)

D (CEO wealth) per 10 in 2583 (189) 2554 (201)FV(c)

Volatility(d) 521 (424) 261 (369)ln (rm market value) 0019 (0006) 0019 (0006)

Stock market return variablesLagged annual rm stock 0006 (0009) 0003 (0008)

market returnLagged annual SampP 500 re- 20035 (0023) 20029 (0023)

turn

Regression diagnosticsTotal sample size 5179 5179Firms 427 427Adjusted R2 0313 0311

Dependent variable share of compensation paid in options Standard errors in parentheses Bothregressions in this table include rm xed effects

(a) Log difference in payoff is the log difference in combined rm and CEO after-tax share betweenpaying compensation in options and paying salary

(b) This variable is the dollar change in CEO wealth per $1000 change in rm market value (seeJensen and Murphy 1990)

(c) This variable in the dollar charge in CEO wealth per 10-percent change in rm value (see Hall andLiebman 1998)

(d) Standard deviation of rm market value in trillions

28 Hall amp Liebman

per $1000 change in rm value (Jensen and Murphy 1990) and thedollar change in CEO wealth per 10-percent change in rm value (Halland Liebman 1998) As discussed in Baker and Hall (1998) these twowealth sensitivity measures reect two different concepts of how closelyaligned the managersrsquo incentives are with the interests of the sharehold-ers In the regression they are meant to measure how far the CEOrsquosexisting contract and ownership of stock and stock options is from theoptimal level The negative coefcients on the two variables suggeststhat rms that are below their optimal pay-to-performance sensitivityare more likely to give more stock options in the current period

The second column replaces the tax variable motivated by deferraladvantages with the three statutory rates The corporate after-tax sharehas a positive coefcient as we would expect because the tax advantageof stock options increases as the corporate rate falls The coefcient onthe personal tax rate is small and statistically insignicant The sign ofthe coefcient on the capital gains rate is the opposite of that predictedby tax-avoidance theory

Note that these results rely on time-series variation alone This is not avery convincing test Since the log difference in payoffs mostly rises overthe time period covered in our sample as does the share of compensa-tion paid in options it is possible that the tax variable is simply reectingother factors that were changing over this time period that we have notincluded in our regression Therefore it is important to look at specica-tions that rely on cross-sectional variation in the tax variables

Table 4 contains specications that rely on cross-sectional variation incorporate tax rates The rst column repeats the specication from col-umn (1) of Table 3 but uses the trichotomous measures of rm corporatetax rates in place of the statutory rate in calculating each rmrsquos logdifference in payoff from paying compensation in options rather than insalary and bonus As in the previous table we nd the positive coef-cient on the log difference in payoff just as the theory suggest How-ever the coefcient estimate is now much smaller suggesting that aone-percent increase in the log difference in payoff produces a 01-percentage-point increase in the share of compensation paid in options

Introducing cross-sectional variation in the dependent variable doesnot eliminate the potentially spurious correlation over time between thetax variable and the share of compensation paid in options The secondcolumn of Table 4 isolates the pooled cross-sectional variation in corpo-rate tax rates by adding time dummies and dropping the rm xedeffects In this specication the coefcient on the tax variable is indistin-guishable from zero The coefcient on the institutional variable contin-ues to be large and statistically signicant The third column includes

Taxation of Executive Compensation 29

both rm xed effects and time dummies which tests for a relationshipbetween within-rm changes in the dependent variable and within-rmchanges in each explanatory variable In this specication the coefcienton the tax variable is again indistinguishable from zero The coefcientson the corporate governance variables are also statistically insignicant

The third column contains a specication that treats the Tax ReformAct of 1986 (TRA 86) as a natural experiment For rms and CEOs facingthe statutory corporate tax rate TRA 86 increased the relative payoff topaying compensation in options by 17 percent For a rm facing a zeromarginal tax rate it reduced the relative payoff by 45 percent The reduc-tion occurred because TRA 86 reduced the personal tax rate makingsalary and bonus more attractive while the decreased corporate rate hadno effect on zero-tax-rate rms Thus we would expect rms facing zeroor low marginal tax rates to reduce their use of options after TRA 86while rms facing the statutory rate would be expected to increase themFew rms face zero marginal tax rates year after year Therefore we tryto distinguish between rms that often face low tax rates and rms thatnearly always face the statutory rate We dene low-tax rms as oneswhose average marginal tax rate over 1984 to 1986 was below 02412

Then we ran a difference-in-differences regression to see if high-taxrms increased their use of options more than low-tax rms The interac-tion between post-TRA 86 and high corporate tax rates is the key vari-able The estimated coefcient on this variable is positive but small andnot statistically different from zero at the 95-percent level

It is important to note that in this specication as well the main corpo-rate governance variable the share of the rmrsquos shares owned by institu-tional investors performs as theory would predict The variable appearsto be quite robust Table 5 presents regressions of the share of compensa-tion paid in stock options on the fraction of the rmrsquos shares owned byinstitutional investors Since we have already established that these vari-ables have a strong longitudinal relationship the regressions here aredesigned to focus on the cross-sectional relationship between these twovariables The rst and third columns present regressions using data onlyfrom 1994 and nd a coefcient of roughly 02 (column 1 contains noother covariates while column 3 includes all of the other nontax covari-ates from the regressions in Tables 3 and 4) Since the median share of armrsquos stock owned by institutional investors in our sample rises from 19percent in 1980 to 49 percent in 1994 this coefcient would imply that this

12 This cutoff was chosen after inspecting the distribution of tax rates and essentiallyseparates rms that are always at or near the statutory rate from ones that are sometimesbelow it

TA

BL

E4

Com

pens

atio

nR

egre

ssio

nsth

atR

ely

onC

ross

-Sec

tion

alV

aria

tion

inC

orpo

rate

Tax

Rat

es

(1)

(2)

(3)

(4)

Tax

vari

able

s(a)

ln

(dif

fere

nce

inp

ayof

f)0

099

(00

41)

20

038

(00

36)

20

005

(00

44)

Pos

t-T

RA

862

000

5(0

013

)H

igh

corp

orat

eta

xra

te0

018

(00

17)

(Pos

t-T

RA

86)

3hi

gh0

024

(00

16)

corp

orat

eta

xra

te

Cor

por

ate

gove

rnan

ceva

riab

les

Frac

tion

of

rmrsquos

shar

es0

109

(00

27)

011

4(0

015

)2

000

9(0

030

)0

087

(00

34)

owne

dby

inst

itu

tion

alin

vest

ors

Size

ofth

e

rmrsquos

boar

dof

20

004

(00

01)

20

001

(00

01)

20

002

(00

01)

000

1(0

002

)d

irec

tors

Frac

tion

ofd

irec

tors

wh

o2

012

4(0

033

)2

000

3(0

018

)2

004

1(0

034

)0

001

(00

39)

are

insi

de

orgr

ay

Con

trac

t-th

eory

vari

able

sD

(CE

Ow

ealt

h)p

er$1

000

20

238

(00

94)

20

167

(00

53)

20

257

(00

96)

20

021

(01

47)

inFV

(b)

D(C

EO

wea

lth)

per

10

in2

560

(17

8)2

592

(10

5)2

594

(19

6)2

147

(04

86)

FV(c

)

Vol

atili

ty(d

)2

48(4

24)

162

(03

4)1

12(4

28)

106

(06

09)

ln(

rmm

arke

tva

lue)

003

3(

006)

000

8(0

003

)0

009

(00

06)

001

2(0

007

)

Stoc

km

arke

tre

turn

vari

able

sL

agge

dan

nual

rm

stoc

k2

000

2(0

009

)0

011

(00

10)

000

6(0

009

)2

001

2(0

013

)m

arke

tre

turn

Lag

ged

annu

alSamp

P50

02

005

6(0

023

)2

005

3(0

051

)2

003

9(0

049

)2

015

4(0

040

)re

turn

Reg

ress

ion

dia

gnos

tics

T

otal

sam

ple

size

5179

5179

5179

2534

Firm

s42

742

739

6A

dju

sted

R2

031

009

032

005

Firm

xed

effe

cts

Yes

No

Yes

No

Tim

ed

um

mie

sN

oY

esY

esN

o

Dep

end

ent

vari

able

isfr

acti

onof

ann

ualc

omp

ensa

tion

from

stoc

kop

tion

sSt

and

ard

erro

rsin

par

enth

eses

(a)

All

thre

esp

eci

cati

ons

use

tric

hoto

mou

sco

rpor

ate

tax

rate

sT

he

sam

ple

inco

lum

n(3

)u

ses

only

the

year

s19

84to

1988

sin

cew

ear

eu

sing

1986

asa

nat

ural

exp

erim

ent

(b)

Thi

sva

riab

leis

the

dol

lar

chan

gein

CE

Ow

ealt

hp

er$1

000

chan

gein

rm

mar

ketv

alu

e(s

eeJe

nse

nan

dM

urp

hy

1990

)

(c)

Thi

sva

riab

lein

the

dol

lar

char

gein

CE

Ow

ealt

hp

er10

-per

cent

chan

gein

rm

valu

e(s

eeH

alla

nd

Lie

bman

199

8)

(d)

Stan

dar

dd

evia

tion

ofr

mm

arke

tva

lue

intr

illio

ns

32 Hall amp Liebman

TABLE 5Large Institutional Investors and Stock-Option Compensation

The Cross-sectional Relationship between Institutional Investors andOption Grants

(1) (2) (3) (4)

Coefcient on fraction of 217 (067) 145 (014) 177 (069) 120 (015)rmrsquos shares owned byinstitutional investors

Sample period 1994 1980ndash1994 1994 1980ndash1994Contract-theory and stock No No Yes Yes

market covariatesTime dummies No Yes No Yes

The dependent variable is the share of compensation paid in options

factor alone can explain 6 percentage points of the 23-percentage-pointincrease in the median share of compensation paid in options that oc-curred over this time period Of course this variable is only one measureof the strengthening of corporate governance so the total impact of im-proved corporate governance could be greater The second and fourthcolumns of the table use the entire 1980ndash1994 sample and include timedummies to isolate the cross-sectional correlations Again there is evi-dence of a strong cross-sectional relationship between the two variablesalthough the coefcients are slightly smaller perhaps reecting a weakercorporate-governance role in the earlier period In sum there is a strongrelationship in both longitudinal and cross-sectional data between theinstitutional-investor variable and the options-percentage variable13

7 THE MILLION-DOLLAR RULE

In 1993 section 162(m) of the Internal Revenue Code was enactedlimiting the deductibility of executive compensation in excess of onemillion dollars unless the compensation was performance-related Pro-ponents of this legislation argued that it would slow the growth ofexecutive pay and tighten the link between rm performance and execu-tive compensation

The new provision became effective for tax year 1994 and applies tothe CEO and the other four most highly compensated executives in eachcompany Pay that is performance-related such as stock-option grants

13 The one specication in which the fraction owned by institutional investors is notsignicant is in column (2) of Table 4 a regression that includes both xed effects and timeeffects

Taxation of Executive Compensation 33

and bonuses paid for meeting clear performance goals is not affected bythis regulation provided that it has been approved by shareholdersThus the main impact of the provision is to limit the deductibility ofsalary in excess of a million dollars Woodlock and Antenucci (1997)studied the proxy statements of 376 rms and documented that mostrms responded to the new law by qualifying top executive pay asperformance-related Perry and Zenner (1999) identied 25 rms thatreduced salaries from above $1 million to below $1 million and foundthat 23 of the rms cited section 162(m) as their reason for doing so

In this section of the paper we investigate whether this provision hasaffected the level or structure of executive pay using the Execucompdata base which contains information on the highest-paid executives ineach of the rms in the SampP 500 SampP Midcap 400 and SampP SmallCap600 Between 1993 and 1998 the median salary of the Execucomp execu-tives rose from $230000 to $306000 The median value of total compen-sation (salary bonus and option grants) rose from $470000 to $882000Limiting the sample of CEOs indicates that the total compensation of themedian CEO rose from $11 million to $18 million over the period whilethe median salary rose from $441000 to $550000 Thus executive com-pensation clearly continued to rise at a rapid rate after the implementa-tion of the million-dollar rule though perhaps at a slower rate than itotherwise would have

If the regulation affects behavior it should have the largest impact onrms whose executives are receiving salaries that are above the limit orjust below it Thus we might expect to see rms whose executives arereceiving relatively high salaries to rely more heavily on performance-based pay and less heavily on salary in the years following the regulationPerry and Zenner (1999) and Livingston (1999) both nd that this is thecase This trend can be clearly identied in the Execucomp data set Forexample over 1993 to 1998 the median annualized growth rate of salaryfor executives whose 1993 salary was below $500000 in the previous yearwas 85 percent For executives with salary between $500000 and$700000 it was 64 percent for executives between $700000 and $850000it was 50 percent for executives between $750000 and $1 million it was31 percent and for executives above $1 million it was 00 percent14

The observation that executives at lower levels of salary had highergrowth rates is not very strong evidence of an impact of section 162(m)however because it is possible that low-salary executives would havehad higher growth rates even in the absence of the provision To explore

14 Of the 3137 executives for whom we have complete data between 1993 and 1998 2708had 1993 salaries below $500000 while only 35 had 1993 salaries above $1 million

34 Hall amp Liebman

whether this is the case we use Execucomp data from both before andafter the law change Ideally we would want to have data from manyyears before the change to examine whether there is an underlying pat-tern of faster salary growth for low-salary executives Unfortunately theExecucomp data begin in 1992 and therefore provide us with only the1992ndash1993 growth rate in compensation to use as a baseline15

We run regressions pooling data from 1992ndash1993 through 1997ndash1998of the annual percentage change in different forms of compensation on avariable million that is designed to measure the likelihood that anexecutiversquos compensation will be affected by section 162(m) We denemillion as the minimum of 1 and the executiversquos salary divided by $1million in the previous year Thus any executive with salary at or above$1 million would receive a value of 1 while an executive with a salary of$500000 would be coded as a 0516 We include this variable by itself andalso interacted with a dummy variable (after) that equals 1 for years1994 and after Thus the million variable is designed to pick up anyunderlying relationship between the level of salary and the growth rateof salary while the interacted one will identify any differential growthrate of salary by income in the period after section 162(m) took effect Wealso include a full set of year dummies the natural logarithm of thermrsquos market value and the rmrsquos annual rate of return over the twoprevious years as control variables To reduce the sensitivity of our re-sults to outliers we run quantile regressions and robust regressions

Table 6 contains our results The rst four columns are for all topexecutives while the last four columns focus on CEOs Columns (1) and(2) present estimates from robust regression for the percentage change insalary and the percentage change in total compensation (salary bonusstock grants and stock-option grants) In the salary growth regressionsfor all executives the coefcient on million is negative and signicantindicating that there is an underlying relationship between the level ofexecutive compensation and its subsequent growth rate The coefcienton million 3 after is negative as well suggesting that the negativerelationship between the level of salary and its subsequent growth rateintensied slightly after 1994 The relatively small magnitude suggeststhat an executive with a $1 million dollar salary would see his salarygrow at an annual rate that would be about 06 percent slower than anexecutive earning $500000

15 The HallndashLiebman data set contains only a combined salary and bonus variable makingit impossible to use it to analyze this issue16 We also ran the regressions (not reported) allowing for a nonlinear relationship betweenmillion and compensation growth and obtained substantively similar results

TA

BL

E6

The

Eff

ect

ofth

eM

illi

on-D

olla

rR

ule

onSa

lary

and

Tot

alC

ompe

nsat

ion

All

top

exec

uti

ves

CE

Os

only

Rob

ust

regr

essi

onQ

uan

tile

regr

essi

onR

obu

stre

gres

sion

Qu

anti

lere

gres

sion

D

in

Din

D

in

Din

D

in

Din

D

in

Din

sala

ryto

talc

omp

(a)

sala

ryto

talc

omp

sa

lary

tota

lcom

p

sala

ry

tota

lcom

p

mil

lio

n(b

)2

006

02

035

72

006

32

024

12

008

72

023

12

010

32

024

5(0

005

)(0

032

)(0

006

)(0

027

)(0

010

)(0

073

)(0

011

)(0

056

)m

illi

on

32

001

10

026

20

021

20

015

20

016

014

92

003

40

158

aft

er(c

)(0

005

)(0

033

)(0

006

)(0

029

)(0

010

)(0

076

)(0

011

)(0

057

)N

359

5035

950

359

5035

950

757

17

589

757

17

589

Stan

dar

der

rors

inp

aren

thes

esR

egre

ssio

nsal

soin

clu

de

year

du

mm

ies

stoc

km

arke

tre

turn

and

log

mar

ket

valu

e

(a)

Tot

alco

mp

ensa

tion

incl

ud

essa

lary

bon

us

rest

rict

edst

ock

and

stoc

k-op

tion

gran

ts

(b)

mil

lion

isd

ene

das

min

(1s

alar

yin

pre

viou

sye

ars

div

ided

by$1

mil

lion)

(c)

aft

eris

anin

dic

ator

vari

able

equ

alto

zero

befo

reth

ere

gula

tion

took

plac

e(1

993)

and

one

afte

rwar

d

36 Hall amp Liebman

In the second column the coefcient on million 3 after is essentiallyzero suggesting that any decrease in salary brought about by section162(m) was offset by increases in bonus and stock-option grants Theresults are similar for the median regressions There is some evidence ofa very small slowdown in growth of salary for executives likely to beaffected by section 162(m) but no sign of a change in total compensa-tion Taken together this provides evidence of a very small substitutionof performance-related pay for salary

The last four columns run identical tests but limit the sample to CEOsThe interaction of million 3 after continues to be small and negative inthe salary regressions In contrast the impact of 162(m) on total compen-sation appears to be positive and fairly substantial (about 15 percentagepoints per year) in these regressions That is the increase in bonus andstock options following the million-dollar rule more than offset the verysmall decline in salary

All of these conclusions should be interpreted somewhat cautiouslybecause our data provide us with only one pre-1993 control year Never-theless the data are consistent with section 162(m) having led to a veryminor slowdown in salary growth and one that is dramatically smallerthan what one would conclude without controlling for underlying differ-ential salary growth rates at different levels of salary Since total compen-sation for CEOs did not decline and perhaps increased there is evidenceof a minor substitution of performance-related pay for salary in responseto the regulation

8 EXECUTIVE COMPENSATION AND THEELASTICITY OF TAXABLE INCOME

Even if taxes have only a minor effect on executive compensation thestructure of executive compensation is important for tax policy be-cause high-income taxpayers are sources of signicant revenue and arethe focus of important debates about the efciency cost of taxation

Goolsbee (1999) has recently argued that executive decisions to exer-cise options are highly responsive to intertemporal differences in taxrates created by tax reforms He claims that much of the apparent declinein taxable income for high-income taxpayers between 1992 and 1993documented by Feldstein and Feenberg (1996) can be explained by execu-tives shifting option gains into 1992 in order to avoid the higher marginaltax rates that went into effect in 1993 By extension he implies that thehigh elasticities of taxable income with respect to personal tax ratesestimated by comparing the pre- and post-tax-reform taxable income of

Taxation of Executive Compensation 37

high-income individuals reect intertemporal shifting of income and notpermanent effects of taxation If this is true then the deadweight loss oftaxation is lower as is the cost of progressivity

While Goolsbee presents comprehensive evidence that option gainswere unusually high in 1992 his identication of the impact of tax rates onthe timing of option gains depends on correctly specifying a counter-factual level of options that would have been exercised in the absence ofthe tax change He simply assumes a linear time trend The purpose ofGoolsbeersquos time trend is presumably to account for the increasing use ofoptions over time However since options typically require a vestingperiod of at least a couple of years a much more direct measure of ex-pected option gains can be constructed based on the total value of vestedoptions held by the CEO Since options typically have vesting periods thatexceed two years this total can be treated as exogenous since it will not beaffected by the tax changes

More generally the value of options held by the CEO is likely to havea major effect on his exercise decisions If his options are out of themoney he will not have gains to realize If there has been a recent run-up in the market an optimizing CEO may exercise an unusually largeamount of options in order to diversify his overall portfolio and if in-the-money options are about to expire they will be exercised in thecurrent year

Our evidence suggests that much of the apparent tax shifting caninstead be attributed to stock market performance and the timing of pastoption grants Using our 15-year panel we replicate Goolsbeersquos (1999)result for the 1993 period but show that the impact of taxes is notpresent in other time periods Moreover even the result for the 1993period disappears when appropriate controls for past option grants andstock price appreciation are included

The top three rows of Table 7 present data on the taxable incomes ofCEOs in the HallndashLiebman sample For these results our panel has beennarrowed to include only CEOs whose rmrsquos scal years correspondwith tax years In addition we create a balanced panel for each of thethree tax-reform periods by excluding any rm for which there are notcomplete data for the period surrounding the reform

The fourth row of the table shows that the 1993 tax act raised the topmarginal personal income tax rate from 310 to 396 percent effective fortax year 1993 with a further increase occurring in 1994 due to the uncap-ping of the Medicare payroll tax The top row of the table shows thattaxable income was higher in 1992 than for any other year in our sample(all data are in 1994 dollars) Option gains were particularly high in1992mdashmore than double their 1991 level and 68 percent higher than their

TA

BL

E7

Tax

Ref

orm

Sto

ckM

arke

tP

erfo

rman

cea

ndth

eT

imin

gof

Opt

ion

Gai

ns

Ave

rage

stoc

km

arke

tM

ean

CE

OM

ean

CE

Ore

turn

for

Mea

nva

lue

taxa

ble

sala

ryan

dM

ean

CE

OT

opfe

der

al

rms

inof

opti

ons

(Val

ue

ofop

tion

Sam

ple

inco

me

bonu

sop

tion

gain

sp

erso

nal

sam

ple

for

held

atbe

gga

ins)

(va

lue

Ref

orm

size

Yea

r($

mil

)($

mil

)($

mil

)M

TR

pre

vye

ar(

)of

year

ofop

tion

she

ld)

ER

TA

8127

919

800

874

064

80

163

7025

80

460

354

1981

083

10

672

009

869

245

070

61

3919

820

872

067

90

099

5011

90

686

144

1983

099

00

724

025

250

355

116

62

16

TR

A86

295

1985

119

70

837

024

050

76

116

02

0719

861

448

091

70

345

5042

81

776

194

1987

165

50

959

059

338

519

12

229

266

1988

163

21

058

039

028

07

237

71

6419

891

656

105

20

414

2817

92

345

176

1993

272

1990

209

51

028

061

828

232

288

02

15T

axA

ct19

911

875

102

00

547

312

01

268

82

0319

922

678

107

01

250

3146

64

552

275

1993

232

41

209

074

339

620

64

299

173

1994

196

41

262

037

842

515

45

113

074

Sam

ple

isH

allndash

Lie

bman

rm

sw

ith

Dec

embe

r

scal

year

sth

atha

dco

mp

lete

dat

afo

ron

eof

the

thre

eta

x-re

form

per

iod

sst

ud

ied

Taxation of Executive Compensation 39

1994 level These patterns are consistent with Goolsbeersquos claim that tax-payers responded to the anticipated increase in marginal tax rates byshifting gains into 1992

The evidence from the other two tax reforms is less clear suggestingthat there might be factors other than tax rates determining the timing ofoption gains ERTA 1981 reduced the top marginal tax rate from 70percent in 1980 to 50 percent in 1982 However there is no sign thatoption gains were reduced in 1980 and 1981 and then increased in 198217

There was a big increase in option gains in 1983 however Similarly thehighest year of stock-option gains in the period around 1986 was 1987 astrange year to take gains since taxpayers would have known that thepersonal rate would be 10 percentage points lower if they waited untilthe following year18

What all the peak years of option gains have in common is that theyfollow years of strong stock market performance The fth row of Table 7shows the average stock market return during the previous year for therms in our sample The large value of option gains in 1983 followed a355-percent increase in the stock market The 1992 boom in option gainsfollowed a 466-percent increase in the stock market the more modest1987 peak in option gains followed a 19-percent gain (and a 43-percentgain two years before)

Strong stock market performance raises the value of options availableto exercise The sixth row of the table shows the mean value of optionsheld by CEOs at the beginning of the year All three of the peak years ofoption gains correspond with peak years in terms of the value of optionsavailable for exercise Since stock market appreciation during the yearalso increases the value of options available to exercise later in that yearwe need to allow for current-year stock market appreciation as well

Table 8 contains Goolsbee-style regressions for each of the tax re-forms The rst two columns reprint the results from Goolsbee (1999)The third and fourth columns replicate his results in our sample Inparticular with a specication similar to his we are able to replicate hisnding that the current after-tax share has a positive coefcient and thefollowing yearrsquos after-tax share has a coefcient of roughly negative oneColumns (5) through (8) show that this pattern was unique to the 1993tax reform Indeed in the TRA 86 regressions the coefcient on the

17 As we explained above for executives covered by the maximum tax on earned incomethere was not a decline in tax rates between 1981 and 198218 It is possible that the large stock market decline in October 1997 may have promptedsome people to exercise options earlier than they otherwise would have However anincrease in option exercises typically follows stock market increases not declines

TA

BL

E8

Att

empt

sto

Rep

lica

teth

eT

ax-S

hift

ing

Res

ult

inO

ther

Tim

eP

erio

ds

Goo

lsbe

eG

ools

bee

1993

1993

Ent

ire

Ent

ire

regr

essi

onre

gres

sion

refo

rmre

form

TR

A86

TR

A86

ER

TA

81E

RT

A81

1980

ndash199

419

80ndash1

994

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

ln[1

2ta

x(t)

]1

244

108

20

455

032

62

017

62

017

22

016

72

016

70

026

20

068

(01

34)

(01

27)

(05

37)

(05

42)

(02

10)

(02

09)

(00

90)

(00

91)

(00

67)

(00

65)

ln[1

2ta

x(t1

1)]

20

867

20

764

21

214

20

917

047

80

722

20

103

004

30

176

018

3(0

118

)(0

112

)(0

491

)(0

499

)(0

157

)(0

174

)(0

072

)(0

072

)(0

068

)(0

066

)

Tim

e0

135

007

62

000

52

036

40

041

000

80

094

003

50

056

002

7(0

008

)(0

008

)(0

049

)(0

049

)(0

028

)(0

028

)(0

025

)(0

026

)(0

003

)(0

003

)

ln(m

arke

tva

lue)

066

10

489

023

70

334

039

5(0

015

)(0

058

)(0

039

)(0

042

)(0

017

)

Yea

rs19

91ndash1

995

1991

ndash199

519

90ndash1

994

1990

ndash199

419

85ndash1

989

1985

ndash198

919

80ndash1

983

1980

ndash198

319

80ndash1

994

1980

ndash199

4

Ad

just

edR

20

720

770

610

630

690

760

800

810

630

68

Obs

erva

tion

s13

856

132

371

385

136

41

470

143

31

113

106

23

968

385

9

Dep

end

ent

vari

able

isth

elo

gof

taxa

ble

inco

me

All

regr

essi

ons

con

tain

rm

xe

def

fect

sG

ools

bee

regr

essi

on(2

)al

soco

ntai

nstw

oot

her

vari

able

sno

tsh

own

inth

eta

ble

Stan

dar

der

rors

inp

aren

thes

es

Taxation of Executive Compensation 41

TABLE 9Alternative Explanations for the Timing of Option Exercises

1993 Entirereform TRA 86 ERTA 81 1980ndash1994

(1) (2) (3) (4)

ln[12tax(t)] 0108 0111 20051 0201(0324) (0136) (0048) (0057)

ln[12tax(t11)] 20134 0573 0125 0150(0336) (0156) (0053) (0066)

ln(market value) 0418 0177 0338 0418(0061) (0039) (0046) (0163)

Firmrsquos stock market return (t) 0298 0268 0214 0258(0044) (0045) (0031) (0023)

Firmrsquos stock market return 0162 0206 0031 0128(t21) (0041) (0008) (0029) (0021)

ln(value of options held by 0004 0008 0001 0009CEO at start of year) (0003) (0003) (0002) (0001)

Years 1990ndash1994 1985ndash1989 1980ndash1983 1980ndash1994Adjusted R2 064 071 082 068Observations 1349 1427 1036 3812

The dependent variable is the log of taxable income All regressions contain rm xed effects Stan-dard errors in parentheses

following yearrsquos tax rate is positive and signicant Columns (9) and (10)pool all three reforms and fail to replicate the pattern that would reectintertemporal shifting19

It is possible that the ease with which one can shift the timing ofoption gains is asymmetric It may be harder to shift option gains for-ward in time since some options expire and therefore must be exercisednow In addition options were a much smaller portion of compensationin the early 1980s so perhaps the ERTA 81 results are not surprisingNonetheless these results cast doubt on the proposition that this is thechannel through which spurious estimates of the elasticity of taxableincome have occurred for previous tax reforms and they suggest thatwe should look harder for alternative explanations for the timing ofoption gains

Table 9 attempts to predict options gains using the lagged and currentincrease in a rmrsquos market value and the value of the options held by theCEO at the start of the year The two stock market variables stronglypredict option gains while the value of options held has the correct sign

19 The results also fail to suggest large permanent effects of marginal tax rates on taxableincome

42 Hall amp Liebman

in all of the specications and is signicant in the pooled sample eventhough it is highly correlated with the two stock market return variablesIn contrast the tax-timing explanation fails even in the 1993 period

It is important to realize that our results do not invalidate Goolsbeersquoscritique of claims that aggregate revenue data from 1992 and 1993 provethat elasticities of taxable income are large We have simply given analternative explanation for why taxable incomes were unusually high in1992 However our results do cast doubt on the conclusion that largetiming shifts are ubiquitous and capable of explaining away the morecareful estimates of taxable income elasticities that have been performedusing micro data (eg Feldstein 1995 Auten and Carroll 1995)

One further point is in order It is sometimes argued that the combina-tion of increased use of options and the booming stock market explainswhy federal tax revenues have been so much higher than expected dur-ing the 1990s However while large option gains increase revenue fromthe personal income tax they produce offsetting deductions for cor-porations The net impact on revenue is only the difference in the twomarginal tax rates times the option gain While the gap between thepersonal rate and the effective corporate rate is nontrivial the impact oflarge option exercises on total revenue is substantially smaller than theamount by which the exercises inate personal tax revenue and reducecorporate tax revenue

9 CONCLUSION

We have described the tax rules for executive pay in detail and analyzedhow changes in various tax rates affect the tax advantages of stock optionsrelative to salary and bonus We nd that there is a moderate tax advan-tage to optionsmdashcurrently about $4 per $100 in pretax executive paymdashand that changes in the tax advantage of options over time have had atmost a modest impact on the composition of pay Corporate governancefactors particularly the role of large investors appear to be more impor-tant in explaining the dramatic increase in option pay

In addition our evidence suggests that more direct attempts to use taxpolicy to inuence executive compensation have had little effect Wend that the million-dollar rule led companies to substitute perfor-mance-based pay for salary But our evidence suggests that this substitu-tion was quite modest and there is no evidence that the total level of paywas reduced Overall although the stock-option explosion has dramati-cally increased the link between pay and performance this change isdue almost entirely to nontax factors

Taxation of Executive Compensation 43

REFERENCES

Auerbach Alan J (1996) ldquoThe Dynamic Effects of Tax Law Asymmetriesrdquo Re-view of Economic Studies 53451ndash405

mdashmdashmdash and James Poterba (1987) ldquoTax Loss Carryforwards and Corporate TaxIncentivesrdquo In The Effects of Taxation on Capital Accumulation M Feldstein(ed) Chicago University of Chicago Press

Altshuler Rosanne and Alan Auerbach (1990) ldquoThe Signicance of Tax LawAsymmetries An Empirical Investigationrdquo Quarterly Journal of Economics 10561ndash86

Auten Gerald and Robert Carroll (1995) ldquoThe Effect of Income Taxes on House-hold Behaviorrdquo Ofce of Tax Analysis US Department of the Treasury

Baker George P and Brian J Hall (1998) ldquoCEO Incentives and Firm SizerdquoNBER Working Paper no 6868 December

Core John and Wayne Guay (1999) ldquoThe Use of Equity Grants to ManageOptimal Equity Incentive Levelsrdquo Draft

Feldstein M (1995) ldquoThe Effect of Marginal Tax Rates on Taxable Income APanel Study of the 1986 Tax Reform Actrdquo Journal of Political Economy 103 (no3)551ndash72

mdashmdashmdash and D Feenberg (1996) ldquoThe Effect of Increased Tax Rates on TaxableIncome and Economic Efciency A Preliminary Analysis of the 1993 Tax RateIncreasesrdquo In Tax Policy and the Economy vol 10 James M Poterba (ed)Cambridge MA MIT Press

Gompers Paul A and Andrew Metrick (1998) ldquoInstitutional Investors andEquity Pricesrdquo NBER Working Paper no 6723

Goolsbee Austan (1999) ldquoWhat Happens When You Tax the Rich Evidencefrom Executive Compensationrdquo NBER Working Paper no 6333

Graham John R (1996) ldquoProxies for the Corporate Marginal Tax Raterdquo Journal ofFinancial Economics 42187ndash221

Hall Brian J (1999) ldquoThe Design of Multi-year Stock Option Plansrdquo Journal ofApplied Corporate Finance 12(2)97ndash106

mdashmdashmdash and Jeffrey B Liebman (1998) ldquoAre CEOs Really Paid Like Bureau-cratsrdquo Quarterly Journal of Economics CXII(no 3)653ndash691

Jensen Michael C and William M Meckling (1976) ldquoTheory of the Firm Mana-gerial Behavior Agency Costs and Ownership Structurerdquo Journal of FinancialEconomics III305ndash360

mdashmdashmdash and Kevin J Murphy (1990) ldquoPerformance Pay and Top-ManagementIncentivesrdquo Journal of Political Economy XCVIII225ndash264

Kaplan Steven N (1997) ldquoThe Evolution of US Corporate Governance WeAre All Henry Kravis Nowrdquo Draft

Livingstone Jane R (1999) ldquoExecutive Compensation Contracting and Re-sponses to Increased Tax Costsrdquo Louisiana State University June

Miller Merton H and Myron S Scholes (1982) ldquoExecutive CompensationTaxes and Incentivesrdquo In Financial Economics Essays in Honor of Paul CootnerWilliam F Sharpe and Cathryn M Cootner (eds) Princeton NJ Prentice-Hall

Murphy Kevin J (1999) ldquoExecutive Compensationrdquo forthcoming in OrleyAshenfelter and David Card (eds) Handbook of Labor Economics Vol 3 NorthHolland

Perry Todd and Marc Zenner (1999) ldquoPay for Performance Government Regu-

Taxation of Executive Compensation 11

Stock options do however lead to expense recognition if they are dis-counted (in the money at grant date) or if the exercise price and numberof options are not known at grant date For discount options the differ-ence between the stock price and the market price is expensed over thevesting period For options with variable terms (eg a variable exerciseprice vesting that is tied to performance) the options are marked tomarket and expensed during the time between grant and exercise Practi-tioners claim that the accounting treatment of options plays an impor-tant role in the design of option programs Thus plans that have ldquobadaccountingrdquo but are thought by many to have attractive incentive fea-tures are often not even seriously considered by companies Examples ofsuch potentially attractive plans include indexed options (where a CEOprots only if his rmrsquos share price grows relative to some market orindustry benchmark) and option grants that are explicitly performance-related both of which would lead to expenses against earnings

Unlike options restricted stock and SARs do not generally receivefavorable accounting treatment Restricted stock is generally expensedover the period in which the restrictions lapse (usually the vesting pe-riod) The magnitude of the expense is the difference between the cur-rent stock price and the executiversquos cost (if any) SARs are marked tomarket each period and the difference between the stock price and theexercise price is expensed over the outstanding period of SARs

4 THE TAXATION OF EXECUTIVE PAY CASHVERSUS OPTIONS

Stock-based compensation performs two roles in executive compensa-tion arrangements First it helps align the incentives of the executivewith the interests of the rmrsquos shareholders Second it often enables therm to compensate the CEO in a way that is more advantageous from atax standpoint than paying the executive in salary and bonus

41 Agency Theory and Executive Compensation in the Presenceof TaxationIn standard agency-theory models ( Jensen and Meckling 1976) agencycosts are the result of the separation of ownership and control Managersdo not have the same incentives as the owners The optimal incentivecontract for managers balances the benets of high-powered incentives(linking the fortunes of owners and managers through stock and stockoptions for example) with the costs of loading too much risk on risk-averse managers

TA

BL

E2

Tax

and

Acc

ount

ing

Tre

atm

ent

ofE

xecu

tive

Com

pens

atio

nIn

cent

ive

Stoc

kO

ptio

ns(I

SOs)

Res

tric

ted

Stoc

kan

dSt

ock

App

reci

atio

nR

ight

s(S

AR

s)

Tax

atio

nat

Tax

atio

nat

Acc

ount

ing

Com

pen

sati

onD

escr

ipti

onp

erso

nall

evel

corp

orat

ele

vel

trea

tmen

t

Ince

nti

vest

ock

ISO

sar

eid

enti

calt

oIf

the

usu

alco

ndit

ions

Ifth

eu

sual

cond

itio

nsA

sw

ith

NQ

SOs

ther

eop

tion

sN

QSO

sin

thei

rd

esig

n

are

met

th

ere

isno

tax

are

met

the

reis

gene

r-is

gene

rally

noex

pen

seH

owev

ert

hey

have

anat

gran

tor

exer

cise

At

ally

no

corp

orat

eta

xd

e-re

cogn

itio

nfo

rIS

Os

annu

alca

pof

$100

000

sale

all

opti

onp

ro

tsd

uct

ion

atgr

ant

exer

-(e

ver)

on

the

amou

ntth

atan

dsu

bseq

uen

tca

pita

lci

seo

rsa

le

vest

sT

he

tax

trea

t-ap

prec

iati

onar

eta

xed

men

tof

ISO

sal

soat

the

cap

ital

gain

sd

iffe

rs

rate

Res

tric

ted

stoc

kSi

mila

rto

stoc

kop

-T

hein

div

idu

alis

taxe

dT

heco

mp

any

gene

rall

yR

estr

icte

dst

ock

isge

n-ti

ons

but

shar

esat

the

ord

inar

yra

teas

rece

ives

ap

aral

leld

e-er

ally

exp

ense

dov

er(w

hich

vest

slow

ly)

the

rest

rict

ions

lap

se

du

ctio

neq

ual

toth

eth

eve

stin

gp

erio

dT

hera

ther

than

opti

ons

are

How

ever

the

man

ager

amou

ntof

the

exec

u-

amou

nt

exp

ense

dis

gran

ted

toth

eex

ecu

-m

aych

ose

tobe

taxe

dti

versquos

inco

me

whe

nth

eth

ed

iffe

renc

ebe

twee

nti

veU

nlik

eop

tion

sat

gran

td

ate

inw

hich

exec

uti

veis

taxe

dR

e-th

ecu

rren

tst

ock

pric

eth

esh

ares

typi

cally

case

alls

ubs

equ

ent

ap-

stri

cted

stoc

kis

gen

er-

and

the

exec

uti

versquos

have

voti

ngan

dd

ivi-

pre

ciat

ion

ista

xed

atal

lysu

bjec

tto

the

cost

(if

any)

d

end

righ

ts

the

cap

ital

gain

sra

te

mill

ion-

dol

lar

cap

D

ivid

end

sar

eta

xed

atth

eor

din

ary

rate

Stoc

kap

pre

ciat

ion

Rig

hts

that

repl

icat

eG

ener

ally

have

the

Gen

eral

lyha

veth

eU

nlik

eop

tion

sSA

Rs

righ

tsth

epr

ots

ofst

ock

op-

sam

eta

xtr

eatm

ent

assa

me

tax

trea

tmen

tas

are

mar

ked

tom

arke

tti

ons

Rar

ely

use

d

NQ

SOs

NQ

SOs

each

per

iod

an

dth

esi

nce

stoc

kop

tion

sd

iffe

renc

ebe

twee

nth

eha

vem

ore

favo

rabl

eac

-st

ock

pri

cean

dth

eex

-co

unt

ing

trea

tmen

ter

cise

pri

ceis

exp

ense

d(a

ndop

tion

sp

rovi

de

over

the

outs

tand

ing

the

sam

ebe

ne

tof

per

iod

ofSA

Rs

cash

less

exer

cise

thro

ugh

brok

erlo

ans

)

Not

eW

eth

ank

com

pen

sati

onco

nsul

tant

sSc

ott

Gre

enbe

rgan

dSc

ott

Ols

en(b

oth

ofT

ower

sP

erri

n)an

dFr

eder

icC

ook

(of

Fred

eric

WC

ook

and

Ass

ocia

tes)

for

hel

pfu

lcon

vers

atio

nsin

und

erst

and

ing

the

tax

and

acco

unt

ing

rule

s

14 Hall amp Liebman

The effect of taxation on the optimal contract (more precisely theshare of compensation that is performance-related) is ambiguous (evenin the absence of deductibility and deferral) because there are offsettingeffects First by reducing the share of corporate prots received byshareholders taxes diminish the importance to the shareholders of moti-vating the CEO Taken alone this effect would imply that taxes wouldbe expected to reduce the use of performance-based compensation Sec-ond because the government shares in the income received by the CEOthe variance of the CEOrsquos income is reduced raising his utility andlowering the cost to the rm of providing any given set of incentivesThird because taxes will lead executives to provide less effort for anygiven level of incentive-based pay the level of compensation that mustbe provided to compensate the CEO for effort is reduced for a givenamount of incentive These last two factors make it cheaper for the rmto offer contracts to the CEO with large incentive components when taxrates are higher and should therefore increase the use of incentive-based pay Since the net effect of taxation on the level of incentive-basedpay is ambiguous agency theory provides no strong predictions abouthow taxation should affect the optimal composition of option versuscash compensation

42 The Tax Advantages of DeferralWhile agency theory yields ambiguous predictions about the effect oftaxes on the use of options there are direct tax advantages of options inthat options provide executives with a way to defer compensation andthereby lower their taxes However because options also lead to a defer-ral of corporate tax deductibility the tax advantages from a combined(executive and corporate) perspective are not so straightforward In thissection we analyze and measure the tax advantages of stock optionsrelative to salary and bonus compensation We also analyze the taxadvantages of NQSOs relative to ISOs In particular we show how thetax advantages of options change as personal corporate and capitalgains taxes change We then show how the tax advantages of optionshave changed over time in response to changing tax rates

The crucial tax difference between standard options (NQSOs) andcash payment is that option payouts are deferred and the two forms ofcompensation earn different rates of return over the deferral period Anyanalysis of the relative tax advantages of two compensation instrumentsmust consider the tax consequences both to the employer and to theemployeemdashwhat Scholes and Wolfson (1992) call the global contractingperspective Thus in order to make valid comparisons between the twowe compare the tax burden to the executive while holding constant the

Taxation of Executive Compensation 15

posttax cost to the company By keeping posttax employer costs (in NPVterms) constant any package that is preferred by the employee is tax-advantaged in the global-contracting sense

Under this methodology a comparison of the tax advantages of op-tions and cash involves comparing a pretax cash payment of P with anoption payment that has the equivalent posttax NPV to the companyFor NQSOs it is straightforward to show that a pretax payment of P toan executive has exactly the same cost to the company as putting aside Pfor the purposes of paying stock-option payouts later That is if a com-pany pays an executive P today it will have the same amount of moneyin N years as if it had put aside P today let it accumulate at the rate ofreturn earned by the rm and then paid it out (as compensation in theform of option gains4) with any appreciation in year N

Assume that the pretax prot rate is r and the corporate tax rate is TcThen if the company pays P today it receives a deduction today of PTcwhich yields

PTc [1 1 r (1 2 Tc)]N (1)

in N years since the benets of the tax deduction are invested in thecompany and receive the after-tax corporate rate of return Converselyif the company puts P aside today then it grows at the after-tax corpo-rate rate of return in N years to give a tax deduction of

P [1 1 r (1 2 Tc)]N Tc (2)

which is the exact same value Note that the equivalence of these two taxdeductions is analogous to the tax-benet equivalence of front-loadedand back-loaded IRAs

An executiversquos payoff in N years from option prots is equal to Pcompounded at the after-tax corporate rate times 1 2 Tp where Tp is thepersonal tax rate Thus the combined payoffs in N years of the corporatededuction and the executiversquos posttax payoff is

P [1 1 r (1 2 Tc)]N Tc 1 P [1 1 r (1 2 Tc)]

N (1 2 Tp) (3)

where the rst term is the payoff from the corporate deduction and thesecond term is the executiversquos payoff

4 This analysis is not unique to options Any form of deferred compensation that enablesexecutives to invest inside the rm at a higher posttax return or without paying capitalgains taxes would have similar effects

16 Hall amp Liebman

The payoff to salary and bonus in N years is more complicated sincean assumption must be made about how the executive invests the origi-nal cash compensation For example if the executive invests in an instru-ment with taxable interest (eg bonds) then the N-year payoff from aninvestment of P accumulates at the post-corporate-tax post-personal-taxrate of return and is

P (1 2 Tp) [1 1 r (1 2 Tc)(1 2 Tp)]N (4)

However if an executive invests in (non-dividend-paying) equities theinvestment accumulates at the higher post-corporate-tax rate of returnwith the offsetting disadvantage that the capital gains are taxed at thecapital gains rate in year N So the executive receives

P [1 1 r (1 2 Tc)]N (1 2 Tp) 2 Tcg $P (1 2 Tp) [1 1 r (1 2 Tc)]

N 2P (1 2 Tp) (5)

where Tcg is the capital gains tax rate5Thus putting the equations together if an executive is paid cash P

and invests g in bonds and 1 2 g in equities the N-year payoff (includ-ing the payoff to the corporation from the tax deduction) is

P [1 1 r (1 2 Tc)]N Tc 1 g $(P [1 1 r (1 2 Tc)(1 2 Tp)]

N (1 2 Tp)1(1 2 g) (P [1 1 r (1 2 Tc)]

N (1 2 Tp) 2 Tcg $P (1 2 Tp)[1 1 r (1 2 Tc)]

N 2 P (1 2 Tp)) (6)

The tax advantage of options versus salary and bonus therefore is thedifference between the combined (corporate and executive) payoffs inequation (3) and the combined payoffs in equation (6)

43 The Tax Advantage of OptionsWe now turn to analysis of how changes in various tax rates affectthe tax advantages of options relative to cash compensation We focuson the case in which the cash earned by executives is invested entirelyin equities since equities are tax-favored relative to bonds The impor-tant conclusions of this analysis are not substantively different if we

5 In practice executives hold a combination of equities and bondlike instruments withtaxable interest Because equities are tax-favored in the next section we simplify the analy-sis by assuming that executives hold only equities Alternatively we could have assumedthat executives hold only less risky bonds since bonds more closely match the risk proleof a stream of tax savings None of our empirical results are substantively affected by thissimplication

Taxation of Executive Compensation 17

instead assume that a portion of an executiversquos holdings are invested inbonds

431 The Tax Advantage of Options Changing the Corporate TaxRate Stock options are tax-advantaged relative to cash simply becauseoption payouts are deferred allowing the executive to invest at the pre-personal-tax rate of return (with no capital gains tax at the end) ratherthan investing at the after-tax rate of return or paying the capital gainstax Since the advantage of being able to defer taxes is large when thecorporate tax rate is low the tax advantage of options is larger when thecorporate rate is low

To illustrate this point we dene the tax advantage of options to bethe dollar amount by which total after-tax option payoffs (to the corpora-tion and the executive) exceed the total payoffs from salary as denedby the difference of equation (3) and (6) assuming that P (the paymentto the executive) is equal to $100 and g 5 0 (cash compensation is in-vested in equities) We use a ten-year horizon since most options have aten-year term We then calculate how the tax advantage of optionschanges as the corporate tax rate varies from zero to 100 percent holdingthe personal tax rate and the capital gains rate xed at 40 percent and 20percent respectively (which approximates current rates)

The top panel of Figure 2 shows the tax advantage of options relativeto cashmdashprecisely the NPV of the total tax advantage of options for a$100 payment to the CEOmdashas the corporate tax rate changes As ex-pected the tax advantage of options declines as corporate tax rates in-crease At a zero corporate tax rate the tax advantage of options has anNPV of $7 This value is about $350 at a corporate rate of 40 percentWhen the corporate tax rate is 100 percent the tax advantage of optionscompletely disappears since the benets of deferred compensation fallto zero because the after-tax return on equities (and therefore the dis-count rate) falls to zero The tax advantage of options is the NPV of thetax savings from avoiding the capital gains tax on ten years of apprecia-tion which is why it is relatively modest even at the 40-percent corpo-rate rate

432 The Tax Advantage of Options Changing the Personal TaxRate Using the same assumptions as above (but this time holding thecorporate rate at 40 percent and the capital gains rate at 20 percent andvarying the personal rate) the tax advantage of options declines as thepersonal tax rate rises as shown in the middle panel of Figure 2 The taxadvantage of options is higher at low personal tax rates because theposttax base that gives rise to capital gains is higher and the advantage of

Note Holds constant the personal rate at 40 percent and the capital gains rate at 20 percent

Note Holds constant the corporate rate at 40 percent and the capital gains rate at 20 percent

Note Holds constant the personal rate at 40 percent and the corporate rate at 40 percent

FIGURE 2 Tax Advantage of Options NPV of Tax Advantage of Pay-ing $100 in Options (NQSOs) Rather Than in Cash

Taxation of Executive Compensation 19

avoiding the capital gains taxes is greater The tax advantage of optionstherefore declines linearly as the personal tax rate rises At a 100-percentpersonal tax rate the executive receives nothing in either case and byconstruction the companyrsquos tax deduction benet is the same in NPVterms so the tax advantage of options falls to zero also6

433 The Tax Advantage of Options Changing the Capital Gains TaxRate The tax advantage of options for various capital gains rates isdepicted in the bottom panel of Figure 2 under the same assumptions(this time holding both the personal rate and the corporate rate at 40percent while varying the capital gains rate) The tax advantage of avoid-ing the capital gains tax increases as the capital gains tax rate rises

5 HOW HAS THE TAX ADVANTAGE OF OPTIONSCHANGED OVER TIME

The analysis so far has illustrated how the tax advantage of optionsvaries with the changes in tax rates In this section we show how the taxadvantage of options has changed over time as tax rates have changed

51 Top Marginal Tax Rates over TimeThe evolution on the top marginal tax ratesmdashpersonal corporate andcapital gainsmdashfrom 1980 to 1998 is shown in Figure 3 In all cases thetop rate (the rate for taxpayers with the highest incomes) is shownwhich is not always the highest rate since various anomalies (such asclawbacks of exemptions) sometimes lead to marginal tax rates that arehigher than those for the highest income tax payers

The top personal tax rate was 70 percent in 1980 falling to 50 percentin 1982 and 28 percent in 1988 following the 1986 tax act7 The topmarginal tax rate has since risen to 396 percent but is effectively 425percent since there is a 29-percent Medicare surcharge (paid half by theemployer and half by the employee) that has no upper limit The topcorporate tax rate has had only one signicant change since 1980 fallingfrom 46 percent before the 1986 tax act to 34 percent following the act in1988 The top corporate tax rate was increased to 35 percent in 1993 The

6 This is one place where ignoring bond investments has substantive implications Inparticular in the all-bond case the tax advantage of options is nonmonotonicmdashit rises andthen fallsmdashas personal tax rates rise7 During the early 1980s the maximum tax on earned income limited the marginal tax rateon the earnings of high earners so many executives did not face a decline in personal taxrates between 1981 and 1982 The empirical results in this paper are robust to assumingthat executives faced a marginal tax rate of 50 percent in 1980 and 1981

20 Hall amp Liebman

FIGURE 3 Top Marginal Tax Rates since 1980

top capital gains rate has uctuated between 28 percent (approximately)and 20 percent since 1980 and is currently at 20 percent

52 The Evolution of the Tax Advantage of NQSOsThe tax advantage of options relative to cash dened in the same way asin section 4 from 1980 to 1998 is shown in Figure 4 The calculations aredone using the same assumptions as before (ie N 5 10 years and cashis invested in equities) and the statutory top marginal tax rates shown inFigure 3 are used Due to the fall in corporate and personal tax rates thetax advantage of options increased sharply after the Tax Reform Act of1986 This increase in tax advantage has largely been reversed in the1990s as top personal tax rates have crept back up and the capital gainsrate has been reduced

Two key facts emerge from this analysis First there is currently only amoderate tax advantage to standard nonqualied stock optionsmdashon theorder of $4 per $100 of compensation This is because the tax advantagesto the executive of deferring taxes are largely offset by the tax disad-vantages to the company of not being able to deduct option expensesfrom taxable prots until the executive exercises the options Secondalthough the 1986 tax act substantially increased the tax advantage ofoptions more than half of this increase has been reversed in the 1990s

Taxation of Executive Compensation 21

FIGURE 4 Tax Advantage of Options 1980 to 1998

NPV of tax advantage of $100 in options (NQSOs) rather than cash

53 The Tax (Dis)advantage of ISOsSo far the analysis has focused on the tax advantages of NQSOs sincethey are so much more prevalent than ISOs which account for onlyabout 5 percent of option grants The relative scarcity of ISOs can beexplained both by their tax status and by their per-executive cap of$100000 per year

Under what conditions are ISOs tax-preferred to NQSOs For a trans-fer P to an executive the ISO is always tax-preferred by the executive(since the capital gains rate is lower than the personal rate) while theNQSO is always tax-preferred from the companyrsquos perspective (since anISO is not deductible) The key issue however is the relative advan-tages of the two types of options from a global-contracting perspectiveBecause ISOs are not deductible the company is indifferent betweensetting aside P in the form of NQSOs today and setting aside P(1 2 Tc) inthe form of ISOs (Since the NPV of the tax deduction for NQSOs is PTcthe NPV of the payment P is P 2 PTc or P (1 2 Tc))

Equalizing the posttax cost of NQSOs and ISOs we need only look atthe payoff of the executive to determine the condition under which eachis preferred The payoff of NQSOs in N years is

P [1 1 r (1 2 Tc)]N (1 2 Tp) (7)

22 Hall amp Liebman

The same payoff for ISOs is

P (1 2 Tc) [1 1 r (1 2 Tc)]N (1 2 Tcg) (8)

ISOs therefore are tax-advantaged if

Tp 2 Tcg Tc 2 Tc Tcg (9)

Intuitively ISOs are tax-advantaged only if their advantage (the differ-ence between the personal rate and the capital gains rate) is largeenough to offset their cost (the disadvantage of not deducting at thecorporate rate) Note however that the condition is not a simple com-parison between the corporate rate and the personal rate minus thecapital gains rate (Tp 2 Tcg Tc)

Figure 5 shows how the tax advantage of NQSOs relative to ISOs haschanged since 1980 as tax rates have changed Unlike the very modestchanges (around 3 to 4 percent) in the relative tax advantages betweenoptions and cash shown in Figure 4 the relative tax advantage betweenNQSOs and ISOs has seen enormous swings When the top personalrate was 70 percent in 1980 ISOs were tax-favored by a margin greaterthan 6 percent However by 1982 (following the 1981 tax acts that dra-matically lowered the personal tax rate) ISOs became tax-disadvantaged

FIGURE 5 Tax Advantage of NQSOs Relative to ISOs over Time

NPV of tax advantage of $100 payment to executive NQSOs minus ISOs

Taxation of Executive Compensation 23

by about 5 percent By 1988 (following the 1986 tax act) ISOs became tax-disadvantaged by about 18 percent a dramatic swing The raising ofpersonal rates coupled with a decrease in the capital gains rate has sincereduced the tax disadvantage of ISOs to about $4 per $100 of pretaxcompensation Although hard data on ISOs are hard to come by practi-tioners (mostly compensation consultants) claim that ISOs were morepopular prior to 1982 when they were tax-advantaged Although ISOsare less disadvantageous from a tax standpoint relative to NQSOs thanthey were in the late 1980s they are still disadvantageous so it is notsurprising that they have not made a signicant resurgence in recentyears

6 EXPLAINING THE INCREASE IN OPTION-BASEDCOMPENSATION

In this section we analyze whether changes in tax rates affect the compo-sition of executive compensation and consider the relative importanceof tax factors and corporate governance factors in explaining the in-creasing share of compensation paid in the form of stock options Thenumerous changes in tax rates that have occurred since 1980 provide anopportunity to assess whether executive compensation arrangementsrespond to tax incentives in the way that the tax-avoidance model sug-gests We test this model using a panel data set of CEOs in largepublicly-traded US corporations Our identication of the tax effectsrelies on time-series variation in personal corporate and capital gainstax rates along with cross-sectional variation in corporate tax rates

61 DataWe use a 15-year panel data set of CEOs in the largest publicly-tradedUS corporations which is described in Hall and Liebman (1998) Thedata set covers the years from 1980 through 1994 and combines CEOcompensation information from corporate proxies and 10-K lings withstock price and stock return information from CRSP and accounting datafrom Compustat In addition some compensation data from the 1970swere collected in order to construct measures of the value of stock op-tions held by the CEOs in the rst period of the sample8

The feature that distinguishes our data from most other CEO datasets is that with our panel of yearly proxy data on option grants optiongains and total options held we are able to calculate the total value of

8 We are grateful to David Yermack for providing us with some of the data for 1984 to 1991See Yermack (1995) for a discussion of these data

24 Hall amp Liebman

all stock options held by the CEO at a point of time More importantlysince we have the details about the stock options held (number exer-cise price time to maturity etc) we can precisely calculate the changein the value of a CEOrsquos stock option holdings for a given change in rmvalue

62 Identication IssuesThe numerous tax changes over our sample period and the specic waysin which changes in personal corporate and capital gains rates arepredicted to affect the composition of CEO compensation give us un-usually rich sources of identication However because many of thechanges in tax rates would be expected to affect all of the CEOs in oursample in a similar way we will need to pay particular attention toseparating out the tax effects from underlying time trends and otherfactors that changed over time Controlling for underlying trends is par-ticularly important in this study because we are focusing on an out-comemdashthe use of stock optionsmdashthat has increased rapidly over thepast two decades and that many practitioners believe was importantlyaffected by nontax considerations Thus the tax effects we are trying toexplain are deviations from a rising trend

We take two steps to try to separate out the tax effects from theunderlying trend First we include key nontax factors that could poten-tially explain the increasing reliance on performance-based compensa-tion It has been suggested that the dramatic increase in incentive-basedpay is the result of the remarkably poor shareholder returns during the1970s which spurred the LBO and takeover movements of the 1980sAccording to this story (Kaplan 1997) shareholders became much morepowerful via the rise of institutional investors even as the LBO andtakeover movements waned Because the inuence of institutional inves-tors is thought to be one of the most important mechanisms of strongcorporate governance we use the share of each companyrsquos stock ownedby institutional investors (dened as institutions with more than $100million under management) as an explanatory variable9 The percentageof shares owned by large institutional investors increased from about 20percent to almost 50 percent in 1994 an upward trend that closelymatches the sharp rise in the share of compensation in the form ofoptions Indeed in our sample the annual average (over all of the rmsin our sample) share of stock owned by institutional investors has acorrelation of 09 with the annual average share of compensation paid inoptions

9 This variable is described in Gompers and Metrick (1998)

Taxation of Executive Compensation 25

In addition we include two other variables that proxy for strongercorporate governance The rst is the size of the board Evidence suggeststhat boards with smaller numbers of directors reduce the inuence of theCEO and better represent shareholder interests (Yermack 1995) Boardsize fell by approximately 15 percent between 1980 and 1994 The secondvariable is the fraction of the rmrsquos board members who are inside direc-tors or gray directors (noninsiders who have business dealings with thecompany) Firms with fewer outsiders are less likely to act in the share-holdersrsquo interests and we would predict that they would be less likely tohave performance-based compensation for their CEOs The average per-centage of inside and gray directors has decreased from about 45 percentin 1980 to about 35 percent in 1994

Second we allow for cross-sectional variation in corporate tax ratesCorporate marginal tax rates vary because rms may be eligible for tax-loss carrybacks and carryforwards investment tax credits and the alter-native minimum tax (see Auerbach 1986 Auerbach and Altshuler1990 Auerbach and Poterba 1987) Because these cross-sectional differ-ences imply that different rms should respond differently to a givenchange in tax rates they potentially provide us with a way to isolatethe tax effects Firm-specic corporate marginal tax rates are notori-ously difcult to calculate from publicly-available data We use a tricho-tomous variable that equals zero if the rm has tax-loss carryforwardsand negative earnings in a given year equals 05 times the statutorycorporate rate if the rm has only one of those conditions and equalsthe statutory rate if the rm has neither loss carryforwards nor negativeearnings10

The cross-sectional variation in the corporate tax rate may not be ex-ogenous Firms that perform poorly could face low marginal tax ratesand also be particularly likely or unlikely to provide performance-basedpay Because most of our specications use xed effects the correlationwe must be concerned about is between changes in rm performanceand changes in compensation We deal with this by controlling forlagged rm stock market performance in our regressions This shouldeliminate spurious correlation between tax rates and compensation thatis jointly caused by rm performance

10 Graham (1996) presents evidence that the trichotomous variable performs nearly as wellas his recommended simulated marginal tax rate Plesko (1999) compares marginal taxrates calculated from Compustat data with actual marginal tax rates from rm tax returnsHe nds that simple binary variables are more reliable measures of the marginal tax ratethan is either the Graham or the trichotomous variable We obtain substantively similarresults when we switch from the trichotomous variable to the binary variables recommendby Plesko

26 Hall amp Liebman

63 ResultsThe dependent variable in our regressions is the fraction of a CEOrsquos totalannual compensation (measured as the BlackndashScholes value of stock-option grants plus salary and bonus) that comes from stock-optiongrants We choose this reduced-form specication because our interest isin the composition of CEO compensation However the factors thataffect the composition of compensation might also affect the levels andit will be important to keep this in mind in interpreting our results

Optimal-contracting theory generally makes predictions about the in-centives provided by the total amount of rm stock and stock optionsowned by the CEO We choose to model the ow of option grants ratherthan the ultimate performance sensitivity of the CEOrsquos compensationand stock and stock-option holdings because we have in mind anadjustment-cost model in which it takes time for a CEOrsquos holdings ofstock and stock options to reach the optimal level11 In our regressionswe include two measures of the current performance sensitivity of theCEOrsquos wealth to rm performance as explanatory variables to reect thedistance of the CEOrsquos contract from the optimal contract

Table 3 presents results in which the tax effects are identied solely byvariation in tax rates over time The fraction of annual compensation paidin stock options is regressed on tax variables corporate governance vari-ables contract-theory variables and variables reecting stock market re-turns Both of the regressions in this table include rm xed effects Incolumn (1) the fraction in annual compensation paid in stock options isregressed on the log difference in payoff (the combined payoff to both therm and the CEO) from receiving compensation in stock options ratherthan salary and bonus where the payoff calculations are based on statu-tory marginal tax rates This variable is dened as the log difference be-tween equation (3) and equation (6) in section 4 The coefcient on the taxvariable is large and statistically signicant at the 95-percent level Thepoint estimate implies that a one-percent increase in the payoff differencebetween stock options and salary and bonus results in a 24-percentage-point increase in the share of compensation paid in stock options

The corporate governance variables all have the predicted sign Firmswith a higher fraction of their shares owned by institutional investorsuse more performance-based pay (stock options) Firms with large corpo-rate boards or a large share of inside directors are less likely to use stockoptions

The two contract-theory variables are the dollar change in CEO wealth

11 See Core and Guay (1999) for evidence that rms adjust in this way

TABLE 3Compensation Regressions that Rely on Variation in

Tax Rates over Time

(1) (2)

Tax variablesln (difference in payoff)(a) 2393 (0349)ln (12statutory personal 20015 (0034)

rate)ln (12statutory cap gains 0330 (0171)

rate)ln (12statutory corporate 0399 (0148)

rate)

Corporate governance variablesFraction of rmrsquos shares 0043 (029) 0052 (0028)

owned by institutionalinvestors

Size of the rmrsquos board of 20002 (001) 20003 (0001)directors

Fraction of directors who 20082 (033) 20092 (0030)are inside or gray

Contract-theory variablesD (CEO wealth) per $1000 20261 (0095) 20247 (0087)

in FV(b)

D (CEO wealth) per 10 in 2583 (189) 2554 (201)FV(c)

Volatility(d) 521 (424) 261 (369)ln (rm market value) 0019 (0006) 0019 (0006)

Stock market return variablesLagged annual rm stock 0006 (0009) 0003 (0008)

market returnLagged annual SampP 500 re- 20035 (0023) 20029 (0023)

turn

Regression diagnosticsTotal sample size 5179 5179Firms 427 427Adjusted R2 0313 0311

Dependent variable share of compensation paid in options Standard errors in parentheses Bothregressions in this table include rm xed effects

(a) Log difference in payoff is the log difference in combined rm and CEO after-tax share betweenpaying compensation in options and paying salary

(b) This variable is the dollar change in CEO wealth per $1000 change in rm market value (seeJensen and Murphy 1990)

(c) This variable in the dollar charge in CEO wealth per 10-percent change in rm value (see Hall andLiebman 1998)

(d) Standard deviation of rm market value in trillions

28 Hall amp Liebman

per $1000 change in rm value (Jensen and Murphy 1990) and thedollar change in CEO wealth per 10-percent change in rm value (Halland Liebman 1998) As discussed in Baker and Hall (1998) these twowealth sensitivity measures reect two different concepts of how closelyaligned the managersrsquo incentives are with the interests of the sharehold-ers In the regression they are meant to measure how far the CEOrsquosexisting contract and ownership of stock and stock options is from theoptimal level The negative coefcients on the two variables suggeststhat rms that are below their optimal pay-to-performance sensitivityare more likely to give more stock options in the current period

The second column replaces the tax variable motivated by deferraladvantages with the three statutory rates The corporate after-tax sharehas a positive coefcient as we would expect because the tax advantageof stock options increases as the corporate rate falls The coefcient onthe personal tax rate is small and statistically insignicant The sign ofthe coefcient on the capital gains rate is the opposite of that predictedby tax-avoidance theory

Note that these results rely on time-series variation alone This is not avery convincing test Since the log difference in payoffs mostly rises overthe time period covered in our sample as does the share of compensa-tion paid in options it is possible that the tax variable is simply reectingother factors that were changing over this time period that we have notincluded in our regression Therefore it is important to look at specica-tions that rely on cross-sectional variation in the tax variables

Table 4 contains specications that rely on cross-sectional variation incorporate tax rates The rst column repeats the specication from col-umn (1) of Table 3 but uses the trichotomous measures of rm corporatetax rates in place of the statutory rate in calculating each rmrsquos logdifference in payoff from paying compensation in options rather than insalary and bonus As in the previous table we nd the positive coef-cient on the log difference in payoff just as the theory suggest How-ever the coefcient estimate is now much smaller suggesting that aone-percent increase in the log difference in payoff produces a 01-percentage-point increase in the share of compensation paid in options

Introducing cross-sectional variation in the dependent variable doesnot eliminate the potentially spurious correlation over time between thetax variable and the share of compensation paid in options The secondcolumn of Table 4 isolates the pooled cross-sectional variation in corpo-rate tax rates by adding time dummies and dropping the rm xedeffects In this specication the coefcient on the tax variable is indistin-guishable from zero The coefcient on the institutional variable contin-ues to be large and statistically signicant The third column includes

Taxation of Executive Compensation 29

both rm xed effects and time dummies which tests for a relationshipbetween within-rm changes in the dependent variable and within-rmchanges in each explanatory variable In this specication the coefcienton the tax variable is again indistinguishable from zero The coefcientson the corporate governance variables are also statistically insignicant

The third column contains a specication that treats the Tax ReformAct of 1986 (TRA 86) as a natural experiment For rms and CEOs facingthe statutory corporate tax rate TRA 86 increased the relative payoff topaying compensation in options by 17 percent For a rm facing a zeromarginal tax rate it reduced the relative payoff by 45 percent The reduc-tion occurred because TRA 86 reduced the personal tax rate makingsalary and bonus more attractive while the decreased corporate rate hadno effect on zero-tax-rate rms Thus we would expect rms facing zeroor low marginal tax rates to reduce their use of options after TRA 86while rms facing the statutory rate would be expected to increase themFew rms face zero marginal tax rates year after year Therefore we tryto distinguish between rms that often face low tax rates and rms thatnearly always face the statutory rate We dene low-tax rms as oneswhose average marginal tax rate over 1984 to 1986 was below 02412

Then we ran a difference-in-differences regression to see if high-taxrms increased their use of options more than low-tax rms The interac-tion between post-TRA 86 and high corporate tax rates is the key vari-able The estimated coefcient on this variable is positive but small andnot statistically different from zero at the 95-percent level

It is important to note that in this specication as well the main corpo-rate governance variable the share of the rmrsquos shares owned by institu-tional investors performs as theory would predict The variable appearsto be quite robust Table 5 presents regressions of the share of compensa-tion paid in stock options on the fraction of the rmrsquos shares owned byinstitutional investors Since we have already established that these vari-ables have a strong longitudinal relationship the regressions here aredesigned to focus on the cross-sectional relationship between these twovariables The rst and third columns present regressions using data onlyfrom 1994 and nd a coefcient of roughly 02 (column 1 contains noother covariates while column 3 includes all of the other nontax covari-ates from the regressions in Tables 3 and 4) Since the median share of armrsquos stock owned by institutional investors in our sample rises from 19percent in 1980 to 49 percent in 1994 this coefcient would imply that this

12 This cutoff was chosen after inspecting the distribution of tax rates and essentiallyseparates rms that are always at or near the statutory rate from ones that are sometimesbelow it

TA

BL

E4

Com

pens

atio

nR

egre

ssio

nsth

atR

ely

onC

ross

-Sec

tion

alV

aria

tion

inC

orpo

rate

Tax

Rat

es

(1)

(2)

(3)

(4)

Tax

vari

able

s(a)

ln

(dif

fere

nce

inp

ayof

f)0

099

(00

41)

20

038

(00

36)

20

005

(00

44)

Pos

t-T

RA

862

000

5(0

013

)H

igh

corp

orat

eta

xra

te0

018

(00

17)

(Pos

t-T

RA

86)

3hi

gh0

024

(00

16)

corp

orat

eta

xra

te

Cor

por

ate

gove

rnan

ceva

riab

les

Frac

tion

of

rmrsquos

shar

es0

109

(00

27)

011

4(0

015

)2

000

9(0

030

)0

087

(00

34)

owne

dby

inst

itu

tion

alin

vest

ors

Size

ofth

e

rmrsquos

boar

dof

20

004

(00

01)

20

001

(00

01)

20

002

(00

01)

000

1(0

002

)d

irec

tors

Frac

tion

ofd

irec

tors

wh

o2

012

4(0

033

)2

000

3(0

018

)2

004

1(0

034

)0

001

(00

39)

are

insi

de

orgr

ay

Con

trac

t-th

eory

vari

able

sD

(CE

Ow

ealt

h)p

er$1

000

20

238

(00

94)

20

167

(00

53)

20

257

(00

96)

20

021

(01

47)

inFV

(b)

D(C

EO

wea

lth)

per

10

in2

560

(17

8)2

592

(10

5)2

594

(19

6)2

147

(04

86)

FV(c

)

Vol

atili

ty(d

)2

48(4

24)

162

(03

4)1

12(4

28)

106

(06

09)

ln(

rmm

arke

tva

lue)

003

3(

006)

000

8(0

003

)0

009

(00

06)

001

2(0

007

)

Stoc

km

arke

tre

turn

vari

able

sL

agge

dan

nual

rm

stoc

k2

000

2(0

009

)0

011

(00

10)

000

6(0

009

)2

001

2(0

013

)m

arke

tre

turn

Lag

ged

annu

alSamp

P50

02

005

6(0

023

)2

005

3(0

051

)2

003

9(0

049

)2

015

4(0

040

)re

turn

Reg

ress

ion

dia

gnos

tics

T

otal

sam

ple

size

5179

5179

5179

2534

Firm

s42

742

739

6A

dju

sted

R2

031

009

032

005

Firm

xed

effe

cts

Yes

No

Yes

No

Tim

ed

um

mie

sN

oY

esY

esN

o

Dep

end

ent

vari

able

isfr

acti

onof

ann

ualc

omp

ensa

tion

from

stoc

kop

tion

sSt

and

ard

erro

rsin

par

enth

eses

(a)

All

thre

esp

eci

cati

ons

use

tric

hoto

mou

sco

rpor

ate

tax

rate

sT

he

sam

ple

inco

lum

n(3

)u

ses

only

the

year

s19

84to

1988

sin

cew

ear

eu

sing

1986

asa

nat

ural

exp

erim

ent

(b)

Thi

sva

riab

leis

the

dol

lar

chan

gein

CE

Ow

ealt

hp

er$1

000

chan

gein

rm

mar

ketv

alu

e(s

eeJe

nse

nan

dM

urp

hy

1990

)

(c)

Thi

sva

riab

lein

the

dol

lar

char

gein

CE

Ow

ealt

hp

er10

-per

cent

chan

gein

rm

valu

e(s

eeH

alla

nd

Lie

bman

199

8)

(d)

Stan

dar

dd

evia

tion

ofr

mm

arke

tva

lue

intr

illio

ns

32 Hall amp Liebman

TABLE 5Large Institutional Investors and Stock-Option Compensation

The Cross-sectional Relationship between Institutional Investors andOption Grants

(1) (2) (3) (4)

Coefcient on fraction of 217 (067) 145 (014) 177 (069) 120 (015)rmrsquos shares owned byinstitutional investors

Sample period 1994 1980ndash1994 1994 1980ndash1994Contract-theory and stock No No Yes Yes

market covariatesTime dummies No Yes No Yes

The dependent variable is the share of compensation paid in options

factor alone can explain 6 percentage points of the 23-percentage-pointincrease in the median share of compensation paid in options that oc-curred over this time period Of course this variable is only one measureof the strengthening of corporate governance so the total impact of im-proved corporate governance could be greater The second and fourthcolumns of the table use the entire 1980ndash1994 sample and include timedummies to isolate the cross-sectional correlations Again there is evi-dence of a strong cross-sectional relationship between the two variablesalthough the coefcients are slightly smaller perhaps reecting a weakercorporate-governance role in the earlier period In sum there is a strongrelationship in both longitudinal and cross-sectional data between theinstitutional-investor variable and the options-percentage variable13

7 THE MILLION-DOLLAR RULE

In 1993 section 162(m) of the Internal Revenue Code was enactedlimiting the deductibility of executive compensation in excess of onemillion dollars unless the compensation was performance-related Pro-ponents of this legislation argued that it would slow the growth ofexecutive pay and tighten the link between rm performance and execu-tive compensation

The new provision became effective for tax year 1994 and applies tothe CEO and the other four most highly compensated executives in eachcompany Pay that is performance-related such as stock-option grants

13 The one specication in which the fraction owned by institutional investors is notsignicant is in column (2) of Table 4 a regression that includes both xed effects and timeeffects

Taxation of Executive Compensation 33

and bonuses paid for meeting clear performance goals is not affected bythis regulation provided that it has been approved by shareholdersThus the main impact of the provision is to limit the deductibility ofsalary in excess of a million dollars Woodlock and Antenucci (1997)studied the proxy statements of 376 rms and documented that mostrms responded to the new law by qualifying top executive pay asperformance-related Perry and Zenner (1999) identied 25 rms thatreduced salaries from above $1 million to below $1 million and foundthat 23 of the rms cited section 162(m) as their reason for doing so

In this section of the paper we investigate whether this provision hasaffected the level or structure of executive pay using the Execucompdata base which contains information on the highest-paid executives ineach of the rms in the SampP 500 SampP Midcap 400 and SampP SmallCap600 Between 1993 and 1998 the median salary of the Execucomp execu-tives rose from $230000 to $306000 The median value of total compen-sation (salary bonus and option grants) rose from $470000 to $882000Limiting the sample of CEOs indicates that the total compensation of themedian CEO rose from $11 million to $18 million over the period whilethe median salary rose from $441000 to $550000 Thus executive com-pensation clearly continued to rise at a rapid rate after the implementa-tion of the million-dollar rule though perhaps at a slower rate than itotherwise would have

If the regulation affects behavior it should have the largest impact onrms whose executives are receiving salaries that are above the limit orjust below it Thus we might expect to see rms whose executives arereceiving relatively high salaries to rely more heavily on performance-based pay and less heavily on salary in the years following the regulationPerry and Zenner (1999) and Livingston (1999) both nd that this is thecase This trend can be clearly identied in the Execucomp data set Forexample over 1993 to 1998 the median annualized growth rate of salaryfor executives whose 1993 salary was below $500000 in the previous yearwas 85 percent For executives with salary between $500000 and$700000 it was 64 percent for executives between $700000 and $850000it was 50 percent for executives between $750000 and $1 million it was31 percent and for executives above $1 million it was 00 percent14

The observation that executives at lower levels of salary had highergrowth rates is not very strong evidence of an impact of section 162(m)however because it is possible that low-salary executives would havehad higher growth rates even in the absence of the provision To explore

14 Of the 3137 executives for whom we have complete data between 1993 and 1998 2708had 1993 salaries below $500000 while only 35 had 1993 salaries above $1 million

34 Hall amp Liebman

whether this is the case we use Execucomp data from both before andafter the law change Ideally we would want to have data from manyyears before the change to examine whether there is an underlying pat-tern of faster salary growth for low-salary executives Unfortunately theExecucomp data begin in 1992 and therefore provide us with only the1992ndash1993 growth rate in compensation to use as a baseline15

We run regressions pooling data from 1992ndash1993 through 1997ndash1998of the annual percentage change in different forms of compensation on avariable million that is designed to measure the likelihood that anexecutiversquos compensation will be affected by section 162(m) We denemillion as the minimum of 1 and the executiversquos salary divided by $1million in the previous year Thus any executive with salary at or above$1 million would receive a value of 1 while an executive with a salary of$500000 would be coded as a 0516 We include this variable by itself andalso interacted with a dummy variable (after) that equals 1 for years1994 and after Thus the million variable is designed to pick up anyunderlying relationship between the level of salary and the growth rateof salary while the interacted one will identify any differential growthrate of salary by income in the period after section 162(m) took effect Wealso include a full set of year dummies the natural logarithm of thermrsquos market value and the rmrsquos annual rate of return over the twoprevious years as control variables To reduce the sensitivity of our re-sults to outliers we run quantile regressions and robust regressions

Table 6 contains our results The rst four columns are for all topexecutives while the last four columns focus on CEOs Columns (1) and(2) present estimates from robust regression for the percentage change insalary and the percentage change in total compensation (salary bonusstock grants and stock-option grants) In the salary growth regressionsfor all executives the coefcient on million is negative and signicantindicating that there is an underlying relationship between the level ofexecutive compensation and its subsequent growth rate The coefcienton million 3 after is negative as well suggesting that the negativerelationship between the level of salary and its subsequent growth rateintensied slightly after 1994 The relatively small magnitude suggeststhat an executive with a $1 million dollar salary would see his salarygrow at an annual rate that would be about 06 percent slower than anexecutive earning $500000

15 The HallndashLiebman data set contains only a combined salary and bonus variable makingit impossible to use it to analyze this issue16 We also ran the regressions (not reported) allowing for a nonlinear relationship betweenmillion and compensation growth and obtained substantively similar results

TA

BL

E6

The

Eff

ect

ofth

eM

illi

on-D

olla

rR

ule

onSa

lary

and

Tot

alC

ompe

nsat

ion

All

top

exec

uti

ves

CE

Os

only

Rob

ust

regr

essi

onQ

uan

tile

regr

essi

onR

obu

stre

gres

sion

Qu

anti

lere

gres

sion

D

in

Din

D

in

Din

D

in

Din

D

in

Din

sala

ryto

talc

omp

(a)

sala

ryto

talc

omp

sa

lary

tota

lcom

p

sala

ry

tota

lcom

p

mil

lio

n(b

)2

006

02

035

72

006

32

024

12

008

72

023

12

010

32

024

5(0

005

)(0

032

)(0

006

)(0

027

)(0

010

)(0

073

)(0

011

)(0

056

)m

illi

on

32

001

10

026

20

021

20

015

20

016

014

92

003

40

158

aft

er(c

)(0

005

)(0

033

)(0

006

)(0

029

)(0

010

)(0

076

)(0

011

)(0

057

)N

359

5035

950

359

5035

950

757

17

589

757

17

589

Stan

dar

der

rors

inp

aren

thes

esR

egre

ssio

nsal

soin

clu

de

year

du

mm

ies

stoc

km

arke

tre

turn

and

log

mar

ket

valu

e

(a)

Tot

alco

mp

ensa

tion

incl

ud

essa

lary

bon

us

rest

rict

edst

ock

and

stoc

k-op

tion

gran

ts

(b)

mil

lion

isd

ene

das

min

(1s

alar

yin

pre

viou

sye

ars

div

ided

by$1

mil

lion)

(c)

aft

eris

anin

dic

ator

vari

able

equ

alto

zero

befo

reth

ere

gula

tion

took

plac

e(1

993)

and

one

afte

rwar

d

36 Hall amp Liebman

In the second column the coefcient on million 3 after is essentiallyzero suggesting that any decrease in salary brought about by section162(m) was offset by increases in bonus and stock-option grants Theresults are similar for the median regressions There is some evidence ofa very small slowdown in growth of salary for executives likely to beaffected by section 162(m) but no sign of a change in total compensa-tion Taken together this provides evidence of a very small substitutionof performance-related pay for salary

The last four columns run identical tests but limit the sample to CEOsThe interaction of million 3 after continues to be small and negative inthe salary regressions In contrast the impact of 162(m) on total compen-sation appears to be positive and fairly substantial (about 15 percentagepoints per year) in these regressions That is the increase in bonus andstock options following the million-dollar rule more than offset the verysmall decline in salary

All of these conclusions should be interpreted somewhat cautiouslybecause our data provide us with only one pre-1993 control year Never-theless the data are consistent with section 162(m) having led to a veryminor slowdown in salary growth and one that is dramatically smallerthan what one would conclude without controlling for underlying differ-ential salary growth rates at different levels of salary Since total compen-sation for CEOs did not decline and perhaps increased there is evidenceof a minor substitution of performance-related pay for salary in responseto the regulation

8 EXECUTIVE COMPENSATION AND THEELASTICITY OF TAXABLE INCOME

Even if taxes have only a minor effect on executive compensation thestructure of executive compensation is important for tax policy be-cause high-income taxpayers are sources of signicant revenue and arethe focus of important debates about the efciency cost of taxation

Goolsbee (1999) has recently argued that executive decisions to exer-cise options are highly responsive to intertemporal differences in taxrates created by tax reforms He claims that much of the apparent declinein taxable income for high-income taxpayers between 1992 and 1993documented by Feldstein and Feenberg (1996) can be explained by execu-tives shifting option gains into 1992 in order to avoid the higher marginaltax rates that went into effect in 1993 By extension he implies that thehigh elasticities of taxable income with respect to personal tax ratesestimated by comparing the pre- and post-tax-reform taxable income of

Taxation of Executive Compensation 37

high-income individuals reect intertemporal shifting of income and notpermanent effects of taxation If this is true then the deadweight loss oftaxation is lower as is the cost of progressivity

While Goolsbee presents comprehensive evidence that option gainswere unusually high in 1992 his identication of the impact of tax rates onthe timing of option gains depends on correctly specifying a counter-factual level of options that would have been exercised in the absence ofthe tax change He simply assumes a linear time trend The purpose ofGoolsbeersquos time trend is presumably to account for the increasing use ofoptions over time However since options typically require a vestingperiod of at least a couple of years a much more direct measure of ex-pected option gains can be constructed based on the total value of vestedoptions held by the CEO Since options typically have vesting periods thatexceed two years this total can be treated as exogenous since it will not beaffected by the tax changes

More generally the value of options held by the CEO is likely to havea major effect on his exercise decisions If his options are out of themoney he will not have gains to realize If there has been a recent run-up in the market an optimizing CEO may exercise an unusually largeamount of options in order to diversify his overall portfolio and if in-the-money options are about to expire they will be exercised in thecurrent year

Our evidence suggests that much of the apparent tax shifting caninstead be attributed to stock market performance and the timing of pastoption grants Using our 15-year panel we replicate Goolsbeersquos (1999)result for the 1993 period but show that the impact of taxes is notpresent in other time periods Moreover even the result for the 1993period disappears when appropriate controls for past option grants andstock price appreciation are included

The top three rows of Table 7 present data on the taxable incomes ofCEOs in the HallndashLiebman sample For these results our panel has beennarrowed to include only CEOs whose rmrsquos scal years correspondwith tax years In addition we create a balanced panel for each of thethree tax-reform periods by excluding any rm for which there are notcomplete data for the period surrounding the reform

The fourth row of the table shows that the 1993 tax act raised the topmarginal personal income tax rate from 310 to 396 percent effective fortax year 1993 with a further increase occurring in 1994 due to the uncap-ping of the Medicare payroll tax The top row of the table shows thattaxable income was higher in 1992 than for any other year in our sample(all data are in 1994 dollars) Option gains were particularly high in1992mdashmore than double their 1991 level and 68 percent higher than their

TA

BL

E7

Tax

Ref

orm

Sto

ckM

arke

tP

erfo

rman

cea

ndth

eT

imin

gof

Opt

ion

Gai

ns

Ave

rage

stoc

km

arke

tM

ean

CE

OM

ean

CE

Ore

turn

for

Mea

nva

lue

taxa

ble

sala

ryan

dM

ean

CE

OT

opfe

der

al

rms

inof

opti

ons

(Val

ue

ofop

tion

Sam

ple

inco

me

bonu

sop

tion

gain

sp

erso

nal

sam

ple

for

held

atbe

gga

ins)

(va

lue

Ref

orm

size

Yea

r($

mil

)($

mil

)($

mil

)M

TR

pre

vye

ar(

)of

year

ofop

tion

she

ld)

ER

TA

8127

919

800

874

064

80

163

7025

80

460

354

1981

083

10

672

009

869

245

070

61

3919

820

872

067

90

099

5011

90

686

144

1983

099

00

724

025

250

355

116

62

16

TR

A86

295

1985

119

70

837

024

050

76

116

02

0719

861

448

091

70

345

5042

81

776

194

1987

165

50

959

059

338

519

12

229

266

1988

163

21

058

039

028

07

237

71

6419

891

656

105

20

414

2817

92

345

176

1993

272

1990

209

51

028

061

828

232

288

02

15T

axA

ct19

911

875

102

00

547

312

01

268

82

0319

922

678

107

01

250

3146

64

552

275

1993

232

41

209

074

339

620

64

299

173

1994

196

41

262

037

842

515

45

113

074

Sam

ple

isH

allndash

Lie

bman

rm

sw

ith

Dec

embe

r

scal

year

sth

atha

dco

mp

lete

dat

afo

ron

eof

the

thre

eta

x-re

form

per

iod

sst

ud

ied

Taxation of Executive Compensation 39

1994 level These patterns are consistent with Goolsbeersquos claim that tax-payers responded to the anticipated increase in marginal tax rates byshifting gains into 1992

The evidence from the other two tax reforms is less clear suggestingthat there might be factors other than tax rates determining the timing ofoption gains ERTA 1981 reduced the top marginal tax rate from 70percent in 1980 to 50 percent in 1982 However there is no sign thatoption gains were reduced in 1980 and 1981 and then increased in 198217

There was a big increase in option gains in 1983 however Similarly thehighest year of stock-option gains in the period around 1986 was 1987 astrange year to take gains since taxpayers would have known that thepersonal rate would be 10 percentage points lower if they waited untilthe following year18

What all the peak years of option gains have in common is that theyfollow years of strong stock market performance The fth row of Table 7shows the average stock market return during the previous year for therms in our sample The large value of option gains in 1983 followed a355-percent increase in the stock market The 1992 boom in option gainsfollowed a 466-percent increase in the stock market the more modest1987 peak in option gains followed a 19-percent gain (and a 43-percentgain two years before)

Strong stock market performance raises the value of options availableto exercise The sixth row of the table shows the mean value of optionsheld by CEOs at the beginning of the year All three of the peak years ofoption gains correspond with peak years in terms of the value of optionsavailable for exercise Since stock market appreciation during the yearalso increases the value of options available to exercise later in that yearwe need to allow for current-year stock market appreciation as well

Table 8 contains Goolsbee-style regressions for each of the tax re-forms The rst two columns reprint the results from Goolsbee (1999)The third and fourth columns replicate his results in our sample Inparticular with a specication similar to his we are able to replicate hisnding that the current after-tax share has a positive coefcient and thefollowing yearrsquos after-tax share has a coefcient of roughly negative oneColumns (5) through (8) show that this pattern was unique to the 1993tax reform Indeed in the TRA 86 regressions the coefcient on the

17 As we explained above for executives covered by the maximum tax on earned incomethere was not a decline in tax rates between 1981 and 198218 It is possible that the large stock market decline in October 1997 may have promptedsome people to exercise options earlier than they otherwise would have However anincrease in option exercises typically follows stock market increases not declines

TA

BL

E8

Att

empt

sto

Rep

lica

teth

eT

ax-S

hift

ing

Res

ult

inO

ther

Tim

eP

erio

ds

Goo

lsbe

eG

ools

bee

1993

1993

Ent

ire

Ent

ire

regr

essi

onre

gres

sion

refo

rmre

form

TR

A86

TR

A86

ER

TA

81E

RT

A81

1980

ndash199

419

80ndash1

994

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

ln[1

2ta

x(t)

]1

244

108

20

455

032

62

017

62

017

22

016

72

016

70

026

20

068

(01

34)

(01

27)

(05

37)

(05

42)

(02

10)

(02

09)

(00

90)

(00

91)

(00

67)

(00

65)

ln[1

2ta

x(t1

1)]

20

867

20

764

21

214

20

917

047

80

722

20

103

004

30

176

018

3(0

118

)(0

112

)(0

491

)(0

499

)(0

157

)(0

174

)(0

072

)(0

072

)(0

068

)(0

066

)

Tim

e0

135

007

62

000

52

036

40

041

000

80

094

003

50

056

002

7(0

008

)(0

008

)(0

049

)(0

049

)(0

028

)(0

028

)(0

025

)(0

026

)(0

003

)(0

003

)

ln(m

arke

tva

lue)

066

10

489

023

70

334

039

5(0

015

)(0

058

)(0

039

)(0

042

)(0

017

)

Yea

rs19

91ndash1

995

1991

ndash199

519

90ndash1

994

1990

ndash199

419

85ndash1

989

1985

ndash198

919

80ndash1

983

1980

ndash198

319

80ndash1

994

1980

ndash199

4

Ad

just

edR

20

720

770

610

630

690

760

800

810

630

68

Obs

erva

tion

s13

856

132

371

385

136

41

470

143

31

113

106

23

968

385

9

Dep

end

ent

vari

able

isth

elo

gof

taxa

ble

inco

me

All

regr

essi

ons

con

tain

rm

xe

def

fect

sG

ools

bee

regr

essi

on(2

)al

soco

ntai

nstw

oot

her

vari

able

sno

tsh

own

inth

eta

ble

Stan

dar

der

rors

inp

aren

thes

es

Taxation of Executive Compensation 41

TABLE 9Alternative Explanations for the Timing of Option Exercises

1993 Entirereform TRA 86 ERTA 81 1980ndash1994

(1) (2) (3) (4)

ln[12tax(t)] 0108 0111 20051 0201(0324) (0136) (0048) (0057)

ln[12tax(t11)] 20134 0573 0125 0150(0336) (0156) (0053) (0066)

ln(market value) 0418 0177 0338 0418(0061) (0039) (0046) (0163)

Firmrsquos stock market return (t) 0298 0268 0214 0258(0044) (0045) (0031) (0023)

Firmrsquos stock market return 0162 0206 0031 0128(t21) (0041) (0008) (0029) (0021)

ln(value of options held by 0004 0008 0001 0009CEO at start of year) (0003) (0003) (0002) (0001)

Years 1990ndash1994 1985ndash1989 1980ndash1983 1980ndash1994Adjusted R2 064 071 082 068Observations 1349 1427 1036 3812

The dependent variable is the log of taxable income All regressions contain rm xed effects Stan-dard errors in parentheses

following yearrsquos tax rate is positive and signicant Columns (9) and (10)pool all three reforms and fail to replicate the pattern that would reectintertemporal shifting19

It is possible that the ease with which one can shift the timing ofoption gains is asymmetric It may be harder to shift option gains for-ward in time since some options expire and therefore must be exercisednow In addition options were a much smaller portion of compensationin the early 1980s so perhaps the ERTA 81 results are not surprisingNonetheless these results cast doubt on the proposition that this is thechannel through which spurious estimates of the elasticity of taxableincome have occurred for previous tax reforms and they suggest thatwe should look harder for alternative explanations for the timing ofoption gains

Table 9 attempts to predict options gains using the lagged and currentincrease in a rmrsquos market value and the value of the options held by theCEO at the start of the year The two stock market variables stronglypredict option gains while the value of options held has the correct sign

19 The results also fail to suggest large permanent effects of marginal tax rates on taxableincome

42 Hall amp Liebman

in all of the specications and is signicant in the pooled sample eventhough it is highly correlated with the two stock market return variablesIn contrast the tax-timing explanation fails even in the 1993 period

It is important to realize that our results do not invalidate Goolsbeersquoscritique of claims that aggregate revenue data from 1992 and 1993 provethat elasticities of taxable income are large We have simply given analternative explanation for why taxable incomes were unusually high in1992 However our results do cast doubt on the conclusion that largetiming shifts are ubiquitous and capable of explaining away the morecareful estimates of taxable income elasticities that have been performedusing micro data (eg Feldstein 1995 Auten and Carroll 1995)

One further point is in order It is sometimes argued that the combina-tion of increased use of options and the booming stock market explainswhy federal tax revenues have been so much higher than expected dur-ing the 1990s However while large option gains increase revenue fromthe personal income tax they produce offsetting deductions for cor-porations The net impact on revenue is only the difference in the twomarginal tax rates times the option gain While the gap between thepersonal rate and the effective corporate rate is nontrivial the impact oflarge option exercises on total revenue is substantially smaller than theamount by which the exercises inate personal tax revenue and reducecorporate tax revenue

9 CONCLUSION

We have described the tax rules for executive pay in detail and analyzedhow changes in various tax rates affect the tax advantages of stock optionsrelative to salary and bonus We nd that there is a moderate tax advan-tage to optionsmdashcurrently about $4 per $100 in pretax executive paymdashand that changes in the tax advantage of options over time have had atmost a modest impact on the composition of pay Corporate governancefactors particularly the role of large investors appear to be more impor-tant in explaining the dramatic increase in option pay

In addition our evidence suggests that more direct attempts to use taxpolicy to inuence executive compensation have had little effect Wend that the million-dollar rule led companies to substitute perfor-mance-based pay for salary But our evidence suggests that this substitu-tion was quite modest and there is no evidence that the total level of paywas reduced Overall although the stock-option explosion has dramati-cally increased the link between pay and performance this change isdue almost entirely to nontax factors

Taxation of Executive Compensation 43

REFERENCES

Auerbach Alan J (1996) ldquoThe Dynamic Effects of Tax Law Asymmetriesrdquo Re-view of Economic Studies 53451ndash405

mdashmdashmdash and James Poterba (1987) ldquoTax Loss Carryforwards and Corporate TaxIncentivesrdquo In The Effects of Taxation on Capital Accumulation M Feldstein(ed) Chicago University of Chicago Press

Altshuler Rosanne and Alan Auerbach (1990) ldquoThe Signicance of Tax LawAsymmetries An Empirical Investigationrdquo Quarterly Journal of Economics 10561ndash86

Auten Gerald and Robert Carroll (1995) ldquoThe Effect of Income Taxes on House-hold Behaviorrdquo Ofce of Tax Analysis US Department of the Treasury

Baker George P and Brian J Hall (1998) ldquoCEO Incentives and Firm SizerdquoNBER Working Paper no 6868 December

Core John and Wayne Guay (1999) ldquoThe Use of Equity Grants to ManageOptimal Equity Incentive Levelsrdquo Draft

Feldstein M (1995) ldquoThe Effect of Marginal Tax Rates on Taxable Income APanel Study of the 1986 Tax Reform Actrdquo Journal of Political Economy 103 (no3)551ndash72

mdashmdashmdash and D Feenberg (1996) ldquoThe Effect of Increased Tax Rates on TaxableIncome and Economic Efciency A Preliminary Analysis of the 1993 Tax RateIncreasesrdquo In Tax Policy and the Economy vol 10 James M Poterba (ed)Cambridge MA MIT Press

Gompers Paul A and Andrew Metrick (1998) ldquoInstitutional Investors andEquity Pricesrdquo NBER Working Paper no 6723

Goolsbee Austan (1999) ldquoWhat Happens When You Tax the Rich Evidencefrom Executive Compensationrdquo NBER Working Paper no 6333

Graham John R (1996) ldquoProxies for the Corporate Marginal Tax Raterdquo Journal ofFinancial Economics 42187ndash221

Hall Brian J (1999) ldquoThe Design of Multi-year Stock Option Plansrdquo Journal ofApplied Corporate Finance 12(2)97ndash106

mdashmdashmdash and Jeffrey B Liebman (1998) ldquoAre CEOs Really Paid Like Bureau-cratsrdquo Quarterly Journal of Economics CXII(no 3)653ndash691

Jensen Michael C and William M Meckling (1976) ldquoTheory of the Firm Mana-gerial Behavior Agency Costs and Ownership Structurerdquo Journal of FinancialEconomics III305ndash360

mdashmdashmdash and Kevin J Murphy (1990) ldquoPerformance Pay and Top-ManagementIncentivesrdquo Journal of Political Economy XCVIII225ndash264

Kaplan Steven N (1997) ldquoThe Evolution of US Corporate Governance WeAre All Henry Kravis Nowrdquo Draft

Livingstone Jane R (1999) ldquoExecutive Compensation Contracting and Re-sponses to Increased Tax Costsrdquo Louisiana State University June

Miller Merton H and Myron S Scholes (1982) ldquoExecutive CompensationTaxes and Incentivesrdquo In Financial Economics Essays in Honor of Paul CootnerWilliam F Sharpe and Cathryn M Cootner (eds) Princeton NJ Prentice-Hall

Murphy Kevin J (1999) ldquoExecutive Compensationrdquo forthcoming in OrleyAshenfelter and David Card (eds) Handbook of Labor Economics Vol 3 NorthHolland

Perry Todd and Marc Zenner (1999) ldquoPay for Performance Government Regu-

TA

BL

E2

Tax

and

Acc

ount

ing

Tre

atm

ent

ofE

xecu

tive

Com

pens

atio

nIn

cent

ive

Stoc

kO

ptio

ns(I

SOs)

Res

tric

ted

Stoc

kan

dSt

ock

App

reci

atio

nR

ight

s(S

AR

s)

Tax

atio

nat

Tax

atio

nat

Acc

ount

ing

Com

pen

sati

onD

escr

ipti

onp

erso

nall

evel

corp

orat

ele

vel

trea

tmen

t

Ince

nti

vest

ock

ISO

sar

eid

enti

calt

oIf

the

usu

alco

ndit

ions

Ifth

eu

sual

cond

itio

nsA

sw

ith

NQ

SOs

ther

eop

tion

sN

QSO

sin

thei

rd

esig

n

are

met

th

ere

isno

tax

are

met

the

reis

gene

r-is

gene

rally

noex

pen

seH

owev

ert

hey

have

anat

gran

tor

exer

cise

At

ally

no

corp

orat

eta

xd

e-re

cogn

itio

nfo

rIS

Os

annu

alca

pof

$100

000

sale

all

opti

onp

ro

tsd

uct

ion

atgr

ant

exer

-(e

ver)

on

the

amou

ntth

atan

dsu

bseq

uen

tca

pita

lci

seo

rsa

le

vest

sT

he

tax

trea

t-ap

prec

iati

onar

eta

xed

men

tof

ISO

sal

soat

the

cap

ital

gain

sd

iffe

rs

rate

Res

tric

ted

stoc

kSi

mila

rto

stoc

kop

-T

hein

div

idu

alis

taxe

dT

heco

mp

any

gene

rall

yR

estr

icte

dst

ock

isge

n-ti

ons

but

shar

esat

the

ord

inar

yra

teas

rece

ives

ap

aral

leld

e-er

ally

exp

ense

dov

er(w

hich

vest

slow

ly)

the

rest

rict

ions

lap

se

du

ctio

neq

ual

toth

eth

eve

stin

gp

erio

dT

hera

ther

than

opti

ons

are

How

ever

the

man

ager

amou

ntof

the

exec

u-

amou

nt

exp

ense

dis

gran

ted

toth

eex

ecu

-m

aych

ose

tobe

taxe

dti

versquos

inco

me

whe

nth

eth

ed

iffe

renc

ebe

twee

nti

veU

nlik

eop

tion

sat

gran

td

ate

inw

hich

exec

uti

veis

taxe

dR

e-th

ecu

rren

tst

ock

pric

eth

esh

ares

typi

cally

case

alls

ubs

equ

ent

ap-

stri

cted

stoc

kis

gen

er-

and

the

exec

uti

versquos

have

voti

ngan

dd

ivi-

pre

ciat

ion

ista

xed

atal

lysu

bjec

tto

the

cost

(if

any)

d

end

righ

ts

the

cap

ital

gain

sra

te

mill

ion-

dol

lar

cap

D

ivid

end

sar

eta

xed

atth

eor

din

ary

rate

Stoc

kap

pre

ciat

ion

Rig

hts

that

repl

icat

eG

ener

ally

have

the

Gen

eral

lyha

veth

eU

nlik

eop

tion

sSA

Rs

righ

tsth

epr

ots

ofst

ock

op-

sam

eta

xtr

eatm

ent

assa

me

tax

trea

tmen

tas

are

mar

ked

tom

arke

tti

ons

Rar

ely

use

d

NQ

SOs

NQ

SOs

each

per

iod

an

dth

esi

nce

stoc

kop

tion

sd

iffe

renc

ebe

twee

nth

eha

vem

ore

favo

rabl

eac

-st

ock

pri

cean

dth

eex

-co

unt

ing

trea

tmen

ter

cise

pri

ceis

exp

ense

d(a

ndop

tion

sp

rovi

de

over

the

outs

tand

ing

the

sam

ebe

ne

tof

per

iod

ofSA

Rs

cash

less

exer

cise

thro

ugh

brok

erlo

ans

)

Not

eW

eth

ank

com

pen

sati

onco

nsul

tant

sSc

ott

Gre

enbe

rgan

dSc

ott

Ols

en(b

oth

ofT

ower

sP

erri

n)an

dFr

eder

icC

ook

(of

Fred

eric

WC

ook

and

Ass

ocia

tes)

for

hel

pfu

lcon

vers

atio

nsin

und

erst

and

ing

the

tax

and

acco

unt

ing

rule

s

14 Hall amp Liebman

The effect of taxation on the optimal contract (more precisely theshare of compensation that is performance-related) is ambiguous (evenin the absence of deductibility and deferral) because there are offsettingeffects First by reducing the share of corporate prots received byshareholders taxes diminish the importance to the shareholders of moti-vating the CEO Taken alone this effect would imply that taxes wouldbe expected to reduce the use of performance-based compensation Sec-ond because the government shares in the income received by the CEOthe variance of the CEOrsquos income is reduced raising his utility andlowering the cost to the rm of providing any given set of incentivesThird because taxes will lead executives to provide less effort for anygiven level of incentive-based pay the level of compensation that mustbe provided to compensate the CEO for effort is reduced for a givenamount of incentive These last two factors make it cheaper for the rmto offer contracts to the CEO with large incentive components when taxrates are higher and should therefore increase the use of incentive-based pay Since the net effect of taxation on the level of incentive-basedpay is ambiguous agency theory provides no strong predictions abouthow taxation should affect the optimal composition of option versuscash compensation

42 The Tax Advantages of DeferralWhile agency theory yields ambiguous predictions about the effect oftaxes on the use of options there are direct tax advantages of options inthat options provide executives with a way to defer compensation andthereby lower their taxes However because options also lead to a defer-ral of corporate tax deductibility the tax advantages from a combined(executive and corporate) perspective are not so straightforward In thissection we analyze and measure the tax advantages of stock optionsrelative to salary and bonus compensation We also analyze the taxadvantages of NQSOs relative to ISOs In particular we show how thetax advantages of options change as personal corporate and capitalgains taxes change We then show how the tax advantages of optionshave changed over time in response to changing tax rates

The crucial tax difference between standard options (NQSOs) andcash payment is that option payouts are deferred and the two forms ofcompensation earn different rates of return over the deferral period Anyanalysis of the relative tax advantages of two compensation instrumentsmust consider the tax consequences both to the employer and to theemployeemdashwhat Scholes and Wolfson (1992) call the global contractingperspective Thus in order to make valid comparisons between the twowe compare the tax burden to the executive while holding constant the

Taxation of Executive Compensation 15

posttax cost to the company By keeping posttax employer costs (in NPVterms) constant any package that is preferred by the employee is tax-advantaged in the global-contracting sense

Under this methodology a comparison of the tax advantages of op-tions and cash involves comparing a pretax cash payment of P with anoption payment that has the equivalent posttax NPV to the companyFor NQSOs it is straightforward to show that a pretax payment of P toan executive has exactly the same cost to the company as putting aside Pfor the purposes of paying stock-option payouts later That is if a com-pany pays an executive P today it will have the same amount of moneyin N years as if it had put aside P today let it accumulate at the rate ofreturn earned by the rm and then paid it out (as compensation in theform of option gains4) with any appreciation in year N

Assume that the pretax prot rate is r and the corporate tax rate is TcThen if the company pays P today it receives a deduction today of PTcwhich yields

PTc [1 1 r (1 2 Tc)]N (1)

in N years since the benets of the tax deduction are invested in thecompany and receive the after-tax corporate rate of return Converselyif the company puts P aside today then it grows at the after-tax corpo-rate rate of return in N years to give a tax deduction of

P [1 1 r (1 2 Tc)]N Tc (2)

which is the exact same value Note that the equivalence of these two taxdeductions is analogous to the tax-benet equivalence of front-loadedand back-loaded IRAs

An executiversquos payoff in N years from option prots is equal to Pcompounded at the after-tax corporate rate times 1 2 Tp where Tp is thepersonal tax rate Thus the combined payoffs in N years of the corporatededuction and the executiversquos posttax payoff is

P [1 1 r (1 2 Tc)]N Tc 1 P [1 1 r (1 2 Tc)]

N (1 2 Tp) (3)

where the rst term is the payoff from the corporate deduction and thesecond term is the executiversquos payoff

4 This analysis is not unique to options Any form of deferred compensation that enablesexecutives to invest inside the rm at a higher posttax return or without paying capitalgains taxes would have similar effects

16 Hall amp Liebman

The payoff to salary and bonus in N years is more complicated sincean assumption must be made about how the executive invests the origi-nal cash compensation For example if the executive invests in an instru-ment with taxable interest (eg bonds) then the N-year payoff from aninvestment of P accumulates at the post-corporate-tax post-personal-taxrate of return and is

P (1 2 Tp) [1 1 r (1 2 Tc)(1 2 Tp)]N (4)

However if an executive invests in (non-dividend-paying) equities theinvestment accumulates at the higher post-corporate-tax rate of returnwith the offsetting disadvantage that the capital gains are taxed at thecapital gains rate in year N So the executive receives

P [1 1 r (1 2 Tc)]N (1 2 Tp) 2 Tcg $P (1 2 Tp) [1 1 r (1 2 Tc)]

N 2P (1 2 Tp) (5)

where Tcg is the capital gains tax rate5Thus putting the equations together if an executive is paid cash P

and invests g in bonds and 1 2 g in equities the N-year payoff (includ-ing the payoff to the corporation from the tax deduction) is

P [1 1 r (1 2 Tc)]N Tc 1 g $(P [1 1 r (1 2 Tc)(1 2 Tp)]

N (1 2 Tp)1(1 2 g) (P [1 1 r (1 2 Tc)]

N (1 2 Tp) 2 Tcg $P (1 2 Tp)[1 1 r (1 2 Tc)]

N 2 P (1 2 Tp)) (6)

The tax advantage of options versus salary and bonus therefore is thedifference between the combined (corporate and executive) payoffs inequation (3) and the combined payoffs in equation (6)

43 The Tax Advantage of OptionsWe now turn to analysis of how changes in various tax rates affectthe tax advantages of options relative to cash compensation We focuson the case in which the cash earned by executives is invested entirelyin equities since equities are tax-favored relative to bonds The impor-tant conclusions of this analysis are not substantively different if we

5 In practice executives hold a combination of equities and bondlike instruments withtaxable interest Because equities are tax-favored in the next section we simplify the analy-sis by assuming that executives hold only equities Alternatively we could have assumedthat executives hold only less risky bonds since bonds more closely match the risk proleof a stream of tax savings None of our empirical results are substantively affected by thissimplication

Taxation of Executive Compensation 17

instead assume that a portion of an executiversquos holdings are invested inbonds

431 The Tax Advantage of Options Changing the Corporate TaxRate Stock options are tax-advantaged relative to cash simply becauseoption payouts are deferred allowing the executive to invest at the pre-personal-tax rate of return (with no capital gains tax at the end) ratherthan investing at the after-tax rate of return or paying the capital gainstax Since the advantage of being able to defer taxes is large when thecorporate tax rate is low the tax advantage of options is larger when thecorporate rate is low

To illustrate this point we dene the tax advantage of options to bethe dollar amount by which total after-tax option payoffs (to the corpora-tion and the executive) exceed the total payoffs from salary as denedby the difference of equation (3) and (6) assuming that P (the paymentto the executive) is equal to $100 and g 5 0 (cash compensation is in-vested in equities) We use a ten-year horizon since most options have aten-year term We then calculate how the tax advantage of optionschanges as the corporate tax rate varies from zero to 100 percent holdingthe personal tax rate and the capital gains rate xed at 40 percent and 20percent respectively (which approximates current rates)

The top panel of Figure 2 shows the tax advantage of options relativeto cashmdashprecisely the NPV of the total tax advantage of options for a$100 payment to the CEOmdashas the corporate tax rate changes As ex-pected the tax advantage of options declines as corporate tax rates in-crease At a zero corporate tax rate the tax advantage of options has anNPV of $7 This value is about $350 at a corporate rate of 40 percentWhen the corporate tax rate is 100 percent the tax advantage of optionscompletely disappears since the benets of deferred compensation fallto zero because the after-tax return on equities (and therefore the dis-count rate) falls to zero The tax advantage of options is the NPV of thetax savings from avoiding the capital gains tax on ten years of apprecia-tion which is why it is relatively modest even at the 40-percent corpo-rate rate

432 The Tax Advantage of Options Changing the Personal TaxRate Using the same assumptions as above (but this time holding thecorporate rate at 40 percent and the capital gains rate at 20 percent andvarying the personal rate) the tax advantage of options declines as thepersonal tax rate rises as shown in the middle panel of Figure 2 The taxadvantage of options is higher at low personal tax rates because theposttax base that gives rise to capital gains is higher and the advantage of

Note Holds constant the personal rate at 40 percent and the capital gains rate at 20 percent

Note Holds constant the corporate rate at 40 percent and the capital gains rate at 20 percent

Note Holds constant the personal rate at 40 percent and the corporate rate at 40 percent

FIGURE 2 Tax Advantage of Options NPV of Tax Advantage of Pay-ing $100 in Options (NQSOs) Rather Than in Cash

Taxation of Executive Compensation 19

avoiding the capital gains taxes is greater The tax advantage of optionstherefore declines linearly as the personal tax rate rises At a 100-percentpersonal tax rate the executive receives nothing in either case and byconstruction the companyrsquos tax deduction benet is the same in NPVterms so the tax advantage of options falls to zero also6

433 The Tax Advantage of Options Changing the Capital Gains TaxRate The tax advantage of options for various capital gains rates isdepicted in the bottom panel of Figure 2 under the same assumptions(this time holding both the personal rate and the corporate rate at 40percent while varying the capital gains rate) The tax advantage of avoid-ing the capital gains tax increases as the capital gains tax rate rises

5 HOW HAS THE TAX ADVANTAGE OF OPTIONSCHANGED OVER TIME

The analysis so far has illustrated how the tax advantage of optionsvaries with the changes in tax rates In this section we show how the taxadvantage of options has changed over time as tax rates have changed

51 Top Marginal Tax Rates over TimeThe evolution on the top marginal tax ratesmdashpersonal corporate andcapital gainsmdashfrom 1980 to 1998 is shown in Figure 3 In all cases thetop rate (the rate for taxpayers with the highest incomes) is shownwhich is not always the highest rate since various anomalies (such asclawbacks of exemptions) sometimes lead to marginal tax rates that arehigher than those for the highest income tax payers

The top personal tax rate was 70 percent in 1980 falling to 50 percentin 1982 and 28 percent in 1988 following the 1986 tax act7 The topmarginal tax rate has since risen to 396 percent but is effectively 425percent since there is a 29-percent Medicare surcharge (paid half by theemployer and half by the employee) that has no upper limit The topcorporate tax rate has had only one signicant change since 1980 fallingfrom 46 percent before the 1986 tax act to 34 percent following the act in1988 The top corporate tax rate was increased to 35 percent in 1993 The

6 This is one place where ignoring bond investments has substantive implications Inparticular in the all-bond case the tax advantage of options is nonmonotonicmdashit rises andthen fallsmdashas personal tax rates rise7 During the early 1980s the maximum tax on earned income limited the marginal tax rateon the earnings of high earners so many executives did not face a decline in personal taxrates between 1981 and 1982 The empirical results in this paper are robust to assumingthat executives faced a marginal tax rate of 50 percent in 1980 and 1981

20 Hall amp Liebman

FIGURE 3 Top Marginal Tax Rates since 1980

top capital gains rate has uctuated between 28 percent (approximately)and 20 percent since 1980 and is currently at 20 percent

52 The Evolution of the Tax Advantage of NQSOsThe tax advantage of options relative to cash dened in the same way asin section 4 from 1980 to 1998 is shown in Figure 4 The calculations aredone using the same assumptions as before (ie N 5 10 years and cashis invested in equities) and the statutory top marginal tax rates shown inFigure 3 are used Due to the fall in corporate and personal tax rates thetax advantage of options increased sharply after the Tax Reform Act of1986 This increase in tax advantage has largely been reversed in the1990s as top personal tax rates have crept back up and the capital gainsrate has been reduced

Two key facts emerge from this analysis First there is currently only amoderate tax advantage to standard nonqualied stock optionsmdashon theorder of $4 per $100 of compensation This is because the tax advantagesto the executive of deferring taxes are largely offset by the tax disad-vantages to the company of not being able to deduct option expensesfrom taxable prots until the executive exercises the options Secondalthough the 1986 tax act substantially increased the tax advantage ofoptions more than half of this increase has been reversed in the 1990s

Taxation of Executive Compensation 21

FIGURE 4 Tax Advantage of Options 1980 to 1998

NPV of tax advantage of $100 in options (NQSOs) rather than cash

53 The Tax (Dis)advantage of ISOsSo far the analysis has focused on the tax advantages of NQSOs sincethey are so much more prevalent than ISOs which account for onlyabout 5 percent of option grants The relative scarcity of ISOs can beexplained both by their tax status and by their per-executive cap of$100000 per year

Under what conditions are ISOs tax-preferred to NQSOs For a trans-fer P to an executive the ISO is always tax-preferred by the executive(since the capital gains rate is lower than the personal rate) while theNQSO is always tax-preferred from the companyrsquos perspective (since anISO is not deductible) The key issue however is the relative advan-tages of the two types of options from a global-contracting perspectiveBecause ISOs are not deductible the company is indifferent betweensetting aside P in the form of NQSOs today and setting aside P(1 2 Tc) inthe form of ISOs (Since the NPV of the tax deduction for NQSOs is PTcthe NPV of the payment P is P 2 PTc or P (1 2 Tc))

Equalizing the posttax cost of NQSOs and ISOs we need only look atthe payoff of the executive to determine the condition under which eachis preferred The payoff of NQSOs in N years is

P [1 1 r (1 2 Tc)]N (1 2 Tp) (7)

22 Hall amp Liebman

The same payoff for ISOs is

P (1 2 Tc) [1 1 r (1 2 Tc)]N (1 2 Tcg) (8)

ISOs therefore are tax-advantaged if

Tp 2 Tcg Tc 2 Tc Tcg (9)

Intuitively ISOs are tax-advantaged only if their advantage (the differ-ence between the personal rate and the capital gains rate) is largeenough to offset their cost (the disadvantage of not deducting at thecorporate rate) Note however that the condition is not a simple com-parison between the corporate rate and the personal rate minus thecapital gains rate (Tp 2 Tcg Tc)

Figure 5 shows how the tax advantage of NQSOs relative to ISOs haschanged since 1980 as tax rates have changed Unlike the very modestchanges (around 3 to 4 percent) in the relative tax advantages betweenoptions and cash shown in Figure 4 the relative tax advantage betweenNQSOs and ISOs has seen enormous swings When the top personalrate was 70 percent in 1980 ISOs were tax-favored by a margin greaterthan 6 percent However by 1982 (following the 1981 tax acts that dra-matically lowered the personal tax rate) ISOs became tax-disadvantaged

FIGURE 5 Tax Advantage of NQSOs Relative to ISOs over Time

NPV of tax advantage of $100 payment to executive NQSOs minus ISOs

Taxation of Executive Compensation 23

by about 5 percent By 1988 (following the 1986 tax act) ISOs became tax-disadvantaged by about 18 percent a dramatic swing The raising ofpersonal rates coupled with a decrease in the capital gains rate has sincereduced the tax disadvantage of ISOs to about $4 per $100 of pretaxcompensation Although hard data on ISOs are hard to come by practi-tioners (mostly compensation consultants) claim that ISOs were morepopular prior to 1982 when they were tax-advantaged Although ISOsare less disadvantageous from a tax standpoint relative to NQSOs thanthey were in the late 1980s they are still disadvantageous so it is notsurprising that they have not made a signicant resurgence in recentyears

6 EXPLAINING THE INCREASE IN OPTION-BASEDCOMPENSATION

In this section we analyze whether changes in tax rates affect the compo-sition of executive compensation and consider the relative importanceof tax factors and corporate governance factors in explaining the in-creasing share of compensation paid in the form of stock options Thenumerous changes in tax rates that have occurred since 1980 provide anopportunity to assess whether executive compensation arrangementsrespond to tax incentives in the way that the tax-avoidance model sug-gests We test this model using a panel data set of CEOs in largepublicly-traded US corporations Our identication of the tax effectsrelies on time-series variation in personal corporate and capital gainstax rates along with cross-sectional variation in corporate tax rates

61 DataWe use a 15-year panel data set of CEOs in the largest publicly-tradedUS corporations which is described in Hall and Liebman (1998) Thedata set covers the years from 1980 through 1994 and combines CEOcompensation information from corporate proxies and 10-K lings withstock price and stock return information from CRSP and accounting datafrom Compustat In addition some compensation data from the 1970swere collected in order to construct measures of the value of stock op-tions held by the CEOs in the rst period of the sample8

The feature that distinguishes our data from most other CEO datasets is that with our panel of yearly proxy data on option grants optiongains and total options held we are able to calculate the total value of

8 We are grateful to David Yermack for providing us with some of the data for 1984 to 1991See Yermack (1995) for a discussion of these data

24 Hall amp Liebman

all stock options held by the CEO at a point of time More importantlysince we have the details about the stock options held (number exer-cise price time to maturity etc) we can precisely calculate the changein the value of a CEOrsquos stock option holdings for a given change in rmvalue

62 Identication IssuesThe numerous tax changes over our sample period and the specic waysin which changes in personal corporate and capital gains rates arepredicted to affect the composition of CEO compensation give us un-usually rich sources of identication However because many of thechanges in tax rates would be expected to affect all of the CEOs in oursample in a similar way we will need to pay particular attention toseparating out the tax effects from underlying time trends and otherfactors that changed over time Controlling for underlying trends is par-ticularly important in this study because we are focusing on an out-comemdashthe use of stock optionsmdashthat has increased rapidly over thepast two decades and that many practitioners believe was importantlyaffected by nontax considerations Thus the tax effects we are trying toexplain are deviations from a rising trend

We take two steps to try to separate out the tax effects from theunderlying trend First we include key nontax factors that could poten-tially explain the increasing reliance on performance-based compensa-tion It has been suggested that the dramatic increase in incentive-basedpay is the result of the remarkably poor shareholder returns during the1970s which spurred the LBO and takeover movements of the 1980sAccording to this story (Kaplan 1997) shareholders became much morepowerful via the rise of institutional investors even as the LBO andtakeover movements waned Because the inuence of institutional inves-tors is thought to be one of the most important mechanisms of strongcorporate governance we use the share of each companyrsquos stock ownedby institutional investors (dened as institutions with more than $100million under management) as an explanatory variable9 The percentageof shares owned by large institutional investors increased from about 20percent to almost 50 percent in 1994 an upward trend that closelymatches the sharp rise in the share of compensation in the form ofoptions Indeed in our sample the annual average (over all of the rmsin our sample) share of stock owned by institutional investors has acorrelation of 09 with the annual average share of compensation paid inoptions

9 This variable is described in Gompers and Metrick (1998)

Taxation of Executive Compensation 25

In addition we include two other variables that proxy for strongercorporate governance The rst is the size of the board Evidence suggeststhat boards with smaller numbers of directors reduce the inuence of theCEO and better represent shareholder interests (Yermack 1995) Boardsize fell by approximately 15 percent between 1980 and 1994 The secondvariable is the fraction of the rmrsquos board members who are inside direc-tors or gray directors (noninsiders who have business dealings with thecompany) Firms with fewer outsiders are less likely to act in the share-holdersrsquo interests and we would predict that they would be less likely tohave performance-based compensation for their CEOs The average per-centage of inside and gray directors has decreased from about 45 percentin 1980 to about 35 percent in 1994

Second we allow for cross-sectional variation in corporate tax ratesCorporate marginal tax rates vary because rms may be eligible for tax-loss carrybacks and carryforwards investment tax credits and the alter-native minimum tax (see Auerbach 1986 Auerbach and Altshuler1990 Auerbach and Poterba 1987) Because these cross-sectional differ-ences imply that different rms should respond differently to a givenchange in tax rates they potentially provide us with a way to isolatethe tax effects Firm-specic corporate marginal tax rates are notori-ously difcult to calculate from publicly-available data We use a tricho-tomous variable that equals zero if the rm has tax-loss carryforwardsand negative earnings in a given year equals 05 times the statutorycorporate rate if the rm has only one of those conditions and equalsthe statutory rate if the rm has neither loss carryforwards nor negativeearnings10

The cross-sectional variation in the corporate tax rate may not be ex-ogenous Firms that perform poorly could face low marginal tax ratesand also be particularly likely or unlikely to provide performance-basedpay Because most of our specications use xed effects the correlationwe must be concerned about is between changes in rm performanceand changes in compensation We deal with this by controlling forlagged rm stock market performance in our regressions This shouldeliminate spurious correlation between tax rates and compensation thatis jointly caused by rm performance

10 Graham (1996) presents evidence that the trichotomous variable performs nearly as wellas his recommended simulated marginal tax rate Plesko (1999) compares marginal taxrates calculated from Compustat data with actual marginal tax rates from rm tax returnsHe nds that simple binary variables are more reliable measures of the marginal tax ratethan is either the Graham or the trichotomous variable We obtain substantively similarresults when we switch from the trichotomous variable to the binary variables recommendby Plesko

26 Hall amp Liebman

63 ResultsThe dependent variable in our regressions is the fraction of a CEOrsquos totalannual compensation (measured as the BlackndashScholes value of stock-option grants plus salary and bonus) that comes from stock-optiongrants We choose this reduced-form specication because our interest isin the composition of CEO compensation However the factors thataffect the composition of compensation might also affect the levels andit will be important to keep this in mind in interpreting our results

Optimal-contracting theory generally makes predictions about the in-centives provided by the total amount of rm stock and stock optionsowned by the CEO We choose to model the ow of option grants ratherthan the ultimate performance sensitivity of the CEOrsquos compensationand stock and stock-option holdings because we have in mind anadjustment-cost model in which it takes time for a CEOrsquos holdings ofstock and stock options to reach the optimal level11 In our regressionswe include two measures of the current performance sensitivity of theCEOrsquos wealth to rm performance as explanatory variables to reect thedistance of the CEOrsquos contract from the optimal contract

Table 3 presents results in which the tax effects are identied solely byvariation in tax rates over time The fraction of annual compensation paidin stock options is regressed on tax variables corporate governance vari-ables contract-theory variables and variables reecting stock market re-turns Both of the regressions in this table include rm xed effects Incolumn (1) the fraction in annual compensation paid in stock options isregressed on the log difference in payoff (the combined payoff to both therm and the CEO) from receiving compensation in stock options ratherthan salary and bonus where the payoff calculations are based on statu-tory marginal tax rates This variable is dened as the log difference be-tween equation (3) and equation (6) in section 4 The coefcient on the taxvariable is large and statistically signicant at the 95-percent level Thepoint estimate implies that a one-percent increase in the payoff differencebetween stock options and salary and bonus results in a 24-percentage-point increase in the share of compensation paid in stock options

The corporate governance variables all have the predicted sign Firmswith a higher fraction of their shares owned by institutional investorsuse more performance-based pay (stock options) Firms with large corpo-rate boards or a large share of inside directors are less likely to use stockoptions

The two contract-theory variables are the dollar change in CEO wealth

11 See Core and Guay (1999) for evidence that rms adjust in this way

TABLE 3Compensation Regressions that Rely on Variation in

Tax Rates over Time

(1) (2)

Tax variablesln (difference in payoff)(a) 2393 (0349)ln (12statutory personal 20015 (0034)

rate)ln (12statutory cap gains 0330 (0171)

rate)ln (12statutory corporate 0399 (0148)

rate)

Corporate governance variablesFraction of rmrsquos shares 0043 (029) 0052 (0028)

owned by institutionalinvestors

Size of the rmrsquos board of 20002 (001) 20003 (0001)directors

Fraction of directors who 20082 (033) 20092 (0030)are inside or gray

Contract-theory variablesD (CEO wealth) per $1000 20261 (0095) 20247 (0087)

in FV(b)

D (CEO wealth) per 10 in 2583 (189) 2554 (201)FV(c)

Volatility(d) 521 (424) 261 (369)ln (rm market value) 0019 (0006) 0019 (0006)

Stock market return variablesLagged annual rm stock 0006 (0009) 0003 (0008)

market returnLagged annual SampP 500 re- 20035 (0023) 20029 (0023)

turn

Regression diagnosticsTotal sample size 5179 5179Firms 427 427Adjusted R2 0313 0311

Dependent variable share of compensation paid in options Standard errors in parentheses Bothregressions in this table include rm xed effects

(a) Log difference in payoff is the log difference in combined rm and CEO after-tax share betweenpaying compensation in options and paying salary

(b) This variable is the dollar change in CEO wealth per $1000 change in rm market value (seeJensen and Murphy 1990)

(c) This variable in the dollar charge in CEO wealth per 10-percent change in rm value (see Hall andLiebman 1998)

(d) Standard deviation of rm market value in trillions

28 Hall amp Liebman

per $1000 change in rm value (Jensen and Murphy 1990) and thedollar change in CEO wealth per 10-percent change in rm value (Halland Liebman 1998) As discussed in Baker and Hall (1998) these twowealth sensitivity measures reect two different concepts of how closelyaligned the managersrsquo incentives are with the interests of the sharehold-ers In the regression they are meant to measure how far the CEOrsquosexisting contract and ownership of stock and stock options is from theoptimal level The negative coefcients on the two variables suggeststhat rms that are below their optimal pay-to-performance sensitivityare more likely to give more stock options in the current period

The second column replaces the tax variable motivated by deferraladvantages with the three statutory rates The corporate after-tax sharehas a positive coefcient as we would expect because the tax advantageof stock options increases as the corporate rate falls The coefcient onthe personal tax rate is small and statistically insignicant The sign ofthe coefcient on the capital gains rate is the opposite of that predictedby tax-avoidance theory

Note that these results rely on time-series variation alone This is not avery convincing test Since the log difference in payoffs mostly rises overthe time period covered in our sample as does the share of compensa-tion paid in options it is possible that the tax variable is simply reectingother factors that were changing over this time period that we have notincluded in our regression Therefore it is important to look at specica-tions that rely on cross-sectional variation in the tax variables

Table 4 contains specications that rely on cross-sectional variation incorporate tax rates The rst column repeats the specication from col-umn (1) of Table 3 but uses the trichotomous measures of rm corporatetax rates in place of the statutory rate in calculating each rmrsquos logdifference in payoff from paying compensation in options rather than insalary and bonus As in the previous table we nd the positive coef-cient on the log difference in payoff just as the theory suggest How-ever the coefcient estimate is now much smaller suggesting that aone-percent increase in the log difference in payoff produces a 01-percentage-point increase in the share of compensation paid in options

Introducing cross-sectional variation in the dependent variable doesnot eliminate the potentially spurious correlation over time between thetax variable and the share of compensation paid in options The secondcolumn of Table 4 isolates the pooled cross-sectional variation in corpo-rate tax rates by adding time dummies and dropping the rm xedeffects In this specication the coefcient on the tax variable is indistin-guishable from zero The coefcient on the institutional variable contin-ues to be large and statistically signicant The third column includes

Taxation of Executive Compensation 29

both rm xed effects and time dummies which tests for a relationshipbetween within-rm changes in the dependent variable and within-rmchanges in each explanatory variable In this specication the coefcienton the tax variable is again indistinguishable from zero The coefcientson the corporate governance variables are also statistically insignicant

The third column contains a specication that treats the Tax ReformAct of 1986 (TRA 86) as a natural experiment For rms and CEOs facingthe statutory corporate tax rate TRA 86 increased the relative payoff topaying compensation in options by 17 percent For a rm facing a zeromarginal tax rate it reduced the relative payoff by 45 percent The reduc-tion occurred because TRA 86 reduced the personal tax rate makingsalary and bonus more attractive while the decreased corporate rate hadno effect on zero-tax-rate rms Thus we would expect rms facing zeroor low marginal tax rates to reduce their use of options after TRA 86while rms facing the statutory rate would be expected to increase themFew rms face zero marginal tax rates year after year Therefore we tryto distinguish between rms that often face low tax rates and rms thatnearly always face the statutory rate We dene low-tax rms as oneswhose average marginal tax rate over 1984 to 1986 was below 02412

Then we ran a difference-in-differences regression to see if high-taxrms increased their use of options more than low-tax rms The interac-tion between post-TRA 86 and high corporate tax rates is the key vari-able The estimated coefcient on this variable is positive but small andnot statistically different from zero at the 95-percent level

It is important to note that in this specication as well the main corpo-rate governance variable the share of the rmrsquos shares owned by institu-tional investors performs as theory would predict The variable appearsto be quite robust Table 5 presents regressions of the share of compensa-tion paid in stock options on the fraction of the rmrsquos shares owned byinstitutional investors Since we have already established that these vari-ables have a strong longitudinal relationship the regressions here aredesigned to focus on the cross-sectional relationship between these twovariables The rst and third columns present regressions using data onlyfrom 1994 and nd a coefcient of roughly 02 (column 1 contains noother covariates while column 3 includes all of the other nontax covari-ates from the regressions in Tables 3 and 4) Since the median share of armrsquos stock owned by institutional investors in our sample rises from 19percent in 1980 to 49 percent in 1994 this coefcient would imply that this

12 This cutoff was chosen after inspecting the distribution of tax rates and essentiallyseparates rms that are always at or near the statutory rate from ones that are sometimesbelow it

TA

BL

E4

Com

pens

atio

nR

egre

ssio

nsth

atR

ely

onC

ross

-Sec

tion

alV

aria

tion

inC

orpo

rate

Tax

Rat

es

(1)

(2)

(3)

(4)

Tax

vari

able

s(a)

ln

(dif

fere

nce

inp

ayof

f)0

099

(00

41)

20

038

(00

36)

20

005

(00

44)

Pos

t-T

RA

862

000

5(0

013

)H

igh

corp

orat

eta

xra

te0

018

(00

17)

(Pos

t-T

RA

86)

3hi

gh0

024

(00

16)

corp

orat

eta

xra

te

Cor

por

ate

gove

rnan

ceva

riab

les

Frac

tion

of

rmrsquos

shar

es0

109

(00

27)

011

4(0

015

)2

000

9(0

030

)0

087

(00

34)

owne

dby

inst

itu

tion

alin

vest

ors

Size

ofth

e

rmrsquos

boar

dof

20

004

(00

01)

20

001

(00

01)

20

002

(00

01)

000

1(0

002

)d

irec

tors

Frac

tion

ofd

irec

tors

wh

o2

012

4(0

033

)2

000

3(0

018

)2

004

1(0

034

)0

001

(00

39)

are

insi

de

orgr

ay

Con

trac

t-th

eory

vari

able

sD

(CE

Ow

ealt

h)p

er$1

000

20

238

(00

94)

20

167

(00

53)

20

257

(00

96)

20

021

(01

47)

inFV

(b)

D(C

EO

wea

lth)

per

10

in2

560

(17

8)2

592

(10

5)2

594

(19

6)2

147

(04

86)

FV(c

)

Vol

atili

ty(d

)2

48(4

24)

162

(03

4)1

12(4

28)

106

(06

09)

ln(

rmm

arke

tva

lue)

003

3(

006)

000

8(0

003

)0

009

(00

06)

001

2(0

007

)

Stoc

km

arke

tre

turn

vari

able

sL

agge

dan

nual

rm

stoc

k2

000

2(0

009

)0

011

(00

10)

000

6(0

009

)2

001

2(0

013

)m

arke

tre

turn

Lag

ged

annu

alSamp

P50

02

005

6(0

023

)2

005

3(0

051

)2

003

9(0

049

)2

015

4(0

040

)re

turn

Reg

ress

ion

dia

gnos

tics

T

otal

sam

ple

size

5179

5179

5179

2534

Firm

s42

742

739

6A

dju

sted

R2

031

009

032

005

Firm

xed

effe

cts

Yes

No

Yes

No

Tim

ed

um

mie

sN

oY

esY

esN

o

Dep

end

ent

vari

able

isfr

acti

onof

ann

ualc

omp

ensa

tion

from

stoc

kop

tion

sSt

and

ard

erro

rsin

par

enth

eses

(a)

All

thre

esp

eci

cati

ons

use

tric

hoto

mou

sco

rpor

ate

tax

rate

sT

he

sam

ple

inco

lum

n(3

)u

ses

only

the

year

s19

84to

1988

sin

cew

ear

eu

sing

1986

asa

nat

ural

exp

erim

ent

(b)

Thi

sva

riab

leis

the

dol

lar

chan

gein

CE

Ow

ealt

hp

er$1

000

chan

gein

rm

mar

ketv

alu

e(s

eeJe

nse

nan

dM

urp

hy

1990

)

(c)

Thi

sva

riab

lein

the

dol

lar

char

gein

CE

Ow

ealt

hp

er10

-per

cent

chan

gein

rm

valu

e(s

eeH

alla

nd

Lie

bman

199

8)

(d)

Stan

dar

dd

evia

tion

ofr

mm

arke

tva

lue

intr

illio

ns

32 Hall amp Liebman

TABLE 5Large Institutional Investors and Stock-Option Compensation

The Cross-sectional Relationship between Institutional Investors andOption Grants

(1) (2) (3) (4)

Coefcient on fraction of 217 (067) 145 (014) 177 (069) 120 (015)rmrsquos shares owned byinstitutional investors

Sample period 1994 1980ndash1994 1994 1980ndash1994Contract-theory and stock No No Yes Yes

market covariatesTime dummies No Yes No Yes

The dependent variable is the share of compensation paid in options

factor alone can explain 6 percentage points of the 23-percentage-pointincrease in the median share of compensation paid in options that oc-curred over this time period Of course this variable is only one measureof the strengthening of corporate governance so the total impact of im-proved corporate governance could be greater The second and fourthcolumns of the table use the entire 1980ndash1994 sample and include timedummies to isolate the cross-sectional correlations Again there is evi-dence of a strong cross-sectional relationship between the two variablesalthough the coefcients are slightly smaller perhaps reecting a weakercorporate-governance role in the earlier period In sum there is a strongrelationship in both longitudinal and cross-sectional data between theinstitutional-investor variable and the options-percentage variable13

7 THE MILLION-DOLLAR RULE

In 1993 section 162(m) of the Internal Revenue Code was enactedlimiting the deductibility of executive compensation in excess of onemillion dollars unless the compensation was performance-related Pro-ponents of this legislation argued that it would slow the growth ofexecutive pay and tighten the link between rm performance and execu-tive compensation

The new provision became effective for tax year 1994 and applies tothe CEO and the other four most highly compensated executives in eachcompany Pay that is performance-related such as stock-option grants

13 The one specication in which the fraction owned by institutional investors is notsignicant is in column (2) of Table 4 a regression that includes both xed effects and timeeffects

Taxation of Executive Compensation 33

and bonuses paid for meeting clear performance goals is not affected bythis regulation provided that it has been approved by shareholdersThus the main impact of the provision is to limit the deductibility ofsalary in excess of a million dollars Woodlock and Antenucci (1997)studied the proxy statements of 376 rms and documented that mostrms responded to the new law by qualifying top executive pay asperformance-related Perry and Zenner (1999) identied 25 rms thatreduced salaries from above $1 million to below $1 million and foundthat 23 of the rms cited section 162(m) as their reason for doing so

In this section of the paper we investigate whether this provision hasaffected the level or structure of executive pay using the Execucompdata base which contains information on the highest-paid executives ineach of the rms in the SampP 500 SampP Midcap 400 and SampP SmallCap600 Between 1993 and 1998 the median salary of the Execucomp execu-tives rose from $230000 to $306000 The median value of total compen-sation (salary bonus and option grants) rose from $470000 to $882000Limiting the sample of CEOs indicates that the total compensation of themedian CEO rose from $11 million to $18 million over the period whilethe median salary rose from $441000 to $550000 Thus executive com-pensation clearly continued to rise at a rapid rate after the implementa-tion of the million-dollar rule though perhaps at a slower rate than itotherwise would have

If the regulation affects behavior it should have the largest impact onrms whose executives are receiving salaries that are above the limit orjust below it Thus we might expect to see rms whose executives arereceiving relatively high salaries to rely more heavily on performance-based pay and less heavily on salary in the years following the regulationPerry and Zenner (1999) and Livingston (1999) both nd that this is thecase This trend can be clearly identied in the Execucomp data set Forexample over 1993 to 1998 the median annualized growth rate of salaryfor executives whose 1993 salary was below $500000 in the previous yearwas 85 percent For executives with salary between $500000 and$700000 it was 64 percent for executives between $700000 and $850000it was 50 percent for executives between $750000 and $1 million it was31 percent and for executives above $1 million it was 00 percent14

The observation that executives at lower levels of salary had highergrowth rates is not very strong evidence of an impact of section 162(m)however because it is possible that low-salary executives would havehad higher growth rates even in the absence of the provision To explore

14 Of the 3137 executives for whom we have complete data between 1993 and 1998 2708had 1993 salaries below $500000 while only 35 had 1993 salaries above $1 million

34 Hall amp Liebman

whether this is the case we use Execucomp data from both before andafter the law change Ideally we would want to have data from manyyears before the change to examine whether there is an underlying pat-tern of faster salary growth for low-salary executives Unfortunately theExecucomp data begin in 1992 and therefore provide us with only the1992ndash1993 growth rate in compensation to use as a baseline15

We run regressions pooling data from 1992ndash1993 through 1997ndash1998of the annual percentage change in different forms of compensation on avariable million that is designed to measure the likelihood that anexecutiversquos compensation will be affected by section 162(m) We denemillion as the minimum of 1 and the executiversquos salary divided by $1million in the previous year Thus any executive with salary at or above$1 million would receive a value of 1 while an executive with a salary of$500000 would be coded as a 0516 We include this variable by itself andalso interacted with a dummy variable (after) that equals 1 for years1994 and after Thus the million variable is designed to pick up anyunderlying relationship between the level of salary and the growth rateof salary while the interacted one will identify any differential growthrate of salary by income in the period after section 162(m) took effect Wealso include a full set of year dummies the natural logarithm of thermrsquos market value and the rmrsquos annual rate of return over the twoprevious years as control variables To reduce the sensitivity of our re-sults to outliers we run quantile regressions and robust regressions

Table 6 contains our results The rst four columns are for all topexecutives while the last four columns focus on CEOs Columns (1) and(2) present estimates from robust regression for the percentage change insalary and the percentage change in total compensation (salary bonusstock grants and stock-option grants) In the salary growth regressionsfor all executives the coefcient on million is negative and signicantindicating that there is an underlying relationship between the level ofexecutive compensation and its subsequent growth rate The coefcienton million 3 after is negative as well suggesting that the negativerelationship between the level of salary and its subsequent growth rateintensied slightly after 1994 The relatively small magnitude suggeststhat an executive with a $1 million dollar salary would see his salarygrow at an annual rate that would be about 06 percent slower than anexecutive earning $500000

15 The HallndashLiebman data set contains only a combined salary and bonus variable makingit impossible to use it to analyze this issue16 We also ran the regressions (not reported) allowing for a nonlinear relationship betweenmillion and compensation growth and obtained substantively similar results

TA

BL

E6

The

Eff

ect

ofth

eM

illi

on-D

olla

rR

ule

onSa

lary

and

Tot

alC

ompe

nsat

ion

All

top

exec

uti

ves

CE

Os

only

Rob

ust

regr

essi

onQ

uan

tile

regr

essi

onR

obu

stre

gres

sion

Qu

anti

lere

gres

sion

D

in

Din

D

in

Din

D

in

Din

D

in

Din

sala

ryto

talc

omp

(a)

sala

ryto

talc

omp

sa

lary

tota

lcom

p

sala

ry

tota

lcom

p

mil

lio

n(b

)2

006

02

035

72

006

32

024

12

008

72

023

12

010

32

024

5(0

005

)(0

032

)(0

006

)(0

027

)(0

010

)(0

073

)(0

011

)(0

056

)m

illi

on

32

001

10

026

20

021

20

015

20

016

014

92

003

40

158

aft

er(c

)(0

005

)(0

033

)(0

006

)(0

029

)(0

010

)(0

076

)(0

011

)(0

057

)N

359

5035

950

359

5035

950

757

17

589

757

17

589

Stan

dar

der

rors

inp

aren

thes

esR

egre

ssio

nsal

soin

clu

de

year

du

mm

ies

stoc

km

arke

tre

turn

and

log

mar

ket

valu

e

(a)

Tot

alco

mp

ensa

tion

incl

ud

essa

lary

bon

us

rest

rict

edst

ock

and

stoc

k-op

tion

gran

ts

(b)

mil

lion

isd

ene

das

min

(1s

alar

yin

pre

viou

sye

ars

div

ided

by$1

mil

lion)

(c)

aft

eris

anin

dic

ator

vari

able

equ

alto

zero

befo

reth

ere

gula

tion

took

plac

e(1

993)

and

one

afte

rwar

d

36 Hall amp Liebman

In the second column the coefcient on million 3 after is essentiallyzero suggesting that any decrease in salary brought about by section162(m) was offset by increases in bonus and stock-option grants Theresults are similar for the median regressions There is some evidence ofa very small slowdown in growth of salary for executives likely to beaffected by section 162(m) but no sign of a change in total compensa-tion Taken together this provides evidence of a very small substitutionof performance-related pay for salary

The last four columns run identical tests but limit the sample to CEOsThe interaction of million 3 after continues to be small and negative inthe salary regressions In contrast the impact of 162(m) on total compen-sation appears to be positive and fairly substantial (about 15 percentagepoints per year) in these regressions That is the increase in bonus andstock options following the million-dollar rule more than offset the verysmall decline in salary

All of these conclusions should be interpreted somewhat cautiouslybecause our data provide us with only one pre-1993 control year Never-theless the data are consistent with section 162(m) having led to a veryminor slowdown in salary growth and one that is dramatically smallerthan what one would conclude without controlling for underlying differ-ential salary growth rates at different levels of salary Since total compen-sation for CEOs did not decline and perhaps increased there is evidenceof a minor substitution of performance-related pay for salary in responseto the regulation

8 EXECUTIVE COMPENSATION AND THEELASTICITY OF TAXABLE INCOME

Even if taxes have only a minor effect on executive compensation thestructure of executive compensation is important for tax policy be-cause high-income taxpayers are sources of signicant revenue and arethe focus of important debates about the efciency cost of taxation

Goolsbee (1999) has recently argued that executive decisions to exer-cise options are highly responsive to intertemporal differences in taxrates created by tax reforms He claims that much of the apparent declinein taxable income for high-income taxpayers between 1992 and 1993documented by Feldstein and Feenberg (1996) can be explained by execu-tives shifting option gains into 1992 in order to avoid the higher marginaltax rates that went into effect in 1993 By extension he implies that thehigh elasticities of taxable income with respect to personal tax ratesestimated by comparing the pre- and post-tax-reform taxable income of

Taxation of Executive Compensation 37

high-income individuals reect intertemporal shifting of income and notpermanent effects of taxation If this is true then the deadweight loss oftaxation is lower as is the cost of progressivity

While Goolsbee presents comprehensive evidence that option gainswere unusually high in 1992 his identication of the impact of tax rates onthe timing of option gains depends on correctly specifying a counter-factual level of options that would have been exercised in the absence ofthe tax change He simply assumes a linear time trend The purpose ofGoolsbeersquos time trend is presumably to account for the increasing use ofoptions over time However since options typically require a vestingperiod of at least a couple of years a much more direct measure of ex-pected option gains can be constructed based on the total value of vestedoptions held by the CEO Since options typically have vesting periods thatexceed two years this total can be treated as exogenous since it will not beaffected by the tax changes

More generally the value of options held by the CEO is likely to havea major effect on his exercise decisions If his options are out of themoney he will not have gains to realize If there has been a recent run-up in the market an optimizing CEO may exercise an unusually largeamount of options in order to diversify his overall portfolio and if in-the-money options are about to expire they will be exercised in thecurrent year

Our evidence suggests that much of the apparent tax shifting caninstead be attributed to stock market performance and the timing of pastoption grants Using our 15-year panel we replicate Goolsbeersquos (1999)result for the 1993 period but show that the impact of taxes is notpresent in other time periods Moreover even the result for the 1993period disappears when appropriate controls for past option grants andstock price appreciation are included

The top three rows of Table 7 present data on the taxable incomes ofCEOs in the HallndashLiebman sample For these results our panel has beennarrowed to include only CEOs whose rmrsquos scal years correspondwith tax years In addition we create a balanced panel for each of thethree tax-reform periods by excluding any rm for which there are notcomplete data for the period surrounding the reform

The fourth row of the table shows that the 1993 tax act raised the topmarginal personal income tax rate from 310 to 396 percent effective fortax year 1993 with a further increase occurring in 1994 due to the uncap-ping of the Medicare payroll tax The top row of the table shows thattaxable income was higher in 1992 than for any other year in our sample(all data are in 1994 dollars) Option gains were particularly high in1992mdashmore than double their 1991 level and 68 percent higher than their

TA

BL

E7

Tax

Ref

orm

Sto

ckM

arke

tP

erfo

rman

cea

ndth

eT

imin

gof

Opt

ion

Gai

ns

Ave

rage

stoc

km

arke

tM

ean

CE

OM

ean

CE

Ore

turn

for

Mea

nva

lue

taxa

ble

sala

ryan

dM

ean

CE

OT

opfe

der

al

rms

inof

opti

ons

(Val

ue

ofop

tion

Sam

ple

inco

me

bonu

sop

tion

gain

sp

erso

nal

sam

ple

for

held

atbe

gga

ins)

(va

lue

Ref

orm

size

Yea

r($

mil

)($

mil

)($

mil

)M

TR

pre

vye

ar(

)of

year

ofop

tion

she

ld)

ER

TA

8127

919

800

874

064

80

163

7025

80

460

354

1981

083

10

672

009

869

245

070

61

3919

820

872

067

90

099

5011

90

686

144

1983

099

00

724

025

250

355

116

62

16

TR

A86

295

1985

119

70

837

024

050

76

116

02

0719

861

448

091

70

345

5042

81

776

194

1987

165

50

959

059

338

519

12

229

266

1988

163

21

058

039

028

07

237

71

6419

891

656

105

20

414

2817

92

345

176

1993

272

1990

209

51

028

061

828

232

288

02

15T

axA

ct19

911

875

102

00

547

312

01

268

82

0319

922

678

107

01

250

3146

64

552

275

1993

232

41

209

074

339

620

64

299

173

1994

196

41

262

037

842

515

45

113

074

Sam

ple

isH

allndash

Lie

bman

rm

sw

ith

Dec

embe

r

scal

year

sth

atha

dco

mp

lete

dat

afo

ron

eof

the

thre

eta

x-re

form

per

iod

sst

ud

ied

Taxation of Executive Compensation 39

1994 level These patterns are consistent with Goolsbeersquos claim that tax-payers responded to the anticipated increase in marginal tax rates byshifting gains into 1992

The evidence from the other two tax reforms is less clear suggestingthat there might be factors other than tax rates determining the timing ofoption gains ERTA 1981 reduced the top marginal tax rate from 70percent in 1980 to 50 percent in 1982 However there is no sign thatoption gains were reduced in 1980 and 1981 and then increased in 198217

There was a big increase in option gains in 1983 however Similarly thehighest year of stock-option gains in the period around 1986 was 1987 astrange year to take gains since taxpayers would have known that thepersonal rate would be 10 percentage points lower if they waited untilthe following year18

What all the peak years of option gains have in common is that theyfollow years of strong stock market performance The fth row of Table 7shows the average stock market return during the previous year for therms in our sample The large value of option gains in 1983 followed a355-percent increase in the stock market The 1992 boom in option gainsfollowed a 466-percent increase in the stock market the more modest1987 peak in option gains followed a 19-percent gain (and a 43-percentgain two years before)

Strong stock market performance raises the value of options availableto exercise The sixth row of the table shows the mean value of optionsheld by CEOs at the beginning of the year All three of the peak years ofoption gains correspond with peak years in terms of the value of optionsavailable for exercise Since stock market appreciation during the yearalso increases the value of options available to exercise later in that yearwe need to allow for current-year stock market appreciation as well

Table 8 contains Goolsbee-style regressions for each of the tax re-forms The rst two columns reprint the results from Goolsbee (1999)The third and fourth columns replicate his results in our sample Inparticular with a specication similar to his we are able to replicate hisnding that the current after-tax share has a positive coefcient and thefollowing yearrsquos after-tax share has a coefcient of roughly negative oneColumns (5) through (8) show that this pattern was unique to the 1993tax reform Indeed in the TRA 86 regressions the coefcient on the

17 As we explained above for executives covered by the maximum tax on earned incomethere was not a decline in tax rates between 1981 and 198218 It is possible that the large stock market decline in October 1997 may have promptedsome people to exercise options earlier than they otherwise would have However anincrease in option exercises typically follows stock market increases not declines

TA

BL

E8

Att

empt

sto

Rep

lica

teth

eT

ax-S

hift

ing

Res

ult

inO

ther

Tim

eP

erio

ds

Goo

lsbe

eG

ools

bee

1993

1993

Ent

ire

Ent

ire

regr

essi

onre

gres

sion

refo

rmre

form

TR

A86

TR

A86

ER

TA

81E

RT

A81

1980

ndash199

419

80ndash1

994

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

ln[1

2ta

x(t)

]1

244

108

20

455

032

62

017

62

017

22

016

72

016

70

026

20

068

(01

34)

(01

27)

(05

37)

(05

42)

(02

10)

(02

09)

(00

90)

(00

91)

(00

67)

(00

65)

ln[1

2ta

x(t1

1)]

20

867

20

764

21

214

20

917

047

80

722

20

103

004

30

176

018

3(0

118

)(0

112

)(0

491

)(0

499

)(0

157

)(0

174

)(0

072

)(0

072

)(0

068

)(0

066

)

Tim

e0

135

007

62

000

52

036

40

041

000

80

094

003

50

056

002

7(0

008

)(0

008

)(0

049

)(0

049

)(0

028

)(0

028

)(0

025

)(0

026

)(0

003

)(0

003

)

ln(m

arke

tva

lue)

066

10

489

023

70

334

039

5(0

015

)(0

058

)(0

039

)(0

042

)(0

017

)

Yea

rs19

91ndash1

995

1991

ndash199

519

90ndash1

994

1990

ndash199

419

85ndash1

989

1985

ndash198

919

80ndash1

983

1980

ndash198

319

80ndash1

994

1980

ndash199

4

Ad

just

edR

20

720

770

610

630

690

760

800

810

630

68

Obs

erva

tion

s13

856

132

371

385

136

41

470

143

31

113

106

23

968

385

9

Dep

end

ent

vari

able

isth

elo

gof

taxa

ble

inco

me

All

regr

essi

ons

con

tain

rm

xe

def

fect

sG

ools

bee

regr

essi

on(2

)al

soco

ntai

nstw

oot

her

vari

able

sno

tsh

own

inth

eta

ble

Stan

dar

der

rors

inp

aren

thes

es

Taxation of Executive Compensation 41

TABLE 9Alternative Explanations for the Timing of Option Exercises

1993 Entirereform TRA 86 ERTA 81 1980ndash1994

(1) (2) (3) (4)

ln[12tax(t)] 0108 0111 20051 0201(0324) (0136) (0048) (0057)

ln[12tax(t11)] 20134 0573 0125 0150(0336) (0156) (0053) (0066)

ln(market value) 0418 0177 0338 0418(0061) (0039) (0046) (0163)

Firmrsquos stock market return (t) 0298 0268 0214 0258(0044) (0045) (0031) (0023)

Firmrsquos stock market return 0162 0206 0031 0128(t21) (0041) (0008) (0029) (0021)

ln(value of options held by 0004 0008 0001 0009CEO at start of year) (0003) (0003) (0002) (0001)

Years 1990ndash1994 1985ndash1989 1980ndash1983 1980ndash1994Adjusted R2 064 071 082 068Observations 1349 1427 1036 3812

The dependent variable is the log of taxable income All regressions contain rm xed effects Stan-dard errors in parentheses

following yearrsquos tax rate is positive and signicant Columns (9) and (10)pool all three reforms and fail to replicate the pattern that would reectintertemporal shifting19

It is possible that the ease with which one can shift the timing ofoption gains is asymmetric It may be harder to shift option gains for-ward in time since some options expire and therefore must be exercisednow In addition options were a much smaller portion of compensationin the early 1980s so perhaps the ERTA 81 results are not surprisingNonetheless these results cast doubt on the proposition that this is thechannel through which spurious estimates of the elasticity of taxableincome have occurred for previous tax reforms and they suggest thatwe should look harder for alternative explanations for the timing ofoption gains

Table 9 attempts to predict options gains using the lagged and currentincrease in a rmrsquos market value and the value of the options held by theCEO at the start of the year The two stock market variables stronglypredict option gains while the value of options held has the correct sign

19 The results also fail to suggest large permanent effects of marginal tax rates on taxableincome

42 Hall amp Liebman

in all of the specications and is signicant in the pooled sample eventhough it is highly correlated with the two stock market return variablesIn contrast the tax-timing explanation fails even in the 1993 period

It is important to realize that our results do not invalidate Goolsbeersquoscritique of claims that aggregate revenue data from 1992 and 1993 provethat elasticities of taxable income are large We have simply given analternative explanation for why taxable incomes were unusually high in1992 However our results do cast doubt on the conclusion that largetiming shifts are ubiquitous and capable of explaining away the morecareful estimates of taxable income elasticities that have been performedusing micro data (eg Feldstein 1995 Auten and Carroll 1995)

One further point is in order It is sometimes argued that the combina-tion of increased use of options and the booming stock market explainswhy federal tax revenues have been so much higher than expected dur-ing the 1990s However while large option gains increase revenue fromthe personal income tax they produce offsetting deductions for cor-porations The net impact on revenue is only the difference in the twomarginal tax rates times the option gain While the gap between thepersonal rate and the effective corporate rate is nontrivial the impact oflarge option exercises on total revenue is substantially smaller than theamount by which the exercises inate personal tax revenue and reducecorporate tax revenue

9 CONCLUSION

We have described the tax rules for executive pay in detail and analyzedhow changes in various tax rates affect the tax advantages of stock optionsrelative to salary and bonus We nd that there is a moderate tax advan-tage to optionsmdashcurrently about $4 per $100 in pretax executive paymdashand that changes in the tax advantage of options over time have had atmost a modest impact on the composition of pay Corporate governancefactors particularly the role of large investors appear to be more impor-tant in explaining the dramatic increase in option pay

In addition our evidence suggests that more direct attempts to use taxpolicy to inuence executive compensation have had little effect Wend that the million-dollar rule led companies to substitute perfor-mance-based pay for salary But our evidence suggests that this substitu-tion was quite modest and there is no evidence that the total level of paywas reduced Overall although the stock-option explosion has dramati-cally increased the link between pay and performance this change isdue almost entirely to nontax factors

Taxation of Executive Compensation 43

REFERENCES

Auerbach Alan J (1996) ldquoThe Dynamic Effects of Tax Law Asymmetriesrdquo Re-view of Economic Studies 53451ndash405

mdashmdashmdash and James Poterba (1987) ldquoTax Loss Carryforwards and Corporate TaxIncentivesrdquo In The Effects of Taxation on Capital Accumulation M Feldstein(ed) Chicago University of Chicago Press

Altshuler Rosanne and Alan Auerbach (1990) ldquoThe Signicance of Tax LawAsymmetries An Empirical Investigationrdquo Quarterly Journal of Economics 10561ndash86

Auten Gerald and Robert Carroll (1995) ldquoThe Effect of Income Taxes on House-hold Behaviorrdquo Ofce of Tax Analysis US Department of the Treasury

Baker George P and Brian J Hall (1998) ldquoCEO Incentives and Firm SizerdquoNBER Working Paper no 6868 December

Core John and Wayne Guay (1999) ldquoThe Use of Equity Grants to ManageOptimal Equity Incentive Levelsrdquo Draft

Feldstein M (1995) ldquoThe Effect of Marginal Tax Rates on Taxable Income APanel Study of the 1986 Tax Reform Actrdquo Journal of Political Economy 103 (no3)551ndash72

mdashmdashmdash and D Feenberg (1996) ldquoThe Effect of Increased Tax Rates on TaxableIncome and Economic Efciency A Preliminary Analysis of the 1993 Tax RateIncreasesrdquo In Tax Policy and the Economy vol 10 James M Poterba (ed)Cambridge MA MIT Press

Gompers Paul A and Andrew Metrick (1998) ldquoInstitutional Investors andEquity Pricesrdquo NBER Working Paper no 6723

Goolsbee Austan (1999) ldquoWhat Happens When You Tax the Rich Evidencefrom Executive Compensationrdquo NBER Working Paper no 6333

Graham John R (1996) ldquoProxies for the Corporate Marginal Tax Raterdquo Journal ofFinancial Economics 42187ndash221

Hall Brian J (1999) ldquoThe Design of Multi-year Stock Option Plansrdquo Journal ofApplied Corporate Finance 12(2)97ndash106

mdashmdashmdash and Jeffrey B Liebman (1998) ldquoAre CEOs Really Paid Like Bureau-cratsrdquo Quarterly Journal of Economics CXII(no 3)653ndash691

Jensen Michael C and William M Meckling (1976) ldquoTheory of the Firm Mana-gerial Behavior Agency Costs and Ownership Structurerdquo Journal of FinancialEconomics III305ndash360

mdashmdashmdash and Kevin J Murphy (1990) ldquoPerformance Pay and Top-ManagementIncentivesrdquo Journal of Political Economy XCVIII225ndash264

Kaplan Steven N (1997) ldquoThe Evolution of US Corporate Governance WeAre All Henry Kravis Nowrdquo Draft

Livingstone Jane R (1999) ldquoExecutive Compensation Contracting and Re-sponses to Increased Tax Costsrdquo Louisiana State University June

Miller Merton H and Myron S Scholes (1982) ldquoExecutive CompensationTaxes and Incentivesrdquo In Financial Economics Essays in Honor of Paul CootnerWilliam F Sharpe and Cathryn M Cootner (eds) Princeton NJ Prentice-Hall

Murphy Kevin J (1999) ldquoExecutive Compensationrdquo forthcoming in OrleyAshenfelter and David Card (eds) Handbook of Labor Economics Vol 3 NorthHolland

Perry Todd and Marc Zenner (1999) ldquoPay for Performance Government Regu-

Stoc

kap

pre

ciat

ion

Rig

hts

that

repl

icat

eG

ener

ally

have

the

Gen

eral

lyha

veth

eU

nlik

eop

tion

sSA

Rs

righ

tsth

epr

ots

ofst

ock

op-

sam

eta

xtr

eatm

ent

assa

me

tax

trea

tmen

tas

are

mar

ked

tom

arke

tti

ons

Rar

ely

use

d

NQ

SOs

NQ

SOs

each

per

iod

an

dth

esi

nce

stoc

kop

tion

sd

iffe

renc

ebe

twee

nth

eha

vem

ore

favo

rabl

eac

-st

ock

pri

cean

dth

eex

-co

unt

ing

trea

tmen

ter

cise

pri

ceis

exp

ense

d(a

ndop

tion

sp

rovi

de

over

the

outs

tand

ing

the

sam

ebe

ne

tof

per

iod

ofSA

Rs

cash

less

exer

cise

thro

ugh

brok

erlo

ans

)

Not

eW

eth

ank

com

pen

sati

onco

nsul

tant

sSc

ott

Gre

enbe

rgan

dSc

ott

Ols

en(b

oth

ofT

ower

sP

erri

n)an

dFr

eder

icC

ook

(of

Fred

eric

WC

ook

and

Ass

ocia

tes)

for

hel

pfu

lcon

vers

atio

nsin

und

erst

and

ing

the

tax

and

acco

unt

ing

rule

s

14 Hall amp Liebman

The effect of taxation on the optimal contract (more precisely theshare of compensation that is performance-related) is ambiguous (evenin the absence of deductibility and deferral) because there are offsettingeffects First by reducing the share of corporate prots received byshareholders taxes diminish the importance to the shareholders of moti-vating the CEO Taken alone this effect would imply that taxes wouldbe expected to reduce the use of performance-based compensation Sec-ond because the government shares in the income received by the CEOthe variance of the CEOrsquos income is reduced raising his utility andlowering the cost to the rm of providing any given set of incentivesThird because taxes will lead executives to provide less effort for anygiven level of incentive-based pay the level of compensation that mustbe provided to compensate the CEO for effort is reduced for a givenamount of incentive These last two factors make it cheaper for the rmto offer contracts to the CEO with large incentive components when taxrates are higher and should therefore increase the use of incentive-based pay Since the net effect of taxation on the level of incentive-basedpay is ambiguous agency theory provides no strong predictions abouthow taxation should affect the optimal composition of option versuscash compensation

42 The Tax Advantages of DeferralWhile agency theory yields ambiguous predictions about the effect oftaxes on the use of options there are direct tax advantages of options inthat options provide executives with a way to defer compensation andthereby lower their taxes However because options also lead to a defer-ral of corporate tax deductibility the tax advantages from a combined(executive and corporate) perspective are not so straightforward In thissection we analyze and measure the tax advantages of stock optionsrelative to salary and bonus compensation We also analyze the taxadvantages of NQSOs relative to ISOs In particular we show how thetax advantages of options change as personal corporate and capitalgains taxes change We then show how the tax advantages of optionshave changed over time in response to changing tax rates

The crucial tax difference between standard options (NQSOs) andcash payment is that option payouts are deferred and the two forms ofcompensation earn different rates of return over the deferral period Anyanalysis of the relative tax advantages of two compensation instrumentsmust consider the tax consequences both to the employer and to theemployeemdashwhat Scholes and Wolfson (1992) call the global contractingperspective Thus in order to make valid comparisons between the twowe compare the tax burden to the executive while holding constant the

Taxation of Executive Compensation 15

posttax cost to the company By keeping posttax employer costs (in NPVterms) constant any package that is preferred by the employee is tax-advantaged in the global-contracting sense

Under this methodology a comparison of the tax advantages of op-tions and cash involves comparing a pretax cash payment of P with anoption payment that has the equivalent posttax NPV to the companyFor NQSOs it is straightforward to show that a pretax payment of P toan executive has exactly the same cost to the company as putting aside Pfor the purposes of paying stock-option payouts later That is if a com-pany pays an executive P today it will have the same amount of moneyin N years as if it had put aside P today let it accumulate at the rate ofreturn earned by the rm and then paid it out (as compensation in theform of option gains4) with any appreciation in year N

Assume that the pretax prot rate is r and the corporate tax rate is TcThen if the company pays P today it receives a deduction today of PTcwhich yields

PTc [1 1 r (1 2 Tc)]N (1)

in N years since the benets of the tax deduction are invested in thecompany and receive the after-tax corporate rate of return Converselyif the company puts P aside today then it grows at the after-tax corpo-rate rate of return in N years to give a tax deduction of

P [1 1 r (1 2 Tc)]N Tc (2)

which is the exact same value Note that the equivalence of these two taxdeductions is analogous to the tax-benet equivalence of front-loadedand back-loaded IRAs

An executiversquos payoff in N years from option prots is equal to Pcompounded at the after-tax corporate rate times 1 2 Tp where Tp is thepersonal tax rate Thus the combined payoffs in N years of the corporatededuction and the executiversquos posttax payoff is

P [1 1 r (1 2 Tc)]N Tc 1 P [1 1 r (1 2 Tc)]

N (1 2 Tp) (3)

where the rst term is the payoff from the corporate deduction and thesecond term is the executiversquos payoff

4 This analysis is not unique to options Any form of deferred compensation that enablesexecutives to invest inside the rm at a higher posttax return or without paying capitalgains taxes would have similar effects

16 Hall amp Liebman

The payoff to salary and bonus in N years is more complicated sincean assumption must be made about how the executive invests the origi-nal cash compensation For example if the executive invests in an instru-ment with taxable interest (eg bonds) then the N-year payoff from aninvestment of P accumulates at the post-corporate-tax post-personal-taxrate of return and is

P (1 2 Tp) [1 1 r (1 2 Tc)(1 2 Tp)]N (4)

However if an executive invests in (non-dividend-paying) equities theinvestment accumulates at the higher post-corporate-tax rate of returnwith the offsetting disadvantage that the capital gains are taxed at thecapital gains rate in year N So the executive receives

P [1 1 r (1 2 Tc)]N (1 2 Tp) 2 Tcg $P (1 2 Tp) [1 1 r (1 2 Tc)]

N 2P (1 2 Tp) (5)

where Tcg is the capital gains tax rate5Thus putting the equations together if an executive is paid cash P

and invests g in bonds and 1 2 g in equities the N-year payoff (includ-ing the payoff to the corporation from the tax deduction) is

P [1 1 r (1 2 Tc)]N Tc 1 g $(P [1 1 r (1 2 Tc)(1 2 Tp)]

N (1 2 Tp)1(1 2 g) (P [1 1 r (1 2 Tc)]

N (1 2 Tp) 2 Tcg $P (1 2 Tp)[1 1 r (1 2 Tc)]

N 2 P (1 2 Tp)) (6)

The tax advantage of options versus salary and bonus therefore is thedifference between the combined (corporate and executive) payoffs inequation (3) and the combined payoffs in equation (6)

43 The Tax Advantage of OptionsWe now turn to analysis of how changes in various tax rates affectthe tax advantages of options relative to cash compensation We focuson the case in which the cash earned by executives is invested entirelyin equities since equities are tax-favored relative to bonds The impor-tant conclusions of this analysis are not substantively different if we

5 In practice executives hold a combination of equities and bondlike instruments withtaxable interest Because equities are tax-favored in the next section we simplify the analy-sis by assuming that executives hold only equities Alternatively we could have assumedthat executives hold only less risky bonds since bonds more closely match the risk proleof a stream of tax savings None of our empirical results are substantively affected by thissimplication

Taxation of Executive Compensation 17

instead assume that a portion of an executiversquos holdings are invested inbonds

431 The Tax Advantage of Options Changing the Corporate TaxRate Stock options are tax-advantaged relative to cash simply becauseoption payouts are deferred allowing the executive to invest at the pre-personal-tax rate of return (with no capital gains tax at the end) ratherthan investing at the after-tax rate of return or paying the capital gainstax Since the advantage of being able to defer taxes is large when thecorporate tax rate is low the tax advantage of options is larger when thecorporate rate is low

To illustrate this point we dene the tax advantage of options to bethe dollar amount by which total after-tax option payoffs (to the corpora-tion and the executive) exceed the total payoffs from salary as denedby the difference of equation (3) and (6) assuming that P (the paymentto the executive) is equal to $100 and g 5 0 (cash compensation is in-vested in equities) We use a ten-year horizon since most options have aten-year term We then calculate how the tax advantage of optionschanges as the corporate tax rate varies from zero to 100 percent holdingthe personal tax rate and the capital gains rate xed at 40 percent and 20percent respectively (which approximates current rates)

The top panel of Figure 2 shows the tax advantage of options relativeto cashmdashprecisely the NPV of the total tax advantage of options for a$100 payment to the CEOmdashas the corporate tax rate changes As ex-pected the tax advantage of options declines as corporate tax rates in-crease At a zero corporate tax rate the tax advantage of options has anNPV of $7 This value is about $350 at a corporate rate of 40 percentWhen the corporate tax rate is 100 percent the tax advantage of optionscompletely disappears since the benets of deferred compensation fallto zero because the after-tax return on equities (and therefore the dis-count rate) falls to zero The tax advantage of options is the NPV of thetax savings from avoiding the capital gains tax on ten years of apprecia-tion which is why it is relatively modest even at the 40-percent corpo-rate rate

432 The Tax Advantage of Options Changing the Personal TaxRate Using the same assumptions as above (but this time holding thecorporate rate at 40 percent and the capital gains rate at 20 percent andvarying the personal rate) the tax advantage of options declines as thepersonal tax rate rises as shown in the middle panel of Figure 2 The taxadvantage of options is higher at low personal tax rates because theposttax base that gives rise to capital gains is higher and the advantage of

Note Holds constant the personal rate at 40 percent and the capital gains rate at 20 percent

Note Holds constant the corporate rate at 40 percent and the capital gains rate at 20 percent

Note Holds constant the personal rate at 40 percent and the corporate rate at 40 percent

FIGURE 2 Tax Advantage of Options NPV of Tax Advantage of Pay-ing $100 in Options (NQSOs) Rather Than in Cash

Taxation of Executive Compensation 19

avoiding the capital gains taxes is greater The tax advantage of optionstherefore declines linearly as the personal tax rate rises At a 100-percentpersonal tax rate the executive receives nothing in either case and byconstruction the companyrsquos tax deduction benet is the same in NPVterms so the tax advantage of options falls to zero also6

433 The Tax Advantage of Options Changing the Capital Gains TaxRate The tax advantage of options for various capital gains rates isdepicted in the bottom panel of Figure 2 under the same assumptions(this time holding both the personal rate and the corporate rate at 40percent while varying the capital gains rate) The tax advantage of avoid-ing the capital gains tax increases as the capital gains tax rate rises

5 HOW HAS THE TAX ADVANTAGE OF OPTIONSCHANGED OVER TIME

The analysis so far has illustrated how the tax advantage of optionsvaries with the changes in tax rates In this section we show how the taxadvantage of options has changed over time as tax rates have changed

51 Top Marginal Tax Rates over TimeThe evolution on the top marginal tax ratesmdashpersonal corporate andcapital gainsmdashfrom 1980 to 1998 is shown in Figure 3 In all cases thetop rate (the rate for taxpayers with the highest incomes) is shownwhich is not always the highest rate since various anomalies (such asclawbacks of exemptions) sometimes lead to marginal tax rates that arehigher than those for the highest income tax payers

The top personal tax rate was 70 percent in 1980 falling to 50 percentin 1982 and 28 percent in 1988 following the 1986 tax act7 The topmarginal tax rate has since risen to 396 percent but is effectively 425percent since there is a 29-percent Medicare surcharge (paid half by theemployer and half by the employee) that has no upper limit The topcorporate tax rate has had only one signicant change since 1980 fallingfrom 46 percent before the 1986 tax act to 34 percent following the act in1988 The top corporate tax rate was increased to 35 percent in 1993 The

6 This is one place where ignoring bond investments has substantive implications Inparticular in the all-bond case the tax advantage of options is nonmonotonicmdashit rises andthen fallsmdashas personal tax rates rise7 During the early 1980s the maximum tax on earned income limited the marginal tax rateon the earnings of high earners so many executives did not face a decline in personal taxrates between 1981 and 1982 The empirical results in this paper are robust to assumingthat executives faced a marginal tax rate of 50 percent in 1980 and 1981

20 Hall amp Liebman

FIGURE 3 Top Marginal Tax Rates since 1980

top capital gains rate has uctuated between 28 percent (approximately)and 20 percent since 1980 and is currently at 20 percent

52 The Evolution of the Tax Advantage of NQSOsThe tax advantage of options relative to cash dened in the same way asin section 4 from 1980 to 1998 is shown in Figure 4 The calculations aredone using the same assumptions as before (ie N 5 10 years and cashis invested in equities) and the statutory top marginal tax rates shown inFigure 3 are used Due to the fall in corporate and personal tax rates thetax advantage of options increased sharply after the Tax Reform Act of1986 This increase in tax advantage has largely been reversed in the1990s as top personal tax rates have crept back up and the capital gainsrate has been reduced

Two key facts emerge from this analysis First there is currently only amoderate tax advantage to standard nonqualied stock optionsmdashon theorder of $4 per $100 of compensation This is because the tax advantagesto the executive of deferring taxes are largely offset by the tax disad-vantages to the company of not being able to deduct option expensesfrom taxable prots until the executive exercises the options Secondalthough the 1986 tax act substantially increased the tax advantage ofoptions more than half of this increase has been reversed in the 1990s

Taxation of Executive Compensation 21

FIGURE 4 Tax Advantage of Options 1980 to 1998

NPV of tax advantage of $100 in options (NQSOs) rather than cash

53 The Tax (Dis)advantage of ISOsSo far the analysis has focused on the tax advantages of NQSOs sincethey are so much more prevalent than ISOs which account for onlyabout 5 percent of option grants The relative scarcity of ISOs can beexplained both by their tax status and by their per-executive cap of$100000 per year

Under what conditions are ISOs tax-preferred to NQSOs For a trans-fer P to an executive the ISO is always tax-preferred by the executive(since the capital gains rate is lower than the personal rate) while theNQSO is always tax-preferred from the companyrsquos perspective (since anISO is not deductible) The key issue however is the relative advan-tages of the two types of options from a global-contracting perspectiveBecause ISOs are not deductible the company is indifferent betweensetting aside P in the form of NQSOs today and setting aside P(1 2 Tc) inthe form of ISOs (Since the NPV of the tax deduction for NQSOs is PTcthe NPV of the payment P is P 2 PTc or P (1 2 Tc))

Equalizing the posttax cost of NQSOs and ISOs we need only look atthe payoff of the executive to determine the condition under which eachis preferred The payoff of NQSOs in N years is

P [1 1 r (1 2 Tc)]N (1 2 Tp) (7)

22 Hall amp Liebman

The same payoff for ISOs is

P (1 2 Tc) [1 1 r (1 2 Tc)]N (1 2 Tcg) (8)

ISOs therefore are tax-advantaged if

Tp 2 Tcg Tc 2 Tc Tcg (9)

Intuitively ISOs are tax-advantaged only if their advantage (the differ-ence between the personal rate and the capital gains rate) is largeenough to offset their cost (the disadvantage of not deducting at thecorporate rate) Note however that the condition is not a simple com-parison between the corporate rate and the personal rate minus thecapital gains rate (Tp 2 Tcg Tc)

Figure 5 shows how the tax advantage of NQSOs relative to ISOs haschanged since 1980 as tax rates have changed Unlike the very modestchanges (around 3 to 4 percent) in the relative tax advantages betweenoptions and cash shown in Figure 4 the relative tax advantage betweenNQSOs and ISOs has seen enormous swings When the top personalrate was 70 percent in 1980 ISOs were tax-favored by a margin greaterthan 6 percent However by 1982 (following the 1981 tax acts that dra-matically lowered the personal tax rate) ISOs became tax-disadvantaged

FIGURE 5 Tax Advantage of NQSOs Relative to ISOs over Time

NPV of tax advantage of $100 payment to executive NQSOs minus ISOs

Taxation of Executive Compensation 23

by about 5 percent By 1988 (following the 1986 tax act) ISOs became tax-disadvantaged by about 18 percent a dramatic swing The raising ofpersonal rates coupled with a decrease in the capital gains rate has sincereduced the tax disadvantage of ISOs to about $4 per $100 of pretaxcompensation Although hard data on ISOs are hard to come by practi-tioners (mostly compensation consultants) claim that ISOs were morepopular prior to 1982 when they were tax-advantaged Although ISOsare less disadvantageous from a tax standpoint relative to NQSOs thanthey were in the late 1980s they are still disadvantageous so it is notsurprising that they have not made a signicant resurgence in recentyears

6 EXPLAINING THE INCREASE IN OPTION-BASEDCOMPENSATION

In this section we analyze whether changes in tax rates affect the compo-sition of executive compensation and consider the relative importanceof tax factors and corporate governance factors in explaining the in-creasing share of compensation paid in the form of stock options Thenumerous changes in tax rates that have occurred since 1980 provide anopportunity to assess whether executive compensation arrangementsrespond to tax incentives in the way that the tax-avoidance model sug-gests We test this model using a panel data set of CEOs in largepublicly-traded US corporations Our identication of the tax effectsrelies on time-series variation in personal corporate and capital gainstax rates along with cross-sectional variation in corporate tax rates

61 DataWe use a 15-year panel data set of CEOs in the largest publicly-tradedUS corporations which is described in Hall and Liebman (1998) Thedata set covers the years from 1980 through 1994 and combines CEOcompensation information from corporate proxies and 10-K lings withstock price and stock return information from CRSP and accounting datafrom Compustat In addition some compensation data from the 1970swere collected in order to construct measures of the value of stock op-tions held by the CEOs in the rst period of the sample8

The feature that distinguishes our data from most other CEO datasets is that with our panel of yearly proxy data on option grants optiongains and total options held we are able to calculate the total value of

8 We are grateful to David Yermack for providing us with some of the data for 1984 to 1991See Yermack (1995) for a discussion of these data

24 Hall amp Liebman

all stock options held by the CEO at a point of time More importantlysince we have the details about the stock options held (number exer-cise price time to maturity etc) we can precisely calculate the changein the value of a CEOrsquos stock option holdings for a given change in rmvalue

62 Identication IssuesThe numerous tax changes over our sample period and the specic waysin which changes in personal corporate and capital gains rates arepredicted to affect the composition of CEO compensation give us un-usually rich sources of identication However because many of thechanges in tax rates would be expected to affect all of the CEOs in oursample in a similar way we will need to pay particular attention toseparating out the tax effects from underlying time trends and otherfactors that changed over time Controlling for underlying trends is par-ticularly important in this study because we are focusing on an out-comemdashthe use of stock optionsmdashthat has increased rapidly over thepast two decades and that many practitioners believe was importantlyaffected by nontax considerations Thus the tax effects we are trying toexplain are deviations from a rising trend

We take two steps to try to separate out the tax effects from theunderlying trend First we include key nontax factors that could poten-tially explain the increasing reliance on performance-based compensa-tion It has been suggested that the dramatic increase in incentive-basedpay is the result of the remarkably poor shareholder returns during the1970s which spurred the LBO and takeover movements of the 1980sAccording to this story (Kaplan 1997) shareholders became much morepowerful via the rise of institutional investors even as the LBO andtakeover movements waned Because the inuence of institutional inves-tors is thought to be one of the most important mechanisms of strongcorporate governance we use the share of each companyrsquos stock ownedby institutional investors (dened as institutions with more than $100million under management) as an explanatory variable9 The percentageof shares owned by large institutional investors increased from about 20percent to almost 50 percent in 1994 an upward trend that closelymatches the sharp rise in the share of compensation in the form ofoptions Indeed in our sample the annual average (over all of the rmsin our sample) share of stock owned by institutional investors has acorrelation of 09 with the annual average share of compensation paid inoptions

9 This variable is described in Gompers and Metrick (1998)

Taxation of Executive Compensation 25

In addition we include two other variables that proxy for strongercorporate governance The rst is the size of the board Evidence suggeststhat boards with smaller numbers of directors reduce the inuence of theCEO and better represent shareholder interests (Yermack 1995) Boardsize fell by approximately 15 percent between 1980 and 1994 The secondvariable is the fraction of the rmrsquos board members who are inside direc-tors or gray directors (noninsiders who have business dealings with thecompany) Firms with fewer outsiders are less likely to act in the share-holdersrsquo interests and we would predict that they would be less likely tohave performance-based compensation for their CEOs The average per-centage of inside and gray directors has decreased from about 45 percentin 1980 to about 35 percent in 1994

Second we allow for cross-sectional variation in corporate tax ratesCorporate marginal tax rates vary because rms may be eligible for tax-loss carrybacks and carryforwards investment tax credits and the alter-native minimum tax (see Auerbach 1986 Auerbach and Altshuler1990 Auerbach and Poterba 1987) Because these cross-sectional differ-ences imply that different rms should respond differently to a givenchange in tax rates they potentially provide us with a way to isolatethe tax effects Firm-specic corporate marginal tax rates are notori-ously difcult to calculate from publicly-available data We use a tricho-tomous variable that equals zero if the rm has tax-loss carryforwardsand negative earnings in a given year equals 05 times the statutorycorporate rate if the rm has only one of those conditions and equalsthe statutory rate if the rm has neither loss carryforwards nor negativeearnings10

The cross-sectional variation in the corporate tax rate may not be ex-ogenous Firms that perform poorly could face low marginal tax ratesand also be particularly likely or unlikely to provide performance-basedpay Because most of our specications use xed effects the correlationwe must be concerned about is between changes in rm performanceand changes in compensation We deal with this by controlling forlagged rm stock market performance in our regressions This shouldeliminate spurious correlation between tax rates and compensation thatis jointly caused by rm performance

10 Graham (1996) presents evidence that the trichotomous variable performs nearly as wellas his recommended simulated marginal tax rate Plesko (1999) compares marginal taxrates calculated from Compustat data with actual marginal tax rates from rm tax returnsHe nds that simple binary variables are more reliable measures of the marginal tax ratethan is either the Graham or the trichotomous variable We obtain substantively similarresults when we switch from the trichotomous variable to the binary variables recommendby Plesko

26 Hall amp Liebman

63 ResultsThe dependent variable in our regressions is the fraction of a CEOrsquos totalannual compensation (measured as the BlackndashScholes value of stock-option grants plus salary and bonus) that comes from stock-optiongrants We choose this reduced-form specication because our interest isin the composition of CEO compensation However the factors thataffect the composition of compensation might also affect the levels andit will be important to keep this in mind in interpreting our results

Optimal-contracting theory generally makes predictions about the in-centives provided by the total amount of rm stock and stock optionsowned by the CEO We choose to model the ow of option grants ratherthan the ultimate performance sensitivity of the CEOrsquos compensationand stock and stock-option holdings because we have in mind anadjustment-cost model in which it takes time for a CEOrsquos holdings ofstock and stock options to reach the optimal level11 In our regressionswe include two measures of the current performance sensitivity of theCEOrsquos wealth to rm performance as explanatory variables to reect thedistance of the CEOrsquos contract from the optimal contract

Table 3 presents results in which the tax effects are identied solely byvariation in tax rates over time The fraction of annual compensation paidin stock options is regressed on tax variables corporate governance vari-ables contract-theory variables and variables reecting stock market re-turns Both of the regressions in this table include rm xed effects Incolumn (1) the fraction in annual compensation paid in stock options isregressed on the log difference in payoff (the combined payoff to both therm and the CEO) from receiving compensation in stock options ratherthan salary and bonus where the payoff calculations are based on statu-tory marginal tax rates This variable is dened as the log difference be-tween equation (3) and equation (6) in section 4 The coefcient on the taxvariable is large and statistically signicant at the 95-percent level Thepoint estimate implies that a one-percent increase in the payoff differencebetween stock options and salary and bonus results in a 24-percentage-point increase in the share of compensation paid in stock options

The corporate governance variables all have the predicted sign Firmswith a higher fraction of their shares owned by institutional investorsuse more performance-based pay (stock options) Firms with large corpo-rate boards or a large share of inside directors are less likely to use stockoptions

The two contract-theory variables are the dollar change in CEO wealth

11 See Core and Guay (1999) for evidence that rms adjust in this way

TABLE 3Compensation Regressions that Rely on Variation in

Tax Rates over Time

(1) (2)

Tax variablesln (difference in payoff)(a) 2393 (0349)ln (12statutory personal 20015 (0034)

rate)ln (12statutory cap gains 0330 (0171)

rate)ln (12statutory corporate 0399 (0148)

rate)

Corporate governance variablesFraction of rmrsquos shares 0043 (029) 0052 (0028)

owned by institutionalinvestors

Size of the rmrsquos board of 20002 (001) 20003 (0001)directors

Fraction of directors who 20082 (033) 20092 (0030)are inside or gray

Contract-theory variablesD (CEO wealth) per $1000 20261 (0095) 20247 (0087)

in FV(b)

D (CEO wealth) per 10 in 2583 (189) 2554 (201)FV(c)

Volatility(d) 521 (424) 261 (369)ln (rm market value) 0019 (0006) 0019 (0006)

Stock market return variablesLagged annual rm stock 0006 (0009) 0003 (0008)

market returnLagged annual SampP 500 re- 20035 (0023) 20029 (0023)

turn

Regression diagnosticsTotal sample size 5179 5179Firms 427 427Adjusted R2 0313 0311

Dependent variable share of compensation paid in options Standard errors in parentheses Bothregressions in this table include rm xed effects

(a) Log difference in payoff is the log difference in combined rm and CEO after-tax share betweenpaying compensation in options and paying salary

(b) This variable is the dollar change in CEO wealth per $1000 change in rm market value (seeJensen and Murphy 1990)

(c) This variable in the dollar charge in CEO wealth per 10-percent change in rm value (see Hall andLiebman 1998)

(d) Standard deviation of rm market value in trillions

28 Hall amp Liebman

per $1000 change in rm value (Jensen and Murphy 1990) and thedollar change in CEO wealth per 10-percent change in rm value (Halland Liebman 1998) As discussed in Baker and Hall (1998) these twowealth sensitivity measures reect two different concepts of how closelyaligned the managersrsquo incentives are with the interests of the sharehold-ers In the regression they are meant to measure how far the CEOrsquosexisting contract and ownership of stock and stock options is from theoptimal level The negative coefcients on the two variables suggeststhat rms that are below their optimal pay-to-performance sensitivityare more likely to give more stock options in the current period

The second column replaces the tax variable motivated by deferraladvantages with the three statutory rates The corporate after-tax sharehas a positive coefcient as we would expect because the tax advantageof stock options increases as the corporate rate falls The coefcient onthe personal tax rate is small and statistically insignicant The sign ofthe coefcient on the capital gains rate is the opposite of that predictedby tax-avoidance theory

Note that these results rely on time-series variation alone This is not avery convincing test Since the log difference in payoffs mostly rises overthe time period covered in our sample as does the share of compensa-tion paid in options it is possible that the tax variable is simply reectingother factors that were changing over this time period that we have notincluded in our regression Therefore it is important to look at specica-tions that rely on cross-sectional variation in the tax variables

Table 4 contains specications that rely on cross-sectional variation incorporate tax rates The rst column repeats the specication from col-umn (1) of Table 3 but uses the trichotomous measures of rm corporatetax rates in place of the statutory rate in calculating each rmrsquos logdifference in payoff from paying compensation in options rather than insalary and bonus As in the previous table we nd the positive coef-cient on the log difference in payoff just as the theory suggest How-ever the coefcient estimate is now much smaller suggesting that aone-percent increase in the log difference in payoff produces a 01-percentage-point increase in the share of compensation paid in options

Introducing cross-sectional variation in the dependent variable doesnot eliminate the potentially spurious correlation over time between thetax variable and the share of compensation paid in options The secondcolumn of Table 4 isolates the pooled cross-sectional variation in corpo-rate tax rates by adding time dummies and dropping the rm xedeffects In this specication the coefcient on the tax variable is indistin-guishable from zero The coefcient on the institutional variable contin-ues to be large and statistically signicant The third column includes

Taxation of Executive Compensation 29

both rm xed effects and time dummies which tests for a relationshipbetween within-rm changes in the dependent variable and within-rmchanges in each explanatory variable In this specication the coefcienton the tax variable is again indistinguishable from zero The coefcientson the corporate governance variables are also statistically insignicant

The third column contains a specication that treats the Tax ReformAct of 1986 (TRA 86) as a natural experiment For rms and CEOs facingthe statutory corporate tax rate TRA 86 increased the relative payoff topaying compensation in options by 17 percent For a rm facing a zeromarginal tax rate it reduced the relative payoff by 45 percent The reduc-tion occurred because TRA 86 reduced the personal tax rate makingsalary and bonus more attractive while the decreased corporate rate hadno effect on zero-tax-rate rms Thus we would expect rms facing zeroor low marginal tax rates to reduce their use of options after TRA 86while rms facing the statutory rate would be expected to increase themFew rms face zero marginal tax rates year after year Therefore we tryto distinguish between rms that often face low tax rates and rms thatnearly always face the statutory rate We dene low-tax rms as oneswhose average marginal tax rate over 1984 to 1986 was below 02412

Then we ran a difference-in-differences regression to see if high-taxrms increased their use of options more than low-tax rms The interac-tion between post-TRA 86 and high corporate tax rates is the key vari-able The estimated coefcient on this variable is positive but small andnot statistically different from zero at the 95-percent level

It is important to note that in this specication as well the main corpo-rate governance variable the share of the rmrsquos shares owned by institu-tional investors performs as theory would predict The variable appearsto be quite robust Table 5 presents regressions of the share of compensa-tion paid in stock options on the fraction of the rmrsquos shares owned byinstitutional investors Since we have already established that these vari-ables have a strong longitudinal relationship the regressions here aredesigned to focus on the cross-sectional relationship between these twovariables The rst and third columns present regressions using data onlyfrom 1994 and nd a coefcient of roughly 02 (column 1 contains noother covariates while column 3 includes all of the other nontax covari-ates from the regressions in Tables 3 and 4) Since the median share of armrsquos stock owned by institutional investors in our sample rises from 19percent in 1980 to 49 percent in 1994 this coefcient would imply that this

12 This cutoff was chosen after inspecting the distribution of tax rates and essentiallyseparates rms that are always at or near the statutory rate from ones that are sometimesbelow it

TA

BL

E4

Com

pens

atio

nR

egre

ssio

nsth

atR

ely

onC

ross

-Sec

tion

alV

aria

tion

inC

orpo

rate

Tax

Rat

es

(1)

(2)

(3)

(4)

Tax

vari

able

s(a)

ln

(dif

fere

nce

inp

ayof

f)0

099

(00

41)

20

038

(00

36)

20

005

(00

44)

Pos

t-T

RA

862

000

5(0

013

)H

igh

corp

orat

eta

xra

te0

018

(00

17)

(Pos

t-T

RA

86)

3hi

gh0

024

(00

16)

corp

orat

eta

xra

te

Cor

por

ate

gove

rnan

ceva

riab

les

Frac

tion

of

rmrsquos

shar

es0

109

(00

27)

011

4(0

015

)2

000

9(0

030

)0

087

(00

34)

owne

dby

inst

itu

tion

alin

vest

ors

Size

ofth

e

rmrsquos

boar

dof

20

004

(00

01)

20

001

(00

01)

20

002

(00

01)

000

1(0

002

)d

irec

tors

Frac

tion

ofd

irec

tors

wh

o2

012

4(0

033

)2

000

3(0

018

)2

004

1(0

034

)0

001

(00

39)

are

insi

de

orgr

ay

Con

trac

t-th

eory

vari

able

sD

(CE

Ow

ealt

h)p

er$1

000

20

238

(00

94)

20

167

(00

53)

20

257

(00

96)

20

021

(01

47)

inFV

(b)

D(C

EO

wea

lth)

per

10

in2

560

(17

8)2

592

(10

5)2

594

(19

6)2

147

(04

86)

FV(c

)

Vol

atili

ty(d

)2

48(4

24)

162

(03

4)1

12(4

28)

106

(06

09)

ln(

rmm

arke

tva

lue)

003

3(

006)

000

8(0

003

)0

009

(00

06)

001

2(0

007

)

Stoc

km

arke

tre

turn

vari

able

sL

agge

dan

nual

rm

stoc

k2

000

2(0

009

)0

011

(00

10)

000

6(0

009

)2

001

2(0

013

)m

arke

tre

turn

Lag

ged

annu

alSamp

P50

02

005

6(0

023

)2

005

3(0

051

)2

003

9(0

049

)2

015

4(0

040

)re

turn

Reg

ress

ion

dia

gnos

tics

T

otal

sam

ple

size

5179

5179

5179

2534

Firm

s42

742

739

6A

dju

sted

R2

031

009

032

005

Firm

xed

effe

cts

Yes

No

Yes

No

Tim

ed

um

mie

sN

oY

esY

esN

o

Dep

end

ent

vari

able

isfr

acti

onof

ann

ualc

omp

ensa

tion

from

stoc

kop

tion

sSt

and

ard

erro

rsin

par

enth

eses

(a)

All

thre

esp

eci

cati

ons

use

tric

hoto

mou

sco

rpor

ate

tax

rate

sT

he

sam

ple

inco

lum

n(3

)u

ses

only

the

year

s19

84to

1988

sin

cew

ear

eu

sing

1986

asa

nat

ural

exp

erim

ent

(b)

Thi

sva

riab

leis

the

dol

lar

chan

gein

CE

Ow

ealt

hp

er$1

000

chan

gein

rm

mar

ketv

alu

e(s

eeJe

nse

nan

dM

urp

hy

1990

)

(c)

Thi

sva

riab

lein

the

dol

lar

char

gein

CE

Ow

ealt

hp

er10

-per

cent

chan

gein

rm

valu

e(s

eeH

alla

nd

Lie

bman

199

8)

(d)

Stan

dar

dd

evia

tion

ofr

mm

arke

tva

lue

intr

illio

ns

32 Hall amp Liebman

TABLE 5Large Institutional Investors and Stock-Option Compensation

The Cross-sectional Relationship between Institutional Investors andOption Grants

(1) (2) (3) (4)

Coefcient on fraction of 217 (067) 145 (014) 177 (069) 120 (015)rmrsquos shares owned byinstitutional investors

Sample period 1994 1980ndash1994 1994 1980ndash1994Contract-theory and stock No No Yes Yes

market covariatesTime dummies No Yes No Yes

The dependent variable is the share of compensation paid in options

factor alone can explain 6 percentage points of the 23-percentage-pointincrease in the median share of compensation paid in options that oc-curred over this time period Of course this variable is only one measureof the strengthening of corporate governance so the total impact of im-proved corporate governance could be greater The second and fourthcolumns of the table use the entire 1980ndash1994 sample and include timedummies to isolate the cross-sectional correlations Again there is evi-dence of a strong cross-sectional relationship between the two variablesalthough the coefcients are slightly smaller perhaps reecting a weakercorporate-governance role in the earlier period In sum there is a strongrelationship in both longitudinal and cross-sectional data between theinstitutional-investor variable and the options-percentage variable13

7 THE MILLION-DOLLAR RULE

In 1993 section 162(m) of the Internal Revenue Code was enactedlimiting the deductibility of executive compensation in excess of onemillion dollars unless the compensation was performance-related Pro-ponents of this legislation argued that it would slow the growth ofexecutive pay and tighten the link between rm performance and execu-tive compensation

The new provision became effective for tax year 1994 and applies tothe CEO and the other four most highly compensated executives in eachcompany Pay that is performance-related such as stock-option grants

13 The one specication in which the fraction owned by institutional investors is notsignicant is in column (2) of Table 4 a regression that includes both xed effects and timeeffects

Taxation of Executive Compensation 33

and bonuses paid for meeting clear performance goals is not affected bythis regulation provided that it has been approved by shareholdersThus the main impact of the provision is to limit the deductibility ofsalary in excess of a million dollars Woodlock and Antenucci (1997)studied the proxy statements of 376 rms and documented that mostrms responded to the new law by qualifying top executive pay asperformance-related Perry and Zenner (1999) identied 25 rms thatreduced salaries from above $1 million to below $1 million and foundthat 23 of the rms cited section 162(m) as their reason for doing so

In this section of the paper we investigate whether this provision hasaffected the level or structure of executive pay using the Execucompdata base which contains information on the highest-paid executives ineach of the rms in the SampP 500 SampP Midcap 400 and SampP SmallCap600 Between 1993 and 1998 the median salary of the Execucomp execu-tives rose from $230000 to $306000 The median value of total compen-sation (salary bonus and option grants) rose from $470000 to $882000Limiting the sample of CEOs indicates that the total compensation of themedian CEO rose from $11 million to $18 million over the period whilethe median salary rose from $441000 to $550000 Thus executive com-pensation clearly continued to rise at a rapid rate after the implementa-tion of the million-dollar rule though perhaps at a slower rate than itotherwise would have

If the regulation affects behavior it should have the largest impact onrms whose executives are receiving salaries that are above the limit orjust below it Thus we might expect to see rms whose executives arereceiving relatively high salaries to rely more heavily on performance-based pay and less heavily on salary in the years following the regulationPerry and Zenner (1999) and Livingston (1999) both nd that this is thecase This trend can be clearly identied in the Execucomp data set Forexample over 1993 to 1998 the median annualized growth rate of salaryfor executives whose 1993 salary was below $500000 in the previous yearwas 85 percent For executives with salary between $500000 and$700000 it was 64 percent for executives between $700000 and $850000it was 50 percent for executives between $750000 and $1 million it was31 percent and for executives above $1 million it was 00 percent14

The observation that executives at lower levels of salary had highergrowth rates is not very strong evidence of an impact of section 162(m)however because it is possible that low-salary executives would havehad higher growth rates even in the absence of the provision To explore

14 Of the 3137 executives for whom we have complete data between 1993 and 1998 2708had 1993 salaries below $500000 while only 35 had 1993 salaries above $1 million

34 Hall amp Liebman

whether this is the case we use Execucomp data from both before andafter the law change Ideally we would want to have data from manyyears before the change to examine whether there is an underlying pat-tern of faster salary growth for low-salary executives Unfortunately theExecucomp data begin in 1992 and therefore provide us with only the1992ndash1993 growth rate in compensation to use as a baseline15

We run regressions pooling data from 1992ndash1993 through 1997ndash1998of the annual percentage change in different forms of compensation on avariable million that is designed to measure the likelihood that anexecutiversquos compensation will be affected by section 162(m) We denemillion as the minimum of 1 and the executiversquos salary divided by $1million in the previous year Thus any executive with salary at or above$1 million would receive a value of 1 while an executive with a salary of$500000 would be coded as a 0516 We include this variable by itself andalso interacted with a dummy variable (after) that equals 1 for years1994 and after Thus the million variable is designed to pick up anyunderlying relationship between the level of salary and the growth rateof salary while the interacted one will identify any differential growthrate of salary by income in the period after section 162(m) took effect Wealso include a full set of year dummies the natural logarithm of thermrsquos market value and the rmrsquos annual rate of return over the twoprevious years as control variables To reduce the sensitivity of our re-sults to outliers we run quantile regressions and robust regressions

Table 6 contains our results The rst four columns are for all topexecutives while the last four columns focus on CEOs Columns (1) and(2) present estimates from robust regression for the percentage change insalary and the percentage change in total compensation (salary bonusstock grants and stock-option grants) In the salary growth regressionsfor all executives the coefcient on million is negative and signicantindicating that there is an underlying relationship between the level ofexecutive compensation and its subsequent growth rate The coefcienton million 3 after is negative as well suggesting that the negativerelationship between the level of salary and its subsequent growth rateintensied slightly after 1994 The relatively small magnitude suggeststhat an executive with a $1 million dollar salary would see his salarygrow at an annual rate that would be about 06 percent slower than anexecutive earning $500000

15 The HallndashLiebman data set contains only a combined salary and bonus variable makingit impossible to use it to analyze this issue16 We also ran the regressions (not reported) allowing for a nonlinear relationship betweenmillion and compensation growth and obtained substantively similar results

TA

BL

E6

The

Eff

ect

ofth

eM

illi

on-D

olla

rR

ule

onSa

lary

and

Tot

alC

ompe

nsat

ion

All

top

exec

uti

ves

CE

Os

only

Rob

ust

regr

essi

onQ

uan

tile

regr

essi

onR

obu

stre

gres

sion

Qu

anti

lere

gres

sion

D

in

Din

D

in

Din

D

in

Din

D

in

Din

sala

ryto

talc

omp

(a)

sala

ryto

talc

omp

sa

lary

tota

lcom

p

sala

ry

tota

lcom

p

mil

lio

n(b

)2

006

02

035

72

006

32

024

12

008

72

023

12

010

32

024

5(0

005

)(0

032

)(0

006

)(0

027

)(0

010

)(0

073

)(0

011

)(0

056

)m

illi

on

32

001

10

026

20

021

20

015

20

016

014

92

003

40

158

aft

er(c

)(0

005

)(0

033

)(0

006

)(0

029

)(0

010

)(0

076

)(0

011

)(0

057

)N

359

5035

950

359

5035

950

757

17

589

757

17

589

Stan

dar

der

rors

inp

aren

thes

esR

egre

ssio

nsal

soin

clu

de

year

du

mm

ies

stoc

km

arke

tre

turn

and

log

mar

ket

valu

e

(a)

Tot

alco

mp

ensa

tion

incl

ud

essa

lary

bon

us

rest

rict

edst

ock

and

stoc

k-op

tion

gran

ts

(b)

mil

lion

isd

ene

das

min

(1s

alar

yin

pre

viou

sye

ars

div

ided

by$1

mil

lion)

(c)

aft

eris

anin

dic

ator

vari

able

equ

alto

zero

befo

reth

ere

gula

tion

took

plac

e(1

993)

and

one

afte

rwar

d

36 Hall amp Liebman

In the second column the coefcient on million 3 after is essentiallyzero suggesting that any decrease in salary brought about by section162(m) was offset by increases in bonus and stock-option grants Theresults are similar for the median regressions There is some evidence ofa very small slowdown in growth of salary for executives likely to beaffected by section 162(m) but no sign of a change in total compensa-tion Taken together this provides evidence of a very small substitutionof performance-related pay for salary

The last four columns run identical tests but limit the sample to CEOsThe interaction of million 3 after continues to be small and negative inthe salary regressions In contrast the impact of 162(m) on total compen-sation appears to be positive and fairly substantial (about 15 percentagepoints per year) in these regressions That is the increase in bonus andstock options following the million-dollar rule more than offset the verysmall decline in salary

All of these conclusions should be interpreted somewhat cautiouslybecause our data provide us with only one pre-1993 control year Never-theless the data are consistent with section 162(m) having led to a veryminor slowdown in salary growth and one that is dramatically smallerthan what one would conclude without controlling for underlying differ-ential salary growth rates at different levels of salary Since total compen-sation for CEOs did not decline and perhaps increased there is evidenceof a minor substitution of performance-related pay for salary in responseto the regulation

8 EXECUTIVE COMPENSATION AND THEELASTICITY OF TAXABLE INCOME

Even if taxes have only a minor effect on executive compensation thestructure of executive compensation is important for tax policy be-cause high-income taxpayers are sources of signicant revenue and arethe focus of important debates about the efciency cost of taxation

Goolsbee (1999) has recently argued that executive decisions to exer-cise options are highly responsive to intertemporal differences in taxrates created by tax reforms He claims that much of the apparent declinein taxable income for high-income taxpayers between 1992 and 1993documented by Feldstein and Feenberg (1996) can be explained by execu-tives shifting option gains into 1992 in order to avoid the higher marginaltax rates that went into effect in 1993 By extension he implies that thehigh elasticities of taxable income with respect to personal tax ratesestimated by comparing the pre- and post-tax-reform taxable income of

Taxation of Executive Compensation 37

high-income individuals reect intertemporal shifting of income and notpermanent effects of taxation If this is true then the deadweight loss oftaxation is lower as is the cost of progressivity

While Goolsbee presents comprehensive evidence that option gainswere unusually high in 1992 his identication of the impact of tax rates onthe timing of option gains depends on correctly specifying a counter-factual level of options that would have been exercised in the absence ofthe tax change He simply assumes a linear time trend The purpose ofGoolsbeersquos time trend is presumably to account for the increasing use ofoptions over time However since options typically require a vestingperiod of at least a couple of years a much more direct measure of ex-pected option gains can be constructed based on the total value of vestedoptions held by the CEO Since options typically have vesting periods thatexceed two years this total can be treated as exogenous since it will not beaffected by the tax changes

More generally the value of options held by the CEO is likely to havea major effect on his exercise decisions If his options are out of themoney he will not have gains to realize If there has been a recent run-up in the market an optimizing CEO may exercise an unusually largeamount of options in order to diversify his overall portfolio and if in-the-money options are about to expire they will be exercised in thecurrent year

Our evidence suggests that much of the apparent tax shifting caninstead be attributed to stock market performance and the timing of pastoption grants Using our 15-year panel we replicate Goolsbeersquos (1999)result for the 1993 period but show that the impact of taxes is notpresent in other time periods Moreover even the result for the 1993period disappears when appropriate controls for past option grants andstock price appreciation are included

The top three rows of Table 7 present data on the taxable incomes ofCEOs in the HallndashLiebman sample For these results our panel has beennarrowed to include only CEOs whose rmrsquos scal years correspondwith tax years In addition we create a balanced panel for each of thethree tax-reform periods by excluding any rm for which there are notcomplete data for the period surrounding the reform

The fourth row of the table shows that the 1993 tax act raised the topmarginal personal income tax rate from 310 to 396 percent effective fortax year 1993 with a further increase occurring in 1994 due to the uncap-ping of the Medicare payroll tax The top row of the table shows thattaxable income was higher in 1992 than for any other year in our sample(all data are in 1994 dollars) Option gains were particularly high in1992mdashmore than double their 1991 level and 68 percent higher than their

TA

BL

E7

Tax

Ref

orm

Sto

ckM

arke

tP

erfo

rman

cea

ndth

eT

imin

gof

Opt

ion

Gai

ns

Ave

rage

stoc

km

arke

tM

ean

CE

OM

ean

CE

Ore

turn

for

Mea

nva

lue

taxa

ble

sala

ryan

dM

ean

CE

OT

opfe

der

al

rms

inof

opti

ons

(Val

ue

ofop

tion

Sam

ple

inco

me

bonu

sop

tion

gain

sp

erso

nal

sam

ple

for

held

atbe

gga

ins)

(va

lue

Ref

orm

size

Yea

r($

mil

)($

mil

)($

mil

)M

TR

pre

vye

ar(

)of

year

ofop

tion

she

ld)

ER

TA

8127

919

800

874

064

80

163

7025

80

460

354

1981

083

10

672

009

869

245

070

61

3919

820

872

067

90

099

5011

90

686

144

1983

099

00

724

025

250

355

116

62

16

TR

A86

295

1985

119

70

837

024

050

76

116

02

0719

861

448

091

70

345

5042

81

776

194

1987

165

50

959

059

338

519

12

229

266

1988

163

21

058

039

028

07

237

71

6419

891

656

105

20

414

2817

92

345

176

1993

272

1990

209

51

028

061

828

232

288

02

15T

axA

ct19

911

875

102

00

547

312

01

268

82

0319

922

678

107

01

250

3146

64

552

275

1993

232

41

209

074

339

620

64

299

173

1994

196

41

262

037

842

515

45

113

074

Sam

ple

isH

allndash

Lie

bman

rm

sw

ith

Dec

embe

r

scal

year

sth

atha

dco

mp

lete

dat

afo

ron

eof

the

thre

eta

x-re

form

per

iod

sst

ud

ied

Taxation of Executive Compensation 39

1994 level These patterns are consistent with Goolsbeersquos claim that tax-payers responded to the anticipated increase in marginal tax rates byshifting gains into 1992

The evidence from the other two tax reforms is less clear suggestingthat there might be factors other than tax rates determining the timing ofoption gains ERTA 1981 reduced the top marginal tax rate from 70percent in 1980 to 50 percent in 1982 However there is no sign thatoption gains were reduced in 1980 and 1981 and then increased in 198217

There was a big increase in option gains in 1983 however Similarly thehighest year of stock-option gains in the period around 1986 was 1987 astrange year to take gains since taxpayers would have known that thepersonal rate would be 10 percentage points lower if they waited untilthe following year18

What all the peak years of option gains have in common is that theyfollow years of strong stock market performance The fth row of Table 7shows the average stock market return during the previous year for therms in our sample The large value of option gains in 1983 followed a355-percent increase in the stock market The 1992 boom in option gainsfollowed a 466-percent increase in the stock market the more modest1987 peak in option gains followed a 19-percent gain (and a 43-percentgain two years before)

Strong stock market performance raises the value of options availableto exercise The sixth row of the table shows the mean value of optionsheld by CEOs at the beginning of the year All three of the peak years ofoption gains correspond with peak years in terms of the value of optionsavailable for exercise Since stock market appreciation during the yearalso increases the value of options available to exercise later in that yearwe need to allow for current-year stock market appreciation as well

Table 8 contains Goolsbee-style regressions for each of the tax re-forms The rst two columns reprint the results from Goolsbee (1999)The third and fourth columns replicate his results in our sample Inparticular with a specication similar to his we are able to replicate hisnding that the current after-tax share has a positive coefcient and thefollowing yearrsquos after-tax share has a coefcient of roughly negative oneColumns (5) through (8) show that this pattern was unique to the 1993tax reform Indeed in the TRA 86 regressions the coefcient on the

17 As we explained above for executives covered by the maximum tax on earned incomethere was not a decline in tax rates between 1981 and 198218 It is possible that the large stock market decline in October 1997 may have promptedsome people to exercise options earlier than they otherwise would have However anincrease in option exercises typically follows stock market increases not declines

TA

BL

E8

Att

empt

sto

Rep

lica

teth

eT

ax-S

hift

ing

Res

ult

inO

ther

Tim

eP

erio

ds

Goo

lsbe

eG

ools

bee

1993

1993

Ent

ire

Ent

ire

regr

essi

onre

gres

sion

refo

rmre

form

TR

A86

TR

A86

ER

TA

81E

RT

A81

1980

ndash199

419

80ndash1

994

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

ln[1

2ta

x(t)

]1

244

108

20

455

032

62

017

62

017

22

016

72

016

70

026

20

068

(01

34)

(01

27)

(05

37)

(05

42)

(02

10)

(02

09)

(00

90)

(00

91)

(00

67)

(00

65)

ln[1

2ta

x(t1

1)]

20

867

20

764

21

214

20

917

047

80

722

20

103

004

30

176

018

3(0

118

)(0

112

)(0

491

)(0

499

)(0

157

)(0

174

)(0

072

)(0

072

)(0

068

)(0

066

)

Tim

e0

135

007

62

000

52

036

40

041

000

80

094

003

50

056

002

7(0

008

)(0

008

)(0

049

)(0

049

)(0

028

)(0

028

)(0

025

)(0

026

)(0

003

)(0

003

)

ln(m

arke

tva

lue)

066

10

489

023

70

334

039

5(0

015

)(0

058

)(0

039

)(0

042

)(0

017

)

Yea

rs19

91ndash1

995

1991

ndash199

519

90ndash1

994

1990

ndash199

419

85ndash1

989

1985

ndash198

919

80ndash1

983

1980

ndash198

319

80ndash1

994

1980

ndash199

4

Ad

just

edR

20

720

770

610

630

690

760

800

810

630

68

Obs

erva

tion

s13

856

132

371

385

136

41

470

143

31

113

106

23

968

385

9

Dep

end

ent

vari

able

isth

elo

gof

taxa

ble

inco

me

All

regr

essi

ons

con

tain

rm

xe

def

fect

sG

ools

bee

regr

essi

on(2

)al

soco

ntai

nstw

oot

her

vari

able

sno

tsh

own

inth

eta

ble

Stan

dar

der

rors

inp

aren

thes

es

Taxation of Executive Compensation 41

TABLE 9Alternative Explanations for the Timing of Option Exercises

1993 Entirereform TRA 86 ERTA 81 1980ndash1994

(1) (2) (3) (4)

ln[12tax(t)] 0108 0111 20051 0201(0324) (0136) (0048) (0057)

ln[12tax(t11)] 20134 0573 0125 0150(0336) (0156) (0053) (0066)

ln(market value) 0418 0177 0338 0418(0061) (0039) (0046) (0163)

Firmrsquos stock market return (t) 0298 0268 0214 0258(0044) (0045) (0031) (0023)

Firmrsquos stock market return 0162 0206 0031 0128(t21) (0041) (0008) (0029) (0021)

ln(value of options held by 0004 0008 0001 0009CEO at start of year) (0003) (0003) (0002) (0001)

Years 1990ndash1994 1985ndash1989 1980ndash1983 1980ndash1994Adjusted R2 064 071 082 068Observations 1349 1427 1036 3812

The dependent variable is the log of taxable income All regressions contain rm xed effects Stan-dard errors in parentheses

following yearrsquos tax rate is positive and signicant Columns (9) and (10)pool all three reforms and fail to replicate the pattern that would reectintertemporal shifting19

It is possible that the ease with which one can shift the timing ofoption gains is asymmetric It may be harder to shift option gains for-ward in time since some options expire and therefore must be exercisednow In addition options were a much smaller portion of compensationin the early 1980s so perhaps the ERTA 81 results are not surprisingNonetheless these results cast doubt on the proposition that this is thechannel through which spurious estimates of the elasticity of taxableincome have occurred for previous tax reforms and they suggest thatwe should look harder for alternative explanations for the timing ofoption gains

Table 9 attempts to predict options gains using the lagged and currentincrease in a rmrsquos market value and the value of the options held by theCEO at the start of the year The two stock market variables stronglypredict option gains while the value of options held has the correct sign

19 The results also fail to suggest large permanent effects of marginal tax rates on taxableincome

42 Hall amp Liebman

in all of the specications and is signicant in the pooled sample eventhough it is highly correlated with the two stock market return variablesIn contrast the tax-timing explanation fails even in the 1993 period

It is important to realize that our results do not invalidate Goolsbeersquoscritique of claims that aggregate revenue data from 1992 and 1993 provethat elasticities of taxable income are large We have simply given analternative explanation for why taxable incomes were unusually high in1992 However our results do cast doubt on the conclusion that largetiming shifts are ubiquitous and capable of explaining away the morecareful estimates of taxable income elasticities that have been performedusing micro data (eg Feldstein 1995 Auten and Carroll 1995)

One further point is in order It is sometimes argued that the combina-tion of increased use of options and the booming stock market explainswhy federal tax revenues have been so much higher than expected dur-ing the 1990s However while large option gains increase revenue fromthe personal income tax they produce offsetting deductions for cor-porations The net impact on revenue is only the difference in the twomarginal tax rates times the option gain While the gap between thepersonal rate and the effective corporate rate is nontrivial the impact oflarge option exercises on total revenue is substantially smaller than theamount by which the exercises inate personal tax revenue and reducecorporate tax revenue

9 CONCLUSION

We have described the tax rules for executive pay in detail and analyzedhow changes in various tax rates affect the tax advantages of stock optionsrelative to salary and bonus We nd that there is a moderate tax advan-tage to optionsmdashcurrently about $4 per $100 in pretax executive paymdashand that changes in the tax advantage of options over time have had atmost a modest impact on the composition of pay Corporate governancefactors particularly the role of large investors appear to be more impor-tant in explaining the dramatic increase in option pay

In addition our evidence suggests that more direct attempts to use taxpolicy to inuence executive compensation have had little effect Wend that the million-dollar rule led companies to substitute perfor-mance-based pay for salary But our evidence suggests that this substitu-tion was quite modest and there is no evidence that the total level of paywas reduced Overall although the stock-option explosion has dramati-cally increased the link between pay and performance this change isdue almost entirely to nontax factors

Taxation of Executive Compensation 43

REFERENCES

Auerbach Alan J (1996) ldquoThe Dynamic Effects of Tax Law Asymmetriesrdquo Re-view of Economic Studies 53451ndash405

mdashmdashmdash and James Poterba (1987) ldquoTax Loss Carryforwards and Corporate TaxIncentivesrdquo In The Effects of Taxation on Capital Accumulation M Feldstein(ed) Chicago University of Chicago Press

Altshuler Rosanne and Alan Auerbach (1990) ldquoThe Signicance of Tax LawAsymmetries An Empirical Investigationrdquo Quarterly Journal of Economics 10561ndash86

Auten Gerald and Robert Carroll (1995) ldquoThe Effect of Income Taxes on House-hold Behaviorrdquo Ofce of Tax Analysis US Department of the Treasury

Baker George P and Brian J Hall (1998) ldquoCEO Incentives and Firm SizerdquoNBER Working Paper no 6868 December

Core John and Wayne Guay (1999) ldquoThe Use of Equity Grants to ManageOptimal Equity Incentive Levelsrdquo Draft

Feldstein M (1995) ldquoThe Effect of Marginal Tax Rates on Taxable Income APanel Study of the 1986 Tax Reform Actrdquo Journal of Political Economy 103 (no3)551ndash72

mdashmdashmdash and D Feenberg (1996) ldquoThe Effect of Increased Tax Rates on TaxableIncome and Economic Efciency A Preliminary Analysis of the 1993 Tax RateIncreasesrdquo In Tax Policy and the Economy vol 10 James M Poterba (ed)Cambridge MA MIT Press

Gompers Paul A and Andrew Metrick (1998) ldquoInstitutional Investors andEquity Pricesrdquo NBER Working Paper no 6723

Goolsbee Austan (1999) ldquoWhat Happens When You Tax the Rich Evidencefrom Executive Compensationrdquo NBER Working Paper no 6333

Graham John R (1996) ldquoProxies for the Corporate Marginal Tax Raterdquo Journal ofFinancial Economics 42187ndash221

Hall Brian J (1999) ldquoThe Design of Multi-year Stock Option Plansrdquo Journal ofApplied Corporate Finance 12(2)97ndash106

mdashmdashmdash and Jeffrey B Liebman (1998) ldquoAre CEOs Really Paid Like Bureau-cratsrdquo Quarterly Journal of Economics CXII(no 3)653ndash691

Jensen Michael C and William M Meckling (1976) ldquoTheory of the Firm Mana-gerial Behavior Agency Costs and Ownership Structurerdquo Journal of FinancialEconomics III305ndash360

mdashmdashmdash and Kevin J Murphy (1990) ldquoPerformance Pay and Top-ManagementIncentivesrdquo Journal of Political Economy XCVIII225ndash264

Kaplan Steven N (1997) ldquoThe Evolution of US Corporate Governance WeAre All Henry Kravis Nowrdquo Draft

Livingstone Jane R (1999) ldquoExecutive Compensation Contracting and Re-sponses to Increased Tax Costsrdquo Louisiana State University June

Miller Merton H and Myron S Scholes (1982) ldquoExecutive CompensationTaxes and Incentivesrdquo In Financial Economics Essays in Honor of Paul CootnerWilliam F Sharpe and Cathryn M Cootner (eds) Princeton NJ Prentice-Hall

Murphy Kevin J (1999) ldquoExecutive Compensationrdquo forthcoming in OrleyAshenfelter and David Card (eds) Handbook of Labor Economics Vol 3 NorthHolland

Perry Todd and Marc Zenner (1999) ldquoPay for Performance Government Regu-

14 Hall amp Liebman

The effect of taxation on the optimal contract (more precisely theshare of compensation that is performance-related) is ambiguous (evenin the absence of deductibility and deferral) because there are offsettingeffects First by reducing the share of corporate prots received byshareholders taxes diminish the importance to the shareholders of moti-vating the CEO Taken alone this effect would imply that taxes wouldbe expected to reduce the use of performance-based compensation Sec-ond because the government shares in the income received by the CEOthe variance of the CEOrsquos income is reduced raising his utility andlowering the cost to the rm of providing any given set of incentivesThird because taxes will lead executives to provide less effort for anygiven level of incentive-based pay the level of compensation that mustbe provided to compensate the CEO for effort is reduced for a givenamount of incentive These last two factors make it cheaper for the rmto offer contracts to the CEO with large incentive components when taxrates are higher and should therefore increase the use of incentive-based pay Since the net effect of taxation on the level of incentive-basedpay is ambiguous agency theory provides no strong predictions abouthow taxation should affect the optimal composition of option versuscash compensation

42 The Tax Advantages of DeferralWhile agency theory yields ambiguous predictions about the effect oftaxes on the use of options there are direct tax advantages of options inthat options provide executives with a way to defer compensation andthereby lower their taxes However because options also lead to a defer-ral of corporate tax deductibility the tax advantages from a combined(executive and corporate) perspective are not so straightforward In thissection we analyze and measure the tax advantages of stock optionsrelative to salary and bonus compensation We also analyze the taxadvantages of NQSOs relative to ISOs In particular we show how thetax advantages of options change as personal corporate and capitalgains taxes change We then show how the tax advantages of optionshave changed over time in response to changing tax rates

The crucial tax difference between standard options (NQSOs) andcash payment is that option payouts are deferred and the two forms ofcompensation earn different rates of return over the deferral period Anyanalysis of the relative tax advantages of two compensation instrumentsmust consider the tax consequences both to the employer and to theemployeemdashwhat Scholes and Wolfson (1992) call the global contractingperspective Thus in order to make valid comparisons between the twowe compare the tax burden to the executive while holding constant the

Taxation of Executive Compensation 15

posttax cost to the company By keeping posttax employer costs (in NPVterms) constant any package that is preferred by the employee is tax-advantaged in the global-contracting sense

Under this methodology a comparison of the tax advantages of op-tions and cash involves comparing a pretax cash payment of P with anoption payment that has the equivalent posttax NPV to the companyFor NQSOs it is straightforward to show that a pretax payment of P toan executive has exactly the same cost to the company as putting aside Pfor the purposes of paying stock-option payouts later That is if a com-pany pays an executive P today it will have the same amount of moneyin N years as if it had put aside P today let it accumulate at the rate ofreturn earned by the rm and then paid it out (as compensation in theform of option gains4) with any appreciation in year N

Assume that the pretax prot rate is r and the corporate tax rate is TcThen if the company pays P today it receives a deduction today of PTcwhich yields

PTc [1 1 r (1 2 Tc)]N (1)

in N years since the benets of the tax deduction are invested in thecompany and receive the after-tax corporate rate of return Converselyif the company puts P aside today then it grows at the after-tax corpo-rate rate of return in N years to give a tax deduction of

P [1 1 r (1 2 Tc)]N Tc (2)

which is the exact same value Note that the equivalence of these two taxdeductions is analogous to the tax-benet equivalence of front-loadedand back-loaded IRAs

An executiversquos payoff in N years from option prots is equal to Pcompounded at the after-tax corporate rate times 1 2 Tp where Tp is thepersonal tax rate Thus the combined payoffs in N years of the corporatededuction and the executiversquos posttax payoff is

P [1 1 r (1 2 Tc)]N Tc 1 P [1 1 r (1 2 Tc)]

N (1 2 Tp) (3)

where the rst term is the payoff from the corporate deduction and thesecond term is the executiversquos payoff

4 This analysis is not unique to options Any form of deferred compensation that enablesexecutives to invest inside the rm at a higher posttax return or without paying capitalgains taxes would have similar effects

16 Hall amp Liebman

The payoff to salary and bonus in N years is more complicated sincean assumption must be made about how the executive invests the origi-nal cash compensation For example if the executive invests in an instru-ment with taxable interest (eg bonds) then the N-year payoff from aninvestment of P accumulates at the post-corporate-tax post-personal-taxrate of return and is

P (1 2 Tp) [1 1 r (1 2 Tc)(1 2 Tp)]N (4)

However if an executive invests in (non-dividend-paying) equities theinvestment accumulates at the higher post-corporate-tax rate of returnwith the offsetting disadvantage that the capital gains are taxed at thecapital gains rate in year N So the executive receives

P [1 1 r (1 2 Tc)]N (1 2 Tp) 2 Tcg $P (1 2 Tp) [1 1 r (1 2 Tc)]

N 2P (1 2 Tp) (5)

where Tcg is the capital gains tax rate5Thus putting the equations together if an executive is paid cash P

and invests g in bonds and 1 2 g in equities the N-year payoff (includ-ing the payoff to the corporation from the tax deduction) is

P [1 1 r (1 2 Tc)]N Tc 1 g $(P [1 1 r (1 2 Tc)(1 2 Tp)]

N (1 2 Tp)1(1 2 g) (P [1 1 r (1 2 Tc)]

N (1 2 Tp) 2 Tcg $P (1 2 Tp)[1 1 r (1 2 Tc)]

N 2 P (1 2 Tp)) (6)

The tax advantage of options versus salary and bonus therefore is thedifference between the combined (corporate and executive) payoffs inequation (3) and the combined payoffs in equation (6)

43 The Tax Advantage of OptionsWe now turn to analysis of how changes in various tax rates affectthe tax advantages of options relative to cash compensation We focuson the case in which the cash earned by executives is invested entirelyin equities since equities are tax-favored relative to bonds The impor-tant conclusions of this analysis are not substantively different if we

5 In practice executives hold a combination of equities and bondlike instruments withtaxable interest Because equities are tax-favored in the next section we simplify the analy-sis by assuming that executives hold only equities Alternatively we could have assumedthat executives hold only less risky bonds since bonds more closely match the risk proleof a stream of tax savings None of our empirical results are substantively affected by thissimplication

Taxation of Executive Compensation 17

instead assume that a portion of an executiversquos holdings are invested inbonds

431 The Tax Advantage of Options Changing the Corporate TaxRate Stock options are tax-advantaged relative to cash simply becauseoption payouts are deferred allowing the executive to invest at the pre-personal-tax rate of return (with no capital gains tax at the end) ratherthan investing at the after-tax rate of return or paying the capital gainstax Since the advantage of being able to defer taxes is large when thecorporate tax rate is low the tax advantage of options is larger when thecorporate rate is low

To illustrate this point we dene the tax advantage of options to bethe dollar amount by which total after-tax option payoffs (to the corpora-tion and the executive) exceed the total payoffs from salary as denedby the difference of equation (3) and (6) assuming that P (the paymentto the executive) is equal to $100 and g 5 0 (cash compensation is in-vested in equities) We use a ten-year horizon since most options have aten-year term We then calculate how the tax advantage of optionschanges as the corporate tax rate varies from zero to 100 percent holdingthe personal tax rate and the capital gains rate xed at 40 percent and 20percent respectively (which approximates current rates)

The top panel of Figure 2 shows the tax advantage of options relativeto cashmdashprecisely the NPV of the total tax advantage of options for a$100 payment to the CEOmdashas the corporate tax rate changes As ex-pected the tax advantage of options declines as corporate tax rates in-crease At a zero corporate tax rate the tax advantage of options has anNPV of $7 This value is about $350 at a corporate rate of 40 percentWhen the corporate tax rate is 100 percent the tax advantage of optionscompletely disappears since the benets of deferred compensation fallto zero because the after-tax return on equities (and therefore the dis-count rate) falls to zero The tax advantage of options is the NPV of thetax savings from avoiding the capital gains tax on ten years of apprecia-tion which is why it is relatively modest even at the 40-percent corpo-rate rate

432 The Tax Advantage of Options Changing the Personal TaxRate Using the same assumptions as above (but this time holding thecorporate rate at 40 percent and the capital gains rate at 20 percent andvarying the personal rate) the tax advantage of options declines as thepersonal tax rate rises as shown in the middle panel of Figure 2 The taxadvantage of options is higher at low personal tax rates because theposttax base that gives rise to capital gains is higher and the advantage of

Note Holds constant the personal rate at 40 percent and the capital gains rate at 20 percent

Note Holds constant the corporate rate at 40 percent and the capital gains rate at 20 percent

Note Holds constant the personal rate at 40 percent and the corporate rate at 40 percent

FIGURE 2 Tax Advantage of Options NPV of Tax Advantage of Pay-ing $100 in Options (NQSOs) Rather Than in Cash

Taxation of Executive Compensation 19

avoiding the capital gains taxes is greater The tax advantage of optionstherefore declines linearly as the personal tax rate rises At a 100-percentpersonal tax rate the executive receives nothing in either case and byconstruction the companyrsquos tax deduction benet is the same in NPVterms so the tax advantage of options falls to zero also6

433 The Tax Advantage of Options Changing the Capital Gains TaxRate The tax advantage of options for various capital gains rates isdepicted in the bottom panel of Figure 2 under the same assumptions(this time holding both the personal rate and the corporate rate at 40percent while varying the capital gains rate) The tax advantage of avoid-ing the capital gains tax increases as the capital gains tax rate rises

5 HOW HAS THE TAX ADVANTAGE OF OPTIONSCHANGED OVER TIME

The analysis so far has illustrated how the tax advantage of optionsvaries with the changes in tax rates In this section we show how the taxadvantage of options has changed over time as tax rates have changed

51 Top Marginal Tax Rates over TimeThe evolution on the top marginal tax ratesmdashpersonal corporate andcapital gainsmdashfrom 1980 to 1998 is shown in Figure 3 In all cases thetop rate (the rate for taxpayers with the highest incomes) is shownwhich is not always the highest rate since various anomalies (such asclawbacks of exemptions) sometimes lead to marginal tax rates that arehigher than those for the highest income tax payers

The top personal tax rate was 70 percent in 1980 falling to 50 percentin 1982 and 28 percent in 1988 following the 1986 tax act7 The topmarginal tax rate has since risen to 396 percent but is effectively 425percent since there is a 29-percent Medicare surcharge (paid half by theemployer and half by the employee) that has no upper limit The topcorporate tax rate has had only one signicant change since 1980 fallingfrom 46 percent before the 1986 tax act to 34 percent following the act in1988 The top corporate tax rate was increased to 35 percent in 1993 The

6 This is one place where ignoring bond investments has substantive implications Inparticular in the all-bond case the tax advantage of options is nonmonotonicmdashit rises andthen fallsmdashas personal tax rates rise7 During the early 1980s the maximum tax on earned income limited the marginal tax rateon the earnings of high earners so many executives did not face a decline in personal taxrates between 1981 and 1982 The empirical results in this paper are robust to assumingthat executives faced a marginal tax rate of 50 percent in 1980 and 1981

20 Hall amp Liebman

FIGURE 3 Top Marginal Tax Rates since 1980

top capital gains rate has uctuated between 28 percent (approximately)and 20 percent since 1980 and is currently at 20 percent

52 The Evolution of the Tax Advantage of NQSOsThe tax advantage of options relative to cash dened in the same way asin section 4 from 1980 to 1998 is shown in Figure 4 The calculations aredone using the same assumptions as before (ie N 5 10 years and cashis invested in equities) and the statutory top marginal tax rates shown inFigure 3 are used Due to the fall in corporate and personal tax rates thetax advantage of options increased sharply after the Tax Reform Act of1986 This increase in tax advantage has largely been reversed in the1990s as top personal tax rates have crept back up and the capital gainsrate has been reduced

Two key facts emerge from this analysis First there is currently only amoderate tax advantage to standard nonqualied stock optionsmdashon theorder of $4 per $100 of compensation This is because the tax advantagesto the executive of deferring taxes are largely offset by the tax disad-vantages to the company of not being able to deduct option expensesfrom taxable prots until the executive exercises the options Secondalthough the 1986 tax act substantially increased the tax advantage ofoptions more than half of this increase has been reversed in the 1990s

Taxation of Executive Compensation 21

FIGURE 4 Tax Advantage of Options 1980 to 1998

NPV of tax advantage of $100 in options (NQSOs) rather than cash

53 The Tax (Dis)advantage of ISOsSo far the analysis has focused on the tax advantages of NQSOs sincethey are so much more prevalent than ISOs which account for onlyabout 5 percent of option grants The relative scarcity of ISOs can beexplained both by their tax status and by their per-executive cap of$100000 per year

Under what conditions are ISOs tax-preferred to NQSOs For a trans-fer P to an executive the ISO is always tax-preferred by the executive(since the capital gains rate is lower than the personal rate) while theNQSO is always tax-preferred from the companyrsquos perspective (since anISO is not deductible) The key issue however is the relative advan-tages of the two types of options from a global-contracting perspectiveBecause ISOs are not deductible the company is indifferent betweensetting aside P in the form of NQSOs today and setting aside P(1 2 Tc) inthe form of ISOs (Since the NPV of the tax deduction for NQSOs is PTcthe NPV of the payment P is P 2 PTc or P (1 2 Tc))

Equalizing the posttax cost of NQSOs and ISOs we need only look atthe payoff of the executive to determine the condition under which eachis preferred The payoff of NQSOs in N years is

P [1 1 r (1 2 Tc)]N (1 2 Tp) (7)

22 Hall amp Liebman

The same payoff for ISOs is

P (1 2 Tc) [1 1 r (1 2 Tc)]N (1 2 Tcg) (8)

ISOs therefore are tax-advantaged if

Tp 2 Tcg Tc 2 Tc Tcg (9)

Intuitively ISOs are tax-advantaged only if their advantage (the differ-ence between the personal rate and the capital gains rate) is largeenough to offset their cost (the disadvantage of not deducting at thecorporate rate) Note however that the condition is not a simple com-parison between the corporate rate and the personal rate minus thecapital gains rate (Tp 2 Tcg Tc)

Figure 5 shows how the tax advantage of NQSOs relative to ISOs haschanged since 1980 as tax rates have changed Unlike the very modestchanges (around 3 to 4 percent) in the relative tax advantages betweenoptions and cash shown in Figure 4 the relative tax advantage betweenNQSOs and ISOs has seen enormous swings When the top personalrate was 70 percent in 1980 ISOs were tax-favored by a margin greaterthan 6 percent However by 1982 (following the 1981 tax acts that dra-matically lowered the personal tax rate) ISOs became tax-disadvantaged

FIGURE 5 Tax Advantage of NQSOs Relative to ISOs over Time

NPV of tax advantage of $100 payment to executive NQSOs minus ISOs

Taxation of Executive Compensation 23

by about 5 percent By 1988 (following the 1986 tax act) ISOs became tax-disadvantaged by about 18 percent a dramatic swing The raising ofpersonal rates coupled with a decrease in the capital gains rate has sincereduced the tax disadvantage of ISOs to about $4 per $100 of pretaxcompensation Although hard data on ISOs are hard to come by practi-tioners (mostly compensation consultants) claim that ISOs were morepopular prior to 1982 when they were tax-advantaged Although ISOsare less disadvantageous from a tax standpoint relative to NQSOs thanthey were in the late 1980s they are still disadvantageous so it is notsurprising that they have not made a signicant resurgence in recentyears

6 EXPLAINING THE INCREASE IN OPTION-BASEDCOMPENSATION

In this section we analyze whether changes in tax rates affect the compo-sition of executive compensation and consider the relative importanceof tax factors and corporate governance factors in explaining the in-creasing share of compensation paid in the form of stock options Thenumerous changes in tax rates that have occurred since 1980 provide anopportunity to assess whether executive compensation arrangementsrespond to tax incentives in the way that the tax-avoidance model sug-gests We test this model using a panel data set of CEOs in largepublicly-traded US corporations Our identication of the tax effectsrelies on time-series variation in personal corporate and capital gainstax rates along with cross-sectional variation in corporate tax rates

61 DataWe use a 15-year panel data set of CEOs in the largest publicly-tradedUS corporations which is described in Hall and Liebman (1998) Thedata set covers the years from 1980 through 1994 and combines CEOcompensation information from corporate proxies and 10-K lings withstock price and stock return information from CRSP and accounting datafrom Compustat In addition some compensation data from the 1970swere collected in order to construct measures of the value of stock op-tions held by the CEOs in the rst period of the sample8

The feature that distinguishes our data from most other CEO datasets is that with our panel of yearly proxy data on option grants optiongains and total options held we are able to calculate the total value of

8 We are grateful to David Yermack for providing us with some of the data for 1984 to 1991See Yermack (1995) for a discussion of these data

24 Hall amp Liebman

all stock options held by the CEO at a point of time More importantlysince we have the details about the stock options held (number exer-cise price time to maturity etc) we can precisely calculate the changein the value of a CEOrsquos stock option holdings for a given change in rmvalue

62 Identication IssuesThe numerous tax changes over our sample period and the specic waysin which changes in personal corporate and capital gains rates arepredicted to affect the composition of CEO compensation give us un-usually rich sources of identication However because many of thechanges in tax rates would be expected to affect all of the CEOs in oursample in a similar way we will need to pay particular attention toseparating out the tax effects from underlying time trends and otherfactors that changed over time Controlling for underlying trends is par-ticularly important in this study because we are focusing on an out-comemdashthe use of stock optionsmdashthat has increased rapidly over thepast two decades and that many practitioners believe was importantlyaffected by nontax considerations Thus the tax effects we are trying toexplain are deviations from a rising trend

We take two steps to try to separate out the tax effects from theunderlying trend First we include key nontax factors that could poten-tially explain the increasing reliance on performance-based compensa-tion It has been suggested that the dramatic increase in incentive-basedpay is the result of the remarkably poor shareholder returns during the1970s which spurred the LBO and takeover movements of the 1980sAccording to this story (Kaplan 1997) shareholders became much morepowerful via the rise of institutional investors even as the LBO andtakeover movements waned Because the inuence of institutional inves-tors is thought to be one of the most important mechanisms of strongcorporate governance we use the share of each companyrsquos stock ownedby institutional investors (dened as institutions with more than $100million under management) as an explanatory variable9 The percentageof shares owned by large institutional investors increased from about 20percent to almost 50 percent in 1994 an upward trend that closelymatches the sharp rise in the share of compensation in the form ofoptions Indeed in our sample the annual average (over all of the rmsin our sample) share of stock owned by institutional investors has acorrelation of 09 with the annual average share of compensation paid inoptions

9 This variable is described in Gompers and Metrick (1998)

Taxation of Executive Compensation 25

In addition we include two other variables that proxy for strongercorporate governance The rst is the size of the board Evidence suggeststhat boards with smaller numbers of directors reduce the inuence of theCEO and better represent shareholder interests (Yermack 1995) Boardsize fell by approximately 15 percent between 1980 and 1994 The secondvariable is the fraction of the rmrsquos board members who are inside direc-tors or gray directors (noninsiders who have business dealings with thecompany) Firms with fewer outsiders are less likely to act in the share-holdersrsquo interests and we would predict that they would be less likely tohave performance-based compensation for their CEOs The average per-centage of inside and gray directors has decreased from about 45 percentin 1980 to about 35 percent in 1994

Second we allow for cross-sectional variation in corporate tax ratesCorporate marginal tax rates vary because rms may be eligible for tax-loss carrybacks and carryforwards investment tax credits and the alter-native minimum tax (see Auerbach 1986 Auerbach and Altshuler1990 Auerbach and Poterba 1987) Because these cross-sectional differ-ences imply that different rms should respond differently to a givenchange in tax rates they potentially provide us with a way to isolatethe tax effects Firm-specic corporate marginal tax rates are notori-ously difcult to calculate from publicly-available data We use a tricho-tomous variable that equals zero if the rm has tax-loss carryforwardsand negative earnings in a given year equals 05 times the statutorycorporate rate if the rm has only one of those conditions and equalsthe statutory rate if the rm has neither loss carryforwards nor negativeearnings10

The cross-sectional variation in the corporate tax rate may not be ex-ogenous Firms that perform poorly could face low marginal tax ratesand also be particularly likely or unlikely to provide performance-basedpay Because most of our specications use xed effects the correlationwe must be concerned about is between changes in rm performanceand changes in compensation We deal with this by controlling forlagged rm stock market performance in our regressions This shouldeliminate spurious correlation between tax rates and compensation thatis jointly caused by rm performance

10 Graham (1996) presents evidence that the trichotomous variable performs nearly as wellas his recommended simulated marginal tax rate Plesko (1999) compares marginal taxrates calculated from Compustat data with actual marginal tax rates from rm tax returnsHe nds that simple binary variables are more reliable measures of the marginal tax ratethan is either the Graham or the trichotomous variable We obtain substantively similarresults when we switch from the trichotomous variable to the binary variables recommendby Plesko

26 Hall amp Liebman

63 ResultsThe dependent variable in our regressions is the fraction of a CEOrsquos totalannual compensation (measured as the BlackndashScholes value of stock-option grants plus salary and bonus) that comes from stock-optiongrants We choose this reduced-form specication because our interest isin the composition of CEO compensation However the factors thataffect the composition of compensation might also affect the levels andit will be important to keep this in mind in interpreting our results

Optimal-contracting theory generally makes predictions about the in-centives provided by the total amount of rm stock and stock optionsowned by the CEO We choose to model the ow of option grants ratherthan the ultimate performance sensitivity of the CEOrsquos compensationand stock and stock-option holdings because we have in mind anadjustment-cost model in which it takes time for a CEOrsquos holdings ofstock and stock options to reach the optimal level11 In our regressionswe include two measures of the current performance sensitivity of theCEOrsquos wealth to rm performance as explanatory variables to reect thedistance of the CEOrsquos contract from the optimal contract

Table 3 presents results in which the tax effects are identied solely byvariation in tax rates over time The fraction of annual compensation paidin stock options is regressed on tax variables corporate governance vari-ables contract-theory variables and variables reecting stock market re-turns Both of the regressions in this table include rm xed effects Incolumn (1) the fraction in annual compensation paid in stock options isregressed on the log difference in payoff (the combined payoff to both therm and the CEO) from receiving compensation in stock options ratherthan salary and bonus where the payoff calculations are based on statu-tory marginal tax rates This variable is dened as the log difference be-tween equation (3) and equation (6) in section 4 The coefcient on the taxvariable is large and statistically signicant at the 95-percent level Thepoint estimate implies that a one-percent increase in the payoff differencebetween stock options and salary and bonus results in a 24-percentage-point increase in the share of compensation paid in stock options

The corporate governance variables all have the predicted sign Firmswith a higher fraction of their shares owned by institutional investorsuse more performance-based pay (stock options) Firms with large corpo-rate boards or a large share of inside directors are less likely to use stockoptions

The two contract-theory variables are the dollar change in CEO wealth

11 See Core and Guay (1999) for evidence that rms adjust in this way

TABLE 3Compensation Regressions that Rely on Variation in

Tax Rates over Time

(1) (2)

Tax variablesln (difference in payoff)(a) 2393 (0349)ln (12statutory personal 20015 (0034)

rate)ln (12statutory cap gains 0330 (0171)

rate)ln (12statutory corporate 0399 (0148)

rate)

Corporate governance variablesFraction of rmrsquos shares 0043 (029) 0052 (0028)

owned by institutionalinvestors

Size of the rmrsquos board of 20002 (001) 20003 (0001)directors

Fraction of directors who 20082 (033) 20092 (0030)are inside or gray

Contract-theory variablesD (CEO wealth) per $1000 20261 (0095) 20247 (0087)

in FV(b)

D (CEO wealth) per 10 in 2583 (189) 2554 (201)FV(c)

Volatility(d) 521 (424) 261 (369)ln (rm market value) 0019 (0006) 0019 (0006)

Stock market return variablesLagged annual rm stock 0006 (0009) 0003 (0008)

market returnLagged annual SampP 500 re- 20035 (0023) 20029 (0023)

turn

Regression diagnosticsTotal sample size 5179 5179Firms 427 427Adjusted R2 0313 0311

Dependent variable share of compensation paid in options Standard errors in parentheses Bothregressions in this table include rm xed effects

(a) Log difference in payoff is the log difference in combined rm and CEO after-tax share betweenpaying compensation in options and paying salary

(b) This variable is the dollar change in CEO wealth per $1000 change in rm market value (seeJensen and Murphy 1990)

(c) This variable in the dollar charge in CEO wealth per 10-percent change in rm value (see Hall andLiebman 1998)

(d) Standard deviation of rm market value in trillions

28 Hall amp Liebman

per $1000 change in rm value (Jensen and Murphy 1990) and thedollar change in CEO wealth per 10-percent change in rm value (Halland Liebman 1998) As discussed in Baker and Hall (1998) these twowealth sensitivity measures reect two different concepts of how closelyaligned the managersrsquo incentives are with the interests of the sharehold-ers In the regression they are meant to measure how far the CEOrsquosexisting contract and ownership of stock and stock options is from theoptimal level The negative coefcients on the two variables suggeststhat rms that are below their optimal pay-to-performance sensitivityare more likely to give more stock options in the current period

The second column replaces the tax variable motivated by deferraladvantages with the three statutory rates The corporate after-tax sharehas a positive coefcient as we would expect because the tax advantageof stock options increases as the corporate rate falls The coefcient onthe personal tax rate is small and statistically insignicant The sign ofthe coefcient on the capital gains rate is the opposite of that predictedby tax-avoidance theory

Note that these results rely on time-series variation alone This is not avery convincing test Since the log difference in payoffs mostly rises overthe time period covered in our sample as does the share of compensa-tion paid in options it is possible that the tax variable is simply reectingother factors that were changing over this time period that we have notincluded in our regression Therefore it is important to look at specica-tions that rely on cross-sectional variation in the tax variables

Table 4 contains specications that rely on cross-sectional variation incorporate tax rates The rst column repeats the specication from col-umn (1) of Table 3 but uses the trichotomous measures of rm corporatetax rates in place of the statutory rate in calculating each rmrsquos logdifference in payoff from paying compensation in options rather than insalary and bonus As in the previous table we nd the positive coef-cient on the log difference in payoff just as the theory suggest How-ever the coefcient estimate is now much smaller suggesting that aone-percent increase in the log difference in payoff produces a 01-percentage-point increase in the share of compensation paid in options

Introducing cross-sectional variation in the dependent variable doesnot eliminate the potentially spurious correlation over time between thetax variable and the share of compensation paid in options The secondcolumn of Table 4 isolates the pooled cross-sectional variation in corpo-rate tax rates by adding time dummies and dropping the rm xedeffects In this specication the coefcient on the tax variable is indistin-guishable from zero The coefcient on the institutional variable contin-ues to be large and statistically signicant The third column includes

Taxation of Executive Compensation 29

both rm xed effects and time dummies which tests for a relationshipbetween within-rm changes in the dependent variable and within-rmchanges in each explanatory variable In this specication the coefcienton the tax variable is again indistinguishable from zero The coefcientson the corporate governance variables are also statistically insignicant

The third column contains a specication that treats the Tax ReformAct of 1986 (TRA 86) as a natural experiment For rms and CEOs facingthe statutory corporate tax rate TRA 86 increased the relative payoff topaying compensation in options by 17 percent For a rm facing a zeromarginal tax rate it reduced the relative payoff by 45 percent The reduc-tion occurred because TRA 86 reduced the personal tax rate makingsalary and bonus more attractive while the decreased corporate rate hadno effect on zero-tax-rate rms Thus we would expect rms facing zeroor low marginal tax rates to reduce their use of options after TRA 86while rms facing the statutory rate would be expected to increase themFew rms face zero marginal tax rates year after year Therefore we tryto distinguish between rms that often face low tax rates and rms thatnearly always face the statutory rate We dene low-tax rms as oneswhose average marginal tax rate over 1984 to 1986 was below 02412

Then we ran a difference-in-differences regression to see if high-taxrms increased their use of options more than low-tax rms The interac-tion between post-TRA 86 and high corporate tax rates is the key vari-able The estimated coefcient on this variable is positive but small andnot statistically different from zero at the 95-percent level

It is important to note that in this specication as well the main corpo-rate governance variable the share of the rmrsquos shares owned by institu-tional investors performs as theory would predict The variable appearsto be quite robust Table 5 presents regressions of the share of compensa-tion paid in stock options on the fraction of the rmrsquos shares owned byinstitutional investors Since we have already established that these vari-ables have a strong longitudinal relationship the regressions here aredesigned to focus on the cross-sectional relationship between these twovariables The rst and third columns present regressions using data onlyfrom 1994 and nd a coefcient of roughly 02 (column 1 contains noother covariates while column 3 includes all of the other nontax covari-ates from the regressions in Tables 3 and 4) Since the median share of armrsquos stock owned by institutional investors in our sample rises from 19percent in 1980 to 49 percent in 1994 this coefcient would imply that this

12 This cutoff was chosen after inspecting the distribution of tax rates and essentiallyseparates rms that are always at or near the statutory rate from ones that are sometimesbelow it

TA

BL

E4

Com

pens

atio

nR

egre

ssio

nsth

atR

ely

onC

ross

-Sec

tion

alV

aria

tion

inC

orpo

rate

Tax

Rat

es

(1)

(2)

(3)

(4)

Tax

vari

able

s(a)

ln

(dif

fere

nce

inp

ayof

f)0

099

(00

41)

20

038

(00

36)

20

005

(00

44)

Pos

t-T

RA

862

000

5(0

013

)H

igh

corp

orat

eta

xra

te0

018

(00

17)

(Pos

t-T

RA

86)

3hi

gh0

024

(00

16)

corp

orat

eta

xra

te

Cor

por

ate

gove

rnan

ceva

riab

les

Frac

tion

of

rmrsquos

shar

es0

109

(00

27)

011

4(0

015

)2

000

9(0

030

)0

087

(00

34)

owne

dby

inst

itu

tion

alin

vest

ors

Size

ofth

e

rmrsquos

boar

dof

20

004

(00

01)

20

001

(00

01)

20

002

(00

01)

000

1(0

002

)d

irec

tors

Frac

tion

ofd

irec

tors

wh

o2

012

4(0

033

)2

000

3(0

018

)2

004

1(0

034

)0

001

(00

39)

are

insi

de

orgr

ay

Con

trac

t-th

eory

vari

able

sD

(CE

Ow

ealt

h)p

er$1

000

20

238

(00

94)

20

167

(00

53)

20

257

(00

96)

20

021

(01

47)

inFV

(b)

D(C

EO

wea

lth)

per

10

in2

560

(17

8)2

592

(10

5)2

594

(19

6)2

147

(04

86)

FV(c

)

Vol

atili

ty(d

)2

48(4

24)

162

(03

4)1

12(4

28)

106

(06

09)

ln(

rmm

arke

tva

lue)

003

3(

006)

000

8(0

003

)0

009

(00

06)

001

2(0

007

)

Stoc

km

arke

tre

turn

vari

able

sL

agge

dan

nual

rm

stoc

k2

000

2(0

009

)0

011

(00

10)

000

6(0

009

)2

001

2(0

013

)m

arke

tre

turn

Lag

ged

annu

alSamp

P50

02

005

6(0

023

)2

005

3(0

051

)2

003

9(0

049

)2

015

4(0

040

)re

turn

Reg

ress

ion

dia

gnos

tics

T

otal

sam

ple

size

5179

5179

5179

2534

Firm

s42

742

739

6A

dju

sted

R2

031

009

032

005

Firm

xed

effe

cts

Yes

No

Yes

No

Tim

ed

um

mie

sN

oY

esY

esN

o

Dep

end

ent

vari

able

isfr

acti

onof

ann

ualc

omp

ensa

tion

from

stoc

kop

tion

sSt

and

ard

erro

rsin

par

enth

eses

(a)

All

thre

esp

eci

cati

ons

use

tric

hoto

mou

sco

rpor

ate

tax

rate

sT

he

sam

ple

inco

lum

n(3

)u

ses

only

the

year

s19

84to

1988

sin

cew

ear

eu

sing

1986

asa

nat

ural

exp

erim

ent

(b)

Thi

sva

riab

leis

the

dol

lar

chan

gein

CE

Ow

ealt

hp

er$1

000

chan

gein

rm

mar

ketv

alu

e(s

eeJe

nse

nan

dM

urp

hy

1990

)

(c)

Thi

sva

riab

lein

the

dol

lar

char

gein

CE

Ow

ealt

hp

er10

-per

cent

chan

gein

rm

valu

e(s

eeH

alla

nd

Lie

bman

199

8)

(d)

Stan

dar

dd

evia

tion

ofr

mm

arke

tva

lue

intr

illio

ns

32 Hall amp Liebman

TABLE 5Large Institutional Investors and Stock-Option Compensation

The Cross-sectional Relationship between Institutional Investors andOption Grants

(1) (2) (3) (4)

Coefcient on fraction of 217 (067) 145 (014) 177 (069) 120 (015)rmrsquos shares owned byinstitutional investors

Sample period 1994 1980ndash1994 1994 1980ndash1994Contract-theory and stock No No Yes Yes

market covariatesTime dummies No Yes No Yes

The dependent variable is the share of compensation paid in options

factor alone can explain 6 percentage points of the 23-percentage-pointincrease in the median share of compensation paid in options that oc-curred over this time period Of course this variable is only one measureof the strengthening of corporate governance so the total impact of im-proved corporate governance could be greater The second and fourthcolumns of the table use the entire 1980ndash1994 sample and include timedummies to isolate the cross-sectional correlations Again there is evi-dence of a strong cross-sectional relationship between the two variablesalthough the coefcients are slightly smaller perhaps reecting a weakercorporate-governance role in the earlier period In sum there is a strongrelationship in both longitudinal and cross-sectional data between theinstitutional-investor variable and the options-percentage variable13

7 THE MILLION-DOLLAR RULE

In 1993 section 162(m) of the Internal Revenue Code was enactedlimiting the deductibility of executive compensation in excess of onemillion dollars unless the compensation was performance-related Pro-ponents of this legislation argued that it would slow the growth ofexecutive pay and tighten the link between rm performance and execu-tive compensation

The new provision became effective for tax year 1994 and applies tothe CEO and the other four most highly compensated executives in eachcompany Pay that is performance-related such as stock-option grants

13 The one specication in which the fraction owned by institutional investors is notsignicant is in column (2) of Table 4 a regression that includes both xed effects and timeeffects

Taxation of Executive Compensation 33

and bonuses paid for meeting clear performance goals is not affected bythis regulation provided that it has been approved by shareholdersThus the main impact of the provision is to limit the deductibility ofsalary in excess of a million dollars Woodlock and Antenucci (1997)studied the proxy statements of 376 rms and documented that mostrms responded to the new law by qualifying top executive pay asperformance-related Perry and Zenner (1999) identied 25 rms thatreduced salaries from above $1 million to below $1 million and foundthat 23 of the rms cited section 162(m) as their reason for doing so

In this section of the paper we investigate whether this provision hasaffected the level or structure of executive pay using the Execucompdata base which contains information on the highest-paid executives ineach of the rms in the SampP 500 SampP Midcap 400 and SampP SmallCap600 Between 1993 and 1998 the median salary of the Execucomp execu-tives rose from $230000 to $306000 The median value of total compen-sation (salary bonus and option grants) rose from $470000 to $882000Limiting the sample of CEOs indicates that the total compensation of themedian CEO rose from $11 million to $18 million over the period whilethe median salary rose from $441000 to $550000 Thus executive com-pensation clearly continued to rise at a rapid rate after the implementa-tion of the million-dollar rule though perhaps at a slower rate than itotherwise would have

If the regulation affects behavior it should have the largest impact onrms whose executives are receiving salaries that are above the limit orjust below it Thus we might expect to see rms whose executives arereceiving relatively high salaries to rely more heavily on performance-based pay and less heavily on salary in the years following the regulationPerry and Zenner (1999) and Livingston (1999) both nd that this is thecase This trend can be clearly identied in the Execucomp data set Forexample over 1993 to 1998 the median annualized growth rate of salaryfor executives whose 1993 salary was below $500000 in the previous yearwas 85 percent For executives with salary between $500000 and$700000 it was 64 percent for executives between $700000 and $850000it was 50 percent for executives between $750000 and $1 million it was31 percent and for executives above $1 million it was 00 percent14

The observation that executives at lower levels of salary had highergrowth rates is not very strong evidence of an impact of section 162(m)however because it is possible that low-salary executives would havehad higher growth rates even in the absence of the provision To explore

14 Of the 3137 executives for whom we have complete data between 1993 and 1998 2708had 1993 salaries below $500000 while only 35 had 1993 salaries above $1 million

34 Hall amp Liebman

whether this is the case we use Execucomp data from both before andafter the law change Ideally we would want to have data from manyyears before the change to examine whether there is an underlying pat-tern of faster salary growth for low-salary executives Unfortunately theExecucomp data begin in 1992 and therefore provide us with only the1992ndash1993 growth rate in compensation to use as a baseline15

We run regressions pooling data from 1992ndash1993 through 1997ndash1998of the annual percentage change in different forms of compensation on avariable million that is designed to measure the likelihood that anexecutiversquos compensation will be affected by section 162(m) We denemillion as the minimum of 1 and the executiversquos salary divided by $1million in the previous year Thus any executive with salary at or above$1 million would receive a value of 1 while an executive with a salary of$500000 would be coded as a 0516 We include this variable by itself andalso interacted with a dummy variable (after) that equals 1 for years1994 and after Thus the million variable is designed to pick up anyunderlying relationship between the level of salary and the growth rateof salary while the interacted one will identify any differential growthrate of salary by income in the period after section 162(m) took effect Wealso include a full set of year dummies the natural logarithm of thermrsquos market value and the rmrsquos annual rate of return over the twoprevious years as control variables To reduce the sensitivity of our re-sults to outliers we run quantile regressions and robust regressions

Table 6 contains our results The rst four columns are for all topexecutives while the last four columns focus on CEOs Columns (1) and(2) present estimates from robust regression for the percentage change insalary and the percentage change in total compensation (salary bonusstock grants and stock-option grants) In the salary growth regressionsfor all executives the coefcient on million is negative and signicantindicating that there is an underlying relationship between the level ofexecutive compensation and its subsequent growth rate The coefcienton million 3 after is negative as well suggesting that the negativerelationship between the level of salary and its subsequent growth rateintensied slightly after 1994 The relatively small magnitude suggeststhat an executive with a $1 million dollar salary would see his salarygrow at an annual rate that would be about 06 percent slower than anexecutive earning $500000

15 The HallndashLiebman data set contains only a combined salary and bonus variable makingit impossible to use it to analyze this issue16 We also ran the regressions (not reported) allowing for a nonlinear relationship betweenmillion and compensation growth and obtained substantively similar results

TA

BL

E6

The

Eff

ect

ofth

eM

illi

on-D

olla

rR

ule

onSa

lary

and

Tot

alC

ompe

nsat

ion

All

top

exec

uti

ves

CE

Os

only

Rob

ust

regr

essi

onQ

uan

tile

regr

essi

onR

obu

stre

gres

sion

Qu

anti

lere

gres

sion

D

in

Din

D

in

Din

D

in

Din

D

in

Din

sala

ryto

talc

omp

(a)

sala

ryto

talc

omp

sa

lary

tota

lcom

p

sala

ry

tota

lcom

p

mil

lio

n(b

)2

006

02

035

72

006

32

024

12

008

72

023

12

010

32

024

5(0

005

)(0

032

)(0

006

)(0

027

)(0

010

)(0

073

)(0

011

)(0

056

)m

illi

on

32

001

10

026

20

021

20

015

20

016

014

92

003

40

158

aft

er(c

)(0

005

)(0

033

)(0

006

)(0

029

)(0

010

)(0

076

)(0

011

)(0

057

)N

359

5035

950

359

5035

950

757

17

589

757

17

589

Stan

dar

der

rors

inp

aren

thes

esR

egre

ssio

nsal

soin

clu

de

year

du

mm

ies

stoc

km

arke

tre

turn

and

log

mar

ket

valu

e

(a)

Tot

alco

mp

ensa

tion

incl

ud

essa

lary

bon

us

rest

rict

edst

ock

and

stoc

k-op

tion

gran

ts

(b)

mil

lion

isd

ene

das

min

(1s

alar

yin

pre

viou

sye

ars

div

ided

by$1

mil

lion)

(c)

aft

eris

anin

dic

ator

vari

able

equ

alto

zero

befo

reth

ere

gula

tion

took

plac

e(1

993)

and

one

afte

rwar

d

36 Hall amp Liebman

In the second column the coefcient on million 3 after is essentiallyzero suggesting that any decrease in salary brought about by section162(m) was offset by increases in bonus and stock-option grants Theresults are similar for the median regressions There is some evidence ofa very small slowdown in growth of salary for executives likely to beaffected by section 162(m) but no sign of a change in total compensa-tion Taken together this provides evidence of a very small substitutionof performance-related pay for salary

The last four columns run identical tests but limit the sample to CEOsThe interaction of million 3 after continues to be small and negative inthe salary regressions In contrast the impact of 162(m) on total compen-sation appears to be positive and fairly substantial (about 15 percentagepoints per year) in these regressions That is the increase in bonus andstock options following the million-dollar rule more than offset the verysmall decline in salary

All of these conclusions should be interpreted somewhat cautiouslybecause our data provide us with only one pre-1993 control year Never-theless the data are consistent with section 162(m) having led to a veryminor slowdown in salary growth and one that is dramatically smallerthan what one would conclude without controlling for underlying differ-ential salary growth rates at different levels of salary Since total compen-sation for CEOs did not decline and perhaps increased there is evidenceof a minor substitution of performance-related pay for salary in responseto the regulation

8 EXECUTIVE COMPENSATION AND THEELASTICITY OF TAXABLE INCOME

Even if taxes have only a minor effect on executive compensation thestructure of executive compensation is important for tax policy be-cause high-income taxpayers are sources of signicant revenue and arethe focus of important debates about the efciency cost of taxation

Goolsbee (1999) has recently argued that executive decisions to exer-cise options are highly responsive to intertemporal differences in taxrates created by tax reforms He claims that much of the apparent declinein taxable income for high-income taxpayers between 1992 and 1993documented by Feldstein and Feenberg (1996) can be explained by execu-tives shifting option gains into 1992 in order to avoid the higher marginaltax rates that went into effect in 1993 By extension he implies that thehigh elasticities of taxable income with respect to personal tax ratesestimated by comparing the pre- and post-tax-reform taxable income of

Taxation of Executive Compensation 37

high-income individuals reect intertemporal shifting of income and notpermanent effects of taxation If this is true then the deadweight loss oftaxation is lower as is the cost of progressivity

While Goolsbee presents comprehensive evidence that option gainswere unusually high in 1992 his identication of the impact of tax rates onthe timing of option gains depends on correctly specifying a counter-factual level of options that would have been exercised in the absence ofthe tax change He simply assumes a linear time trend The purpose ofGoolsbeersquos time trend is presumably to account for the increasing use ofoptions over time However since options typically require a vestingperiod of at least a couple of years a much more direct measure of ex-pected option gains can be constructed based on the total value of vestedoptions held by the CEO Since options typically have vesting periods thatexceed two years this total can be treated as exogenous since it will not beaffected by the tax changes

More generally the value of options held by the CEO is likely to havea major effect on his exercise decisions If his options are out of themoney he will not have gains to realize If there has been a recent run-up in the market an optimizing CEO may exercise an unusually largeamount of options in order to diversify his overall portfolio and if in-the-money options are about to expire they will be exercised in thecurrent year

Our evidence suggests that much of the apparent tax shifting caninstead be attributed to stock market performance and the timing of pastoption grants Using our 15-year panel we replicate Goolsbeersquos (1999)result for the 1993 period but show that the impact of taxes is notpresent in other time periods Moreover even the result for the 1993period disappears when appropriate controls for past option grants andstock price appreciation are included

The top three rows of Table 7 present data on the taxable incomes ofCEOs in the HallndashLiebman sample For these results our panel has beennarrowed to include only CEOs whose rmrsquos scal years correspondwith tax years In addition we create a balanced panel for each of thethree tax-reform periods by excluding any rm for which there are notcomplete data for the period surrounding the reform

The fourth row of the table shows that the 1993 tax act raised the topmarginal personal income tax rate from 310 to 396 percent effective fortax year 1993 with a further increase occurring in 1994 due to the uncap-ping of the Medicare payroll tax The top row of the table shows thattaxable income was higher in 1992 than for any other year in our sample(all data are in 1994 dollars) Option gains were particularly high in1992mdashmore than double their 1991 level and 68 percent higher than their

TA

BL

E7

Tax

Ref

orm

Sto

ckM

arke

tP

erfo

rman

cea

ndth

eT

imin

gof

Opt

ion

Gai

ns

Ave

rage

stoc

km

arke

tM

ean

CE

OM

ean

CE

Ore

turn

for

Mea

nva

lue

taxa

ble

sala

ryan

dM

ean

CE

OT

opfe

der

al

rms

inof

opti

ons

(Val

ue

ofop

tion

Sam

ple

inco

me

bonu

sop

tion

gain

sp

erso

nal

sam

ple

for

held

atbe

gga

ins)

(va

lue

Ref

orm

size

Yea

r($

mil

)($

mil

)($

mil

)M

TR

pre

vye

ar(

)of

year

ofop

tion

she

ld)

ER

TA

8127

919

800

874

064

80

163

7025

80

460

354

1981

083

10

672

009

869

245

070

61

3919

820

872

067

90

099

5011

90

686

144

1983

099

00

724

025

250

355

116

62

16

TR

A86

295

1985

119

70

837

024

050

76

116

02

0719

861

448

091

70

345

5042

81

776

194

1987

165

50

959

059

338

519

12

229

266

1988

163

21

058

039

028

07

237

71

6419

891

656

105

20

414

2817

92

345

176

1993

272

1990

209

51

028

061

828

232

288

02

15T

axA

ct19

911

875

102

00

547

312

01

268

82

0319

922

678

107

01

250

3146

64

552

275

1993

232

41

209

074

339

620

64

299

173

1994

196

41

262

037

842

515

45

113

074

Sam

ple

isH

allndash

Lie

bman

rm

sw

ith

Dec

embe

r

scal

year

sth

atha

dco

mp

lete

dat

afo

ron

eof

the

thre

eta

x-re

form

per

iod

sst

ud

ied

Taxation of Executive Compensation 39

1994 level These patterns are consistent with Goolsbeersquos claim that tax-payers responded to the anticipated increase in marginal tax rates byshifting gains into 1992

The evidence from the other two tax reforms is less clear suggestingthat there might be factors other than tax rates determining the timing ofoption gains ERTA 1981 reduced the top marginal tax rate from 70percent in 1980 to 50 percent in 1982 However there is no sign thatoption gains were reduced in 1980 and 1981 and then increased in 198217

There was a big increase in option gains in 1983 however Similarly thehighest year of stock-option gains in the period around 1986 was 1987 astrange year to take gains since taxpayers would have known that thepersonal rate would be 10 percentage points lower if they waited untilthe following year18

What all the peak years of option gains have in common is that theyfollow years of strong stock market performance The fth row of Table 7shows the average stock market return during the previous year for therms in our sample The large value of option gains in 1983 followed a355-percent increase in the stock market The 1992 boom in option gainsfollowed a 466-percent increase in the stock market the more modest1987 peak in option gains followed a 19-percent gain (and a 43-percentgain two years before)

Strong stock market performance raises the value of options availableto exercise The sixth row of the table shows the mean value of optionsheld by CEOs at the beginning of the year All three of the peak years ofoption gains correspond with peak years in terms of the value of optionsavailable for exercise Since stock market appreciation during the yearalso increases the value of options available to exercise later in that yearwe need to allow for current-year stock market appreciation as well

Table 8 contains Goolsbee-style regressions for each of the tax re-forms The rst two columns reprint the results from Goolsbee (1999)The third and fourth columns replicate his results in our sample Inparticular with a specication similar to his we are able to replicate hisnding that the current after-tax share has a positive coefcient and thefollowing yearrsquos after-tax share has a coefcient of roughly negative oneColumns (5) through (8) show that this pattern was unique to the 1993tax reform Indeed in the TRA 86 regressions the coefcient on the

17 As we explained above for executives covered by the maximum tax on earned incomethere was not a decline in tax rates between 1981 and 198218 It is possible that the large stock market decline in October 1997 may have promptedsome people to exercise options earlier than they otherwise would have However anincrease in option exercises typically follows stock market increases not declines

TA

BL

E8

Att

empt

sto

Rep

lica

teth

eT

ax-S

hift

ing

Res

ult

inO

ther

Tim

eP

erio

ds

Goo

lsbe

eG

ools

bee

1993

1993

Ent

ire

Ent

ire

regr

essi

onre

gres

sion

refo

rmre

form

TR

A86

TR

A86

ER

TA

81E

RT

A81

1980

ndash199

419

80ndash1

994

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

ln[1

2ta

x(t)

]1

244

108

20

455

032

62

017

62

017

22

016

72

016

70

026

20

068

(01

34)

(01

27)

(05

37)

(05

42)

(02

10)

(02

09)

(00

90)

(00

91)

(00

67)

(00

65)

ln[1

2ta

x(t1

1)]

20

867

20

764

21

214

20

917

047

80

722

20

103

004

30

176

018

3(0

118

)(0

112

)(0

491

)(0

499

)(0

157

)(0

174

)(0

072

)(0

072

)(0

068

)(0

066

)

Tim

e0

135

007

62

000

52

036

40

041

000

80

094

003

50

056

002

7(0

008

)(0

008

)(0

049

)(0

049

)(0

028

)(0

028

)(0

025

)(0

026

)(0

003

)(0

003

)

ln(m

arke

tva

lue)

066

10

489

023

70

334

039

5(0

015

)(0

058

)(0

039

)(0

042

)(0

017

)

Yea

rs19

91ndash1

995

1991

ndash199

519

90ndash1

994

1990

ndash199

419

85ndash1

989

1985

ndash198

919

80ndash1

983

1980

ndash198

319

80ndash1

994

1980

ndash199

4

Ad

just

edR

20

720

770

610

630

690

760

800

810

630

68

Obs

erva

tion

s13

856

132

371

385

136

41

470

143

31

113

106

23

968

385

9

Dep

end

ent

vari

able

isth

elo

gof

taxa

ble

inco

me

All

regr

essi

ons

con

tain

rm

xe

def

fect

sG

ools

bee

regr

essi

on(2

)al

soco

ntai

nstw

oot

her

vari

able

sno

tsh

own

inth

eta

ble

Stan

dar

der

rors

inp

aren

thes

es

Taxation of Executive Compensation 41

TABLE 9Alternative Explanations for the Timing of Option Exercises

1993 Entirereform TRA 86 ERTA 81 1980ndash1994

(1) (2) (3) (4)

ln[12tax(t)] 0108 0111 20051 0201(0324) (0136) (0048) (0057)

ln[12tax(t11)] 20134 0573 0125 0150(0336) (0156) (0053) (0066)

ln(market value) 0418 0177 0338 0418(0061) (0039) (0046) (0163)

Firmrsquos stock market return (t) 0298 0268 0214 0258(0044) (0045) (0031) (0023)

Firmrsquos stock market return 0162 0206 0031 0128(t21) (0041) (0008) (0029) (0021)

ln(value of options held by 0004 0008 0001 0009CEO at start of year) (0003) (0003) (0002) (0001)

Years 1990ndash1994 1985ndash1989 1980ndash1983 1980ndash1994Adjusted R2 064 071 082 068Observations 1349 1427 1036 3812

The dependent variable is the log of taxable income All regressions contain rm xed effects Stan-dard errors in parentheses

following yearrsquos tax rate is positive and signicant Columns (9) and (10)pool all three reforms and fail to replicate the pattern that would reectintertemporal shifting19

It is possible that the ease with which one can shift the timing ofoption gains is asymmetric It may be harder to shift option gains for-ward in time since some options expire and therefore must be exercisednow In addition options were a much smaller portion of compensationin the early 1980s so perhaps the ERTA 81 results are not surprisingNonetheless these results cast doubt on the proposition that this is thechannel through which spurious estimates of the elasticity of taxableincome have occurred for previous tax reforms and they suggest thatwe should look harder for alternative explanations for the timing ofoption gains

Table 9 attempts to predict options gains using the lagged and currentincrease in a rmrsquos market value and the value of the options held by theCEO at the start of the year The two stock market variables stronglypredict option gains while the value of options held has the correct sign

19 The results also fail to suggest large permanent effects of marginal tax rates on taxableincome

42 Hall amp Liebman

in all of the specications and is signicant in the pooled sample eventhough it is highly correlated with the two stock market return variablesIn contrast the tax-timing explanation fails even in the 1993 period

It is important to realize that our results do not invalidate Goolsbeersquoscritique of claims that aggregate revenue data from 1992 and 1993 provethat elasticities of taxable income are large We have simply given analternative explanation for why taxable incomes were unusually high in1992 However our results do cast doubt on the conclusion that largetiming shifts are ubiquitous and capable of explaining away the morecareful estimates of taxable income elasticities that have been performedusing micro data (eg Feldstein 1995 Auten and Carroll 1995)

One further point is in order It is sometimes argued that the combina-tion of increased use of options and the booming stock market explainswhy federal tax revenues have been so much higher than expected dur-ing the 1990s However while large option gains increase revenue fromthe personal income tax they produce offsetting deductions for cor-porations The net impact on revenue is only the difference in the twomarginal tax rates times the option gain While the gap between thepersonal rate and the effective corporate rate is nontrivial the impact oflarge option exercises on total revenue is substantially smaller than theamount by which the exercises inate personal tax revenue and reducecorporate tax revenue

9 CONCLUSION

We have described the tax rules for executive pay in detail and analyzedhow changes in various tax rates affect the tax advantages of stock optionsrelative to salary and bonus We nd that there is a moderate tax advan-tage to optionsmdashcurrently about $4 per $100 in pretax executive paymdashand that changes in the tax advantage of options over time have had atmost a modest impact on the composition of pay Corporate governancefactors particularly the role of large investors appear to be more impor-tant in explaining the dramatic increase in option pay

In addition our evidence suggests that more direct attempts to use taxpolicy to inuence executive compensation have had little effect Wend that the million-dollar rule led companies to substitute perfor-mance-based pay for salary But our evidence suggests that this substitu-tion was quite modest and there is no evidence that the total level of paywas reduced Overall although the stock-option explosion has dramati-cally increased the link between pay and performance this change isdue almost entirely to nontax factors

Taxation of Executive Compensation 43

REFERENCES

Auerbach Alan J (1996) ldquoThe Dynamic Effects of Tax Law Asymmetriesrdquo Re-view of Economic Studies 53451ndash405

mdashmdashmdash and James Poterba (1987) ldquoTax Loss Carryforwards and Corporate TaxIncentivesrdquo In The Effects of Taxation on Capital Accumulation M Feldstein(ed) Chicago University of Chicago Press

Altshuler Rosanne and Alan Auerbach (1990) ldquoThe Signicance of Tax LawAsymmetries An Empirical Investigationrdquo Quarterly Journal of Economics 10561ndash86

Auten Gerald and Robert Carroll (1995) ldquoThe Effect of Income Taxes on House-hold Behaviorrdquo Ofce of Tax Analysis US Department of the Treasury

Baker George P and Brian J Hall (1998) ldquoCEO Incentives and Firm SizerdquoNBER Working Paper no 6868 December

Core John and Wayne Guay (1999) ldquoThe Use of Equity Grants to ManageOptimal Equity Incentive Levelsrdquo Draft

Feldstein M (1995) ldquoThe Effect of Marginal Tax Rates on Taxable Income APanel Study of the 1986 Tax Reform Actrdquo Journal of Political Economy 103 (no3)551ndash72

mdashmdashmdash and D Feenberg (1996) ldquoThe Effect of Increased Tax Rates on TaxableIncome and Economic Efciency A Preliminary Analysis of the 1993 Tax RateIncreasesrdquo In Tax Policy and the Economy vol 10 James M Poterba (ed)Cambridge MA MIT Press

Gompers Paul A and Andrew Metrick (1998) ldquoInstitutional Investors andEquity Pricesrdquo NBER Working Paper no 6723

Goolsbee Austan (1999) ldquoWhat Happens When You Tax the Rich Evidencefrom Executive Compensationrdquo NBER Working Paper no 6333

Graham John R (1996) ldquoProxies for the Corporate Marginal Tax Raterdquo Journal ofFinancial Economics 42187ndash221

Hall Brian J (1999) ldquoThe Design of Multi-year Stock Option Plansrdquo Journal ofApplied Corporate Finance 12(2)97ndash106

mdashmdashmdash and Jeffrey B Liebman (1998) ldquoAre CEOs Really Paid Like Bureau-cratsrdquo Quarterly Journal of Economics CXII(no 3)653ndash691

Jensen Michael C and William M Meckling (1976) ldquoTheory of the Firm Mana-gerial Behavior Agency Costs and Ownership Structurerdquo Journal of FinancialEconomics III305ndash360

mdashmdashmdash and Kevin J Murphy (1990) ldquoPerformance Pay and Top-ManagementIncentivesrdquo Journal of Political Economy XCVIII225ndash264

Kaplan Steven N (1997) ldquoThe Evolution of US Corporate Governance WeAre All Henry Kravis Nowrdquo Draft

Livingstone Jane R (1999) ldquoExecutive Compensation Contracting and Re-sponses to Increased Tax Costsrdquo Louisiana State University June

Miller Merton H and Myron S Scholes (1982) ldquoExecutive CompensationTaxes and Incentivesrdquo In Financial Economics Essays in Honor of Paul CootnerWilliam F Sharpe and Cathryn M Cootner (eds) Princeton NJ Prentice-Hall

Murphy Kevin J (1999) ldquoExecutive Compensationrdquo forthcoming in OrleyAshenfelter and David Card (eds) Handbook of Labor Economics Vol 3 NorthHolland

Perry Todd and Marc Zenner (1999) ldquoPay for Performance Government Regu-

Taxation of Executive Compensation 15

posttax cost to the company By keeping posttax employer costs (in NPVterms) constant any package that is preferred by the employee is tax-advantaged in the global-contracting sense

Under this methodology a comparison of the tax advantages of op-tions and cash involves comparing a pretax cash payment of P with anoption payment that has the equivalent posttax NPV to the companyFor NQSOs it is straightforward to show that a pretax payment of P toan executive has exactly the same cost to the company as putting aside Pfor the purposes of paying stock-option payouts later That is if a com-pany pays an executive P today it will have the same amount of moneyin N years as if it had put aside P today let it accumulate at the rate ofreturn earned by the rm and then paid it out (as compensation in theform of option gains4) with any appreciation in year N

Assume that the pretax prot rate is r and the corporate tax rate is TcThen if the company pays P today it receives a deduction today of PTcwhich yields

PTc [1 1 r (1 2 Tc)]N (1)

in N years since the benets of the tax deduction are invested in thecompany and receive the after-tax corporate rate of return Converselyif the company puts P aside today then it grows at the after-tax corpo-rate rate of return in N years to give a tax deduction of

P [1 1 r (1 2 Tc)]N Tc (2)

which is the exact same value Note that the equivalence of these two taxdeductions is analogous to the tax-benet equivalence of front-loadedand back-loaded IRAs

An executiversquos payoff in N years from option prots is equal to Pcompounded at the after-tax corporate rate times 1 2 Tp where Tp is thepersonal tax rate Thus the combined payoffs in N years of the corporatededuction and the executiversquos posttax payoff is

P [1 1 r (1 2 Tc)]N Tc 1 P [1 1 r (1 2 Tc)]

N (1 2 Tp) (3)

where the rst term is the payoff from the corporate deduction and thesecond term is the executiversquos payoff

4 This analysis is not unique to options Any form of deferred compensation that enablesexecutives to invest inside the rm at a higher posttax return or without paying capitalgains taxes would have similar effects

16 Hall amp Liebman

The payoff to salary and bonus in N years is more complicated sincean assumption must be made about how the executive invests the origi-nal cash compensation For example if the executive invests in an instru-ment with taxable interest (eg bonds) then the N-year payoff from aninvestment of P accumulates at the post-corporate-tax post-personal-taxrate of return and is

P (1 2 Tp) [1 1 r (1 2 Tc)(1 2 Tp)]N (4)

However if an executive invests in (non-dividend-paying) equities theinvestment accumulates at the higher post-corporate-tax rate of returnwith the offsetting disadvantage that the capital gains are taxed at thecapital gains rate in year N So the executive receives

P [1 1 r (1 2 Tc)]N (1 2 Tp) 2 Tcg $P (1 2 Tp) [1 1 r (1 2 Tc)]

N 2P (1 2 Tp) (5)

where Tcg is the capital gains tax rate5Thus putting the equations together if an executive is paid cash P

and invests g in bonds and 1 2 g in equities the N-year payoff (includ-ing the payoff to the corporation from the tax deduction) is

P [1 1 r (1 2 Tc)]N Tc 1 g $(P [1 1 r (1 2 Tc)(1 2 Tp)]

N (1 2 Tp)1(1 2 g) (P [1 1 r (1 2 Tc)]

N (1 2 Tp) 2 Tcg $P (1 2 Tp)[1 1 r (1 2 Tc)]

N 2 P (1 2 Tp)) (6)

The tax advantage of options versus salary and bonus therefore is thedifference between the combined (corporate and executive) payoffs inequation (3) and the combined payoffs in equation (6)

43 The Tax Advantage of OptionsWe now turn to analysis of how changes in various tax rates affectthe tax advantages of options relative to cash compensation We focuson the case in which the cash earned by executives is invested entirelyin equities since equities are tax-favored relative to bonds The impor-tant conclusions of this analysis are not substantively different if we

5 In practice executives hold a combination of equities and bondlike instruments withtaxable interest Because equities are tax-favored in the next section we simplify the analy-sis by assuming that executives hold only equities Alternatively we could have assumedthat executives hold only less risky bonds since bonds more closely match the risk proleof a stream of tax savings None of our empirical results are substantively affected by thissimplication

Taxation of Executive Compensation 17

instead assume that a portion of an executiversquos holdings are invested inbonds

431 The Tax Advantage of Options Changing the Corporate TaxRate Stock options are tax-advantaged relative to cash simply becauseoption payouts are deferred allowing the executive to invest at the pre-personal-tax rate of return (with no capital gains tax at the end) ratherthan investing at the after-tax rate of return or paying the capital gainstax Since the advantage of being able to defer taxes is large when thecorporate tax rate is low the tax advantage of options is larger when thecorporate rate is low

To illustrate this point we dene the tax advantage of options to bethe dollar amount by which total after-tax option payoffs (to the corpora-tion and the executive) exceed the total payoffs from salary as denedby the difference of equation (3) and (6) assuming that P (the paymentto the executive) is equal to $100 and g 5 0 (cash compensation is in-vested in equities) We use a ten-year horizon since most options have aten-year term We then calculate how the tax advantage of optionschanges as the corporate tax rate varies from zero to 100 percent holdingthe personal tax rate and the capital gains rate xed at 40 percent and 20percent respectively (which approximates current rates)

The top panel of Figure 2 shows the tax advantage of options relativeto cashmdashprecisely the NPV of the total tax advantage of options for a$100 payment to the CEOmdashas the corporate tax rate changes As ex-pected the tax advantage of options declines as corporate tax rates in-crease At a zero corporate tax rate the tax advantage of options has anNPV of $7 This value is about $350 at a corporate rate of 40 percentWhen the corporate tax rate is 100 percent the tax advantage of optionscompletely disappears since the benets of deferred compensation fallto zero because the after-tax return on equities (and therefore the dis-count rate) falls to zero The tax advantage of options is the NPV of thetax savings from avoiding the capital gains tax on ten years of apprecia-tion which is why it is relatively modest even at the 40-percent corpo-rate rate

432 The Tax Advantage of Options Changing the Personal TaxRate Using the same assumptions as above (but this time holding thecorporate rate at 40 percent and the capital gains rate at 20 percent andvarying the personal rate) the tax advantage of options declines as thepersonal tax rate rises as shown in the middle panel of Figure 2 The taxadvantage of options is higher at low personal tax rates because theposttax base that gives rise to capital gains is higher and the advantage of

Note Holds constant the personal rate at 40 percent and the capital gains rate at 20 percent

Note Holds constant the corporate rate at 40 percent and the capital gains rate at 20 percent

Note Holds constant the personal rate at 40 percent and the corporate rate at 40 percent

FIGURE 2 Tax Advantage of Options NPV of Tax Advantage of Pay-ing $100 in Options (NQSOs) Rather Than in Cash

Taxation of Executive Compensation 19

avoiding the capital gains taxes is greater The tax advantage of optionstherefore declines linearly as the personal tax rate rises At a 100-percentpersonal tax rate the executive receives nothing in either case and byconstruction the companyrsquos tax deduction benet is the same in NPVterms so the tax advantage of options falls to zero also6

433 The Tax Advantage of Options Changing the Capital Gains TaxRate The tax advantage of options for various capital gains rates isdepicted in the bottom panel of Figure 2 under the same assumptions(this time holding both the personal rate and the corporate rate at 40percent while varying the capital gains rate) The tax advantage of avoid-ing the capital gains tax increases as the capital gains tax rate rises

5 HOW HAS THE TAX ADVANTAGE OF OPTIONSCHANGED OVER TIME

The analysis so far has illustrated how the tax advantage of optionsvaries with the changes in tax rates In this section we show how the taxadvantage of options has changed over time as tax rates have changed

51 Top Marginal Tax Rates over TimeThe evolution on the top marginal tax ratesmdashpersonal corporate andcapital gainsmdashfrom 1980 to 1998 is shown in Figure 3 In all cases thetop rate (the rate for taxpayers with the highest incomes) is shownwhich is not always the highest rate since various anomalies (such asclawbacks of exemptions) sometimes lead to marginal tax rates that arehigher than those for the highest income tax payers

The top personal tax rate was 70 percent in 1980 falling to 50 percentin 1982 and 28 percent in 1988 following the 1986 tax act7 The topmarginal tax rate has since risen to 396 percent but is effectively 425percent since there is a 29-percent Medicare surcharge (paid half by theemployer and half by the employee) that has no upper limit The topcorporate tax rate has had only one signicant change since 1980 fallingfrom 46 percent before the 1986 tax act to 34 percent following the act in1988 The top corporate tax rate was increased to 35 percent in 1993 The

6 This is one place where ignoring bond investments has substantive implications Inparticular in the all-bond case the tax advantage of options is nonmonotonicmdashit rises andthen fallsmdashas personal tax rates rise7 During the early 1980s the maximum tax on earned income limited the marginal tax rateon the earnings of high earners so many executives did not face a decline in personal taxrates between 1981 and 1982 The empirical results in this paper are robust to assumingthat executives faced a marginal tax rate of 50 percent in 1980 and 1981

20 Hall amp Liebman

FIGURE 3 Top Marginal Tax Rates since 1980

top capital gains rate has uctuated between 28 percent (approximately)and 20 percent since 1980 and is currently at 20 percent

52 The Evolution of the Tax Advantage of NQSOsThe tax advantage of options relative to cash dened in the same way asin section 4 from 1980 to 1998 is shown in Figure 4 The calculations aredone using the same assumptions as before (ie N 5 10 years and cashis invested in equities) and the statutory top marginal tax rates shown inFigure 3 are used Due to the fall in corporate and personal tax rates thetax advantage of options increased sharply after the Tax Reform Act of1986 This increase in tax advantage has largely been reversed in the1990s as top personal tax rates have crept back up and the capital gainsrate has been reduced

Two key facts emerge from this analysis First there is currently only amoderate tax advantage to standard nonqualied stock optionsmdashon theorder of $4 per $100 of compensation This is because the tax advantagesto the executive of deferring taxes are largely offset by the tax disad-vantages to the company of not being able to deduct option expensesfrom taxable prots until the executive exercises the options Secondalthough the 1986 tax act substantially increased the tax advantage ofoptions more than half of this increase has been reversed in the 1990s

Taxation of Executive Compensation 21

FIGURE 4 Tax Advantage of Options 1980 to 1998

NPV of tax advantage of $100 in options (NQSOs) rather than cash

53 The Tax (Dis)advantage of ISOsSo far the analysis has focused on the tax advantages of NQSOs sincethey are so much more prevalent than ISOs which account for onlyabout 5 percent of option grants The relative scarcity of ISOs can beexplained both by their tax status and by their per-executive cap of$100000 per year

Under what conditions are ISOs tax-preferred to NQSOs For a trans-fer P to an executive the ISO is always tax-preferred by the executive(since the capital gains rate is lower than the personal rate) while theNQSO is always tax-preferred from the companyrsquos perspective (since anISO is not deductible) The key issue however is the relative advan-tages of the two types of options from a global-contracting perspectiveBecause ISOs are not deductible the company is indifferent betweensetting aside P in the form of NQSOs today and setting aside P(1 2 Tc) inthe form of ISOs (Since the NPV of the tax deduction for NQSOs is PTcthe NPV of the payment P is P 2 PTc or P (1 2 Tc))

Equalizing the posttax cost of NQSOs and ISOs we need only look atthe payoff of the executive to determine the condition under which eachis preferred The payoff of NQSOs in N years is

P [1 1 r (1 2 Tc)]N (1 2 Tp) (7)

22 Hall amp Liebman

The same payoff for ISOs is

P (1 2 Tc) [1 1 r (1 2 Tc)]N (1 2 Tcg) (8)

ISOs therefore are tax-advantaged if

Tp 2 Tcg Tc 2 Tc Tcg (9)

Intuitively ISOs are tax-advantaged only if their advantage (the differ-ence between the personal rate and the capital gains rate) is largeenough to offset their cost (the disadvantage of not deducting at thecorporate rate) Note however that the condition is not a simple com-parison between the corporate rate and the personal rate minus thecapital gains rate (Tp 2 Tcg Tc)

Figure 5 shows how the tax advantage of NQSOs relative to ISOs haschanged since 1980 as tax rates have changed Unlike the very modestchanges (around 3 to 4 percent) in the relative tax advantages betweenoptions and cash shown in Figure 4 the relative tax advantage betweenNQSOs and ISOs has seen enormous swings When the top personalrate was 70 percent in 1980 ISOs were tax-favored by a margin greaterthan 6 percent However by 1982 (following the 1981 tax acts that dra-matically lowered the personal tax rate) ISOs became tax-disadvantaged

FIGURE 5 Tax Advantage of NQSOs Relative to ISOs over Time

NPV of tax advantage of $100 payment to executive NQSOs minus ISOs

Taxation of Executive Compensation 23

by about 5 percent By 1988 (following the 1986 tax act) ISOs became tax-disadvantaged by about 18 percent a dramatic swing The raising ofpersonal rates coupled with a decrease in the capital gains rate has sincereduced the tax disadvantage of ISOs to about $4 per $100 of pretaxcompensation Although hard data on ISOs are hard to come by practi-tioners (mostly compensation consultants) claim that ISOs were morepopular prior to 1982 when they were tax-advantaged Although ISOsare less disadvantageous from a tax standpoint relative to NQSOs thanthey were in the late 1980s they are still disadvantageous so it is notsurprising that they have not made a signicant resurgence in recentyears

6 EXPLAINING THE INCREASE IN OPTION-BASEDCOMPENSATION

In this section we analyze whether changes in tax rates affect the compo-sition of executive compensation and consider the relative importanceof tax factors and corporate governance factors in explaining the in-creasing share of compensation paid in the form of stock options Thenumerous changes in tax rates that have occurred since 1980 provide anopportunity to assess whether executive compensation arrangementsrespond to tax incentives in the way that the tax-avoidance model sug-gests We test this model using a panel data set of CEOs in largepublicly-traded US corporations Our identication of the tax effectsrelies on time-series variation in personal corporate and capital gainstax rates along with cross-sectional variation in corporate tax rates

61 DataWe use a 15-year panel data set of CEOs in the largest publicly-tradedUS corporations which is described in Hall and Liebman (1998) Thedata set covers the years from 1980 through 1994 and combines CEOcompensation information from corporate proxies and 10-K lings withstock price and stock return information from CRSP and accounting datafrom Compustat In addition some compensation data from the 1970swere collected in order to construct measures of the value of stock op-tions held by the CEOs in the rst period of the sample8

The feature that distinguishes our data from most other CEO datasets is that with our panel of yearly proxy data on option grants optiongains and total options held we are able to calculate the total value of

8 We are grateful to David Yermack for providing us with some of the data for 1984 to 1991See Yermack (1995) for a discussion of these data

24 Hall amp Liebman

all stock options held by the CEO at a point of time More importantlysince we have the details about the stock options held (number exer-cise price time to maturity etc) we can precisely calculate the changein the value of a CEOrsquos stock option holdings for a given change in rmvalue

62 Identication IssuesThe numerous tax changes over our sample period and the specic waysin which changes in personal corporate and capital gains rates arepredicted to affect the composition of CEO compensation give us un-usually rich sources of identication However because many of thechanges in tax rates would be expected to affect all of the CEOs in oursample in a similar way we will need to pay particular attention toseparating out the tax effects from underlying time trends and otherfactors that changed over time Controlling for underlying trends is par-ticularly important in this study because we are focusing on an out-comemdashthe use of stock optionsmdashthat has increased rapidly over thepast two decades and that many practitioners believe was importantlyaffected by nontax considerations Thus the tax effects we are trying toexplain are deviations from a rising trend

We take two steps to try to separate out the tax effects from theunderlying trend First we include key nontax factors that could poten-tially explain the increasing reliance on performance-based compensa-tion It has been suggested that the dramatic increase in incentive-basedpay is the result of the remarkably poor shareholder returns during the1970s which spurred the LBO and takeover movements of the 1980sAccording to this story (Kaplan 1997) shareholders became much morepowerful via the rise of institutional investors even as the LBO andtakeover movements waned Because the inuence of institutional inves-tors is thought to be one of the most important mechanisms of strongcorporate governance we use the share of each companyrsquos stock ownedby institutional investors (dened as institutions with more than $100million under management) as an explanatory variable9 The percentageof shares owned by large institutional investors increased from about 20percent to almost 50 percent in 1994 an upward trend that closelymatches the sharp rise in the share of compensation in the form ofoptions Indeed in our sample the annual average (over all of the rmsin our sample) share of stock owned by institutional investors has acorrelation of 09 with the annual average share of compensation paid inoptions

9 This variable is described in Gompers and Metrick (1998)

Taxation of Executive Compensation 25

In addition we include two other variables that proxy for strongercorporate governance The rst is the size of the board Evidence suggeststhat boards with smaller numbers of directors reduce the inuence of theCEO and better represent shareholder interests (Yermack 1995) Boardsize fell by approximately 15 percent between 1980 and 1994 The secondvariable is the fraction of the rmrsquos board members who are inside direc-tors or gray directors (noninsiders who have business dealings with thecompany) Firms with fewer outsiders are less likely to act in the share-holdersrsquo interests and we would predict that they would be less likely tohave performance-based compensation for their CEOs The average per-centage of inside and gray directors has decreased from about 45 percentin 1980 to about 35 percent in 1994

Second we allow for cross-sectional variation in corporate tax ratesCorporate marginal tax rates vary because rms may be eligible for tax-loss carrybacks and carryforwards investment tax credits and the alter-native minimum tax (see Auerbach 1986 Auerbach and Altshuler1990 Auerbach and Poterba 1987) Because these cross-sectional differ-ences imply that different rms should respond differently to a givenchange in tax rates they potentially provide us with a way to isolatethe tax effects Firm-specic corporate marginal tax rates are notori-ously difcult to calculate from publicly-available data We use a tricho-tomous variable that equals zero if the rm has tax-loss carryforwardsand negative earnings in a given year equals 05 times the statutorycorporate rate if the rm has only one of those conditions and equalsthe statutory rate if the rm has neither loss carryforwards nor negativeearnings10

The cross-sectional variation in the corporate tax rate may not be ex-ogenous Firms that perform poorly could face low marginal tax ratesand also be particularly likely or unlikely to provide performance-basedpay Because most of our specications use xed effects the correlationwe must be concerned about is between changes in rm performanceand changes in compensation We deal with this by controlling forlagged rm stock market performance in our regressions This shouldeliminate spurious correlation between tax rates and compensation thatis jointly caused by rm performance

10 Graham (1996) presents evidence that the trichotomous variable performs nearly as wellas his recommended simulated marginal tax rate Plesko (1999) compares marginal taxrates calculated from Compustat data with actual marginal tax rates from rm tax returnsHe nds that simple binary variables are more reliable measures of the marginal tax ratethan is either the Graham or the trichotomous variable We obtain substantively similarresults when we switch from the trichotomous variable to the binary variables recommendby Plesko

26 Hall amp Liebman

63 ResultsThe dependent variable in our regressions is the fraction of a CEOrsquos totalannual compensation (measured as the BlackndashScholes value of stock-option grants plus salary and bonus) that comes from stock-optiongrants We choose this reduced-form specication because our interest isin the composition of CEO compensation However the factors thataffect the composition of compensation might also affect the levels andit will be important to keep this in mind in interpreting our results

Optimal-contracting theory generally makes predictions about the in-centives provided by the total amount of rm stock and stock optionsowned by the CEO We choose to model the ow of option grants ratherthan the ultimate performance sensitivity of the CEOrsquos compensationand stock and stock-option holdings because we have in mind anadjustment-cost model in which it takes time for a CEOrsquos holdings ofstock and stock options to reach the optimal level11 In our regressionswe include two measures of the current performance sensitivity of theCEOrsquos wealth to rm performance as explanatory variables to reect thedistance of the CEOrsquos contract from the optimal contract

Table 3 presents results in which the tax effects are identied solely byvariation in tax rates over time The fraction of annual compensation paidin stock options is regressed on tax variables corporate governance vari-ables contract-theory variables and variables reecting stock market re-turns Both of the regressions in this table include rm xed effects Incolumn (1) the fraction in annual compensation paid in stock options isregressed on the log difference in payoff (the combined payoff to both therm and the CEO) from receiving compensation in stock options ratherthan salary and bonus where the payoff calculations are based on statu-tory marginal tax rates This variable is dened as the log difference be-tween equation (3) and equation (6) in section 4 The coefcient on the taxvariable is large and statistically signicant at the 95-percent level Thepoint estimate implies that a one-percent increase in the payoff differencebetween stock options and salary and bonus results in a 24-percentage-point increase in the share of compensation paid in stock options

The corporate governance variables all have the predicted sign Firmswith a higher fraction of their shares owned by institutional investorsuse more performance-based pay (stock options) Firms with large corpo-rate boards or a large share of inside directors are less likely to use stockoptions

The two contract-theory variables are the dollar change in CEO wealth

11 See Core and Guay (1999) for evidence that rms adjust in this way

TABLE 3Compensation Regressions that Rely on Variation in

Tax Rates over Time

(1) (2)

Tax variablesln (difference in payoff)(a) 2393 (0349)ln (12statutory personal 20015 (0034)

rate)ln (12statutory cap gains 0330 (0171)

rate)ln (12statutory corporate 0399 (0148)

rate)

Corporate governance variablesFraction of rmrsquos shares 0043 (029) 0052 (0028)

owned by institutionalinvestors

Size of the rmrsquos board of 20002 (001) 20003 (0001)directors

Fraction of directors who 20082 (033) 20092 (0030)are inside or gray

Contract-theory variablesD (CEO wealth) per $1000 20261 (0095) 20247 (0087)

in FV(b)

D (CEO wealth) per 10 in 2583 (189) 2554 (201)FV(c)

Volatility(d) 521 (424) 261 (369)ln (rm market value) 0019 (0006) 0019 (0006)

Stock market return variablesLagged annual rm stock 0006 (0009) 0003 (0008)

market returnLagged annual SampP 500 re- 20035 (0023) 20029 (0023)

turn

Regression diagnosticsTotal sample size 5179 5179Firms 427 427Adjusted R2 0313 0311

Dependent variable share of compensation paid in options Standard errors in parentheses Bothregressions in this table include rm xed effects

(a) Log difference in payoff is the log difference in combined rm and CEO after-tax share betweenpaying compensation in options and paying salary

(b) This variable is the dollar change in CEO wealth per $1000 change in rm market value (seeJensen and Murphy 1990)

(c) This variable in the dollar charge in CEO wealth per 10-percent change in rm value (see Hall andLiebman 1998)

(d) Standard deviation of rm market value in trillions

28 Hall amp Liebman

per $1000 change in rm value (Jensen and Murphy 1990) and thedollar change in CEO wealth per 10-percent change in rm value (Halland Liebman 1998) As discussed in Baker and Hall (1998) these twowealth sensitivity measures reect two different concepts of how closelyaligned the managersrsquo incentives are with the interests of the sharehold-ers In the regression they are meant to measure how far the CEOrsquosexisting contract and ownership of stock and stock options is from theoptimal level The negative coefcients on the two variables suggeststhat rms that are below their optimal pay-to-performance sensitivityare more likely to give more stock options in the current period

The second column replaces the tax variable motivated by deferraladvantages with the three statutory rates The corporate after-tax sharehas a positive coefcient as we would expect because the tax advantageof stock options increases as the corporate rate falls The coefcient onthe personal tax rate is small and statistically insignicant The sign ofthe coefcient on the capital gains rate is the opposite of that predictedby tax-avoidance theory

Note that these results rely on time-series variation alone This is not avery convincing test Since the log difference in payoffs mostly rises overthe time period covered in our sample as does the share of compensa-tion paid in options it is possible that the tax variable is simply reectingother factors that were changing over this time period that we have notincluded in our regression Therefore it is important to look at specica-tions that rely on cross-sectional variation in the tax variables

Table 4 contains specications that rely on cross-sectional variation incorporate tax rates The rst column repeats the specication from col-umn (1) of Table 3 but uses the trichotomous measures of rm corporatetax rates in place of the statutory rate in calculating each rmrsquos logdifference in payoff from paying compensation in options rather than insalary and bonus As in the previous table we nd the positive coef-cient on the log difference in payoff just as the theory suggest How-ever the coefcient estimate is now much smaller suggesting that aone-percent increase in the log difference in payoff produces a 01-percentage-point increase in the share of compensation paid in options

Introducing cross-sectional variation in the dependent variable doesnot eliminate the potentially spurious correlation over time between thetax variable and the share of compensation paid in options The secondcolumn of Table 4 isolates the pooled cross-sectional variation in corpo-rate tax rates by adding time dummies and dropping the rm xedeffects In this specication the coefcient on the tax variable is indistin-guishable from zero The coefcient on the institutional variable contin-ues to be large and statistically signicant The third column includes

Taxation of Executive Compensation 29

both rm xed effects and time dummies which tests for a relationshipbetween within-rm changes in the dependent variable and within-rmchanges in each explanatory variable In this specication the coefcienton the tax variable is again indistinguishable from zero The coefcientson the corporate governance variables are also statistically insignicant

The third column contains a specication that treats the Tax ReformAct of 1986 (TRA 86) as a natural experiment For rms and CEOs facingthe statutory corporate tax rate TRA 86 increased the relative payoff topaying compensation in options by 17 percent For a rm facing a zeromarginal tax rate it reduced the relative payoff by 45 percent The reduc-tion occurred because TRA 86 reduced the personal tax rate makingsalary and bonus more attractive while the decreased corporate rate hadno effect on zero-tax-rate rms Thus we would expect rms facing zeroor low marginal tax rates to reduce their use of options after TRA 86while rms facing the statutory rate would be expected to increase themFew rms face zero marginal tax rates year after year Therefore we tryto distinguish between rms that often face low tax rates and rms thatnearly always face the statutory rate We dene low-tax rms as oneswhose average marginal tax rate over 1984 to 1986 was below 02412

Then we ran a difference-in-differences regression to see if high-taxrms increased their use of options more than low-tax rms The interac-tion between post-TRA 86 and high corporate tax rates is the key vari-able The estimated coefcient on this variable is positive but small andnot statistically different from zero at the 95-percent level

It is important to note that in this specication as well the main corpo-rate governance variable the share of the rmrsquos shares owned by institu-tional investors performs as theory would predict The variable appearsto be quite robust Table 5 presents regressions of the share of compensa-tion paid in stock options on the fraction of the rmrsquos shares owned byinstitutional investors Since we have already established that these vari-ables have a strong longitudinal relationship the regressions here aredesigned to focus on the cross-sectional relationship between these twovariables The rst and third columns present regressions using data onlyfrom 1994 and nd a coefcient of roughly 02 (column 1 contains noother covariates while column 3 includes all of the other nontax covari-ates from the regressions in Tables 3 and 4) Since the median share of armrsquos stock owned by institutional investors in our sample rises from 19percent in 1980 to 49 percent in 1994 this coefcient would imply that this

12 This cutoff was chosen after inspecting the distribution of tax rates and essentiallyseparates rms that are always at or near the statutory rate from ones that are sometimesbelow it

TA

BL

E4

Com

pens

atio

nR

egre

ssio

nsth

atR

ely

onC

ross

-Sec

tion

alV

aria

tion

inC

orpo

rate

Tax

Rat

es

(1)

(2)

(3)

(4)

Tax

vari

able

s(a)

ln

(dif

fere

nce

inp

ayof

f)0

099

(00

41)

20

038

(00

36)

20

005

(00

44)

Pos

t-T

RA

862

000

5(0

013

)H

igh

corp

orat

eta

xra

te0

018

(00

17)

(Pos

t-T

RA

86)

3hi

gh0

024

(00

16)

corp

orat

eta

xra

te

Cor

por

ate

gove

rnan

ceva

riab

les

Frac

tion

of

rmrsquos

shar

es0

109

(00

27)

011

4(0

015

)2

000

9(0

030

)0

087

(00

34)

owne

dby

inst

itu

tion

alin

vest

ors

Size

ofth

e

rmrsquos

boar

dof

20

004

(00

01)

20

001

(00

01)

20

002

(00

01)

000

1(0

002

)d

irec

tors

Frac

tion

ofd

irec

tors

wh

o2

012

4(0

033

)2

000

3(0

018

)2

004

1(0

034

)0

001

(00

39)

are

insi

de

orgr

ay

Con

trac

t-th

eory

vari

able

sD

(CE

Ow

ealt

h)p

er$1

000

20

238

(00

94)

20

167

(00

53)

20

257

(00

96)

20

021

(01

47)

inFV

(b)

D(C

EO

wea

lth)

per

10

in2

560

(17

8)2

592

(10

5)2

594

(19

6)2

147

(04

86)

FV(c

)

Vol

atili

ty(d

)2

48(4

24)

162

(03

4)1

12(4

28)

106

(06

09)

ln(

rmm

arke

tva

lue)

003

3(

006)

000

8(0

003

)0

009

(00

06)

001

2(0

007

)

Stoc

km

arke

tre

turn

vari

able

sL

agge

dan

nual

rm

stoc

k2

000

2(0

009

)0

011

(00

10)

000

6(0

009

)2

001

2(0

013

)m

arke

tre

turn

Lag

ged

annu

alSamp

P50

02

005

6(0

023

)2

005

3(0

051

)2

003

9(0

049

)2

015

4(0

040

)re

turn

Reg

ress

ion

dia

gnos

tics

T

otal

sam

ple

size

5179

5179

5179

2534

Firm

s42

742

739

6A

dju

sted

R2

031

009

032

005

Firm

xed

effe

cts

Yes

No

Yes

No

Tim

ed

um

mie

sN

oY

esY

esN

o

Dep

end

ent

vari

able

isfr

acti

onof

ann

ualc

omp

ensa

tion

from

stoc

kop

tion

sSt

and

ard

erro

rsin

par

enth

eses

(a)

All

thre

esp

eci

cati

ons

use

tric

hoto

mou

sco

rpor

ate

tax

rate

sT

he

sam

ple

inco

lum

n(3

)u

ses

only

the

year

s19

84to

1988

sin

cew

ear

eu

sing

1986

asa

nat

ural

exp

erim

ent

(b)

Thi

sva

riab

leis

the

dol

lar

chan

gein

CE

Ow

ealt

hp

er$1

000

chan

gein

rm

mar

ketv

alu

e(s

eeJe

nse

nan

dM

urp

hy

1990

)

(c)

Thi

sva

riab

lein

the

dol

lar

char

gein

CE

Ow

ealt

hp

er10

-per

cent

chan

gein

rm

valu

e(s

eeH

alla

nd

Lie

bman

199

8)

(d)

Stan

dar

dd

evia

tion

ofr

mm

arke

tva

lue

intr

illio

ns

32 Hall amp Liebman

TABLE 5Large Institutional Investors and Stock-Option Compensation

The Cross-sectional Relationship between Institutional Investors andOption Grants

(1) (2) (3) (4)

Coefcient on fraction of 217 (067) 145 (014) 177 (069) 120 (015)rmrsquos shares owned byinstitutional investors

Sample period 1994 1980ndash1994 1994 1980ndash1994Contract-theory and stock No No Yes Yes

market covariatesTime dummies No Yes No Yes

The dependent variable is the share of compensation paid in options

factor alone can explain 6 percentage points of the 23-percentage-pointincrease in the median share of compensation paid in options that oc-curred over this time period Of course this variable is only one measureof the strengthening of corporate governance so the total impact of im-proved corporate governance could be greater The second and fourthcolumns of the table use the entire 1980ndash1994 sample and include timedummies to isolate the cross-sectional correlations Again there is evi-dence of a strong cross-sectional relationship between the two variablesalthough the coefcients are slightly smaller perhaps reecting a weakercorporate-governance role in the earlier period In sum there is a strongrelationship in both longitudinal and cross-sectional data between theinstitutional-investor variable and the options-percentage variable13

7 THE MILLION-DOLLAR RULE

In 1993 section 162(m) of the Internal Revenue Code was enactedlimiting the deductibility of executive compensation in excess of onemillion dollars unless the compensation was performance-related Pro-ponents of this legislation argued that it would slow the growth ofexecutive pay and tighten the link between rm performance and execu-tive compensation

The new provision became effective for tax year 1994 and applies tothe CEO and the other four most highly compensated executives in eachcompany Pay that is performance-related such as stock-option grants

13 The one specication in which the fraction owned by institutional investors is notsignicant is in column (2) of Table 4 a regression that includes both xed effects and timeeffects

Taxation of Executive Compensation 33

and bonuses paid for meeting clear performance goals is not affected bythis regulation provided that it has been approved by shareholdersThus the main impact of the provision is to limit the deductibility ofsalary in excess of a million dollars Woodlock and Antenucci (1997)studied the proxy statements of 376 rms and documented that mostrms responded to the new law by qualifying top executive pay asperformance-related Perry and Zenner (1999) identied 25 rms thatreduced salaries from above $1 million to below $1 million and foundthat 23 of the rms cited section 162(m) as their reason for doing so

In this section of the paper we investigate whether this provision hasaffected the level or structure of executive pay using the Execucompdata base which contains information on the highest-paid executives ineach of the rms in the SampP 500 SampP Midcap 400 and SampP SmallCap600 Between 1993 and 1998 the median salary of the Execucomp execu-tives rose from $230000 to $306000 The median value of total compen-sation (salary bonus and option grants) rose from $470000 to $882000Limiting the sample of CEOs indicates that the total compensation of themedian CEO rose from $11 million to $18 million over the period whilethe median salary rose from $441000 to $550000 Thus executive com-pensation clearly continued to rise at a rapid rate after the implementa-tion of the million-dollar rule though perhaps at a slower rate than itotherwise would have

If the regulation affects behavior it should have the largest impact onrms whose executives are receiving salaries that are above the limit orjust below it Thus we might expect to see rms whose executives arereceiving relatively high salaries to rely more heavily on performance-based pay and less heavily on salary in the years following the regulationPerry and Zenner (1999) and Livingston (1999) both nd that this is thecase This trend can be clearly identied in the Execucomp data set Forexample over 1993 to 1998 the median annualized growth rate of salaryfor executives whose 1993 salary was below $500000 in the previous yearwas 85 percent For executives with salary between $500000 and$700000 it was 64 percent for executives between $700000 and $850000it was 50 percent for executives between $750000 and $1 million it was31 percent and for executives above $1 million it was 00 percent14

The observation that executives at lower levels of salary had highergrowth rates is not very strong evidence of an impact of section 162(m)however because it is possible that low-salary executives would havehad higher growth rates even in the absence of the provision To explore

14 Of the 3137 executives for whom we have complete data between 1993 and 1998 2708had 1993 salaries below $500000 while only 35 had 1993 salaries above $1 million

34 Hall amp Liebman

whether this is the case we use Execucomp data from both before andafter the law change Ideally we would want to have data from manyyears before the change to examine whether there is an underlying pat-tern of faster salary growth for low-salary executives Unfortunately theExecucomp data begin in 1992 and therefore provide us with only the1992ndash1993 growth rate in compensation to use as a baseline15

We run regressions pooling data from 1992ndash1993 through 1997ndash1998of the annual percentage change in different forms of compensation on avariable million that is designed to measure the likelihood that anexecutiversquos compensation will be affected by section 162(m) We denemillion as the minimum of 1 and the executiversquos salary divided by $1million in the previous year Thus any executive with salary at or above$1 million would receive a value of 1 while an executive with a salary of$500000 would be coded as a 0516 We include this variable by itself andalso interacted with a dummy variable (after) that equals 1 for years1994 and after Thus the million variable is designed to pick up anyunderlying relationship between the level of salary and the growth rateof salary while the interacted one will identify any differential growthrate of salary by income in the period after section 162(m) took effect Wealso include a full set of year dummies the natural logarithm of thermrsquos market value and the rmrsquos annual rate of return over the twoprevious years as control variables To reduce the sensitivity of our re-sults to outliers we run quantile regressions and robust regressions

Table 6 contains our results The rst four columns are for all topexecutives while the last four columns focus on CEOs Columns (1) and(2) present estimates from robust regression for the percentage change insalary and the percentage change in total compensation (salary bonusstock grants and stock-option grants) In the salary growth regressionsfor all executives the coefcient on million is negative and signicantindicating that there is an underlying relationship between the level ofexecutive compensation and its subsequent growth rate The coefcienton million 3 after is negative as well suggesting that the negativerelationship between the level of salary and its subsequent growth rateintensied slightly after 1994 The relatively small magnitude suggeststhat an executive with a $1 million dollar salary would see his salarygrow at an annual rate that would be about 06 percent slower than anexecutive earning $500000

15 The HallndashLiebman data set contains only a combined salary and bonus variable makingit impossible to use it to analyze this issue16 We also ran the regressions (not reported) allowing for a nonlinear relationship betweenmillion and compensation growth and obtained substantively similar results

TA

BL

E6

The

Eff

ect

ofth

eM

illi

on-D

olla

rR

ule

onSa

lary

and

Tot

alC

ompe

nsat

ion

All

top

exec

uti

ves

CE

Os

only

Rob

ust

regr

essi

onQ

uan

tile

regr

essi

onR

obu

stre

gres

sion

Qu

anti

lere

gres

sion

D

in

Din

D

in

Din

D

in

Din

D

in

Din

sala

ryto

talc

omp

(a)

sala

ryto

talc

omp

sa

lary

tota

lcom

p

sala

ry

tota

lcom

p

mil

lio

n(b

)2

006

02

035

72

006

32

024

12

008

72

023

12

010

32

024

5(0

005

)(0

032

)(0

006

)(0

027

)(0

010

)(0

073

)(0

011

)(0

056

)m

illi

on

32

001

10

026

20

021

20

015

20

016

014

92

003

40

158

aft

er(c

)(0

005

)(0

033

)(0

006

)(0

029

)(0

010

)(0

076

)(0

011

)(0

057

)N

359

5035

950

359

5035

950

757

17

589

757

17

589

Stan

dar

der

rors

inp

aren

thes

esR

egre

ssio

nsal

soin

clu

de

year

du

mm

ies

stoc

km

arke

tre

turn

and

log

mar

ket

valu

e

(a)

Tot

alco

mp

ensa

tion

incl

ud

essa

lary

bon

us

rest

rict

edst

ock

and

stoc

k-op

tion

gran

ts

(b)

mil

lion

isd

ene

das

min

(1s

alar

yin

pre

viou

sye

ars

div

ided

by$1

mil

lion)

(c)

aft

eris

anin

dic

ator

vari

able

equ

alto

zero

befo

reth

ere

gula

tion

took

plac

e(1

993)

and

one

afte

rwar

d

36 Hall amp Liebman

In the second column the coefcient on million 3 after is essentiallyzero suggesting that any decrease in salary brought about by section162(m) was offset by increases in bonus and stock-option grants Theresults are similar for the median regressions There is some evidence ofa very small slowdown in growth of salary for executives likely to beaffected by section 162(m) but no sign of a change in total compensa-tion Taken together this provides evidence of a very small substitutionof performance-related pay for salary

The last four columns run identical tests but limit the sample to CEOsThe interaction of million 3 after continues to be small and negative inthe salary regressions In contrast the impact of 162(m) on total compen-sation appears to be positive and fairly substantial (about 15 percentagepoints per year) in these regressions That is the increase in bonus andstock options following the million-dollar rule more than offset the verysmall decline in salary

All of these conclusions should be interpreted somewhat cautiouslybecause our data provide us with only one pre-1993 control year Never-theless the data are consistent with section 162(m) having led to a veryminor slowdown in salary growth and one that is dramatically smallerthan what one would conclude without controlling for underlying differ-ential salary growth rates at different levels of salary Since total compen-sation for CEOs did not decline and perhaps increased there is evidenceof a minor substitution of performance-related pay for salary in responseto the regulation

8 EXECUTIVE COMPENSATION AND THEELASTICITY OF TAXABLE INCOME

Even if taxes have only a minor effect on executive compensation thestructure of executive compensation is important for tax policy be-cause high-income taxpayers are sources of signicant revenue and arethe focus of important debates about the efciency cost of taxation

Goolsbee (1999) has recently argued that executive decisions to exer-cise options are highly responsive to intertemporal differences in taxrates created by tax reforms He claims that much of the apparent declinein taxable income for high-income taxpayers between 1992 and 1993documented by Feldstein and Feenberg (1996) can be explained by execu-tives shifting option gains into 1992 in order to avoid the higher marginaltax rates that went into effect in 1993 By extension he implies that thehigh elasticities of taxable income with respect to personal tax ratesestimated by comparing the pre- and post-tax-reform taxable income of

Taxation of Executive Compensation 37

high-income individuals reect intertemporal shifting of income and notpermanent effects of taxation If this is true then the deadweight loss oftaxation is lower as is the cost of progressivity

While Goolsbee presents comprehensive evidence that option gainswere unusually high in 1992 his identication of the impact of tax rates onthe timing of option gains depends on correctly specifying a counter-factual level of options that would have been exercised in the absence ofthe tax change He simply assumes a linear time trend The purpose ofGoolsbeersquos time trend is presumably to account for the increasing use ofoptions over time However since options typically require a vestingperiod of at least a couple of years a much more direct measure of ex-pected option gains can be constructed based on the total value of vestedoptions held by the CEO Since options typically have vesting periods thatexceed two years this total can be treated as exogenous since it will not beaffected by the tax changes

More generally the value of options held by the CEO is likely to havea major effect on his exercise decisions If his options are out of themoney he will not have gains to realize If there has been a recent run-up in the market an optimizing CEO may exercise an unusually largeamount of options in order to diversify his overall portfolio and if in-the-money options are about to expire they will be exercised in thecurrent year

Our evidence suggests that much of the apparent tax shifting caninstead be attributed to stock market performance and the timing of pastoption grants Using our 15-year panel we replicate Goolsbeersquos (1999)result for the 1993 period but show that the impact of taxes is notpresent in other time periods Moreover even the result for the 1993period disappears when appropriate controls for past option grants andstock price appreciation are included

The top three rows of Table 7 present data on the taxable incomes ofCEOs in the HallndashLiebman sample For these results our panel has beennarrowed to include only CEOs whose rmrsquos scal years correspondwith tax years In addition we create a balanced panel for each of thethree tax-reform periods by excluding any rm for which there are notcomplete data for the period surrounding the reform

The fourth row of the table shows that the 1993 tax act raised the topmarginal personal income tax rate from 310 to 396 percent effective fortax year 1993 with a further increase occurring in 1994 due to the uncap-ping of the Medicare payroll tax The top row of the table shows thattaxable income was higher in 1992 than for any other year in our sample(all data are in 1994 dollars) Option gains were particularly high in1992mdashmore than double their 1991 level and 68 percent higher than their

TA

BL

E7

Tax

Ref

orm

Sto

ckM

arke

tP

erfo

rman

cea

ndth

eT

imin

gof

Opt

ion

Gai

ns

Ave

rage

stoc

km

arke

tM

ean

CE

OM

ean

CE

Ore

turn

for

Mea

nva

lue

taxa

ble

sala

ryan

dM

ean

CE

OT

opfe

der

al

rms

inof

opti

ons

(Val

ue

ofop

tion

Sam

ple

inco

me

bonu

sop

tion

gain

sp

erso

nal

sam

ple

for

held

atbe

gga

ins)

(va

lue

Ref

orm

size

Yea

r($

mil

)($

mil

)($

mil

)M

TR

pre

vye

ar(

)of

year

ofop

tion

she

ld)

ER

TA

8127

919

800

874

064

80

163

7025

80

460

354

1981

083

10

672

009

869

245

070

61

3919

820

872

067

90

099

5011

90

686

144

1983

099

00

724

025

250

355

116

62

16

TR

A86

295

1985

119

70

837

024

050

76

116

02

0719

861

448

091

70

345

5042

81

776

194

1987

165

50

959

059

338

519

12

229

266

1988

163

21

058

039

028

07

237

71

6419

891

656

105

20

414

2817

92

345

176

1993

272

1990

209

51

028

061

828

232

288

02

15T

axA

ct19

911

875

102

00

547

312

01

268

82

0319

922

678

107

01

250

3146

64

552

275

1993

232

41

209

074

339

620

64

299

173

1994

196

41

262

037

842

515

45

113

074

Sam

ple

isH

allndash

Lie

bman

rm

sw

ith

Dec

embe

r

scal

year

sth

atha

dco

mp

lete

dat

afo

ron

eof

the

thre

eta

x-re

form

per

iod

sst

ud

ied

Taxation of Executive Compensation 39

1994 level These patterns are consistent with Goolsbeersquos claim that tax-payers responded to the anticipated increase in marginal tax rates byshifting gains into 1992

The evidence from the other two tax reforms is less clear suggestingthat there might be factors other than tax rates determining the timing ofoption gains ERTA 1981 reduced the top marginal tax rate from 70percent in 1980 to 50 percent in 1982 However there is no sign thatoption gains were reduced in 1980 and 1981 and then increased in 198217

There was a big increase in option gains in 1983 however Similarly thehighest year of stock-option gains in the period around 1986 was 1987 astrange year to take gains since taxpayers would have known that thepersonal rate would be 10 percentage points lower if they waited untilthe following year18

What all the peak years of option gains have in common is that theyfollow years of strong stock market performance The fth row of Table 7shows the average stock market return during the previous year for therms in our sample The large value of option gains in 1983 followed a355-percent increase in the stock market The 1992 boom in option gainsfollowed a 466-percent increase in the stock market the more modest1987 peak in option gains followed a 19-percent gain (and a 43-percentgain two years before)

Strong stock market performance raises the value of options availableto exercise The sixth row of the table shows the mean value of optionsheld by CEOs at the beginning of the year All three of the peak years ofoption gains correspond with peak years in terms of the value of optionsavailable for exercise Since stock market appreciation during the yearalso increases the value of options available to exercise later in that yearwe need to allow for current-year stock market appreciation as well

Table 8 contains Goolsbee-style regressions for each of the tax re-forms The rst two columns reprint the results from Goolsbee (1999)The third and fourth columns replicate his results in our sample Inparticular with a specication similar to his we are able to replicate hisnding that the current after-tax share has a positive coefcient and thefollowing yearrsquos after-tax share has a coefcient of roughly negative oneColumns (5) through (8) show that this pattern was unique to the 1993tax reform Indeed in the TRA 86 regressions the coefcient on the

17 As we explained above for executives covered by the maximum tax on earned incomethere was not a decline in tax rates between 1981 and 198218 It is possible that the large stock market decline in October 1997 may have promptedsome people to exercise options earlier than they otherwise would have However anincrease in option exercises typically follows stock market increases not declines

TA

BL

E8

Att

empt

sto

Rep

lica

teth

eT

ax-S

hift

ing

Res

ult

inO

ther

Tim

eP

erio

ds

Goo

lsbe

eG

ools

bee

1993

1993

Ent

ire

Ent

ire

regr

essi

onre

gres

sion

refo

rmre

form

TR

A86

TR

A86

ER

TA

81E

RT

A81

1980

ndash199

419

80ndash1

994

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

ln[1

2ta

x(t)

]1

244

108

20

455

032

62

017

62

017

22

016

72

016

70

026

20

068

(01

34)

(01

27)

(05

37)

(05

42)

(02

10)

(02

09)

(00

90)

(00

91)

(00

67)

(00

65)

ln[1

2ta

x(t1

1)]

20

867

20

764

21

214

20

917

047

80

722

20

103

004

30

176

018

3(0

118

)(0

112

)(0

491

)(0

499

)(0

157

)(0

174

)(0

072

)(0

072

)(0

068

)(0

066

)

Tim

e0

135

007

62

000

52

036

40

041

000

80

094

003

50

056

002

7(0

008

)(0

008

)(0

049

)(0

049

)(0

028

)(0

028

)(0

025

)(0

026

)(0

003

)(0

003

)

ln(m

arke

tva

lue)

066

10

489

023

70

334

039

5(0

015

)(0

058

)(0

039

)(0

042

)(0

017

)

Yea

rs19

91ndash1

995

1991

ndash199

519

90ndash1

994

1990

ndash199

419

85ndash1

989

1985

ndash198

919

80ndash1

983

1980

ndash198

319

80ndash1

994

1980

ndash199

4

Ad

just

edR

20

720

770

610

630

690

760

800

810

630

68

Obs

erva

tion

s13

856

132

371

385

136

41

470

143

31

113

106

23

968

385

9

Dep

end

ent

vari

able

isth

elo

gof

taxa

ble

inco

me

All

regr

essi

ons

con

tain

rm

xe

def

fect

sG

ools

bee

regr

essi

on(2

)al

soco

ntai

nstw

oot

her

vari

able

sno

tsh

own

inth

eta

ble

Stan

dar

der

rors

inp

aren

thes

es

Taxation of Executive Compensation 41

TABLE 9Alternative Explanations for the Timing of Option Exercises

1993 Entirereform TRA 86 ERTA 81 1980ndash1994

(1) (2) (3) (4)

ln[12tax(t)] 0108 0111 20051 0201(0324) (0136) (0048) (0057)

ln[12tax(t11)] 20134 0573 0125 0150(0336) (0156) (0053) (0066)

ln(market value) 0418 0177 0338 0418(0061) (0039) (0046) (0163)

Firmrsquos stock market return (t) 0298 0268 0214 0258(0044) (0045) (0031) (0023)

Firmrsquos stock market return 0162 0206 0031 0128(t21) (0041) (0008) (0029) (0021)

ln(value of options held by 0004 0008 0001 0009CEO at start of year) (0003) (0003) (0002) (0001)

Years 1990ndash1994 1985ndash1989 1980ndash1983 1980ndash1994Adjusted R2 064 071 082 068Observations 1349 1427 1036 3812

The dependent variable is the log of taxable income All regressions contain rm xed effects Stan-dard errors in parentheses

following yearrsquos tax rate is positive and signicant Columns (9) and (10)pool all three reforms and fail to replicate the pattern that would reectintertemporal shifting19

It is possible that the ease with which one can shift the timing ofoption gains is asymmetric It may be harder to shift option gains for-ward in time since some options expire and therefore must be exercisednow In addition options were a much smaller portion of compensationin the early 1980s so perhaps the ERTA 81 results are not surprisingNonetheless these results cast doubt on the proposition that this is thechannel through which spurious estimates of the elasticity of taxableincome have occurred for previous tax reforms and they suggest thatwe should look harder for alternative explanations for the timing ofoption gains

Table 9 attempts to predict options gains using the lagged and currentincrease in a rmrsquos market value and the value of the options held by theCEO at the start of the year The two stock market variables stronglypredict option gains while the value of options held has the correct sign

19 The results also fail to suggest large permanent effects of marginal tax rates on taxableincome

42 Hall amp Liebman

in all of the specications and is signicant in the pooled sample eventhough it is highly correlated with the two stock market return variablesIn contrast the tax-timing explanation fails even in the 1993 period

It is important to realize that our results do not invalidate Goolsbeersquoscritique of claims that aggregate revenue data from 1992 and 1993 provethat elasticities of taxable income are large We have simply given analternative explanation for why taxable incomes were unusually high in1992 However our results do cast doubt on the conclusion that largetiming shifts are ubiquitous and capable of explaining away the morecareful estimates of taxable income elasticities that have been performedusing micro data (eg Feldstein 1995 Auten and Carroll 1995)

One further point is in order It is sometimes argued that the combina-tion of increased use of options and the booming stock market explainswhy federal tax revenues have been so much higher than expected dur-ing the 1990s However while large option gains increase revenue fromthe personal income tax they produce offsetting deductions for cor-porations The net impact on revenue is only the difference in the twomarginal tax rates times the option gain While the gap between thepersonal rate and the effective corporate rate is nontrivial the impact oflarge option exercises on total revenue is substantially smaller than theamount by which the exercises inate personal tax revenue and reducecorporate tax revenue

9 CONCLUSION

We have described the tax rules for executive pay in detail and analyzedhow changes in various tax rates affect the tax advantages of stock optionsrelative to salary and bonus We nd that there is a moderate tax advan-tage to optionsmdashcurrently about $4 per $100 in pretax executive paymdashand that changes in the tax advantage of options over time have had atmost a modest impact on the composition of pay Corporate governancefactors particularly the role of large investors appear to be more impor-tant in explaining the dramatic increase in option pay

In addition our evidence suggests that more direct attempts to use taxpolicy to inuence executive compensation have had little effect Wend that the million-dollar rule led companies to substitute perfor-mance-based pay for salary But our evidence suggests that this substitu-tion was quite modest and there is no evidence that the total level of paywas reduced Overall although the stock-option explosion has dramati-cally increased the link between pay and performance this change isdue almost entirely to nontax factors

Taxation of Executive Compensation 43

REFERENCES

Auerbach Alan J (1996) ldquoThe Dynamic Effects of Tax Law Asymmetriesrdquo Re-view of Economic Studies 53451ndash405

mdashmdashmdash and James Poterba (1987) ldquoTax Loss Carryforwards and Corporate TaxIncentivesrdquo In The Effects of Taxation on Capital Accumulation M Feldstein(ed) Chicago University of Chicago Press

Altshuler Rosanne and Alan Auerbach (1990) ldquoThe Signicance of Tax LawAsymmetries An Empirical Investigationrdquo Quarterly Journal of Economics 10561ndash86

Auten Gerald and Robert Carroll (1995) ldquoThe Effect of Income Taxes on House-hold Behaviorrdquo Ofce of Tax Analysis US Department of the Treasury

Baker George P and Brian J Hall (1998) ldquoCEO Incentives and Firm SizerdquoNBER Working Paper no 6868 December

Core John and Wayne Guay (1999) ldquoThe Use of Equity Grants to ManageOptimal Equity Incentive Levelsrdquo Draft

Feldstein M (1995) ldquoThe Effect of Marginal Tax Rates on Taxable Income APanel Study of the 1986 Tax Reform Actrdquo Journal of Political Economy 103 (no3)551ndash72

mdashmdashmdash and D Feenberg (1996) ldquoThe Effect of Increased Tax Rates on TaxableIncome and Economic Efciency A Preliminary Analysis of the 1993 Tax RateIncreasesrdquo In Tax Policy and the Economy vol 10 James M Poterba (ed)Cambridge MA MIT Press

Gompers Paul A and Andrew Metrick (1998) ldquoInstitutional Investors andEquity Pricesrdquo NBER Working Paper no 6723

Goolsbee Austan (1999) ldquoWhat Happens When You Tax the Rich Evidencefrom Executive Compensationrdquo NBER Working Paper no 6333

Graham John R (1996) ldquoProxies for the Corporate Marginal Tax Raterdquo Journal ofFinancial Economics 42187ndash221

Hall Brian J (1999) ldquoThe Design of Multi-year Stock Option Plansrdquo Journal ofApplied Corporate Finance 12(2)97ndash106

mdashmdashmdash and Jeffrey B Liebman (1998) ldquoAre CEOs Really Paid Like Bureau-cratsrdquo Quarterly Journal of Economics CXII(no 3)653ndash691

Jensen Michael C and William M Meckling (1976) ldquoTheory of the Firm Mana-gerial Behavior Agency Costs and Ownership Structurerdquo Journal of FinancialEconomics III305ndash360

mdashmdashmdash and Kevin J Murphy (1990) ldquoPerformance Pay and Top-ManagementIncentivesrdquo Journal of Political Economy XCVIII225ndash264

Kaplan Steven N (1997) ldquoThe Evolution of US Corporate Governance WeAre All Henry Kravis Nowrdquo Draft

Livingstone Jane R (1999) ldquoExecutive Compensation Contracting and Re-sponses to Increased Tax Costsrdquo Louisiana State University June

Miller Merton H and Myron S Scholes (1982) ldquoExecutive CompensationTaxes and Incentivesrdquo In Financial Economics Essays in Honor of Paul CootnerWilliam F Sharpe and Cathryn M Cootner (eds) Princeton NJ Prentice-Hall

Murphy Kevin J (1999) ldquoExecutive Compensationrdquo forthcoming in OrleyAshenfelter and David Card (eds) Handbook of Labor Economics Vol 3 NorthHolland

Perry Todd and Marc Zenner (1999) ldquoPay for Performance Government Regu-

16 Hall amp Liebman

The payoff to salary and bonus in N years is more complicated sincean assumption must be made about how the executive invests the origi-nal cash compensation For example if the executive invests in an instru-ment with taxable interest (eg bonds) then the N-year payoff from aninvestment of P accumulates at the post-corporate-tax post-personal-taxrate of return and is

P (1 2 Tp) [1 1 r (1 2 Tc)(1 2 Tp)]N (4)

However if an executive invests in (non-dividend-paying) equities theinvestment accumulates at the higher post-corporate-tax rate of returnwith the offsetting disadvantage that the capital gains are taxed at thecapital gains rate in year N So the executive receives

P [1 1 r (1 2 Tc)]N (1 2 Tp) 2 Tcg $P (1 2 Tp) [1 1 r (1 2 Tc)]

N 2P (1 2 Tp) (5)

where Tcg is the capital gains tax rate5Thus putting the equations together if an executive is paid cash P

and invests g in bonds and 1 2 g in equities the N-year payoff (includ-ing the payoff to the corporation from the tax deduction) is

P [1 1 r (1 2 Tc)]N Tc 1 g $(P [1 1 r (1 2 Tc)(1 2 Tp)]

N (1 2 Tp)1(1 2 g) (P [1 1 r (1 2 Tc)]

N (1 2 Tp) 2 Tcg $P (1 2 Tp)[1 1 r (1 2 Tc)]

N 2 P (1 2 Tp)) (6)

The tax advantage of options versus salary and bonus therefore is thedifference between the combined (corporate and executive) payoffs inequation (3) and the combined payoffs in equation (6)

43 The Tax Advantage of OptionsWe now turn to analysis of how changes in various tax rates affectthe tax advantages of options relative to cash compensation We focuson the case in which the cash earned by executives is invested entirelyin equities since equities are tax-favored relative to bonds The impor-tant conclusions of this analysis are not substantively different if we

5 In practice executives hold a combination of equities and bondlike instruments withtaxable interest Because equities are tax-favored in the next section we simplify the analy-sis by assuming that executives hold only equities Alternatively we could have assumedthat executives hold only less risky bonds since bonds more closely match the risk proleof a stream of tax savings None of our empirical results are substantively affected by thissimplication

Taxation of Executive Compensation 17

instead assume that a portion of an executiversquos holdings are invested inbonds

431 The Tax Advantage of Options Changing the Corporate TaxRate Stock options are tax-advantaged relative to cash simply becauseoption payouts are deferred allowing the executive to invest at the pre-personal-tax rate of return (with no capital gains tax at the end) ratherthan investing at the after-tax rate of return or paying the capital gainstax Since the advantage of being able to defer taxes is large when thecorporate tax rate is low the tax advantage of options is larger when thecorporate rate is low

To illustrate this point we dene the tax advantage of options to bethe dollar amount by which total after-tax option payoffs (to the corpora-tion and the executive) exceed the total payoffs from salary as denedby the difference of equation (3) and (6) assuming that P (the paymentto the executive) is equal to $100 and g 5 0 (cash compensation is in-vested in equities) We use a ten-year horizon since most options have aten-year term We then calculate how the tax advantage of optionschanges as the corporate tax rate varies from zero to 100 percent holdingthe personal tax rate and the capital gains rate xed at 40 percent and 20percent respectively (which approximates current rates)

The top panel of Figure 2 shows the tax advantage of options relativeto cashmdashprecisely the NPV of the total tax advantage of options for a$100 payment to the CEOmdashas the corporate tax rate changes As ex-pected the tax advantage of options declines as corporate tax rates in-crease At a zero corporate tax rate the tax advantage of options has anNPV of $7 This value is about $350 at a corporate rate of 40 percentWhen the corporate tax rate is 100 percent the tax advantage of optionscompletely disappears since the benets of deferred compensation fallto zero because the after-tax return on equities (and therefore the dis-count rate) falls to zero The tax advantage of options is the NPV of thetax savings from avoiding the capital gains tax on ten years of apprecia-tion which is why it is relatively modest even at the 40-percent corpo-rate rate

432 The Tax Advantage of Options Changing the Personal TaxRate Using the same assumptions as above (but this time holding thecorporate rate at 40 percent and the capital gains rate at 20 percent andvarying the personal rate) the tax advantage of options declines as thepersonal tax rate rises as shown in the middle panel of Figure 2 The taxadvantage of options is higher at low personal tax rates because theposttax base that gives rise to capital gains is higher and the advantage of

Note Holds constant the personal rate at 40 percent and the capital gains rate at 20 percent

Note Holds constant the corporate rate at 40 percent and the capital gains rate at 20 percent

Note Holds constant the personal rate at 40 percent and the corporate rate at 40 percent

FIGURE 2 Tax Advantage of Options NPV of Tax Advantage of Pay-ing $100 in Options (NQSOs) Rather Than in Cash

Taxation of Executive Compensation 19

avoiding the capital gains taxes is greater The tax advantage of optionstherefore declines linearly as the personal tax rate rises At a 100-percentpersonal tax rate the executive receives nothing in either case and byconstruction the companyrsquos tax deduction benet is the same in NPVterms so the tax advantage of options falls to zero also6

433 The Tax Advantage of Options Changing the Capital Gains TaxRate The tax advantage of options for various capital gains rates isdepicted in the bottom panel of Figure 2 under the same assumptions(this time holding both the personal rate and the corporate rate at 40percent while varying the capital gains rate) The tax advantage of avoid-ing the capital gains tax increases as the capital gains tax rate rises

5 HOW HAS THE TAX ADVANTAGE OF OPTIONSCHANGED OVER TIME

The analysis so far has illustrated how the tax advantage of optionsvaries with the changes in tax rates In this section we show how the taxadvantage of options has changed over time as tax rates have changed

51 Top Marginal Tax Rates over TimeThe evolution on the top marginal tax ratesmdashpersonal corporate andcapital gainsmdashfrom 1980 to 1998 is shown in Figure 3 In all cases thetop rate (the rate for taxpayers with the highest incomes) is shownwhich is not always the highest rate since various anomalies (such asclawbacks of exemptions) sometimes lead to marginal tax rates that arehigher than those for the highest income tax payers

The top personal tax rate was 70 percent in 1980 falling to 50 percentin 1982 and 28 percent in 1988 following the 1986 tax act7 The topmarginal tax rate has since risen to 396 percent but is effectively 425percent since there is a 29-percent Medicare surcharge (paid half by theemployer and half by the employee) that has no upper limit The topcorporate tax rate has had only one signicant change since 1980 fallingfrom 46 percent before the 1986 tax act to 34 percent following the act in1988 The top corporate tax rate was increased to 35 percent in 1993 The

6 This is one place where ignoring bond investments has substantive implications Inparticular in the all-bond case the tax advantage of options is nonmonotonicmdashit rises andthen fallsmdashas personal tax rates rise7 During the early 1980s the maximum tax on earned income limited the marginal tax rateon the earnings of high earners so many executives did not face a decline in personal taxrates between 1981 and 1982 The empirical results in this paper are robust to assumingthat executives faced a marginal tax rate of 50 percent in 1980 and 1981

20 Hall amp Liebman

FIGURE 3 Top Marginal Tax Rates since 1980

top capital gains rate has uctuated between 28 percent (approximately)and 20 percent since 1980 and is currently at 20 percent

52 The Evolution of the Tax Advantage of NQSOsThe tax advantage of options relative to cash dened in the same way asin section 4 from 1980 to 1998 is shown in Figure 4 The calculations aredone using the same assumptions as before (ie N 5 10 years and cashis invested in equities) and the statutory top marginal tax rates shown inFigure 3 are used Due to the fall in corporate and personal tax rates thetax advantage of options increased sharply after the Tax Reform Act of1986 This increase in tax advantage has largely been reversed in the1990s as top personal tax rates have crept back up and the capital gainsrate has been reduced

Two key facts emerge from this analysis First there is currently only amoderate tax advantage to standard nonqualied stock optionsmdashon theorder of $4 per $100 of compensation This is because the tax advantagesto the executive of deferring taxes are largely offset by the tax disad-vantages to the company of not being able to deduct option expensesfrom taxable prots until the executive exercises the options Secondalthough the 1986 tax act substantially increased the tax advantage ofoptions more than half of this increase has been reversed in the 1990s

Taxation of Executive Compensation 21

FIGURE 4 Tax Advantage of Options 1980 to 1998

NPV of tax advantage of $100 in options (NQSOs) rather than cash

53 The Tax (Dis)advantage of ISOsSo far the analysis has focused on the tax advantages of NQSOs sincethey are so much more prevalent than ISOs which account for onlyabout 5 percent of option grants The relative scarcity of ISOs can beexplained both by their tax status and by their per-executive cap of$100000 per year

Under what conditions are ISOs tax-preferred to NQSOs For a trans-fer P to an executive the ISO is always tax-preferred by the executive(since the capital gains rate is lower than the personal rate) while theNQSO is always tax-preferred from the companyrsquos perspective (since anISO is not deductible) The key issue however is the relative advan-tages of the two types of options from a global-contracting perspectiveBecause ISOs are not deductible the company is indifferent betweensetting aside P in the form of NQSOs today and setting aside P(1 2 Tc) inthe form of ISOs (Since the NPV of the tax deduction for NQSOs is PTcthe NPV of the payment P is P 2 PTc or P (1 2 Tc))

Equalizing the posttax cost of NQSOs and ISOs we need only look atthe payoff of the executive to determine the condition under which eachis preferred The payoff of NQSOs in N years is

P [1 1 r (1 2 Tc)]N (1 2 Tp) (7)

22 Hall amp Liebman

The same payoff for ISOs is

P (1 2 Tc) [1 1 r (1 2 Tc)]N (1 2 Tcg) (8)

ISOs therefore are tax-advantaged if

Tp 2 Tcg Tc 2 Tc Tcg (9)

Intuitively ISOs are tax-advantaged only if their advantage (the differ-ence between the personal rate and the capital gains rate) is largeenough to offset their cost (the disadvantage of not deducting at thecorporate rate) Note however that the condition is not a simple com-parison between the corporate rate and the personal rate minus thecapital gains rate (Tp 2 Tcg Tc)

Figure 5 shows how the tax advantage of NQSOs relative to ISOs haschanged since 1980 as tax rates have changed Unlike the very modestchanges (around 3 to 4 percent) in the relative tax advantages betweenoptions and cash shown in Figure 4 the relative tax advantage betweenNQSOs and ISOs has seen enormous swings When the top personalrate was 70 percent in 1980 ISOs were tax-favored by a margin greaterthan 6 percent However by 1982 (following the 1981 tax acts that dra-matically lowered the personal tax rate) ISOs became tax-disadvantaged

FIGURE 5 Tax Advantage of NQSOs Relative to ISOs over Time

NPV of tax advantage of $100 payment to executive NQSOs minus ISOs

Taxation of Executive Compensation 23

by about 5 percent By 1988 (following the 1986 tax act) ISOs became tax-disadvantaged by about 18 percent a dramatic swing The raising ofpersonal rates coupled with a decrease in the capital gains rate has sincereduced the tax disadvantage of ISOs to about $4 per $100 of pretaxcompensation Although hard data on ISOs are hard to come by practi-tioners (mostly compensation consultants) claim that ISOs were morepopular prior to 1982 when they were tax-advantaged Although ISOsare less disadvantageous from a tax standpoint relative to NQSOs thanthey were in the late 1980s they are still disadvantageous so it is notsurprising that they have not made a signicant resurgence in recentyears

6 EXPLAINING THE INCREASE IN OPTION-BASEDCOMPENSATION

In this section we analyze whether changes in tax rates affect the compo-sition of executive compensation and consider the relative importanceof tax factors and corporate governance factors in explaining the in-creasing share of compensation paid in the form of stock options Thenumerous changes in tax rates that have occurred since 1980 provide anopportunity to assess whether executive compensation arrangementsrespond to tax incentives in the way that the tax-avoidance model sug-gests We test this model using a panel data set of CEOs in largepublicly-traded US corporations Our identication of the tax effectsrelies on time-series variation in personal corporate and capital gainstax rates along with cross-sectional variation in corporate tax rates

61 DataWe use a 15-year panel data set of CEOs in the largest publicly-tradedUS corporations which is described in Hall and Liebman (1998) Thedata set covers the years from 1980 through 1994 and combines CEOcompensation information from corporate proxies and 10-K lings withstock price and stock return information from CRSP and accounting datafrom Compustat In addition some compensation data from the 1970swere collected in order to construct measures of the value of stock op-tions held by the CEOs in the rst period of the sample8

The feature that distinguishes our data from most other CEO datasets is that with our panel of yearly proxy data on option grants optiongains and total options held we are able to calculate the total value of

8 We are grateful to David Yermack for providing us with some of the data for 1984 to 1991See Yermack (1995) for a discussion of these data

24 Hall amp Liebman

all stock options held by the CEO at a point of time More importantlysince we have the details about the stock options held (number exer-cise price time to maturity etc) we can precisely calculate the changein the value of a CEOrsquos stock option holdings for a given change in rmvalue

62 Identication IssuesThe numerous tax changes over our sample period and the specic waysin which changes in personal corporate and capital gains rates arepredicted to affect the composition of CEO compensation give us un-usually rich sources of identication However because many of thechanges in tax rates would be expected to affect all of the CEOs in oursample in a similar way we will need to pay particular attention toseparating out the tax effects from underlying time trends and otherfactors that changed over time Controlling for underlying trends is par-ticularly important in this study because we are focusing on an out-comemdashthe use of stock optionsmdashthat has increased rapidly over thepast two decades and that many practitioners believe was importantlyaffected by nontax considerations Thus the tax effects we are trying toexplain are deviations from a rising trend

We take two steps to try to separate out the tax effects from theunderlying trend First we include key nontax factors that could poten-tially explain the increasing reliance on performance-based compensa-tion It has been suggested that the dramatic increase in incentive-basedpay is the result of the remarkably poor shareholder returns during the1970s which spurred the LBO and takeover movements of the 1980sAccording to this story (Kaplan 1997) shareholders became much morepowerful via the rise of institutional investors even as the LBO andtakeover movements waned Because the inuence of institutional inves-tors is thought to be one of the most important mechanisms of strongcorporate governance we use the share of each companyrsquos stock ownedby institutional investors (dened as institutions with more than $100million under management) as an explanatory variable9 The percentageof shares owned by large institutional investors increased from about 20percent to almost 50 percent in 1994 an upward trend that closelymatches the sharp rise in the share of compensation in the form ofoptions Indeed in our sample the annual average (over all of the rmsin our sample) share of stock owned by institutional investors has acorrelation of 09 with the annual average share of compensation paid inoptions

9 This variable is described in Gompers and Metrick (1998)

Taxation of Executive Compensation 25

In addition we include two other variables that proxy for strongercorporate governance The rst is the size of the board Evidence suggeststhat boards with smaller numbers of directors reduce the inuence of theCEO and better represent shareholder interests (Yermack 1995) Boardsize fell by approximately 15 percent between 1980 and 1994 The secondvariable is the fraction of the rmrsquos board members who are inside direc-tors or gray directors (noninsiders who have business dealings with thecompany) Firms with fewer outsiders are less likely to act in the share-holdersrsquo interests and we would predict that they would be less likely tohave performance-based compensation for their CEOs The average per-centage of inside and gray directors has decreased from about 45 percentin 1980 to about 35 percent in 1994

Second we allow for cross-sectional variation in corporate tax ratesCorporate marginal tax rates vary because rms may be eligible for tax-loss carrybacks and carryforwards investment tax credits and the alter-native minimum tax (see Auerbach 1986 Auerbach and Altshuler1990 Auerbach and Poterba 1987) Because these cross-sectional differ-ences imply that different rms should respond differently to a givenchange in tax rates they potentially provide us with a way to isolatethe tax effects Firm-specic corporate marginal tax rates are notori-ously difcult to calculate from publicly-available data We use a tricho-tomous variable that equals zero if the rm has tax-loss carryforwardsand negative earnings in a given year equals 05 times the statutorycorporate rate if the rm has only one of those conditions and equalsthe statutory rate if the rm has neither loss carryforwards nor negativeearnings10

The cross-sectional variation in the corporate tax rate may not be ex-ogenous Firms that perform poorly could face low marginal tax ratesand also be particularly likely or unlikely to provide performance-basedpay Because most of our specications use xed effects the correlationwe must be concerned about is between changes in rm performanceand changes in compensation We deal with this by controlling forlagged rm stock market performance in our regressions This shouldeliminate spurious correlation between tax rates and compensation thatis jointly caused by rm performance

10 Graham (1996) presents evidence that the trichotomous variable performs nearly as wellas his recommended simulated marginal tax rate Plesko (1999) compares marginal taxrates calculated from Compustat data with actual marginal tax rates from rm tax returnsHe nds that simple binary variables are more reliable measures of the marginal tax ratethan is either the Graham or the trichotomous variable We obtain substantively similarresults when we switch from the trichotomous variable to the binary variables recommendby Plesko

26 Hall amp Liebman

63 ResultsThe dependent variable in our regressions is the fraction of a CEOrsquos totalannual compensation (measured as the BlackndashScholes value of stock-option grants plus salary and bonus) that comes from stock-optiongrants We choose this reduced-form specication because our interest isin the composition of CEO compensation However the factors thataffect the composition of compensation might also affect the levels andit will be important to keep this in mind in interpreting our results

Optimal-contracting theory generally makes predictions about the in-centives provided by the total amount of rm stock and stock optionsowned by the CEO We choose to model the ow of option grants ratherthan the ultimate performance sensitivity of the CEOrsquos compensationand stock and stock-option holdings because we have in mind anadjustment-cost model in which it takes time for a CEOrsquos holdings ofstock and stock options to reach the optimal level11 In our regressionswe include two measures of the current performance sensitivity of theCEOrsquos wealth to rm performance as explanatory variables to reect thedistance of the CEOrsquos contract from the optimal contract

Table 3 presents results in which the tax effects are identied solely byvariation in tax rates over time The fraction of annual compensation paidin stock options is regressed on tax variables corporate governance vari-ables contract-theory variables and variables reecting stock market re-turns Both of the regressions in this table include rm xed effects Incolumn (1) the fraction in annual compensation paid in stock options isregressed on the log difference in payoff (the combined payoff to both therm and the CEO) from receiving compensation in stock options ratherthan salary and bonus where the payoff calculations are based on statu-tory marginal tax rates This variable is dened as the log difference be-tween equation (3) and equation (6) in section 4 The coefcient on the taxvariable is large and statistically signicant at the 95-percent level Thepoint estimate implies that a one-percent increase in the payoff differencebetween stock options and salary and bonus results in a 24-percentage-point increase in the share of compensation paid in stock options

The corporate governance variables all have the predicted sign Firmswith a higher fraction of their shares owned by institutional investorsuse more performance-based pay (stock options) Firms with large corpo-rate boards or a large share of inside directors are less likely to use stockoptions

The two contract-theory variables are the dollar change in CEO wealth

11 See Core and Guay (1999) for evidence that rms adjust in this way

TABLE 3Compensation Regressions that Rely on Variation in

Tax Rates over Time

(1) (2)

Tax variablesln (difference in payoff)(a) 2393 (0349)ln (12statutory personal 20015 (0034)

rate)ln (12statutory cap gains 0330 (0171)

rate)ln (12statutory corporate 0399 (0148)

rate)

Corporate governance variablesFraction of rmrsquos shares 0043 (029) 0052 (0028)

owned by institutionalinvestors

Size of the rmrsquos board of 20002 (001) 20003 (0001)directors

Fraction of directors who 20082 (033) 20092 (0030)are inside or gray

Contract-theory variablesD (CEO wealth) per $1000 20261 (0095) 20247 (0087)

in FV(b)

D (CEO wealth) per 10 in 2583 (189) 2554 (201)FV(c)

Volatility(d) 521 (424) 261 (369)ln (rm market value) 0019 (0006) 0019 (0006)

Stock market return variablesLagged annual rm stock 0006 (0009) 0003 (0008)

market returnLagged annual SampP 500 re- 20035 (0023) 20029 (0023)

turn

Regression diagnosticsTotal sample size 5179 5179Firms 427 427Adjusted R2 0313 0311

Dependent variable share of compensation paid in options Standard errors in parentheses Bothregressions in this table include rm xed effects

(a) Log difference in payoff is the log difference in combined rm and CEO after-tax share betweenpaying compensation in options and paying salary

(b) This variable is the dollar change in CEO wealth per $1000 change in rm market value (seeJensen and Murphy 1990)

(c) This variable in the dollar charge in CEO wealth per 10-percent change in rm value (see Hall andLiebman 1998)

(d) Standard deviation of rm market value in trillions

28 Hall amp Liebman

per $1000 change in rm value (Jensen and Murphy 1990) and thedollar change in CEO wealth per 10-percent change in rm value (Halland Liebman 1998) As discussed in Baker and Hall (1998) these twowealth sensitivity measures reect two different concepts of how closelyaligned the managersrsquo incentives are with the interests of the sharehold-ers In the regression they are meant to measure how far the CEOrsquosexisting contract and ownership of stock and stock options is from theoptimal level The negative coefcients on the two variables suggeststhat rms that are below their optimal pay-to-performance sensitivityare more likely to give more stock options in the current period

The second column replaces the tax variable motivated by deferraladvantages with the three statutory rates The corporate after-tax sharehas a positive coefcient as we would expect because the tax advantageof stock options increases as the corporate rate falls The coefcient onthe personal tax rate is small and statistically insignicant The sign ofthe coefcient on the capital gains rate is the opposite of that predictedby tax-avoidance theory

Note that these results rely on time-series variation alone This is not avery convincing test Since the log difference in payoffs mostly rises overthe time period covered in our sample as does the share of compensa-tion paid in options it is possible that the tax variable is simply reectingother factors that were changing over this time period that we have notincluded in our regression Therefore it is important to look at specica-tions that rely on cross-sectional variation in the tax variables

Table 4 contains specications that rely on cross-sectional variation incorporate tax rates The rst column repeats the specication from col-umn (1) of Table 3 but uses the trichotomous measures of rm corporatetax rates in place of the statutory rate in calculating each rmrsquos logdifference in payoff from paying compensation in options rather than insalary and bonus As in the previous table we nd the positive coef-cient on the log difference in payoff just as the theory suggest How-ever the coefcient estimate is now much smaller suggesting that aone-percent increase in the log difference in payoff produces a 01-percentage-point increase in the share of compensation paid in options

Introducing cross-sectional variation in the dependent variable doesnot eliminate the potentially spurious correlation over time between thetax variable and the share of compensation paid in options The secondcolumn of Table 4 isolates the pooled cross-sectional variation in corpo-rate tax rates by adding time dummies and dropping the rm xedeffects In this specication the coefcient on the tax variable is indistin-guishable from zero The coefcient on the institutional variable contin-ues to be large and statistically signicant The third column includes

Taxation of Executive Compensation 29

both rm xed effects and time dummies which tests for a relationshipbetween within-rm changes in the dependent variable and within-rmchanges in each explanatory variable In this specication the coefcienton the tax variable is again indistinguishable from zero The coefcientson the corporate governance variables are also statistically insignicant

The third column contains a specication that treats the Tax ReformAct of 1986 (TRA 86) as a natural experiment For rms and CEOs facingthe statutory corporate tax rate TRA 86 increased the relative payoff topaying compensation in options by 17 percent For a rm facing a zeromarginal tax rate it reduced the relative payoff by 45 percent The reduc-tion occurred because TRA 86 reduced the personal tax rate makingsalary and bonus more attractive while the decreased corporate rate hadno effect on zero-tax-rate rms Thus we would expect rms facing zeroor low marginal tax rates to reduce their use of options after TRA 86while rms facing the statutory rate would be expected to increase themFew rms face zero marginal tax rates year after year Therefore we tryto distinguish between rms that often face low tax rates and rms thatnearly always face the statutory rate We dene low-tax rms as oneswhose average marginal tax rate over 1984 to 1986 was below 02412

Then we ran a difference-in-differences regression to see if high-taxrms increased their use of options more than low-tax rms The interac-tion between post-TRA 86 and high corporate tax rates is the key vari-able The estimated coefcient on this variable is positive but small andnot statistically different from zero at the 95-percent level

It is important to note that in this specication as well the main corpo-rate governance variable the share of the rmrsquos shares owned by institu-tional investors performs as theory would predict The variable appearsto be quite robust Table 5 presents regressions of the share of compensa-tion paid in stock options on the fraction of the rmrsquos shares owned byinstitutional investors Since we have already established that these vari-ables have a strong longitudinal relationship the regressions here aredesigned to focus on the cross-sectional relationship between these twovariables The rst and third columns present regressions using data onlyfrom 1994 and nd a coefcient of roughly 02 (column 1 contains noother covariates while column 3 includes all of the other nontax covari-ates from the regressions in Tables 3 and 4) Since the median share of armrsquos stock owned by institutional investors in our sample rises from 19percent in 1980 to 49 percent in 1994 this coefcient would imply that this

12 This cutoff was chosen after inspecting the distribution of tax rates and essentiallyseparates rms that are always at or near the statutory rate from ones that are sometimesbelow it

TA

BL

E4

Com

pens

atio

nR

egre

ssio

nsth

atR

ely

onC

ross

-Sec

tion

alV

aria

tion

inC

orpo

rate

Tax

Rat

es

(1)

(2)

(3)

(4)

Tax

vari

able

s(a)

ln

(dif

fere

nce

inp

ayof

f)0

099

(00

41)

20

038

(00

36)

20

005

(00

44)

Pos

t-T

RA

862

000

5(0

013

)H

igh

corp

orat

eta

xra

te0

018

(00

17)

(Pos

t-T

RA

86)

3hi

gh0

024

(00

16)

corp

orat

eta

xra

te

Cor

por

ate

gove

rnan

ceva

riab

les

Frac

tion

of

rmrsquos

shar

es0

109

(00

27)

011

4(0

015

)2

000

9(0

030

)0

087

(00

34)

owne

dby

inst

itu

tion

alin

vest

ors

Size

ofth

e

rmrsquos

boar

dof

20

004

(00

01)

20

001

(00

01)

20

002

(00

01)

000

1(0

002

)d

irec

tors

Frac

tion

ofd

irec

tors

wh

o2

012

4(0

033

)2

000

3(0

018

)2

004

1(0

034

)0

001

(00

39)

are

insi

de

orgr

ay

Con

trac

t-th

eory

vari

able

sD

(CE

Ow

ealt

h)p

er$1

000

20

238

(00

94)

20

167

(00

53)

20

257

(00

96)

20

021

(01

47)

inFV

(b)

D(C

EO

wea

lth)

per

10

in2

560

(17

8)2

592

(10

5)2

594

(19

6)2

147

(04

86)

FV(c

)

Vol

atili

ty(d

)2

48(4

24)

162

(03

4)1

12(4

28)

106

(06

09)

ln(

rmm

arke

tva

lue)

003

3(

006)

000

8(0

003

)0

009

(00

06)

001

2(0

007

)

Stoc

km

arke

tre

turn

vari

able

sL

agge

dan

nual

rm

stoc

k2

000

2(0

009

)0

011

(00

10)

000

6(0

009

)2

001

2(0

013

)m

arke

tre

turn

Lag

ged

annu

alSamp

P50

02

005

6(0

023

)2

005

3(0

051

)2

003

9(0

049

)2

015

4(0

040

)re

turn

Reg

ress

ion

dia

gnos

tics

T

otal

sam

ple

size

5179

5179

5179

2534

Firm

s42

742

739

6A

dju

sted

R2

031

009

032

005

Firm

xed

effe

cts

Yes

No

Yes

No

Tim

ed

um

mie

sN

oY

esY

esN

o

Dep

end

ent

vari

able

isfr

acti

onof

ann

ualc

omp

ensa

tion

from

stoc

kop

tion

sSt

and

ard

erro

rsin

par

enth

eses

(a)

All

thre

esp

eci

cati

ons

use

tric

hoto

mou

sco

rpor

ate

tax

rate

sT

he

sam

ple

inco

lum

n(3

)u

ses

only

the

year

s19

84to

1988

sin

cew

ear

eu

sing

1986

asa

nat

ural

exp

erim

ent

(b)

Thi

sva

riab

leis

the

dol

lar

chan

gein

CE

Ow

ealt

hp

er$1

000

chan

gein

rm

mar

ketv

alu

e(s

eeJe

nse

nan

dM

urp

hy

1990

)

(c)

Thi

sva

riab

lein

the

dol

lar

char

gein

CE

Ow

ealt

hp

er10

-per

cent

chan

gein

rm

valu

e(s

eeH

alla

nd

Lie

bman

199

8)

(d)

Stan

dar

dd

evia

tion

ofr

mm

arke

tva

lue

intr

illio

ns

32 Hall amp Liebman

TABLE 5Large Institutional Investors and Stock-Option Compensation

The Cross-sectional Relationship between Institutional Investors andOption Grants

(1) (2) (3) (4)

Coefcient on fraction of 217 (067) 145 (014) 177 (069) 120 (015)rmrsquos shares owned byinstitutional investors

Sample period 1994 1980ndash1994 1994 1980ndash1994Contract-theory and stock No No Yes Yes

market covariatesTime dummies No Yes No Yes

The dependent variable is the share of compensation paid in options

factor alone can explain 6 percentage points of the 23-percentage-pointincrease in the median share of compensation paid in options that oc-curred over this time period Of course this variable is only one measureof the strengthening of corporate governance so the total impact of im-proved corporate governance could be greater The second and fourthcolumns of the table use the entire 1980ndash1994 sample and include timedummies to isolate the cross-sectional correlations Again there is evi-dence of a strong cross-sectional relationship between the two variablesalthough the coefcients are slightly smaller perhaps reecting a weakercorporate-governance role in the earlier period In sum there is a strongrelationship in both longitudinal and cross-sectional data between theinstitutional-investor variable and the options-percentage variable13

7 THE MILLION-DOLLAR RULE

In 1993 section 162(m) of the Internal Revenue Code was enactedlimiting the deductibility of executive compensation in excess of onemillion dollars unless the compensation was performance-related Pro-ponents of this legislation argued that it would slow the growth ofexecutive pay and tighten the link between rm performance and execu-tive compensation

The new provision became effective for tax year 1994 and applies tothe CEO and the other four most highly compensated executives in eachcompany Pay that is performance-related such as stock-option grants

13 The one specication in which the fraction owned by institutional investors is notsignicant is in column (2) of Table 4 a regression that includes both xed effects and timeeffects

Taxation of Executive Compensation 33

and bonuses paid for meeting clear performance goals is not affected bythis regulation provided that it has been approved by shareholdersThus the main impact of the provision is to limit the deductibility ofsalary in excess of a million dollars Woodlock and Antenucci (1997)studied the proxy statements of 376 rms and documented that mostrms responded to the new law by qualifying top executive pay asperformance-related Perry and Zenner (1999) identied 25 rms thatreduced salaries from above $1 million to below $1 million and foundthat 23 of the rms cited section 162(m) as their reason for doing so

In this section of the paper we investigate whether this provision hasaffected the level or structure of executive pay using the Execucompdata base which contains information on the highest-paid executives ineach of the rms in the SampP 500 SampP Midcap 400 and SampP SmallCap600 Between 1993 and 1998 the median salary of the Execucomp execu-tives rose from $230000 to $306000 The median value of total compen-sation (salary bonus and option grants) rose from $470000 to $882000Limiting the sample of CEOs indicates that the total compensation of themedian CEO rose from $11 million to $18 million over the period whilethe median salary rose from $441000 to $550000 Thus executive com-pensation clearly continued to rise at a rapid rate after the implementa-tion of the million-dollar rule though perhaps at a slower rate than itotherwise would have

If the regulation affects behavior it should have the largest impact onrms whose executives are receiving salaries that are above the limit orjust below it Thus we might expect to see rms whose executives arereceiving relatively high salaries to rely more heavily on performance-based pay and less heavily on salary in the years following the regulationPerry and Zenner (1999) and Livingston (1999) both nd that this is thecase This trend can be clearly identied in the Execucomp data set Forexample over 1993 to 1998 the median annualized growth rate of salaryfor executives whose 1993 salary was below $500000 in the previous yearwas 85 percent For executives with salary between $500000 and$700000 it was 64 percent for executives between $700000 and $850000it was 50 percent for executives between $750000 and $1 million it was31 percent and for executives above $1 million it was 00 percent14

The observation that executives at lower levels of salary had highergrowth rates is not very strong evidence of an impact of section 162(m)however because it is possible that low-salary executives would havehad higher growth rates even in the absence of the provision To explore

14 Of the 3137 executives for whom we have complete data between 1993 and 1998 2708had 1993 salaries below $500000 while only 35 had 1993 salaries above $1 million

34 Hall amp Liebman

whether this is the case we use Execucomp data from both before andafter the law change Ideally we would want to have data from manyyears before the change to examine whether there is an underlying pat-tern of faster salary growth for low-salary executives Unfortunately theExecucomp data begin in 1992 and therefore provide us with only the1992ndash1993 growth rate in compensation to use as a baseline15

We run regressions pooling data from 1992ndash1993 through 1997ndash1998of the annual percentage change in different forms of compensation on avariable million that is designed to measure the likelihood that anexecutiversquos compensation will be affected by section 162(m) We denemillion as the minimum of 1 and the executiversquos salary divided by $1million in the previous year Thus any executive with salary at or above$1 million would receive a value of 1 while an executive with a salary of$500000 would be coded as a 0516 We include this variable by itself andalso interacted with a dummy variable (after) that equals 1 for years1994 and after Thus the million variable is designed to pick up anyunderlying relationship between the level of salary and the growth rateof salary while the interacted one will identify any differential growthrate of salary by income in the period after section 162(m) took effect Wealso include a full set of year dummies the natural logarithm of thermrsquos market value and the rmrsquos annual rate of return over the twoprevious years as control variables To reduce the sensitivity of our re-sults to outliers we run quantile regressions and robust regressions

Table 6 contains our results The rst four columns are for all topexecutives while the last four columns focus on CEOs Columns (1) and(2) present estimates from robust regression for the percentage change insalary and the percentage change in total compensation (salary bonusstock grants and stock-option grants) In the salary growth regressionsfor all executives the coefcient on million is negative and signicantindicating that there is an underlying relationship between the level ofexecutive compensation and its subsequent growth rate The coefcienton million 3 after is negative as well suggesting that the negativerelationship between the level of salary and its subsequent growth rateintensied slightly after 1994 The relatively small magnitude suggeststhat an executive with a $1 million dollar salary would see his salarygrow at an annual rate that would be about 06 percent slower than anexecutive earning $500000

15 The HallndashLiebman data set contains only a combined salary and bonus variable makingit impossible to use it to analyze this issue16 We also ran the regressions (not reported) allowing for a nonlinear relationship betweenmillion and compensation growth and obtained substantively similar results

TA

BL

E6

The

Eff

ect

ofth

eM

illi

on-D

olla

rR

ule

onSa

lary

and

Tot

alC

ompe

nsat

ion

All

top

exec

uti

ves

CE

Os

only

Rob

ust

regr

essi

onQ

uan

tile

regr

essi

onR

obu

stre

gres

sion

Qu

anti

lere

gres

sion

D

in

Din

D

in

Din

D

in

Din

D

in

Din

sala

ryto

talc

omp

(a)

sala

ryto

talc

omp

sa

lary

tota

lcom

p

sala

ry

tota

lcom

p

mil

lio

n(b

)2

006

02

035

72

006

32

024

12

008

72

023

12

010

32

024

5(0

005

)(0

032

)(0

006

)(0

027

)(0

010

)(0

073

)(0

011

)(0

056

)m

illi

on

32

001

10

026

20

021

20

015

20

016

014

92

003

40

158

aft

er(c

)(0

005

)(0

033

)(0

006

)(0

029

)(0

010

)(0

076

)(0

011

)(0

057

)N

359

5035

950

359

5035

950

757

17

589

757

17

589

Stan

dar

der

rors

inp

aren

thes

esR

egre

ssio

nsal

soin

clu

de

year

du

mm

ies

stoc

km

arke

tre

turn

and

log

mar

ket

valu

e

(a)

Tot

alco

mp

ensa

tion

incl

ud

essa

lary

bon

us

rest

rict

edst

ock

and

stoc

k-op

tion

gran

ts

(b)

mil

lion

isd

ene

das

min

(1s

alar

yin

pre

viou

sye

ars

div

ided

by$1

mil

lion)

(c)

aft

eris

anin

dic

ator

vari

able

equ

alto

zero

befo

reth

ere

gula

tion

took

plac

e(1

993)

and

one

afte

rwar

d

36 Hall amp Liebman

In the second column the coefcient on million 3 after is essentiallyzero suggesting that any decrease in salary brought about by section162(m) was offset by increases in bonus and stock-option grants Theresults are similar for the median regressions There is some evidence ofa very small slowdown in growth of salary for executives likely to beaffected by section 162(m) but no sign of a change in total compensa-tion Taken together this provides evidence of a very small substitutionof performance-related pay for salary

The last four columns run identical tests but limit the sample to CEOsThe interaction of million 3 after continues to be small and negative inthe salary regressions In contrast the impact of 162(m) on total compen-sation appears to be positive and fairly substantial (about 15 percentagepoints per year) in these regressions That is the increase in bonus andstock options following the million-dollar rule more than offset the verysmall decline in salary

All of these conclusions should be interpreted somewhat cautiouslybecause our data provide us with only one pre-1993 control year Never-theless the data are consistent with section 162(m) having led to a veryminor slowdown in salary growth and one that is dramatically smallerthan what one would conclude without controlling for underlying differ-ential salary growth rates at different levels of salary Since total compen-sation for CEOs did not decline and perhaps increased there is evidenceof a minor substitution of performance-related pay for salary in responseto the regulation

8 EXECUTIVE COMPENSATION AND THEELASTICITY OF TAXABLE INCOME

Even if taxes have only a minor effect on executive compensation thestructure of executive compensation is important for tax policy be-cause high-income taxpayers are sources of signicant revenue and arethe focus of important debates about the efciency cost of taxation

Goolsbee (1999) has recently argued that executive decisions to exer-cise options are highly responsive to intertemporal differences in taxrates created by tax reforms He claims that much of the apparent declinein taxable income for high-income taxpayers between 1992 and 1993documented by Feldstein and Feenberg (1996) can be explained by execu-tives shifting option gains into 1992 in order to avoid the higher marginaltax rates that went into effect in 1993 By extension he implies that thehigh elasticities of taxable income with respect to personal tax ratesestimated by comparing the pre- and post-tax-reform taxable income of

Taxation of Executive Compensation 37

high-income individuals reect intertemporal shifting of income and notpermanent effects of taxation If this is true then the deadweight loss oftaxation is lower as is the cost of progressivity

While Goolsbee presents comprehensive evidence that option gainswere unusually high in 1992 his identication of the impact of tax rates onthe timing of option gains depends on correctly specifying a counter-factual level of options that would have been exercised in the absence ofthe tax change He simply assumes a linear time trend The purpose ofGoolsbeersquos time trend is presumably to account for the increasing use ofoptions over time However since options typically require a vestingperiod of at least a couple of years a much more direct measure of ex-pected option gains can be constructed based on the total value of vestedoptions held by the CEO Since options typically have vesting periods thatexceed two years this total can be treated as exogenous since it will not beaffected by the tax changes

More generally the value of options held by the CEO is likely to havea major effect on his exercise decisions If his options are out of themoney he will not have gains to realize If there has been a recent run-up in the market an optimizing CEO may exercise an unusually largeamount of options in order to diversify his overall portfolio and if in-the-money options are about to expire they will be exercised in thecurrent year

Our evidence suggests that much of the apparent tax shifting caninstead be attributed to stock market performance and the timing of pastoption grants Using our 15-year panel we replicate Goolsbeersquos (1999)result for the 1993 period but show that the impact of taxes is notpresent in other time periods Moreover even the result for the 1993period disappears when appropriate controls for past option grants andstock price appreciation are included

The top three rows of Table 7 present data on the taxable incomes ofCEOs in the HallndashLiebman sample For these results our panel has beennarrowed to include only CEOs whose rmrsquos scal years correspondwith tax years In addition we create a balanced panel for each of thethree tax-reform periods by excluding any rm for which there are notcomplete data for the period surrounding the reform

The fourth row of the table shows that the 1993 tax act raised the topmarginal personal income tax rate from 310 to 396 percent effective fortax year 1993 with a further increase occurring in 1994 due to the uncap-ping of the Medicare payroll tax The top row of the table shows thattaxable income was higher in 1992 than for any other year in our sample(all data are in 1994 dollars) Option gains were particularly high in1992mdashmore than double their 1991 level and 68 percent higher than their

TA

BL

E7

Tax

Ref

orm

Sto

ckM

arke

tP

erfo

rman

cea

ndth

eT

imin

gof

Opt

ion

Gai

ns

Ave

rage

stoc

km

arke

tM

ean

CE

OM

ean

CE

Ore

turn

for

Mea

nva

lue

taxa

ble

sala

ryan

dM

ean

CE

OT

opfe

der

al

rms

inof

opti

ons

(Val

ue

ofop

tion

Sam

ple

inco

me

bonu

sop

tion

gain

sp

erso

nal

sam

ple

for

held

atbe

gga

ins)

(va

lue

Ref

orm

size

Yea

r($

mil

)($

mil

)($

mil

)M

TR

pre

vye

ar(

)of

year

ofop

tion

she

ld)

ER

TA

8127

919

800

874

064

80

163

7025

80

460

354

1981

083

10

672

009

869

245

070

61

3919

820

872

067

90

099

5011

90

686

144

1983

099

00

724

025

250

355

116

62

16

TR

A86

295

1985

119

70

837

024

050

76

116

02

0719

861

448

091

70

345

5042

81

776

194

1987

165

50

959

059

338

519

12

229

266

1988

163

21

058

039

028

07

237

71

6419

891

656

105

20

414

2817

92

345

176

1993

272

1990

209

51

028

061

828

232

288

02

15T

axA

ct19

911

875

102

00

547

312

01

268

82

0319

922

678

107

01

250

3146

64

552

275

1993

232

41

209

074

339

620

64

299

173

1994

196

41

262

037

842

515

45

113

074

Sam

ple

isH

allndash

Lie

bman

rm

sw

ith

Dec

embe

r

scal

year

sth

atha

dco

mp

lete

dat

afo

ron

eof

the

thre

eta

x-re

form

per

iod

sst

ud

ied

Taxation of Executive Compensation 39

1994 level These patterns are consistent with Goolsbeersquos claim that tax-payers responded to the anticipated increase in marginal tax rates byshifting gains into 1992

The evidence from the other two tax reforms is less clear suggestingthat there might be factors other than tax rates determining the timing ofoption gains ERTA 1981 reduced the top marginal tax rate from 70percent in 1980 to 50 percent in 1982 However there is no sign thatoption gains were reduced in 1980 and 1981 and then increased in 198217

There was a big increase in option gains in 1983 however Similarly thehighest year of stock-option gains in the period around 1986 was 1987 astrange year to take gains since taxpayers would have known that thepersonal rate would be 10 percentage points lower if they waited untilthe following year18

What all the peak years of option gains have in common is that theyfollow years of strong stock market performance The fth row of Table 7shows the average stock market return during the previous year for therms in our sample The large value of option gains in 1983 followed a355-percent increase in the stock market The 1992 boom in option gainsfollowed a 466-percent increase in the stock market the more modest1987 peak in option gains followed a 19-percent gain (and a 43-percentgain two years before)

Strong stock market performance raises the value of options availableto exercise The sixth row of the table shows the mean value of optionsheld by CEOs at the beginning of the year All three of the peak years ofoption gains correspond with peak years in terms of the value of optionsavailable for exercise Since stock market appreciation during the yearalso increases the value of options available to exercise later in that yearwe need to allow for current-year stock market appreciation as well

Table 8 contains Goolsbee-style regressions for each of the tax re-forms The rst two columns reprint the results from Goolsbee (1999)The third and fourth columns replicate his results in our sample Inparticular with a specication similar to his we are able to replicate hisnding that the current after-tax share has a positive coefcient and thefollowing yearrsquos after-tax share has a coefcient of roughly negative oneColumns (5) through (8) show that this pattern was unique to the 1993tax reform Indeed in the TRA 86 regressions the coefcient on the

17 As we explained above for executives covered by the maximum tax on earned incomethere was not a decline in tax rates between 1981 and 198218 It is possible that the large stock market decline in October 1997 may have promptedsome people to exercise options earlier than they otherwise would have However anincrease in option exercises typically follows stock market increases not declines

TA

BL

E8

Att

empt

sto

Rep

lica

teth

eT

ax-S

hift

ing

Res

ult

inO

ther

Tim

eP

erio

ds

Goo

lsbe

eG

ools

bee

1993

1993

Ent

ire

Ent

ire

regr

essi

onre

gres

sion

refo

rmre

form

TR

A86

TR

A86

ER

TA

81E

RT

A81

1980

ndash199

419

80ndash1

994

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

ln[1

2ta

x(t)

]1

244

108

20

455

032

62

017

62

017

22

016

72

016

70

026

20

068

(01

34)

(01

27)

(05

37)

(05

42)

(02

10)

(02

09)

(00

90)

(00

91)

(00

67)

(00

65)

ln[1

2ta

x(t1

1)]

20

867

20

764

21

214

20

917

047

80

722

20

103

004

30

176

018

3(0

118

)(0

112

)(0

491

)(0

499

)(0

157

)(0

174

)(0

072

)(0

072

)(0

068

)(0

066

)

Tim

e0

135

007

62

000

52

036

40

041

000

80

094

003

50

056

002

7(0

008

)(0

008

)(0

049

)(0

049

)(0

028

)(0

028

)(0

025

)(0

026

)(0

003

)(0

003

)

ln(m

arke

tva

lue)

066

10

489

023

70

334

039

5(0

015

)(0

058

)(0

039

)(0

042

)(0

017

)

Yea

rs19

91ndash1

995

1991

ndash199

519

90ndash1

994

1990

ndash199

419

85ndash1

989

1985

ndash198

919

80ndash1

983

1980

ndash198

319

80ndash1

994

1980

ndash199

4

Ad

just

edR

20

720

770

610

630

690

760

800

810

630

68

Obs

erva

tion

s13

856

132

371

385

136

41

470

143

31

113

106

23

968

385

9

Dep

end

ent

vari

able

isth

elo

gof

taxa

ble

inco

me

All

regr

essi

ons

con

tain

rm

xe

def

fect

sG

ools

bee

regr

essi

on(2

)al

soco

ntai

nstw

oot

her

vari

able

sno

tsh

own

inth

eta

ble

Stan

dar

der

rors

inp

aren

thes

es

Taxation of Executive Compensation 41

TABLE 9Alternative Explanations for the Timing of Option Exercises

1993 Entirereform TRA 86 ERTA 81 1980ndash1994

(1) (2) (3) (4)

ln[12tax(t)] 0108 0111 20051 0201(0324) (0136) (0048) (0057)

ln[12tax(t11)] 20134 0573 0125 0150(0336) (0156) (0053) (0066)

ln(market value) 0418 0177 0338 0418(0061) (0039) (0046) (0163)

Firmrsquos stock market return (t) 0298 0268 0214 0258(0044) (0045) (0031) (0023)

Firmrsquos stock market return 0162 0206 0031 0128(t21) (0041) (0008) (0029) (0021)

ln(value of options held by 0004 0008 0001 0009CEO at start of year) (0003) (0003) (0002) (0001)

Years 1990ndash1994 1985ndash1989 1980ndash1983 1980ndash1994Adjusted R2 064 071 082 068Observations 1349 1427 1036 3812

The dependent variable is the log of taxable income All regressions contain rm xed effects Stan-dard errors in parentheses

following yearrsquos tax rate is positive and signicant Columns (9) and (10)pool all three reforms and fail to replicate the pattern that would reectintertemporal shifting19

It is possible that the ease with which one can shift the timing ofoption gains is asymmetric It may be harder to shift option gains for-ward in time since some options expire and therefore must be exercisednow In addition options were a much smaller portion of compensationin the early 1980s so perhaps the ERTA 81 results are not surprisingNonetheless these results cast doubt on the proposition that this is thechannel through which spurious estimates of the elasticity of taxableincome have occurred for previous tax reforms and they suggest thatwe should look harder for alternative explanations for the timing ofoption gains

Table 9 attempts to predict options gains using the lagged and currentincrease in a rmrsquos market value and the value of the options held by theCEO at the start of the year The two stock market variables stronglypredict option gains while the value of options held has the correct sign

19 The results also fail to suggest large permanent effects of marginal tax rates on taxableincome

42 Hall amp Liebman

in all of the specications and is signicant in the pooled sample eventhough it is highly correlated with the two stock market return variablesIn contrast the tax-timing explanation fails even in the 1993 period

It is important to realize that our results do not invalidate Goolsbeersquoscritique of claims that aggregate revenue data from 1992 and 1993 provethat elasticities of taxable income are large We have simply given analternative explanation for why taxable incomes were unusually high in1992 However our results do cast doubt on the conclusion that largetiming shifts are ubiquitous and capable of explaining away the morecareful estimates of taxable income elasticities that have been performedusing micro data (eg Feldstein 1995 Auten and Carroll 1995)

One further point is in order It is sometimes argued that the combina-tion of increased use of options and the booming stock market explainswhy federal tax revenues have been so much higher than expected dur-ing the 1990s However while large option gains increase revenue fromthe personal income tax they produce offsetting deductions for cor-porations The net impact on revenue is only the difference in the twomarginal tax rates times the option gain While the gap between thepersonal rate and the effective corporate rate is nontrivial the impact oflarge option exercises on total revenue is substantially smaller than theamount by which the exercises inate personal tax revenue and reducecorporate tax revenue

9 CONCLUSION

We have described the tax rules for executive pay in detail and analyzedhow changes in various tax rates affect the tax advantages of stock optionsrelative to salary and bonus We nd that there is a moderate tax advan-tage to optionsmdashcurrently about $4 per $100 in pretax executive paymdashand that changes in the tax advantage of options over time have had atmost a modest impact on the composition of pay Corporate governancefactors particularly the role of large investors appear to be more impor-tant in explaining the dramatic increase in option pay

In addition our evidence suggests that more direct attempts to use taxpolicy to inuence executive compensation have had little effect Wend that the million-dollar rule led companies to substitute perfor-mance-based pay for salary But our evidence suggests that this substitu-tion was quite modest and there is no evidence that the total level of paywas reduced Overall although the stock-option explosion has dramati-cally increased the link between pay and performance this change isdue almost entirely to nontax factors

Taxation of Executive Compensation 43

REFERENCES

Auerbach Alan J (1996) ldquoThe Dynamic Effects of Tax Law Asymmetriesrdquo Re-view of Economic Studies 53451ndash405

mdashmdashmdash and James Poterba (1987) ldquoTax Loss Carryforwards and Corporate TaxIncentivesrdquo In The Effects of Taxation on Capital Accumulation M Feldstein(ed) Chicago University of Chicago Press

Altshuler Rosanne and Alan Auerbach (1990) ldquoThe Signicance of Tax LawAsymmetries An Empirical Investigationrdquo Quarterly Journal of Economics 10561ndash86

Auten Gerald and Robert Carroll (1995) ldquoThe Effect of Income Taxes on House-hold Behaviorrdquo Ofce of Tax Analysis US Department of the Treasury

Baker George P and Brian J Hall (1998) ldquoCEO Incentives and Firm SizerdquoNBER Working Paper no 6868 December

Core John and Wayne Guay (1999) ldquoThe Use of Equity Grants to ManageOptimal Equity Incentive Levelsrdquo Draft

Feldstein M (1995) ldquoThe Effect of Marginal Tax Rates on Taxable Income APanel Study of the 1986 Tax Reform Actrdquo Journal of Political Economy 103 (no3)551ndash72

mdashmdashmdash and D Feenberg (1996) ldquoThe Effect of Increased Tax Rates on TaxableIncome and Economic Efciency A Preliminary Analysis of the 1993 Tax RateIncreasesrdquo In Tax Policy and the Economy vol 10 James M Poterba (ed)Cambridge MA MIT Press

Gompers Paul A and Andrew Metrick (1998) ldquoInstitutional Investors andEquity Pricesrdquo NBER Working Paper no 6723

Goolsbee Austan (1999) ldquoWhat Happens When You Tax the Rich Evidencefrom Executive Compensationrdquo NBER Working Paper no 6333

Graham John R (1996) ldquoProxies for the Corporate Marginal Tax Raterdquo Journal ofFinancial Economics 42187ndash221

Hall Brian J (1999) ldquoThe Design of Multi-year Stock Option Plansrdquo Journal ofApplied Corporate Finance 12(2)97ndash106

mdashmdashmdash and Jeffrey B Liebman (1998) ldquoAre CEOs Really Paid Like Bureau-cratsrdquo Quarterly Journal of Economics CXII(no 3)653ndash691

Jensen Michael C and William M Meckling (1976) ldquoTheory of the Firm Mana-gerial Behavior Agency Costs and Ownership Structurerdquo Journal of FinancialEconomics III305ndash360

mdashmdashmdash and Kevin J Murphy (1990) ldquoPerformance Pay and Top-ManagementIncentivesrdquo Journal of Political Economy XCVIII225ndash264

Kaplan Steven N (1997) ldquoThe Evolution of US Corporate Governance WeAre All Henry Kravis Nowrdquo Draft

Livingstone Jane R (1999) ldquoExecutive Compensation Contracting and Re-sponses to Increased Tax Costsrdquo Louisiana State University June

Miller Merton H and Myron S Scholes (1982) ldquoExecutive CompensationTaxes and Incentivesrdquo In Financial Economics Essays in Honor of Paul CootnerWilliam F Sharpe and Cathryn M Cootner (eds) Princeton NJ Prentice-Hall

Murphy Kevin J (1999) ldquoExecutive Compensationrdquo forthcoming in OrleyAshenfelter and David Card (eds) Handbook of Labor Economics Vol 3 NorthHolland

Perry Todd and Marc Zenner (1999) ldquoPay for Performance Government Regu-

Taxation of Executive Compensation 17

instead assume that a portion of an executiversquos holdings are invested inbonds

431 The Tax Advantage of Options Changing the Corporate TaxRate Stock options are tax-advantaged relative to cash simply becauseoption payouts are deferred allowing the executive to invest at the pre-personal-tax rate of return (with no capital gains tax at the end) ratherthan investing at the after-tax rate of return or paying the capital gainstax Since the advantage of being able to defer taxes is large when thecorporate tax rate is low the tax advantage of options is larger when thecorporate rate is low

To illustrate this point we dene the tax advantage of options to bethe dollar amount by which total after-tax option payoffs (to the corpora-tion and the executive) exceed the total payoffs from salary as denedby the difference of equation (3) and (6) assuming that P (the paymentto the executive) is equal to $100 and g 5 0 (cash compensation is in-vested in equities) We use a ten-year horizon since most options have aten-year term We then calculate how the tax advantage of optionschanges as the corporate tax rate varies from zero to 100 percent holdingthe personal tax rate and the capital gains rate xed at 40 percent and 20percent respectively (which approximates current rates)

The top panel of Figure 2 shows the tax advantage of options relativeto cashmdashprecisely the NPV of the total tax advantage of options for a$100 payment to the CEOmdashas the corporate tax rate changes As ex-pected the tax advantage of options declines as corporate tax rates in-crease At a zero corporate tax rate the tax advantage of options has anNPV of $7 This value is about $350 at a corporate rate of 40 percentWhen the corporate tax rate is 100 percent the tax advantage of optionscompletely disappears since the benets of deferred compensation fallto zero because the after-tax return on equities (and therefore the dis-count rate) falls to zero The tax advantage of options is the NPV of thetax savings from avoiding the capital gains tax on ten years of apprecia-tion which is why it is relatively modest even at the 40-percent corpo-rate rate

432 The Tax Advantage of Options Changing the Personal TaxRate Using the same assumptions as above (but this time holding thecorporate rate at 40 percent and the capital gains rate at 20 percent andvarying the personal rate) the tax advantage of options declines as thepersonal tax rate rises as shown in the middle panel of Figure 2 The taxadvantage of options is higher at low personal tax rates because theposttax base that gives rise to capital gains is higher and the advantage of

Note Holds constant the personal rate at 40 percent and the capital gains rate at 20 percent

Note Holds constant the corporate rate at 40 percent and the capital gains rate at 20 percent

Note Holds constant the personal rate at 40 percent and the corporate rate at 40 percent

FIGURE 2 Tax Advantage of Options NPV of Tax Advantage of Pay-ing $100 in Options (NQSOs) Rather Than in Cash

Taxation of Executive Compensation 19

avoiding the capital gains taxes is greater The tax advantage of optionstherefore declines linearly as the personal tax rate rises At a 100-percentpersonal tax rate the executive receives nothing in either case and byconstruction the companyrsquos tax deduction benet is the same in NPVterms so the tax advantage of options falls to zero also6

433 The Tax Advantage of Options Changing the Capital Gains TaxRate The tax advantage of options for various capital gains rates isdepicted in the bottom panel of Figure 2 under the same assumptions(this time holding both the personal rate and the corporate rate at 40percent while varying the capital gains rate) The tax advantage of avoid-ing the capital gains tax increases as the capital gains tax rate rises

5 HOW HAS THE TAX ADVANTAGE OF OPTIONSCHANGED OVER TIME

The analysis so far has illustrated how the tax advantage of optionsvaries with the changes in tax rates In this section we show how the taxadvantage of options has changed over time as tax rates have changed

51 Top Marginal Tax Rates over TimeThe evolution on the top marginal tax ratesmdashpersonal corporate andcapital gainsmdashfrom 1980 to 1998 is shown in Figure 3 In all cases thetop rate (the rate for taxpayers with the highest incomes) is shownwhich is not always the highest rate since various anomalies (such asclawbacks of exemptions) sometimes lead to marginal tax rates that arehigher than those for the highest income tax payers

The top personal tax rate was 70 percent in 1980 falling to 50 percentin 1982 and 28 percent in 1988 following the 1986 tax act7 The topmarginal tax rate has since risen to 396 percent but is effectively 425percent since there is a 29-percent Medicare surcharge (paid half by theemployer and half by the employee) that has no upper limit The topcorporate tax rate has had only one signicant change since 1980 fallingfrom 46 percent before the 1986 tax act to 34 percent following the act in1988 The top corporate tax rate was increased to 35 percent in 1993 The

6 This is one place where ignoring bond investments has substantive implications Inparticular in the all-bond case the tax advantage of options is nonmonotonicmdashit rises andthen fallsmdashas personal tax rates rise7 During the early 1980s the maximum tax on earned income limited the marginal tax rateon the earnings of high earners so many executives did not face a decline in personal taxrates between 1981 and 1982 The empirical results in this paper are robust to assumingthat executives faced a marginal tax rate of 50 percent in 1980 and 1981

20 Hall amp Liebman

FIGURE 3 Top Marginal Tax Rates since 1980

top capital gains rate has uctuated between 28 percent (approximately)and 20 percent since 1980 and is currently at 20 percent

52 The Evolution of the Tax Advantage of NQSOsThe tax advantage of options relative to cash dened in the same way asin section 4 from 1980 to 1998 is shown in Figure 4 The calculations aredone using the same assumptions as before (ie N 5 10 years and cashis invested in equities) and the statutory top marginal tax rates shown inFigure 3 are used Due to the fall in corporate and personal tax rates thetax advantage of options increased sharply after the Tax Reform Act of1986 This increase in tax advantage has largely been reversed in the1990s as top personal tax rates have crept back up and the capital gainsrate has been reduced

Two key facts emerge from this analysis First there is currently only amoderate tax advantage to standard nonqualied stock optionsmdashon theorder of $4 per $100 of compensation This is because the tax advantagesto the executive of deferring taxes are largely offset by the tax disad-vantages to the company of not being able to deduct option expensesfrom taxable prots until the executive exercises the options Secondalthough the 1986 tax act substantially increased the tax advantage ofoptions more than half of this increase has been reversed in the 1990s

Taxation of Executive Compensation 21

FIGURE 4 Tax Advantage of Options 1980 to 1998

NPV of tax advantage of $100 in options (NQSOs) rather than cash

53 The Tax (Dis)advantage of ISOsSo far the analysis has focused on the tax advantages of NQSOs sincethey are so much more prevalent than ISOs which account for onlyabout 5 percent of option grants The relative scarcity of ISOs can beexplained both by their tax status and by their per-executive cap of$100000 per year

Under what conditions are ISOs tax-preferred to NQSOs For a trans-fer P to an executive the ISO is always tax-preferred by the executive(since the capital gains rate is lower than the personal rate) while theNQSO is always tax-preferred from the companyrsquos perspective (since anISO is not deductible) The key issue however is the relative advan-tages of the two types of options from a global-contracting perspectiveBecause ISOs are not deductible the company is indifferent betweensetting aside P in the form of NQSOs today and setting aside P(1 2 Tc) inthe form of ISOs (Since the NPV of the tax deduction for NQSOs is PTcthe NPV of the payment P is P 2 PTc or P (1 2 Tc))

Equalizing the posttax cost of NQSOs and ISOs we need only look atthe payoff of the executive to determine the condition under which eachis preferred The payoff of NQSOs in N years is

P [1 1 r (1 2 Tc)]N (1 2 Tp) (7)

22 Hall amp Liebman

The same payoff for ISOs is

P (1 2 Tc) [1 1 r (1 2 Tc)]N (1 2 Tcg) (8)

ISOs therefore are tax-advantaged if

Tp 2 Tcg Tc 2 Tc Tcg (9)

Intuitively ISOs are tax-advantaged only if their advantage (the differ-ence between the personal rate and the capital gains rate) is largeenough to offset their cost (the disadvantage of not deducting at thecorporate rate) Note however that the condition is not a simple com-parison between the corporate rate and the personal rate minus thecapital gains rate (Tp 2 Tcg Tc)

Figure 5 shows how the tax advantage of NQSOs relative to ISOs haschanged since 1980 as tax rates have changed Unlike the very modestchanges (around 3 to 4 percent) in the relative tax advantages betweenoptions and cash shown in Figure 4 the relative tax advantage betweenNQSOs and ISOs has seen enormous swings When the top personalrate was 70 percent in 1980 ISOs were tax-favored by a margin greaterthan 6 percent However by 1982 (following the 1981 tax acts that dra-matically lowered the personal tax rate) ISOs became tax-disadvantaged

FIGURE 5 Tax Advantage of NQSOs Relative to ISOs over Time

NPV of tax advantage of $100 payment to executive NQSOs minus ISOs

Taxation of Executive Compensation 23

by about 5 percent By 1988 (following the 1986 tax act) ISOs became tax-disadvantaged by about 18 percent a dramatic swing The raising ofpersonal rates coupled with a decrease in the capital gains rate has sincereduced the tax disadvantage of ISOs to about $4 per $100 of pretaxcompensation Although hard data on ISOs are hard to come by practi-tioners (mostly compensation consultants) claim that ISOs were morepopular prior to 1982 when they were tax-advantaged Although ISOsare less disadvantageous from a tax standpoint relative to NQSOs thanthey were in the late 1980s they are still disadvantageous so it is notsurprising that they have not made a signicant resurgence in recentyears

6 EXPLAINING THE INCREASE IN OPTION-BASEDCOMPENSATION

In this section we analyze whether changes in tax rates affect the compo-sition of executive compensation and consider the relative importanceof tax factors and corporate governance factors in explaining the in-creasing share of compensation paid in the form of stock options Thenumerous changes in tax rates that have occurred since 1980 provide anopportunity to assess whether executive compensation arrangementsrespond to tax incentives in the way that the tax-avoidance model sug-gests We test this model using a panel data set of CEOs in largepublicly-traded US corporations Our identication of the tax effectsrelies on time-series variation in personal corporate and capital gainstax rates along with cross-sectional variation in corporate tax rates

61 DataWe use a 15-year panel data set of CEOs in the largest publicly-tradedUS corporations which is described in Hall and Liebman (1998) Thedata set covers the years from 1980 through 1994 and combines CEOcompensation information from corporate proxies and 10-K lings withstock price and stock return information from CRSP and accounting datafrom Compustat In addition some compensation data from the 1970swere collected in order to construct measures of the value of stock op-tions held by the CEOs in the rst period of the sample8

The feature that distinguishes our data from most other CEO datasets is that with our panel of yearly proxy data on option grants optiongains and total options held we are able to calculate the total value of

8 We are grateful to David Yermack for providing us with some of the data for 1984 to 1991See Yermack (1995) for a discussion of these data

24 Hall amp Liebman

all stock options held by the CEO at a point of time More importantlysince we have the details about the stock options held (number exer-cise price time to maturity etc) we can precisely calculate the changein the value of a CEOrsquos stock option holdings for a given change in rmvalue

62 Identication IssuesThe numerous tax changes over our sample period and the specic waysin which changes in personal corporate and capital gains rates arepredicted to affect the composition of CEO compensation give us un-usually rich sources of identication However because many of thechanges in tax rates would be expected to affect all of the CEOs in oursample in a similar way we will need to pay particular attention toseparating out the tax effects from underlying time trends and otherfactors that changed over time Controlling for underlying trends is par-ticularly important in this study because we are focusing on an out-comemdashthe use of stock optionsmdashthat has increased rapidly over thepast two decades and that many practitioners believe was importantlyaffected by nontax considerations Thus the tax effects we are trying toexplain are deviations from a rising trend

We take two steps to try to separate out the tax effects from theunderlying trend First we include key nontax factors that could poten-tially explain the increasing reliance on performance-based compensa-tion It has been suggested that the dramatic increase in incentive-basedpay is the result of the remarkably poor shareholder returns during the1970s which spurred the LBO and takeover movements of the 1980sAccording to this story (Kaplan 1997) shareholders became much morepowerful via the rise of institutional investors even as the LBO andtakeover movements waned Because the inuence of institutional inves-tors is thought to be one of the most important mechanisms of strongcorporate governance we use the share of each companyrsquos stock ownedby institutional investors (dened as institutions with more than $100million under management) as an explanatory variable9 The percentageof shares owned by large institutional investors increased from about 20percent to almost 50 percent in 1994 an upward trend that closelymatches the sharp rise in the share of compensation in the form ofoptions Indeed in our sample the annual average (over all of the rmsin our sample) share of stock owned by institutional investors has acorrelation of 09 with the annual average share of compensation paid inoptions

9 This variable is described in Gompers and Metrick (1998)

Taxation of Executive Compensation 25

In addition we include two other variables that proxy for strongercorporate governance The rst is the size of the board Evidence suggeststhat boards with smaller numbers of directors reduce the inuence of theCEO and better represent shareholder interests (Yermack 1995) Boardsize fell by approximately 15 percent between 1980 and 1994 The secondvariable is the fraction of the rmrsquos board members who are inside direc-tors or gray directors (noninsiders who have business dealings with thecompany) Firms with fewer outsiders are less likely to act in the share-holdersrsquo interests and we would predict that they would be less likely tohave performance-based compensation for their CEOs The average per-centage of inside and gray directors has decreased from about 45 percentin 1980 to about 35 percent in 1994

Second we allow for cross-sectional variation in corporate tax ratesCorporate marginal tax rates vary because rms may be eligible for tax-loss carrybacks and carryforwards investment tax credits and the alter-native minimum tax (see Auerbach 1986 Auerbach and Altshuler1990 Auerbach and Poterba 1987) Because these cross-sectional differ-ences imply that different rms should respond differently to a givenchange in tax rates they potentially provide us with a way to isolatethe tax effects Firm-specic corporate marginal tax rates are notori-ously difcult to calculate from publicly-available data We use a tricho-tomous variable that equals zero if the rm has tax-loss carryforwardsand negative earnings in a given year equals 05 times the statutorycorporate rate if the rm has only one of those conditions and equalsthe statutory rate if the rm has neither loss carryforwards nor negativeearnings10

The cross-sectional variation in the corporate tax rate may not be ex-ogenous Firms that perform poorly could face low marginal tax ratesand also be particularly likely or unlikely to provide performance-basedpay Because most of our specications use xed effects the correlationwe must be concerned about is between changes in rm performanceand changes in compensation We deal with this by controlling forlagged rm stock market performance in our regressions This shouldeliminate spurious correlation between tax rates and compensation thatis jointly caused by rm performance

10 Graham (1996) presents evidence that the trichotomous variable performs nearly as wellas his recommended simulated marginal tax rate Plesko (1999) compares marginal taxrates calculated from Compustat data with actual marginal tax rates from rm tax returnsHe nds that simple binary variables are more reliable measures of the marginal tax ratethan is either the Graham or the trichotomous variable We obtain substantively similarresults when we switch from the trichotomous variable to the binary variables recommendby Plesko

26 Hall amp Liebman

63 ResultsThe dependent variable in our regressions is the fraction of a CEOrsquos totalannual compensation (measured as the BlackndashScholes value of stock-option grants plus salary and bonus) that comes from stock-optiongrants We choose this reduced-form specication because our interest isin the composition of CEO compensation However the factors thataffect the composition of compensation might also affect the levels andit will be important to keep this in mind in interpreting our results

Optimal-contracting theory generally makes predictions about the in-centives provided by the total amount of rm stock and stock optionsowned by the CEO We choose to model the ow of option grants ratherthan the ultimate performance sensitivity of the CEOrsquos compensationand stock and stock-option holdings because we have in mind anadjustment-cost model in which it takes time for a CEOrsquos holdings ofstock and stock options to reach the optimal level11 In our regressionswe include two measures of the current performance sensitivity of theCEOrsquos wealth to rm performance as explanatory variables to reect thedistance of the CEOrsquos contract from the optimal contract

Table 3 presents results in which the tax effects are identied solely byvariation in tax rates over time The fraction of annual compensation paidin stock options is regressed on tax variables corporate governance vari-ables contract-theory variables and variables reecting stock market re-turns Both of the regressions in this table include rm xed effects Incolumn (1) the fraction in annual compensation paid in stock options isregressed on the log difference in payoff (the combined payoff to both therm and the CEO) from receiving compensation in stock options ratherthan salary and bonus where the payoff calculations are based on statu-tory marginal tax rates This variable is dened as the log difference be-tween equation (3) and equation (6) in section 4 The coefcient on the taxvariable is large and statistically signicant at the 95-percent level Thepoint estimate implies that a one-percent increase in the payoff differencebetween stock options and salary and bonus results in a 24-percentage-point increase in the share of compensation paid in stock options

The corporate governance variables all have the predicted sign Firmswith a higher fraction of their shares owned by institutional investorsuse more performance-based pay (stock options) Firms with large corpo-rate boards or a large share of inside directors are less likely to use stockoptions

The two contract-theory variables are the dollar change in CEO wealth

11 See Core and Guay (1999) for evidence that rms adjust in this way

TABLE 3Compensation Regressions that Rely on Variation in

Tax Rates over Time

(1) (2)

Tax variablesln (difference in payoff)(a) 2393 (0349)ln (12statutory personal 20015 (0034)

rate)ln (12statutory cap gains 0330 (0171)

rate)ln (12statutory corporate 0399 (0148)

rate)

Corporate governance variablesFraction of rmrsquos shares 0043 (029) 0052 (0028)

owned by institutionalinvestors

Size of the rmrsquos board of 20002 (001) 20003 (0001)directors

Fraction of directors who 20082 (033) 20092 (0030)are inside or gray

Contract-theory variablesD (CEO wealth) per $1000 20261 (0095) 20247 (0087)

in FV(b)

D (CEO wealth) per 10 in 2583 (189) 2554 (201)FV(c)

Volatility(d) 521 (424) 261 (369)ln (rm market value) 0019 (0006) 0019 (0006)

Stock market return variablesLagged annual rm stock 0006 (0009) 0003 (0008)

market returnLagged annual SampP 500 re- 20035 (0023) 20029 (0023)

turn

Regression diagnosticsTotal sample size 5179 5179Firms 427 427Adjusted R2 0313 0311

Dependent variable share of compensation paid in options Standard errors in parentheses Bothregressions in this table include rm xed effects

(a) Log difference in payoff is the log difference in combined rm and CEO after-tax share betweenpaying compensation in options and paying salary

(b) This variable is the dollar change in CEO wealth per $1000 change in rm market value (seeJensen and Murphy 1990)

(c) This variable in the dollar charge in CEO wealth per 10-percent change in rm value (see Hall andLiebman 1998)

(d) Standard deviation of rm market value in trillions

28 Hall amp Liebman

per $1000 change in rm value (Jensen and Murphy 1990) and thedollar change in CEO wealth per 10-percent change in rm value (Halland Liebman 1998) As discussed in Baker and Hall (1998) these twowealth sensitivity measures reect two different concepts of how closelyaligned the managersrsquo incentives are with the interests of the sharehold-ers In the regression they are meant to measure how far the CEOrsquosexisting contract and ownership of stock and stock options is from theoptimal level The negative coefcients on the two variables suggeststhat rms that are below their optimal pay-to-performance sensitivityare more likely to give more stock options in the current period

The second column replaces the tax variable motivated by deferraladvantages with the three statutory rates The corporate after-tax sharehas a positive coefcient as we would expect because the tax advantageof stock options increases as the corporate rate falls The coefcient onthe personal tax rate is small and statistically insignicant The sign ofthe coefcient on the capital gains rate is the opposite of that predictedby tax-avoidance theory

Note that these results rely on time-series variation alone This is not avery convincing test Since the log difference in payoffs mostly rises overthe time period covered in our sample as does the share of compensa-tion paid in options it is possible that the tax variable is simply reectingother factors that were changing over this time period that we have notincluded in our regression Therefore it is important to look at specica-tions that rely on cross-sectional variation in the tax variables

Table 4 contains specications that rely on cross-sectional variation incorporate tax rates The rst column repeats the specication from col-umn (1) of Table 3 but uses the trichotomous measures of rm corporatetax rates in place of the statutory rate in calculating each rmrsquos logdifference in payoff from paying compensation in options rather than insalary and bonus As in the previous table we nd the positive coef-cient on the log difference in payoff just as the theory suggest How-ever the coefcient estimate is now much smaller suggesting that aone-percent increase in the log difference in payoff produces a 01-percentage-point increase in the share of compensation paid in options

Introducing cross-sectional variation in the dependent variable doesnot eliminate the potentially spurious correlation over time between thetax variable and the share of compensation paid in options The secondcolumn of Table 4 isolates the pooled cross-sectional variation in corpo-rate tax rates by adding time dummies and dropping the rm xedeffects In this specication the coefcient on the tax variable is indistin-guishable from zero The coefcient on the institutional variable contin-ues to be large and statistically signicant The third column includes

Taxation of Executive Compensation 29

both rm xed effects and time dummies which tests for a relationshipbetween within-rm changes in the dependent variable and within-rmchanges in each explanatory variable In this specication the coefcienton the tax variable is again indistinguishable from zero The coefcientson the corporate governance variables are also statistically insignicant

The third column contains a specication that treats the Tax ReformAct of 1986 (TRA 86) as a natural experiment For rms and CEOs facingthe statutory corporate tax rate TRA 86 increased the relative payoff topaying compensation in options by 17 percent For a rm facing a zeromarginal tax rate it reduced the relative payoff by 45 percent The reduc-tion occurred because TRA 86 reduced the personal tax rate makingsalary and bonus more attractive while the decreased corporate rate hadno effect on zero-tax-rate rms Thus we would expect rms facing zeroor low marginal tax rates to reduce their use of options after TRA 86while rms facing the statutory rate would be expected to increase themFew rms face zero marginal tax rates year after year Therefore we tryto distinguish between rms that often face low tax rates and rms thatnearly always face the statutory rate We dene low-tax rms as oneswhose average marginal tax rate over 1984 to 1986 was below 02412

Then we ran a difference-in-differences regression to see if high-taxrms increased their use of options more than low-tax rms The interac-tion between post-TRA 86 and high corporate tax rates is the key vari-able The estimated coefcient on this variable is positive but small andnot statistically different from zero at the 95-percent level

It is important to note that in this specication as well the main corpo-rate governance variable the share of the rmrsquos shares owned by institu-tional investors performs as theory would predict The variable appearsto be quite robust Table 5 presents regressions of the share of compensa-tion paid in stock options on the fraction of the rmrsquos shares owned byinstitutional investors Since we have already established that these vari-ables have a strong longitudinal relationship the regressions here aredesigned to focus on the cross-sectional relationship between these twovariables The rst and third columns present regressions using data onlyfrom 1994 and nd a coefcient of roughly 02 (column 1 contains noother covariates while column 3 includes all of the other nontax covari-ates from the regressions in Tables 3 and 4) Since the median share of armrsquos stock owned by institutional investors in our sample rises from 19percent in 1980 to 49 percent in 1994 this coefcient would imply that this

12 This cutoff was chosen after inspecting the distribution of tax rates and essentiallyseparates rms that are always at or near the statutory rate from ones that are sometimesbelow it

TA

BL

E4

Com

pens

atio

nR

egre

ssio

nsth

atR

ely

onC

ross

-Sec

tion

alV

aria

tion

inC

orpo

rate

Tax

Rat

es

(1)

(2)

(3)

(4)

Tax

vari

able

s(a)

ln

(dif

fere

nce

inp

ayof

f)0

099

(00

41)

20

038

(00

36)

20

005

(00

44)

Pos

t-T

RA

862

000

5(0

013

)H

igh

corp

orat

eta

xra

te0

018

(00

17)

(Pos

t-T

RA

86)

3hi

gh0

024

(00

16)

corp

orat

eta

xra

te

Cor

por

ate

gove

rnan

ceva

riab

les

Frac

tion

of

rmrsquos

shar

es0

109

(00

27)

011

4(0

015

)2

000

9(0

030

)0

087

(00

34)

owne

dby

inst

itu

tion

alin

vest

ors

Size

ofth

e

rmrsquos

boar

dof

20

004

(00

01)

20

001

(00

01)

20

002

(00

01)

000

1(0

002

)d

irec

tors

Frac

tion

ofd

irec

tors

wh

o2

012

4(0

033

)2

000

3(0

018

)2

004

1(0

034

)0

001

(00

39)

are

insi

de

orgr

ay

Con

trac

t-th

eory

vari

able

sD

(CE

Ow

ealt

h)p

er$1

000

20

238

(00

94)

20

167

(00

53)

20

257

(00

96)

20

021

(01

47)

inFV

(b)

D(C

EO

wea

lth)

per

10

in2

560

(17

8)2

592

(10

5)2

594

(19

6)2

147

(04

86)

FV(c

)

Vol

atili

ty(d

)2

48(4

24)

162

(03

4)1

12(4

28)

106

(06

09)

ln(

rmm

arke

tva

lue)

003

3(

006)

000

8(0

003

)0

009

(00

06)

001

2(0

007

)

Stoc

km

arke

tre

turn

vari

able

sL

agge

dan

nual

rm

stoc

k2

000

2(0

009

)0

011

(00

10)

000

6(0

009

)2

001

2(0

013

)m

arke

tre

turn

Lag

ged

annu

alSamp

P50

02

005

6(0

023

)2

005

3(0

051

)2

003

9(0

049

)2

015

4(0

040

)re

turn

Reg

ress

ion

dia

gnos

tics

T

otal

sam

ple

size

5179

5179

5179

2534

Firm

s42

742

739

6A

dju

sted

R2

031

009

032

005

Firm

xed

effe

cts

Yes

No

Yes

No

Tim

ed

um

mie

sN

oY

esY

esN

o

Dep

end

ent

vari

able

isfr

acti

onof

ann

ualc

omp

ensa

tion

from

stoc

kop

tion

sSt

and

ard

erro

rsin

par

enth

eses

(a)

All

thre

esp

eci

cati

ons

use

tric

hoto

mou

sco

rpor

ate

tax

rate

sT

he

sam

ple

inco

lum

n(3

)u

ses

only

the

year

s19

84to

1988

sin

cew

ear

eu

sing

1986

asa

nat

ural

exp

erim

ent

(b)

Thi

sva

riab

leis

the

dol

lar

chan

gein

CE

Ow

ealt

hp

er$1

000

chan

gein

rm

mar

ketv

alu

e(s

eeJe

nse

nan

dM

urp

hy

1990

)

(c)

Thi

sva

riab

lein

the

dol

lar

char

gein

CE

Ow

ealt

hp

er10

-per

cent

chan

gein

rm

valu

e(s

eeH

alla

nd

Lie

bman

199

8)

(d)

Stan

dar

dd

evia

tion

ofr

mm

arke

tva

lue

intr

illio

ns

32 Hall amp Liebman

TABLE 5Large Institutional Investors and Stock-Option Compensation

The Cross-sectional Relationship between Institutional Investors andOption Grants

(1) (2) (3) (4)

Coefcient on fraction of 217 (067) 145 (014) 177 (069) 120 (015)rmrsquos shares owned byinstitutional investors

Sample period 1994 1980ndash1994 1994 1980ndash1994Contract-theory and stock No No Yes Yes

market covariatesTime dummies No Yes No Yes

The dependent variable is the share of compensation paid in options

factor alone can explain 6 percentage points of the 23-percentage-pointincrease in the median share of compensation paid in options that oc-curred over this time period Of course this variable is only one measureof the strengthening of corporate governance so the total impact of im-proved corporate governance could be greater The second and fourthcolumns of the table use the entire 1980ndash1994 sample and include timedummies to isolate the cross-sectional correlations Again there is evi-dence of a strong cross-sectional relationship between the two variablesalthough the coefcients are slightly smaller perhaps reecting a weakercorporate-governance role in the earlier period In sum there is a strongrelationship in both longitudinal and cross-sectional data between theinstitutional-investor variable and the options-percentage variable13

7 THE MILLION-DOLLAR RULE

In 1993 section 162(m) of the Internal Revenue Code was enactedlimiting the deductibility of executive compensation in excess of onemillion dollars unless the compensation was performance-related Pro-ponents of this legislation argued that it would slow the growth ofexecutive pay and tighten the link between rm performance and execu-tive compensation

The new provision became effective for tax year 1994 and applies tothe CEO and the other four most highly compensated executives in eachcompany Pay that is performance-related such as stock-option grants

13 The one specication in which the fraction owned by institutional investors is notsignicant is in column (2) of Table 4 a regression that includes both xed effects and timeeffects

Taxation of Executive Compensation 33

and bonuses paid for meeting clear performance goals is not affected bythis regulation provided that it has been approved by shareholdersThus the main impact of the provision is to limit the deductibility ofsalary in excess of a million dollars Woodlock and Antenucci (1997)studied the proxy statements of 376 rms and documented that mostrms responded to the new law by qualifying top executive pay asperformance-related Perry and Zenner (1999) identied 25 rms thatreduced salaries from above $1 million to below $1 million and foundthat 23 of the rms cited section 162(m) as their reason for doing so

In this section of the paper we investigate whether this provision hasaffected the level or structure of executive pay using the Execucompdata base which contains information on the highest-paid executives ineach of the rms in the SampP 500 SampP Midcap 400 and SampP SmallCap600 Between 1993 and 1998 the median salary of the Execucomp execu-tives rose from $230000 to $306000 The median value of total compen-sation (salary bonus and option grants) rose from $470000 to $882000Limiting the sample of CEOs indicates that the total compensation of themedian CEO rose from $11 million to $18 million over the period whilethe median salary rose from $441000 to $550000 Thus executive com-pensation clearly continued to rise at a rapid rate after the implementa-tion of the million-dollar rule though perhaps at a slower rate than itotherwise would have

If the regulation affects behavior it should have the largest impact onrms whose executives are receiving salaries that are above the limit orjust below it Thus we might expect to see rms whose executives arereceiving relatively high salaries to rely more heavily on performance-based pay and less heavily on salary in the years following the regulationPerry and Zenner (1999) and Livingston (1999) both nd that this is thecase This trend can be clearly identied in the Execucomp data set Forexample over 1993 to 1998 the median annualized growth rate of salaryfor executives whose 1993 salary was below $500000 in the previous yearwas 85 percent For executives with salary between $500000 and$700000 it was 64 percent for executives between $700000 and $850000it was 50 percent for executives between $750000 and $1 million it was31 percent and for executives above $1 million it was 00 percent14

The observation that executives at lower levels of salary had highergrowth rates is not very strong evidence of an impact of section 162(m)however because it is possible that low-salary executives would havehad higher growth rates even in the absence of the provision To explore

14 Of the 3137 executives for whom we have complete data between 1993 and 1998 2708had 1993 salaries below $500000 while only 35 had 1993 salaries above $1 million

34 Hall amp Liebman

whether this is the case we use Execucomp data from both before andafter the law change Ideally we would want to have data from manyyears before the change to examine whether there is an underlying pat-tern of faster salary growth for low-salary executives Unfortunately theExecucomp data begin in 1992 and therefore provide us with only the1992ndash1993 growth rate in compensation to use as a baseline15

We run regressions pooling data from 1992ndash1993 through 1997ndash1998of the annual percentage change in different forms of compensation on avariable million that is designed to measure the likelihood that anexecutiversquos compensation will be affected by section 162(m) We denemillion as the minimum of 1 and the executiversquos salary divided by $1million in the previous year Thus any executive with salary at or above$1 million would receive a value of 1 while an executive with a salary of$500000 would be coded as a 0516 We include this variable by itself andalso interacted with a dummy variable (after) that equals 1 for years1994 and after Thus the million variable is designed to pick up anyunderlying relationship between the level of salary and the growth rateof salary while the interacted one will identify any differential growthrate of salary by income in the period after section 162(m) took effect Wealso include a full set of year dummies the natural logarithm of thermrsquos market value and the rmrsquos annual rate of return over the twoprevious years as control variables To reduce the sensitivity of our re-sults to outliers we run quantile regressions and robust regressions

Table 6 contains our results The rst four columns are for all topexecutives while the last four columns focus on CEOs Columns (1) and(2) present estimates from robust regression for the percentage change insalary and the percentage change in total compensation (salary bonusstock grants and stock-option grants) In the salary growth regressionsfor all executives the coefcient on million is negative and signicantindicating that there is an underlying relationship between the level ofexecutive compensation and its subsequent growth rate The coefcienton million 3 after is negative as well suggesting that the negativerelationship between the level of salary and its subsequent growth rateintensied slightly after 1994 The relatively small magnitude suggeststhat an executive with a $1 million dollar salary would see his salarygrow at an annual rate that would be about 06 percent slower than anexecutive earning $500000

15 The HallndashLiebman data set contains only a combined salary and bonus variable makingit impossible to use it to analyze this issue16 We also ran the regressions (not reported) allowing for a nonlinear relationship betweenmillion and compensation growth and obtained substantively similar results

TA

BL

E6

The

Eff

ect

ofth

eM

illi

on-D

olla

rR

ule

onSa

lary

and

Tot

alC

ompe

nsat

ion

All

top

exec

uti

ves

CE

Os

only

Rob

ust

regr

essi

onQ

uan

tile

regr

essi

onR

obu

stre

gres

sion

Qu

anti

lere

gres

sion

D

in

Din

D

in

Din

D

in

Din

D

in

Din

sala

ryto

talc

omp

(a)

sala

ryto

talc

omp

sa

lary

tota

lcom

p

sala

ry

tota

lcom

p

mil

lio

n(b

)2

006

02

035

72

006

32

024

12

008

72

023

12

010

32

024

5(0

005

)(0

032

)(0

006

)(0

027

)(0

010

)(0

073

)(0

011

)(0

056

)m

illi

on

32

001

10

026

20

021

20

015

20

016

014

92

003

40

158

aft

er(c

)(0

005

)(0

033

)(0

006

)(0

029

)(0

010

)(0

076

)(0

011

)(0

057

)N

359

5035

950

359

5035

950

757

17

589

757

17

589

Stan

dar

der

rors

inp

aren

thes

esR

egre

ssio

nsal

soin

clu

de

year

du

mm

ies

stoc

km

arke

tre

turn

and

log

mar

ket

valu

e

(a)

Tot

alco

mp

ensa

tion

incl

ud

essa

lary

bon

us

rest

rict

edst

ock

and

stoc

k-op

tion

gran

ts

(b)

mil

lion

isd

ene

das

min

(1s

alar

yin

pre

viou

sye

ars

div

ided

by$1

mil

lion)

(c)

aft

eris

anin

dic

ator

vari

able

equ

alto

zero

befo

reth

ere

gula

tion

took

plac

e(1

993)

and

one

afte

rwar

d

36 Hall amp Liebman

In the second column the coefcient on million 3 after is essentiallyzero suggesting that any decrease in salary brought about by section162(m) was offset by increases in bonus and stock-option grants Theresults are similar for the median regressions There is some evidence ofa very small slowdown in growth of salary for executives likely to beaffected by section 162(m) but no sign of a change in total compensa-tion Taken together this provides evidence of a very small substitutionof performance-related pay for salary

The last four columns run identical tests but limit the sample to CEOsThe interaction of million 3 after continues to be small and negative inthe salary regressions In contrast the impact of 162(m) on total compen-sation appears to be positive and fairly substantial (about 15 percentagepoints per year) in these regressions That is the increase in bonus andstock options following the million-dollar rule more than offset the verysmall decline in salary

All of these conclusions should be interpreted somewhat cautiouslybecause our data provide us with only one pre-1993 control year Never-theless the data are consistent with section 162(m) having led to a veryminor slowdown in salary growth and one that is dramatically smallerthan what one would conclude without controlling for underlying differ-ential salary growth rates at different levels of salary Since total compen-sation for CEOs did not decline and perhaps increased there is evidenceof a minor substitution of performance-related pay for salary in responseto the regulation

8 EXECUTIVE COMPENSATION AND THEELASTICITY OF TAXABLE INCOME

Even if taxes have only a minor effect on executive compensation thestructure of executive compensation is important for tax policy be-cause high-income taxpayers are sources of signicant revenue and arethe focus of important debates about the efciency cost of taxation

Goolsbee (1999) has recently argued that executive decisions to exer-cise options are highly responsive to intertemporal differences in taxrates created by tax reforms He claims that much of the apparent declinein taxable income for high-income taxpayers between 1992 and 1993documented by Feldstein and Feenberg (1996) can be explained by execu-tives shifting option gains into 1992 in order to avoid the higher marginaltax rates that went into effect in 1993 By extension he implies that thehigh elasticities of taxable income with respect to personal tax ratesestimated by comparing the pre- and post-tax-reform taxable income of

Taxation of Executive Compensation 37

high-income individuals reect intertemporal shifting of income and notpermanent effects of taxation If this is true then the deadweight loss oftaxation is lower as is the cost of progressivity

While Goolsbee presents comprehensive evidence that option gainswere unusually high in 1992 his identication of the impact of tax rates onthe timing of option gains depends on correctly specifying a counter-factual level of options that would have been exercised in the absence ofthe tax change He simply assumes a linear time trend The purpose ofGoolsbeersquos time trend is presumably to account for the increasing use ofoptions over time However since options typically require a vestingperiod of at least a couple of years a much more direct measure of ex-pected option gains can be constructed based on the total value of vestedoptions held by the CEO Since options typically have vesting periods thatexceed two years this total can be treated as exogenous since it will not beaffected by the tax changes

More generally the value of options held by the CEO is likely to havea major effect on his exercise decisions If his options are out of themoney he will not have gains to realize If there has been a recent run-up in the market an optimizing CEO may exercise an unusually largeamount of options in order to diversify his overall portfolio and if in-the-money options are about to expire they will be exercised in thecurrent year

Our evidence suggests that much of the apparent tax shifting caninstead be attributed to stock market performance and the timing of pastoption grants Using our 15-year panel we replicate Goolsbeersquos (1999)result for the 1993 period but show that the impact of taxes is notpresent in other time periods Moreover even the result for the 1993period disappears when appropriate controls for past option grants andstock price appreciation are included

The top three rows of Table 7 present data on the taxable incomes ofCEOs in the HallndashLiebman sample For these results our panel has beennarrowed to include only CEOs whose rmrsquos scal years correspondwith tax years In addition we create a balanced panel for each of thethree tax-reform periods by excluding any rm for which there are notcomplete data for the period surrounding the reform

The fourth row of the table shows that the 1993 tax act raised the topmarginal personal income tax rate from 310 to 396 percent effective fortax year 1993 with a further increase occurring in 1994 due to the uncap-ping of the Medicare payroll tax The top row of the table shows thattaxable income was higher in 1992 than for any other year in our sample(all data are in 1994 dollars) Option gains were particularly high in1992mdashmore than double their 1991 level and 68 percent higher than their

TA

BL

E7

Tax

Ref

orm

Sto

ckM

arke

tP

erfo

rman

cea

ndth

eT

imin

gof

Opt

ion

Gai

ns

Ave

rage

stoc

km

arke

tM

ean

CE

OM

ean

CE

Ore

turn

for

Mea

nva

lue

taxa

ble

sala

ryan

dM

ean

CE

OT

opfe

der

al

rms

inof

opti

ons

(Val

ue

ofop

tion

Sam

ple

inco

me

bonu

sop

tion

gain

sp

erso

nal

sam

ple

for

held

atbe

gga

ins)

(va

lue

Ref

orm

size

Yea

r($

mil

)($

mil

)($

mil

)M

TR

pre

vye

ar(

)of

year

ofop

tion

she

ld)

ER

TA

8127

919

800

874

064

80

163

7025

80

460

354

1981

083

10

672

009

869

245

070

61

3919

820

872

067

90

099

5011

90

686

144

1983

099

00

724

025

250

355

116

62

16

TR

A86

295

1985

119

70

837

024

050

76

116

02

0719

861

448

091

70

345

5042

81

776

194

1987

165

50

959

059

338

519

12

229

266

1988

163

21

058

039

028

07

237

71

6419

891

656

105

20

414

2817

92

345

176

1993

272

1990

209

51

028

061

828

232

288

02

15T

axA

ct19

911

875

102

00

547

312

01

268

82

0319

922

678

107

01

250

3146

64

552

275

1993

232

41

209

074

339

620

64

299

173

1994

196

41

262

037

842

515

45

113

074

Sam

ple

isH

allndash

Lie

bman

rm

sw

ith

Dec

embe

r

scal

year

sth

atha

dco

mp

lete

dat

afo

ron

eof

the

thre

eta

x-re

form

per

iod

sst

ud

ied

Taxation of Executive Compensation 39

1994 level These patterns are consistent with Goolsbeersquos claim that tax-payers responded to the anticipated increase in marginal tax rates byshifting gains into 1992

The evidence from the other two tax reforms is less clear suggestingthat there might be factors other than tax rates determining the timing ofoption gains ERTA 1981 reduced the top marginal tax rate from 70percent in 1980 to 50 percent in 1982 However there is no sign thatoption gains were reduced in 1980 and 1981 and then increased in 198217

There was a big increase in option gains in 1983 however Similarly thehighest year of stock-option gains in the period around 1986 was 1987 astrange year to take gains since taxpayers would have known that thepersonal rate would be 10 percentage points lower if they waited untilthe following year18

What all the peak years of option gains have in common is that theyfollow years of strong stock market performance The fth row of Table 7shows the average stock market return during the previous year for therms in our sample The large value of option gains in 1983 followed a355-percent increase in the stock market The 1992 boom in option gainsfollowed a 466-percent increase in the stock market the more modest1987 peak in option gains followed a 19-percent gain (and a 43-percentgain two years before)

Strong stock market performance raises the value of options availableto exercise The sixth row of the table shows the mean value of optionsheld by CEOs at the beginning of the year All three of the peak years ofoption gains correspond with peak years in terms of the value of optionsavailable for exercise Since stock market appreciation during the yearalso increases the value of options available to exercise later in that yearwe need to allow for current-year stock market appreciation as well

Table 8 contains Goolsbee-style regressions for each of the tax re-forms The rst two columns reprint the results from Goolsbee (1999)The third and fourth columns replicate his results in our sample Inparticular with a specication similar to his we are able to replicate hisnding that the current after-tax share has a positive coefcient and thefollowing yearrsquos after-tax share has a coefcient of roughly negative oneColumns (5) through (8) show that this pattern was unique to the 1993tax reform Indeed in the TRA 86 regressions the coefcient on the

17 As we explained above for executives covered by the maximum tax on earned incomethere was not a decline in tax rates between 1981 and 198218 It is possible that the large stock market decline in October 1997 may have promptedsome people to exercise options earlier than they otherwise would have However anincrease in option exercises typically follows stock market increases not declines

TA

BL

E8

Att

empt

sto

Rep

lica

teth

eT

ax-S

hift

ing

Res

ult

inO

ther

Tim

eP

erio

ds

Goo

lsbe

eG

ools

bee

1993

1993

Ent

ire

Ent

ire

regr

essi

onre

gres

sion

refo

rmre

form

TR

A86

TR

A86

ER

TA

81E

RT

A81

1980

ndash199

419

80ndash1

994

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

ln[1

2ta

x(t)

]1

244

108

20

455

032

62

017

62

017

22

016

72

016

70

026

20

068

(01

34)

(01

27)

(05

37)

(05

42)

(02

10)

(02

09)

(00

90)

(00

91)

(00

67)

(00

65)

ln[1

2ta

x(t1

1)]

20

867

20

764

21

214

20

917

047

80

722

20

103

004

30

176

018

3(0

118

)(0

112

)(0

491

)(0

499

)(0

157

)(0

174

)(0

072

)(0

072

)(0

068

)(0

066

)

Tim

e0

135

007

62

000

52

036

40

041

000

80

094

003

50

056

002

7(0

008

)(0

008

)(0

049

)(0

049

)(0

028

)(0

028

)(0

025

)(0

026

)(0

003

)(0

003

)

ln(m

arke

tva

lue)

066

10

489

023

70

334

039

5(0

015

)(0

058

)(0

039

)(0

042

)(0

017

)

Yea

rs19

91ndash1

995

1991

ndash199

519

90ndash1

994

1990

ndash199

419

85ndash1

989

1985

ndash198

919

80ndash1

983

1980

ndash198

319

80ndash1

994

1980

ndash199

4

Ad

just

edR

20

720

770

610

630

690

760

800

810

630

68

Obs

erva

tion

s13

856

132

371

385

136

41

470

143

31

113

106

23

968

385

9

Dep

end

ent

vari

able

isth

elo

gof

taxa

ble

inco

me

All

regr

essi

ons

con

tain

rm

xe

def

fect

sG

ools

bee

regr

essi

on(2

)al

soco

ntai

nstw

oot

her

vari

able

sno

tsh

own

inth

eta

ble

Stan

dar

der

rors

inp

aren

thes

es

Taxation of Executive Compensation 41

TABLE 9Alternative Explanations for the Timing of Option Exercises

1993 Entirereform TRA 86 ERTA 81 1980ndash1994

(1) (2) (3) (4)

ln[12tax(t)] 0108 0111 20051 0201(0324) (0136) (0048) (0057)

ln[12tax(t11)] 20134 0573 0125 0150(0336) (0156) (0053) (0066)

ln(market value) 0418 0177 0338 0418(0061) (0039) (0046) (0163)

Firmrsquos stock market return (t) 0298 0268 0214 0258(0044) (0045) (0031) (0023)

Firmrsquos stock market return 0162 0206 0031 0128(t21) (0041) (0008) (0029) (0021)

ln(value of options held by 0004 0008 0001 0009CEO at start of year) (0003) (0003) (0002) (0001)

Years 1990ndash1994 1985ndash1989 1980ndash1983 1980ndash1994Adjusted R2 064 071 082 068Observations 1349 1427 1036 3812

The dependent variable is the log of taxable income All regressions contain rm xed effects Stan-dard errors in parentheses

following yearrsquos tax rate is positive and signicant Columns (9) and (10)pool all three reforms and fail to replicate the pattern that would reectintertemporal shifting19

It is possible that the ease with which one can shift the timing ofoption gains is asymmetric It may be harder to shift option gains for-ward in time since some options expire and therefore must be exercisednow In addition options were a much smaller portion of compensationin the early 1980s so perhaps the ERTA 81 results are not surprisingNonetheless these results cast doubt on the proposition that this is thechannel through which spurious estimates of the elasticity of taxableincome have occurred for previous tax reforms and they suggest thatwe should look harder for alternative explanations for the timing ofoption gains

Table 9 attempts to predict options gains using the lagged and currentincrease in a rmrsquos market value and the value of the options held by theCEO at the start of the year The two stock market variables stronglypredict option gains while the value of options held has the correct sign

19 The results also fail to suggest large permanent effects of marginal tax rates on taxableincome

42 Hall amp Liebman

in all of the specications and is signicant in the pooled sample eventhough it is highly correlated with the two stock market return variablesIn contrast the tax-timing explanation fails even in the 1993 period

It is important to realize that our results do not invalidate Goolsbeersquoscritique of claims that aggregate revenue data from 1992 and 1993 provethat elasticities of taxable income are large We have simply given analternative explanation for why taxable incomes were unusually high in1992 However our results do cast doubt on the conclusion that largetiming shifts are ubiquitous and capable of explaining away the morecareful estimates of taxable income elasticities that have been performedusing micro data (eg Feldstein 1995 Auten and Carroll 1995)

One further point is in order It is sometimes argued that the combina-tion of increased use of options and the booming stock market explainswhy federal tax revenues have been so much higher than expected dur-ing the 1990s However while large option gains increase revenue fromthe personal income tax they produce offsetting deductions for cor-porations The net impact on revenue is only the difference in the twomarginal tax rates times the option gain While the gap between thepersonal rate and the effective corporate rate is nontrivial the impact oflarge option exercises on total revenue is substantially smaller than theamount by which the exercises inate personal tax revenue and reducecorporate tax revenue

9 CONCLUSION

We have described the tax rules for executive pay in detail and analyzedhow changes in various tax rates affect the tax advantages of stock optionsrelative to salary and bonus We nd that there is a moderate tax advan-tage to optionsmdashcurrently about $4 per $100 in pretax executive paymdashand that changes in the tax advantage of options over time have had atmost a modest impact on the composition of pay Corporate governancefactors particularly the role of large investors appear to be more impor-tant in explaining the dramatic increase in option pay

In addition our evidence suggests that more direct attempts to use taxpolicy to inuence executive compensation have had little effect Wend that the million-dollar rule led companies to substitute perfor-mance-based pay for salary But our evidence suggests that this substitu-tion was quite modest and there is no evidence that the total level of paywas reduced Overall although the stock-option explosion has dramati-cally increased the link between pay and performance this change isdue almost entirely to nontax factors

Taxation of Executive Compensation 43

REFERENCES

Auerbach Alan J (1996) ldquoThe Dynamic Effects of Tax Law Asymmetriesrdquo Re-view of Economic Studies 53451ndash405

mdashmdashmdash and James Poterba (1987) ldquoTax Loss Carryforwards and Corporate TaxIncentivesrdquo In The Effects of Taxation on Capital Accumulation M Feldstein(ed) Chicago University of Chicago Press

Altshuler Rosanne and Alan Auerbach (1990) ldquoThe Signicance of Tax LawAsymmetries An Empirical Investigationrdquo Quarterly Journal of Economics 10561ndash86

Auten Gerald and Robert Carroll (1995) ldquoThe Effect of Income Taxes on House-hold Behaviorrdquo Ofce of Tax Analysis US Department of the Treasury

Baker George P and Brian J Hall (1998) ldquoCEO Incentives and Firm SizerdquoNBER Working Paper no 6868 December

Core John and Wayne Guay (1999) ldquoThe Use of Equity Grants to ManageOptimal Equity Incentive Levelsrdquo Draft

Feldstein M (1995) ldquoThe Effect of Marginal Tax Rates on Taxable Income APanel Study of the 1986 Tax Reform Actrdquo Journal of Political Economy 103 (no3)551ndash72

mdashmdashmdash and D Feenberg (1996) ldquoThe Effect of Increased Tax Rates on TaxableIncome and Economic Efciency A Preliminary Analysis of the 1993 Tax RateIncreasesrdquo In Tax Policy and the Economy vol 10 James M Poterba (ed)Cambridge MA MIT Press

Gompers Paul A and Andrew Metrick (1998) ldquoInstitutional Investors andEquity Pricesrdquo NBER Working Paper no 6723

Goolsbee Austan (1999) ldquoWhat Happens When You Tax the Rich Evidencefrom Executive Compensationrdquo NBER Working Paper no 6333

Graham John R (1996) ldquoProxies for the Corporate Marginal Tax Raterdquo Journal ofFinancial Economics 42187ndash221

Hall Brian J (1999) ldquoThe Design of Multi-year Stock Option Plansrdquo Journal ofApplied Corporate Finance 12(2)97ndash106

mdashmdashmdash and Jeffrey B Liebman (1998) ldquoAre CEOs Really Paid Like Bureau-cratsrdquo Quarterly Journal of Economics CXII(no 3)653ndash691

Jensen Michael C and William M Meckling (1976) ldquoTheory of the Firm Mana-gerial Behavior Agency Costs and Ownership Structurerdquo Journal of FinancialEconomics III305ndash360

mdashmdashmdash and Kevin J Murphy (1990) ldquoPerformance Pay and Top-ManagementIncentivesrdquo Journal of Political Economy XCVIII225ndash264

Kaplan Steven N (1997) ldquoThe Evolution of US Corporate Governance WeAre All Henry Kravis Nowrdquo Draft

Livingstone Jane R (1999) ldquoExecutive Compensation Contracting and Re-sponses to Increased Tax Costsrdquo Louisiana State University June

Miller Merton H and Myron S Scholes (1982) ldquoExecutive CompensationTaxes and Incentivesrdquo In Financial Economics Essays in Honor of Paul CootnerWilliam F Sharpe and Cathryn M Cootner (eds) Princeton NJ Prentice-Hall

Murphy Kevin J (1999) ldquoExecutive Compensationrdquo forthcoming in OrleyAshenfelter and David Card (eds) Handbook of Labor Economics Vol 3 NorthHolland

Perry Todd and Marc Zenner (1999) ldquoPay for Performance Government Regu-

Note Holds constant the personal rate at 40 percent and the capital gains rate at 20 percent

Note Holds constant the corporate rate at 40 percent and the capital gains rate at 20 percent

Note Holds constant the personal rate at 40 percent and the corporate rate at 40 percent

FIGURE 2 Tax Advantage of Options NPV of Tax Advantage of Pay-ing $100 in Options (NQSOs) Rather Than in Cash

Taxation of Executive Compensation 19

avoiding the capital gains taxes is greater The tax advantage of optionstherefore declines linearly as the personal tax rate rises At a 100-percentpersonal tax rate the executive receives nothing in either case and byconstruction the companyrsquos tax deduction benet is the same in NPVterms so the tax advantage of options falls to zero also6

433 The Tax Advantage of Options Changing the Capital Gains TaxRate The tax advantage of options for various capital gains rates isdepicted in the bottom panel of Figure 2 under the same assumptions(this time holding both the personal rate and the corporate rate at 40percent while varying the capital gains rate) The tax advantage of avoid-ing the capital gains tax increases as the capital gains tax rate rises

5 HOW HAS THE TAX ADVANTAGE OF OPTIONSCHANGED OVER TIME

The analysis so far has illustrated how the tax advantage of optionsvaries with the changes in tax rates In this section we show how the taxadvantage of options has changed over time as tax rates have changed

51 Top Marginal Tax Rates over TimeThe evolution on the top marginal tax ratesmdashpersonal corporate andcapital gainsmdashfrom 1980 to 1998 is shown in Figure 3 In all cases thetop rate (the rate for taxpayers with the highest incomes) is shownwhich is not always the highest rate since various anomalies (such asclawbacks of exemptions) sometimes lead to marginal tax rates that arehigher than those for the highest income tax payers

The top personal tax rate was 70 percent in 1980 falling to 50 percentin 1982 and 28 percent in 1988 following the 1986 tax act7 The topmarginal tax rate has since risen to 396 percent but is effectively 425percent since there is a 29-percent Medicare surcharge (paid half by theemployer and half by the employee) that has no upper limit The topcorporate tax rate has had only one signicant change since 1980 fallingfrom 46 percent before the 1986 tax act to 34 percent following the act in1988 The top corporate tax rate was increased to 35 percent in 1993 The

6 This is one place where ignoring bond investments has substantive implications Inparticular in the all-bond case the tax advantage of options is nonmonotonicmdashit rises andthen fallsmdashas personal tax rates rise7 During the early 1980s the maximum tax on earned income limited the marginal tax rateon the earnings of high earners so many executives did not face a decline in personal taxrates between 1981 and 1982 The empirical results in this paper are robust to assumingthat executives faced a marginal tax rate of 50 percent in 1980 and 1981

20 Hall amp Liebman

FIGURE 3 Top Marginal Tax Rates since 1980

top capital gains rate has uctuated between 28 percent (approximately)and 20 percent since 1980 and is currently at 20 percent

52 The Evolution of the Tax Advantage of NQSOsThe tax advantage of options relative to cash dened in the same way asin section 4 from 1980 to 1998 is shown in Figure 4 The calculations aredone using the same assumptions as before (ie N 5 10 years and cashis invested in equities) and the statutory top marginal tax rates shown inFigure 3 are used Due to the fall in corporate and personal tax rates thetax advantage of options increased sharply after the Tax Reform Act of1986 This increase in tax advantage has largely been reversed in the1990s as top personal tax rates have crept back up and the capital gainsrate has been reduced

Two key facts emerge from this analysis First there is currently only amoderate tax advantage to standard nonqualied stock optionsmdashon theorder of $4 per $100 of compensation This is because the tax advantagesto the executive of deferring taxes are largely offset by the tax disad-vantages to the company of not being able to deduct option expensesfrom taxable prots until the executive exercises the options Secondalthough the 1986 tax act substantially increased the tax advantage ofoptions more than half of this increase has been reversed in the 1990s

Taxation of Executive Compensation 21

FIGURE 4 Tax Advantage of Options 1980 to 1998

NPV of tax advantage of $100 in options (NQSOs) rather than cash

53 The Tax (Dis)advantage of ISOsSo far the analysis has focused on the tax advantages of NQSOs sincethey are so much more prevalent than ISOs which account for onlyabout 5 percent of option grants The relative scarcity of ISOs can beexplained both by their tax status and by their per-executive cap of$100000 per year

Under what conditions are ISOs tax-preferred to NQSOs For a trans-fer P to an executive the ISO is always tax-preferred by the executive(since the capital gains rate is lower than the personal rate) while theNQSO is always tax-preferred from the companyrsquos perspective (since anISO is not deductible) The key issue however is the relative advan-tages of the two types of options from a global-contracting perspectiveBecause ISOs are not deductible the company is indifferent betweensetting aside P in the form of NQSOs today and setting aside P(1 2 Tc) inthe form of ISOs (Since the NPV of the tax deduction for NQSOs is PTcthe NPV of the payment P is P 2 PTc or P (1 2 Tc))

Equalizing the posttax cost of NQSOs and ISOs we need only look atthe payoff of the executive to determine the condition under which eachis preferred The payoff of NQSOs in N years is

P [1 1 r (1 2 Tc)]N (1 2 Tp) (7)

22 Hall amp Liebman

The same payoff for ISOs is

P (1 2 Tc) [1 1 r (1 2 Tc)]N (1 2 Tcg) (8)

ISOs therefore are tax-advantaged if

Tp 2 Tcg Tc 2 Tc Tcg (9)

Intuitively ISOs are tax-advantaged only if their advantage (the differ-ence between the personal rate and the capital gains rate) is largeenough to offset their cost (the disadvantage of not deducting at thecorporate rate) Note however that the condition is not a simple com-parison between the corporate rate and the personal rate minus thecapital gains rate (Tp 2 Tcg Tc)

Figure 5 shows how the tax advantage of NQSOs relative to ISOs haschanged since 1980 as tax rates have changed Unlike the very modestchanges (around 3 to 4 percent) in the relative tax advantages betweenoptions and cash shown in Figure 4 the relative tax advantage betweenNQSOs and ISOs has seen enormous swings When the top personalrate was 70 percent in 1980 ISOs were tax-favored by a margin greaterthan 6 percent However by 1982 (following the 1981 tax acts that dra-matically lowered the personal tax rate) ISOs became tax-disadvantaged

FIGURE 5 Tax Advantage of NQSOs Relative to ISOs over Time

NPV of tax advantage of $100 payment to executive NQSOs minus ISOs

Taxation of Executive Compensation 23

by about 5 percent By 1988 (following the 1986 tax act) ISOs became tax-disadvantaged by about 18 percent a dramatic swing The raising ofpersonal rates coupled with a decrease in the capital gains rate has sincereduced the tax disadvantage of ISOs to about $4 per $100 of pretaxcompensation Although hard data on ISOs are hard to come by practi-tioners (mostly compensation consultants) claim that ISOs were morepopular prior to 1982 when they were tax-advantaged Although ISOsare less disadvantageous from a tax standpoint relative to NQSOs thanthey were in the late 1980s they are still disadvantageous so it is notsurprising that they have not made a signicant resurgence in recentyears

6 EXPLAINING THE INCREASE IN OPTION-BASEDCOMPENSATION

In this section we analyze whether changes in tax rates affect the compo-sition of executive compensation and consider the relative importanceof tax factors and corporate governance factors in explaining the in-creasing share of compensation paid in the form of stock options Thenumerous changes in tax rates that have occurred since 1980 provide anopportunity to assess whether executive compensation arrangementsrespond to tax incentives in the way that the tax-avoidance model sug-gests We test this model using a panel data set of CEOs in largepublicly-traded US corporations Our identication of the tax effectsrelies on time-series variation in personal corporate and capital gainstax rates along with cross-sectional variation in corporate tax rates

61 DataWe use a 15-year panel data set of CEOs in the largest publicly-tradedUS corporations which is described in Hall and Liebman (1998) Thedata set covers the years from 1980 through 1994 and combines CEOcompensation information from corporate proxies and 10-K lings withstock price and stock return information from CRSP and accounting datafrom Compustat In addition some compensation data from the 1970swere collected in order to construct measures of the value of stock op-tions held by the CEOs in the rst period of the sample8

The feature that distinguishes our data from most other CEO datasets is that with our panel of yearly proxy data on option grants optiongains and total options held we are able to calculate the total value of

8 We are grateful to David Yermack for providing us with some of the data for 1984 to 1991See Yermack (1995) for a discussion of these data

24 Hall amp Liebman

all stock options held by the CEO at a point of time More importantlysince we have the details about the stock options held (number exer-cise price time to maturity etc) we can precisely calculate the changein the value of a CEOrsquos stock option holdings for a given change in rmvalue

62 Identication IssuesThe numerous tax changes over our sample period and the specic waysin which changes in personal corporate and capital gains rates arepredicted to affect the composition of CEO compensation give us un-usually rich sources of identication However because many of thechanges in tax rates would be expected to affect all of the CEOs in oursample in a similar way we will need to pay particular attention toseparating out the tax effects from underlying time trends and otherfactors that changed over time Controlling for underlying trends is par-ticularly important in this study because we are focusing on an out-comemdashthe use of stock optionsmdashthat has increased rapidly over thepast two decades and that many practitioners believe was importantlyaffected by nontax considerations Thus the tax effects we are trying toexplain are deviations from a rising trend

We take two steps to try to separate out the tax effects from theunderlying trend First we include key nontax factors that could poten-tially explain the increasing reliance on performance-based compensa-tion It has been suggested that the dramatic increase in incentive-basedpay is the result of the remarkably poor shareholder returns during the1970s which spurred the LBO and takeover movements of the 1980sAccording to this story (Kaplan 1997) shareholders became much morepowerful via the rise of institutional investors even as the LBO andtakeover movements waned Because the inuence of institutional inves-tors is thought to be one of the most important mechanisms of strongcorporate governance we use the share of each companyrsquos stock ownedby institutional investors (dened as institutions with more than $100million under management) as an explanatory variable9 The percentageof shares owned by large institutional investors increased from about 20percent to almost 50 percent in 1994 an upward trend that closelymatches the sharp rise in the share of compensation in the form ofoptions Indeed in our sample the annual average (over all of the rmsin our sample) share of stock owned by institutional investors has acorrelation of 09 with the annual average share of compensation paid inoptions

9 This variable is described in Gompers and Metrick (1998)

Taxation of Executive Compensation 25

In addition we include two other variables that proxy for strongercorporate governance The rst is the size of the board Evidence suggeststhat boards with smaller numbers of directors reduce the inuence of theCEO and better represent shareholder interests (Yermack 1995) Boardsize fell by approximately 15 percent between 1980 and 1994 The secondvariable is the fraction of the rmrsquos board members who are inside direc-tors or gray directors (noninsiders who have business dealings with thecompany) Firms with fewer outsiders are less likely to act in the share-holdersrsquo interests and we would predict that they would be less likely tohave performance-based compensation for their CEOs The average per-centage of inside and gray directors has decreased from about 45 percentin 1980 to about 35 percent in 1994

Second we allow for cross-sectional variation in corporate tax ratesCorporate marginal tax rates vary because rms may be eligible for tax-loss carrybacks and carryforwards investment tax credits and the alter-native minimum tax (see Auerbach 1986 Auerbach and Altshuler1990 Auerbach and Poterba 1987) Because these cross-sectional differ-ences imply that different rms should respond differently to a givenchange in tax rates they potentially provide us with a way to isolatethe tax effects Firm-specic corporate marginal tax rates are notori-ously difcult to calculate from publicly-available data We use a tricho-tomous variable that equals zero if the rm has tax-loss carryforwardsand negative earnings in a given year equals 05 times the statutorycorporate rate if the rm has only one of those conditions and equalsthe statutory rate if the rm has neither loss carryforwards nor negativeearnings10

The cross-sectional variation in the corporate tax rate may not be ex-ogenous Firms that perform poorly could face low marginal tax ratesand also be particularly likely or unlikely to provide performance-basedpay Because most of our specications use xed effects the correlationwe must be concerned about is between changes in rm performanceand changes in compensation We deal with this by controlling forlagged rm stock market performance in our regressions This shouldeliminate spurious correlation between tax rates and compensation thatis jointly caused by rm performance

10 Graham (1996) presents evidence that the trichotomous variable performs nearly as wellas his recommended simulated marginal tax rate Plesko (1999) compares marginal taxrates calculated from Compustat data with actual marginal tax rates from rm tax returnsHe nds that simple binary variables are more reliable measures of the marginal tax ratethan is either the Graham or the trichotomous variable We obtain substantively similarresults when we switch from the trichotomous variable to the binary variables recommendby Plesko

26 Hall amp Liebman

63 ResultsThe dependent variable in our regressions is the fraction of a CEOrsquos totalannual compensation (measured as the BlackndashScholes value of stock-option grants plus salary and bonus) that comes from stock-optiongrants We choose this reduced-form specication because our interest isin the composition of CEO compensation However the factors thataffect the composition of compensation might also affect the levels andit will be important to keep this in mind in interpreting our results

Optimal-contracting theory generally makes predictions about the in-centives provided by the total amount of rm stock and stock optionsowned by the CEO We choose to model the ow of option grants ratherthan the ultimate performance sensitivity of the CEOrsquos compensationand stock and stock-option holdings because we have in mind anadjustment-cost model in which it takes time for a CEOrsquos holdings ofstock and stock options to reach the optimal level11 In our regressionswe include two measures of the current performance sensitivity of theCEOrsquos wealth to rm performance as explanatory variables to reect thedistance of the CEOrsquos contract from the optimal contract

Table 3 presents results in which the tax effects are identied solely byvariation in tax rates over time The fraction of annual compensation paidin stock options is regressed on tax variables corporate governance vari-ables contract-theory variables and variables reecting stock market re-turns Both of the regressions in this table include rm xed effects Incolumn (1) the fraction in annual compensation paid in stock options isregressed on the log difference in payoff (the combined payoff to both therm and the CEO) from receiving compensation in stock options ratherthan salary and bonus where the payoff calculations are based on statu-tory marginal tax rates This variable is dened as the log difference be-tween equation (3) and equation (6) in section 4 The coefcient on the taxvariable is large and statistically signicant at the 95-percent level Thepoint estimate implies that a one-percent increase in the payoff differencebetween stock options and salary and bonus results in a 24-percentage-point increase in the share of compensation paid in stock options

The corporate governance variables all have the predicted sign Firmswith a higher fraction of their shares owned by institutional investorsuse more performance-based pay (stock options) Firms with large corpo-rate boards or a large share of inside directors are less likely to use stockoptions

The two contract-theory variables are the dollar change in CEO wealth

11 See Core and Guay (1999) for evidence that rms adjust in this way

TABLE 3Compensation Regressions that Rely on Variation in

Tax Rates over Time

(1) (2)

Tax variablesln (difference in payoff)(a) 2393 (0349)ln (12statutory personal 20015 (0034)

rate)ln (12statutory cap gains 0330 (0171)

rate)ln (12statutory corporate 0399 (0148)

rate)

Corporate governance variablesFraction of rmrsquos shares 0043 (029) 0052 (0028)

owned by institutionalinvestors

Size of the rmrsquos board of 20002 (001) 20003 (0001)directors

Fraction of directors who 20082 (033) 20092 (0030)are inside or gray

Contract-theory variablesD (CEO wealth) per $1000 20261 (0095) 20247 (0087)

in FV(b)

D (CEO wealth) per 10 in 2583 (189) 2554 (201)FV(c)

Volatility(d) 521 (424) 261 (369)ln (rm market value) 0019 (0006) 0019 (0006)

Stock market return variablesLagged annual rm stock 0006 (0009) 0003 (0008)

market returnLagged annual SampP 500 re- 20035 (0023) 20029 (0023)

turn

Regression diagnosticsTotal sample size 5179 5179Firms 427 427Adjusted R2 0313 0311

Dependent variable share of compensation paid in options Standard errors in parentheses Bothregressions in this table include rm xed effects

(a) Log difference in payoff is the log difference in combined rm and CEO after-tax share betweenpaying compensation in options and paying salary

(b) This variable is the dollar change in CEO wealth per $1000 change in rm market value (seeJensen and Murphy 1990)

(c) This variable in the dollar charge in CEO wealth per 10-percent change in rm value (see Hall andLiebman 1998)

(d) Standard deviation of rm market value in trillions

28 Hall amp Liebman

per $1000 change in rm value (Jensen and Murphy 1990) and thedollar change in CEO wealth per 10-percent change in rm value (Halland Liebman 1998) As discussed in Baker and Hall (1998) these twowealth sensitivity measures reect two different concepts of how closelyaligned the managersrsquo incentives are with the interests of the sharehold-ers In the regression they are meant to measure how far the CEOrsquosexisting contract and ownership of stock and stock options is from theoptimal level The negative coefcients on the two variables suggeststhat rms that are below their optimal pay-to-performance sensitivityare more likely to give more stock options in the current period

The second column replaces the tax variable motivated by deferraladvantages with the three statutory rates The corporate after-tax sharehas a positive coefcient as we would expect because the tax advantageof stock options increases as the corporate rate falls The coefcient onthe personal tax rate is small and statistically insignicant The sign ofthe coefcient on the capital gains rate is the opposite of that predictedby tax-avoidance theory

Note that these results rely on time-series variation alone This is not avery convincing test Since the log difference in payoffs mostly rises overthe time period covered in our sample as does the share of compensa-tion paid in options it is possible that the tax variable is simply reectingother factors that were changing over this time period that we have notincluded in our regression Therefore it is important to look at specica-tions that rely on cross-sectional variation in the tax variables

Table 4 contains specications that rely on cross-sectional variation incorporate tax rates The rst column repeats the specication from col-umn (1) of Table 3 but uses the trichotomous measures of rm corporatetax rates in place of the statutory rate in calculating each rmrsquos logdifference in payoff from paying compensation in options rather than insalary and bonus As in the previous table we nd the positive coef-cient on the log difference in payoff just as the theory suggest How-ever the coefcient estimate is now much smaller suggesting that aone-percent increase in the log difference in payoff produces a 01-percentage-point increase in the share of compensation paid in options

Introducing cross-sectional variation in the dependent variable doesnot eliminate the potentially spurious correlation over time between thetax variable and the share of compensation paid in options The secondcolumn of Table 4 isolates the pooled cross-sectional variation in corpo-rate tax rates by adding time dummies and dropping the rm xedeffects In this specication the coefcient on the tax variable is indistin-guishable from zero The coefcient on the institutional variable contin-ues to be large and statistically signicant The third column includes

Taxation of Executive Compensation 29

both rm xed effects and time dummies which tests for a relationshipbetween within-rm changes in the dependent variable and within-rmchanges in each explanatory variable In this specication the coefcienton the tax variable is again indistinguishable from zero The coefcientson the corporate governance variables are also statistically insignicant

The third column contains a specication that treats the Tax ReformAct of 1986 (TRA 86) as a natural experiment For rms and CEOs facingthe statutory corporate tax rate TRA 86 increased the relative payoff topaying compensation in options by 17 percent For a rm facing a zeromarginal tax rate it reduced the relative payoff by 45 percent The reduc-tion occurred because TRA 86 reduced the personal tax rate makingsalary and bonus more attractive while the decreased corporate rate hadno effect on zero-tax-rate rms Thus we would expect rms facing zeroor low marginal tax rates to reduce their use of options after TRA 86while rms facing the statutory rate would be expected to increase themFew rms face zero marginal tax rates year after year Therefore we tryto distinguish between rms that often face low tax rates and rms thatnearly always face the statutory rate We dene low-tax rms as oneswhose average marginal tax rate over 1984 to 1986 was below 02412

Then we ran a difference-in-differences regression to see if high-taxrms increased their use of options more than low-tax rms The interac-tion between post-TRA 86 and high corporate tax rates is the key vari-able The estimated coefcient on this variable is positive but small andnot statistically different from zero at the 95-percent level

It is important to note that in this specication as well the main corpo-rate governance variable the share of the rmrsquos shares owned by institu-tional investors performs as theory would predict The variable appearsto be quite robust Table 5 presents regressions of the share of compensa-tion paid in stock options on the fraction of the rmrsquos shares owned byinstitutional investors Since we have already established that these vari-ables have a strong longitudinal relationship the regressions here aredesigned to focus on the cross-sectional relationship between these twovariables The rst and third columns present regressions using data onlyfrom 1994 and nd a coefcient of roughly 02 (column 1 contains noother covariates while column 3 includes all of the other nontax covari-ates from the regressions in Tables 3 and 4) Since the median share of armrsquos stock owned by institutional investors in our sample rises from 19percent in 1980 to 49 percent in 1994 this coefcient would imply that this

12 This cutoff was chosen after inspecting the distribution of tax rates and essentiallyseparates rms that are always at or near the statutory rate from ones that are sometimesbelow it

TA

BL

E4

Com

pens

atio

nR

egre

ssio

nsth

atR

ely

onC

ross

-Sec

tion

alV

aria

tion

inC

orpo

rate

Tax

Rat

es

(1)

(2)

(3)

(4)

Tax

vari

able

s(a)

ln

(dif

fere

nce

inp

ayof

f)0

099

(00

41)

20

038

(00

36)

20

005

(00

44)

Pos

t-T

RA

862

000

5(0

013

)H

igh

corp

orat

eta

xra

te0

018

(00

17)

(Pos

t-T

RA

86)

3hi

gh0

024

(00

16)

corp

orat

eta

xra

te

Cor

por

ate

gove

rnan

ceva

riab

les

Frac

tion

of

rmrsquos

shar

es0

109

(00

27)

011

4(0

015

)2

000

9(0

030

)0

087

(00

34)

owne

dby

inst

itu

tion

alin

vest

ors

Size

ofth

e

rmrsquos

boar

dof

20

004

(00

01)

20

001

(00

01)

20

002

(00

01)

000

1(0

002

)d

irec

tors

Frac

tion

ofd

irec

tors

wh

o2

012

4(0

033

)2

000

3(0

018

)2

004

1(0

034

)0

001

(00

39)

are

insi

de

orgr

ay

Con

trac

t-th

eory

vari

able

sD

(CE

Ow

ealt

h)p

er$1

000

20

238

(00

94)

20

167

(00

53)

20

257

(00

96)

20

021

(01

47)

inFV

(b)

D(C

EO

wea

lth)

per

10

in2

560

(17

8)2

592

(10

5)2

594

(19

6)2

147

(04

86)

FV(c

)

Vol

atili

ty(d

)2

48(4

24)

162

(03

4)1

12(4

28)

106

(06

09)

ln(

rmm

arke

tva

lue)

003

3(

006)

000

8(0

003

)0

009

(00

06)

001

2(0

007

)

Stoc

km

arke

tre

turn

vari

able

sL

agge

dan

nual

rm

stoc

k2

000

2(0

009

)0

011

(00

10)

000

6(0

009

)2

001

2(0

013

)m

arke

tre

turn

Lag

ged

annu

alSamp

P50

02

005

6(0

023

)2

005

3(0

051

)2

003

9(0

049

)2

015

4(0

040

)re

turn

Reg

ress

ion

dia

gnos

tics

T

otal

sam

ple

size

5179

5179

5179

2534

Firm

s42

742

739

6A

dju

sted

R2

031

009

032

005

Firm

xed

effe

cts

Yes

No

Yes

No

Tim

ed

um

mie

sN

oY

esY

esN

o

Dep

end

ent

vari

able

isfr

acti

onof

ann

ualc

omp

ensa

tion

from

stoc

kop

tion

sSt

and

ard

erro

rsin

par

enth

eses

(a)

All

thre

esp

eci

cati

ons

use

tric

hoto

mou

sco

rpor

ate

tax

rate

sT

he

sam

ple

inco

lum

n(3

)u

ses

only

the

year

s19

84to

1988

sin

cew

ear

eu

sing

1986

asa

nat

ural

exp

erim

ent

(b)

Thi

sva

riab

leis

the

dol

lar

chan

gein

CE

Ow

ealt

hp

er$1

000

chan

gein

rm

mar

ketv

alu

e(s

eeJe

nse

nan

dM

urp

hy

1990

)

(c)

Thi

sva

riab

lein

the

dol

lar

char

gein

CE

Ow

ealt

hp

er10

-per

cent

chan

gein

rm

valu

e(s

eeH

alla

nd

Lie

bman

199

8)

(d)

Stan

dar

dd

evia

tion

ofr

mm

arke

tva

lue

intr

illio

ns

32 Hall amp Liebman

TABLE 5Large Institutional Investors and Stock-Option Compensation

The Cross-sectional Relationship between Institutional Investors andOption Grants

(1) (2) (3) (4)

Coefcient on fraction of 217 (067) 145 (014) 177 (069) 120 (015)rmrsquos shares owned byinstitutional investors

Sample period 1994 1980ndash1994 1994 1980ndash1994Contract-theory and stock No No Yes Yes

market covariatesTime dummies No Yes No Yes

The dependent variable is the share of compensation paid in options

factor alone can explain 6 percentage points of the 23-percentage-pointincrease in the median share of compensation paid in options that oc-curred over this time period Of course this variable is only one measureof the strengthening of corporate governance so the total impact of im-proved corporate governance could be greater The second and fourthcolumns of the table use the entire 1980ndash1994 sample and include timedummies to isolate the cross-sectional correlations Again there is evi-dence of a strong cross-sectional relationship between the two variablesalthough the coefcients are slightly smaller perhaps reecting a weakercorporate-governance role in the earlier period In sum there is a strongrelationship in both longitudinal and cross-sectional data between theinstitutional-investor variable and the options-percentage variable13

7 THE MILLION-DOLLAR RULE

In 1993 section 162(m) of the Internal Revenue Code was enactedlimiting the deductibility of executive compensation in excess of onemillion dollars unless the compensation was performance-related Pro-ponents of this legislation argued that it would slow the growth ofexecutive pay and tighten the link between rm performance and execu-tive compensation

The new provision became effective for tax year 1994 and applies tothe CEO and the other four most highly compensated executives in eachcompany Pay that is performance-related such as stock-option grants

13 The one specication in which the fraction owned by institutional investors is notsignicant is in column (2) of Table 4 a regression that includes both xed effects and timeeffects

Taxation of Executive Compensation 33

and bonuses paid for meeting clear performance goals is not affected bythis regulation provided that it has been approved by shareholdersThus the main impact of the provision is to limit the deductibility ofsalary in excess of a million dollars Woodlock and Antenucci (1997)studied the proxy statements of 376 rms and documented that mostrms responded to the new law by qualifying top executive pay asperformance-related Perry and Zenner (1999) identied 25 rms thatreduced salaries from above $1 million to below $1 million and foundthat 23 of the rms cited section 162(m) as their reason for doing so

In this section of the paper we investigate whether this provision hasaffected the level or structure of executive pay using the Execucompdata base which contains information on the highest-paid executives ineach of the rms in the SampP 500 SampP Midcap 400 and SampP SmallCap600 Between 1993 and 1998 the median salary of the Execucomp execu-tives rose from $230000 to $306000 The median value of total compen-sation (salary bonus and option grants) rose from $470000 to $882000Limiting the sample of CEOs indicates that the total compensation of themedian CEO rose from $11 million to $18 million over the period whilethe median salary rose from $441000 to $550000 Thus executive com-pensation clearly continued to rise at a rapid rate after the implementa-tion of the million-dollar rule though perhaps at a slower rate than itotherwise would have

If the regulation affects behavior it should have the largest impact onrms whose executives are receiving salaries that are above the limit orjust below it Thus we might expect to see rms whose executives arereceiving relatively high salaries to rely more heavily on performance-based pay and less heavily on salary in the years following the regulationPerry and Zenner (1999) and Livingston (1999) both nd that this is thecase This trend can be clearly identied in the Execucomp data set Forexample over 1993 to 1998 the median annualized growth rate of salaryfor executives whose 1993 salary was below $500000 in the previous yearwas 85 percent For executives with salary between $500000 and$700000 it was 64 percent for executives between $700000 and $850000it was 50 percent for executives between $750000 and $1 million it was31 percent and for executives above $1 million it was 00 percent14

The observation that executives at lower levels of salary had highergrowth rates is not very strong evidence of an impact of section 162(m)however because it is possible that low-salary executives would havehad higher growth rates even in the absence of the provision To explore

14 Of the 3137 executives for whom we have complete data between 1993 and 1998 2708had 1993 salaries below $500000 while only 35 had 1993 salaries above $1 million

34 Hall amp Liebman

whether this is the case we use Execucomp data from both before andafter the law change Ideally we would want to have data from manyyears before the change to examine whether there is an underlying pat-tern of faster salary growth for low-salary executives Unfortunately theExecucomp data begin in 1992 and therefore provide us with only the1992ndash1993 growth rate in compensation to use as a baseline15

We run regressions pooling data from 1992ndash1993 through 1997ndash1998of the annual percentage change in different forms of compensation on avariable million that is designed to measure the likelihood that anexecutiversquos compensation will be affected by section 162(m) We denemillion as the minimum of 1 and the executiversquos salary divided by $1million in the previous year Thus any executive with salary at or above$1 million would receive a value of 1 while an executive with a salary of$500000 would be coded as a 0516 We include this variable by itself andalso interacted with a dummy variable (after) that equals 1 for years1994 and after Thus the million variable is designed to pick up anyunderlying relationship between the level of salary and the growth rateof salary while the interacted one will identify any differential growthrate of salary by income in the period after section 162(m) took effect Wealso include a full set of year dummies the natural logarithm of thermrsquos market value and the rmrsquos annual rate of return over the twoprevious years as control variables To reduce the sensitivity of our re-sults to outliers we run quantile regressions and robust regressions

Table 6 contains our results The rst four columns are for all topexecutives while the last four columns focus on CEOs Columns (1) and(2) present estimates from robust regression for the percentage change insalary and the percentage change in total compensation (salary bonusstock grants and stock-option grants) In the salary growth regressionsfor all executives the coefcient on million is negative and signicantindicating that there is an underlying relationship between the level ofexecutive compensation and its subsequent growth rate The coefcienton million 3 after is negative as well suggesting that the negativerelationship between the level of salary and its subsequent growth rateintensied slightly after 1994 The relatively small magnitude suggeststhat an executive with a $1 million dollar salary would see his salarygrow at an annual rate that would be about 06 percent slower than anexecutive earning $500000

15 The HallndashLiebman data set contains only a combined salary and bonus variable makingit impossible to use it to analyze this issue16 We also ran the regressions (not reported) allowing for a nonlinear relationship betweenmillion and compensation growth and obtained substantively similar results

TA

BL

E6

The

Eff

ect

ofth

eM

illi

on-D

olla

rR

ule

onSa

lary

and

Tot

alC

ompe

nsat

ion

All

top

exec

uti

ves

CE

Os

only

Rob

ust

regr

essi

onQ

uan

tile

regr

essi

onR

obu

stre

gres

sion

Qu

anti

lere

gres

sion

D

in

Din

D

in

Din

D

in

Din

D

in

Din

sala

ryto

talc

omp

(a)

sala

ryto

talc

omp

sa

lary

tota

lcom

p

sala

ry

tota

lcom

p

mil

lio

n(b

)2

006

02

035

72

006

32

024

12

008

72

023

12

010

32

024

5(0

005

)(0

032

)(0

006

)(0

027

)(0

010

)(0

073

)(0

011

)(0

056

)m

illi

on

32

001

10

026

20

021

20

015

20

016

014

92

003

40

158

aft

er(c

)(0

005

)(0

033

)(0

006

)(0

029

)(0

010

)(0

076

)(0

011

)(0

057

)N

359

5035

950

359

5035

950

757

17

589

757

17

589

Stan

dar

der

rors

inp

aren

thes

esR

egre

ssio

nsal

soin

clu

de

year

du

mm

ies

stoc

km

arke

tre

turn

and

log

mar

ket

valu

e

(a)

Tot

alco

mp

ensa

tion

incl

ud

essa

lary

bon

us

rest

rict

edst

ock

and

stoc

k-op

tion

gran

ts

(b)

mil

lion

isd

ene

das

min

(1s

alar

yin

pre

viou

sye

ars

div

ided

by$1

mil

lion)

(c)

aft

eris

anin

dic

ator

vari

able

equ

alto

zero

befo

reth

ere

gula

tion

took

plac

e(1

993)

and

one

afte

rwar

d

36 Hall amp Liebman

In the second column the coefcient on million 3 after is essentiallyzero suggesting that any decrease in salary brought about by section162(m) was offset by increases in bonus and stock-option grants Theresults are similar for the median regressions There is some evidence ofa very small slowdown in growth of salary for executives likely to beaffected by section 162(m) but no sign of a change in total compensa-tion Taken together this provides evidence of a very small substitutionof performance-related pay for salary

The last four columns run identical tests but limit the sample to CEOsThe interaction of million 3 after continues to be small and negative inthe salary regressions In contrast the impact of 162(m) on total compen-sation appears to be positive and fairly substantial (about 15 percentagepoints per year) in these regressions That is the increase in bonus andstock options following the million-dollar rule more than offset the verysmall decline in salary

All of these conclusions should be interpreted somewhat cautiouslybecause our data provide us with only one pre-1993 control year Never-theless the data are consistent with section 162(m) having led to a veryminor slowdown in salary growth and one that is dramatically smallerthan what one would conclude without controlling for underlying differ-ential salary growth rates at different levels of salary Since total compen-sation for CEOs did not decline and perhaps increased there is evidenceof a minor substitution of performance-related pay for salary in responseto the regulation

8 EXECUTIVE COMPENSATION AND THEELASTICITY OF TAXABLE INCOME

Even if taxes have only a minor effect on executive compensation thestructure of executive compensation is important for tax policy be-cause high-income taxpayers are sources of signicant revenue and arethe focus of important debates about the efciency cost of taxation

Goolsbee (1999) has recently argued that executive decisions to exer-cise options are highly responsive to intertemporal differences in taxrates created by tax reforms He claims that much of the apparent declinein taxable income for high-income taxpayers between 1992 and 1993documented by Feldstein and Feenberg (1996) can be explained by execu-tives shifting option gains into 1992 in order to avoid the higher marginaltax rates that went into effect in 1993 By extension he implies that thehigh elasticities of taxable income with respect to personal tax ratesestimated by comparing the pre- and post-tax-reform taxable income of

Taxation of Executive Compensation 37

high-income individuals reect intertemporal shifting of income and notpermanent effects of taxation If this is true then the deadweight loss oftaxation is lower as is the cost of progressivity

While Goolsbee presents comprehensive evidence that option gainswere unusually high in 1992 his identication of the impact of tax rates onthe timing of option gains depends on correctly specifying a counter-factual level of options that would have been exercised in the absence ofthe tax change He simply assumes a linear time trend The purpose ofGoolsbeersquos time trend is presumably to account for the increasing use ofoptions over time However since options typically require a vestingperiod of at least a couple of years a much more direct measure of ex-pected option gains can be constructed based on the total value of vestedoptions held by the CEO Since options typically have vesting periods thatexceed two years this total can be treated as exogenous since it will not beaffected by the tax changes

More generally the value of options held by the CEO is likely to havea major effect on his exercise decisions If his options are out of themoney he will not have gains to realize If there has been a recent run-up in the market an optimizing CEO may exercise an unusually largeamount of options in order to diversify his overall portfolio and if in-the-money options are about to expire they will be exercised in thecurrent year

Our evidence suggests that much of the apparent tax shifting caninstead be attributed to stock market performance and the timing of pastoption grants Using our 15-year panel we replicate Goolsbeersquos (1999)result for the 1993 period but show that the impact of taxes is notpresent in other time periods Moreover even the result for the 1993period disappears when appropriate controls for past option grants andstock price appreciation are included

The top three rows of Table 7 present data on the taxable incomes ofCEOs in the HallndashLiebman sample For these results our panel has beennarrowed to include only CEOs whose rmrsquos scal years correspondwith tax years In addition we create a balanced panel for each of thethree tax-reform periods by excluding any rm for which there are notcomplete data for the period surrounding the reform

The fourth row of the table shows that the 1993 tax act raised the topmarginal personal income tax rate from 310 to 396 percent effective fortax year 1993 with a further increase occurring in 1994 due to the uncap-ping of the Medicare payroll tax The top row of the table shows thattaxable income was higher in 1992 than for any other year in our sample(all data are in 1994 dollars) Option gains were particularly high in1992mdashmore than double their 1991 level and 68 percent higher than their

TA

BL

E7

Tax

Ref

orm

Sto

ckM

arke

tP

erfo

rman

cea

ndth

eT

imin

gof

Opt

ion

Gai

ns

Ave

rage

stoc

km

arke

tM

ean

CE

OM

ean

CE

Ore

turn

for

Mea

nva

lue

taxa

ble

sala

ryan

dM

ean

CE

OT

opfe

der

al

rms

inof

opti

ons

(Val

ue

ofop

tion

Sam

ple

inco

me

bonu

sop

tion

gain

sp

erso

nal

sam

ple

for

held

atbe

gga

ins)

(va

lue

Ref

orm

size

Yea

r($

mil

)($

mil

)($

mil

)M

TR

pre

vye

ar(

)of

year

ofop

tion

she

ld)

ER

TA

8127

919

800

874

064

80

163

7025

80

460

354

1981

083

10

672

009

869

245

070

61

3919

820

872

067

90

099

5011

90

686

144

1983

099

00

724

025

250

355

116

62

16

TR

A86

295

1985

119

70

837

024

050

76

116

02

0719

861

448

091

70

345

5042

81

776

194

1987

165

50

959

059

338

519

12

229

266

1988

163

21

058

039

028

07

237

71

6419

891

656

105

20

414

2817

92

345

176

1993

272

1990

209

51

028

061

828

232

288

02

15T

axA

ct19

911

875

102

00

547

312

01

268

82

0319

922

678

107

01

250

3146

64

552

275

1993

232

41

209

074

339

620

64

299

173

1994

196

41

262

037

842

515

45

113

074

Sam

ple

isH

allndash

Lie

bman

rm

sw

ith

Dec

embe

r

scal

year

sth

atha

dco

mp

lete

dat

afo

ron

eof

the

thre

eta

x-re

form

per

iod

sst

ud

ied

Taxation of Executive Compensation 39

1994 level These patterns are consistent with Goolsbeersquos claim that tax-payers responded to the anticipated increase in marginal tax rates byshifting gains into 1992

The evidence from the other two tax reforms is less clear suggestingthat there might be factors other than tax rates determining the timing ofoption gains ERTA 1981 reduced the top marginal tax rate from 70percent in 1980 to 50 percent in 1982 However there is no sign thatoption gains were reduced in 1980 and 1981 and then increased in 198217

There was a big increase in option gains in 1983 however Similarly thehighest year of stock-option gains in the period around 1986 was 1987 astrange year to take gains since taxpayers would have known that thepersonal rate would be 10 percentage points lower if they waited untilthe following year18

What all the peak years of option gains have in common is that theyfollow years of strong stock market performance The fth row of Table 7shows the average stock market return during the previous year for therms in our sample The large value of option gains in 1983 followed a355-percent increase in the stock market The 1992 boom in option gainsfollowed a 466-percent increase in the stock market the more modest1987 peak in option gains followed a 19-percent gain (and a 43-percentgain two years before)

Strong stock market performance raises the value of options availableto exercise The sixth row of the table shows the mean value of optionsheld by CEOs at the beginning of the year All three of the peak years ofoption gains correspond with peak years in terms of the value of optionsavailable for exercise Since stock market appreciation during the yearalso increases the value of options available to exercise later in that yearwe need to allow for current-year stock market appreciation as well

Table 8 contains Goolsbee-style regressions for each of the tax re-forms The rst two columns reprint the results from Goolsbee (1999)The third and fourth columns replicate his results in our sample Inparticular with a specication similar to his we are able to replicate hisnding that the current after-tax share has a positive coefcient and thefollowing yearrsquos after-tax share has a coefcient of roughly negative oneColumns (5) through (8) show that this pattern was unique to the 1993tax reform Indeed in the TRA 86 regressions the coefcient on the

17 As we explained above for executives covered by the maximum tax on earned incomethere was not a decline in tax rates between 1981 and 198218 It is possible that the large stock market decline in October 1997 may have promptedsome people to exercise options earlier than they otherwise would have However anincrease in option exercises typically follows stock market increases not declines

TA

BL

E8

Att

empt

sto

Rep

lica

teth

eT

ax-S

hift

ing

Res

ult

inO

ther

Tim

eP

erio

ds

Goo

lsbe

eG

ools

bee

1993

1993

Ent

ire

Ent

ire

regr

essi

onre

gres

sion

refo

rmre

form

TR

A86

TR

A86

ER

TA

81E

RT

A81

1980

ndash199

419

80ndash1

994

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

ln[1

2ta

x(t)

]1

244

108

20

455

032

62

017

62

017

22

016

72

016

70

026

20

068

(01

34)

(01

27)

(05

37)

(05

42)

(02

10)

(02

09)

(00

90)

(00

91)

(00

67)

(00

65)

ln[1

2ta

x(t1

1)]

20

867

20

764

21

214

20

917

047

80

722

20

103

004

30

176

018

3(0

118

)(0

112

)(0

491

)(0

499

)(0

157

)(0

174

)(0

072

)(0

072

)(0

068

)(0

066

)

Tim

e0

135

007

62

000

52

036

40

041

000

80

094

003

50

056

002

7(0

008

)(0

008

)(0

049

)(0

049

)(0

028

)(0

028

)(0

025

)(0

026

)(0

003

)(0

003

)

ln(m

arke

tva

lue)

066

10

489

023

70

334

039

5(0

015

)(0

058

)(0

039

)(0

042

)(0

017

)

Yea

rs19

91ndash1

995

1991

ndash199

519

90ndash1

994

1990

ndash199

419

85ndash1

989

1985

ndash198

919

80ndash1

983

1980

ndash198

319

80ndash1

994

1980

ndash199

4

Ad

just

edR

20

720

770

610

630

690

760

800

810

630

68

Obs

erva

tion

s13

856

132

371

385

136

41

470

143

31

113

106

23

968

385

9

Dep

end

ent

vari

able

isth

elo

gof

taxa

ble

inco

me

All

regr

essi

ons

con

tain

rm

xe

def

fect

sG

ools

bee

regr

essi

on(2

)al

soco

ntai

nstw

oot

her

vari

able

sno

tsh

own

inth

eta

ble

Stan

dar

der

rors

inp

aren

thes

es

Taxation of Executive Compensation 41

TABLE 9Alternative Explanations for the Timing of Option Exercises

1993 Entirereform TRA 86 ERTA 81 1980ndash1994

(1) (2) (3) (4)

ln[12tax(t)] 0108 0111 20051 0201(0324) (0136) (0048) (0057)

ln[12tax(t11)] 20134 0573 0125 0150(0336) (0156) (0053) (0066)

ln(market value) 0418 0177 0338 0418(0061) (0039) (0046) (0163)

Firmrsquos stock market return (t) 0298 0268 0214 0258(0044) (0045) (0031) (0023)

Firmrsquos stock market return 0162 0206 0031 0128(t21) (0041) (0008) (0029) (0021)

ln(value of options held by 0004 0008 0001 0009CEO at start of year) (0003) (0003) (0002) (0001)

Years 1990ndash1994 1985ndash1989 1980ndash1983 1980ndash1994Adjusted R2 064 071 082 068Observations 1349 1427 1036 3812

The dependent variable is the log of taxable income All regressions contain rm xed effects Stan-dard errors in parentheses

following yearrsquos tax rate is positive and signicant Columns (9) and (10)pool all three reforms and fail to replicate the pattern that would reectintertemporal shifting19

It is possible that the ease with which one can shift the timing ofoption gains is asymmetric It may be harder to shift option gains for-ward in time since some options expire and therefore must be exercisednow In addition options were a much smaller portion of compensationin the early 1980s so perhaps the ERTA 81 results are not surprisingNonetheless these results cast doubt on the proposition that this is thechannel through which spurious estimates of the elasticity of taxableincome have occurred for previous tax reforms and they suggest thatwe should look harder for alternative explanations for the timing ofoption gains

Table 9 attempts to predict options gains using the lagged and currentincrease in a rmrsquos market value and the value of the options held by theCEO at the start of the year The two stock market variables stronglypredict option gains while the value of options held has the correct sign

19 The results also fail to suggest large permanent effects of marginal tax rates on taxableincome

42 Hall amp Liebman

in all of the specications and is signicant in the pooled sample eventhough it is highly correlated with the two stock market return variablesIn contrast the tax-timing explanation fails even in the 1993 period

It is important to realize that our results do not invalidate Goolsbeersquoscritique of claims that aggregate revenue data from 1992 and 1993 provethat elasticities of taxable income are large We have simply given analternative explanation for why taxable incomes were unusually high in1992 However our results do cast doubt on the conclusion that largetiming shifts are ubiquitous and capable of explaining away the morecareful estimates of taxable income elasticities that have been performedusing micro data (eg Feldstein 1995 Auten and Carroll 1995)

One further point is in order It is sometimes argued that the combina-tion of increased use of options and the booming stock market explainswhy federal tax revenues have been so much higher than expected dur-ing the 1990s However while large option gains increase revenue fromthe personal income tax they produce offsetting deductions for cor-porations The net impact on revenue is only the difference in the twomarginal tax rates times the option gain While the gap between thepersonal rate and the effective corporate rate is nontrivial the impact oflarge option exercises on total revenue is substantially smaller than theamount by which the exercises inate personal tax revenue and reducecorporate tax revenue

9 CONCLUSION

We have described the tax rules for executive pay in detail and analyzedhow changes in various tax rates affect the tax advantages of stock optionsrelative to salary and bonus We nd that there is a moderate tax advan-tage to optionsmdashcurrently about $4 per $100 in pretax executive paymdashand that changes in the tax advantage of options over time have had atmost a modest impact on the composition of pay Corporate governancefactors particularly the role of large investors appear to be more impor-tant in explaining the dramatic increase in option pay

In addition our evidence suggests that more direct attempts to use taxpolicy to inuence executive compensation have had little effect Wend that the million-dollar rule led companies to substitute perfor-mance-based pay for salary But our evidence suggests that this substitu-tion was quite modest and there is no evidence that the total level of paywas reduced Overall although the stock-option explosion has dramati-cally increased the link between pay and performance this change isdue almost entirely to nontax factors

Taxation of Executive Compensation 43

REFERENCES

Auerbach Alan J (1996) ldquoThe Dynamic Effects of Tax Law Asymmetriesrdquo Re-view of Economic Studies 53451ndash405

mdashmdashmdash and James Poterba (1987) ldquoTax Loss Carryforwards and Corporate TaxIncentivesrdquo In The Effects of Taxation on Capital Accumulation M Feldstein(ed) Chicago University of Chicago Press

Altshuler Rosanne and Alan Auerbach (1990) ldquoThe Signicance of Tax LawAsymmetries An Empirical Investigationrdquo Quarterly Journal of Economics 10561ndash86

Auten Gerald and Robert Carroll (1995) ldquoThe Effect of Income Taxes on House-hold Behaviorrdquo Ofce of Tax Analysis US Department of the Treasury

Baker George P and Brian J Hall (1998) ldquoCEO Incentives and Firm SizerdquoNBER Working Paper no 6868 December

Core John and Wayne Guay (1999) ldquoThe Use of Equity Grants to ManageOptimal Equity Incentive Levelsrdquo Draft

Feldstein M (1995) ldquoThe Effect of Marginal Tax Rates on Taxable Income APanel Study of the 1986 Tax Reform Actrdquo Journal of Political Economy 103 (no3)551ndash72

mdashmdashmdash and D Feenberg (1996) ldquoThe Effect of Increased Tax Rates on TaxableIncome and Economic Efciency A Preliminary Analysis of the 1993 Tax RateIncreasesrdquo In Tax Policy and the Economy vol 10 James M Poterba (ed)Cambridge MA MIT Press

Gompers Paul A and Andrew Metrick (1998) ldquoInstitutional Investors andEquity Pricesrdquo NBER Working Paper no 6723

Goolsbee Austan (1999) ldquoWhat Happens When You Tax the Rich Evidencefrom Executive Compensationrdquo NBER Working Paper no 6333

Graham John R (1996) ldquoProxies for the Corporate Marginal Tax Raterdquo Journal ofFinancial Economics 42187ndash221

Hall Brian J (1999) ldquoThe Design of Multi-year Stock Option Plansrdquo Journal ofApplied Corporate Finance 12(2)97ndash106

mdashmdashmdash and Jeffrey B Liebman (1998) ldquoAre CEOs Really Paid Like Bureau-cratsrdquo Quarterly Journal of Economics CXII(no 3)653ndash691

Jensen Michael C and William M Meckling (1976) ldquoTheory of the Firm Mana-gerial Behavior Agency Costs and Ownership Structurerdquo Journal of FinancialEconomics III305ndash360

mdashmdashmdash and Kevin J Murphy (1990) ldquoPerformance Pay and Top-ManagementIncentivesrdquo Journal of Political Economy XCVIII225ndash264

Kaplan Steven N (1997) ldquoThe Evolution of US Corporate Governance WeAre All Henry Kravis Nowrdquo Draft

Livingstone Jane R (1999) ldquoExecutive Compensation Contracting and Re-sponses to Increased Tax Costsrdquo Louisiana State University June

Miller Merton H and Myron S Scholes (1982) ldquoExecutive CompensationTaxes and Incentivesrdquo In Financial Economics Essays in Honor of Paul CootnerWilliam F Sharpe and Cathryn M Cootner (eds) Princeton NJ Prentice-Hall

Murphy Kevin J (1999) ldquoExecutive Compensationrdquo forthcoming in OrleyAshenfelter and David Card (eds) Handbook of Labor Economics Vol 3 NorthHolland

Perry Todd and Marc Zenner (1999) ldquoPay for Performance Government Regu-

Taxation of Executive Compensation 19

avoiding the capital gains taxes is greater The tax advantage of optionstherefore declines linearly as the personal tax rate rises At a 100-percentpersonal tax rate the executive receives nothing in either case and byconstruction the companyrsquos tax deduction benet is the same in NPVterms so the tax advantage of options falls to zero also6

433 The Tax Advantage of Options Changing the Capital Gains TaxRate The tax advantage of options for various capital gains rates isdepicted in the bottom panel of Figure 2 under the same assumptions(this time holding both the personal rate and the corporate rate at 40percent while varying the capital gains rate) The tax advantage of avoid-ing the capital gains tax increases as the capital gains tax rate rises

5 HOW HAS THE TAX ADVANTAGE OF OPTIONSCHANGED OVER TIME

The analysis so far has illustrated how the tax advantage of optionsvaries with the changes in tax rates In this section we show how the taxadvantage of options has changed over time as tax rates have changed

51 Top Marginal Tax Rates over TimeThe evolution on the top marginal tax ratesmdashpersonal corporate andcapital gainsmdashfrom 1980 to 1998 is shown in Figure 3 In all cases thetop rate (the rate for taxpayers with the highest incomes) is shownwhich is not always the highest rate since various anomalies (such asclawbacks of exemptions) sometimes lead to marginal tax rates that arehigher than those for the highest income tax payers

The top personal tax rate was 70 percent in 1980 falling to 50 percentin 1982 and 28 percent in 1988 following the 1986 tax act7 The topmarginal tax rate has since risen to 396 percent but is effectively 425percent since there is a 29-percent Medicare surcharge (paid half by theemployer and half by the employee) that has no upper limit The topcorporate tax rate has had only one signicant change since 1980 fallingfrom 46 percent before the 1986 tax act to 34 percent following the act in1988 The top corporate tax rate was increased to 35 percent in 1993 The

6 This is one place where ignoring bond investments has substantive implications Inparticular in the all-bond case the tax advantage of options is nonmonotonicmdashit rises andthen fallsmdashas personal tax rates rise7 During the early 1980s the maximum tax on earned income limited the marginal tax rateon the earnings of high earners so many executives did not face a decline in personal taxrates between 1981 and 1982 The empirical results in this paper are robust to assumingthat executives faced a marginal tax rate of 50 percent in 1980 and 1981

20 Hall amp Liebman

FIGURE 3 Top Marginal Tax Rates since 1980

top capital gains rate has uctuated between 28 percent (approximately)and 20 percent since 1980 and is currently at 20 percent

52 The Evolution of the Tax Advantage of NQSOsThe tax advantage of options relative to cash dened in the same way asin section 4 from 1980 to 1998 is shown in Figure 4 The calculations aredone using the same assumptions as before (ie N 5 10 years and cashis invested in equities) and the statutory top marginal tax rates shown inFigure 3 are used Due to the fall in corporate and personal tax rates thetax advantage of options increased sharply after the Tax Reform Act of1986 This increase in tax advantage has largely been reversed in the1990s as top personal tax rates have crept back up and the capital gainsrate has been reduced

Two key facts emerge from this analysis First there is currently only amoderate tax advantage to standard nonqualied stock optionsmdashon theorder of $4 per $100 of compensation This is because the tax advantagesto the executive of deferring taxes are largely offset by the tax disad-vantages to the company of not being able to deduct option expensesfrom taxable prots until the executive exercises the options Secondalthough the 1986 tax act substantially increased the tax advantage ofoptions more than half of this increase has been reversed in the 1990s

Taxation of Executive Compensation 21

FIGURE 4 Tax Advantage of Options 1980 to 1998

NPV of tax advantage of $100 in options (NQSOs) rather than cash

53 The Tax (Dis)advantage of ISOsSo far the analysis has focused on the tax advantages of NQSOs sincethey are so much more prevalent than ISOs which account for onlyabout 5 percent of option grants The relative scarcity of ISOs can beexplained both by their tax status and by their per-executive cap of$100000 per year

Under what conditions are ISOs tax-preferred to NQSOs For a trans-fer P to an executive the ISO is always tax-preferred by the executive(since the capital gains rate is lower than the personal rate) while theNQSO is always tax-preferred from the companyrsquos perspective (since anISO is not deductible) The key issue however is the relative advan-tages of the two types of options from a global-contracting perspectiveBecause ISOs are not deductible the company is indifferent betweensetting aside P in the form of NQSOs today and setting aside P(1 2 Tc) inthe form of ISOs (Since the NPV of the tax deduction for NQSOs is PTcthe NPV of the payment P is P 2 PTc or P (1 2 Tc))

Equalizing the posttax cost of NQSOs and ISOs we need only look atthe payoff of the executive to determine the condition under which eachis preferred The payoff of NQSOs in N years is

P [1 1 r (1 2 Tc)]N (1 2 Tp) (7)

22 Hall amp Liebman

The same payoff for ISOs is

P (1 2 Tc) [1 1 r (1 2 Tc)]N (1 2 Tcg) (8)

ISOs therefore are tax-advantaged if

Tp 2 Tcg Tc 2 Tc Tcg (9)

Intuitively ISOs are tax-advantaged only if their advantage (the differ-ence between the personal rate and the capital gains rate) is largeenough to offset their cost (the disadvantage of not deducting at thecorporate rate) Note however that the condition is not a simple com-parison between the corporate rate and the personal rate minus thecapital gains rate (Tp 2 Tcg Tc)

Figure 5 shows how the tax advantage of NQSOs relative to ISOs haschanged since 1980 as tax rates have changed Unlike the very modestchanges (around 3 to 4 percent) in the relative tax advantages betweenoptions and cash shown in Figure 4 the relative tax advantage betweenNQSOs and ISOs has seen enormous swings When the top personalrate was 70 percent in 1980 ISOs were tax-favored by a margin greaterthan 6 percent However by 1982 (following the 1981 tax acts that dra-matically lowered the personal tax rate) ISOs became tax-disadvantaged

FIGURE 5 Tax Advantage of NQSOs Relative to ISOs over Time

NPV of tax advantage of $100 payment to executive NQSOs minus ISOs

Taxation of Executive Compensation 23

by about 5 percent By 1988 (following the 1986 tax act) ISOs became tax-disadvantaged by about 18 percent a dramatic swing The raising ofpersonal rates coupled with a decrease in the capital gains rate has sincereduced the tax disadvantage of ISOs to about $4 per $100 of pretaxcompensation Although hard data on ISOs are hard to come by practi-tioners (mostly compensation consultants) claim that ISOs were morepopular prior to 1982 when they were tax-advantaged Although ISOsare less disadvantageous from a tax standpoint relative to NQSOs thanthey were in the late 1980s they are still disadvantageous so it is notsurprising that they have not made a signicant resurgence in recentyears

6 EXPLAINING THE INCREASE IN OPTION-BASEDCOMPENSATION

In this section we analyze whether changes in tax rates affect the compo-sition of executive compensation and consider the relative importanceof tax factors and corporate governance factors in explaining the in-creasing share of compensation paid in the form of stock options Thenumerous changes in tax rates that have occurred since 1980 provide anopportunity to assess whether executive compensation arrangementsrespond to tax incentives in the way that the tax-avoidance model sug-gests We test this model using a panel data set of CEOs in largepublicly-traded US corporations Our identication of the tax effectsrelies on time-series variation in personal corporate and capital gainstax rates along with cross-sectional variation in corporate tax rates

61 DataWe use a 15-year panel data set of CEOs in the largest publicly-tradedUS corporations which is described in Hall and Liebman (1998) Thedata set covers the years from 1980 through 1994 and combines CEOcompensation information from corporate proxies and 10-K lings withstock price and stock return information from CRSP and accounting datafrom Compustat In addition some compensation data from the 1970swere collected in order to construct measures of the value of stock op-tions held by the CEOs in the rst period of the sample8

The feature that distinguishes our data from most other CEO datasets is that with our panel of yearly proxy data on option grants optiongains and total options held we are able to calculate the total value of

8 We are grateful to David Yermack for providing us with some of the data for 1984 to 1991See Yermack (1995) for a discussion of these data

24 Hall amp Liebman

all stock options held by the CEO at a point of time More importantlysince we have the details about the stock options held (number exer-cise price time to maturity etc) we can precisely calculate the changein the value of a CEOrsquos stock option holdings for a given change in rmvalue

62 Identication IssuesThe numerous tax changes over our sample period and the specic waysin which changes in personal corporate and capital gains rates arepredicted to affect the composition of CEO compensation give us un-usually rich sources of identication However because many of thechanges in tax rates would be expected to affect all of the CEOs in oursample in a similar way we will need to pay particular attention toseparating out the tax effects from underlying time trends and otherfactors that changed over time Controlling for underlying trends is par-ticularly important in this study because we are focusing on an out-comemdashthe use of stock optionsmdashthat has increased rapidly over thepast two decades and that many practitioners believe was importantlyaffected by nontax considerations Thus the tax effects we are trying toexplain are deviations from a rising trend

We take two steps to try to separate out the tax effects from theunderlying trend First we include key nontax factors that could poten-tially explain the increasing reliance on performance-based compensa-tion It has been suggested that the dramatic increase in incentive-basedpay is the result of the remarkably poor shareholder returns during the1970s which spurred the LBO and takeover movements of the 1980sAccording to this story (Kaplan 1997) shareholders became much morepowerful via the rise of institutional investors even as the LBO andtakeover movements waned Because the inuence of institutional inves-tors is thought to be one of the most important mechanisms of strongcorporate governance we use the share of each companyrsquos stock ownedby institutional investors (dened as institutions with more than $100million under management) as an explanatory variable9 The percentageof shares owned by large institutional investors increased from about 20percent to almost 50 percent in 1994 an upward trend that closelymatches the sharp rise in the share of compensation in the form ofoptions Indeed in our sample the annual average (over all of the rmsin our sample) share of stock owned by institutional investors has acorrelation of 09 with the annual average share of compensation paid inoptions

9 This variable is described in Gompers and Metrick (1998)

Taxation of Executive Compensation 25

In addition we include two other variables that proxy for strongercorporate governance The rst is the size of the board Evidence suggeststhat boards with smaller numbers of directors reduce the inuence of theCEO and better represent shareholder interests (Yermack 1995) Boardsize fell by approximately 15 percent between 1980 and 1994 The secondvariable is the fraction of the rmrsquos board members who are inside direc-tors or gray directors (noninsiders who have business dealings with thecompany) Firms with fewer outsiders are less likely to act in the share-holdersrsquo interests and we would predict that they would be less likely tohave performance-based compensation for their CEOs The average per-centage of inside and gray directors has decreased from about 45 percentin 1980 to about 35 percent in 1994

Second we allow for cross-sectional variation in corporate tax ratesCorporate marginal tax rates vary because rms may be eligible for tax-loss carrybacks and carryforwards investment tax credits and the alter-native minimum tax (see Auerbach 1986 Auerbach and Altshuler1990 Auerbach and Poterba 1987) Because these cross-sectional differ-ences imply that different rms should respond differently to a givenchange in tax rates they potentially provide us with a way to isolatethe tax effects Firm-specic corporate marginal tax rates are notori-ously difcult to calculate from publicly-available data We use a tricho-tomous variable that equals zero if the rm has tax-loss carryforwardsand negative earnings in a given year equals 05 times the statutorycorporate rate if the rm has only one of those conditions and equalsthe statutory rate if the rm has neither loss carryforwards nor negativeearnings10

The cross-sectional variation in the corporate tax rate may not be ex-ogenous Firms that perform poorly could face low marginal tax ratesand also be particularly likely or unlikely to provide performance-basedpay Because most of our specications use xed effects the correlationwe must be concerned about is between changes in rm performanceand changes in compensation We deal with this by controlling forlagged rm stock market performance in our regressions This shouldeliminate spurious correlation between tax rates and compensation thatis jointly caused by rm performance

10 Graham (1996) presents evidence that the trichotomous variable performs nearly as wellas his recommended simulated marginal tax rate Plesko (1999) compares marginal taxrates calculated from Compustat data with actual marginal tax rates from rm tax returnsHe nds that simple binary variables are more reliable measures of the marginal tax ratethan is either the Graham or the trichotomous variable We obtain substantively similarresults when we switch from the trichotomous variable to the binary variables recommendby Plesko

26 Hall amp Liebman

63 ResultsThe dependent variable in our regressions is the fraction of a CEOrsquos totalannual compensation (measured as the BlackndashScholes value of stock-option grants plus salary and bonus) that comes from stock-optiongrants We choose this reduced-form specication because our interest isin the composition of CEO compensation However the factors thataffect the composition of compensation might also affect the levels andit will be important to keep this in mind in interpreting our results

Optimal-contracting theory generally makes predictions about the in-centives provided by the total amount of rm stock and stock optionsowned by the CEO We choose to model the ow of option grants ratherthan the ultimate performance sensitivity of the CEOrsquos compensationand stock and stock-option holdings because we have in mind anadjustment-cost model in which it takes time for a CEOrsquos holdings ofstock and stock options to reach the optimal level11 In our regressionswe include two measures of the current performance sensitivity of theCEOrsquos wealth to rm performance as explanatory variables to reect thedistance of the CEOrsquos contract from the optimal contract

Table 3 presents results in which the tax effects are identied solely byvariation in tax rates over time The fraction of annual compensation paidin stock options is regressed on tax variables corporate governance vari-ables contract-theory variables and variables reecting stock market re-turns Both of the regressions in this table include rm xed effects Incolumn (1) the fraction in annual compensation paid in stock options isregressed on the log difference in payoff (the combined payoff to both therm and the CEO) from receiving compensation in stock options ratherthan salary and bonus where the payoff calculations are based on statu-tory marginal tax rates This variable is dened as the log difference be-tween equation (3) and equation (6) in section 4 The coefcient on the taxvariable is large and statistically signicant at the 95-percent level Thepoint estimate implies that a one-percent increase in the payoff differencebetween stock options and salary and bonus results in a 24-percentage-point increase in the share of compensation paid in stock options

The corporate governance variables all have the predicted sign Firmswith a higher fraction of their shares owned by institutional investorsuse more performance-based pay (stock options) Firms with large corpo-rate boards or a large share of inside directors are less likely to use stockoptions

The two contract-theory variables are the dollar change in CEO wealth

11 See Core and Guay (1999) for evidence that rms adjust in this way

TABLE 3Compensation Regressions that Rely on Variation in

Tax Rates over Time

(1) (2)

Tax variablesln (difference in payoff)(a) 2393 (0349)ln (12statutory personal 20015 (0034)

rate)ln (12statutory cap gains 0330 (0171)

rate)ln (12statutory corporate 0399 (0148)

rate)

Corporate governance variablesFraction of rmrsquos shares 0043 (029) 0052 (0028)

owned by institutionalinvestors

Size of the rmrsquos board of 20002 (001) 20003 (0001)directors

Fraction of directors who 20082 (033) 20092 (0030)are inside or gray

Contract-theory variablesD (CEO wealth) per $1000 20261 (0095) 20247 (0087)

in FV(b)

D (CEO wealth) per 10 in 2583 (189) 2554 (201)FV(c)

Volatility(d) 521 (424) 261 (369)ln (rm market value) 0019 (0006) 0019 (0006)

Stock market return variablesLagged annual rm stock 0006 (0009) 0003 (0008)

market returnLagged annual SampP 500 re- 20035 (0023) 20029 (0023)

turn

Regression diagnosticsTotal sample size 5179 5179Firms 427 427Adjusted R2 0313 0311

Dependent variable share of compensation paid in options Standard errors in parentheses Bothregressions in this table include rm xed effects

(a) Log difference in payoff is the log difference in combined rm and CEO after-tax share betweenpaying compensation in options and paying salary

(b) This variable is the dollar change in CEO wealth per $1000 change in rm market value (seeJensen and Murphy 1990)

(c) This variable in the dollar charge in CEO wealth per 10-percent change in rm value (see Hall andLiebman 1998)

(d) Standard deviation of rm market value in trillions

28 Hall amp Liebman

per $1000 change in rm value (Jensen and Murphy 1990) and thedollar change in CEO wealth per 10-percent change in rm value (Halland Liebman 1998) As discussed in Baker and Hall (1998) these twowealth sensitivity measures reect two different concepts of how closelyaligned the managersrsquo incentives are with the interests of the sharehold-ers In the regression they are meant to measure how far the CEOrsquosexisting contract and ownership of stock and stock options is from theoptimal level The negative coefcients on the two variables suggeststhat rms that are below their optimal pay-to-performance sensitivityare more likely to give more stock options in the current period

The second column replaces the tax variable motivated by deferraladvantages with the three statutory rates The corporate after-tax sharehas a positive coefcient as we would expect because the tax advantageof stock options increases as the corporate rate falls The coefcient onthe personal tax rate is small and statistically insignicant The sign ofthe coefcient on the capital gains rate is the opposite of that predictedby tax-avoidance theory

Note that these results rely on time-series variation alone This is not avery convincing test Since the log difference in payoffs mostly rises overthe time period covered in our sample as does the share of compensa-tion paid in options it is possible that the tax variable is simply reectingother factors that were changing over this time period that we have notincluded in our regression Therefore it is important to look at specica-tions that rely on cross-sectional variation in the tax variables

Table 4 contains specications that rely on cross-sectional variation incorporate tax rates The rst column repeats the specication from col-umn (1) of Table 3 but uses the trichotomous measures of rm corporatetax rates in place of the statutory rate in calculating each rmrsquos logdifference in payoff from paying compensation in options rather than insalary and bonus As in the previous table we nd the positive coef-cient on the log difference in payoff just as the theory suggest How-ever the coefcient estimate is now much smaller suggesting that aone-percent increase in the log difference in payoff produces a 01-percentage-point increase in the share of compensation paid in options

Introducing cross-sectional variation in the dependent variable doesnot eliminate the potentially spurious correlation over time between thetax variable and the share of compensation paid in options The secondcolumn of Table 4 isolates the pooled cross-sectional variation in corpo-rate tax rates by adding time dummies and dropping the rm xedeffects In this specication the coefcient on the tax variable is indistin-guishable from zero The coefcient on the institutional variable contin-ues to be large and statistically signicant The third column includes

Taxation of Executive Compensation 29

both rm xed effects and time dummies which tests for a relationshipbetween within-rm changes in the dependent variable and within-rmchanges in each explanatory variable In this specication the coefcienton the tax variable is again indistinguishable from zero The coefcientson the corporate governance variables are also statistically insignicant

The third column contains a specication that treats the Tax ReformAct of 1986 (TRA 86) as a natural experiment For rms and CEOs facingthe statutory corporate tax rate TRA 86 increased the relative payoff topaying compensation in options by 17 percent For a rm facing a zeromarginal tax rate it reduced the relative payoff by 45 percent The reduc-tion occurred because TRA 86 reduced the personal tax rate makingsalary and bonus more attractive while the decreased corporate rate hadno effect on zero-tax-rate rms Thus we would expect rms facing zeroor low marginal tax rates to reduce their use of options after TRA 86while rms facing the statutory rate would be expected to increase themFew rms face zero marginal tax rates year after year Therefore we tryto distinguish between rms that often face low tax rates and rms thatnearly always face the statutory rate We dene low-tax rms as oneswhose average marginal tax rate over 1984 to 1986 was below 02412

Then we ran a difference-in-differences regression to see if high-taxrms increased their use of options more than low-tax rms The interac-tion between post-TRA 86 and high corporate tax rates is the key vari-able The estimated coefcient on this variable is positive but small andnot statistically different from zero at the 95-percent level

It is important to note that in this specication as well the main corpo-rate governance variable the share of the rmrsquos shares owned by institu-tional investors performs as theory would predict The variable appearsto be quite robust Table 5 presents regressions of the share of compensa-tion paid in stock options on the fraction of the rmrsquos shares owned byinstitutional investors Since we have already established that these vari-ables have a strong longitudinal relationship the regressions here aredesigned to focus on the cross-sectional relationship between these twovariables The rst and third columns present regressions using data onlyfrom 1994 and nd a coefcient of roughly 02 (column 1 contains noother covariates while column 3 includes all of the other nontax covari-ates from the regressions in Tables 3 and 4) Since the median share of armrsquos stock owned by institutional investors in our sample rises from 19percent in 1980 to 49 percent in 1994 this coefcient would imply that this

12 This cutoff was chosen after inspecting the distribution of tax rates and essentiallyseparates rms that are always at or near the statutory rate from ones that are sometimesbelow it

TA

BL

E4

Com

pens

atio

nR

egre

ssio

nsth

atR

ely

onC

ross

-Sec

tion

alV

aria

tion

inC

orpo

rate

Tax

Rat

es

(1)

(2)

(3)

(4)

Tax

vari

able

s(a)

ln

(dif

fere

nce

inp

ayof

f)0

099

(00

41)

20

038

(00

36)

20

005

(00

44)

Pos

t-T

RA

862

000

5(0

013

)H

igh

corp

orat

eta

xra

te0

018

(00

17)

(Pos

t-T

RA

86)

3hi

gh0

024

(00

16)

corp

orat

eta

xra

te

Cor

por

ate

gove

rnan

ceva

riab

les

Frac

tion

of

rmrsquos

shar

es0

109

(00

27)

011

4(0

015

)2

000

9(0

030

)0

087

(00

34)

owne

dby

inst

itu

tion

alin

vest

ors

Size

ofth

e

rmrsquos

boar

dof

20

004

(00

01)

20

001

(00

01)

20

002

(00

01)

000

1(0

002

)d

irec

tors

Frac

tion

ofd

irec

tors

wh

o2

012

4(0

033

)2

000

3(0

018

)2

004

1(0

034

)0

001

(00

39)

are

insi

de

orgr

ay

Con

trac

t-th

eory

vari

able

sD

(CE

Ow

ealt

h)p

er$1

000

20

238

(00

94)

20

167

(00

53)

20

257

(00

96)

20

021

(01

47)

inFV

(b)

D(C

EO

wea

lth)

per

10

in2

560

(17

8)2

592

(10

5)2

594

(19

6)2

147

(04

86)

FV(c

)

Vol

atili

ty(d

)2

48(4

24)

162

(03

4)1

12(4

28)

106

(06

09)

ln(

rmm

arke

tva

lue)

003

3(

006)

000

8(0

003

)0

009

(00

06)

001

2(0

007

)

Stoc

km

arke

tre

turn

vari

able

sL

agge

dan

nual

rm

stoc

k2

000

2(0

009

)0

011

(00

10)

000

6(0

009

)2

001

2(0

013

)m

arke

tre

turn

Lag

ged

annu

alSamp

P50

02

005

6(0

023

)2

005

3(0

051

)2

003

9(0

049

)2

015

4(0

040

)re

turn

Reg

ress

ion

dia

gnos

tics

T

otal

sam

ple

size

5179

5179

5179

2534

Firm

s42

742

739

6A

dju

sted

R2

031

009

032

005

Firm

xed

effe

cts

Yes

No

Yes

No

Tim

ed

um

mie

sN

oY

esY

esN

o

Dep

end

ent

vari

able

isfr

acti

onof

ann

ualc

omp

ensa

tion

from

stoc

kop

tion

sSt

and

ard

erro

rsin

par

enth

eses

(a)

All

thre

esp

eci

cati

ons

use

tric

hoto

mou

sco

rpor

ate

tax

rate

sT

he

sam

ple

inco

lum

n(3

)u

ses

only

the

year

s19

84to

1988

sin

cew

ear

eu

sing

1986

asa

nat

ural

exp

erim

ent

(b)

Thi

sva

riab

leis

the

dol

lar

chan

gein

CE

Ow

ealt

hp

er$1

000

chan

gein

rm

mar

ketv

alu

e(s

eeJe

nse

nan

dM

urp

hy

1990

)

(c)

Thi

sva

riab

lein

the

dol

lar

char

gein

CE

Ow

ealt

hp

er10

-per

cent

chan

gein

rm

valu

e(s

eeH

alla

nd

Lie

bman

199

8)

(d)

Stan

dar

dd

evia

tion

ofr

mm

arke

tva

lue

intr

illio

ns

32 Hall amp Liebman

TABLE 5Large Institutional Investors and Stock-Option Compensation

The Cross-sectional Relationship between Institutional Investors andOption Grants

(1) (2) (3) (4)

Coefcient on fraction of 217 (067) 145 (014) 177 (069) 120 (015)rmrsquos shares owned byinstitutional investors

Sample period 1994 1980ndash1994 1994 1980ndash1994Contract-theory and stock No No Yes Yes

market covariatesTime dummies No Yes No Yes

The dependent variable is the share of compensation paid in options

factor alone can explain 6 percentage points of the 23-percentage-pointincrease in the median share of compensation paid in options that oc-curred over this time period Of course this variable is only one measureof the strengthening of corporate governance so the total impact of im-proved corporate governance could be greater The second and fourthcolumns of the table use the entire 1980ndash1994 sample and include timedummies to isolate the cross-sectional correlations Again there is evi-dence of a strong cross-sectional relationship between the two variablesalthough the coefcients are slightly smaller perhaps reecting a weakercorporate-governance role in the earlier period In sum there is a strongrelationship in both longitudinal and cross-sectional data between theinstitutional-investor variable and the options-percentage variable13

7 THE MILLION-DOLLAR RULE

In 1993 section 162(m) of the Internal Revenue Code was enactedlimiting the deductibility of executive compensation in excess of onemillion dollars unless the compensation was performance-related Pro-ponents of this legislation argued that it would slow the growth ofexecutive pay and tighten the link between rm performance and execu-tive compensation

The new provision became effective for tax year 1994 and applies tothe CEO and the other four most highly compensated executives in eachcompany Pay that is performance-related such as stock-option grants

13 The one specication in which the fraction owned by institutional investors is notsignicant is in column (2) of Table 4 a regression that includes both xed effects and timeeffects

Taxation of Executive Compensation 33

and bonuses paid for meeting clear performance goals is not affected bythis regulation provided that it has been approved by shareholdersThus the main impact of the provision is to limit the deductibility ofsalary in excess of a million dollars Woodlock and Antenucci (1997)studied the proxy statements of 376 rms and documented that mostrms responded to the new law by qualifying top executive pay asperformance-related Perry and Zenner (1999) identied 25 rms thatreduced salaries from above $1 million to below $1 million and foundthat 23 of the rms cited section 162(m) as their reason for doing so

In this section of the paper we investigate whether this provision hasaffected the level or structure of executive pay using the Execucompdata base which contains information on the highest-paid executives ineach of the rms in the SampP 500 SampP Midcap 400 and SampP SmallCap600 Between 1993 and 1998 the median salary of the Execucomp execu-tives rose from $230000 to $306000 The median value of total compen-sation (salary bonus and option grants) rose from $470000 to $882000Limiting the sample of CEOs indicates that the total compensation of themedian CEO rose from $11 million to $18 million over the period whilethe median salary rose from $441000 to $550000 Thus executive com-pensation clearly continued to rise at a rapid rate after the implementa-tion of the million-dollar rule though perhaps at a slower rate than itotherwise would have

If the regulation affects behavior it should have the largest impact onrms whose executives are receiving salaries that are above the limit orjust below it Thus we might expect to see rms whose executives arereceiving relatively high salaries to rely more heavily on performance-based pay and less heavily on salary in the years following the regulationPerry and Zenner (1999) and Livingston (1999) both nd that this is thecase This trend can be clearly identied in the Execucomp data set Forexample over 1993 to 1998 the median annualized growth rate of salaryfor executives whose 1993 salary was below $500000 in the previous yearwas 85 percent For executives with salary between $500000 and$700000 it was 64 percent for executives between $700000 and $850000it was 50 percent for executives between $750000 and $1 million it was31 percent and for executives above $1 million it was 00 percent14

The observation that executives at lower levels of salary had highergrowth rates is not very strong evidence of an impact of section 162(m)however because it is possible that low-salary executives would havehad higher growth rates even in the absence of the provision To explore

14 Of the 3137 executives for whom we have complete data between 1993 and 1998 2708had 1993 salaries below $500000 while only 35 had 1993 salaries above $1 million

34 Hall amp Liebman

whether this is the case we use Execucomp data from both before andafter the law change Ideally we would want to have data from manyyears before the change to examine whether there is an underlying pat-tern of faster salary growth for low-salary executives Unfortunately theExecucomp data begin in 1992 and therefore provide us with only the1992ndash1993 growth rate in compensation to use as a baseline15

We run regressions pooling data from 1992ndash1993 through 1997ndash1998of the annual percentage change in different forms of compensation on avariable million that is designed to measure the likelihood that anexecutiversquos compensation will be affected by section 162(m) We denemillion as the minimum of 1 and the executiversquos salary divided by $1million in the previous year Thus any executive with salary at or above$1 million would receive a value of 1 while an executive with a salary of$500000 would be coded as a 0516 We include this variable by itself andalso interacted with a dummy variable (after) that equals 1 for years1994 and after Thus the million variable is designed to pick up anyunderlying relationship between the level of salary and the growth rateof salary while the interacted one will identify any differential growthrate of salary by income in the period after section 162(m) took effect Wealso include a full set of year dummies the natural logarithm of thermrsquos market value and the rmrsquos annual rate of return over the twoprevious years as control variables To reduce the sensitivity of our re-sults to outliers we run quantile regressions and robust regressions

Table 6 contains our results The rst four columns are for all topexecutives while the last four columns focus on CEOs Columns (1) and(2) present estimates from robust regression for the percentage change insalary and the percentage change in total compensation (salary bonusstock grants and stock-option grants) In the salary growth regressionsfor all executives the coefcient on million is negative and signicantindicating that there is an underlying relationship between the level ofexecutive compensation and its subsequent growth rate The coefcienton million 3 after is negative as well suggesting that the negativerelationship between the level of salary and its subsequent growth rateintensied slightly after 1994 The relatively small magnitude suggeststhat an executive with a $1 million dollar salary would see his salarygrow at an annual rate that would be about 06 percent slower than anexecutive earning $500000

15 The HallndashLiebman data set contains only a combined salary and bonus variable makingit impossible to use it to analyze this issue16 We also ran the regressions (not reported) allowing for a nonlinear relationship betweenmillion and compensation growth and obtained substantively similar results

TA

BL

E6

The

Eff

ect

ofth

eM

illi

on-D

olla

rR

ule

onSa

lary

and

Tot

alC

ompe

nsat

ion

All

top

exec

uti

ves

CE

Os

only

Rob

ust

regr

essi

onQ

uan

tile

regr

essi

onR

obu

stre

gres

sion

Qu

anti

lere

gres

sion

D

in

Din

D

in

Din

D

in

Din

D

in

Din

sala

ryto

talc

omp

(a)

sala

ryto

talc

omp

sa

lary

tota

lcom

p

sala

ry

tota

lcom

p

mil

lio

n(b

)2

006

02

035

72

006

32

024

12

008

72

023

12

010

32

024

5(0

005

)(0

032

)(0

006

)(0

027

)(0

010

)(0

073

)(0

011

)(0

056

)m

illi

on

32

001

10

026

20

021

20

015

20

016

014

92

003

40

158

aft

er(c

)(0

005

)(0

033

)(0

006

)(0

029

)(0

010

)(0

076

)(0

011

)(0

057

)N

359

5035

950

359

5035

950

757

17

589

757

17

589

Stan

dar

der

rors

inp

aren

thes

esR

egre

ssio

nsal

soin

clu

de

year

du

mm

ies

stoc

km

arke

tre

turn

and

log

mar

ket

valu

e

(a)

Tot

alco

mp

ensa

tion

incl

ud

essa

lary

bon

us

rest

rict

edst

ock

and

stoc

k-op

tion

gran

ts

(b)

mil

lion

isd

ene

das

min

(1s

alar

yin

pre

viou

sye

ars

div

ided

by$1

mil

lion)

(c)

aft

eris

anin

dic

ator

vari

able

equ

alto

zero

befo

reth

ere

gula

tion

took

plac

e(1

993)

and

one

afte

rwar

d

36 Hall amp Liebman

In the second column the coefcient on million 3 after is essentiallyzero suggesting that any decrease in salary brought about by section162(m) was offset by increases in bonus and stock-option grants Theresults are similar for the median regressions There is some evidence ofa very small slowdown in growth of salary for executives likely to beaffected by section 162(m) but no sign of a change in total compensa-tion Taken together this provides evidence of a very small substitutionof performance-related pay for salary

The last four columns run identical tests but limit the sample to CEOsThe interaction of million 3 after continues to be small and negative inthe salary regressions In contrast the impact of 162(m) on total compen-sation appears to be positive and fairly substantial (about 15 percentagepoints per year) in these regressions That is the increase in bonus andstock options following the million-dollar rule more than offset the verysmall decline in salary

All of these conclusions should be interpreted somewhat cautiouslybecause our data provide us with only one pre-1993 control year Never-theless the data are consistent with section 162(m) having led to a veryminor slowdown in salary growth and one that is dramatically smallerthan what one would conclude without controlling for underlying differ-ential salary growth rates at different levels of salary Since total compen-sation for CEOs did not decline and perhaps increased there is evidenceof a minor substitution of performance-related pay for salary in responseto the regulation

8 EXECUTIVE COMPENSATION AND THEELASTICITY OF TAXABLE INCOME

Even if taxes have only a minor effect on executive compensation thestructure of executive compensation is important for tax policy be-cause high-income taxpayers are sources of signicant revenue and arethe focus of important debates about the efciency cost of taxation

Goolsbee (1999) has recently argued that executive decisions to exer-cise options are highly responsive to intertemporal differences in taxrates created by tax reforms He claims that much of the apparent declinein taxable income for high-income taxpayers between 1992 and 1993documented by Feldstein and Feenberg (1996) can be explained by execu-tives shifting option gains into 1992 in order to avoid the higher marginaltax rates that went into effect in 1993 By extension he implies that thehigh elasticities of taxable income with respect to personal tax ratesestimated by comparing the pre- and post-tax-reform taxable income of

Taxation of Executive Compensation 37

high-income individuals reect intertemporal shifting of income and notpermanent effects of taxation If this is true then the deadweight loss oftaxation is lower as is the cost of progressivity

While Goolsbee presents comprehensive evidence that option gainswere unusually high in 1992 his identication of the impact of tax rates onthe timing of option gains depends on correctly specifying a counter-factual level of options that would have been exercised in the absence ofthe tax change He simply assumes a linear time trend The purpose ofGoolsbeersquos time trend is presumably to account for the increasing use ofoptions over time However since options typically require a vestingperiod of at least a couple of years a much more direct measure of ex-pected option gains can be constructed based on the total value of vestedoptions held by the CEO Since options typically have vesting periods thatexceed two years this total can be treated as exogenous since it will not beaffected by the tax changes

More generally the value of options held by the CEO is likely to havea major effect on his exercise decisions If his options are out of themoney he will not have gains to realize If there has been a recent run-up in the market an optimizing CEO may exercise an unusually largeamount of options in order to diversify his overall portfolio and if in-the-money options are about to expire they will be exercised in thecurrent year

Our evidence suggests that much of the apparent tax shifting caninstead be attributed to stock market performance and the timing of pastoption grants Using our 15-year panel we replicate Goolsbeersquos (1999)result for the 1993 period but show that the impact of taxes is notpresent in other time periods Moreover even the result for the 1993period disappears when appropriate controls for past option grants andstock price appreciation are included

The top three rows of Table 7 present data on the taxable incomes ofCEOs in the HallndashLiebman sample For these results our panel has beennarrowed to include only CEOs whose rmrsquos scal years correspondwith tax years In addition we create a balanced panel for each of thethree tax-reform periods by excluding any rm for which there are notcomplete data for the period surrounding the reform

The fourth row of the table shows that the 1993 tax act raised the topmarginal personal income tax rate from 310 to 396 percent effective fortax year 1993 with a further increase occurring in 1994 due to the uncap-ping of the Medicare payroll tax The top row of the table shows thattaxable income was higher in 1992 than for any other year in our sample(all data are in 1994 dollars) Option gains were particularly high in1992mdashmore than double their 1991 level and 68 percent higher than their

TA

BL

E7

Tax

Ref

orm

Sto

ckM

arke

tP

erfo

rman

cea

ndth

eT

imin

gof

Opt

ion

Gai

ns

Ave

rage

stoc

km

arke

tM

ean

CE

OM

ean

CE

Ore

turn

for

Mea

nva

lue

taxa

ble

sala

ryan

dM

ean

CE

OT

opfe

der

al

rms

inof

opti

ons

(Val

ue

ofop

tion

Sam

ple

inco

me

bonu

sop

tion

gain

sp

erso

nal

sam

ple

for

held

atbe

gga

ins)

(va

lue

Ref

orm

size

Yea

r($

mil

)($

mil

)($

mil

)M

TR

pre

vye

ar(

)of

year

ofop

tion

she

ld)

ER

TA

8127

919

800

874

064

80

163

7025

80

460

354

1981

083

10

672

009

869

245

070

61

3919

820

872

067

90

099

5011

90

686

144

1983

099

00

724

025

250

355

116

62

16

TR

A86

295

1985

119

70

837

024

050

76

116

02

0719

861

448

091

70

345

5042

81

776

194

1987

165

50

959

059

338

519

12

229

266

1988

163

21

058

039

028

07

237

71

6419

891

656

105

20

414

2817

92

345

176

1993

272

1990

209

51

028

061

828

232

288

02

15T

axA

ct19

911

875

102

00

547

312

01

268

82

0319

922

678

107

01

250

3146

64

552

275

1993

232

41

209

074

339

620

64

299

173

1994

196

41

262

037

842

515

45

113

074

Sam

ple

isH

allndash

Lie

bman

rm

sw

ith

Dec

embe

r

scal

year

sth

atha

dco

mp

lete

dat

afo

ron

eof

the

thre

eta

x-re

form

per

iod

sst

ud

ied

Taxation of Executive Compensation 39

1994 level These patterns are consistent with Goolsbeersquos claim that tax-payers responded to the anticipated increase in marginal tax rates byshifting gains into 1992

The evidence from the other two tax reforms is less clear suggestingthat there might be factors other than tax rates determining the timing ofoption gains ERTA 1981 reduced the top marginal tax rate from 70percent in 1980 to 50 percent in 1982 However there is no sign thatoption gains were reduced in 1980 and 1981 and then increased in 198217

There was a big increase in option gains in 1983 however Similarly thehighest year of stock-option gains in the period around 1986 was 1987 astrange year to take gains since taxpayers would have known that thepersonal rate would be 10 percentage points lower if they waited untilthe following year18

What all the peak years of option gains have in common is that theyfollow years of strong stock market performance The fth row of Table 7shows the average stock market return during the previous year for therms in our sample The large value of option gains in 1983 followed a355-percent increase in the stock market The 1992 boom in option gainsfollowed a 466-percent increase in the stock market the more modest1987 peak in option gains followed a 19-percent gain (and a 43-percentgain two years before)

Strong stock market performance raises the value of options availableto exercise The sixth row of the table shows the mean value of optionsheld by CEOs at the beginning of the year All three of the peak years ofoption gains correspond with peak years in terms of the value of optionsavailable for exercise Since stock market appreciation during the yearalso increases the value of options available to exercise later in that yearwe need to allow for current-year stock market appreciation as well

Table 8 contains Goolsbee-style regressions for each of the tax re-forms The rst two columns reprint the results from Goolsbee (1999)The third and fourth columns replicate his results in our sample Inparticular with a specication similar to his we are able to replicate hisnding that the current after-tax share has a positive coefcient and thefollowing yearrsquos after-tax share has a coefcient of roughly negative oneColumns (5) through (8) show that this pattern was unique to the 1993tax reform Indeed in the TRA 86 regressions the coefcient on the

17 As we explained above for executives covered by the maximum tax on earned incomethere was not a decline in tax rates between 1981 and 198218 It is possible that the large stock market decline in October 1997 may have promptedsome people to exercise options earlier than they otherwise would have However anincrease in option exercises typically follows stock market increases not declines

TA

BL

E8

Att

empt

sto

Rep

lica

teth

eT

ax-S

hift

ing

Res

ult

inO

ther

Tim

eP

erio

ds

Goo

lsbe

eG

ools

bee

1993

1993

Ent

ire

Ent

ire

regr

essi

onre

gres

sion

refo

rmre

form

TR

A86

TR

A86

ER

TA

81E

RT

A81

1980

ndash199

419

80ndash1

994

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

ln[1

2ta

x(t)

]1

244

108

20

455

032

62

017

62

017

22

016

72

016

70

026

20

068

(01

34)

(01

27)

(05

37)

(05

42)

(02

10)

(02

09)

(00

90)

(00

91)

(00

67)

(00

65)

ln[1

2ta

x(t1

1)]

20

867

20

764

21

214

20

917

047

80

722

20

103

004

30

176

018

3(0

118

)(0

112

)(0

491

)(0

499

)(0

157

)(0

174

)(0

072

)(0

072

)(0

068

)(0

066

)

Tim

e0

135

007

62

000

52

036

40

041

000

80

094

003

50

056

002

7(0

008

)(0

008

)(0

049

)(0

049

)(0

028

)(0

028

)(0

025

)(0

026

)(0

003

)(0

003

)

ln(m

arke

tva

lue)

066

10

489

023

70

334

039

5(0

015

)(0

058

)(0

039

)(0

042

)(0

017

)

Yea

rs19

91ndash1

995

1991

ndash199

519

90ndash1

994

1990

ndash199

419

85ndash1

989

1985

ndash198

919

80ndash1

983

1980

ndash198

319

80ndash1

994

1980

ndash199

4

Ad

just

edR

20

720

770

610

630

690

760

800

810

630

68

Obs

erva

tion

s13

856

132

371

385

136

41

470

143

31

113

106

23

968

385

9

Dep

end

ent

vari

able

isth

elo

gof

taxa

ble

inco

me

All

regr

essi

ons

con

tain

rm

xe

def

fect

sG

ools

bee

regr

essi

on(2

)al

soco

ntai

nstw

oot

her

vari

able

sno

tsh

own

inth

eta

ble

Stan

dar

der

rors

inp

aren

thes

es

Taxation of Executive Compensation 41

TABLE 9Alternative Explanations for the Timing of Option Exercises

1993 Entirereform TRA 86 ERTA 81 1980ndash1994

(1) (2) (3) (4)

ln[12tax(t)] 0108 0111 20051 0201(0324) (0136) (0048) (0057)

ln[12tax(t11)] 20134 0573 0125 0150(0336) (0156) (0053) (0066)

ln(market value) 0418 0177 0338 0418(0061) (0039) (0046) (0163)

Firmrsquos stock market return (t) 0298 0268 0214 0258(0044) (0045) (0031) (0023)

Firmrsquos stock market return 0162 0206 0031 0128(t21) (0041) (0008) (0029) (0021)

ln(value of options held by 0004 0008 0001 0009CEO at start of year) (0003) (0003) (0002) (0001)

Years 1990ndash1994 1985ndash1989 1980ndash1983 1980ndash1994Adjusted R2 064 071 082 068Observations 1349 1427 1036 3812

The dependent variable is the log of taxable income All regressions contain rm xed effects Stan-dard errors in parentheses

following yearrsquos tax rate is positive and signicant Columns (9) and (10)pool all three reforms and fail to replicate the pattern that would reectintertemporal shifting19

It is possible that the ease with which one can shift the timing ofoption gains is asymmetric It may be harder to shift option gains for-ward in time since some options expire and therefore must be exercisednow In addition options were a much smaller portion of compensationin the early 1980s so perhaps the ERTA 81 results are not surprisingNonetheless these results cast doubt on the proposition that this is thechannel through which spurious estimates of the elasticity of taxableincome have occurred for previous tax reforms and they suggest thatwe should look harder for alternative explanations for the timing ofoption gains

Table 9 attempts to predict options gains using the lagged and currentincrease in a rmrsquos market value and the value of the options held by theCEO at the start of the year The two stock market variables stronglypredict option gains while the value of options held has the correct sign

19 The results also fail to suggest large permanent effects of marginal tax rates on taxableincome

42 Hall amp Liebman

in all of the specications and is signicant in the pooled sample eventhough it is highly correlated with the two stock market return variablesIn contrast the tax-timing explanation fails even in the 1993 period

It is important to realize that our results do not invalidate Goolsbeersquoscritique of claims that aggregate revenue data from 1992 and 1993 provethat elasticities of taxable income are large We have simply given analternative explanation for why taxable incomes were unusually high in1992 However our results do cast doubt on the conclusion that largetiming shifts are ubiquitous and capable of explaining away the morecareful estimates of taxable income elasticities that have been performedusing micro data (eg Feldstein 1995 Auten and Carroll 1995)

One further point is in order It is sometimes argued that the combina-tion of increased use of options and the booming stock market explainswhy federal tax revenues have been so much higher than expected dur-ing the 1990s However while large option gains increase revenue fromthe personal income tax they produce offsetting deductions for cor-porations The net impact on revenue is only the difference in the twomarginal tax rates times the option gain While the gap between thepersonal rate and the effective corporate rate is nontrivial the impact oflarge option exercises on total revenue is substantially smaller than theamount by which the exercises inate personal tax revenue and reducecorporate tax revenue

9 CONCLUSION

We have described the tax rules for executive pay in detail and analyzedhow changes in various tax rates affect the tax advantages of stock optionsrelative to salary and bonus We nd that there is a moderate tax advan-tage to optionsmdashcurrently about $4 per $100 in pretax executive paymdashand that changes in the tax advantage of options over time have had atmost a modest impact on the composition of pay Corporate governancefactors particularly the role of large investors appear to be more impor-tant in explaining the dramatic increase in option pay

In addition our evidence suggests that more direct attempts to use taxpolicy to inuence executive compensation have had little effect Wend that the million-dollar rule led companies to substitute perfor-mance-based pay for salary But our evidence suggests that this substitu-tion was quite modest and there is no evidence that the total level of paywas reduced Overall although the stock-option explosion has dramati-cally increased the link between pay and performance this change isdue almost entirely to nontax factors

Taxation of Executive Compensation 43

REFERENCES

Auerbach Alan J (1996) ldquoThe Dynamic Effects of Tax Law Asymmetriesrdquo Re-view of Economic Studies 53451ndash405

mdashmdashmdash and James Poterba (1987) ldquoTax Loss Carryforwards and Corporate TaxIncentivesrdquo In The Effects of Taxation on Capital Accumulation M Feldstein(ed) Chicago University of Chicago Press

Altshuler Rosanne and Alan Auerbach (1990) ldquoThe Signicance of Tax LawAsymmetries An Empirical Investigationrdquo Quarterly Journal of Economics 10561ndash86

Auten Gerald and Robert Carroll (1995) ldquoThe Effect of Income Taxes on House-hold Behaviorrdquo Ofce of Tax Analysis US Department of the Treasury

Baker George P and Brian J Hall (1998) ldquoCEO Incentives and Firm SizerdquoNBER Working Paper no 6868 December

Core John and Wayne Guay (1999) ldquoThe Use of Equity Grants to ManageOptimal Equity Incentive Levelsrdquo Draft

Feldstein M (1995) ldquoThe Effect of Marginal Tax Rates on Taxable Income APanel Study of the 1986 Tax Reform Actrdquo Journal of Political Economy 103 (no3)551ndash72

mdashmdashmdash and D Feenberg (1996) ldquoThe Effect of Increased Tax Rates on TaxableIncome and Economic Efciency A Preliminary Analysis of the 1993 Tax RateIncreasesrdquo In Tax Policy and the Economy vol 10 James M Poterba (ed)Cambridge MA MIT Press

Gompers Paul A and Andrew Metrick (1998) ldquoInstitutional Investors andEquity Pricesrdquo NBER Working Paper no 6723

Goolsbee Austan (1999) ldquoWhat Happens When You Tax the Rich Evidencefrom Executive Compensationrdquo NBER Working Paper no 6333

Graham John R (1996) ldquoProxies for the Corporate Marginal Tax Raterdquo Journal ofFinancial Economics 42187ndash221

Hall Brian J (1999) ldquoThe Design of Multi-year Stock Option Plansrdquo Journal ofApplied Corporate Finance 12(2)97ndash106

mdashmdashmdash and Jeffrey B Liebman (1998) ldquoAre CEOs Really Paid Like Bureau-cratsrdquo Quarterly Journal of Economics CXII(no 3)653ndash691

Jensen Michael C and William M Meckling (1976) ldquoTheory of the Firm Mana-gerial Behavior Agency Costs and Ownership Structurerdquo Journal of FinancialEconomics III305ndash360

mdashmdashmdash and Kevin J Murphy (1990) ldquoPerformance Pay and Top-ManagementIncentivesrdquo Journal of Political Economy XCVIII225ndash264

Kaplan Steven N (1997) ldquoThe Evolution of US Corporate Governance WeAre All Henry Kravis Nowrdquo Draft

Livingstone Jane R (1999) ldquoExecutive Compensation Contracting and Re-sponses to Increased Tax Costsrdquo Louisiana State University June

Miller Merton H and Myron S Scholes (1982) ldquoExecutive CompensationTaxes and Incentivesrdquo In Financial Economics Essays in Honor of Paul CootnerWilliam F Sharpe and Cathryn M Cootner (eds) Princeton NJ Prentice-Hall

Murphy Kevin J (1999) ldquoExecutive Compensationrdquo forthcoming in OrleyAshenfelter and David Card (eds) Handbook of Labor Economics Vol 3 NorthHolland

Perry Todd and Marc Zenner (1999) ldquoPay for Performance Government Regu-

20 Hall amp Liebman

FIGURE 3 Top Marginal Tax Rates since 1980

top capital gains rate has uctuated between 28 percent (approximately)and 20 percent since 1980 and is currently at 20 percent

52 The Evolution of the Tax Advantage of NQSOsThe tax advantage of options relative to cash dened in the same way asin section 4 from 1980 to 1998 is shown in Figure 4 The calculations aredone using the same assumptions as before (ie N 5 10 years and cashis invested in equities) and the statutory top marginal tax rates shown inFigure 3 are used Due to the fall in corporate and personal tax rates thetax advantage of options increased sharply after the Tax Reform Act of1986 This increase in tax advantage has largely been reversed in the1990s as top personal tax rates have crept back up and the capital gainsrate has been reduced

Two key facts emerge from this analysis First there is currently only amoderate tax advantage to standard nonqualied stock optionsmdashon theorder of $4 per $100 of compensation This is because the tax advantagesto the executive of deferring taxes are largely offset by the tax disad-vantages to the company of not being able to deduct option expensesfrom taxable prots until the executive exercises the options Secondalthough the 1986 tax act substantially increased the tax advantage ofoptions more than half of this increase has been reversed in the 1990s

Taxation of Executive Compensation 21

FIGURE 4 Tax Advantage of Options 1980 to 1998

NPV of tax advantage of $100 in options (NQSOs) rather than cash

53 The Tax (Dis)advantage of ISOsSo far the analysis has focused on the tax advantages of NQSOs sincethey are so much more prevalent than ISOs which account for onlyabout 5 percent of option grants The relative scarcity of ISOs can beexplained both by their tax status and by their per-executive cap of$100000 per year

Under what conditions are ISOs tax-preferred to NQSOs For a trans-fer P to an executive the ISO is always tax-preferred by the executive(since the capital gains rate is lower than the personal rate) while theNQSO is always tax-preferred from the companyrsquos perspective (since anISO is not deductible) The key issue however is the relative advan-tages of the two types of options from a global-contracting perspectiveBecause ISOs are not deductible the company is indifferent betweensetting aside P in the form of NQSOs today and setting aside P(1 2 Tc) inthe form of ISOs (Since the NPV of the tax deduction for NQSOs is PTcthe NPV of the payment P is P 2 PTc or P (1 2 Tc))

Equalizing the posttax cost of NQSOs and ISOs we need only look atthe payoff of the executive to determine the condition under which eachis preferred The payoff of NQSOs in N years is

P [1 1 r (1 2 Tc)]N (1 2 Tp) (7)

22 Hall amp Liebman

The same payoff for ISOs is

P (1 2 Tc) [1 1 r (1 2 Tc)]N (1 2 Tcg) (8)

ISOs therefore are tax-advantaged if

Tp 2 Tcg Tc 2 Tc Tcg (9)

Intuitively ISOs are tax-advantaged only if their advantage (the differ-ence between the personal rate and the capital gains rate) is largeenough to offset their cost (the disadvantage of not deducting at thecorporate rate) Note however that the condition is not a simple com-parison between the corporate rate and the personal rate minus thecapital gains rate (Tp 2 Tcg Tc)

Figure 5 shows how the tax advantage of NQSOs relative to ISOs haschanged since 1980 as tax rates have changed Unlike the very modestchanges (around 3 to 4 percent) in the relative tax advantages betweenoptions and cash shown in Figure 4 the relative tax advantage betweenNQSOs and ISOs has seen enormous swings When the top personalrate was 70 percent in 1980 ISOs were tax-favored by a margin greaterthan 6 percent However by 1982 (following the 1981 tax acts that dra-matically lowered the personal tax rate) ISOs became tax-disadvantaged

FIGURE 5 Tax Advantage of NQSOs Relative to ISOs over Time

NPV of tax advantage of $100 payment to executive NQSOs minus ISOs

Taxation of Executive Compensation 23

by about 5 percent By 1988 (following the 1986 tax act) ISOs became tax-disadvantaged by about 18 percent a dramatic swing The raising ofpersonal rates coupled with a decrease in the capital gains rate has sincereduced the tax disadvantage of ISOs to about $4 per $100 of pretaxcompensation Although hard data on ISOs are hard to come by practi-tioners (mostly compensation consultants) claim that ISOs were morepopular prior to 1982 when they were tax-advantaged Although ISOsare less disadvantageous from a tax standpoint relative to NQSOs thanthey were in the late 1980s they are still disadvantageous so it is notsurprising that they have not made a signicant resurgence in recentyears

6 EXPLAINING THE INCREASE IN OPTION-BASEDCOMPENSATION

In this section we analyze whether changes in tax rates affect the compo-sition of executive compensation and consider the relative importanceof tax factors and corporate governance factors in explaining the in-creasing share of compensation paid in the form of stock options Thenumerous changes in tax rates that have occurred since 1980 provide anopportunity to assess whether executive compensation arrangementsrespond to tax incentives in the way that the tax-avoidance model sug-gests We test this model using a panel data set of CEOs in largepublicly-traded US corporations Our identication of the tax effectsrelies on time-series variation in personal corporate and capital gainstax rates along with cross-sectional variation in corporate tax rates

61 DataWe use a 15-year panel data set of CEOs in the largest publicly-tradedUS corporations which is described in Hall and Liebman (1998) Thedata set covers the years from 1980 through 1994 and combines CEOcompensation information from corporate proxies and 10-K lings withstock price and stock return information from CRSP and accounting datafrom Compustat In addition some compensation data from the 1970swere collected in order to construct measures of the value of stock op-tions held by the CEOs in the rst period of the sample8

The feature that distinguishes our data from most other CEO datasets is that with our panel of yearly proxy data on option grants optiongains and total options held we are able to calculate the total value of

8 We are grateful to David Yermack for providing us with some of the data for 1984 to 1991See Yermack (1995) for a discussion of these data

24 Hall amp Liebman

all stock options held by the CEO at a point of time More importantlysince we have the details about the stock options held (number exer-cise price time to maturity etc) we can precisely calculate the changein the value of a CEOrsquos stock option holdings for a given change in rmvalue

62 Identication IssuesThe numerous tax changes over our sample period and the specic waysin which changes in personal corporate and capital gains rates arepredicted to affect the composition of CEO compensation give us un-usually rich sources of identication However because many of thechanges in tax rates would be expected to affect all of the CEOs in oursample in a similar way we will need to pay particular attention toseparating out the tax effects from underlying time trends and otherfactors that changed over time Controlling for underlying trends is par-ticularly important in this study because we are focusing on an out-comemdashthe use of stock optionsmdashthat has increased rapidly over thepast two decades and that many practitioners believe was importantlyaffected by nontax considerations Thus the tax effects we are trying toexplain are deviations from a rising trend

We take two steps to try to separate out the tax effects from theunderlying trend First we include key nontax factors that could poten-tially explain the increasing reliance on performance-based compensa-tion It has been suggested that the dramatic increase in incentive-basedpay is the result of the remarkably poor shareholder returns during the1970s which spurred the LBO and takeover movements of the 1980sAccording to this story (Kaplan 1997) shareholders became much morepowerful via the rise of institutional investors even as the LBO andtakeover movements waned Because the inuence of institutional inves-tors is thought to be one of the most important mechanisms of strongcorporate governance we use the share of each companyrsquos stock ownedby institutional investors (dened as institutions with more than $100million under management) as an explanatory variable9 The percentageof shares owned by large institutional investors increased from about 20percent to almost 50 percent in 1994 an upward trend that closelymatches the sharp rise in the share of compensation in the form ofoptions Indeed in our sample the annual average (over all of the rmsin our sample) share of stock owned by institutional investors has acorrelation of 09 with the annual average share of compensation paid inoptions

9 This variable is described in Gompers and Metrick (1998)

Taxation of Executive Compensation 25

In addition we include two other variables that proxy for strongercorporate governance The rst is the size of the board Evidence suggeststhat boards with smaller numbers of directors reduce the inuence of theCEO and better represent shareholder interests (Yermack 1995) Boardsize fell by approximately 15 percent between 1980 and 1994 The secondvariable is the fraction of the rmrsquos board members who are inside direc-tors or gray directors (noninsiders who have business dealings with thecompany) Firms with fewer outsiders are less likely to act in the share-holdersrsquo interests and we would predict that they would be less likely tohave performance-based compensation for their CEOs The average per-centage of inside and gray directors has decreased from about 45 percentin 1980 to about 35 percent in 1994

Second we allow for cross-sectional variation in corporate tax ratesCorporate marginal tax rates vary because rms may be eligible for tax-loss carrybacks and carryforwards investment tax credits and the alter-native minimum tax (see Auerbach 1986 Auerbach and Altshuler1990 Auerbach and Poterba 1987) Because these cross-sectional differ-ences imply that different rms should respond differently to a givenchange in tax rates they potentially provide us with a way to isolatethe tax effects Firm-specic corporate marginal tax rates are notori-ously difcult to calculate from publicly-available data We use a tricho-tomous variable that equals zero if the rm has tax-loss carryforwardsand negative earnings in a given year equals 05 times the statutorycorporate rate if the rm has only one of those conditions and equalsthe statutory rate if the rm has neither loss carryforwards nor negativeearnings10

The cross-sectional variation in the corporate tax rate may not be ex-ogenous Firms that perform poorly could face low marginal tax ratesand also be particularly likely or unlikely to provide performance-basedpay Because most of our specications use xed effects the correlationwe must be concerned about is between changes in rm performanceand changes in compensation We deal with this by controlling forlagged rm stock market performance in our regressions This shouldeliminate spurious correlation between tax rates and compensation thatis jointly caused by rm performance

10 Graham (1996) presents evidence that the trichotomous variable performs nearly as wellas his recommended simulated marginal tax rate Plesko (1999) compares marginal taxrates calculated from Compustat data with actual marginal tax rates from rm tax returnsHe nds that simple binary variables are more reliable measures of the marginal tax ratethan is either the Graham or the trichotomous variable We obtain substantively similarresults when we switch from the trichotomous variable to the binary variables recommendby Plesko

26 Hall amp Liebman

63 ResultsThe dependent variable in our regressions is the fraction of a CEOrsquos totalannual compensation (measured as the BlackndashScholes value of stock-option grants plus salary and bonus) that comes from stock-optiongrants We choose this reduced-form specication because our interest isin the composition of CEO compensation However the factors thataffect the composition of compensation might also affect the levels andit will be important to keep this in mind in interpreting our results

Optimal-contracting theory generally makes predictions about the in-centives provided by the total amount of rm stock and stock optionsowned by the CEO We choose to model the ow of option grants ratherthan the ultimate performance sensitivity of the CEOrsquos compensationand stock and stock-option holdings because we have in mind anadjustment-cost model in which it takes time for a CEOrsquos holdings ofstock and stock options to reach the optimal level11 In our regressionswe include two measures of the current performance sensitivity of theCEOrsquos wealth to rm performance as explanatory variables to reect thedistance of the CEOrsquos contract from the optimal contract

Table 3 presents results in which the tax effects are identied solely byvariation in tax rates over time The fraction of annual compensation paidin stock options is regressed on tax variables corporate governance vari-ables contract-theory variables and variables reecting stock market re-turns Both of the regressions in this table include rm xed effects Incolumn (1) the fraction in annual compensation paid in stock options isregressed on the log difference in payoff (the combined payoff to both therm and the CEO) from receiving compensation in stock options ratherthan salary and bonus where the payoff calculations are based on statu-tory marginal tax rates This variable is dened as the log difference be-tween equation (3) and equation (6) in section 4 The coefcient on the taxvariable is large and statistically signicant at the 95-percent level Thepoint estimate implies that a one-percent increase in the payoff differencebetween stock options and salary and bonus results in a 24-percentage-point increase in the share of compensation paid in stock options

The corporate governance variables all have the predicted sign Firmswith a higher fraction of their shares owned by institutional investorsuse more performance-based pay (stock options) Firms with large corpo-rate boards or a large share of inside directors are less likely to use stockoptions

The two contract-theory variables are the dollar change in CEO wealth

11 See Core and Guay (1999) for evidence that rms adjust in this way

TABLE 3Compensation Regressions that Rely on Variation in

Tax Rates over Time

(1) (2)

Tax variablesln (difference in payoff)(a) 2393 (0349)ln (12statutory personal 20015 (0034)

rate)ln (12statutory cap gains 0330 (0171)

rate)ln (12statutory corporate 0399 (0148)

rate)

Corporate governance variablesFraction of rmrsquos shares 0043 (029) 0052 (0028)

owned by institutionalinvestors

Size of the rmrsquos board of 20002 (001) 20003 (0001)directors

Fraction of directors who 20082 (033) 20092 (0030)are inside or gray

Contract-theory variablesD (CEO wealth) per $1000 20261 (0095) 20247 (0087)

in FV(b)

D (CEO wealth) per 10 in 2583 (189) 2554 (201)FV(c)

Volatility(d) 521 (424) 261 (369)ln (rm market value) 0019 (0006) 0019 (0006)

Stock market return variablesLagged annual rm stock 0006 (0009) 0003 (0008)

market returnLagged annual SampP 500 re- 20035 (0023) 20029 (0023)

turn

Regression diagnosticsTotal sample size 5179 5179Firms 427 427Adjusted R2 0313 0311

Dependent variable share of compensation paid in options Standard errors in parentheses Bothregressions in this table include rm xed effects

(a) Log difference in payoff is the log difference in combined rm and CEO after-tax share betweenpaying compensation in options and paying salary

(b) This variable is the dollar change in CEO wealth per $1000 change in rm market value (seeJensen and Murphy 1990)

(c) This variable in the dollar charge in CEO wealth per 10-percent change in rm value (see Hall andLiebman 1998)

(d) Standard deviation of rm market value in trillions

28 Hall amp Liebman

per $1000 change in rm value (Jensen and Murphy 1990) and thedollar change in CEO wealth per 10-percent change in rm value (Halland Liebman 1998) As discussed in Baker and Hall (1998) these twowealth sensitivity measures reect two different concepts of how closelyaligned the managersrsquo incentives are with the interests of the sharehold-ers In the regression they are meant to measure how far the CEOrsquosexisting contract and ownership of stock and stock options is from theoptimal level The negative coefcients on the two variables suggeststhat rms that are below their optimal pay-to-performance sensitivityare more likely to give more stock options in the current period

The second column replaces the tax variable motivated by deferraladvantages with the three statutory rates The corporate after-tax sharehas a positive coefcient as we would expect because the tax advantageof stock options increases as the corporate rate falls The coefcient onthe personal tax rate is small and statistically insignicant The sign ofthe coefcient on the capital gains rate is the opposite of that predictedby tax-avoidance theory

Note that these results rely on time-series variation alone This is not avery convincing test Since the log difference in payoffs mostly rises overthe time period covered in our sample as does the share of compensa-tion paid in options it is possible that the tax variable is simply reectingother factors that were changing over this time period that we have notincluded in our regression Therefore it is important to look at specica-tions that rely on cross-sectional variation in the tax variables

Table 4 contains specications that rely on cross-sectional variation incorporate tax rates The rst column repeats the specication from col-umn (1) of Table 3 but uses the trichotomous measures of rm corporatetax rates in place of the statutory rate in calculating each rmrsquos logdifference in payoff from paying compensation in options rather than insalary and bonus As in the previous table we nd the positive coef-cient on the log difference in payoff just as the theory suggest How-ever the coefcient estimate is now much smaller suggesting that aone-percent increase in the log difference in payoff produces a 01-percentage-point increase in the share of compensation paid in options

Introducing cross-sectional variation in the dependent variable doesnot eliminate the potentially spurious correlation over time between thetax variable and the share of compensation paid in options The secondcolumn of Table 4 isolates the pooled cross-sectional variation in corpo-rate tax rates by adding time dummies and dropping the rm xedeffects In this specication the coefcient on the tax variable is indistin-guishable from zero The coefcient on the institutional variable contin-ues to be large and statistically signicant The third column includes

Taxation of Executive Compensation 29

both rm xed effects and time dummies which tests for a relationshipbetween within-rm changes in the dependent variable and within-rmchanges in each explanatory variable In this specication the coefcienton the tax variable is again indistinguishable from zero The coefcientson the corporate governance variables are also statistically insignicant

The third column contains a specication that treats the Tax ReformAct of 1986 (TRA 86) as a natural experiment For rms and CEOs facingthe statutory corporate tax rate TRA 86 increased the relative payoff topaying compensation in options by 17 percent For a rm facing a zeromarginal tax rate it reduced the relative payoff by 45 percent The reduc-tion occurred because TRA 86 reduced the personal tax rate makingsalary and bonus more attractive while the decreased corporate rate hadno effect on zero-tax-rate rms Thus we would expect rms facing zeroor low marginal tax rates to reduce their use of options after TRA 86while rms facing the statutory rate would be expected to increase themFew rms face zero marginal tax rates year after year Therefore we tryto distinguish between rms that often face low tax rates and rms thatnearly always face the statutory rate We dene low-tax rms as oneswhose average marginal tax rate over 1984 to 1986 was below 02412

Then we ran a difference-in-differences regression to see if high-taxrms increased their use of options more than low-tax rms The interac-tion between post-TRA 86 and high corporate tax rates is the key vari-able The estimated coefcient on this variable is positive but small andnot statistically different from zero at the 95-percent level

It is important to note that in this specication as well the main corpo-rate governance variable the share of the rmrsquos shares owned by institu-tional investors performs as theory would predict The variable appearsto be quite robust Table 5 presents regressions of the share of compensa-tion paid in stock options on the fraction of the rmrsquos shares owned byinstitutional investors Since we have already established that these vari-ables have a strong longitudinal relationship the regressions here aredesigned to focus on the cross-sectional relationship between these twovariables The rst and third columns present regressions using data onlyfrom 1994 and nd a coefcient of roughly 02 (column 1 contains noother covariates while column 3 includes all of the other nontax covari-ates from the regressions in Tables 3 and 4) Since the median share of armrsquos stock owned by institutional investors in our sample rises from 19percent in 1980 to 49 percent in 1994 this coefcient would imply that this

12 This cutoff was chosen after inspecting the distribution of tax rates and essentiallyseparates rms that are always at or near the statutory rate from ones that are sometimesbelow it

TA

BL

E4

Com

pens

atio

nR

egre

ssio

nsth

atR

ely

onC

ross

-Sec

tion

alV

aria

tion

inC

orpo

rate

Tax

Rat

es

(1)

(2)

(3)

(4)

Tax

vari

able

s(a)

ln

(dif

fere

nce

inp

ayof

f)0

099

(00

41)

20

038

(00

36)

20

005

(00

44)

Pos

t-T

RA

862

000

5(0

013

)H

igh

corp

orat

eta

xra

te0

018

(00

17)

(Pos

t-T

RA

86)

3hi

gh0

024

(00

16)

corp

orat

eta

xra

te

Cor

por

ate

gove

rnan

ceva

riab

les

Frac

tion

of

rmrsquos

shar

es0

109

(00

27)

011

4(0

015

)2

000

9(0

030

)0

087

(00

34)

owne

dby

inst

itu

tion

alin

vest

ors

Size

ofth

e

rmrsquos

boar

dof

20

004

(00

01)

20

001

(00

01)

20

002

(00

01)

000

1(0

002

)d

irec

tors

Frac

tion

ofd

irec

tors

wh

o2

012

4(0

033

)2

000

3(0

018

)2

004

1(0

034

)0

001

(00

39)

are

insi

de

orgr

ay

Con

trac

t-th

eory

vari

able

sD

(CE

Ow

ealt

h)p

er$1

000

20

238

(00

94)

20

167

(00

53)

20

257

(00

96)

20

021

(01

47)

inFV

(b)

D(C

EO

wea

lth)

per

10

in2

560

(17

8)2

592

(10

5)2

594

(19

6)2

147

(04

86)

FV(c

)

Vol

atili

ty(d

)2

48(4

24)

162

(03

4)1

12(4

28)

106

(06

09)

ln(

rmm

arke

tva

lue)

003

3(

006)

000

8(0

003

)0

009

(00

06)

001

2(0

007

)

Stoc

km

arke

tre

turn

vari

able

sL

agge

dan

nual

rm

stoc

k2

000

2(0

009

)0

011

(00

10)

000

6(0

009

)2

001

2(0

013

)m

arke

tre

turn

Lag

ged

annu

alSamp

P50

02

005

6(0

023

)2

005

3(0

051

)2

003

9(0

049

)2

015

4(0

040

)re

turn

Reg

ress

ion

dia

gnos

tics

T

otal

sam

ple

size

5179

5179

5179

2534

Firm

s42

742

739

6A

dju

sted

R2

031

009

032

005

Firm

xed

effe

cts

Yes

No

Yes

No

Tim

ed

um

mie

sN

oY

esY

esN

o

Dep

end

ent

vari

able

isfr

acti

onof

ann

ualc

omp

ensa

tion

from

stoc

kop

tion

sSt

and

ard

erro

rsin

par

enth

eses

(a)

All

thre

esp

eci

cati

ons

use

tric

hoto

mou

sco

rpor

ate

tax

rate

sT

he

sam

ple

inco

lum

n(3

)u

ses

only

the

year

s19

84to

1988

sin

cew

ear

eu

sing

1986

asa

nat

ural

exp

erim

ent

(b)

Thi

sva

riab

leis

the

dol

lar

chan

gein

CE

Ow

ealt

hp

er$1

000

chan

gein

rm

mar

ketv

alu

e(s

eeJe

nse

nan

dM

urp

hy

1990

)

(c)

Thi

sva

riab

lein

the

dol

lar

char

gein

CE

Ow

ealt

hp

er10

-per

cent

chan

gein

rm

valu

e(s

eeH

alla

nd

Lie

bman

199

8)

(d)

Stan

dar

dd

evia

tion

ofr

mm

arke

tva

lue

intr

illio

ns

32 Hall amp Liebman

TABLE 5Large Institutional Investors and Stock-Option Compensation

The Cross-sectional Relationship between Institutional Investors andOption Grants

(1) (2) (3) (4)

Coefcient on fraction of 217 (067) 145 (014) 177 (069) 120 (015)rmrsquos shares owned byinstitutional investors

Sample period 1994 1980ndash1994 1994 1980ndash1994Contract-theory and stock No No Yes Yes

market covariatesTime dummies No Yes No Yes

The dependent variable is the share of compensation paid in options

factor alone can explain 6 percentage points of the 23-percentage-pointincrease in the median share of compensation paid in options that oc-curred over this time period Of course this variable is only one measureof the strengthening of corporate governance so the total impact of im-proved corporate governance could be greater The second and fourthcolumns of the table use the entire 1980ndash1994 sample and include timedummies to isolate the cross-sectional correlations Again there is evi-dence of a strong cross-sectional relationship between the two variablesalthough the coefcients are slightly smaller perhaps reecting a weakercorporate-governance role in the earlier period In sum there is a strongrelationship in both longitudinal and cross-sectional data between theinstitutional-investor variable and the options-percentage variable13

7 THE MILLION-DOLLAR RULE

In 1993 section 162(m) of the Internal Revenue Code was enactedlimiting the deductibility of executive compensation in excess of onemillion dollars unless the compensation was performance-related Pro-ponents of this legislation argued that it would slow the growth ofexecutive pay and tighten the link between rm performance and execu-tive compensation

The new provision became effective for tax year 1994 and applies tothe CEO and the other four most highly compensated executives in eachcompany Pay that is performance-related such as stock-option grants

13 The one specication in which the fraction owned by institutional investors is notsignicant is in column (2) of Table 4 a regression that includes both xed effects and timeeffects

Taxation of Executive Compensation 33

and bonuses paid for meeting clear performance goals is not affected bythis regulation provided that it has been approved by shareholdersThus the main impact of the provision is to limit the deductibility ofsalary in excess of a million dollars Woodlock and Antenucci (1997)studied the proxy statements of 376 rms and documented that mostrms responded to the new law by qualifying top executive pay asperformance-related Perry and Zenner (1999) identied 25 rms thatreduced salaries from above $1 million to below $1 million and foundthat 23 of the rms cited section 162(m) as their reason for doing so

In this section of the paper we investigate whether this provision hasaffected the level or structure of executive pay using the Execucompdata base which contains information on the highest-paid executives ineach of the rms in the SampP 500 SampP Midcap 400 and SampP SmallCap600 Between 1993 and 1998 the median salary of the Execucomp execu-tives rose from $230000 to $306000 The median value of total compen-sation (salary bonus and option grants) rose from $470000 to $882000Limiting the sample of CEOs indicates that the total compensation of themedian CEO rose from $11 million to $18 million over the period whilethe median salary rose from $441000 to $550000 Thus executive com-pensation clearly continued to rise at a rapid rate after the implementa-tion of the million-dollar rule though perhaps at a slower rate than itotherwise would have

If the regulation affects behavior it should have the largest impact onrms whose executives are receiving salaries that are above the limit orjust below it Thus we might expect to see rms whose executives arereceiving relatively high salaries to rely more heavily on performance-based pay and less heavily on salary in the years following the regulationPerry and Zenner (1999) and Livingston (1999) both nd that this is thecase This trend can be clearly identied in the Execucomp data set Forexample over 1993 to 1998 the median annualized growth rate of salaryfor executives whose 1993 salary was below $500000 in the previous yearwas 85 percent For executives with salary between $500000 and$700000 it was 64 percent for executives between $700000 and $850000it was 50 percent for executives between $750000 and $1 million it was31 percent and for executives above $1 million it was 00 percent14

The observation that executives at lower levels of salary had highergrowth rates is not very strong evidence of an impact of section 162(m)however because it is possible that low-salary executives would havehad higher growth rates even in the absence of the provision To explore

14 Of the 3137 executives for whom we have complete data between 1993 and 1998 2708had 1993 salaries below $500000 while only 35 had 1993 salaries above $1 million

34 Hall amp Liebman

whether this is the case we use Execucomp data from both before andafter the law change Ideally we would want to have data from manyyears before the change to examine whether there is an underlying pat-tern of faster salary growth for low-salary executives Unfortunately theExecucomp data begin in 1992 and therefore provide us with only the1992ndash1993 growth rate in compensation to use as a baseline15

We run regressions pooling data from 1992ndash1993 through 1997ndash1998of the annual percentage change in different forms of compensation on avariable million that is designed to measure the likelihood that anexecutiversquos compensation will be affected by section 162(m) We denemillion as the minimum of 1 and the executiversquos salary divided by $1million in the previous year Thus any executive with salary at or above$1 million would receive a value of 1 while an executive with a salary of$500000 would be coded as a 0516 We include this variable by itself andalso interacted with a dummy variable (after) that equals 1 for years1994 and after Thus the million variable is designed to pick up anyunderlying relationship between the level of salary and the growth rateof salary while the interacted one will identify any differential growthrate of salary by income in the period after section 162(m) took effect Wealso include a full set of year dummies the natural logarithm of thermrsquos market value and the rmrsquos annual rate of return over the twoprevious years as control variables To reduce the sensitivity of our re-sults to outliers we run quantile regressions and robust regressions

Table 6 contains our results The rst four columns are for all topexecutives while the last four columns focus on CEOs Columns (1) and(2) present estimates from robust regression for the percentage change insalary and the percentage change in total compensation (salary bonusstock grants and stock-option grants) In the salary growth regressionsfor all executives the coefcient on million is negative and signicantindicating that there is an underlying relationship between the level ofexecutive compensation and its subsequent growth rate The coefcienton million 3 after is negative as well suggesting that the negativerelationship between the level of salary and its subsequent growth rateintensied slightly after 1994 The relatively small magnitude suggeststhat an executive with a $1 million dollar salary would see his salarygrow at an annual rate that would be about 06 percent slower than anexecutive earning $500000

15 The HallndashLiebman data set contains only a combined salary and bonus variable makingit impossible to use it to analyze this issue16 We also ran the regressions (not reported) allowing for a nonlinear relationship betweenmillion and compensation growth and obtained substantively similar results

TA

BL

E6

The

Eff

ect

ofth

eM

illi

on-D

olla

rR

ule

onSa

lary

and

Tot

alC

ompe

nsat

ion

All

top

exec

uti

ves

CE

Os

only

Rob

ust

regr

essi

onQ

uan

tile

regr

essi

onR

obu

stre

gres

sion

Qu

anti

lere

gres

sion

D

in

Din

D

in

Din

D

in

Din

D

in

Din

sala

ryto

talc

omp

(a)

sala

ryto

talc

omp

sa

lary

tota

lcom

p

sala

ry

tota

lcom

p

mil

lio

n(b

)2

006

02

035

72

006

32

024

12

008

72

023

12

010

32

024

5(0

005

)(0

032

)(0

006

)(0

027

)(0

010

)(0

073

)(0

011

)(0

056

)m

illi

on

32

001

10

026

20

021

20

015

20

016

014

92

003

40

158

aft

er(c

)(0

005

)(0

033

)(0

006

)(0

029

)(0

010

)(0

076

)(0

011

)(0

057

)N

359

5035

950

359

5035

950

757

17

589

757

17

589

Stan

dar

der

rors

inp

aren

thes

esR

egre

ssio

nsal

soin

clu

de

year

du

mm

ies

stoc

km

arke

tre

turn

and

log

mar

ket

valu

e

(a)

Tot

alco

mp

ensa

tion

incl

ud

essa

lary

bon

us

rest

rict

edst

ock

and

stoc

k-op

tion

gran

ts

(b)

mil

lion

isd

ene

das

min

(1s

alar

yin

pre

viou

sye

ars

div

ided

by$1

mil

lion)

(c)

aft

eris

anin

dic

ator

vari

able

equ

alto

zero

befo

reth

ere

gula

tion

took

plac

e(1

993)

and

one

afte

rwar

d

36 Hall amp Liebman

In the second column the coefcient on million 3 after is essentiallyzero suggesting that any decrease in salary brought about by section162(m) was offset by increases in bonus and stock-option grants Theresults are similar for the median regressions There is some evidence ofa very small slowdown in growth of salary for executives likely to beaffected by section 162(m) but no sign of a change in total compensa-tion Taken together this provides evidence of a very small substitutionof performance-related pay for salary

The last four columns run identical tests but limit the sample to CEOsThe interaction of million 3 after continues to be small and negative inthe salary regressions In contrast the impact of 162(m) on total compen-sation appears to be positive and fairly substantial (about 15 percentagepoints per year) in these regressions That is the increase in bonus andstock options following the million-dollar rule more than offset the verysmall decline in salary

All of these conclusions should be interpreted somewhat cautiouslybecause our data provide us with only one pre-1993 control year Never-theless the data are consistent with section 162(m) having led to a veryminor slowdown in salary growth and one that is dramatically smallerthan what one would conclude without controlling for underlying differ-ential salary growth rates at different levels of salary Since total compen-sation for CEOs did not decline and perhaps increased there is evidenceof a minor substitution of performance-related pay for salary in responseto the regulation

8 EXECUTIVE COMPENSATION AND THEELASTICITY OF TAXABLE INCOME

Even if taxes have only a minor effect on executive compensation thestructure of executive compensation is important for tax policy be-cause high-income taxpayers are sources of signicant revenue and arethe focus of important debates about the efciency cost of taxation

Goolsbee (1999) has recently argued that executive decisions to exer-cise options are highly responsive to intertemporal differences in taxrates created by tax reforms He claims that much of the apparent declinein taxable income for high-income taxpayers between 1992 and 1993documented by Feldstein and Feenberg (1996) can be explained by execu-tives shifting option gains into 1992 in order to avoid the higher marginaltax rates that went into effect in 1993 By extension he implies that thehigh elasticities of taxable income with respect to personal tax ratesestimated by comparing the pre- and post-tax-reform taxable income of

Taxation of Executive Compensation 37

high-income individuals reect intertemporal shifting of income and notpermanent effects of taxation If this is true then the deadweight loss oftaxation is lower as is the cost of progressivity

While Goolsbee presents comprehensive evidence that option gainswere unusually high in 1992 his identication of the impact of tax rates onthe timing of option gains depends on correctly specifying a counter-factual level of options that would have been exercised in the absence ofthe tax change He simply assumes a linear time trend The purpose ofGoolsbeersquos time trend is presumably to account for the increasing use ofoptions over time However since options typically require a vestingperiod of at least a couple of years a much more direct measure of ex-pected option gains can be constructed based on the total value of vestedoptions held by the CEO Since options typically have vesting periods thatexceed two years this total can be treated as exogenous since it will not beaffected by the tax changes

More generally the value of options held by the CEO is likely to havea major effect on his exercise decisions If his options are out of themoney he will not have gains to realize If there has been a recent run-up in the market an optimizing CEO may exercise an unusually largeamount of options in order to diversify his overall portfolio and if in-the-money options are about to expire they will be exercised in thecurrent year

Our evidence suggests that much of the apparent tax shifting caninstead be attributed to stock market performance and the timing of pastoption grants Using our 15-year panel we replicate Goolsbeersquos (1999)result for the 1993 period but show that the impact of taxes is notpresent in other time periods Moreover even the result for the 1993period disappears when appropriate controls for past option grants andstock price appreciation are included

The top three rows of Table 7 present data on the taxable incomes ofCEOs in the HallndashLiebman sample For these results our panel has beennarrowed to include only CEOs whose rmrsquos scal years correspondwith tax years In addition we create a balanced panel for each of thethree tax-reform periods by excluding any rm for which there are notcomplete data for the period surrounding the reform

The fourth row of the table shows that the 1993 tax act raised the topmarginal personal income tax rate from 310 to 396 percent effective fortax year 1993 with a further increase occurring in 1994 due to the uncap-ping of the Medicare payroll tax The top row of the table shows thattaxable income was higher in 1992 than for any other year in our sample(all data are in 1994 dollars) Option gains were particularly high in1992mdashmore than double their 1991 level and 68 percent higher than their

TA

BL

E7

Tax

Ref

orm

Sto

ckM

arke

tP

erfo

rman

cea

ndth

eT

imin

gof

Opt

ion

Gai

ns

Ave

rage

stoc

km

arke

tM

ean

CE

OM

ean

CE

Ore

turn

for

Mea

nva

lue

taxa

ble

sala

ryan

dM

ean

CE

OT

opfe

der

al

rms

inof

opti

ons

(Val

ue

ofop

tion

Sam

ple

inco

me

bonu

sop

tion

gain

sp

erso

nal

sam

ple

for

held

atbe

gga

ins)

(va

lue

Ref

orm

size

Yea

r($

mil

)($

mil

)($

mil

)M

TR

pre

vye

ar(

)of

year

ofop

tion

she

ld)

ER

TA

8127

919

800

874

064

80

163

7025

80

460

354

1981

083

10

672

009

869

245

070

61

3919

820

872

067

90

099

5011

90

686

144

1983

099

00

724

025

250

355

116

62

16

TR

A86

295

1985

119

70

837

024

050

76

116

02

0719

861

448

091

70

345

5042

81

776

194

1987

165

50

959

059

338

519

12

229

266

1988

163

21

058

039

028

07

237

71

6419

891

656

105

20

414

2817

92

345

176

1993

272

1990

209

51

028

061

828

232

288

02

15T

axA

ct19

911

875

102

00

547

312

01

268

82

0319

922

678

107

01

250

3146

64

552

275

1993

232

41

209

074

339

620

64

299

173

1994

196

41

262

037

842

515

45

113

074

Sam

ple

isH

allndash

Lie

bman

rm

sw

ith

Dec

embe

r

scal

year

sth

atha

dco

mp

lete

dat

afo

ron

eof

the

thre

eta

x-re

form

per

iod

sst

ud

ied

Taxation of Executive Compensation 39

1994 level These patterns are consistent with Goolsbeersquos claim that tax-payers responded to the anticipated increase in marginal tax rates byshifting gains into 1992

The evidence from the other two tax reforms is less clear suggestingthat there might be factors other than tax rates determining the timing ofoption gains ERTA 1981 reduced the top marginal tax rate from 70percent in 1980 to 50 percent in 1982 However there is no sign thatoption gains were reduced in 1980 and 1981 and then increased in 198217

There was a big increase in option gains in 1983 however Similarly thehighest year of stock-option gains in the period around 1986 was 1987 astrange year to take gains since taxpayers would have known that thepersonal rate would be 10 percentage points lower if they waited untilthe following year18

What all the peak years of option gains have in common is that theyfollow years of strong stock market performance The fth row of Table 7shows the average stock market return during the previous year for therms in our sample The large value of option gains in 1983 followed a355-percent increase in the stock market The 1992 boom in option gainsfollowed a 466-percent increase in the stock market the more modest1987 peak in option gains followed a 19-percent gain (and a 43-percentgain two years before)

Strong stock market performance raises the value of options availableto exercise The sixth row of the table shows the mean value of optionsheld by CEOs at the beginning of the year All three of the peak years ofoption gains correspond with peak years in terms of the value of optionsavailable for exercise Since stock market appreciation during the yearalso increases the value of options available to exercise later in that yearwe need to allow for current-year stock market appreciation as well

Table 8 contains Goolsbee-style regressions for each of the tax re-forms The rst two columns reprint the results from Goolsbee (1999)The third and fourth columns replicate his results in our sample Inparticular with a specication similar to his we are able to replicate hisnding that the current after-tax share has a positive coefcient and thefollowing yearrsquos after-tax share has a coefcient of roughly negative oneColumns (5) through (8) show that this pattern was unique to the 1993tax reform Indeed in the TRA 86 regressions the coefcient on the

17 As we explained above for executives covered by the maximum tax on earned incomethere was not a decline in tax rates between 1981 and 198218 It is possible that the large stock market decline in October 1997 may have promptedsome people to exercise options earlier than they otherwise would have However anincrease in option exercises typically follows stock market increases not declines

TA

BL

E8

Att

empt

sto

Rep

lica

teth

eT

ax-S

hift

ing

Res

ult

inO

ther

Tim

eP

erio

ds

Goo

lsbe

eG

ools

bee

1993

1993

Ent

ire

Ent

ire

regr

essi

onre

gres

sion

refo

rmre

form

TR

A86

TR

A86

ER

TA

81E

RT

A81

1980

ndash199

419

80ndash1

994

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

ln[1

2ta

x(t)

]1

244

108

20

455

032

62

017

62

017

22

016

72

016

70

026

20

068

(01

34)

(01

27)

(05

37)

(05

42)

(02

10)

(02

09)

(00

90)

(00

91)

(00

67)

(00

65)

ln[1

2ta

x(t1

1)]

20

867

20

764

21

214

20

917

047

80

722

20

103

004

30

176

018

3(0

118

)(0

112

)(0

491

)(0

499

)(0

157

)(0

174

)(0

072

)(0

072

)(0

068

)(0

066

)

Tim

e0

135

007

62

000

52

036

40

041

000

80

094

003

50

056

002

7(0

008

)(0

008

)(0

049

)(0

049

)(0

028

)(0

028

)(0

025

)(0

026

)(0

003

)(0

003

)

ln(m

arke

tva

lue)

066

10

489

023

70

334

039

5(0

015

)(0

058

)(0

039

)(0

042

)(0

017

)

Yea

rs19

91ndash1

995

1991

ndash199

519

90ndash1

994

1990

ndash199

419

85ndash1

989

1985

ndash198

919

80ndash1

983

1980

ndash198

319

80ndash1

994

1980

ndash199

4

Ad

just

edR

20

720

770

610

630

690

760

800

810

630

68

Obs

erva

tion

s13

856

132

371

385

136

41

470

143

31

113

106

23

968

385

9

Dep

end

ent

vari

able

isth

elo

gof

taxa

ble

inco

me

All

regr

essi

ons

con

tain

rm

xe

def

fect

sG

ools

bee

regr

essi

on(2

)al

soco

ntai

nstw

oot

her

vari

able

sno

tsh

own

inth

eta

ble

Stan

dar

der

rors

inp

aren

thes

es

Taxation of Executive Compensation 41

TABLE 9Alternative Explanations for the Timing of Option Exercises

1993 Entirereform TRA 86 ERTA 81 1980ndash1994

(1) (2) (3) (4)

ln[12tax(t)] 0108 0111 20051 0201(0324) (0136) (0048) (0057)

ln[12tax(t11)] 20134 0573 0125 0150(0336) (0156) (0053) (0066)

ln(market value) 0418 0177 0338 0418(0061) (0039) (0046) (0163)

Firmrsquos stock market return (t) 0298 0268 0214 0258(0044) (0045) (0031) (0023)

Firmrsquos stock market return 0162 0206 0031 0128(t21) (0041) (0008) (0029) (0021)

ln(value of options held by 0004 0008 0001 0009CEO at start of year) (0003) (0003) (0002) (0001)

Years 1990ndash1994 1985ndash1989 1980ndash1983 1980ndash1994Adjusted R2 064 071 082 068Observations 1349 1427 1036 3812

The dependent variable is the log of taxable income All regressions contain rm xed effects Stan-dard errors in parentheses

following yearrsquos tax rate is positive and signicant Columns (9) and (10)pool all three reforms and fail to replicate the pattern that would reectintertemporal shifting19

It is possible that the ease with which one can shift the timing ofoption gains is asymmetric It may be harder to shift option gains for-ward in time since some options expire and therefore must be exercisednow In addition options were a much smaller portion of compensationin the early 1980s so perhaps the ERTA 81 results are not surprisingNonetheless these results cast doubt on the proposition that this is thechannel through which spurious estimates of the elasticity of taxableincome have occurred for previous tax reforms and they suggest thatwe should look harder for alternative explanations for the timing ofoption gains

Table 9 attempts to predict options gains using the lagged and currentincrease in a rmrsquos market value and the value of the options held by theCEO at the start of the year The two stock market variables stronglypredict option gains while the value of options held has the correct sign

19 The results also fail to suggest large permanent effects of marginal tax rates on taxableincome

42 Hall amp Liebman

in all of the specications and is signicant in the pooled sample eventhough it is highly correlated with the two stock market return variablesIn contrast the tax-timing explanation fails even in the 1993 period

It is important to realize that our results do not invalidate Goolsbeersquoscritique of claims that aggregate revenue data from 1992 and 1993 provethat elasticities of taxable income are large We have simply given analternative explanation for why taxable incomes were unusually high in1992 However our results do cast doubt on the conclusion that largetiming shifts are ubiquitous and capable of explaining away the morecareful estimates of taxable income elasticities that have been performedusing micro data (eg Feldstein 1995 Auten and Carroll 1995)

One further point is in order It is sometimes argued that the combina-tion of increased use of options and the booming stock market explainswhy federal tax revenues have been so much higher than expected dur-ing the 1990s However while large option gains increase revenue fromthe personal income tax they produce offsetting deductions for cor-porations The net impact on revenue is only the difference in the twomarginal tax rates times the option gain While the gap between thepersonal rate and the effective corporate rate is nontrivial the impact oflarge option exercises on total revenue is substantially smaller than theamount by which the exercises inate personal tax revenue and reducecorporate tax revenue

9 CONCLUSION

We have described the tax rules for executive pay in detail and analyzedhow changes in various tax rates affect the tax advantages of stock optionsrelative to salary and bonus We nd that there is a moderate tax advan-tage to optionsmdashcurrently about $4 per $100 in pretax executive paymdashand that changes in the tax advantage of options over time have had atmost a modest impact on the composition of pay Corporate governancefactors particularly the role of large investors appear to be more impor-tant in explaining the dramatic increase in option pay

In addition our evidence suggests that more direct attempts to use taxpolicy to inuence executive compensation have had little effect Wend that the million-dollar rule led companies to substitute perfor-mance-based pay for salary But our evidence suggests that this substitu-tion was quite modest and there is no evidence that the total level of paywas reduced Overall although the stock-option explosion has dramati-cally increased the link between pay and performance this change isdue almost entirely to nontax factors

Taxation of Executive Compensation 43

REFERENCES

Auerbach Alan J (1996) ldquoThe Dynamic Effects of Tax Law Asymmetriesrdquo Re-view of Economic Studies 53451ndash405

mdashmdashmdash and James Poterba (1987) ldquoTax Loss Carryforwards and Corporate TaxIncentivesrdquo In The Effects of Taxation on Capital Accumulation M Feldstein(ed) Chicago University of Chicago Press

Altshuler Rosanne and Alan Auerbach (1990) ldquoThe Signicance of Tax LawAsymmetries An Empirical Investigationrdquo Quarterly Journal of Economics 10561ndash86

Auten Gerald and Robert Carroll (1995) ldquoThe Effect of Income Taxes on House-hold Behaviorrdquo Ofce of Tax Analysis US Department of the Treasury

Baker George P and Brian J Hall (1998) ldquoCEO Incentives and Firm SizerdquoNBER Working Paper no 6868 December

Core John and Wayne Guay (1999) ldquoThe Use of Equity Grants to ManageOptimal Equity Incentive Levelsrdquo Draft

Feldstein M (1995) ldquoThe Effect of Marginal Tax Rates on Taxable Income APanel Study of the 1986 Tax Reform Actrdquo Journal of Political Economy 103 (no3)551ndash72

mdashmdashmdash and D Feenberg (1996) ldquoThe Effect of Increased Tax Rates on TaxableIncome and Economic Efciency A Preliminary Analysis of the 1993 Tax RateIncreasesrdquo In Tax Policy and the Economy vol 10 James M Poterba (ed)Cambridge MA MIT Press

Gompers Paul A and Andrew Metrick (1998) ldquoInstitutional Investors andEquity Pricesrdquo NBER Working Paper no 6723

Goolsbee Austan (1999) ldquoWhat Happens When You Tax the Rich Evidencefrom Executive Compensationrdquo NBER Working Paper no 6333

Graham John R (1996) ldquoProxies for the Corporate Marginal Tax Raterdquo Journal ofFinancial Economics 42187ndash221

Hall Brian J (1999) ldquoThe Design of Multi-year Stock Option Plansrdquo Journal ofApplied Corporate Finance 12(2)97ndash106

mdashmdashmdash and Jeffrey B Liebman (1998) ldquoAre CEOs Really Paid Like Bureau-cratsrdquo Quarterly Journal of Economics CXII(no 3)653ndash691

Jensen Michael C and William M Meckling (1976) ldquoTheory of the Firm Mana-gerial Behavior Agency Costs and Ownership Structurerdquo Journal of FinancialEconomics III305ndash360

mdashmdashmdash and Kevin J Murphy (1990) ldquoPerformance Pay and Top-ManagementIncentivesrdquo Journal of Political Economy XCVIII225ndash264

Kaplan Steven N (1997) ldquoThe Evolution of US Corporate Governance WeAre All Henry Kravis Nowrdquo Draft

Livingstone Jane R (1999) ldquoExecutive Compensation Contracting and Re-sponses to Increased Tax Costsrdquo Louisiana State University June

Miller Merton H and Myron S Scholes (1982) ldquoExecutive CompensationTaxes and Incentivesrdquo In Financial Economics Essays in Honor of Paul CootnerWilliam F Sharpe and Cathryn M Cootner (eds) Princeton NJ Prentice-Hall

Murphy Kevin J (1999) ldquoExecutive Compensationrdquo forthcoming in OrleyAshenfelter and David Card (eds) Handbook of Labor Economics Vol 3 NorthHolland

Perry Todd and Marc Zenner (1999) ldquoPay for Performance Government Regu-

Taxation of Executive Compensation 21

FIGURE 4 Tax Advantage of Options 1980 to 1998

NPV of tax advantage of $100 in options (NQSOs) rather than cash

53 The Tax (Dis)advantage of ISOsSo far the analysis has focused on the tax advantages of NQSOs sincethey are so much more prevalent than ISOs which account for onlyabout 5 percent of option grants The relative scarcity of ISOs can beexplained both by their tax status and by their per-executive cap of$100000 per year

Under what conditions are ISOs tax-preferred to NQSOs For a trans-fer P to an executive the ISO is always tax-preferred by the executive(since the capital gains rate is lower than the personal rate) while theNQSO is always tax-preferred from the companyrsquos perspective (since anISO is not deductible) The key issue however is the relative advan-tages of the two types of options from a global-contracting perspectiveBecause ISOs are not deductible the company is indifferent betweensetting aside P in the form of NQSOs today and setting aside P(1 2 Tc) inthe form of ISOs (Since the NPV of the tax deduction for NQSOs is PTcthe NPV of the payment P is P 2 PTc or P (1 2 Tc))

Equalizing the posttax cost of NQSOs and ISOs we need only look atthe payoff of the executive to determine the condition under which eachis preferred The payoff of NQSOs in N years is

P [1 1 r (1 2 Tc)]N (1 2 Tp) (7)

22 Hall amp Liebman

The same payoff for ISOs is

P (1 2 Tc) [1 1 r (1 2 Tc)]N (1 2 Tcg) (8)

ISOs therefore are tax-advantaged if

Tp 2 Tcg Tc 2 Tc Tcg (9)

Intuitively ISOs are tax-advantaged only if their advantage (the differ-ence between the personal rate and the capital gains rate) is largeenough to offset their cost (the disadvantage of not deducting at thecorporate rate) Note however that the condition is not a simple com-parison between the corporate rate and the personal rate minus thecapital gains rate (Tp 2 Tcg Tc)

Figure 5 shows how the tax advantage of NQSOs relative to ISOs haschanged since 1980 as tax rates have changed Unlike the very modestchanges (around 3 to 4 percent) in the relative tax advantages betweenoptions and cash shown in Figure 4 the relative tax advantage betweenNQSOs and ISOs has seen enormous swings When the top personalrate was 70 percent in 1980 ISOs were tax-favored by a margin greaterthan 6 percent However by 1982 (following the 1981 tax acts that dra-matically lowered the personal tax rate) ISOs became tax-disadvantaged

FIGURE 5 Tax Advantage of NQSOs Relative to ISOs over Time

NPV of tax advantage of $100 payment to executive NQSOs minus ISOs

Taxation of Executive Compensation 23

by about 5 percent By 1988 (following the 1986 tax act) ISOs became tax-disadvantaged by about 18 percent a dramatic swing The raising ofpersonal rates coupled with a decrease in the capital gains rate has sincereduced the tax disadvantage of ISOs to about $4 per $100 of pretaxcompensation Although hard data on ISOs are hard to come by practi-tioners (mostly compensation consultants) claim that ISOs were morepopular prior to 1982 when they were tax-advantaged Although ISOsare less disadvantageous from a tax standpoint relative to NQSOs thanthey were in the late 1980s they are still disadvantageous so it is notsurprising that they have not made a signicant resurgence in recentyears

6 EXPLAINING THE INCREASE IN OPTION-BASEDCOMPENSATION

In this section we analyze whether changes in tax rates affect the compo-sition of executive compensation and consider the relative importanceof tax factors and corporate governance factors in explaining the in-creasing share of compensation paid in the form of stock options Thenumerous changes in tax rates that have occurred since 1980 provide anopportunity to assess whether executive compensation arrangementsrespond to tax incentives in the way that the tax-avoidance model sug-gests We test this model using a panel data set of CEOs in largepublicly-traded US corporations Our identication of the tax effectsrelies on time-series variation in personal corporate and capital gainstax rates along with cross-sectional variation in corporate tax rates

61 DataWe use a 15-year panel data set of CEOs in the largest publicly-tradedUS corporations which is described in Hall and Liebman (1998) Thedata set covers the years from 1980 through 1994 and combines CEOcompensation information from corporate proxies and 10-K lings withstock price and stock return information from CRSP and accounting datafrom Compustat In addition some compensation data from the 1970swere collected in order to construct measures of the value of stock op-tions held by the CEOs in the rst period of the sample8

The feature that distinguishes our data from most other CEO datasets is that with our panel of yearly proxy data on option grants optiongains and total options held we are able to calculate the total value of

8 We are grateful to David Yermack for providing us with some of the data for 1984 to 1991See Yermack (1995) for a discussion of these data

24 Hall amp Liebman

all stock options held by the CEO at a point of time More importantlysince we have the details about the stock options held (number exer-cise price time to maturity etc) we can precisely calculate the changein the value of a CEOrsquos stock option holdings for a given change in rmvalue

62 Identication IssuesThe numerous tax changes over our sample period and the specic waysin which changes in personal corporate and capital gains rates arepredicted to affect the composition of CEO compensation give us un-usually rich sources of identication However because many of thechanges in tax rates would be expected to affect all of the CEOs in oursample in a similar way we will need to pay particular attention toseparating out the tax effects from underlying time trends and otherfactors that changed over time Controlling for underlying trends is par-ticularly important in this study because we are focusing on an out-comemdashthe use of stock optionsmdashthat has increased rapidly over thepast two decades and that many practitioners believe was importantlyaffected by nontax considerations Thus the tax effects we are trying toexplain are deviations from a rising trend

We take two steps to try to separate out the tax effects from theunderlying trend First we include key nontax factors that could poten-tially explain the increasing reliance on performance-based compensa-tion It has been suggested that the dramatic increase in incentive-basedpay is the result of the remarkably poor shareholder returns during the1970s which spurred the LBO and takeover movements of the 1980sAccording to this story (Kaplan 1997) shareholders became much morepowerful via the rise of institutional investors even as the LBO andtakeover movements waned Because the inuence of institutional inves-tors is thought to be one of the most important mechanisms of strongcorporate governance we use the share of each companyrsquos stock ownedby institutional investors (dened as institutions with more than $100million under management) as an explanatory variable9 The percentageof shares owned by large institutional investors increased from about 20percent to almost 50 percent in 1994 an upward trend that closelymatches the sharp rise in the share of compensation in the form ofoptions Indeed in our sample the annual average (over all of the rmsin our sample) share of stock owned by institutional investors has acorrelation of 09 with the annual average share of compensation paid inoptions

9 This variable is described in Gompers and Metrick (1998)

Taxation of Executive Compensation 25

In addition we include two other variables that proxy for strongercorporate governance The rst is the size of the board Evidence suggeststhat boards with smaller numbers of directors reduce the inuence of theCEO and better represent shareholder interests (Yermack 1995) Boardsize fell by approximately 15 percent between 1980 and 1994 The secondvariable is the fraction of the rmrsquos board members who are inside direc-tors or gray directors (noninsiders who have business dealings with thecompany) Firms with fewer outsiders are less likely to act in the share-holdersrsquo interests and we would predict that they would be less likely tohave performance-based compensation for their CEOs The average per-centage of inside and gray directors has decreased from about 45 percentin 1980 to about 35 percent in 1994

Second we allow for cross-sectional variation in corporate tax ratesCorporate marginal tax rates vary because rms may be eligible for tax-loss carrybacks and carryforwards investment tax credits and the alter-native minimum tax (see Auerbach 1986 Auerbach and Altshuler1990 Auerbach and Poterba 1987) Because these cross-sectional differ-ences imply that different rms should respond differently to a givenchange in tax rates they potentially provide us with a way to isolatethe tax effects Firm-specic corporate marginal tax rates are notori-ously difcult to calculate from publicly-available data We use a tricho-tomous variable that equals zero if the rm has tax-loss carryforwardsand negative earnings in a given year equals 05 times the statutorycorporate rate if the rm has only one of those conditions and equalsthe statutory rate if the rm has neither loss carryforwards nor negativeearnings10

The cross-sectional variation in the corporate tax rate may not be ex-ogenous Firms that perform poorly could face low marginal tax ratesand also be particularly likely or unlikely to provide performance-basedpay Because most of our specications use xed effects the correlationwe must be concerned about is between changes in rm performanceand changes in compensation We deal with this by controlling forlagged rm stock market performance in our regressions This shouldeliminate spurious correlation between tax rates and compensation thatis jointly caused by rm performance

10 Graham (1996) presents evidence that the trichotomous variable performs nearly as wellas his recommended simulated marginal tax rate Plesko (1999) compares marginal taxrates calculated from Compustat data with actual marginal tax rates from rm tax returnsHe nds that simple binary variables are more reliable measures of the marginal tax ratethan is either the Graham or the trichotomous variable We obtain substantively similarresults when we switch from the trichotomous variable to the binary variables recommendby Plesko

26 Hall amp Liebman

63 ResultsThe dependent variable in our regressions is the fraction of a CEOrsquos totalannual compensation (measured as the BlackndashScholes value of stock-option grants plus salary and bonus) that comes from stock-optiongrants We choose this reduced-form specication because our interest isin the composition of CEO compensation However the factors thataffect the composition of compensation might also affect the levels andit will be important to keep this in mind in interpreting our results

Optimal-contracting theory generally makes predictions about the in-centives provided by the total amount of rm stock and stock optionsowned by the CEO We choose to model the ow of option grants ratherthan the ultimate performance sensitivity of the CEOrsquos compensationand stock and stock-option holdings because we have in mind anadjustment-cost model in which it takes time for a CEOrsquos holdings ofstock and stock options to reach the optimal level11 In our regressionswe include two measures of the current performance sensitivity of theCEOrsquos wealth to rm performance as explanatory variables to reect thedistance of the CEOrsquos contract from the optimal contract

Table 3 presents results in which the tax effects are identied solely byvariation in tax rates over time The fraction of annual compensation paidin stock options is regressed on tax variables corporate governance vari-ables contract-theory variables and variables reecting stock market re-turns Both of the regressions in this table include rm xed effects Incolumn (1) the fraction in annual compensation paid in stock options isregressed on the log difference in payoff (the combined payoff to both therm and the CEO) from receiving compensation in stock options ratherthan salary and bonus where the payoff calculations are based on statu-tory marginal tax rates This variable is dened as the log difference be-tween equation (3) and equation (6) in section 4 The coefcient on the taxvariable is large and statistically signicant at the 95-percent level Thepoint estimate implies that a one-percent increase in the payoff differencebetween stock options and salary and bonus results in a 24-percentage-point increase in the share of compensation paid in stock options

The corporate governance variables all have the predicted sign Firmswith a higher fraction of their shares owned by institutional investorsuse more performance-based pay (stock options) Firms with large corpo-rate boards or a large share of inside directors are less likely to use stockoptions

The two contract-theory variables are the dollar change in CEO wealth

11 See Core and Guay (1999) for evidence that rms adjust in this way

TABLE 3Compensation Regressions that Rely on Variation in

Tax Rates over Time

(1) (2)

Tax variablesln (difference in payoff)(a) 2393 (0349)ln (12statutory personal 20015 (0034)

rate)ln (12statutory cap gains 0330 (0171)

rate)ln (12statutory corporate 0399 (0148)

rate)

Corporate governance variablesFraction of rmrsquos shares 0043 (029) 0052 (0028)

owned by institutionalinvestors

Size of the rmrsquos board of 20002 (001) 20003 (0001)directors

Fraction of directors who 20082 (033) 20092 (0030)are inside or gray

Contract-theory variablesD (CEO wealth) per $1000 20261 (0095) 20247 (0087)

in FV(b)

D (CEO wealth) per 10 in 2583 (189) 2554 (201)FV(c)

Volatility(d) 521 (424) 261 (369)ln (rm market value) 0019 (0006) 0019 (0006)

Stock market return variablesLagged annual rm stock 0006 (0009) 0003 (0008)

market returnLagged annual SampP 500 re- 20035 (0023) 20029 (0023)

turn

Regression diagnosticsTotal sample size 5179 5179Firms 427 427Adjusted R2 0313 0311

Dependent variable share of compensation paid in options Standard errors in parentheses Bothregressions in this table include rm xed effects

(a) Log difference in payoff is the log difference in combined rm and CEO after-tax share betweenpaying compensation in options and paying salary

(b) This variable is the dollar change in CEO wealth per $1000 change in rm market value (seeJensen and Murphy 1990)

(c) This variable in the dollar charge in CEO wealth per 10-percent change in rm value (see Hall andLiebman 1998)

(d) Standard deviation of rm market value in trillions

28 Hall amp Liebman

per $1000 change in rm value (Jensen and Murphy 1990) and thedollar change in CEO wealth per 10-percent change in rm value (Halland Liebman 1998) As discussed in Baker and Hall (1998) these twowealth sensitivity measures reect two different concepts of how closelyaligned the managersrsquo incentives are with the interests of the sharehold-ers In the regression they are meant to measure how far the CEOrsquosexisting contract and ownership of stock and stock options is from theoptimal level The negative coefcients on the two variables suggeststhat rms that are below their optimal pay-to-performance sensitivityare more likely to give more stock options in the current period

The second column replaces the tax variable motivated by deferraladvantages with the three statutory rates The corporate after-tax sharehas a positive coefcient as we would expect because the tax advantageof stock options increases as the corporate rate falls The coefcient onthe personal tax rate is small and statistically insignicant The sign ofthe coefcient on the capital gains rate is the opposite of that predictedby tax-avoidance theory

Note that these results rely on time-series variation alone This is not avery convincing test Since the log difference in payoffs mostly rises overthe time period covered in our sample as does the share of compensa-tion paid in options it is possible that the tax variable is simply reectingother factors that were changing over this time period that we have notincluded in our regression Therefore it is important to look at specica-tions that rely on cross-sectional variation in the tax variables

Table 4 contains specications that rely on cross-sectional variation incorporate tax rates The rst column repeats the specication from col-umn (1) of Table 3 but uses the trichotomous measures of rm corporatetax rates in place of the statutory rate in calculating each rmrsquos logdifference in payoff from paying compensation in options rather than insalary and bonus As in the previous table we nd the positive coef-cient on the log difference in payoff just as the theory suggest How-ever the coefcient estimate is now much smaller suggesting that aone-percent increase in the log difference in payoff produces a 01-percentage-point increase in the share of compensation paid in options

Introducing cross-sectional variation in the dependent variable doesnot eliminate the potentially spurious correlation over time between thetax variable and the share of compensation paid in options The secondcolumn of Table 4 isolates the pooled cross-sectional variation in corpo-rate tax rates by adding time dummies and dropping the rm xedeffects In this specication the coefcient on the tax variable is indistin-guishable from zero The coefcient on the institutional variable contin-ues to be large and statistically signicant The third column includes

Taxation of Executive Compensation 29

both rm xed effects and time dummies which tests for a relationshipbetween within-rm changes in the dependent variable and within-rmchanges in each explanatory variable In this specication the coefcienton the tax variable is again indistinguishable from zero The coefcientson the corporate governance variables are also statistically insignicant

The third column contains a specication that treats the Tax ReformAct of 1986 (TRA 86) as a natural experiment For rms and CEOs facingthe statutory corporate tax rate TRA 86 increased the relative payoff topaying compensation in options by 17 percent For a rm facing a zeromarginal tax rate it reduced the relative payoff by 45 percent The reduc-tion occurred because TRA 86 reduced the personal tax rate makingsalary and bonus more attractive while the decreased corporate rate hadno effect on zero-tax-rate rms Thus we would expect rms facing zeroor low marginal tax rates to reduce their use of options after TRA 86while rms facing the statutory rate would be expected to increase themFew rms face zero marginal tax rates year after year Therefore we tryto distinguish between rms that often face low tax rates and rms thatnearly always face the statutory rate We dene low-tax rms as oneswhose average marginal tax rate over 1984 to 1986 was below 02412

Then we ran a difference-in-differences regression to see if high-taxrms increased their use of options more than low-tax rms The interac-tion between post-TRA 86 and high corporate tax rates is the key vari-able The estimated coefcient on this variable is positive but small andnot statistically different from zero at the 95-percent level

It is important to note that in this specication as well the main corpo-rate governance variable the share of the rmrsquos shares owned by institu-tional investors performs as theory would predict The variable appearsto be quite robust Table 5 presents regressions of the share of compensa-tion paid in stock options on the fraction of the rmrsquos shares owned byinstitutional investors Since we have already established that these vari-ables have a strong longitudinal relationship the regressions here aredesigned to focus on the cross-sectional relationship between these twovariables The rst and third columns present regressions using data onlyfrom 1994 and nd a coefcient of roughly 02 (column 1 contains noother covariates while column 3 includes all of the other nontax covari-ates from the regressions in Tables 3 and 4) Since the median share of armrsquos stock owned by institutional investors in our sample rises from 19percent in 1980 to 49 percent in 1994 this coefcient would imply that this

12 This cutoff was chosen after inspecting the distribution of tax rates and essentiallyseparates rms that are always at or near the statutory rate from ones that are sometimesbelow it

TA

BL

E4

Com

pens

atio

nR

egre

ssio

nsth

atR

ely

onC

ross

-Sec

tion

alV

aria

tion

inC

orpo

rate

Tax

Rat

es

(1)

(2)

(3)

(4)

Tax

vari

able

s(a)

ln

(dif

fere

nce

inp

ayof

f)0

099

(00

41)

20

038

(00

36)

20

005

(00

44)

Pos

t-T

RA

862

000

5(0

013

)H

igh

corp

orat

eta

xra

te0

018

(00

17)

(Pos

t-T

RA

86)

3hi

gh0

024

(00

16)

corp

orat

eta

xra

te

Cor

por

ate

gove

rnan

ceva

riab

les

Frac

tion

of

rmrsquos

shar

es0

109

(00

27)

011

4(0

015

)2

000

9(0

030

)0

087

(00

34)

owne

dby

inst

itu

tion

alin

vest

ors

Size

ofth

e

rmrsquos

boar

dof

20

004

(00

01)

20

001

(00

01)

20

002

(00

01)

000

1(0

002

)d

irec

tors

Frac

tion

ofd

irec

tors

wh

o2

012

4(0

033

)2

000

3(0

018

)2

004

1(0

034

)0

001

(00

39)

are

insi

de

orgr

ay

Con

trac

t-th

eory

vari

able

sD

(CE

Ow

ealt

h)p

er$1

000

20

238

(00

94)

20

167

(00

53)

20

257

(00

96)

20

021

(01

47)

inFV

(b)

D(C

EO

wea

lth)

per

10

in2

560

(17

8)2

592

(10

5)2

594

(19

6)2

147

(04

86)

FV(c

)

Vol

atili

ty(d

)2

48(4

24)

162

(03

4)1

12(4

28)

106

(06

09)

ln(

rmm

arke

tva

lue)

003

3(

006)

000

8(0

003

)0

009

(00

06)

001

2(0

007

)

Stoc

km

arke

tre

turn

vari

able

sL

agge

dan

nual

rm

stoc

k2

000

2(0

009

)0

011

(00

10)

000

6(0

009

)2

001

2(0

013

)m

arke

tre

turn

Lag

ged

annu

alSamp

P50

02

005

6(0

023

)2

005

3(0

051

)2

003

9(0

049

)2

015

4(0

040

)re

turn

Reg

ress

ion

dia

gnos

tics

T

otal

sam

ple

size

5179

5179

5179

2534

Firm

s42

742

739

6A

dju

sted

R2

031

009

032

005

Firm

xed

effe

cts

Yes

No

Yes

No

Tim

ed

um

mie

sN

oY

esY

esN

o

Dep

end

ent

vari

able

isfr

acti

onof

ann

ualc

omp

ensa

tion

from

stoc

kop

tion

sSt

and

ard

erro

rsin

par

enth

eses

(a)

All

thre

esp

eci

cati

ons

use

tric

hoto

mou

sco

rpor

ate

tax

rate

sT

he

sam

ple

inco

lum

n(3

)u

ses

only

the

year

s19

84to

1988

sin

cew

ear

eu

sing

1986

asa

nat

ural

exp

erim

ent

(b)

Thi

sva

riab

leis

the

dol

lar

chan

gein

CE

Ow

ealt

hp

er$1

000

chan

gein

rm

mar

ketv

alu

e(s

eeJe

nse

nan

dM

urp

hy

1990

)

(c)

Thi

sva

riab

lein

the

dol

lar

char

gein

CE

Ow

ealt

hp

er10

-per

cent

chan

gein

rm

valu

e(s

eeH

alla

nd

Lie

bman

199

8)

(d)

Stan

dar

dd

evia

tion

ofr

mm

arke

tva

lue

intr

illio

ns

32 Hall amp Liebman

TABLE 5Large Institutional Investors and Stock-Option Compensation

The Cross-sectional Relationship between Institutional Investors andOption Grants

(1) (2) (3) (4)

Coefcient on fraction of 217 (067) 145 (014) 177 (069) 120 (015)rmrsquos shares owned byinstitutional investors

Sample period 1994 1980ndash1994 1994 1980ndash1994Contract-theory and stock No No Yes Yes

market covariatesTime dummies No Yes No Yes

The dependent variable is the share of compensation paid in options

factor alone can explain 6 percentage points of the 23-percentage-pointincrease in the median share of compensation paid in options that oc-curred over this time period Of course this variable is only one measureof the strengthening of corporate governance so the total impact of im-proved corporate governance could be greater The second and fourthcolumns of the table use the entire 1980ndash1994 sample and include timedummies to isolate the cross-sectional correlations Again there is evi-dence of a strong cross-sectional relationship between the two variablesalthough the coefcients are slightly smaller perhaps reecting a weakercorporate-governance role in the earlier period In sum there is a strongrelationship in both longitudinal and cross-sectional data between theinstitutional-investor variable and the options-percentage variable13

7 THE MILLION-DOLLAR RULE

In 1993 section 162(m) of the Internal Revenue Code was enactedlimiting the deductibility of executive compensation in excess of onemillion dollars unless the compensation was performance-related Pro-ponents of this legislation argued that it would slow the growth ofexecutive pay and tighten the link between rm performance and execu-tive compensation

The new provision became effective for tax year 1994 and applies tothe CEO and the other four most highly compensated executives in eachcompany Pay that is performance-related such as stock-option grants

13 The one specication in which the fraction owned by institutional investors is notsignicant is in column (2) of Table 4 a regression that includes both xed effects and timeeffects

Taxation of Executive Compensation 33

and bonuses paid for meeting clear performance goals is not affected bythis regulation provided that it has been approved by shareholdersThus the main impact of the provision is to limit the deductibility ofsalary in excess of a million dollars Woodlock and Antenucci (1997)studied the proxy statements of 376 rms and documented that mostrms responded to the new law by qualifying top executive pay asperformance-related Perry and Zenner (1999) identied 25 rms thatreduced salaries from above $1 million to below $1 million and foundthat 23 of the rms cited section 162(m) as their reason for doing so

In this section of the paper we investigate whether this provision hasaffected the level or structure of executive pay using the Execucompdata base which contains information on the highest-paid executives ineach of the rms in the SampP 500 SampP Midcap 400 and SampP SmallCap600 Between 1993 and 1998 the median salary of the Execucomp execu-tives rose from $230000 to $306000 The median value of total compen-sation (salary bonus and option grants) rose from $470000 to $882000Limiting the sample of CEOs indicates that the total compensation of themedian CEO rose from $11 million to $18 million over the period whilethe median salary rose from $441000 to $550000 Thus executive com-pensation clearly continued to rise at a rapid rate after the implementa-tion of the million-dollar rule though perhaps at a slower rate than itotherwise would have

If the regulation affects behavior it should have the largest impact onrms whose executives are receiving salaries that are above the limit orjust below it Thus we might expect to see rms whose executives arereceiving relatively high salaries to rely more heavily on performance-based pay and less heavily on salary in the years following the regulationPerry and Zenner (1999) and Livingston (1999) both nd that this is thecase This trend can be clearly identied in the Execucomp data set Forexample over 1993 to 1998 the median annualized growth rate of salaryfor executives whose 1993 salary was below $500000 in the previous yearwas 85 percent For executives with salary between $500000 and$700000 it was 64 percent for executives between $700000 and $850000it was 50 percent for executives between $750000 and $1 million it was31 percent and for executives above $1 million it was 00 percent14

The observation that executives at lower levels of salary had highergrowth rates is not very strong evidence of an impact of section 162(m)however because it is possible that low-salary executives would havehad higher growth rates even in the absence of the provision To explore

14 Of the 3137 executives for whom we have complete data between 1993 and 1998 2708had 1993 salaries below $500000 while only 35 had 1993 salaries above $1 million

34 Hall amp Liebman

whether this is the case we use Execucomp data from both before andafter the law change Ideally we would want to have data from manyyears before the change to examine whether there is an underlying pat-tern of faster salary growth for low-salary executives Unfortunately theExecucomp data begin in 1992 and therefore provide us with only the1992ndash1993 growth rate in compensation to use as a baseline15

We run regressions pooling data from 1992ndash1993 through 1997ndash1998of the annual percentage change in different forms of compensation on avariable million that is designed to measure the likelihood that anexecutiversquos compensation will be affected by section 162(m) We denemillion as the minimum of 1 and the executiversquos salary divided by $1million in the previous year Thus any executive with salary at or above$1 million would receive a value of 1 while an executive with a salary of$500000 would be coded as a 0516 We include this variable by itself andalso interacted with a dummy variable (after) that equals 1 for years1994 and after Thus the million variable is designed to pick up anyunderlying relationship between the level of salary and the growth rateof salary while the interacted one will identify any differential growthrate of salary by income in the period after section 162(m) took effect Wealso include a full set of year dummies the natural logarithm of thermrsquos market value and the rmrsquos annual rate of return over the twoprevious years as control variables To reduce the sensitivity of our re-sults to outliers we run quantile regressions and robust regressions

Table 6 contains our results The rst four columns are for all topexecutives while the last four columns focus on CEOs Columns (1) and(2) present estimates from robust regression for the percentage change insalary and the percentage change in total compensation (salary bonusstock grants and stock-option grants) In the salary growth regressionsfor all executives the coefcient on million is negative and signicantindicating that there is an underlying relationship between the level ofexecutive compensation and its subsequent growth rate The coefcienton million 3 after is negative as well suggesting that the negativerelationship between the level of salary and its subsequent growth rateintensied slightly after 1994 The relatively small magnitude suggeststhat an executive with a $1 million dollar salary would see his salarygrow at an annual rate that would be about 06 percent slower than anexecutive earning $500000

15 The HallndashLiebman data set contains only a combined salary and bonus variable makingit impossible to use it to analyze this issue16 We also ran the regressions (not reported) allowing for a nonlinear relationship betweenmillion and compensation growth and obtained substantively similar results

TA

BL

E6

The

Eff

ect

ofth

eM

illi

on-D

olla

rR

ule

onSa

lary

and

Tot

alC

ompe

nsat

ion

All

top

exec

uti

ves

CE

Os

only

Rob

ust

regr

essi

onQ

uan

tile

regr

essi

onR

obu

stre

gres

sion

Qu

anti

lere

gres

sion

D

in

Din

D

in

Din

D

in

Din

D

in

Din

sala

ryto

talc

omp

(a)

sala

ryto

talc

omp

sa

lary

tota

lcom

p

sala

ry

tota

lcom

p

mil

lio

n(b

)2

006

02

035

72

006

32

024

12

008

72

023

12

010

32

024

5(0

005

)(0

032

)(0

006

)(0

027

)(0

010

)(0

073

)(0

011

)(0

056

)m

illi

on

32

001

10

026

20

021

20

015

20

016

014

92

003

40

158

aft

er(c

)(0

005

)(0

033

)(0

006

)(0

029

)(0

010

)(0

076

)(0

011

)(0

057

)N

359

5035

950

359

5035

950

757

17

589

757

17

589

Stan

dar

der

rors

inp

aren

thes

esR

egre

ssio

nsal

soin

clu

de

year

du

mm

ies

stoc

km

arke

tre

turn

and

log

mar

ket

valu

e

(a)

Tot

alco

mp

ensa

tion

incl

ud

essa

lary

bon

us

rest

rict

edst

ock

and

stoc

k-op

tion

gran

ts

(b)

mil

lion

isd

ene

das

min

(1s

alar

yin

pre

viou

sye

ars

div

ided

by$1

mil

lion)

(c)

aft

eris

anin

dic

ator

vari

able

equ

alto

zero

befo

reth

ere

gula

tion

took

plac

e(1

993)

and

one

afte

rwar

d

36 Hall amp Liebman

In the second column the coefcient on million 3 after is essentiallyzero suggesting that any decrease in salary brought about by section162(m) was offset by increases in bonus and stock-option grants Theresults are similar for the median regressions There is some evidence ofa very small slowdown in growth of salary for executives likely to beaffected by section 162(m) but no sign of a change in total compensa-tion Taken together this provides evidence of a very small substitutionof performance-related pay for salary

The last four columns run identical tests but limit the sample to CEOsThe interaction of million 3 after continues to be small and negative inthe salary regressions In contrast the impact of 162(m) on total compen-sation appears to be positive and fairly substantial (about 15 percentagepoints per year) in these regressions That is the increase in bonus andstock options following the million-dollar rule more than offset the verysmall decline in salary

All of these conclusions should be interpreted somewhat cautiouslybecause our data provide us with only one pre-1993 control year Never-theless the data are consistent with section 162(m) having led to a veryminor slowdown in salary growth and one that is dramatically smallerthan what one would conclude without controlling for underlying differ-ential salary growth rates at different levels of salary Since total compen-sation for CEOs did not decline and perhaps increased there is evidenceof a minor substitution of performance-related pay for salary in responseto the regulation

8 EXECUTIVE COMPENSATION AND THEELASTICITY OF TAXABLE INCOME

Even if taxes have only a minor effect on executive compensation thestructure of executive compensation is important for tax policy be-cause high-income taxpayers are sources of signicant revenue and arethe focus of important debates about the efciency cost of taxation

Goolsbee (1999) has recently argued that executive decisions to exer-cise options are highly responsive to intertemporal differences in taxrates created by tax reforms He claims that much of the apparent declinein taxable income for high-income taxpayers between 1992 and 1993documented by Feldstein and Feenberg (1996) can be explained by execu-tives shifting option gains into 1992 in order to avoid the higher marginaltax rates that went into effect in 1993 By extension he implies that thehigh elasticities of taxable income with respect to personal tax ratesestimated by comparing the pre- and post-tax-reform taxable income of

Taxation of Executive Compensation 37

high-income individuals reect intertemporal shifting of income and notpermanent effects of taxation If this is true then the deadweight loss oftaxation is lower as is the cost of progressivity

While Goolsbee presents comprehensive evidence that option gainswere unusually high in 1992 his identication of the impact of tax rates onthe timing of option gains depends on correctly specifying a counter-factual level of options that would have been exercised in the absence ofthe tax change He simply assumes a linear time trend The purpose ofGoolsbeersquos time trend is presumably to account for the increasing use ofoptions over time However since options typically require a vestingperiod of at least a couple of years a much more direct measure of ex-pected option gains can be constructed based on the total value of vestedoptions held by the CEO Since options typically have vesting periods thatexceed two years this total can be treated as exogenous since it will not beaffected by the tax changes

More generally the value of options held by the CEO is likely to havea major effect on his exercise decisions If his options are out of themoney he will not have gains to realize If there has been a recent run-up in the market an optimizing CEO may exercise an unusually largeamount of options in order to diversify his overall portfolio and if in-the-money options are about to expire they will be exercised in thecurrent year

Our evidence suggests that much of the apparent tax shifting caninstead be attributed to stock market performance and the timing of pastoption grants Using our 15-year panel we replicate Goolsbeersquos (1999)result for the 1993 period but show that the impact of taxes is notpresent in other time periods Moreover even the result for the 1993period disappears when appropriate controls for past option grants andstock price appreciation are included

The top three rows of Table 7 present data on the taxable incomes ofCEOs in the HallndashLiebman sample For these results our panel has beennarrowed to include only CEOs whose rmrsquos scal years correspondwith tax years In addition we create a balanced panel for each of thethree tax-reform periods by excluding any rm for which there are notcomplete data for the period surrounding the reform

The fourth row of the table shows that the 1993 tax act raised the topmarginal personal income tax rate from 310 to 396 percent effective fortax year 1993 with a further increase occurring in 1994 due to the uncap-ping of the Medicare payroll tax The top row of the table shows thattaxable income was higher in 1992 than for any other year in our sample(all data are in 1994 dollars) Option gains were particularly high in1992mdashmore than double their 1991 level and 68 percent higher than their

TA

BL

E7

Tax

Ref

orm

Sto

ckM

arke

tP

erfo

rman

cea

ndth

eT

imin

gof

Opt

ion

Gai

ns

Ave

rage

stoc

km

arke

tM

ean

CE

OM

ean

CE

Ore

turn

for

Mea

nva

lue

taxa

ble

sala

ryan

dM

ean

CE

OT

opfe

der

al

rms

inof

opti

ons

(Val

ue

ofop

tion

Sam

ple

inco

me

bonu

sop

tion

gain

sp

erso

nal

sam

ple

for

held

atbe

gga

ins)

(va

lue

Ref

orm

size

Yea

r($

mil

)($

mil

)($

mil

)M

TR

pre

vye

ar(

)of

year

ofop

tion

she

ld)

ER

TA

8127

919

800

874

064

80

163

7025

80

460

354

1981

083

10

672

009

869

245

070

61

3919

820

872

067

90

099

5011

90

686

144

1983

099

00

724

025

250

355

116

62

16

TR

A86

295

1985

119

70

837

024

050

76

116

02

0719

861

448

091

70

345

5042

81

776

194

1987

165

50

959

059

338

519

12

229

266

1988

163

21

058

039

028

07

237

71

6419

891

656

105

20

414

2817

92

345

176

1993

272

1990

209

51

028

061

828

232

288

02

15T

axA

ct19

911

875

102

00

547

312

01

268

82

0319

922

678

107

01

250

3146

64

552

275

1993

232

41

209

074

339

620

64

299

173

1994

196

41

262

037

842

515

45

113

074

Sam

ple

isH

allndash

Lie

bman

rm

sw

ith

Dec

embe

r

scal

year

sth

atha

dco

mp

lete

dat

afo

ron

eof

the

thre

eta

x-re

form

per

iod

sst

ud

ied

Taxation of Executive Compensation 39

1994 level These patterns are consistent with Goolsbeersquos claim that tax-payers responded to the anticipated increase in marginal tax rates byshifting gains into 1992

The evidence from the other two tax reforms is less clear suggestingthat there might be factors other than tax rates determining the timing ofoption gains ERTA 1981 reduced the top marginal tax rate from 70percent in 1980 to 50 percent in 1982 However there is no sign thatoption gains were reduced in 1980 and 1981 and then increased in 198217

There was a big increase in option gains in 1983 however Similarly thehighest year of stock-option gains in the period around 1986 was 1987 astrange year to take gains since taxpayers would have known that thepersonal rate would be 10 percentage points lower if they waited untilthe following year18

What all the peak years of option gains have in common is that theyfollow years of strong stock market performance The fth row of Table 7shows the average stock market return during the previous year for therms in our sample The large value of option gains in 1983 followed a355-percent increase in the stock market The 1992 boom in option gainsfollowed a 466-percent increase in the stock market the more modest1987 peak in option gains followed a 19-percent gain (and a 43-percentgain two years before)

Strong stock market performance raises the value of options availableto exercise The sixth row of the table shows the mean value of optionsheld by CEOs at the beginning of the year All three of the peak years ofoption gains correspond with peak years in terms of the value of optionsavailable for exercise Since stock market appreciation during the yearalso increases the value of options available to exercise later in that yearwe need to allow for current-year stock market appreciation as well

Table 8 contains Goolsbee-style regressions for each of the tax re-forms The rst two columns reprint the results from Goolsbee (1999)The third and fourth columns replicate his results in our sample Inparticular with a specication similar to his we are able to replicate hisnding that the current after-tax share has a positive coefcient and thefollowing yearrsquos after-tax share has a coefcient of roughly negative oneColumns (5) through (8) show that this pattern was unique to the 1993tax reform Indeed in the TRA 86 regressions the coefcient on the

17 As we explained above for executives covered by the maximum tax on earned incomethere was not a decline in tax rates between 1981 and 198218 It is possible that the large stock market decline in October 1997 may have promptedsome people to exercise options earlier than they otherwise would have However anincrease in option exercises typically follows stock market increases not declines

TA

BL

E8

Att

empt

sto

Rep

lica

teth

eT

ax-S

hift

ing

Res

ult

inO

ther

Tim

eP

erio

ds

Goo

lsbe

eG

ools

bee

1993

1993

Ent

ire

Ent

ire

regr

essi

onre

gres

sion

refo

rmre

form

TR

A86

TR

A86

ER

TA

81E

RT

A81

1980

ndash199

419

80ndash1

994

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

ln[1

2ta

x(t)

]1

244

108

20

455

032

62

017

62

017

22

016

72

016

70

026

20

068

(01

34)

(01

27)

(05

37)

(05

42)

(02

10)

(02

09)

(00

90)

(00

91)

(00

67)

(00

65)

ln[1

2ta

x(t1

1)]

20

867

20

764

21

214

20

917

047

80

722

20

103

004

30

176

018

3(0

118

)(0

112

)(0

491

)(0

499

)(0

157

)(0

174

)(0

072

)(0

072

)(0

068

)(0

066

)

Tim

e0

135

007

62

000

52

036

40

041

000

80

094

003

50

056

002

7(0

008

)(0

008

)(0

049

)(0

049

)(0

028

)(0

028

)(0

025

)(0

026

)(0

003

)(0

003

)

ln(m

arke

tva

lue)

066

10

489

023

70

334

039

5(0

015

)(0

058

)(0

039

)(0

042

)(0

017

)

Yea

rs19

91ndash1

995

1991

ndash199

519

90ndash1

994

1990

ndash199

419

85ndash1

989

1985

ndash198

919

80ndash1

983

1980

ndash198

319

80ndash1

994

1980

ndash199

4

Ad

just

edR

20

720

770

610

630

690

760

800

810

630

68

Obs

erva

tion

s13

856

132

371

385

136

41

470

143

31

113

106

23

968

385

9

Dep

end

ent

vari

able

isth

elo

gof

taxa

ble

inco

me

All

regr

essi

ons

con

tain

rm

xe

def

fect

sG

ools

bee

regr

essi

on(2

)al

soco

ntai

nstw

oot

her

vari

able

sno

tsh

own

inth

eta

ble

Stan

dar

der

rors

inp

aren

thes

es

Taxation of Executive Compensation 41

TABLE 9Alternative Explanations for the Timing of Option Exercises

1993 Entirereform TRA 86 ERTA 81 1980ndash1994

(1) (2) (3) (4)

ln[12tax(t)] 0108 0111 20051 0201(0324) (0136) (0048) (0057)

ln[12tax(t11)] 20134 0573 0125 0150(0336) (0156) (0053) (0066)

ln(market value) 0418 0177 0338 0418(0061) (0039) (0046) (0163)

Firmrsquos stock market return (t) 0298 0268 0214 0258(0044) (0045) (0031) (0023)

Firmrsquos stock market return 0162 0206 0031 0128(t21) (0041) (0008) (0029) (0021)

ln(value of options held by 0004 0008 0001 0009CEO at start of year) (0003) (0003) (0002) (0001)

Years 1990ndash1994 1985ndash1989 1980ndash1983 1980ndash1994Adjusted R2 064 071 082 068Observations 1349 1427 1036 3812

The dependent variable is the log of taxable income All regressions contain rm xed effects Stan-dard errors in parentheses

following yearrsquos tax rate is positive and signicant Columns (9) and (10)pool all three reforms and fail to replicate the pattern that would reectintertemporal shifting19

It is possible that the ease with which one can shift the timing ofoption gains is asymmetric It may be harder to shift option gains for-ward in time since some options expire and therefore must be exercisednow In addition options were a much smaller portion of compensationin the early 1980s so perhaps the ERTA 81 results are not surprisingNonetheless these results cast doubt on the proposition that this is thechannel through which spurious estimates of the elasticity of taxableincome have occurred for previous tax reforms and they suggest thatwe should look harder for alternative explanations for the timing ofoption gains

Table 9 attempts to predict options gains using the lagged and currentincrease in a rmrsquos market value and the value of the options held by theCEO at the start of the year The two stock market variables stronglypredict option gains while the value of options held has the correct sign

19 The results also fail to suggest large permanent effects of marginal tax rates on taxableincome

42 Hall amp Liebman

in all of the specications and is signicant in the pooled sample eventhough it is highly correlated with the two stock market return variablesIn contrast the tax-timing explanation fails even in the 1993 period

It is important to realize that our results do not invalidate Goolsbeersquoscritique of claims that aggregate revenue data from 1992 and 1993 provethat elasticities of taxable income are large We have simply given analternative explanation for why taxable incomes were unusually high in1992 However our results do cast doubt on the conclusion that largetiming shifts are ubiquitous and capable of explaining away the morecareful estimates of taxable income elasticities that have been performedusing micro data (eg Feldstein 1995 Auten and Carroll 1995)

One further point is in order It is sometimes argued that the combina-tion of increased use of options and the booming stock market explainswhy federal tax revenues have been so much higher than expected dur-ing the 1990s However while large option gains increase revenue fromthe personal income tax they produce offsetting deductions for cor-porations The net impact on revenue is only the difference in the twomarginal tax rates times the option gain While the gap between thepersonal rate and the effective corporate rate is nontrivial the impact oflarge option exercises on total revenue is substantially smaller than theamount by which the exercises inate personal tax revenue and reducecorporate tax revenue

9 CONCLUSION

We have described the tax rules for executive pay in detail and analyzedhow changes in various tax rates affect the tax advantages of stock optionsrelative to salary and bonus We nd that there is a moderate tax advan-tage to optionsmdashcurrently about $4 per $100 in pretax executive paymdashand that changes in the tax advantage of options over time have had atmost a modest impact on the composition of pay Corporate governancefactors particularly the role of large investors appear to be more impor-tant in explaining the dramatic increase in option pay

In addition our evidence suggests that more direct attempts to use taxpolicy to inuence executive compensation have had little effect Wend that the million-dollar rule led companies to substitute perfor-mance-based pay for salary But our evidence suggests that this substitu-tion was quite modest and there is no evidence that the total level of paywas reduced Overall although the stock-option explosion has dramati-cally increased the link between pay and performance this change isdue almost entirely to nontax factors

Taxation of Executive Compensation 43

REFERENCES

Auerbach Alan J (1996) ldquoThe Dynamic Effects of Tax Law Asymmetriesrdquo Re-view of Economic Studies 53451ndash405

mdashmdashmdash and James Poterba (1987) ldquoTax Loss Carryforwards and Corporate TaxIncentivesrdquo In The Effects of Taxation on Capital Accumulation M Feldstein(ed) Chicago University of Chicago Press

Altshuler Rosanne and Alan Auerbach (1990) ldquoThe Signicance of Tax LawAsymmetries An Empirical Investigationrdquo Quarterly Journal of Economics 10561ndash86

Auten Gerald and Robert Carroll (1995) ldquoThe Effect of Income Taxes on House-hold Behaviorrdquo Ofce of Tax Analysis US Department of the Treasury

Baker George P and Brian J Hall (1998) ldquoCEO Incentives and Firm SizerdquoNBER Working Paper no 6868 December

Core John and Wayne Guay (1999) ldquoThe Use of Equity Grants to ManageOptimal Equity Incentive Levelsrdquo Draft

Feldstein M (1995) ldquoThe Effect of Marginal Tax Rates on Taxable Income APanel Study of the 1986 Tax Reform Actrdquo Journal of Political Economy 103 (no3)551ndash72

mdashmdashmdash and D Feenberg (1996) ldquoThe Effect of Increased Tax Rates on TaxableIncome and Economic Efciency A Preliminary Analysis of the 1993 Tax RateIncreasesrdquo In Tax Policy and the Economy vol 10 James M Poterba (ed)Cambridge MA MIT Press

Gompers Paul A and Andrew Metrick (1998) ldquoInstitutional Investors andEquity Pricesrdquo NBER Working Paper no 6723

Goolsbee Austan (1999) ldquoWhat Happens When You Tax the Rich Evidencefrom Executive Compensationrdquo NBER Working Paper no 6333

Graham John R (1996) ldquoProxies for the Corporate Marginal Tax Raterdquo Journal ofFinancial Economics 42187ndash221

Hall Brian J (1999) ldquoThe Design of Multi-year Stock Option Plansrdquo Journal ofApplied Corporate Finance 12(2)97ndash106

mdashmdashmdash and Jeffrey B Liebman (1998) ldquoAre CEOs Really Paid Like Bureau-cratsrdquo Quarterly Journal of Economics CXII(no 3)653ndash691

Jensen Michael C and William M Meckling (1976) ldquoTheory of the Firm Mana-gerial Behavior Agency Costs and Ownership Structurerdquo Journal of FinancialEconomics III305ndash360

mdashmdashmdash and Kevin J Murphy (1990) ldquoPerformance Pay and Top-ManagementIncentivesrdquo Journal of Political Economy XCVIII225ndash264

Kaplan Steven N (1997) ldquoThe Evolution of US Corporate Governance WeAre All Henry Kravis Nowrdquo Draft

Livingstone Jane R (1999) ldquoExecutive Compensation Contracting and Re-sponses to Increased Tax Costsrdquo Louisiana State University June

Miller Merton H and Myron S Scholes (1982) ldquoExecutive CompensationTaxes and Incentivesrdquo In Financial Economics Essays in Honor of Paul CootnerWilliam F Sharpe and Cathryn M Cootner (eds) Princeton NJ Prentice-Hall

Murphy Kevin J (1999) ldquoExecutive Compensationrdquo forthcoming in OrleyAshenfelter and David Card (eds) Handbook of Labor Economics Vol 3 NorthHolland

Perry Todd and Marc Zenner (1999) ldquoPay for Performance Government Regu-

22 Hall amp Liebman

The same payoff for ISOs is

P (1 2 Tc) [1 1 r (1 2 Tc)]N (1 2 Tcg) (8)

ISOs therefore are tax-advantaged if

Tp 2 Tcg Tc 2 Tc Tcg (9)

Intuitively ISOs are tax-advantaged only if their advantage (the differ-ence between the personal rate and the capital gains rate) is largeenough to offset their cost (the disadvantage of not deducting at thecorporate rate) Note however that the condition is not a simple com-parison between the corporate rate and the personal rate minus thecapital gains rate (Tp 2 Tcg Tc)

Figure 5 shows how the tax advantage of NQSOs relative to ISOs haschanged since 1980 as tax rates have changed Unlike the very modestchanges (around 3 to 4 percent) in the relative tax advantages betweenoptions and cash shown in Figure 4 the relative tax advantage betweenNQSOs and ISOs has seen enormous swings When the top personalrate was 70 percent in 1980 ISOs were tax-favored by a margin greaterthan 6 percent However by 1982 (following the 1981 tax acts that dra-matically lowered the personal tax rate) ISOs became tax-disadvantaged

FIGURE 5 Tax Advantage of NQSOs Relative to ISOs over Time

NPV of tax advantage of $100 payment to executive NQSOs minus ISOs

Taxation of Executive Compensation 23

by about 5 percent By 1988 (following the 1986 tax act) ISOs became tax-disadvantaged by about 18 percent a dramatic swing The raising ofpersonal rates coupled with a decrease in the capital gains rate has sincereduced the tax disadvantage of ISOs to about $4 per $100 of pretaxcompensation Although hard data on ISOs are hard to come by practi-tioners (mostly compensation consultants) claim that ISOs were morepopular prior to 1982 when they were tax-advantaged Although ISOsare less disadvantageous from a tax standpoint relative to NQSOs thanthey were in the late 1980s they are still disadvantageous so it is notsurprising that they have not made a signicant resurgence in recentyears

6 EXPLAINING THE INCREASE IN OPTION-BASEDCOMPENSATION

In this section we analyze whether changes in tax rates affect the compo-sition of executive compensation and consider the relative importanceof tax factors and corporate governance factors in explaining the in-creasing share of compensation paid in the form of stock options Thenumerous changes in tax rates that have occurred since 1980 provide anopportunity to assess whether executive compensation arrangementsrespond to tax incentives in the way that the tax-avoidance model sug-gests We test this model using a panel data set of CEOs in largepublicly-traded US corporations Our identication of the tax effectsrelies on time-series variation in personal corporate and capital gainstax rates along with cross-sectional variation in corporate tax rates

61 DataWe use a 15-year panel data set of CEOs in the largest publicly-tradedUS corporations which is described in Hall and Liebman (1998) Thedata set covers the years from 1980 through 1994 and combines CEOcompensation information from corporate proxies and 10-K lings withstock price and stock return information from CRSP and accounting datafrom Compustat In addition some compensation data from the 1970swere collected in order to construct measures of the value of stock op-tions held by the CEOs in the rst period of the sample8

The feature that distinguishes our data from most other CEO datasets is that with our panel of yearly proxy data on option grants optiongains and total options held we are able to calculate the total value of

8 We are grateful to David Yermack for providing us with some of the data for 1984 to 1991See Yermack (1995) for a discussion of these data

24 Hall amp Liebman

all stock options held by the CEO at a point of time More importantlysince we have the details about the stock options held (number exer-cise price time to maturity etc) we can precisely calculate the changein the value of a CEOrsquos stock option holdings for a given change in rmvalue

62 Identication IssuesThe numerous tax changes over our sample period and the specic waysin which changes in personal corporate and capital gains rates arepredicted to affect the composition of CEO compensation give us un-usually rich sources of identication However because many of thechanges in tax rates would be expected to affect all of the CEOs in oursample in a similar way we will need to pay particular attention toseparating out the tax effects from underlying time trends and otherfactors that changed over time Controlling for underlying trends is par-ticularly important in this study because we are focusing on an out-comemdashthe use of stock optionsmdashthat has increased rapidly over thepast two decades and that many practitioners believe was importantlyaffected by nontax considerations Thus the tax effects we are trying toexplain are deviations from a rising trend

We take two steps to try to separate out the tax effects from theunderlying trend First we include key nontax factors that could poten-tially explain the increasing reliance on performance-based compensa-tion It has been suggested that the dramatic increase in incentive-basedpay is the result of the remarkably poor shareholder returns during the1970s which spurred the LBO and takeover movements of the 1980sAccording to this story (Kaplan 1997) shareholders became much morepowerful via the rise of institutional investors even as the LBO andtakeover movements waned Because the inuence of institutional inves-tors is thought to be one of the most important mechanisms of strongcorporate governance we use the share of each companyrsquos stock ownedby institutional investors (dened as institutions with more than $100million under management) as an explanatory variable9 The percentageof shares owned by large institutional investors increased from about 20percent to almost 50 percent in 1994 an upward trend that closelymatches the sharp rise in the share of compensation in the form ofoptions Indeed in our sample the annual average (over all of the rmsin our sample) share of stock owned by institutional investors has acorrelation of 09 with the annual average share of compensation paid inoptions

9 This variable is described in Gompers and Metrick (1998)

Taxation of Executive Compensation 25

In addition we include two other variables that proxy for strongercorporate governance The rst is the size of the board Evidence suggeststhat boards with smaller numbers of directors reduce the inuence of theCEO and better represent shareholder interests (Yermack 1995) Boardsize fell by approximately 15 percent between 1980 and 1994 The secondvariable is the fraction of the rmrsquos board members who are inside direc-tors or gray directors (noninsiders who have business dealings with thecompany) Firms with fewer outsiders are less likely to act in the share-holdersrsquo interests and we would predict that they would be less likely tohave performance-based compensation for their CEOs The average per-centage of inside and gray directors has decreased from about 45 percentin 1980 to about 35 percent in 1994

Second we allow for cross-sectional variation in corporate tax ratesCorporate marginal tax rates vary because rms may be eligible for tax-loss carrybacks and carryforwards investment tax credits and the alter-native minimum tax (see Auerbach 1986 Auerbach and Altshuler1990 Auerbach and Poterba 1987) Because these cross-sectional differ-ences imply that different rms should respond differently to a givenchange in tax rates they potentially provide us with a way to isolatethe tax effects Firm-specic corporate marginal tax rates are notori-ously difcult to calculate from publicly-available data We use a tricho-tomous variable that equals zero if the rm has tax-loss carryforwardsand negative earnings in a given year equals 05 times the statutorycorporate rate if the rm has only one of those conditions and equalsthe statutory rate if the rm has neither loss carryforwards nor negativeearnings10

The cross-sectional variation in the corporate tax rate may not be ex-ogenous Firms that perform poorly could face low marginal tax ratesand also be particularly likely or unlikely to provide performance-basedpay Because most of our specications use xed effects the correlationwe must be concerned about is between changes in rm performanceand changes in compensation We deal with this by controlling forlagged rm stock market performance in our regressions This shouldeliminate spurious correlation between tax rates and compensation thatis jointly caused by rm performance

10 Graham (1996) presents evidence that the trichotomous variable performs nearly as wellas his recommended simulated marginal tax rate Plesko (1999) compares marginal taxrates calculated from Compustat data with actual marginal tax rates from rm tax returnsHe nds that simple binary variables are more reliable measures of the marginal tax ratethan is either the Graham or the trichotomous variable We obtain substantively similarresults when we switch from the trichotomous variable to the binary variables recommendby Plesko

26 Hall amp Liebman

63 ResultsThe dependent variable in our regressions is the fraction of a CEOrsquos totalannual compensation (measured as the BlackndashScholes value of stock-option grants plus salary and bonus) that comes from stock-optiongrants We choose this reduced-form specication because our interest isin the composition of CEO compensation However the factors thataffect the composition of compensation might also affect the levels andit will be important to keep this in mind in interpreting our results

Optimal-contracting theory generally makes predictions about the in-centives provided by the total amount of rm stock and stock optionsowned by the CEO We choose to model the ow of option grants ratherthan the ultimate performance sensitivity of the CEOrsquos compensationand stock and stock-option holdings because we have in mind anadjustment-cost model in which it takes time for a CEOrsquos holdings ofstock and stock options to reach the optimal level11 In our regressionswe include two measures of the current performance sensitivity of theCEOrsquos wealth to rm performance as explanatory variables to reect thedistance of the CEOrsquos contract from the optimal contract

Table 3 presents results in which the tax effects are identied solely byvariation in tax rates over time The fraction of annual compensation paidin stock options is regressed on tax variables corporate governance vari-ables contract-theory variables and variables reecting stock market re-turns Both of the regressions in this table include rm xed effects Incolumn (1) the fraction in annual compensation paid in stock options isregressed on the log difference in payoff (the combined payoff to both therm and the CEO) from receiving compensation in stock options ratherthan salary and bonus where the payoff calculations are based on statu-tory marginal tax rates This variable is dened as the log difference be-tween equation (3) and equation (6) in section 4 The coefcient on the taxvariable is large and statistically signicant at the 95-percent level Thepoint estimate implies that a one-percent increase in the payoff differencebetween stock options and salary and bonus results in a 24-percentage-point increase in the share of compensation paid in stock options

The corporate governance variables all have the predicted sign Firmswith a higher fraction of their shares owned by institutional investorsuse more performance-based pay (stock options) Firms with large corpo-rate boards or a large share of inside directors are less likely to use stockoptions

The two contract-theory variables are the dollar change in CEO wealth

11 See Core and Guay (1999) for evidence that rms adjust in this way

TABLE 3Compensation Regressions that Rely on Variation in

Tax Rates over Time

(1) (2)

Tax variablesln (difference in payoff)(a) 2393 (0349)ln (12statutory personal 20015 (0034)

rate)ln (12statutory cap gains 0330 (0171)

rate)ln (12statutory corporate 0399 (0148)

rate)

Corporate governance variablesFraction of rmrsquos shares 0043 (029) 0052 (0028)

owned by institutionalinvestors

Size of the rmrsquos board of 20002 (001) 20003 (0001)directors

Fraction of directors who 20082 (033) 20092 (0030)are inside or gray

Contract-theory variablesD (CEO wealth) per $1000 20261 (0095) 20247 (0087)

in FV(b)

D (CEO wealth) per 10 in 2583 (189) 2554 (201)FV(c)

Volatility(d) 521 (424) 261 (369)ln (rm market value) 0019 (0006) 0019 (0006)

Stock market return variablesLagged annual rm stock 0006 (0009) 0003 (0008)

market returnLagged annual SampP 500 re- 20035 (0023) 20029 (0023)

turn

Regression diagnosticsTotal sample size 5179 5179Firms 427 427Adjusted R2 0313 0311

Dependent variable share of compensation paid in options Standard errors in parentheses Bothregressions in this table include rm xed effects

(a) Log difference in payoff is the log difference in combined rm and CEO after-tax share betweenpaying compensation in options and paying salary

(b) This variable is the dollar change in CEO wealth per $1000 change in rm market value (seeJensen and Murphy 1990)

(c) This variable in the dollar charge in CEO wealth per 10-percent change in rm value (see Hall andLiebman 1998)

(d) Standard deviation of rm market value in trillions

28 Hall amp Liebman

per $1000 change in rm value (Jensen and Murphy 1990) and thedollar change in CEO wealth per 10-percent change in rm value (Halland Liebman 1998) As discussed in Baker and Hall (1998) these twowealth sensitivity measures reect two different concepts of how closelyaligned the managersrsquo incentives are with the interests of the sharehold-ers In the regression they are meant to measure how far the CEOrsquosexisting contract and ownership of stock and stock options is from theoptimal level The negative coefcients on the two variables suggeststhat rms that are below their optimal pay-to-performance sensitivityare more likely to give more stock options in the current period

The second column replaces the tax variable motivated by deferraladvantages with the three statutory rates The corporate after-tax sharehas a positive coefcient as we would expect because the tax advantageof stock options increases as the corporate rate falls The coefcient onthe personal tax rate is small and statistically insignicant The sign ofthe coefcient on the capital gains rate is the opposite of that predictedby tax-avoidance theory

Note that these results rely on time-series variation alone This is not avery convincing test Since the log difference in payoffs mostly rises overthe time period covered in our sample as does the share of compensa-tion paid in options it is possible that the tax variable is simply reectingother factors that were changing over this time period that we have notincluded in our regression Therefore it is important to look at specica-tions that rely on cross-sectional variation in the tax variables

Table 4 contains specications that rely on cross-sectional variation incorporate tax rates The rst column repeats the specication from col-umn (1) of Table 3 but uses the trichotomous measures of rm corporatetax rates in place of the statutory rate in calculating each rmrsquos logdifference in payoff from paying compensation in options rather than insalary and bonus As in the previous table we nd the positive coef-cient on the log difference in payoff just as the theory suggest How-ever the coefcient estimate is now much smaller suggesting that aone-percent increase in the log difference in payoff produces a 01-percentage-point increase in the share of compensation paid in options

Introducing cross-sectional variation in the dependent variable doesnot eliminate the potentially spurious correlation over time between thetax variable and the share of compensation paid in options The secondcolumn of Table 4 isolates the pooled cross-sectional variation in corpo-rate tax rates by adding time dummies and dropping the rm xedeffects In this specication the coefcient on the tax variable is indistin-guishable from zero The coefcient on the institutional variable contin-ues to be large and statistically signicant The third column includes

Taxation of Executive Compensation 29

both rm xed effects and time dummies which tests for a relationshipbetween within-rm changes in the dependent variable and within-rmchanges in each explanatory variable In this specication the coefcienton the tax variable is again indistinguishable from zero The coefcientson the corporate governance variables are also statistically insignicant

The third column contains a specication that treats the Tax ReformAct of 1986 (TRA 86) as a natural experiment For rms and CEOs facingthe statutory corporate tax rate TRA 86 increased the relative payoff topaying compensation in options by 17 percent For a rm facing a zeromarginal tax rate it reduced the relative payoff by 45 percent The reduc-tion occurred because TRA 86 reduced the personal tax rate makingsalary and bonus more attractive while the decreased corporate rate hadno effect on zero-tax-rate rms Thus we would expect rms facing zeroor low marginal tax rates to reduce their use of options after TRA 86while rms facing the statutory rate would be expected to increase themFew rms face zero marginal tax rates year after year Therefore we tryto distinguish between rms that often face low tax rates and rms thatnearly always face the statutory rate We dene low-tax rms as oneswhose average marginal tax rate over 1984 to 1986 was below 02412

Then we ran a difference-in-differences regression to see if high-taxrms increased their use of options more than low-tax rms The interac-tion between post-TRA 86 and high corporate tax rates is the key vari-able The estimated coefcient on this variable is positive but small andnot statistically different from zero at the 95-percent level

It is important to note that in this specication as well the main corpo-rate governance variable the share of the rmrsquos shares owned by institu-tional investors performs as theory would predict The variable appearsto be quite robust Table 5 presents regressions of the share of compensa-tion paid in stock options on the fraction of the rmrsquos shares owned byinstitutional investors Since we have already established that these vari-ables have a strong longitudinal relationship the regressions here aredesigned to focus on the cross-sectional relationship between these twovariables The rst and third columns present regressions using data onlyfrom 1994 and nd a coefcient of roughly 02 (column 1 contains noother covariates while column 3 includes all of the other nontax covari-ates from the regressions in Tables 3 and 4) Since the median share of armrsquos stock owned by institutional investors in our sample rises from 19percent in 1980 to 49 percent in 1994 this coefcient would imply that this

12 This cutoff was chosen after inspecting the distribution of tax rates and essentiallyseparates rms that are always at or near the statutory rate from ones that are sometimesbelow it

TA

BL

E4

Com

pens

atio

nR

egre

ssio

nsth

atR

ely

onC

ross

-Sec

tion

alV

aria

tion

inC

orpo

rate

Tax

Rat

es

(1)

(2)

(3)

(4)

Tax

vari

able

s(a)

ln

(dif

fere

nce

inp

ayof

f)0

099

(00

41)

20

038

(00

36)

20

005

(00

44)

Pos

t-T

RA

862

000

5(0

013

)H

igh

corp

orat

eta

xra

te0

018

(00

17)

(Pos

t-T

RA

86)

3hi

gh0

024

(00

16)

corp

orat

eta

xra

te

Cor

por

ate

gove

rnan

ceva

riab

les

Frac

tion

of

rmrsquos

shar

es0

109

(00

27)

011

4(0

015

)2

000

9(0

030

)0

087

(00

34)

owne

dby

inst

itu

tion

alin

vest

ors

Size

ofth

e

rmrsquos

boar

dof

20

004

(00

01)

20

001

(00

01)

20

002

(00

01)

000

1(0

002

)d

irec

tors

Frac

tion

ofd

irec

tors

wh

o2

012

4(0

033

)2

000

3(0

018

)2

004

1(0

034

)0

001

(00

39)

are

insi

de

orgr

ay

Con

trac

t-th

eory

vari

able

sD

(CE

Ow

ealt

h)p

er$1

000

20

238

(00

94)

20

167

(00

53)

20

257

(00

96)

20

021

(01

47)

inFV

(b)

D(C

EO

wea

lth)

per

10

in2

560

(17

8)2

592

(10

5)2

594

(19

6)2

147

(04

86)

FV(c

)

Vol

atili

ty(d

)2

48(4

24)

162

(03

4)1

12(4

28)

106

(06

09)

ln(

rmm

arke

tva

lue)

003

3(

006)

000

8(0

003

)0

009

(00

06)

001

2(0

007

)

Stoc

km

arke

tre

turn

vari

able

sL

agge

dan

nual

rm

stoc

k2

000

2(0

009

)0

011

(00

10)

000

6(0

009

)2

001

2(0

013

)m

arke

tre

turn

Lag

ged

annu

alSamp

P50

02

005

6(0

023

)2

005

3(0

051

)2

003

9(0

049

)2

015

4(0

040

)re

turn

Reg

ress

ion

dia

gnos

tics

T

otal

sam

ple

size

5179

5179

5179

2534

Firm

s42

742

739

6A

dju

sted

R2

031

009

032

005

Firm

xed

effe

cts

Yes

No

Yes

No

Tim

ed

um

mie

sN

oY

esY

esN

o

Dep

end

ent

vari

able

isfr

acti

onof

ann

ualc

omp

ensa

tion

from

stoc

kop

tion

sSt

and

ard

erro

rsin

par

enth

eses

(a)

All

thre

esp

eci

cati

ons

use

tric

hoto

mou

sco

rpor

ate

tax

rate

sT

he

sam

ple

inco

lum

n(3

)u

ses

only

the

year

s19

84to

1988

sin

cew

ear

eu

sing

1986

asa

nat

ural

exp

erim

ent

(b)

Thi

sva

riab

leis

the

dol

lar

chan

gein

CE

Ow

ealt

hp

er$1

000

chan

gein

rm

mar

ketv

alu

e(s

eeJe

nse

nan

dM

urp

hy

1990

)

(c)

Thi

sva

riab

lein

the

dol

lar

char

gein

CE

Ow

ealt

hp

er10

-per

cent

chan

gein

rm

valu

e(s

eeH

alla

nd

Lie

bman

199

8)

(d)

Stan

dar

dd

evia

tion

ofr

mm

arke

tva

lue

intr

illio

ns

32 Hall amp Liebman

TABLE 5Large Institutional Investors and Stock-Option Compensation

The Cross-sectional Relationship between Institutional Investors andOption Grants

(1) (2) (3) (4)

Coefcient on fraction of 217 (067) 145 (014) 177 (069) 120 (015)rmrsquos shares owned byinstitutional investors

Sample period 1994 1980ndash1994 1994 1980ndash1994Contract-theory and stock No No Yes Yes

market covariatesTime dummies No Yes No Yes

The dependent variable is the share of compensation paid in options

factor alone can explain 6 percentage points of the 23-percentage-pointincrease in the median share of compensation paid in options that oc-curred over this time period Of course this variable is only one measureof the strengthening of corporate governance so the total impact of im-proved corporate governance could be greater The second and fourthcolumns of the table use the entire 1980ndash1994 sample and include timedummies to isolate the cross-sectional correlations Again there is evi-dence of a strong cross-sectional relationship between the two variablesalthough the coefcients are slightly smaller perhaps reecting a weakercorporate-governance role in the earlier period In sum there is a strongrelationship in both longitudinal and cross-sectional data between theinstitutional-investor variable and the options-percentage variable13

7 THE MILLION-DOLLAR RULE

In 1993 section 162(m) of the Internal Revenue Code was enactedlimiting the deductibility of executive compensation in excess of onemillion dollars unless the compensation was performance-related Pro-ponents of this legislation argued that it would slow the growth ofexecutive pay and tighten the link between rm performance and execu-tive compensation

The new provision became effective for tax year 1994 and applies tothe CEO and the other four most highly compensated executives in eachcompany Pay that is performance-related such as stock-option grants

13 The one specication in which the fraction owned by institutional investors is notsignicant is in column (2) of Table 4 a regression that includes both xed effects and timeeffects

Taxation of Executive Compensation 33

and bonuses paid for meeting clear performance goals is not affected bythis regulation provided that it has been approved by shareholdersThus the main impact of the provision is to limit the deductibility ofsalary in excess of a million dollars Woodlock and Antenucci (1997)studied the proxy statements of 376 rms and documented that mostrms responded to the new law by qualifying top executive pay asperformance-related Perry and Zenner (1999) identied 25 rms thatreduced salaries from above $1 million to below $1 million and foundthat 23 of the rms cited section 162(m) as their reason for doing so

In this section of the paper we investigate whether this provision hasaffected the level or structure of executive pay using the Execucompdata base which contains information on the highest-paid executives ineach of the rms in the SampP 500 SampP Midcap 400 and SampP SmallCap600 Between 1993 and 1998 the median salary of the Execucomp execu-tives rose from $230000 to $306000 The median value of total compen-sation (salary bonus and option grants) rose from $470000 to $882000Limiting the sample of CEOs indicates that the total compensation of themedian CEO rose from $11 million to $18 million over the period whilethe median salary rose from $441000 to $550000 Thus executive com-pensation clearly continued to rise at a rapid rate after the implementa-tion of the million-dollar rule though perhaps at a slower rate than itotherwise would have

If the regulation affects behavior it should have the largest impact onrms whose executives are receiving salaries that are above the limit orjust below it Thus we might expect to see rms whose executives arereceiving relatively high salaries to rely more heavily on performance-based pay and less heavily on salary in the years following the regulationPerry and Zenner (1999) and Livingston (1999) both nd that this is thecase This trend can be clearly identied in the Execucomp data set Forexample over 1993 to 1998 the median annualized growth rate of salaryfor executives whose 1993 salary was below $500000 in the previous yearwas 85 percent For executives with salary between $500000 and$700000 it was 64 percent for executives between $700000 and $850000it was 50 percent for executives between $750000 and $1 million it was31 percent and for executives above $1 million it was 00 percent14

The observation that executives at lower levels of salary had highergrowth rates is not very strong evidence of an impact of section 162(m)however because it is possible that low-salary executives would havehad higher growth rates even in the absence of the provision To explore

14 Of the 3137 executives for whom we have complete data between 1993 and 1998 2708had 1993 salaries below $500000 while only 35 had 1993 salaries above $1 million

34 Hall amp Liebman

whether this is the case we use Execucomp data from both before andafter the law change Ideally we would want to have data from manyyears before the change to examine whether there is an underlying pat-tern of faster salary growth for low-salary executives Unfortunately theExecucomp data begin in 1992 and therefore provide us with only the1992ndash1993 growth rate in compensation to use as a baseline15

We run regressions pooling data from 1992ndash1993 through 1997ndash1998of the annual percentage change in different forms of compensation on avariable million that is designed to measure the likelihood that anexecutiversquos compensation will be affected by section 162(m) We denemillion as the minimum of 1 and the executiversquos salary divided by $1million in the previous year Thus any executive with salary at or above$1 million would receive a value of 1 while an executive with a salary of$500000 would be coded as a 0516 We include this variable by itself andalso interacted with a dummy variable (after) that equals 1 for years1994 and after Thus the million variable is designed to pick up anyunderlying relationship between the level of salary and the growth rateof salary while the interacted one will identify any differential growthrate of salary by income in the period after section 162(m) took effect Wealso include a full set of year dummies the natural logarithm of thermrsquos market value and the rmrsquos annual rate of return over the twoprevious years as control variables To reduce the sensitivity of our re-sults to outliers we run quantile regressions and robust regressions

Table 6 contains our results The rst four columns are for all topexecutives while the last four columns focus on CEOs Columns (1) and(2) present estimates from robust regression for the percentage change insalary and the percentage change in total compensation (salary bonusstock grants and stock-option grants) In the salary growth regressionsfor all executives the coefcient on million is negative and signicantindicating that there is an underlying relationship between the level ofexecutive compensation and its subsequent growth rate The coefcienton million 3 after is negative as well suggesting that the negativerelationship between the level of salary and its subsequent growth rateintensied slightly after 1994 The relatively small magnitude suggeststhat an executive with a $1 million dollar salary would see his salarygrow at an annual rate that would be about 06 percent slower than anexecutive earning $500000

15 The HallndashLiebman data set contains only a combined salary and bonus variable makingit impossible to use it to analyze this issue16 We also ran the regressions (not reported) allowing for a nonlinear relationship betweenmillion and compensation growth and obtained substantively similar results

TA

BL

E6

The

Eff

ect

ofth

eM

illi

on-D

olla

rR

ule

onSa

lary

and

Tot

alC

ompe

nsat

ion

All

top

exec

uti

ves

CE

Os

only

Rob

ust

regr

essi

onQ

uan

tile

regr

essi

onR

obu

stre

gres

sion

Qu

anti

lere

gres

sion

D

in

Din

D

in

Din

D

in

Din

D

in

Din

sala

ryto

talc

omp

(a)

sala

ryto

talc

omp

sa

lary

tota

lcom

p

sala

ry

tota

lcom

p

mil

lio

n(b

)2

006

02

035

72

006

32

024

12

008

72

023

12

010

32

024

5(0

005

)(0

032

)(0

006

)(0

027

)(0

010

)(0

073

)(0

011

)(0

056

)m

illi

on

32

001

10

026

20

021

20

015

20

016

014

92

003

40

158

aft

er(c

)(0

005

)(0

033

)(0

006

)(0

029

)(0

010

)(0

076

)(0

011

)(0

057

)N

359

5035

950

359

5035

950

757

17

589

757

17

589

Stan

dar

der

rors

inp

aren

thes

esR

egre

ssio

nsal

soin

clu

de

year

du

mm

ies

stoc

km

arke

tre

turn

and

log

mar

ket

valu

e

(a)

Tot

alco

mp

ensa

tion

incl

ud

essa

lary

bon

us

rest

rict

edst

ock

and

stoc

k-op

tion

gran

ts

(b)

mil

lion

isd

ene

das

min

(1s

alar

yin

pre

viou

sye

ars

div

ided

by$1

mil

lion)

(c)

aft

eris

anin

dic

ator

vari

able

equ

alto

zero

befo

reth

ere

gula

tion

took

plac

e(1

993)

and

one

afte

rwar

d

36 Hall amp Liebman

In the second column the coefcient on million 3 after is essentiallyzero suggesting that any decrease in salary brought about by section162(m) was offset by increases in bonus and stock-option grants Theresults are similar for the median regressions There is some evidence ofa very small slowdown in growth of salary for executives likely to beaffected by section 162(m) but no sign of a change in total compensa-tion Taken together this provides evidence of a very small substitutionof performance-related pay for salary

The last four columns run identical tests but limit the sample to CEOsThe interaction of million 3 after continues to be small and negative inthe salary regressions In contrast the impact of 162(m) on total compen-sation appears to be positive and fairly substantial (about 15 percentagepoints per year) in these regressions That is the increase in bonus andstock options following the million-dollar rule more than offset the verysmall decline in salary

All of these conclusions should be interpreted somewhat cautiouslybecause our data provide us with only one pre-1993 control year Never-theless the data are consistent with section 162(m) having led to a veryminor slowdown in salary growth and one that is dramatically smallerthan what one would conclude without controlling for underlying differ-ential salary growth rates at different levels of salary Since total compen-sation for CEOs did not decline and perhaps increased there is evidenceof a minor substitution of performance-related pay for salary in responseto the regulation

8 EXECUTIVE COMPENSATION AND THEELASTICITY OF TAXABLE INCOME

Even if taxes have only a minor effect on executive compensation thestructure of executive compensation is important for tax policy be-cause high-income taxpayers are sources of signicant revenue and arethe focus of important debates about the efciency cost of taxation

Goolsbee (1999) has recently argued that executive decisions to exer-cise options are highly responsive to intertemporal differences in taxrates created by tax reforms He claims that much of the apparent declinein taxable income for high-income taxpayers between 1992 and 1993documented by Feldstein and Feenberg (1996) can be explained by execu-tives shifting option gains into 1992 in order to avoid the higher marginaltax rates that went into effect in 1993 By extension he implies that thehigh elasticities of taxable income with respect to personal tax ratesestimated by comparing the pre- and post-tax-reform taxable income of

Taxation of Executive Compensation 37

high-income individuals reect intertemporal shifting of income and notpermanent effects of taxation If this is true then the deadweight loss oftaxation is lower as is the cost of progressivity

While Goolsbee presents comprehensive evidence that option gainswere unusually high in 1992 his identication of the impact of tax rates onthe timing of option gains depends on correctly specifying a counter-factual level of options that would have been exercised in the absence ofthe tax change He simply assumes a linear time trend The purpose ofGoolsbeersquos time trend is presumably to account for the increasing use ofoptions over time However since options typically require a vestingperiod of at least a couple of years a much more direct measure of ex-pected option gains can be constructed based on the total value of vestedoptions held by the CEO Since options typically have vesting periods thatexceed two years this total can be treated as exogenous since it will not beaffected by the tax changes

More generally the value of options held by the CEO is likely to havea major effect on his exercise decisions If his options are out of themoney he will not have gains to realize If there has been a recent run-up in the market an optimizing CEO may exercise an unusually largeamount of options in order to diversify his overall portfolio and if in-the-money options are about to expire they will be exercised in thecurrent year

Our evidence suggests that much of the apparent tax shifting caninstead be attributed to stock market performance and the timing of pastoption grants Using our 15-year panel we replicate Goolsbeersquos (1999)result for the 1993 period but show that the impact of taxes is notpresent in other time periods Moreover even the result for the 1993period disappears when appropriate controls for past option grants andstock price appreciation are included

The top three rows of Table 7 present data on the taxable incomes ofCEOs in the HallndashLiebman sample For these results our panel has beennarrowed to include only CEOs whose rmrsquos scal years correspondwith tax years In addition we create a balanced panel for each of thethree tax-reform periods by excluding any rm for which there are notcomplete data for the period surrounding the reform

The fourth row of the table shows that the 1993 tax act raised the topmarginal personal income tax rate from 310 to 396 percent effective fortax year 1993 with a further increase occurring in 1994 due to the uncap-ping of the Medicare payroll tax The top row of the table shows thattaxable income was higher in 1992 than for any other year in our sample(all data are in 1994 dollars) Option gains were particularly high in1992mdashmore than double their 1991 level and 68 percent higher than their

TA

BL

E7

Tax

Ref

orm

Sto

ckM

arke

tP

erfo

rman

cea

ndth

eT

imin

gof

Opt

ion

Gai

ns

Ave

rage

stoc

km

arke

tM

ean

CE

OM

ean

CE

Ore

turn

for

Mea

nva

lue

taxa

ble

sala

ryan

dM

ean

CE

OT

opfe

der

al

rms

inof

opti

ons

(Val

ue

ofop

tion

Sam

ple

inco

me

bonu

sop

tion

gain

sp

erso

nal

sam

ple

for

held

atbe

gga

ins)

(va

lue

Ref

orm

size

Yea

r($

mil

)($

mil

)($

mil

)M

TR

pre

vye

ar(

)of

year

ofop

tion

she

ld)

ER

TA

8127

919

800

874

064

80

163

7025

80

460

354

1981

083

10

672

009

869

245

070

61

3919

820

872

067

90

099

5011

90

686

144

1983

099

00

724

025

250

355

116

62

16

TR

A86

295

1985

119

70

837

024

050

76

116

02

0719

861

448

091

70

345

5042

81

776

194

1987

165

50

959

059

338

519

12

229

266

1988

163

21

058

039

028

07

237

71

6419

891

656

105

20

414

2817

92

345

176

1993

272

1990

209

51

028

061

828

232

288

02

15T

axA

ct19

911

875

102

00

547

312

01

268

82

0319

922

678

107

01

250

3146

64

552

275

1993

232

41

209

074

339

620

64

299

173

1994

196

41

262

037

842

515

45

113

074

Sam

ple

isH

allndash

Lie

bman

rm

sw

ith

Dec

embe

r

scal

year

sth

atha

dco

mp

lete

dat

afo

ron

eof

the

thre

eta

x-re

form

per

iod

sst

ud

ied

Taxation of Executive Compensation 39

1994 level These patterns are consistent with Goolsbeersquos claim that tax-payers responded to the anticipated increase in marginal tax rates byshifting gains into 1992

The evidence from the other two tax reforms is less clear suggestingthat there might be factors other than tax rates determining the timing ofoption gains ERTA 1981 reduced the top marginal tax rate from 70percent in 1980 to 50 percent in 1982 However there is no sign thatoption gains were reduced in 1980 and 1981 and then increased in 198217

There was a big increase in option gains in 1983 however Similarly thehighest year of stock-option gains in the period around 1986 was 1987 astrange year to take gains since taxpayers would have known that thepersonal rate would be 10 percentage points lower if they waited untilthe following year18

What all the peak years of option gains have in common is that theyfollow years of strong stock market performance The fth row of Table 7shows the average stock market return during the previous year for therms in our sample The large value of option gains in 1983 followed a355-percent increase in the stock market The 1992 boom in option gainsfollowed a 466-percent increase in the stock market the more modest1987 peak in option gains followed a 19-percent gain (and a 43-percentgain two years before)

Strong stock market performance raises the value of options availableto exercise The sixth row of the table shows the mean value of optionsheld by CEOs at the beginning of the year All three of the peak years ofoption gains correspond with peak years in terms of the value of optionsavailable for exercise Since stock market appreciation during the yearalso increases the value of options available to exercise later in that yearwe need to allow for current-year stock market appreciation as well

Table 8 contains Goolsbee-style regressions for each of the tax re-forms The rst two columns reprint the results from Goolsbee (1999)The third and fourth columns replicate his results in our sample Inparticular with a specication similar to his we are able to replicate hisnding that the current after-tax share has a positive coefcient and thefollowing yearrsquos after-tax share has a coefcient of roughly negative oneColumns (5) through (8) show that this pattern was unique to the 1993tax reform Indeed in the TRA 86 regressions the coefcient on the

17 As we explained above for executives covered by the maximum tax on earned incomethere was not a decline in tax rates between 1981 and 198218 It is possible that the large stock market decline in October 1997 may have promptedsome people to exercise options earlier than they otherwise would have However anincrease in option exercises typically follows stock market increases not declines

TA

BL

E8

Att

empt

sto

Rep

lica

teth

eT

ax-S

hift

ing

Res

ult

inO

ther

Tim

eP

erio

ds

Goo

lsbe

eG

ools

bee

1993

1993

Ent

ire

Ent

ire

regr

essi

onre

gres

sion

refo

rmre

form

TR

A86

TR

A86

ER

TA

81E

RT

A81

1980

ndash199

419

80ndash1

994

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

ln[1

2ta

x(t)

]1

244

108

20

455

032

62

017

62

017

22

016

72

016

70

026

20

068

(01

34)

(01

27)

(05

37)

(05

42)

(02

10)

(02

09)

(00

90)

(00

91)

(00

67)

(00

65)

ln[1

2ta

x(t1

1)]

20

867

20

764

21

214

20

917

047

80

722

20

103

004

30

176

018

3(0

118

)(0

112

)(0

491

)(0

499

)(0

157

)(0

174

)(0

072

)(0

072

)(0

068

)(0

066

)

Tim

e0

135

007

62

000

52

036

40

041

000

80

094

003

50

056

002

7(0

008

)(0

008

)(0

049

)(0

049

)(0

028

)(0

028

)(0

025

)(0

026

)(0

003

)(0

003

)

ln(m

arke

tva

lue)

066

10

489

023

70

334

039

5(0

015

)(0

058

)(0

039

)(0

042

)(0

017

)

Yea

rs19

91ndash1

995

1991

ndash199

519

90ndash1

994

1990

ndash199

419

85ndash1

989

1985

ndash198

919

80ndash1

983

1980

ndash198

319

80ndash1

994

1980

ndash199

4

Ad

just

edR

20

720

770

610

630

690

760

800

810

630

68

Obs

erva

tion

s13

856

132

371

385

136

41

470

143

31

113

106

23

968

385

9

Dep

end

ent

vari

able

isth

elo

gof

taxa

ble

inco

me

All

regr

essi

ons

con

tain

rm

xe

def

fect

sG

ools

bee

regr

essi

on(2

)al

soco

ntai

nstw

oot

her

vari

able

sno

tsh

own

inth

eta

ble

Stan

dar

der

rors

inp

aren

thes

es

Taxation of Executive Compensation 41

TABLE 9Alternative Explanations for the Timing of Option Exercises

1993 Entirereform TRA 86 ERTA 81 1980ndash1994

(1) (2) (3) (4)

ln[12tax(t)] 0108 0111 20051 0201(0324) (0136) (0048) (0057)

ln[12tax(t11)] 20134 0573 0125 0150(0336) (0156) (0053) (0066)

ln(market value) 0418 0177 0338 0418(0061) (0039) (0046) (0163)

Firmrsquos stock market return (t) 0298 0268 0214 0258(0044) (0045) (0031) (0023)

Firmrsquos stock market return 0162 0206 0031 0128(t21) (0041) (0008) (0029) (0021)

ln(value of options held by 0004 0008 0001 0009CEO at start of year) (0003) (0003) (0002) (0001)

Years 1990ndash1994 1985ndash1989 1980ndash1983 1980ndash1994Adjusted R2 064 071 082 068Observations 1349 1427 1036 3812

The dependent variable is the log of taxable income All regressions contain rm xed effects Stan-dard errors in parentheses

following yearrsquos tax rate is positive and signicant Columns (9) and (10)pool all three reforms and fail to replicate the pattern that would reectintertemporal shifting19

It is possible that the ease with which one can shift the timing ofoption gains is asymmetric It may be harder to shift option gains for-ward in time since some options expire and therefore must be exercisednow In addition options were a much smaller portion of compensationin the early 1980s so perhaps the ERTA 81 results are not surprisingNonetheless these results cast doubt on the proposition that this is thechannel through which spurious estimates of the elasticity of taxableincome have occurred for previous tax reforms and they suggest thatwe should look harder for alternative explanations for the timing ofoption gains

Table 9 attempts to predict options gains using the lagged and currentincrease in a rmrsquos market value and the value of the options held by theCEO at the start of the year The two stock market variables stronglypredict option gains while the value of options held has the correct sign

19 The results also fail to suggest large permanent effects of marginal tax rates on taxableincome

42 Hall amp Liebman

in all of the specications and is signicant in the pooled sample eventhough it is highly correlated with the two stock market return variablesIn contrast the tax-timing explanation fails even in the 1993 period

It is important to realize that our results do not invalidate Goolsbeersquoscritique of claims that aggregate revenue data from 1992 and 1993 provethat elasticities of taxable income are large We have simply given analternative explanation for why taxable incomes were unusually high in1992 However our results do cast doubt on the conclusion that largetiming shifts are ubiquitous and capable of explaining away the morecareful estimates of taxable income elasticities that have been performedusing micro data (eg Feldstein 1995 Auten and Carroll 1995)

One further point is in order It is sometimes argued that the combina-tion of increased use of options and the booming stock market explainswhy federal tax revenues have been so much higher than expected dur-ing the 1990s However while large option gains increase revenue fromthe personal income tax they produce offsetting deductions for cor-porations The net impact on revenue is only the difference in the twomarginal tax rates times the option gain While the gap between thepersonal rate and the effective corporate rate is nontrivial the impact oflarge option exercises on total revenue is substantially smaller than theamount by which the exercises inate personal tax revenue and reducecorporate tax revenue

9 CONCLUSION

We have described the tax rules for executive pay in detail and analyzedhow changes in various tax rates affect the tax advantages of stock optionsrelative to salary and bonus We nd that there is a moderate tax advan-tage to optionsmdashcurrently about $4 per $100 in pretax executive paymdashand that changes in the tax advantage of options over time have had atmost a modest impact on the composition of pay Corporate governancefactors particularly the role of large investors appear to be more impor-tant in explaining the dramatic increase in option pay

In addition our evidence suggests that more direct attempts to use taxpolicy to inuence executive compensation have had little effect Wend that the million-dollar rule led companies to substitute perfor-mance-based pay for salary But our evidence suggests that this substitu-tion was quite modest and there is no evidence that the total level of paywas reduced Overall although the stock-option explosion has dramati-cally increased the link between pay and performance this change isdue almost entirely to nontax factors

Taxation of Executive Compensation 43

REFERENCES

Auerbach Alan J (1996) ldquoThe Dynamic Effects of Tax Law Asymmetriesrdquo Re-view of Economic Studies 53451ndash405

mdashmdashmdash and James Poterba (1987) ldquoTax Loss Carryforwards and Corporate TaxIncentivesrdquo In The Effects of Taxation on Capital Accumulation M Feldstein(ed) Chicago University of Chicago Press

Altshuler Rosanne and Alan Auerbach (1990) ldquoThe Signicance of Tax LawAsymmetries An Empirical Investigationrdquo Quarterly Journal of Economics 10561ndash86

Auten Gerald and Robert Carroll (1995) ldquoThe Effect of Income Taxes on House-hold Behaviorrdquo Ofce of Tax Analysis US Department of the Treasury

Baker George P and Brian J Hall (1998) ldquoCEO Incentives and Firm SizerdquoNBER Working Paper no 6868 December

Core John and Wayne Guay (1999) ldquoThe Use of Equity Grants to ManageOptimal Equity Incentive Levelsrdquo Draft

Feldstein M (1995) ldquoThe Effect of Marginal Tax Rates on Taxable Income APanel Study of the 1986 Tax Reform Actrdquo Journal of Political Economy 103 (no3)551ndash72

mdashmdashmdash and D Feenberg (1996) ldquoThe Effect of Increased Tax Rates on TaxableIncome and Economic Efciency A Preliminary Analysis of the 1993 Tax RateIncreasesrdquo In Tax Policy and the Economy vol 10 James M Poterba (ed)Cambridge MA MIT Press

Gompers Paul A and Andrew Metrick (1998) ldquoInstitutional Investors andEquity Pricesrdquo NBER Working Paper no 6723

Goolsbee Austan (1999) ldquoWhat Happens When You Tax the Rich Evidencefrom Executive Compensationrdquo NBER Working Paper no 6333

Graham John R (1996) ldquoProxies for the Corporate Marginal Tax Raterdquo Journal ofFinancial Economics 42187ndash221

Hall Brian J (1999) ldquoThe Design of Multi-year Stock Option Plansrdquo Journal ofApplied Corporate Finance 12(2)97ndash106

mdashmdashmdash and Jeffrey B Liebman (1998) ldquoAre CEOs Really Paid Like Bureau-cratsrdquo Quarterly Journal of Economics CXII(no 3)653ndash691

Jensen Michael C and William M Meckling (1976) ldquoTheory of the Firm Mana-gerial Behavior Agency Costs and Ownership Structurerdquo Journal of FinancialEconomics III305ndash360

mdashmdashmdash and Kevin J Murphy (1990) ldquoPerformance Pay and Top-ManagementIncentivesrdquo Journal of Political Economy XCVIII225ndash264

Kaplan Steven N (1997) ldquoThe Evolution of US Corporate Governance WeAre All Henry Kravis Nowrdquo Draft

Livingstone Jane R (1999) ldquoExecutive Compensation Contracting and Re-sponses to Increased Tax Costsrdquo Louisiana State University June

Miller Merton H and Myron S Scholes (1982) ldquoExecutive CompensationTaxes and Incentivesrdquo In Financial Economics Essays in Honor of Paul CootnerWilliam F Sharpe and Cathryn M Cootner (eds) Princeton NJ Prentice-Hall

Murphy Kevin J (1999) ldquoExecutive Compensationrdquo forthcoming in OrleyAshenfelter and David Card (eds) Handbook of Labor Economics Vol 3 NorthHolland

Perry Todd and Marc Zenner (1999) ldquoPay for Performance Government Regu-

Taxation of Executive Compensation 23

by about 5 percent By 1988 (following the 1986 tax act) ISOs became tax-disadvantaged by about 18 percent a dramatic swing The raising ofpersonal rates coupled with a decrease in the capital gains rate has sincereduced the tax disadvantage of ISOs to about $4 per $100 of pretaxcompensation Although hard data on ISOs are hard to come by practi-tioners (mostly compensation consultants) claim that ISOs were morepopular prior to 1982 when they were tax-advantaged Although ISOsare less disadvantageous from a tax standpoint relative to NQSOs thanthey were in the late 1980s they are still disadvantageous so it is notsurprising that they have not made a signicant resurgence in recentyears

6 EXPLAINING THE INCREASE IN OPTION-BASEDCOMPENSATION

In this section we analyze whether changes in tax rates affect the compo-sition of executive compensation and consider the relative importanceof tax factors and corporate governance factors in explaining the in-creasing share of compensation paid in the form of stock options Thenumerous changes in tax rates that have occurred since 1980 provide anopportunity to assess whether executive compensation arrangementsrespond to tax incentives in the way that the tax-avoidance model sug-gests We test this model using a panel data set of CEOs in largepublicly-traded US corporations Our identication of the tax effectsrelies on time-series variation in personal corporate and capital gainstax rates along with cross-sectional variation in corporate tax rates

61 DataWe use a 15-year panel data set of CEOs in the largest publicly-tradedUS corporations which is described in Hall and Liebman (1998) Thedata set covers the years from 1980 through 1994 and combines CEOcompensation information from corporate proxies and 10-K lings withstock price and stock return information from CRSP and accounting datafrom Compustat In addition some compensation data from the 1970swere collected in order to construct measures of the value of stock op-tions held by the CEOs in the rst period of the sample8

The feature that distinguishes our data from most other CEO datasets is that with our panel of yearly proxy data on option grants optiongains and total options held we are able to calculate the total value of

8 We are grateful to David Yermack for providing us with some of the data for 1984 to 1991See Yermack (1995) for a discussion of these data

24 Hall amp Liebman

all stock options held by the CEO at a point of time More importantlysince we have the details about the stock options held (number exer-cise price time to maturity etc) we can precisely calculate the changein the value of a CEOrsquos stock option holdings for a given change in rmvalue

62 Identication IssuesThe numerous tax changes over our sample period and the specic waysin which changes in personal corporate and capital gains rates arepredicted to affect the composition of CEO compensation give us un-usually rich sources of identication However because many of thechanges in tax rates would be expected to affect all of the CEOs in oursample in a similar way we will need to pay particular attention toseparating out the tax effects from underlying time trends and otherfactors that changed over time Controlling for underlying trends is par-ticularly important in this study because we are focusing on an out-comemdashthe use of stock optionsmdashthat has increased rapidly over thepast two decades and that many practitioners believe was importantlyaffected by nontax considerations Thus the tax effects we are trying toexplain are deviations from a rising trend

We take two steps to try to separate out the tax effects from theunderlying trend First we include key nontax factors that could poten-tially explain the increasing reliance on performance-based compensa-tion It has been suggested that the dramatic increase in incentive-basedpay is the result of the remarkably poor shareholder returns during the1970s which spurred the LBO and takeover movements of the 1980sAccording to this story (Kaplan 1997) shareholders became much morepowerful via the rise of institutional investors even as the LBO andtakeover movements waned Because the inuence of institutional inves-tors is thought to be one of the most important mechanisms of strongcorporate governance we use the share of each companyrsquos stock ownedby institutional investors (dened as institutions with more than $100million under management) as an explanatory variable9 The percentageof shares owned by large institutional investors increased from about 20percent to almost 50 percent in 1994 an upward trend that closelymatches the sharp rise in the share of compensation in the form ofoptions Indeed in our sample the annual average (over all of the rmsin our sample) share of stock owned by institutional investors has acorrelation of 09 with the annual average share of compensation paid inoptions

9 This variable is described in Gompers and Metrick (1998)

Taxation of Executive Compensation 25

In addition we include two other variables that proxy for strongercorporate governance The rst is the size of the board Evidence suggeststhat boards with smaller numbers of directors reduce the inuence of theCEO and better represent shareholder interests (Yermack 1995) Boardsize fell by approximately 15 percent between 1980 and 1994 The secondvariable is the fraction of the rmrsquos board members who are inside direc-tors or gray directors (noninsiders who have business dealings with thecompany) Firms with fewer outsiders are less likely to act in the share-holdersrsquo interests and we would predict that they would be less likely tohave performance-based compensation for their CEOs The average per-centage of inside and gray directors has decreased from about 45 percentin 1980 to about 35 percent in 1994

Second we allow for cross-sectional variation in corporate tax ratesCorporate marginal tax rates vary because rms may be eligible for tax-loss carrybacks and carryforwards investment tax credits and the alter-native minimum tax (see Auerbach 1986 Auerbach and Altshuler1990 Auerbach and Poterba 1987) Because these cross-sectional differ-ences imply that different rms should respond differently to a givenchange in tax rates they potentially provide us with a way to isolatethe tax effects Firm-specic corporate marginal tax rates are notori-ously difcult to calculate from publicly-available data We use a tricho-tomous variable that equals zero if the rm has tax-loss carryforwardsand negative earnings in a given year equals 05 times the statutorycorporate rate if the rm has only one of those conditions and equalsthe statutory rate if the rm has neither loss carryforwards nor negativeearnings10

The cross-sectional variation in the corporate tax rate may not be ex-ogenous Firms that perform poorly could face low marginal tax ratesand also be particularly likely or unlikely to provide performance-basedpay Because most of our specications use xed effects the correlationwe must be concerned about is between changes in rm performanceand changes in compensation We deal with this by controlling forlagged rm stock market performance in our regressions This shouldeliminate spurious correlation between tax rates and compensation thatis jointly caused by rm performance

10 Graham (1996) presents evidence that the trichotomous variable performs nearly as wellas his recommended simulated marginal tax rate Plesko (1999) compares marginal taxrates calculated from Compustat data with actual marginal tax rates from rm tax returnsHe nds that simple binary variables are more reliable measures of the marginal tax ratethan is either the Graham or the trichotomous variable We obtain substantively similarresults when we switch from the trichotomous variable to the binary variables recommendby Plesko

26 Hall amp Liebman

63 ResultsThe dependent variable in our regressions is the fraction of a CEOrsquos totalannual compensation (measured as the BlackndashScholes value of stock-option grants plus salary and bonus) that comes from stock-optiongrants We choose this reduced-form specication because our interest isin the composition of CEO compensation However the factors thataffect the composition of compensation might also affect the levels andit will be important to keep this in mind in interpreting our results

Optimal-contracting theory generally makes predictions about the in-centives provided by the total amount of rm stock and stock optionsowned by the CEO We choose to model the ow of option grants ratherthan the ultimate performance sensitivity of the CEOrsquos compensationand stock and stock-option holdings because we have in mind anadjustment-cost model in which it takes time for a CEOrsquos holdings ofstock and stock options to reach the optimal level11 In our regressionswe include two measures of the current performance sensitivity of theCEOrsquos wealth to rm performance as explanatory variables to reect thedistance of the CEOrsquos contract from the optimal contract

Table 3 presents results in which the tax effects are identied solely byvariation in tax rates over time The fraction of annual compensation paidin stock options is regressed on tax variables corporate governance vari-ables contract-theory variables and variables reecting stock market re-turns Both of the regressions in this table include rm xed effects Incolumn (1) the fraction in annual compensation paid in stock options isregressed on the log difference in payoff (the combined payoff to both therm and the CEO) from receiving compensation in stock options ratherthan salary and bonus where the payoff calculations are based on statu-tory marginal tax rates This variable is dened as the log difference be-tween equation (3) and equation (6) in section 4 The coefcient on the taxvariable is large and statistically signicant at the 95-percent level Thepoint estimate implies that a one-percent increase in the payoff differencebetween stock options and salary and bonus results in a 24-percentage-point increase in the share of compensation paid in stock options

The corporate governance variables all have the predicted sign Firmswith a higher fraction of their shares owned by institutional investorsuse more performance-based pay (stock options) Firms with large corpo-rate boards or a large share of inside directors are less likely to use stockoptions

The two contract-theory variables are the dollar change in CEO wealth

11 See Core and Guay (1999) for evidence that rms adjust in this way

TABLE 3Compensation Regressions that Rely on Variation in

Tax Rates over Time

(1) (2)

Tax variablesln (difference in payoff)(a) 2393 (0349)ln (12statutory personal 20015 (0034)

rate)ln (12statutory cap gains 0330 (0171)

rate)ln (12statutory corporate 0399 (0148)

rate)

Corporate governance variablesFraction of rmrsquos shares 0043 (029) 0052 (0028)

owned by institutionalinvestors

Size of the rmrsquos board of 20002 (001) 20003 (0001)directors

Fraction of directors who 20082 (033) 20092 (0030)are inside or gray

Contract-theory variablesD (CEO wealth) per $1000 20261 (0095) 20247 (0087)

in FV(b)

D (CEO wealth) per 10 in 2583 (189) 2554 (201)FV(c)

Volatility(d) 521 (424) 261 (369)ln (rm market value) 0019 (0006) 0019 (0006)

Stock market return variablesLagged annual rm stock 0006 (0009) 0003 (0008)

market returnLagged annual SampP 500 re- 20035 (0023) 20029 (0023)

turn

Regression diagnosticsTotal sample size 5179 5179Firms 427 427Adjusted R2 0313 0311

Dependent variable share of compensation paid in options Standard errors in parentheses Bothregressions in this table include rm xed effects

(a) Log difference in payoff is the log difference in combined rm and CEO after-tax share betweenpaying compensation in options and paying salary

(b) This variable is the dollar change in CEO wealth per $1000 change in rm market value (seeJensen and Murphy 1990)

(c) This variable in the dollar charge in CEO wealth per 10-percent change in rm value (see Hall andLiebman 1998)

(d) Standard deviation of rm market value in trillions

28 Hall amp Liebman

per $1000 change in rm value (Jensen and Murphy 1990) and thedollar change in CEO wealth per 10-percent change in rm value (Halland Liebman 1998) As discussed in Baker and Hall (1998) these twowealth sensitivity measures reect two different concepts of how closelyaligned the managersrsquo incentives are with the interests of the sharehold-ers In the regression they are meant to measure how far the CEOrsquosexisting contract and ownership of stock and stock options is from theoptimal level The negative coefcients on the two variables suggeststhat rms that are below their optimal pay-to-performance sensitivityare more likely to give more stock options in the current period

The second column replaces the tax variable motivated by deferraladvantages with the three statutory rates The corporate after-tax sharehas a positive coefcient as we would expect because the tax advantageof stock options increases as the corporate rate falls The coefcient onthe personal tax rate is small and statistically insignicant The sign ofthe coefcient on the capital gains rate is the opposite of that predictedby tax-avoidance theory

Note that these results rely on time-series variation alone This is not avery convincing test Since the log difference in payoffs mostly rises overthe time period covered in our sample as does the share of compensa-tion paid in options it is possible that the tax variable is simply reectingother factors that were changing over this time period that we have notincluded in our regression Therefore it is important to look at specica-tions that rely on cross-sectional variation in the tax variables

Table 4 contains specications that rely on cross-sectional variation incorporate tax rates The rst column repeats the specication from col-umn (1) of Table 3 but uses the trichotomous measures of rm corporatetax rates in place of the statutory rate in calculating each rmrsquos logdifference in payoff from paying compensation in options rather than insalary and bonus As in the previous table we nd the positive coef-cient on the log difference in payoff just as the theory suggest How-ever the coefcient estimate is now much smaller suggesting that aone-percent increase in the log difference in payoff produces a 01-percentage-point increase in the share of compensation paid in options

Introducing cross-sectional variation in the dependent variable doesnot eliminate the potentially spurious correlation over time between thetax variable and the share of compensation paid in options The secondcolumn of Table 4 isolates the pooled cross-sectional variation in corpo-rate tax rates by adding time dummies and dropping the rm xedeffects In this specication the coefcient on the tax variable is indistin-guishable from zero The coefcient on the institutional variable contin-ues to be large and statistically signicant The third column includes

Taxation of Executive Compensation 29

both rm xed effects and time dummies which tests for a relationshipbetween within-rm changes in the dependent variable and within-rmchanges in each explanatory variable In this specication the coefcienton the tax variable is again indistinguishable from zero The coefcientson the corporate governance variables are also statistically insignicant

The third column contains a specication that treats the Tax ReformAct of 1986 (TRA 86) as a natural experiment For rms and CEOs facingthe statutory corporate tax rate TRA 86 increased the relative payoff topaying compensation in options by 17 percent For a rm facing a zeromarginal tax rate it reduced the relative payoff by 45 percent The reduc-tion occurred because TRA 86 reduced the personal tax rate makingsalary and bonus more attractive while the decreased corporate rate hadno effect on zero-tax-rate rms Thus we would expect rms facing zeroor low marginal tax rates to reduce their use of options after TRA 86while rms facing the statutory rate would be expected to increase themFew rms face zero marginal tax rates year after year Therefore we tryto distinguish between rms that often face low tax rates and rms thatnearly always face the statutory rate We dene low-tax rms as oneswhose average marginal tax rate over 1984 to 1986 was below 02412

Then we ran a difference-in-differences regression to see if high-taxrms increased their use of options more than low-tax rms The interac-tion between post-TRA 86 and high corporate tax rates is the key vari-able The estimated coefcient on this variable is positive but small andnot statistically different from zero at the 95-percent level

It is important to note that in this specication as well the main corpo-rate governance variable the share of the rmrsquos shares owned by institu-tional investors performs as theory would predict The variable appearsto be quite robust Table 5 presents regressions of the share of compensa-tion paid in stock options on the fraction of the rmrsquos shares owned byinstitutional investors Since we have already established that these vari-ables have a strong longitudinal relationship the regressions here aredesigned to focus on the cross-sectional relationship between these twovariables The rst and third columns present regressions using data onlyfrom 1994 and nd a coefcient of roughly 02 (column 1 contains noother covariates while column 3 includes all of the other nontax covari-ates from the regressions in Tables 3 and 4) Since the median share of armrsquos stock owned by institutional investors in our sample rises from 19percent in 1980 to 49 percent in 1994 this coefcient would imply that this

12 This cutoff was chosen after inspecting the distribution of tax rates and essentiallyseparates rms that are always at or near the statutory rate from ones that are sometimesbelow it

TA

BL

E4

Com

pens

atio

nR

egre

ssio

nsth

atR

ely

onC

ross

-Sec

tion

alV

aria

tion

inC

orpo

rate

Tax

Rat

es

(1)

(2)

(3)

(4)

Tax

vari

able

s(a)

ln

(dif

fere

nce

inp

ayof

f)0

099

(00

41)

20

038

(00

36)

20

005

(00

44)

Pos

t-T

RA

862

000

5(0

013

)H

igh

corp

orat

eta

xra

te0

018

(00

17)

(Pos

t-T

RA

86)

3hi

gh0

024

(00

16)

corp

orat

eta

xra

te

Cor

por

ate

gove

rnan

ceva

riab

les

Frac

tion

of

rmrsquos

shar

es0

109

(00

27)

011

4(0

015

)2

000

9(0

030

)0

087

(00

34)

owne

dby

inst

itu

tion

alin

vest

ors

Size

ofth

e

rmrsquos

boar

dof

20

004

(00

01)

20

001

(00

01)

20

002

(00

01)

000

1(0

002

)d

irec

tors

Frac

tion

ofd

irec

tors

wh

o2

012

4(0

033

)2

000

3(0

018

)2

004

1(0

034

)0

001

(00

39)

are

insi

de

orgr

ay

Con

trac

t-th

eory

vari

able

sD

(CE

Ow

ealt

h)p

er$1

000

20

238

(00

94)

20

167

(00

53)

20

257

(00

96)

20

021

(01

47)

inFV

(b)

D(C

EO

wea

lth)

per

10

in2

560

(17

8)2

592

(10

5)2

594

(19

6)2

147

(04

86)

FV(c

)

Vol

atili

ty(d

)2

48(4

24)

162

(03

4)1

12(4

28)

106

(06

09)

ln(

rmm

arke

tva

lue)

003

3(

006)

000

8(0

003

)0

009

(00

06)

001

2(0

007

)

Stoc

km

arke

tre

turn

vari

able

sL

agge

dan

nual

rm

stoc

k2

000

2(0

009

)0

011

(00

10)

000

6(0

009

)2

001

2(0

013

)m

arke

tre

turn

Lag

ged

annu

alSamp

P50

02

005

6(0

023

)2

005

3(0

051

)2

003

9(0

049

)2

015

4(0

040

)re

turn

Reg

ress

ion

dia

gnos

tics

T

otal

sam

ple

size

5179

5179

5179

2534

Firm

s42

742

739

6A

dju

sted

R2

031

009

032

005

Firm

xed

effe

cts

Yes

No

Yes

No

Tim

ed

um

mie

sN

oY

esY

esN

o

Dep

end

ent

vari

able

isfr

acti

onof

ann

ualc

omp

ensa

tion

from

stoc

kop

tion

sSt

and

ard

erro

rsin

par

enth

eses

(a)

All

thre

esp

eci

cati

ons

use

tric

hoto

mou

sco

rpor

ate

tax

rate

sT

he

sam

ple

inco

lum

n(3

)u

ses

only

the

year

s19

84to

1988

sin

cew

ear

eu

sing

1986

asa

nat

ural

exp

erim

ent

(b)

Thi

sva

riab

leis

the

dol

lar

chan

gein

CE

Ow

ealt

hp

er$1

000

chan

gein

rm

mar

ketv

alu

e(s

eeJe

nse

nan

dM

urp

hy

1990

)

(c)

Thi

sva

riab

lein

the

dol

lar

char

gein

CE

Ow

ealt

hp

er10

-per

cent

chan

gein

rm

valu

e(s

eeH

alla

nd

Lie

bman

199

8)

(d)

Stan

dar

dd

evia

tion

ofr

mm

arke

tva

lue

intr

illio

ns

32 Hall amp Liebman

TABLE 5Large Institutional Investors and Stock-Option Compensation

The Cross-sectional Relationship between Institutional Investors andOption Grants

(1) (2) (3) (4)

Coefcient on fraction of 217 (067) 145 (014) 177 (069) 120 (015)rmrsquos shares owned byinstitutional investors

Sample period 1994 1980ndash1994 1994 1980ndash1994Contract-theory and stock No No Yes Yes

market covariatesTime dummies No Yes No Yes

The dependent variable is the share of compensation paid in options

factor alone can explain 6 percentage points of the 23-percentage-pointincrease in the median share of compensation paid in options that oc-curred over this time period Of course this variable is only one measureof the strengthening of corporate governance so the total impact of im-proved corporate governance could be greater The second and fourthcolumns of the table use the entire 1980ndash1994 sample and include timedummies to isolate the cross-sectional correlations Again there is evi-dence of a strong cross-sectional relationship between the two variablesalthough the coefcients are slightly smaller perhaps reecting a weakercorporate-governance role in the earlier period In sum there is a strongrelationship in both longitudinal and cross-sectional data between theinstitutional-investor variable and the options-percentage variable13

7 THE MILLION-DOLLAR RULE

In 1993 section 162(m) of the Internal Revenue Code was enactedlimiting the deductibility of executive compensation in excess of onemillion dollars unless the compensation was performance-related Pro-ponents of this legislation argued that it would slow the growth ofexecutive pay and tighten the link between rm performance and execu-tive compensation

The new provision became effective for tax year 1994 and applies tothe CEO and the other four most highly compensated executives in eachcompany Pay that is performance-related such as stock-option grants

13 The one specication in which the fraction owned by institutional investors is notsignicant is in column (2) of Table 4 a regression that includes both xed effects and timeeffects

Taxation of Executive Compensation 33

and bonuses paid for meeting clear performance goals is not affected bythis regulation provided that it has been approved by shareholdersThus the main impact of the provision is to limit the deductibility ofsalary in excess of a million dollars Woodlock and Antenucci (1997)studied the proxy statements of 376 rms and documented that mostrms responded to the new law by qualifying top executive pay asperformance-related Perry and Zenner (1999) identied 25 rms thatreduced salaries from above $1 million to below $1 million and foundthat 23 of the rms cited section 162(m) as their reason for doing so

In this section of the paper we investigate whether this provision hasaffected the level or structure of executive pay using the Execucompdata base which contains information on the highest-paid executives ineach of the rms in the SampP 500 SampP Midcap 400 and SampP SmallCap600 Between 1993 and 1998 the median salary of the Execucomp execu-tives rose from $230000 to $306000 The median value of total compen-sation (salary bonus and option grants) rose from $470000 to $882000Limiting the sample of CEOs indicates that the total compensation of themedian CEO rose from $11 million to $18 million over the period whilethe median salary rose from $441000 to $550000 Thus executive com-pensation clearly continued to rise at a rapid rate after the implementa-tion of the million-dollar rule though perhaps at a slower rate than itotherwise would have

If the regulation affects behavior it should have the largest impact onrms whose executives are receiving salaries that are above the limit orjust below it Thus we might expect to see rms whose executives arereceiving relatively high salaries to rely more heavily on performance-based pay and less heavily on salary in the years following the regulationPerry and Zenner (1999) and Livingston (1999) both nd that this is thecase This trend can be clearly identied in the Execucomp data set Forexample over 1993 to 1998 the median annualized growth rate of salaryfor executives whose 1993 salary was below $500000 in the previous yearwas 85 percent For executives with salary between $500000 and$700000 it was 64 percent for executives between $700000 and $850000it was 50 percent for executives between $750000 and $1 million it was31 percent and for executives above $1 million it was 00 percent14

The observation that executives at lower levels of salary had highergrowth rates is not very strong evidence of an impact of section 162(m)however because it is possible that low-salary executives would havehad higher growth rates even in the absence of the provision To explore

14 Of the 3137 executives for whom we have complete data between 1993 and 1998 2708had 1993 salaries below $500000 while only 35 had 1993 salaries above $1 million

34 Hall amp Liebman

whether this is the case we use Execucomp data from both before andafter the law change Ideally we would want to have data from manyyears before the change to examine whether there is an underlying pat-tern of faster salary growth for low-salary executives Unfortunately theExecucomp data begin in 1992 and therefore provide us with only the1992ndash1993 growth rate in compensation to use as a baseline15

We run regressions pooling data from 1992ndash1993 through 1997ndash1998of the annual percentage change in different forms of compensation on avariable million that is designed to measure the likelihood that anexecutiversquos compensation will be affected by section 162(m) We denemillion as the minimum of 1 and the executiversquos salary divided by $1million in the previous year Thus any executive with salary at or above$1 million would receive a value of 1 while an executive with a salary of$500000 would be coded as a 0516 We include this variable by itself andalso interacted with a dummy variable (after) that equals 1 for years1994 and after Thus the million variable is designed to pick up anyunderlying relationship between the level of salary and the growth rateof salary while the interacted one will identify any differential growthrate of salary by income in the period after section 162(m) took effect Wealso include a full set of year dummies the natural logarithm of thermrsquos market value and the rmrsquos annual rate of return over the twoprevious years as control variables To reduce the sensitivity of our re-sults to outliers we run quantile regressions and robust regressions

Table 6 contains our results The rst four columns are for all topexecutives while the last four columns focus on CEOs Columns (1) and(2) present estimates from robust regression for the percentage change insalary and the percentage change in total compensation (salary bonusstock grants and stock-option grants) In the salary growth regressionsfor all executives the coefcient on million is negative and signicantindicating that there is an underlying relationship between the level ofexecutive compensation and its subsequent growth rate The coefcienton million 3 after is negative as well suggesting that the negativerelationship between the level of salary and its subsequent growth rateintensied slightly after 1994 The relatively small magnitude suggeststhat an executive with a $1 million dollar salary would see his salarygrow at an annual rate that would be about 06 percent slower than anexecutive earning $500000

15 The HallndashLiebman data set contains only a combined salary and bonus variable makingit impossible to use it to analyze this issue16 We also ran the regressions (not reported) allowing for a nonlinear relationship betweenmillion and compensation growth and obtained substantively similar results

TA

BL

E6

The

Eff

ect

ofth

eM

illi

on-D

olla

rR

ule

onSa

lary

and

Tot

alC

ompe

nsat

ion

All

top

exec

uti

ves

CE

Os

only

Rob

ust

regr

essi

onQ

uan

tile

regr

essi

onR

obu

stre

gres

sion

Qu

anti

lere

gres

sion

D

in

Din

D

in

Din

D

in

Din

D

in

Din

sala

ryto

talc

omp

(a)

sala

ryto

talc

omp

sa

lary

tota

lcom

p

sala

ry

tota

lcom

p

mil

lio

n(b

)2

006

02

035

72

006

32

024

12

008

72

023

12

010

32

024

5(0

005

)(0

032

)(0

006

)(0

027

)(0

010

)(0

073

)(0

011

)(0

056

)m

illi

on

32

001

10

026

20

021

20

015

20

016

014

92

003

40

158

aft

er(c

)(0

005

)(0

033

)(0

006

)(0

029

)(0

010

)(0

076

)(0

011

)(0

057

)N

359

5035

950

359

5035

950

757

17

589

757

17

589

Stan

dar

der

rors

inp

aren

thes

esR

egre

ssio

nsal

soin

clu

de

year

du

mm

ies

stoc

km

arke

tre

turn

and

log

mar

ket

valu

e

(a)

Tot

alco

mp

ensa

tion

incl

ud

essa

lary

bon

us

rest

rict

edst

ock

and

stoc

k-op

tion

gran

ts

(b)

mil

lion

isd

ene

das

min

(1s

alar

yin

pre

viou

sye

ars

div

ided

by$1

mil

lion)

(c)

aft

eris

anin

dic

ator

vari

able

equ

alto

zero

befo

reth

ere

gula

tion

took

plac

e(1

993)

and

one

afte

rwar

d

36 Hall amp Liebman

In the second column the coefcient on million 3 after is essentiallyzero suggesting that any decrease in salary brought about by section162(m) was offset by increases in bonus and stock-option grants Theresults are similar for the median regressions There is some evidence ofa very small slowdown in growth of salary for executives likely to beaffected by section 162(m) but no sign of a change in total compensa-tion Taken together this provides evidence of a very small substitutionof performance-related pay for salary

The last four columns run identical tests but limit the sample to CEOsThe interaction of million 3 after continues to be small and negative inthe salary regressions In contrast the impact of 162(m) on total compen-sation appears to be positive and fairly substantial (about 15 percentagepoints per year) in these regressions That is the increase in bonus andstock options following the million-dollar rule more than offset the verysmall decline in salary

All of these conclusions should be interpreted somewhat cautiouslybecause our data provide us with only one pre-1993 control year Never-theless the data are consistent with section 162(m) having led to a veryminor slowdown in salary growth and one that is dramatically smallerthan what one would conclude without controlling for underlying differ-ential salary growth rates at different levels of salary Since total compen-sation for CEOs did not decline and perhaps increased there is evidenceof a minor substitution of performance-related pay for salary in responseto the regulation

8 EXECUTIVE COMPENSATION AND THEELASTICITY OF TAXABLE INCOME

Even if taxes have only a minor effect on executive compensation thestructure of executive compensation is important for tax policy be-cause high-income taxpayers are sources of signicant revenue and arethe focus of important debates about the efciency cost of taxation

Goolsbee (1999) has recently argued that executive decisions to exer-cise options are highly responsive to intertemporal differences in taxrates created by tax reforms He claims that much of the apparent declinein taxable income for high-income taxpayers between 1992 and 1993documented by Feldstein and Feenberg (1996) can be explained by execu-tives shifting option gains into 1992 in order to avoid the higher marginaltax rates that went into effect in 1993 By extension he implies that thehigh elasticities of taxable income with respect to personal tax ratesestimated by comparing the pre- and post-tax-reform taxable income of

Taxation of Executive Compensation 37

high-income individuals reect intertemporal shifting of income and notpermanent effects of taxation If this is true then the deadweight loss oftaxation is lower as is the cost of progressivity

While Goolsbee presents comprehensive evidence that option gainswere unusually high in 1992 his identication of the impact of tax rates onthe timing of option gains depends on correctly specifying a counter-factual level of options that would have been exercised in the absence ofthe tax change He simply assumes a linear time trend The purpose ofGoolsbeersquos time trend is presumably to account for the increasing use ofoptions over time However since options typically require a vestingperiod of at least a couple of years a much more direct measure of ex-pected option gains can be constructed based on the total value of vestedoptions held by the CEO Since options typically have vesting periods thatexceed two years this total can be treated as exogenous since it will not beaffected by the tax changes

More generally the value of options held by the CEO is likely to havea major effect on his exercise decisions If his options are out of themoney he will not have gains to realize If there has been a recent run-up in the market an optimizing CEO may exercise an unusually largeamount of options in order to diversify his overall portfolio and if in-the-money options are about to expire they will be exercised in thecurrent year

Our evidence suggests that much of the apparent tax shifting caninstead be attributed to stock market performance and the timing of pastoption grants Using our 15-year panel we replicate Goolsbeersquos (1999)result for the 1993 period but show that the impact of taxes is notpresent in other time periods Moreover even the result for the 1993period disappears when appropriate controls for past option grants andstock price appreciation are included

The top three rows of Table 7 present data on the taxable incomes ofCEOs in the HallndashLiebman sample For these results our panel has beennarrowed to include only CEOs whose rmrsquos scal years correspondwith tax years In addition we create a balanced panel for each of thethree tax-reform periods by excluding any rm for which there are notcomplete data for the period surrounding the reform

The fourth row of the table shows that the 1993 tax act raised the topmarginal personal income tax rate from 310 to 396 percent effective fortax year 1993 with a further increase occurring in 1994 due to the uncap-ping of the Medicare payroll tax The top row of the table shows thattaxable income was higher in 1992 than for any other year in our sample(all data are in 1994 dollars) Option gains were particularly high in1992mdashmore than double their 1991 level and 68 percent higher than their

TA

BL

E7

Tax

Ref

orm

Sto

ckM

arke

tP

erfo

rman

cea

ndth

eT

imin

gof

Opt

ion

Gai

ns

Ave

rage

stoc

km

arke

tM

ean

CE

OM

ean

CE

Ore

turn

for

Mea

nva

lue

taxa

ble

sala

ryan

dM

ean

CE

OT

opfe

der

al

rms

inof

opti

ons

(Val

ue

ofop

tion

Sam

ple

inco

me

bonu

sop

tion

gain

sp

erso

nal

sam

ple

for

held

atbe

gga

ins)

(va

lue

Ref

orm

size

Yea

r($

mil

)($

mil

)($

mil

)M

TR

pre

vye

ar(

)of

year

ofop

tion

she

ld)

ER

TA

8127

919

800

874

064

80

163

7025

80

460

354

1981

083

10

672

009

869

245

070

61

3919

820

872

067

90

099

5011

90

686

144

1983

099

00

724

025

250

355

116

62

16

TR

A86

295

1985

119

70

837

024

050

76

116

02

0719

861

448

091

70

345

5042

81

776

194

1987

165

50

959

059

338

519

12

229

266

1988

163

21

058

039

028

07

237

71

6419

891

656

105

20

414

2817

92

345

176

1993

272

1990

209

51

028

061

828

232

288

02

15T

axA

ct19

911

875

102

00

547

312

01

268

82

0319

922

678

107

01

250

3146

64

552

275

1993

232

41

209

074

339

620

64

299

173

1994

196

41

262

037

842

515

45

113

074

Sam

ple

isH

allndash

Lie

bman

rm

sw

ith

Dec

embe

r

scal

year

sth

atha

dco

mp

lete

dat

afo

ron

eof

the

thre

eta

x-re

form

per

iod

sst

ud

ied

Taxation of Executive Compensation 39

1994 level These patterns are consistent with Goolsbeersquos claim that tax-payers responded to the anticipated increase in marginal tax rates byshifting gains into 1992

The evidence from the other two tax reforms is less clear suggestingthat there might be factors other than tax rates determining the timing ofoption gains ERTA 1981 reduced the top marginal tax rate from 70percent in 1980 to 50 percent in 1982 However there is no sign thatoption gains were reduced in 1980 and 1981 and then increased in 198217

There was a big increase in option gains in 1983 however Similarly thehighest year of stock-option gains in the period around 1986 was 1987 astrange year to take gains since taxpayers would have known that thepersonal rate would be 10 percentage points lower if they waited untilthe following year18

What all the peak years of option gains have in common is that theyfollow years of strong stock market performance The fth row of Table 7shows the average stock market return during the previous year for therms in our sample The large value of option gains in 1983 followed a355-percent increase in the stock market The 1992 boom in option gainsfollowed a 466-percent increase in the stock market the more modest1987 peak in option gains followed a 19-percent gain (and a 43-percentgain two years before)

Strong stock market performance raises the value of options availableto exercise The sixth row of the table shows the mean value of optionsheld by CEOs at the beginning of the year All three of the peak years ofoption gains correspond with peak years in terms of the value of optionsavailable for exercise Since stock market appreciation during the yearalso increases the value of options available to exercise later in that yearwe need to allow for current-year stock market appreciation as well

Table 8 contains Goolsbee-style regressions for each of the tax re-forms The rst two columns reprint the results from Goolsbee (1999)The third and fourth columns replicate his results in our sample Inparticular with a specication similar to his we are able to replicate hisnding that the current after-tax share has a positive coefcient and thefollowing yearrsquos after-tax share has a coefcient of roughly negative oneColumns (5) through (8) show that this pattern was unique to the 1993tax reform Indeed in the TRA 86 regressions the coefcient on the

17 As we explained above for executives covered by the maximum tax on earned incomethere was not a decline in tax rates between 1981 and 198218 It is possible that the large stock market decline in October 1997 may have promptedsome people to exercise options earlier than they otherwise would have However anincrease in option exercises typically follows stock market increases not declines

TA

BL

E8

Att

empt

sto

Rep

lica

teth

eT

ax-S

hift

ing

Res

ult

inO

ther

Tim

eP

erio

ds

Goo

lsbe

eG

ools

bee

1993

1993

Ent

ire

Ent

ire

regr

essi

onre

gres

sion

refo

rmre

form

TR

A86

TR

A86

ER

TA

81E

RT

A81

1980

ndash199

419

80ndash1

994

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

ln[1

2ta

x(t)

]1

244

108

20

455

032

62

017

62

017

22

016

72

016

70

026

20

068

(01

34)

(01

27)

(05

37)

(05

42)

(02

10)

(02

09)

(00

90)

(00

91)

(00

67)

(00

65)

ln[1

2ta

x(t1

1)]

20

867

20

764

21

214

20

917

047

80

722

20

103

004

30

176

018

3(0

118

)(0

112

)(0

491

)(0

499

)(0

157

)(0

174

)(0

072

)(0

072

)(0

068

)(0

066

)

Tim

e0

135

007

62

000

52

036

40

041

000

80

094

003

50

056

002

7(0

008

)(0

008

)(0

049

)(0

049

)(0

028

)(0

028

)(0

025

)(0

026

)(0

003

)(0

003

)

ln(m

arke

tva

lue)

066

10

489

023

70

334

039

5(0

015

)(0

058

)(0

039

)(0

042

)(0

017

)

Yea

rs19

91ndash1

995

1991

ndash199

519

90ndash1

994

1990

ndash199

419

85ndash1

989

1985

ndash198

919

80ndash1

983

1980

ndash198

319

80ndash1

994

1980

ndash199

4

Ad

just

edR

20

720

770

610

630

690

760

800

810

630

68

Obs

erva

tion

s13

856

132

371

385

136

41

470

143

31

113

106

23

968

385

9

Dep

end

ent

vari

able

isth

elo

gof

taxa

ble

inco

me

All

regr

essi

ons

con

tain

rm

xe

def

fect

sG

ools

bee

regr

essi

on(2

)al

soco

ntai

nstw

oot

her

vari

able

sno

tsh

own

inth

eta

ble

Stan

dar

der

rors

inp

aren

thes

es

Taxation of Executive Compensation 41

TABLE 9Alternative Explanations for the Timing of Option Exercises

1993 Entirereform TRA 86 ERTA 81 1980ndash1994

(1) (2) (3) (4)

ln[12tax(t)] 0108 0111 20051 0201(0324) (0136) (0048) (0057)

ln[12tax(t11)] 20134 0573 0125 0150(0336) (0156) (0053) (0066)

ln(market value) 0418 0177 0338 0418(0061) (0039) (0046) (0163)

Firmrsquos stock market return (t) 0298 0268 0214 0258(0044) (0045) (0031) (0023)

Firmrsquos stock market return 0162 0206 0031 0128(t21) (0041) (0008) (0029) (0021)

ln(value of options held by 0004 0008 0001 0009CEO at start of year) (0003) (0003) (0002) (0001)

Years 1990ndash1994 1985ndash1989 1980ndash1983 1980ndash1994Adjusted R2 064 071 082 068Observations 1349 1427 1036 3812

The dependent variable is the log of taxable income All regressions contain rm xed effects Stan-dard errors in parentheses

following yearrsquos tax rate is positive and signicant Columns (9) and (10)pool all three reforms and fail to replicate the pattern that would reectintertemporal shifting19

It is possible that the ease with which one can shift the timing ofoption gains is asymmetric It may be harder to shift option gains for-ward in time since some options expire and therefore must be exercisednow In addition options were a much smaller portion of compensationin the early 1980s so perhaps the ERTA 81 results are not surprisingNonetheless these results cast doubt on the proposition that this is thechannel through which spurious estimates of the elasticity of taxableincome have occurred for previous tax reforms and they suggest thatwe should look harder for alternative explanations for the timing ofoption gains

Table 9 attempts to predict options gains using the lagged and currentincrease in a rmrsquos market value and the value of the options held by theCEO at the start of the year The two stock market variables stronglypredict option gains while the value of options held has the correct sign

19 The results also fail to suggest large permanent effects of marginal tax rates on taxableincome

42 Hall amp Liebman

in all of the specications and is signicant in the pooled sample eventhough it is highly correlated with the two stock market return variablesIn contrast the tax-timing explanation fails even in the 1993 period

It is important to realize that our results do not invalidate Goolsbeersquoscritique of claims that aggregate revenue data from 1992 and 1993 provethat elasticities of taxable income are large We have simply given analternative explanation for why taxable incomes were unusually high in1992 However our results do cast doubt on the conclusion that largetiming shifts are ubiquitous and capable of explaining away the morecareful estimates of taxable income elasticities that have been performedusing micro data (eg Feldstein 1995 Auten and Carroll 1995)

One further point is in order It is sometimes argued that the combina-tion of increased use of options and the booming stock market explainswhy federal tax revenues have been so much higher than expected dur-ing the 1990s However while large option gains increase revenue fromthe personal income tax they produce offsetting deductions for cor-porations The net impact on revenue is only the difference in the twomarginal tax rates times the option gain While the gap between thepersonal rate and the effective corporate rate is nontrivial the impact oflarge option exercises on total revenue is substantially smaller than theamount by which the exercises inate personal tax revenue and reducecorporate tax revenue

9 CONCLUSION

We have described the tax rules for executive pay in detail and analyzedhow changes in various tax rates affect the tax advantages of stock optionsrelative to salary and bonus We nd that there is a moderate tax advan-tage to optionsmdashcurrently about $4 per $100 in pretax executive paymdashand that changes in the tax advantage of options over time have had atmost a modest impact on the composition of pay Corporate governancefactors particularly the role of large investors appear to be more impor-tant in explaining the dramatic increase in option pay

In addition our evidence suggests that more direct attempts to use taxpolicy to inuence executive compensation have had little effect Wend that the million-dollar rule led companies to substitute perfor-mance-based pay for salary But our evidence suggests that this substitu-tion was quite modest and there is no evidence that the total level of paywas reduced Overall although the stock-option explosion has dramati-cally increased the link between pay and performance this change isdue almost entirely to nontax factors

Taxation of Executive Compensation 43

REFERENCES

Auerbach Alan J (1996) ldquoThe Dynamic Effects of Tax Law Asymmetriesrdquo Re-view of Economic Studies 53451ndash405

mdashmdashmdash and James Poterba (1987) ldquoTax Loss Carryforwards and Corporate TaxIncentivesrdquo In The Effects of Taxation on Capital Accumulation M Feldstein(ed) Chicago University of Chicago Press

Altshuler Rosanne and Alan Auerbach (1990) ldquoThe Signicance of Tax LawAsymmetries An Empirical Investigationrdquo Quarterly Journal of Economics 10561ndash86

Auten Gerald and Robert Carroll (1995) ldquoThe Effect of Income Taxes on House-hold Behaviorrdquo Ofce of Tax Analysis US Department of the Treasury

Baker George P and Brian J Hall (1998) ldquoCEO Incentives and Firm SizerdquoNBER Working Paper no 6868 December

Core John and Wayne Guay (1999) ldquoThe Use of Equity Grants to ManageOptimal Equity Incentive Levelsrdquo Draft

Feldstein M (1995) ldquoThe Effect of Marginal Tax Rates on Taxable Income APanel Study of the 1986 Tax Reform Actrdquo Journal of Political Economy 103 (no3)551ndash72

mdashmdashmdash and D Feenberg (1996) ldquoThe Effect of Increased Tax Rates on TaxableIncome and Economic Efciency A Preliminary Analysis of the 1993 Tax RateIncreasesrdquo In Tax Policy and the Economy vol 10 James M Poterba (ed)Cambridge MA MIT Press

Gompers Paul A and Andrew Metrick (1998) ldquoInstitutional Investors andEquity Pricesrdquo NBER Working Paper no 6723

Goolsbee Austan (1999) ldquoWhat Happens When You Tax the Rich Evidencefrom Executive Compensationrdquo NBER Working Paper no 6333

Graham John R (1996) ldquoProxies for the Corporate Marginal Tax Raterdquo Journal ofFinancial Economics 42187ndash221

Hall Brian J (1999) ldquoThe Design of Multi-year Stock Option Plansrdquo Journal ofApplied Corporate Finance 12(2)97ndash106

mdashmdashmdash and Jeffrey B Liebman (1998) ldquoAre CEOs Really Paid Like Bureau-cratsrdquo Quarterly Journal of Economics CXII(no 3)653ndash691

Jensen Michael C and William M Meckling (1976) ldquoTheory of the Firm Mana-gerial Behavior Agency Costs and Ownership Structurerdquo Journal of FinancialEconomics III305ndash360

mdashmdashmdash and Kevin J Murphy (1990) ldquoPerformance Pay and Top-ManagementIncentivesrdquo Journal of Political Economy XCVIII225ndash264

Kaplan Steven N (1997) ldquoThe Evolution of US Corporate Governance WeAre All Henry Kravis Nowrdquo Draft

Livingstone Jane R (1999) ldquoExecutive Compensation Contracting and Re-sponses to Increased Tax Costsrdquo Louisiana State University June

Miller Merton H and Myron S Scholes (1982) ldquoExecutive CompensationTaxes and Incentivesrdquo In Financial Economics Essays in Honor of Paul CootnerWilliam F Sharpe and Cathryn M Cootner (eds) Princeton NJ Prentice-Hall

Murphy Kevin J (1999) ldquoExecutive Compensationrdquo forthcoming in OrleyAshenfelter and David Card (eds) Handbook of Labor Economics Vol 3 NorthHolland

Perry Todd and Marc Zenner (1999) ldquoPay for Performance Government Regu-

24 Hall amp Liebman

all stock options held by the CEO at a point of time More importantlysince we have the details about the stock options held (number exer-cise price time to maturity etc) we can precisely calculate the changein the value of a CEOrsquos stock option holdings for a given change in rmvalue

62 Identication IssuesThe numerous tax changes over our sample period and the specic waysin which changes in personal corporate and capital gains rates arepredicted to affect the composition of CEO compensation give us un-usually rich sources of identication However because many of thechanges in tax rates would be expected to affect all of the CEOs in oursample in a similar way we will need to pay particular attention toseparating out the tax effects from underlying time trends and otherfactors that changed over time Controlling for underlying trends is par-ticularly important in this study because we are focusing on an out-comemdashthe use of stock optionsmdashthat has increased rapidly over thepast two decades and that many practitioners believe was importantlyaffected by nontax considerations Thus the tax effects we are trying toexplain are deviations from a rising trend

We take two steps to try to separate out the tax effects from theunderlying trend First we include key nontax factors that could poten-tially explain the increasing reliance on performance-based compensa-tion It has been suggested that the dramatic increase in incentive-basedpay is the result of the remarkably poor shareholder returns during the1970s which spurred the LBO and takeover movements of the 1980sAccording to this story (Kaplan 1997) shareholders became much morepowerful via the rise of institutional investors even as the LBO andtakeover movements waned Because the inuence of institutional inves-tors is thought to be one of the most important mechanisms of strongcorporate governance we use the share of each companyrsquos stock ownedby institutional investors (dened as institutions with more than $100million under management) as an explanatory variable9 The percentageof shares owned by large institutional investors increased from about 20percent to almost 50 percent in 1994 an upward trend that closelymatches the sharp rise in the share of compensation in the form ofoptions Indeed in our sample the annual average (over all of the rmsin our sample) share of stock owned by institutional investors has acorrelation of 09 with the annual average share of compensation paid inoptions

9 This variable is described in Gompers and Metrick (1998)

Taxation of Executive Compensation 25

In addition we include two other variables that proxy for strongercorporate governance The rst is the size of the board Evidence suggeststhat boards with smaller numbers of directors reduce the inuence of theCEO and better represent shareholder interests (Yermack 1995) Boardsize fell by approximately 15 percent between 1980 and 1994 The secondvariable is the fraction of the rmrsquos board members who are inside direc-tors or gray directors (noninsiders who have business dealings with thecompany) Firms with fewer outsiders are less likely to act in the share-holdersrsquo interests and we would predict that they would be less likely tohave performance-based compensation for their CEOs The average per-centage of inside and gray directors has decreased from about 45 percentin 1980 to about 35 percent in 1994

Second we allow for cross-sectional variation in corporate tax ratesCorporate marginal tax rates vary because rms may be eligible for tax-loss carrybacks and carryforwards investment tax credits and the alter-native minimum tax (see Auerbach 1986 Auerbach and Altshuler1990 Auerbach and Poterba 1987) Because these cross-sectional differ-ences imply that different rms should respond differently to a givenchange in tax rates they potentially provide us with a way to isolatethe tax effects Firm-specic corporate marginal tax rates are notori-ously difcult to calculate from publicly-available data We use a tricho-tomous variable that equals zero if the rm has tax-loss carryforwardsand negative earnings in a given year equals 05 times the statutorycorporate rate if the rm has only one of those conditions and equalsthe statutory rate if the rm has neither loss carryforwards nor negativeearnings10

The cross-sectional variation in the corporate tax rate may not be ex-ogenous Firms that perform poorly could face low marginal tax ratesand also be particularly likely or unlikely to provide performance-basedpay Because most of our specications use xed effects the correlationwe must be concerned about is between changes in rm performanceand changes in compensation We deal with this by controlling forlagged rm stock market performance in our regressions This shouldeliminate spurious correlation between tax rates and compensation thatis jointly caused by rm performance

10 Graham (1996) presents evidence that the trichotomous variable performs nearly as wellas his recommended simulated marginal tax rate Plesko (1999) compares marginal taxrates calculated from Compustat data with actual marginal tax rates from rm tax returnsHe nds that simple binary variables are more reliable measures of the marginal tax ratethan is either the Graham or the trichotomous variable We obtain substantively similarresults when we switch from the trichotomous variable to the binary variables recommendby Plesko

26 Hall amp Liebman

63 ResultsThe dependent variable in our regressions is the fraction of a CEOrsquos totalannual compensation (measured as the BlackndashScholes value of stock-option grants plus salary and bonus) that comes from stock-optiongrants We choose this reduced-form specication because our interest isin the composition of CEO compensation However the factors thataffect the composition of compensation might also affect the levels andit will be important to keep this in mind in interpreting our results

Optimal-contracting theory generally makes predictions about the in-centives provided by the total amount of rm stock and stock optionsowned by the CEO We choose to model the ow of option grants ratherthan the ultimate performance sensitivity of the CEOrsquos compensationand stock and stock-option holdings because we have in mind anadjustment-cost model in which it takes time for a CEOrsquos holdings ofstock and stock options to reach the optimal level11 In our regressionswe include two measures of the current performance sensitivity of theCEOrsquos wealth to rm performance as explanatory variables to reect thedistance of the CEOrsquos contract from the optimal contract

Table 3 presents results in which the tax effects are identied solely byvariation in tax rates over time The fraction of annual compensation paidin stock options is regressed on tax variables corporate governance vari-ables contract-theory variables and variables reecting stock market re-turns Both of the regressions in this table include rm xed effects Incolumn (1) the fraction in annual compensation paid in stock options isregressed on the log difference in payoff (the combined payoff to both therm and the CEO) from receiving compensation in stock options ratherthan salary and bonus where the payoff calculations are based on statu-tory marginal tax rates This variable is dened as the log difference be-tween equation (3) and equation (6) in section 4 The coefcient on the taxvariable is large and statistically signicant at the 95-percent level Thepoint estimate implies that a one-percent increase in the payoff differencebetween stock options and salary and bonus results in a 24-percentage-point increase in the share of compensation paid in stock options

The corporate governance variables all have the predicted sign Firmswith a higher fraction of their shares owned by institutional investorsuse more performance-based pay (stock options) Firms with large corpo-rate boards or a large share of inside directors are less likely to use stockoptions

The two contract-theory variables are the dollar change in CEO wealth

11 See Core and Guay (1999) for evidence that rms adjust in this way

TABLE 3Compensation Regressions that Rely on Variation in

Tax Rates over Time

(1) (2)

Tax variablesln (difference in payoff)(a) 2393 (0349)ln (12statutory personal 20015 (0034)

rate)ln (12statutory cap gains 0330 (0171)

rate)ln (12statutory corporate 0399 (0148)

rate)

Corporate governance variablesFraction of rmrsquos shares 0043 (029) 0052 (0028)

owned by institutionalinvestors

Size of the rmrsquos board of 20002 (001) 20003 (0001)directors

Fraction of directors who 20082 (033) 20092 (0030)are inside or gray

Contract-theory variablesD (CEO wealth) per $1000 20261 (0095) 20247 (0087)

in FV(b)

D (CEO wealth) per 10 in 2583 (189) 2554 (201)FV(c)

Volatility(d) 521 (424) 261 (369)ln (rm market value) 0019 (0006) 0019 (0006)

Stock market return variablesLagged annual rm stock 0006 (0009) 0003 (0008)

market returnLagged annual SampP 500 re- 20035 (0023) 20029 (0023)

turn

Regression diagnosticsTotal sample size 5179 5179Firms 427 427Adjusted R2 0313 0311

Dependent variable share of compensation paid in options Standard errors in parentheses Bothregressions in this table include rm xed effects

(a) Log difference in payoff is the log difference in combined rm and CEO after-tax share betweenpaying compensation in options and paying salary

(b) This variable is the dollar change in CEO wealth per $1000 change in rm market value (seeJensen and Murphy 1990)

(c) This variable in the dollar charge in CEO wealth per 10-percent change in rm value (see Hall andLiebman 1998)

(d) Standard deviation of rm market value in trillions

28 Hall amp Liebman

per $1000 change in rm value (Jensen and Murphy 1990) and thedollar change in CEO wealth per 10-percent change in rm value (Halland Liebman 1998) As discussed in Baker and Hall (1998) these twowealth sensitivity measures reect two different concepts of how closelyaligned the managersrsquo incentives are with the interests of the sharehold-ers In the regression they are meant to measure how far the CEOrsquosexisting contract and ownership of stock and stock options is from theoptimal level The negative coefcients on the two variables suggeststhat rms that are below their optimal pay-to-performance sensitivityare more likely to give more stock options in the current period

The second column replaces the tax variable motivated by deferraladvantages with the three statutory rates The corporate after-tax sharehas a positive coefcient as we would expect because the tax advantageof stock options increases as the corporate rate falls The coefcient onthe personal tax rate is small and statistically insignicant The sign ofthe coefcient on the capital gains rate is the opposite of that predictedby tax-avoidance theory

Note that these results rely on time-series variation alone This is not avery convincing test Since the log difference in payoffs mostly rises overthe time period covered in our sample as does the share of compensa-tion paid in options it is possible that the tax variable is simply reectingother factors that were changing over this time period that we have notincluded in our regression Therefore it is important to look at specica-tions that rely on cross-sectional variation in the tax variables

Table 4 contains specications that rely on cross-sectional variation incorporate tax rates The rst column repeats the specication from col-umn (1) of Table 3 but uses the trichotomous measures of rm corporatetax rates in place of the statutory rate in calculating each rmrsquos logdifference in payoff from paying compensation in options rather than insalary and bonus As in the previous table we nd the positive coef-cient on the log difference in payoff just as the theory suggest How-ever the coefcient estimate is now much smaller suggesting that aone-percent increase in the log difference in payoff produces a 01-percentage-point increase in the share of compensation paid in options

Introducing cross-sectional variation in the dependent variable doesnot eliminate the potentially spurious correlation over time between thetax variable and the share of compensation paid in options The secondcolumn of Table 4 isolates the pooled cross-sectional variation in corpo-rate tax rates by adding time dummies and dropping the rm xedeffects In this specication the coefcient on the tax variable is indistin-guishable from zero The coefcient on the institutional variable contin-ues to be large and statistically signicant The third column includes

Taxation of Executive Compensation 29

both rm xed effects and time dummies which tests for a relationshipbetween within-rm changes in the dependent variable and within-rmchanges in each explanatory variable In this specication the coefcienton the tax variable is again indistinguishable from zero The coefcientson the corporate governance variables are also statistically insignicant

The third column contains a specication that treats the Tax ReformAct of 1986 (TRA 86) as a natural experiment For rms and CEOs facingthe statutory corporate tax rate TRA 86 increased the relative payoff topaying compensation in options by 17 percent For a rm facing a zeromarginal tax rate it reduced the relative payoff by 45 percent The reduc-tion occurred because TRA 86 reduced the personal tax rate makingsalary and bonus more attractive while the decreased corporate rate hadno effect on zero-tax-rate rms Thus we would expect rms facing zeroor low marginal tax rates to reduce their use of options after TRA 86while rms facing the statutory rate would be expected to increase themFew rms face zero marginal tax rates year after year Therefore we tryto distinguish between rms that often face low tax rates and rms thatnearly always face the statutory rate We dene low-tax rms as oneswhose average marginal tax rate over 1984 to 1986 was below 02412

Then we ran a difference-in-differences regression to see if high-taxrms increased their use of options more than low-tax rms The interac-tion between post-TRA 86 and high corporate tax rates is the key vari-able The estimated coefcient on this variable is positive but small andnot statistically different from zero at the 95-percent level

It is important to note that in this specication as well the main corpo-rate governance variable the share of the rmrsquos shares owned by institu-tional investors performs as theory would predict The variable appearsto be quite robust Table 5 presents regressions of the share of compensa-tion paid in stock options on the fraction of the rmrsquos shares owned byinstitutional investors Since we have already established that these vari-ables have a strong longitudinal relationship the regressions here aredesigned to focus on the cross-sectional relationship between these twovariables The rst and third columns present regressions using data onlyfrom 1994 and nd a coefcient of roughly 02 (column 1 contains noother covariates while column 3 includes all of the other nontax covari-ates from the regressions in Tables 3 and 4) Since the median share of armrsquos stock owned by institutional investors in our sample rises from 19percent in 1980 to 49 percent in 1994 this coefcient would imply that this

12 This cutoff was chosen after inspecting the distribution of tax rates and essentiallyseparates rms that are always at or near the statutory rate from ones that are sometimesbelow it

TA

BL

E4

Com

pens

atio

nR

egre

ssio

nsth

atR

ely

onC

ross

-Sec

tion

alV

aria

tion

inC

orpo

rate

Tax

Rat

es

(1)

(2)

(3)

(4)

Tax

vari

able

s(a)

ln

(dif

fere

nce

inp

ayof

f)0

099

(00

41)

20

038

(00

36)

20

005

(00

44)

Pos

t-T

RA

862

000

5(0

013

)H

igh

corp

orat

eta

xra

te0

018

(00

17)

(Pos

t-T

RA

86)

3hi

gh0

024

(00

16)

corp

orat

eta

xra

te

Cor

por

ate

gove

rnan

ceva

riab

les

Frac

tion

of

rmrsquos

shar

es0

109

(00

27)

011

4(0

015

)2

000

9(0

030

)0

087

(00

34)

owne

dby

inst

itu

tion

alin

vest

ors

Size

ofth

e

rmrsquos

boar

dof

20

004

(00

01)

20

001

(00

01)

20

002

(00

01)

000

1(0

002

)d

irec

tors

Frac

tion

ofd

irec

tors

wh

o2

012

4(0

033

)2

000

3(0

018

)2

004

1(0

034

)0

001

(00

39)

are

insi

de

orgr

ay

Con

trac

t-th

eory

vari

able

sD

(CE

Ow

ealt

h)p

er$1

000

20

238

(00

94)

20

167

(00

53)

20

257

(00

96)

20

021

(01

47)

inFV

(b)

D(C

EO

wea

lth)

per

10

in2

560

(17

8)2

592

(10

5)2

594

(19

6)2

147

(04

86)

FV(c

)

Vol

atili

ty(d

)2

48(4

24)

162

(03

4)1

12(4

28)

106

(06

09)

ln(

rmm

arke

tva

lue)

003

3(

006)

000

8(0

003

)0

009

(00

06)

001

2(0

007

)

Stoc

km

arke

tre

turn

vari

able

sL

agge

dan

nual

rm

stoc

k2

000

2(0

009

)0

011

(00

10)

000

6(0

009

)2

001

2(0

013

)m

arke

tre

turn

Lag

ged

annu

alSamp

P50

02

005

6(0

023

)2

005

3(0

051

)2

003

9(0

049

)2

015

4(0

040

)re

turn

Reg

ress

ion

dia

gnos

tics

T

otal

sam

ple

size

5179

5179

5179

2534

Firm

s42

742

739

6A

dju

sted

R2

031

009

032

005

Firm

xed

effe

cts

Yes

No

Yes

No

Tim

ed

um

mie

sN

oY

esY

esN

o

Dep

end

ent

vari

able

isfr

acti

onof

ann

ualc

omp

ensa

tion

from

stoc

kop

tion

sSt

and

ard

erro

rsin

par

enth

eses

(a)

All

thre

esp

eci

cati

ons

use

tric

hoto

mou

sco

rpor

ate

tax

rate

sT

he

sam

ple

inco

lum

n(3

)u

ses

only

the

year

s19

84to

1988

sin

cew

ear

eu

sing

1986

asa

nat

ural

exp

erim

ent

(b)

Thi

sva

riab

leis

the

dol

lar

chan

gein

CE

Ow

ealt

hp

er$1

000

chan

gein

rm

mar

ketv

alu

e(s

eeJe

nse

nan

dM

urp

hy

1990

)

(c)

Thi

sva

riab

lein

the

dol

lar

char

gein

CE

Ow

ealt

hp

er10

-per

cent

chan

gein

rm

valu

e(s

eeH

alla

nd

Lie

bman

199

8)

(d)

Stan

dar

dd

evia

tion

ofr

mm

arke

tva

lue

intr

illio

ns

32 Hall amp Liebman

TABLE 5Large Institutional Investors and Stock-Option Compensation

The Cross-sectional Relationship between Institutional Investors andOption Grants

(1) (2) (3) (4)

Coefcient on fraction of 217 (067) 145 (014) 177 (069) 120 (015)rmrsquos shares owned byinstitutional investors

Sample period 1994 1980ndash1994 1994 1980ndash1994Contract-theory and stock No No Yes Yes

market covariatesTime dummies No Yes No Yes

The dependent variable is the share of compensation paid in options

factor alone can explain 6 percentage points of the 23-percentage-pointincrease in the median share of compensation paid in options that oc-curred over this time period Of course this variable is only one measureof the strengthening of corporate governance so the total impact of im-proved corporate governance could be greater The second and fourthcolumns of the table use the entire 1980ndash1994 sample and include timedummies to isolate the cross-sectional correlations Again there is evi-dence of a strong cross-sectional relationship between the two variablesalthough the coefcients are slightly smaller perhaps reecting a weakercorporate-governance role in the earlier period In sum there is a strongrelationship in both longitudinal and cross-sectional data between theinstitutional-investor variable and the options-percentage variable13

7 THE MILLION-DOLLAR RULE

In 1993 section 162(m) of the Internal Revenue Code was enactedlimiting the deductibility of executive compensation in excess of onemillion dollars unless the compensation was performance-related Pro-ponents of this legislation argued that it would slow the growth ofexecutive pay and tighten the link between rm performance and execu-tive compensation

The new provision became effective for tax year 1994 and applies tothe CEO and the other four most highly compensated executives in eachcompany Pay that is performance-related such as stock-option grants

13 The one specication in which the fraction owned by institutional investors is notsignicant is in column (2) of Table 4 a regression that includes both xed effects and timeeffects

Taxation of Executive Compensation 33

and bonuses paid for meeting clear performance goals is not affected bythis regulation provided that it has been approved by shareholdersThus the main impact of the provision is to limit the deductibility ofsalary in excess of a million dollars Woodlock and Antenucci (1997)studied the proxy statements of 376 rms and documented that mostrms responded to the new law by qualifying top executive pay asperformance-related Perry and Zenner (1999) identied 25 rms thatreduced salaries from above $1 million to below $1 million and foundthat 23 of the rms cited section 162(m) as their reason for doing so

In this section of the paper we investigate whether this provision hasaffected the level or structure of executive pay using the Execucompdata base which contains information on the highest-paid executives ineach of the rms in the SampP 500 SampP Midcap 400 and SampP SmallCap600 Between 1993 and 1998 the median salary of the Execucomp execu-tives rose from $230000 to $306000 The median value of total compen-sation (salary bonus and option grants) rose from $470000 to $882000Limiting the sample of CEOs indicates that the total compensation of themedian CEO rose from $11 million to $18 million over the period whilethe median salary rose from $441000 to $550000 Thus executive com-pensation clearly continued to rise at a rapid rate after the implementa-tion of the million-dollar rule though perhaps at a slower rate than itotherwise would have

If the regulation affects behavior it should have the largest impact onrms whose executives are receiving salaries that are above the limit orjust below it Thus we might expect to see rms whose executives arereceiving relatively high salaries to rely more heavily on performance-based pay and less heavily on salary in the years following the regulationPerry and Zenner (1999) and Livingston (1999) both nd that this is thecase This trend can be clearly identied in the Execucomp data set Forexample over 1993 to 1998 the median annualized growth rate of salaryfor executives whose 1993 salary was below $500000 in the previous yearwas 85 percent For executives with salary between $500000 and$700000 it was 64 percent for executives between $700000 and $850000it was 50 percent for executives between $750000 and $1 million it was31 percent and for executives above $1 million it was 00 percent14

The observation that executives at lower levels of salary had highergrowth rates is not very strong evidence of an impact of section 162(m)however because it is possible that low-salary executives would havehad higher growth rates even in the absence of the provision To explore

14 Of the 3137 executives for whom we have complete data between 1993 and 1998 2708had 1993 salaries below $500000 while only 35 had 1993 salaries above $1 million

34 Hall amp Liebman

whether this is the case we use Execucomp data from both before andafter the law change Ideally we would want to have data from manyyears before the change to examine whether there is an underlying pat-tern of faster salary growth for low-salary executives Unfortunately theExecucomp data begin in 1992 and therefore provide us with only the1992ndash1993 growth rate in compensation to use as a baseline15

We run regressions pooling data from 1992ndash1993 through 1997ndash1998of the annual percentage change in different forms of compensation on avariable million that is designed to measure the likelihood that anexecutiversquos compensation will be affected by section 162(m) We denemillion as the minimum of 1 and the executiversquos salary divided by $1million in the previous year Thus any executive with salary at or above$1 million would receive a value of 1 while an executive with a salary of$500000 would be coded as a 0516 We include this variable by itself andalso interacted with a dummy variable (after) that equals 1 for years1994 and after Thus the million variable is designed to pick up anyunderlying relationship between the level of salary and the growth rateof salary while the interacted one will identify any differential growthrate of salary by income in the period after section 162(m) took effect Wealso include a full set of year dummies the natural logarithm of thermrsquos market value and the rmrsquos annual rate of return over the twoprevious years as control variables To reduce the sensitivity of our re-sults to outliers we run quantile regressions and robust regressions

Table 6 contains our results The rst four columns are for all topexecutives while the last four columns focus on CEOs Columns (1) and(2) present estimates from robust regression for the percentage change insalary and the percentage change in total compensation (salary bonusstock grants and stock-option grants) In the salary growth regressionsfor all executives the coefcient on million is negative and signicantindicating that there is an underlying relationship between the level ofexecutive compensation and its subsequent growth rate The coefcienton million 3 after is negative as well suggesting that the negativerelationship between the level of salary and its subsequent growth rateintensied slightly after 1994 The relatively small magnitude suggeststhat an executive with a $1 million dollar salary would see his salarygrow at an annual rate that would be about 06 percent slower than anexecutive earning $500000

15 The HallndashLiebman data set contains only a combined salary and bonus variable makingit impossible to use it to analyze this issue16 We also ran the regressions (not reported) allowing for a nonlinear relationship betweenmillion and compensation growth and obtained substantively similar results

TA

BL

E6

The

Eff

ect

ofth

eM

illi

on-D

olla

rR

ule

onSa

lary

and

Tot

alC

ompe

nsat

ion

All

top

exec

uti

ves

CE

Os

only

Rob

ust

regr

essi

onQ

uan

tile

regr

essi

onR

obu

stre

gres

sion

Qu

anti

lere

gres

sion

D

in

Din

D

in

Din

D

in

Din

D

in

Din

sala

ryto

talc

omp

(a)

sala

ryto

talc

omp

sa

lary

tota

lcom

p

sala

ry

tota

lcom

p

mil

lio

n(b

)2

006

02

035

72

006

32

024

12

008

72

023

12

010

32

024

5(0

005

)(0

032

)(0

006

)(0

027

)(0

010

)(0

073

)(0

011

)(0

056

)m

illi

on

32

001

10

026

20

021

20

015

20

016

014

92

003

40

158

aft

er(c

)(0

005

)(0

033

)(0

006

)(0

029

)(0

010

)(0

076

)(0

011

)(0

057

)N

359

5035

950

359

5035

950

757

17

589

757

17

589

Stan

dar

der

rors

inp

aren

thes

esR

egre

ssio

nsal

soin

clu

de

year

du

mm

ies

stoc

km

arke

tre

turn

and

log

mar

ket

valu

e

(a)

Tot

alco

mp

ensa

tion

incl

ud

essa

lary

bon

us

rest

rict

edst

ock

and

stoc

k-op

tion

gran

ts

(b)

mil

lion

isd

ene

das

min

(1s

alar

yin

pre

viou

sye

ars

div

ided

by$1

mil

lion)

(c)

aft

eris

anin

dic

ator

vari

able

equ

alto

zero

befo

reth

ere

gula

tion

took

plac

e(1

993)

and

one

afte

rwar

d

36 Hall amp Liebman

In the second column the coefcient on million 3 after is essentiallyzero suggesting that any decrease in salary brought about by section162(m) was offset by increases in bonus and stock-option grants Theresults are similar for the median regressions There is some evidence ofa very small slowdown in growth of salary for executives likely to beaffected by section 162(m) but no sign of a change in total compensa-tion Taken together this provides evidence of a very small substitutionof performance-related pay for salary

The last four columns run identical tests but limit the sample to CEOsThe interaction of million 3 after continues to be small and negative inthe salary regressions In contrast the impact of 162(m) on total compen-sation appears to be positive and fairly substantial (about 15 percentagepoints per year) in these regressions That is the increase in bonus andstock options following the million-dollar rule more than offset the verysmall decline in salary

All of these conclusions should be interpreted somewhat cautiouslybecause our data provide us with only one pre-1993 control year Never-theless the data are consistent with section 162(m) having led to a veryminor slowdown in salary growth and one that is dramatically smallerthan what one would conclude without controlling for underlying differ-ential salary growth rates at different levels of salary Since total compen-sation for CEOs did not decline and perhaps increased there is evidenceof a minor substitution of performance-related pay for salary in responseto the regulation

8 EXECUTIVE COMPENSATION AND THEELASTICITY OF TAXABLE INCOME

Even if taxes have only a minor effect on executive compensation thestructure of executive compensation is important for tax policy be-cause high-income taxpayers are sources of signicant revenue and arethe focus of important debates about the efciency cost of taxation

Goolsbee (1999) has recently argued that executive decisions to exer-cise options are highly responsive to intertemporal differences in taxrates created by tax reforms He claims that much of the apparent declinein taxable income for high-income taxpayers between 1992 and 1993documented by Feldstein and Feenberg (1996) can be explained by execu-tives shifting option gains into 1992 in order to avoid the higher marginaltax rates that went into effect in 1993 By extension he implies that thehigh elasticities of taxable income with respect to personal tax ratesestimated by comparing the pre- and post-tax-reform taxable income of

Taxation of Executive Compensation 37

high-income individuals reect intertemporal shifting of income and notpermanent effects of taxation If this is true then the deadweight loss oftaxation is lower as is the cost of progressivity

While Goolsbee presents comprehensive evidence that option gainswere unusually high in 1992 his identication of the impact of tax rates onthe timing of option gains depends on correctly specifying a counter-factual level of options that would have been exercised in the absence ofthe tax change He simply assumes a linear time trend The purpose ofGoolsbeersquos time trend is presumably to account for the increasing use ofoptions over time However since options typically require a vestingperiod of at least a couple of years a much more direct measure of ex-pected option gains can be constructed based on the total value of vestedoptions held by the CEO Since options typically have vesting periods thatexceed two years this total can be treated as exogenous since it will not beaffected by the tax changes

More generally the value of options held by the CEO is likely to havea major effect on his exercise decisions If his options are out of themoney he will not have gains to realize If there has been a recent run-up in the market an optimizing CEO may exercise an unusually largeamount of options in order to diversify his overall portfolio and if in-the-money options are about to expire they will be exercised in thecurrent year

Our evidence suggests that much of the apparent tax shifting caninstead be attributed to stock market performance and the timing of pastoption grants Using our 15-year panel we replicate Goolsbeersquos (1999)result for the 1993 period but show that the impact of taxes is notpresent in other time periods Moreover even the result for the 1993period disappears when appropriate controls for past option grants andstock price appreciation are included

The top three rows of Table 7 present data on the taxable incomes ofCEOs in the HallndashLiebman sample For these results our panel has beennarrowed to include only CEOs whose rmrsquos scal years correspondwith tax years In addition we create a balanced panel for each of thethree tax-reform periods by excluding any rm for which there are notcomplete data for the period surrounding the reform

The fourth row of the table shows that the 1993 tax act raised the topmarginal personal income tax rate from 310 to 396 percent effective fortax year 1993 with a further increase occurring in 1994 due to the uncap-ping of the Medicare payroll tax The top row of the table shows thattaxable income was higher in 1992 than for any other year in our sample(all data are in 1994 dollars) Option gains were particularly high in1992mdashmore than double their 1991 level and 68 percent higher than their

TA

BL

E7

Tax

Ref

orm

Sto

ckM

arke

tP

erfo

rman

cea

ndth

eT

imin

gof

Opt

ion

Gai

ns

Ave

rage

stoc

km

arke

tM

ean

CE

OM

ean

CE

Ore

turn

for

Mea

nva

lue

taxa

ble

sala

ryan

dM

ean

CE

OT

opfe

der

al

rms

inof

opti

ons

(Val

ue

ofop

tion

Sam

ple

inco

me

bonu

sop

tion

gain

sp

erso

nal

sam

ple

for

held

atbe

gga

ins)

(va

lue

Ref

orm

size

Yea

r($

mil

)($

mil

)($

mil

)M

TR

pre

vye

ar(

)of

year

ofop

tion

she

ld)

ER

TA

8127

919

800

874

064

80

163

7025

80

460

354

1981

083

10

672

009

869

245

070

61

3919

820

872

067

90

099

5011

90

686

144

1983

099

00

724

025

250

355

116

62

16

TR

A86

295

1985

119

70

837

024

050

76

116

02

0719

861

448

091

70

345

5042

81

776

194

1987

165

50

959

059

338

519

12

229

266

1988

163

21

058

039

028

07

237

71

6419

891

656

105

20

414

2817

92

345

176

1993

272

1990

209

51

028

061

828

232

288

02

15T

axA

ct19

911

875

102

00

547

312

01

268

82

0319

922

678

107

01

250

3146

64

552

275

1993

232

41

209

074

339

620

64

299

173

1994

196

41

262

037

842

515

45

113

074

Sam

ple

isH

allndash

Lie

bman

rm

sw

ith

Dec

embe

r

scal

year

sth

atha

dco

mp

lete

dat

afo

ron

eof

the

thre

eta

x-re

form

per

iod

sst

ud

ied

Taxation of Executive Compensation 39

1994 level These patterns are consistent with Goolsbeersquos claim that tax-payers responded to the anticipated increase in marginal tax rates byshifting gains into 1992

The evidence from the other two tax reforms is less clear suggestingthat there might be factors other than tax rates determining the timing ofoption gains ERTA 1981 reduced the top marginal tax rate from 70percent in 1980 to 50 percent in 1982 However there is no sign thatoption gains were reduced in 1980 and 1981 and then increased in 198217

There was a big increase in option gains in 1983 however Similarly thehighest year of stock-option gains in the period around 1986 was 1987 astrange year to take gains since taxpayers would have known that thepersonal rate would be 10 percentage points lower if they waited untilthe following year18

What all the peak years of option gains have in common is that theyfollow years of strong stock market performance The fth row of Table 7shows the average stock market return during the previous year for therms in our sample The large value of option gains in 1983 followed a355-percent increase in the stock market The 1992 boom in option gainsfollowed a 466-percent increase in the stock market the more modest1987 peak in option gains followed a 19-percent gain (and a 43-percentgain two years before)

Strong stock market performance raises the value of options availableto exercise The sixth row of the table shows the mean value of optionsheld by CEOs at the beginning of the year All three of the peak years ofoption gains correspond with peak years in terms of the value of optionsavailable for exercise Since stock market appreciation during the yearalso increases the value of options available to exercise later in that yearwe need to allow for current-year stock market appreciation as well

Table 8 contains Goolsbee-style regressions for each of the tax re-forms The rst two columns reprint the results from Goolsbee (1999)The third and fourth columns replicate his results in our sample Inparticular with a specication similar to his we are able to replicate hisnding that the current after-tax share has a positive coefcient and thefollowing yearrsquos after-tax share has a coefcient of roughly negative oneColumns (5) through (8) show that this pattern was unique to the 1993tax reform Indeed in the TRA 86 regressions the coefcient on the

17 As we explained above for executives covered by the maximum tax on earned incomethere was not a decline in tax rates between 1981 and 198218 It is possible that the large stock market decline in October 1997 may have promptedsome people to exercise options earlier than they otherwise would have However anincrease in option exercises typically follows stock market increases not declines

TA

BL

E8

Att

empt

sto

Rep

lica

teth

eT

ax-S

hift

ing

Res

ult

inO

ther

Tim

eP

erio

ds

Goo

lsbe

eG

ools

bee

1993

1993

Ent

ire

Ent

ire

regr

essi

onre

gres

sion

refo

rmre

form

TR

A86

TR

A86

ER

TA

81E

RT

A81

1980

ndash199

419

80ndash1

994

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

ln[1

2ta

x(t)

]1

244

108

20

455

032

62

017

62

017

22

016

72

016

70

026

20

068

(01

34)

(01

27)

(05

37)

(05

42)

(02

10)

(02

09)

(00

90)

(00

91)

(00

67)

(00

65)

ln[1

2ta

x(t1

1)]

20

867

20

764

21

214

20

917

047

80

722

20

103

004

30

176

018

3(0

118

)(0

112

)(0

491

)(0

499

)(0

157

)(0

174

)(0

072

)(0

072

)(0

068

)(0

066

)

Tim

e0

135

007

62

000

52

036

40

041

000

80

094

003

50

056

002

7(0

008

)(0

008

)(0

049

)(0

049

)(0

028

)(0

028

)(0

025

)(0

026

)(0

003

)(0

003

)

ln(m

arke

tva

lue)

066

10

489

023

70

334

039

5(0

015

)(0

058

)(0

039

)(0

042

)(0

017

)

Yea

rs19

91ndash1

995

1991

ndash199

519

90ndash1

994

1990

ndash199

419

85ndash1

989

1985

ndash198

919

80ndash1

983

1980

ndash198

319

80ndash1

994

1980

ndash199

4

Ad

just

edR

20

720

770

610

630

690

760

800

810

630

68

Obs

erva

tion

s13

856

132

371

385

136

41

470

143

31

113

106

23

968

385

9

Dep

end

ent

vari

able

isth

elo

gof

taxa

ble

inco

me

All

regr

essi

ons

con

tain

rm

xe

def

fect

sG

ools

bee

regr

essi

on(2

)al

soco

ntai

nstw

oot

her

vari

able

sno

tsh

own

inth

eta

ble

Stan

dar

der

rors

inp

aren

thes

es

Taxation of Executive Compensation 41

TABLE 9Alternative Explanations for the Timing of Option Exercises

1993 Entirereform TRA 86 ERTA 81 1980ndash1994

(1) (2) (3) (4)

ln[12tax(t)] 0108 0111 20051 0201(0324) (0136) (0048) (0057)

ln[12tax(t11)] 20134 0573 0125 0150(0336) (0156) (0053) (0066)

ln(market value) 0418 0177 0338 0418(0061) (0039) (0046) (0163)

Firmrsquos stock market return (t) 0298 0268 0214 0258(0044) (0045) (0031) (0023)

Firmrsquos stock market return 0162 0206 0031 0128(t21) (0041) (0008) (0029) (0021)

ln(value of options held by 0004 0008 0001 0009CEO at start of year) (0003) (0003) (0002) (0001)

Years 1990ndash1994 1985ndash1989 1980ndash1983 1980ndash1994Adjusted R2 064 071 082 068Observations 1349 1427 1036 3812

The dependent variable is the log of taxable income All regressions contain rm xed effects Stan-dard errors in parentheses

following yearrsquos tax rate is positive and signicant Columns (9) and (10)pool all three reforms and fail to replicate the pattern that would reectintertemporal shifting19

It is possible that the ease with which one can shift the timing ofoption gains is asymmetric It may be harder to shift option gains for-ward in time since some options expire and therefore must be exercisednow In addition options were a much smaller portion of compensationin the early 1980s so perhaps the ERTA 81 results are not surprisingNonetheless these results cast doubt on the proposition that this is thechannel through which spurious estimates of the elasticity of taxableincome have occurred for previous tax reforms and they suggest thatwe should look harder for alternative explanations for the timing ofoption gains

Table 9 attempts to predict options gains using the lagged and currentincrease in a rmrsquos market value and the value of the options held by theCEO at the start of the year The two stock market variables stronglypredict option gains while the value of options held has the correct sign

19 The results also fail to suggest large permanent effects of marginal tax rates on taxableincome

42 Hall amp Liebman

in all of the specications and is signicant in the pooled sample eventhough it is highly correlated with the two stock market return variablesIn contrast the tax-timing explanation fails even in the 1993 period

It is important to realize that our results do not invalidate Goolsbeersquoscritique of claims that aggregate revenue data from 1992 and 1993 provethat elasticities of taxable income are large We have simply given analternative explanation for why taxable incomes were unusually high in1992 However our results do cast doubt on the conclusion that largetiming shifts are ubiquitous and capable of explaining away the morecareful estimates of taxable income elasticities that have been performedusing micro data (eg Feldstein 1995 Auten and Carroll 1995)

One further point is in order It is sometimes argued that the combina-tion of increased use of options and the booming stock market explainswhy federal tax revenues have been so much higher than expected dur-ing the 1990s However while large option gains increase revenue fromthe personal income tax they produce offsetting deductions for cor-porations The net impact on revenue is only the difference in the twomarginal tax rates times the option gain While the gap between thepersonal rate and the effective corporate rate is nontrivial the impact oflarge option exercises on total revenue is substantially smaller than theamount by which the exercises inate personal tax revenue and reducecorporate tax revenue

9 CONCLUSION

We have described the tax rules for executive pay in detail and analyzedhow changes in various tax rates affect the tax advantages of stock optionsrelative to salary and bonus We nd that there is a moderate tax advan-tage to optionsmdashcurrently about $4 per $100 in pretax executive paymdashand that changes in the tax advantage of options over time have had atmost a modest impact on the composition of pay Corporate governancefactors particularly the role of large investors appear to be more impor-tant in explaining the dramatic increase in option pay

In addition our evidence suggests that more direct attempts to use taxpolicy to inuence executive compensation have had little effect Wend that the million-dollar rule led companies to substitute perfor-mance-based pay for salary But our evidence suggests that this substitu-tion was quite modest and there is no evidence that the total level of paywas reduced Overall although the stock-option explosion has dramati-cally increased the link between pay and performance this change isdue almost entirely to nontax factors

Taxation of Executive Compensation 43

REFERENCES

Auerbach Alan J (1996) ldquoThe Dynamic Effects of Tax Law Asymmetriesrdquo Re-view of Economic Studies 53451ndash405

mdashmdashmdash and James Poterba (1987) ldquoTax Loss Carryforwards and Corporate TaxIncentivesrdquo In The Effects of Taxation on Capital Accumulation M Feldstein(ed) Chicago University of Chicago Press

Altshuler Rosanne and Alan Auerbach (1990) ldquoThe Signicance of Tax LawAsymmetries An Empirical Investigationrdquo Quarterly Journal of Economics 10561ndash86

Auten Gerald and Robert Carroll (1995) ldquoThe Effect of Income Taxes on House-hold Behaviorrdquo Ofce of Tax Analysis US Department of the Treasury

Baker George P and Brian J Hall (1998) ldquoCEO Incentives and Firm SizerdquoNBER Working Paper no 6868 December

Core John and Wayne Guay (1999) ldquoThe Use of Equity Grants to ManageOptimal Equity Incentive Levelsrdquo Draft

Feldstein M (1995) ldquoThe Effect of Marginal Tax Rates on Taxable Income APanel Study of the 1986 Tax Reform Actrdquo Journal of Political Economy 103 (no3)551ndash72

mdashmdashmdash and D Feenberg (1996) ldquoThe Effect of Increased Tax Rates on TaxableIncome and Economic Efciency A Preliminary Analysis of the 1993 Tax RateIncreasesrdquo In Tax Policy and the Economy vol 10 James M Poterba (ed)Cambridge MA MIT Press

Gompers Paul A and Andrew Metrick (1998) ldquoInstitutional Investors andEquity Pricesrdquo NBER Working Paper no 6723

Goolsbee Austan (1999) ldquoWhat Happens When You Tax the Rich Evidencefrom Executive Compensationrdquo NBER Working Paper no 6333

Graham John R (1996) ldquoProxies for the Corporate Marginal Tax Raterdquo Journal ofFinancial Economics 42187ndash221

Hall Brian J (1999) ldquoThe Design of Multi-year Stock Option Plansrdquo Journal ofApplied Corporate Finance 12(2)97ndash106

mdashmdashmdash and Jeffrey B Liebman (1998) ldquoAre CEOs Really Paid Like Bureau-cratsrdquo Quarterly Journal of Economics CXII(no 3)653ndash691

Jensen Michael C and William M Meckling (1976) ldquoTheory of the Firm Mana-gerial Behavior Agency Costs and Ownership Structurerdquo Journal of FinancialEconomics III305ndash360

mdashmdashmdash and Kevin J Murphy (1990) ldquoPerformance Pay and Top-ManagementIncentivesrdquo Journal of Political Economy XCVIII225ndash264

Kaplan Steven N (1997) ldquoThe Evolution of US Corporate Governance WeAre All Henry Kravis Nowrdquo Draft

Livingstone Jane R (1999) ldquoExecutive Compensation Contracting and Re-sponses to Increased Tax Costsrdquo Louisiana State University June

Miller Merton H and Myron S Scholes (1982) ldquoExecutive CompensationTaxes and Incentivesrdquo In Financial Economics Essays in Honor of Paul CootnerWilliam F Sharpe and Cathryn M Cootner (eds) Princeton NJ Prentice-Hall

Murphy Kevin J (1999) ldquoExecutive Compensationrdquo forthcoming in OrleyAshenfelter and David Card (eds) Handbook of Labor Economics Vol 3 NorthHolland

Perry Todd and Marc Zenner (1999) ldquoPay for Performance Government Regu-

Taxation of Executive Compensation 25

In addition we include two other variables that proxy for strongercorporate governance The rst is the size of the board Evidence suggeststhat boards with smaller numbers of directors reduce the inuence of theCEO and better represent shareholder interests (Yermack 1995) Boardsize fell by approximately 15 percent between 1980 and 1994 The secondvariable is the fraction of the rmrsquos board members who are inside direc-tors or gray directors (noninsiders who have business dealings with thecompany) Firms with fewer outsiders are less likely to act in the share-holdersrsquo interests and we would predict that they would be less likely tohave performance-based compensation for their CEOs The average per-centage of inside and gray directors has decreased from about 45 percentin 1980 to about 35 percent in 1994

Second we allow for cross-sectional variation in corporate tax ratesCorporate marginal tax rates vary because rms may be eligible for tax-loss carrybacks and carryforwards investment tax credits and the alter-native minimum tax (see Auerbach 1986 Auerbach and Altshuler1990 Auerbach and Poterba 1987) Because these cross-sectional differ-ences imply that different rms should respond differently to a givenchange in tax rates they potentially provide us with a way to isolatethe tax effects Firm-specic corporate marginal tax rates are notori-ously difcult to calculate from publicly-available data We use a tricho-tomous variable that equals zero if the rm has tax-loss carryforwardsand negative earnings in a given year equals 05 times the statutorycorporate rate if the rm has only one of those conditions and equalsthe statutory rate if the rm has neither loss carryforwards nor negativeearnings10

The cross-sectional variation in the corporate tax rate may not be ex-ogenous Firms that perform poorly could face low marginal tax ratesand also be particularly likely or unlikely to provide performance-basedpay Because most of our specications use xed effects the correlationwe must be concerned about is between changes in rm performanceand changes in compensation We deal with this by controlling forlagged rm stock market performance in our regressions This shouldeliminate spurious correlation between tax rates and compensation thatis jointly caused by rm performance

10 Graham (1996) presents evidence that the trichotomous variable performs nearly as wellas his recommended simulated marginal tax rate Plesko (1999) compares marginal taxrates calculated from Compustat data with actual marginal tax rates from rm tax returnsHe nds that simple binary variables are more reliable measures of the marginal tax ratethan is either the Graham or the trichotomous variable We obtain substantively similarresults when we switch from the trichotomous variable to the binary variables recommendby Plesko

26 Hall amp Liebman

63 ResultsThe dependent variable in our regressions is the fraction of a CEOrsquos totalannual compensation (measured as the BlackndashScholes value of stock-option grants plus salary and bonus) that comes from stock-optiongrants We choose this reduced-form specication because our interest isin the composition of CEO compensation However the factors thataffect the composition of compensation might also affect the levels andit will be important to keep this in mind in interpreting our results

Optimal-contracting theory generally makes predictions about the in-centives provided by the total amount of rm stock and stock optionsowned by the CEO We choose to model the ow of option grants ratherthan the ultimate performance sensitivity of the CEOrsquos compensationand stock and stock-option holdings because we have in mind anadjustment-cost model in which it takes time for a CEOrsquos holdings ofstock and stock options to reach the optimal level11 In our regressionswe include two measures of the current performance sensitivity of theCEOrsquos wealth to rm performance as explanatory variables to reect thedistance of the CEOrsquos contract from the optimal contract

Table 3 presents results in which the tax effects are identied solely byvariation in tax rates over time The fraction of annual compensation paidin stock options is regressed on tax variables corporate governance vari-ables contract-theory variables and variables reecting stock market re-turns Both of the regressions in this table include rm xed effects Incolumn (1) the fraction in annual compensation paid in stock options isregressed on the log difference in payoff (the combined payoff to both therm and the CEO) from receiving compensation in stock options ratherthan salary and bonus where the payoff calculations are based on statu-tory marginal tax rates This variable is dened as the log difference be-tween equation (3) and equation (6) in section 4 The coefcient on the taxvariable is large and statistically signicant at the 95-percent level Thepoint estimate implies that a one-percent increase in the payoff differencebetween stock options and salary and bonus results in a 24-percentage-point increase in the share of compensation paid in stock options

The corporate governance variables all have the predicted sign Firmswith a higher fraction of their shares owned by institutional investorsuse more performance-based pay (stock options) Firms with large corpo-rate boards or a large share of inside directors are less likely to use stockoptions

The two contract-theory variables are the dollar change in CEO wealth

11 See Core and Guay (1999) for evidence that rms adjust in this way

TABLE 3Compensation Regressions that Rely on Variation in

Tax Rates over Time

(1) (2)

Tax variablesln (difference in payoff)(a) 2393 (0349)ln (12statutory personal 20015 (0034)

rate)ln (12statutory cap gains 0330 (0171)

rate)ln (12statutory corporate 0399 (0148)

rate)

Corporate governance variablesFraction of rmrsquos shares 0043 (029) 0052 (0028)

owned by institutionalinvestors

Size of the rmrsquos board of 20002 (001) 20003 (0001)directors

Fraction of directors who 20082 (033) 20092 (0030)are inside or gray

Contract-theory variablesD (CEO wealth) per $1000 20261 (0095) 20247 (0087)

in FV(b)

D (CEO wealth) per 10 in 2583 (189) 2554 (201)FV(c)

Volatility(d) 521 (424) 261 (369)ln (rm market value) 0019 (0006) 0019 (0006)

Stock market return variablesLagged annual rm stock 0006 (0009) 0003 (0008)

market returnLagged annual SampP 500 re- 20035 (0023) 20029 (0023)

turn

Regression diagnosticsTotal sample size 5179 5179Firms 427 427Adjusted R2 0313 0311

Dependent variable share of compensation paid in options Standard errors in parentheses Bothregressions in this table include rm xed effects

(a) Log difference in payoff is the log difference in combined rm and CEO after-tax share betweenpaying compensation in options and paying salary

(b) This variable is the dollar change in CEO wealth per $1000 change in rm market value (seeJensen and Murphy 1990)

(c) This variable in the dollar charge in CEO wealth per 10-percent change in rm value (see Hall andLiebman 1998)

(d) Standard deviation of rm market value in trillions

28 Hall amp Liebman

per $1000 change in rm value (Jensen and Murphy 1990) and thedollar change in CEO wealth per 10-percent change in rm value (Halland Liebman 1998) As discussed in Baker and Hall (1998) these twowealth sensitivity measures reect two different concepts of how closelyaligned the managersrsquo incentives are with the interests of the sharehold-ers In the regression they are meant to measure how far the CEOrsquosexisting contract and ownership of stock and stock options is from theoptimal level The negative coefcients on the two variables suggeststhat rms that are below their optimal pay-to-performance sensitivityare more likely to give more stock options in the current period

The second column replaces the tax variable motivated by deferraladvantages with the three statutory rates The corporate after-tax sharehas a positive coefcient as we would expect because the tax advantageof stock options increases as the corporate rate falls The coefcient onthe personal tax rate is small and statistically insignicant The sign ofthe coefcient on the capital gains rate is the opposite of that predictedby tax-avoidance theory

Note that these results rely on time-series variation alone This is not avery convincing test Since the log difference in payoffs mostly rises overthe time period covered in our sample as does the share of compensa-tion paid in options it is possible that the tax variable is simply reectingother factors that were changing over this time period that we have notincluded in our regression Therefore it is important to look at specica-tions that rely on cross-sectional variation in the tax variables

Table 4 contains specications that rely on cross-sectional variation incorporate tax rates The rst column repeats the specication from col-umn (1) of Table 3 but uses the trichotomous measures of rm corporatetax rates in place of the statutory rate in calculating each rmrsquos logdifference in payoff from paying compensation in options rather than insalary and bonus As in the previous table we nd the positive coef-cient on the log difference in payoff just as the theory suggest How-ever the coefcient estimate is now much smaller suggesting that aone-percent increase in the log difference in payoff produces a 01-percentage-point increase in the share of compensation paid in options

Introducing cross-sectional variation in the dependent variable doesnot eliminate the potentially spurious correlation over time between thetax variable and the share of compensation paid in options The secondcolumn of Table 4 isolates the pooled cross-sectional variation in corpo-rate tax rates by adding time dummies and dropping the rm xedeffects In this specication the coefcient on the tax variable is indistin-guishable from zero The coefcient on the institutional variable contin-ues to be large and statistically signicant The third column includes

Taxation of Executive Compensation 29

both rm xed effects and time dummies which tests for a relationshipbetween within-rm changes in the dependent variable and within-rmchanges in each explanatory variable In this specication the coefcienton the tax variable is again indistinguishable from zero The coefcientson the corporate governance variables are also statistically insignicant

The third column contains a specication that treats the Tax ReformAct of 1986 (TRA 86) as a natural experiment For rms and CEOs facingthe statutory corporate tax rate TRA 86 increased the relative payoff topaying compensation in options by 17 percent For a rm facing a zeromarginal tax rate it reduced the relative payoff by 45 percent The reduc-tion occurred because TRA 86 reduced the personal tax rate makingsalary and bonus more attractive while the decreased corporate rate hadno effect on zero-tax-rate rms Thus we would expect rms facing zeroor low marginal tax rates to reduce their use of options after TRA 86while rms facing the statutory rate would be expected to increase themFew rms face zero marginal tax rates year after year Therefore we tryto distinguish between rms that often face low tax rates and rms thatnearly always face the statutory rate We dene low-tax rms as oneswhose average marginal tax rate over 1984 to 1986 was below 02412

Then we ran a difference-in-differences regression to see if high-taxrms increased their use of options more than low-tax rms The interac-tion between post-TRA 86 and high corporate tax rates is the key vari-able The estimated coefcient on this variable is positive but small andnot statistically different from zero at the 95-percent level

It is important to note that in this specication as well the main corpo-rate governance variable the share of the rmrsquos shares owned by institu-tional investors performs as theory would predict The variable appearsto be quite robust Table 5 presents regressions of the share of compensa-tion paid in stock options on the fraction of the rmrsquos shares owned byinstitutional investors Since we have already established that these vari-ables have a strong longitudinal relationship the regressions here aredesigned to focus on the cross-sectional relationship between these twovariables The rst and third columns present regressions using data onlyfrom 1994 and nd a coefcient of roughly 02 (column 1 contains noother covariates while column 3 includes all of the other nontax covari-ates from the regressions in Tables 3 and 4) Since the median share of armrsquos stock owned by institutional investors in our sample rises from 19percent in 1980 to 49 percent in 1994 this coefcient would imply that this

12 This cutoff was chosen after inspecting the distribution of tax rates and essentiallyseparates rms that are always at or near the statutory rate from ones that are sometimesbelow it

TA

BL

E4

Com

pens

atio

nR

egre

ssio

nsth

atR

ely

onC

ross

-Sec

tion

alV

aria

tion

inC

orpo

rate

Tax

Rat

es

(1)

(2)

(3)

(4)

Tax

vari

able

s(a)

ln

(dif

fere

nce

inp

ayof

f)0

099

(00

41)

20

038

(00

36)

20

005

(00

44)

Pos

t-T

RA

862

000

5(0

013

)H

igh

corp

orat

eta

xra

te0

018

(00

17)

(Pos

t-T

RA

86)

3hi

gh0

024

(00

16)

corp

orat

eta

xra

te

Cor

por

ate

gove

rnan

ceva

riab

les

Frac

tion

of

rmrsquos

shar

es0

109

(00

27)

011

4(0

015

)2

000

9(0

030

)0

087

(00

34)

owne

dby

inst

itu

tion

alin

vest

ors

Size

ofth

e

rmrsquos

boar

dof

20

004

(00

01)

20

001

(00

01)

20

002

(00

01)

000

1(0

002

)d

irec

tors

Frac

tion

ofd

irec

tors

wh

o2

012

4(0

033

)2

000

3(0

018

)2

004

1(0

034

)0

001

(00

39)

are

insi

de

orgr

ay

Con

trac

t-th

eory

vari

able

sD

(CE

Ow

ealt

h)p

er$1

000

20

238

(00

94)

20

167

(00

53)

20

257

(00

96)

20

021

(01

47)

inFV

(b)

D(C

EO

wea

lth)

per

10

in2

560

(17

8)2

592

(10

5)2

594

(19

6)2

147

(04

86)

FV(c

)

Vol

atili

ty(d

)2

48(4

24)

162

(03

4)1

12(4

28)

106

(06

09)

ln(

rmm

arke

tva

lue)

003

3(

006)

000

8(0

003

)0

009

(00

06)

001

2(0

007

)

Stoc

km

arke

tre

turn

vari

able

sL

agge

dan

nual

rm

stoc

k2

000

2(0

009

)0

011

(00

10)

000

6(0

009

)2

001

2(0

013

)m

arke

tre

turn

Lag

ged

annu

alSamp

P50

02

005

6(0

023

)2

005

3(0

051

)2

003

9(0

049

)2

015

4(0

040

)re

turn

Reg

ress

ion

dia

gnos

tics

T

otal

sam

ple

size

5179

5179

5179

2534

Firm

s42

742

739

6A

dju

sted

R2

031

009

032

005

Firm

xed

effe

cts

Yes

No

Yes

No

Tim

ed

um

mie

sN

oY

esY

esN

o

Dep

end

ent

vari

able

isfr

acti

onof

ann

ualc

omp

ensa

tion

from

stoc

kop

tion

sSt

and

ard

erro

rsin

par

enth

eses

(a)

All

thre

esp

eci

cati

ons

use

tric

hoto

mou

sco

rpor

ate

tax

rate

sT

he

sam

ple

inco

lum

n(3

)u

ses

only

the

year

s19

84to

1988

sin

cew

ear

eu

sing

1986

asa

nat

ural

exp

erim

ent

(b)

Thi

sva

riab

leis

the

dol

lar

chan

gein

CE

Ow

ealt

hp

er$1

000

chan

gein

rm

mar

ketv

alu

e(s

eeJe

nse

nan

dM

urp

hy

1990

)

(c)

Thi

sva

riab

lein

the

dol

lar

char

gein

CE

Ow

ealt

hp

er10

-per

cent

chan

gein

rm

valu

e(s

eeH

alla

nd

Lie

bman

199

8)

(d)

Stan

dar

dd

evia

tion

ofr

mm

arke

tva

lue

intr

illio

ns

32 Hall amp Liebman

TABLE 5Large Institutional Investors and Stock-Option Compensation

The Cross-sectional Relationship between Institutional Investors andOption Grants

(1) (2) (3) (4)

Coefcient on fraction of 217 (067) 145 (014) 177 (069) 120 (015)rmrsquos shares owned byinstitutional investors

Sample period 1994 1980ndash1994 1994 1980ndash1994Contract-theory and stock No No Yes Yes

market covariatesTime dummies No Yes No Yes

The dependent variable is the share of compensation paid in options

factor alone can explain 6 percentage points of the 23-percentage-pointincrease in the median share of compensation paid in options that oc-curred over this time period Of course this variable is only one measureof the strengthening of corporate governance so the total impact of im-proved corporate governance could be greater The second and fourthcolumns of the table use the entire 1980ndash1994 sample and include timedummies to isolate the cross-sectional correlations Again there is evi-dence of a strong cross-sectional relationship between the two variablesalthough the coefcients are slightly smaller perhaps reecting a weakercorporate-governance role in the earlier period In sum there is a strongrelationship in both longitudinal and cross-sectional data between theinstitutional-investor variable and the options-percentage variable13

7 THE MILLION-DOLLAR RULE

In 1993 section 162(m) of the Internal Revenue Code was enactedlimiting the deductibility of executive compensation in excess of onemillion dollars unless the compensation was performance-related Pro-ponents of this legislation argued that it would slow the growth ofexecutive pay and tighten the link between rm performance and execu-tive compensation

The new provision became effective for tax year 1994 and applies tothe CEO and the other four most highly compensated executives in eachcompany Pay that is performance-related such as stock-option grants

13 The one specication in which the fraction owned by institutional investors is notsignicant is in column (2) of Table 4 a regression that includes both xed effects and timeeffects

Taxation of Executive Compensation 33

and bonuses paid for meeting clear performance goals is not affected bythis regulation provided that it has been approved by shareholdersThus the main impact of the provision is to limit the deductibility ofsalary in excess of a million dollars Woodlock and Antenucci (1997)studied the proxy statements of 376 rms and documented that mostrms responded to the new law by qualifying top executive pay asperformance-related Perry and Zenner (1999) identied 25 rms thatreduced salaries from above $1 million to below $1 million and foundthat 23 of the rms cited section 162(m) as their reason for doing so

In this section of the paper we investigate whether this provision hasaffected the level or structure of executive pay using the Execucompdata base which contains information on the highest-paid executives ineach of the rms in the SampP 500 SampP Midcap 400 and SampP SmallCap600 Between 1993 and 1998 the median salary of the Execucomp execu-tives rose from $230000 to $306000 The median value of total compen-sation (salary bonus and option grants) rose from $470000 to $882000Limiting the sample of CEOs indicates that the total compensation of themedian CEO rose from $11 million to $18 million over the period whilethe median salary rose from $441000 to $550000 Thus executive com-pensation clearly continued to rise at a rapid rate after the implementa-tion of the million-dollar rule though perhaps at a slower rate than itotherwise would have

If the regulation affects behavior it should have the largest impact onrms whose executives are receiving salaries that are above the limit orjust below it Thus we might expect to see rms whose executives arereceiving relatively high salaries to rely more heavily on performance-based pay and less heavily on salary in the years following the regulationPerry and Zenner (1999) and Livingston (1999) both nd that this is thecase This trend can be clearly identied in the Execucomp data set Forexample over 1993 to 1998 the median annualized growth rate of salaryfor executives whose 1993 salary was below $500000 in the previous yearwas 85 percent For executives with salary between $500000 and$700000 it was 64 percent for executives between $700000 and $850000it was 50 percent for executives between $750000 and $1 million it was31 percent and for executives above $1 million it was 00 percent14

The observation that executives at lower levels of salary had highergrowth rates is not very strong evidence of an impact of section 162(m)however because it is possible that low-salary executives would havehad higher growth rates even in the absence of the provision To explore

14 Of the 3137 executives for whom we have complete data between 1993 and 1998 2708had 1993 salaries below $500000 while only 35 had 1993 salaries above $1 million

34 Hall amp Liebman

whether this is the case we use Execucomp data from both before andafter the law change Ideally we would want to have data from manyyears before the change to examine whether there is an underlying pat-tern of faster salary growth for low-salary executives Unfortunately theExecucomp data begin in 1992 and therefore provide us with only the1992ndash1993 growth rate in compensation to use as a baseline15

We run regressions pooling data from 1992ndash1993 through 1997ndash1998of the annual percentage change in different forms of compensation on avariable million that is designed to measure the likelihood that anexecutiversquos compensation will be affected by section 162(m) We denemillion as the minimum of 1 and the executiversquos salary divided by $1million in the previous year Thus any executive with salary at or above$1 million would receive a value of 1 while an executive with a salary of$500000 would be coded as a 0516 We include this variable by itself andalso interacted with a dummy variable (after) that equals 1 for years1994 and after Thus the million variable is designed to pick up anyunderlying relationship between the level of salary and the growth rateof salary while the interacted one will identify any differential growthrate of salary by income in the period after section 162(m) took effect Wealso include a full set of year dummies the natural logarithm of thermrsquos market value and the rmrsquos annual rate of return over the twoprevious years as control variables To reduce the sensitivity of our re-sults to outliers we run quantile regressions and robust regressions

Table 6 contains our results The rst four columns are for all topexecutives while the last four columns focus on CEOs Columns (1) and(2) present estimates from robust regression for the percentage change insalary and the percentage change in total compensation (salary bonusstock grants and stock-option grants) In the salary growth regressionsfor all executives the coefcient on million is negative and signicantindicating that there is an underlying relationship between the level ofexecutive compensation and its subsequent growth rate The coefcienton million 3 after is negative as well suggesting that the negativerelationship between the level of salary and its subsequent growth rateintensied slightly after 1994 The relatively small magnitude suggeststhat an executive with a $1 million dollar salary would see his salarygrow at an annual rate that would be about 06 percent slower than anexecutive earning $500000

15 The HallndashLiebman data set contains only a combined salary and bonus variable makingit impossible to use it to analyze this issue16 We also ran the regressions (not reported) allowing for a nonlinear relationship betweenmillion and compensation growth and obtained substantively similar results

TA

BL

E6

The

Eff

ect

ofth

eM

illi

on-D

olla

rR

ule

onSa

lary

and

Tot

alC

ompe

nsat

ion

All

top

exec

uti

ves

CE

Os

only

Rob

ust

regr

essi

onQ

uan

tile

regr

essi

onR

obu

stre

gres

sion

Qu

anti

lere

gres

sion

D

in

Din

D

in

Din

D

in

Din

D

in

Din

sala

ryto

talc

omp

(a)

sala

ryto

talc

omp

sa

lary

tota

lcom

p

sala

ry

tota

lcom

p

mil

lio

n(b

)2

006

02

035

72

006

32

024

12

008

72

023

12

010

32

024

5(0

005

)(0

032

)(0

006

)(0

027

)(0

010

)(0

073

)(0

011

)(0

056

)m

illi

on

32

001

10

026

20

021

20

015

20

016

014

92

003

40

158

aft

er(c

)(0

005

)(0

033

)(0

006

)(0

029

)(0

010

)(0

076

)(0

011

)(0

057

)N

359

5035

950

359

5035

950

757

17

589

757

17

589

Stan

dar

der

rors

inp

aren

thes

esR

egre

ssio

nsal

soin

clu

de

year

du

mm

ies

stoc

km

arke

tre

turn

and

log

mar

ket

valu

e

(a)

Tot

alco

mp

ensa

tion

incl

ud

essa

lary

bon

us

rest

rict

edst

ock

and

stoc

k-op

tion

gran

ts

(b)

mil

lion

isd

ene

das

min

(1s

alar

yin

pre

viou

sye

ars

div

ided

by$1

mil

lion)

(c)

aft

eris

anin

dic

ator

vari

able

equ

alto

zero

befo

reth

ere

gula

tion

took

plac

e(1

993)

and

one

afte

rwar

d

36 Hall amp Liebman

In the second column the coefcient on million 3 after is essentiallyzero suggesting that any decrease in salary brought about by section162(m) was offset by increases in bonus and stock-option grants Theresults are similar for the median regressions There is some evidence ofa very small slowdown in growth of salary for executives likely to beaffected by section 162(m) but no sign of a change in total compensa-tion Taken together this provides evidence of a very small substitutionof performance-related pay for salary

The last four columns run identical tests but limit the sample to CEOsThe interaction of million 3 after continues to be small and negative inthe salary regressions In contrast the impact of 162(m) on total compen-sation appears to be positive and fairly substantial (about 15 percentagepoints per year) in these regressions That is the increase in bonus andstock options following the million-dollar rule more than offset the verysmall decline in salary

All of these conclusions should be interpreted somewhat cautiouslybecause our data provide us with only one pre-1993 control year Never-theless the data are consistent with section 162(m) having led to a veryminor slowdown in salary growth and one that is dramatically smallerthan what one would conclude without controlling for underlying differ-ential salary growth rates at different levels of salary Since total compen-sation for CEOs did not decline and perhaps increased there is evidenceof a minor substitution of performance-related pay for salary in responseto the regulation

8 EXECUTIVE COMPENSATION AND THEELASTICITY OF TAXABLE INCOME

Even if taxes have only a minor effect on executive compensation thestructure of executive compensation is important for tax policy be-cause high-income taxpayers are sources of signicant revenue and arethe focus of important debates about the efciency cost of taxation

Goolsbee (1999) has recently argued that executive decisions to exer-cise options are highly responsive to intertemporal differences in taxrates created by tax reforms He claims that much of the apparent declinein taxable income for high-income taxpayers between 1992 and 1993documented by Feldstein and Feenberg (1996) can be explained by execu-tives shifting option gains into 1992 in order to avoid the higher marginaltax rates that went into effect in 1993 By extension he implies that thehigh elasticities of taxable income with respect to personal tax ratesestimated by comparing the pre- and post-tax-reform taxable income of

Taxation of Executive Compensation 37

high-income individuals reect intertemporal shifting of income and notpermanent effects of taxation If this is true then the deadweight loss oftaxation is lower as is the cost of progressivity

While Goolsbee presents comprehensive evidence that option gainswere unusually high in 1992 his identication of the impact of tax rates onthe timing of option gains depends on correctly specifying a counter-factual level of options that would have been exercised in the absence ofthe tax change He simply assumes a linear time trend The purpose ofGoolsbeersquos time trend is presumably to account for the increasing use ofoptions over time However since options typically require a vestingperiod of at least a couple of years a much more direct measure of ex-pected option gains can be constructed based on the total value of vestedoptions held by the CEO Since options typically have vesting periods thatexceed two years this total can be treated as exogenous since it will not beaffected by the tax changes

More generally the value of options held by the CEO is likely to havea major effect on his exercise decisions If his options are out of themoney he will not have gains to realize If there has been a recent run-up in the market an optimizing CEO may exercise an unusually largeamount of options in order to diversify his overall portfolio and if in-the-money options are about to expire they will be exercised in thecurrent year

Our evidence suggests that much of the apparent tax shifting caninstead be attributed to stock market performance and the timing of pastoption grants Using our 15-year panel we replicate Goolsbeersquos (1999)result for the 1993 period but show that the impact of taxes is notpresent in other time periods Moreover even the result for the 1993period disappears when appropriate controls for past option grants andstock price appreciation are included

The top three rows of Table 7 present data on the taxable incomes ofCEOs in the HallndashLiebman sample For these results our panel has beennarrowed to include only CEOs whose rmrsquos scal years correspondwith tax years In addition we create a balanced panel for each of thethree tax-reform periods by excluding any rm for which there are notcomplete data for the period surrounding the reform

The fourth row of the table shows that the 1993 tax act raised the topmarginal personal income tax rate from 310 to 396 percent effective fortax year 1993 with a further increase occurring in 1994 due to the uncap-ping of the Medicare payroll tax The top row of the table shows thattaxable income was higher in 1992 than for any other year in our sample(all data are in 1994 dollars) Option gains were particularly high in1992mdashmore than double their 1991 level and 68 percent higher than their

TA

BL

E7

Tax

Ref

orm

Sto

ckM

arke

tP

erfo

rman

cea

ndth

eT

imin

gof

Opt

ion

Gai

ns

Ave

rage

stoc

km

arke

tM

ean

CE

OM

ean

CE

Ore

turn

for

Mea

nva

lue

taxa

ble

sala

ryan

dM

ean

CE

OT

opfe

der

al

rms

inof

opti

ons

(Val

ue

ofop

tion

Sam

ple

inco

me

bonu

sop

tion

gain

sp

erso

nal

sam

ple

for

held

atbe

gga

ins)

(va

lue

Ref

orm

size

Yea

r($

mil

)($

mil

)($

mil

)M

TR

pre

vye

ar(

)of

year

ofop

tion

she

ld)

ER

TA

8127

919

800

874

064

80

163

7025

80

460

354

1981

083

10

672

009

869

245

070

61

3919

820

872

067

90

099

5011

90

686

144

1983

099

00

724

025

250

355

116

62

16

TR

A86

295

1985

119

70

837

024

050

76

116

02

0719

861

448

091

70

345

5042

81

776

194

1987

165

50

959

059

338

519

12

229

266

1988

163

21

058

039

028

07

237

71

6419

891

656

105

20

414

2817

92

345

176

1993

272

1990

209

51

028

061

828

232

288

02

15T

axA

ct19

911

875

102

00

547

312

01

268

82

0319

922

678

107

01

250

3146

64

552

275

1993

232

41

209

074

339

620

64

299

173

1994

196

41

262

037

842

515

45

113

074

Sam

ple

isH

allndash

Lie

bman

rm

sw

ith

Dec

embe

r

scal

year

sth

atha

dco

mp

lete

dat

afo

ron

eof

the

thre

eta

x-re

form

per

iod

sst

ud

ied

Taxation of Executive Compensation 39

1994 level These patterns are consistent with Goolsbeersquos claim that tax-payers responded to the anticipated increase in marginal tax rates byshifting gains into 1992

The evidence from the other two tax reforms is less clear suggestingthat there might be factors other than tax rates determining the timing ofoption gains ERTA 1981 reduced the top marginal tax rate from 70percent in 1980 to 50 percent in 1982 However there is no sign thatoption gains were reduced in 1980 and 1981 and then increased in 198217

There was a big increase in option gains in 1983 however Similarly thehighest year of stock-option gains in the period around 1986 was 1987 astrange year to take gains since taxpayers would have known that thepersonal rate would be 10 percentage points lower if they waited untilthe following year18

What all the peak years of option gains have in common is that theyfollow years of strong stock market performance The fth row of Table 7shows the average stock market return during the previous year for therms in our sample The large value of option gains in 1983 followed a355-percent increase in the stock market The 1992 boom in option gainsfollowed a 466-percent increase in the stock market the more modest1987 peak in option gains followed a 19-percent gain (and a 43-percentgain two years before)

Strong stock market performance raises the value of options availableto exercise The sixth row of the table shows the mean value of optionsheld by CEOs at the beginning of the year All three of the peak years ofoption gains correspond with peak years in terms of the value of optionsavailable for exercise Since stock market appreciation during the yearalso increases the value of options available to exercise later in that yearwe need to allow for current-year stock market appreciation as well

Table 8 contains Goolsbee-style regressions for each of the tax re-forms The rst two columns reprint the results from Goolsbee (1999)The third and fourth columns replicate his results in our sample Inparticular with a specication similar to his we are able to replicate hisnding that the current after-tax share has a positive coefcient and thefollowing yearrsquos after-tax share has a coefcient of roughly negative oneColumns (5) through (8) show that this pattern was unique to the 1993tax reform Indeed in the TRA 86 regressions the coefcient on the

17 As we explained above for executives covered by the maximum tax on earned incomethere was not a decline in tax rates between 1981 and 198218 It is possible that the large stock market decline in October 1997 may have promptedsome people to exercise options earlier than they otherwise would have However anincrease in option exercises typically follows stock market increases not declines

TA

BL

E8

Att

empt

sto

Rep

lica

teth

eT

ax-S

hift

ing

Res

ult

inO

ther

Tim

eP

erio

ds

Goo

lsbe

eG

ools

bee

1993

1993

Ent

ire

Ent

ire

regr

essi

onre

gres

sion

refo

rmre

form

TR

A86

TR

A86

ER

TA

81E

RT

A81

1980

ndash199

419

80ndash1

994

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

ln[1

2ta

x(t)

]1

244

108

20

455

032

62

017

62

017

22

016

72

016

70

026

20

068

(01

34)

(01

27)

(05

37)

(05

42)

(02

10)

(02

09)

(00

90)

(00

91)

(00

67)

(00

65)

ln[1

2ta

x(t1

1)]

20

867

20

764

21

214

20

917

047

80

722

20

103

004

30

176

018

3(0

118

)(0

112

)(0

491

)(0

499

)(0

157

)(0

174

)(0

072

)(0

072

)(0

068

)(0

066

)

Tim

e0

135

007

62

000

52

036

40

041

000

80

094

003

50

056

002

7(0

008

)(0

008

)(0

049

)(0

049

)(0

028

)(0

028

)(0

025

)(0

026

)(0

003

)(0

003

)

ln(m

arke

tva

lue)

066

10

489

023

70

334

039

5(0

015

)(0

058

)(0

039

)(0

042

)(0

017

)

Yea

rs19

91ndash1

995

1991

ndash199

519

90ndash1

994

1990

ndash199

419

85ndash1

989

1985

ndash198

919

80ndash1

983

1980

ndash198

319

80ndash1

994

1980

ndash199

4

Ad

just

edR

20

720

770

610

630

690

760

800

810

630

68

Obs

erva

tion

s13

856

132

371

385

136

41

470

143

31

113

106

23

968

385

9

Dep

end

ent

vari

able

isth

elo

gof

taxa

ble

inco

me

All

regr

essi

ons

con

tain

rm

xe

def

fect

sG

ools

bee

regr

essi

on(2

)al

soco

ntai

nstw

oot

her

vari

able

sno

tsh

own

inth

eta

ble

Stan

dar

der

rors

inp

aren

thes

es

Taxation of Executive Compensation 41

TABLE 9Alternative Explanations for the Timing of Option Exercises

1993 Entirereform TRA 86 ERTA 81 1980ndash1994

(1) (2) (3) (4)

ln[12tax(t)] 0108 0111 20051 0201(0324) (0136) (0048) (0057)

ln[12tax(t11)] 20134 0573 0125 0150(0336) (0156) (0053) (0066)

ln(market value) 0418 0177 0338 0418(0061) (0039) (0046) (0163)

Firmrsquos stock market return (t) 0298 0268 0214 0258(0044) (0045) (0031) (0023)

Firmrsquos stock market return 0162 0206 0031 0128(t21) (0041) (0008) (0029) (0021)

ln(value of options held by 0004 0008 0001 0009CEO at start of year) (0003) (0003) (0002) (0001)

Years 1990ndash1994 1985ndash1989 1980ndash1983 1980ndash1994Adjusted R2 064 071 082 068Observations 1349 1427 1036 3812

The dependent variable is the log of taxable income All regressions contain rm xed effects Stan-dard errors in parentheses

following yearrsquos tax rate is positive and signicant Columns (9) and (10)pool all three reforms and fail to replicate the pattern that would reectintertemporal shifting19

It is possible that the ease with which one can shift the timing ofoption gains is asymmetric It may be harder to shift option gains for-ward in time since some options expire and therefore must be exercisednow In addition options were a much smaller portion of compensationin the early 1980s so perhaps the ERTA 81 results are not surprisingNonetheless these results cast doubt on the proposition that this is thechannel through which spurious estimates of the elasticity of taxableincome have occurred for previous tax reforms and they suggest thatwe should look harder for alternative explanations for the timing ofoption gains

Table 9 attempts to predict options gains using the lagged and currentincrease in a rmrsquos market value and the value of the options held by theCEO at the start of the year The two stock market variables stronglypredict option gains while the value of options held has the correct sign

19 The results also fail to suggest large permanent effects of marginal tax rates on taxableincome

42 Hall amp Liebman

in all of the specications and is signicant in the pooled sample eventhough it is highly correlated with the two stock market return variablesIn contrast the tax-timing explanation fails even in the 1993 period

It is important to realize that our results do not invalidate Goolsbeersquoscritique of claims that aggregate revenue data from 1992 and 1993 provethat elasticities of taxable income are large We have simply given analternative explanation for why taxable incomes were unusually high in1992 However our results do cast doubt on the conclusion that largetiming shifts are ubiquitous and capable of explaining away the morecareful estimates of taxable income elasticities that have been performedusing micro data (eg Feldstein 1995 Auten and Carroll 1995)

One further point is in order It is sometimes argued that the combina-tion of increased use of options and the booming stock market explainswhy federal tax revenues have been so much higher than expected dur-ing the 1990s However while large option gains increase revenue fromthe personal income tax they produce offsetting deductions for cor-porations The net impact on revenue is only the difference in the twomarginal tax rates times the option gain While the gap between thepersonal rate and the effective corporate rate is nontrivial the impact oflarge option exercises on total revenue is substantially smaller than theamount by which the exercises inate personal tax revenue and reducecorporate tax revenue

9 CONCLUSION

We have described the tax rules for executive pay in detail and analyzedhow changes in various tax rates affect the tax advantages of stock optionsrelative to salary and bonus We nd that there is a moderate tax advan-tage to optionsmdashcurrently about $4 per $100 in pretax executive paymdashand that changes in the tax advantage of options over time have had atmost a modest impact on the composition of pay Corporate governancefactors particularly the role of large investors appear to be more impor-tant in explaining the dramatic increase in option pay

In addition our evidence suggests that more direct attempts to use taxpolicy to inuence executive compensation have had little effect Wend that the million-dollar rule led companies to substitute perfor-mance-based pay for salary But our evidence suggests that this substitu-tion was quite modest and there is no evidence that the total level of paywas reduced Overall although the stock-option explosion has dramati-cally increased the link between pay and performance this change isdue almost entirely to nontax factors

Taxation of Executive Compensation 43

REFERENCES

Auerbach Alan J (1996) ldquoThe Dynamic Effects of Tax Law Asymmetriesrdquo Re-view of Economic Studies 53451ndash405

mdashmdashmdash and James Poterba (1987) ldquoTax Loss Carryforwards and Corporate TaxIncentivesrdquo In The Effects of Taxation on Capital Accumulation M Feldstein(ed) Chicago University of Chicago Press

Altshuler Rosanne and Alan Auerbach (1990) ldquoThe Signicance of Tax LawAsymmetries An Empirical Investigationrdquo Quarterly Journal of Economics 10561ndash86

Auten Gerald and Robert Carroll (1995) ldquoThe Effect of Income Taxes on House-hold Behaviorrdquo Ofce of Tax Analysis US Department of the Treasury

Baker George P and Brian J Hall (1998) ldquoCEO Incentives and Firm SizerdquoNBER Working Paper no 6868 December

Core John and Wayne Guay (1999) ldquoThe Use of Equity Grants to ManageOptimal Equity Incentive Levelsrdquo Draft

Feldstein M (1995) ldquoThe Effect of Marginal Tax Rates on Taxable Income APanel Study of the 1986 Tax Reform Actrdquo Journal of Political Economy 103 (no3)551ndash72

mdashmdashmdash and D Feenberg (1996) ldquoThe Effect of Increased Tax Rates on TaxableIncome and Economic Efciency A Preliminary Analysis of the 1993 Tax RateIncreasesrdquo In Tax Policy and the Economy vol 10 James M Poterba (ed)Cambridge MA MIT Press

Gompers Paul A and Andrew Metrick (1998) ldquoInstitutional Investors andEquity Pricesrdquo NBER Working Paper no 6723

Goolsbee Austan (1999) ldquoWhat Happens When You Tax the Rich Evidencefrom Executive Compensationrdquo NBER Working Paper no 6333

Graham John R (1996) ldquoProxies for the Corporate Marginal Tax Raterdquo Journal ofFinancial Economics 42187ndash221

Hall Brian J (1999) ldquoThe Design of Multi-year Stock Option Plansrdquo Journal ofApplied Corporate Finance 12(2)97ndash106

mdashmdashmdash and Jeffrey B Liebman (1998) ldquoAre CEOs Really Paid Like Bureau-cratsrdquo Quarterly Journal of Economics CXII(no 3)653ndash691

Jensen Michael C and William M Meckling (1976) ldquoTheory of the Firm Mana-gerial Behavior Agency Costs and Ownership Structurerdquo Journal of FinancialEconomics III305ndash360

mdashmdashmdash and Kevin J Murphy (1990) ldquoPerformance Pay and Top-ManagementIncentivesrdquo Journal of Political Economy XCVIII225ndash264

Kaplan Steven N (1997) ldquoThe Evolution of US Corporate Governance WeAre All Henry Kravis Nowrdquo Draft

Livingstone Jane R (1999) ldquoExecutive Compensation Contracting and Re-sponses to Increased Tax Costsrdquo Louisiana State University June

Miller Merton H and Myron S Scholes (1982) ldquoExecutive CompensationTaxes and Incentivesrdquo In Financial Economics Essays in Honor of Paul CootnerWilliam F Sharpe and Cathryn M Cootner (eds) Princeton NJ Prentice-Hall

Murphy Kevin J (1999) ldquoExecutive Compensationrdquo forthcoming in OrleyAshenfelter and David Card (eds) Handbook of Labor Economics Vol 3 NorthHolland

Perry Todd and Marc Zenner (1999) ldquoPay for Performance Government Regu-