capm, arbitrage, and linear factor · pdf filecapm, arbitrage, and linear factor models george...

3.1: CAPM 3.2: Arbitrage 3.3: APT 3.4: Summary

CAPM, Arbitrage, and Linear Factor Models

George Pennacchi

University of Illinois

George Pennacchi University of Illinois

CAPM, Arbitrage, Linear Factor Models 1/ 41


Introduction

We now assume all investors actually choose mean-variancee¢ cient portfolios.

By equating these investors�aggregate asset demands toaggregate asset supply, an equilibrium single risk factor pricingmodel (CAPM) can be derived.

Relaxing CAPM assumptions may allow for multiple riskfactors.

Arbitrage arguments can be used to derive a multifactorpricing model (APT)

Multifactor models are very popular in empirical asset pricing




Review of Mean-Variance Portfolio Choice

Recall that for n risky assets and a risk-free asset, the optimalportfolio weights for the n risky assets are

!� = �V�1��R � Rf e

�(1)

where � � Rp � Rf��R � Rf e

�0 V�1 ��R � Rf e� .The amount invested in the risk-free asset is then 1� e 0!�.Rp is determined by where the particular investor�sindi¤erence curve is tangent to the e¢ cient frontier.

All investors, no matter what their risk aversion, choose therisky assets in the same relative proportions.




Tangency Portfolio

Also recall that the e¢ cient frontier is linear in �p and Rp :

Rp = Rf +��R � Rf e

�0 V�1 ��R � Rf e�� 12 �p (2)

This frontier is tangent to the �risky asset only� frontier,where the following graph denotes this tangency portfolio as!m located at point �m , Rm .




Graph of E¢ cient Frontier




The Tangency Portfolio

Note that the tangency portfolio satis�es e 0!m = 1. Thus

e 0�V�1��R � Rf e

�= 1 (3)

or� = m �

h�R � Rf e

�0V�1e

i�1(4)

so that!m = mV�1(�R � Rf e) (5)




Asset Covariances with the Tangency Portfolio

Now de�ne �M as the n � 1 vector of covariances of thetangency portfolio with each of the n risky assets. It equals

�M = V!m = m(�R � Rf e) (6)

By pre-multiplying equation (6) by !m 0, we also obtain thevariance of the tangency portfolio:

�2m = !m 0V!m = !m 0�M = m!m 0(�R � Rf e) (7)

= m(�Rm � Rf )

where �Rm � !m 0 �R is the expected return on the tangencyportfolio.




Expected Excess Returns

Rearranging (6) and substituting in for 1m =

1�2m(Rm � Rf )

from (7), we have

(�R � Rf e) =1m�M =

�M�2m(�Rm � Rf ) = �(�Rm � Rf ) (8)

where � � �M�2mis the n � 1 vector whose i th element is

Cov (~Rm ;eRi )Var (~Rm)

.

Equation (8) links the excess expected return on the tangencyportfolio, (�Rm � Rf ), to the excess expected returns on theindividual risky assets, (�R � Rf e).




CAPM

The Capital Asset Pricing Model is completed by noting thatthe tangency portfolio, !m , chosen by all investors must bethe equilibrium market portfolio.

Hence, Rm and �2m are the mean and variance of the marketportfolio returns and �M is its covariance with the individualassets.

Aggregate supply can be modeled in di¤erent ways(endowment economy, production economy), but inequilibrium it will equal aggregate demands for the riskyassets in proportions given by !m .

Also Rf < Rmv for assets to be held in positive amounts(!mi > 0).




CAPM: Realized Returns

De�ne asset i�s and the market�s realized returns as~Ri = �Ri + ~� i and ~Rm = �Rm + ~�m where ~� i and ~�m are theunexpected components. Substitute these into (8):

~Ri = Rf + � i (~Rm � ~�m � Rf ) + ~� i (9)

= Rf + � i (~Rm � Rf ) + ~� i � � i ~�m= Rf + � i (~Rm � Rf ) + e"i

where e"i � ~� i � � i ~�m . Note that

Cov(~Rm ;e"i ) = Cov(~Rm ; ~� i )� � iCov(~Rm ; ~�m) (10)

= Cov(~Rm ; ~Ri )�Cov(~Rm ; ~Ri )

Var(~Rm)Cov(~Rm ; ~Rm)

= Cov(~Rm ; ~Ri )� Cov(~Rm ; ~Ri ) = 0




Idiosyncratic Risk

Since Cov(~Rm ;e"i ) = 0, from equation (9) we see that thetotal variance of risky asset i , �2i , equals:

�2i = �2i �2m + �

2"i

(11)

Another implication of Cov(~Rm ;e"i ) = 0 is that equation (9)represents a regression equation.

The orthogonal, mean-zero residual, e"i , is referred to asidiosyncratic, unsystematic, or diversi�able risk.

Since this portion of the asset�s risk can be eliminated by theindividual who invests optimally, there is no �price�or �riskpremium�attached to it.




What Risk is Priced?

To make clear what risk is priced, denote �Mi = Cov(~Rm ; eRi ),which is the i th element of �M . Also let �im be thecorrelation between eRi and ~Rm .Then equation (8) can be rewritten as

�Ri � Rf =�Mi�2m

(�Rm � Rf ) =�Mi�m

(�Rm � Rf )�m

(12)

= �mi�i(�Rm � Rf )

�m= �mi�iSe

where Se � (�Rm�Rf )�m

is the equilibrium excess return on themarket portfolio per unit of market risk.




What Risk is Priced? cont�d

Se � (�Rm�Rf )�m

is known as the market Sharpe ratio.

It represents �the market price�of systematic ornondiversi�able risk, and is also referred to as the slope of thecapital market line, where the capital market line is thee¢ cient frontier that connects the points Rf and

�Rm;�m

�.

Now de�ne !mi as the weight of asset i in the market portfolioand Vi as the i th row of V . Then

@�m@!mi

=12�m

@�2m@!mi

=12�m

@!mV!m

@!mi=

12�m

2Vi!m =1�m

nXj=1

!mj �ij

(13)where �ij is the i ; j th element of V .




What Risk is Priced? cont�d

Since ~Rm =nPj=1!mj~Rj , then Cov(~Ri ; ~Rm) =

nPj=1!mj �ij . Hence,

(13) is@�m@!mi

=1�mCov(~Ri ; ~Rm) = �im�i (14)

Thus, �im�i is the marginal increase in �market risk,��m ,from a marginal increase of asset i in the market portfolio.Thus �im�i is the quantity of asset i�s systematic risk.

We saw in (12) that �im�i multiplied by the price ofsystematic risk, Se , determines an asset�s required riskpremium.




Zero-Beta CAPM (Black, 1972)

Does a CAPM hold when there is no riskless asset?Suppose the economy has I total investors with investor ihaving a proportion Wi of the economy�s total initial wealthand choosing an e¢ cient frontier portfolio !i = a+ b�Rip ,where �Rip re�ects investor i�s risk aversion.Then the risky asset weights of the market portfolio are

!m =IXi=1

Wi!i =

IXi=1

Wi�a+ b�Rip

�(15)

= aIXi=1

Wi + bIXi=1

Wi �Rip = a+ b�Rm

where �Rm �PIi=1Wi �Rip .




Zero-Beta CAPM cont�d

Equation (15) shows that the aggregate market portfolio, !m ,is a frontier portfolio and its expected return, Rm , is aweighted average of the expected returns of the individualinvestors�portfolios.

Consider the covariance between the market portfolio and anarbitrary risky portfolio with weights !0, random return ofeR0p , and mean return of R0p :

Cov�eRm ; eR0p�= !m0V!0= �a+ b�Rm�0 V!0 (16)

=

�&V�1e � �V�1 �R

&� � �2 +�V�1 �R � �V�1e

&� � �2�Rm

�0V!0





=&e 0V�1V!0 � ��R 0V�1V!0

&� � �2

+� �Rm �R 0V�1V!0 � ��Rme 0V�1V!0

&� � �2

=& � �R0p + � �RmR0p � ��Rm

&� � �2

Rearranging (16) gives

R0p =��Rm � &� �Rm � �

+ Cov�eRm ; eR0p� &� � �2

� �Rm � �(17)

Recall from (2.32) that �2m =1� +

�(Rm�� )

2

&��2George Pennacchi University of Illinois




Multiply and divide the second term of (17) by �2m , and addand subtract �

2

� from the top and factor out � from thebottom of the �rst term to obtain:

R 0p =�

�� &� � �2

�2�Rm � �

�

� + Cov�eRm ; eR0p��2m

0B@1�+��Rm � �

�

�2&� � �2

1CA &� � �2

��Rm � �

=�

�� &� � �2

�2�Rm � �

�

� + Cov�eRm ; eR0p��2m

0@Rm � �

�+

&� � �2

�2�Rm � �

�

�1A(18)

From (2.39), the �rst two terms of (18) equal the expectedreturn on the portfolio that has zero covariance with themarket portfolio, call it Rzm . Thus, equation (18) can bewritten as





R0p = Rzm +Cov

�eRm ; eR0p��2m

�Rm � Rzm

�(19)

= Rzm + �0�Rm � Rzm

�Since !0 is any risky-asset portfolio, including a single assetportfolio, (19) is identical to the previous CAPM result (12)except that Rzm replaces Rf .Thus, a single factor CAPM continues to exist when there isno riskless asset.Stephen Ross (1976) derived a similar multifactor relationship,not based on investor preferences but rather the principle ofarbitrage.




Arbitrage

If a portfolio with zero net investment cost can produce gains,but never losses, it is an arbitrage portfolio.

In e¢ cient markets, arbitrage should be temporary:exploitation by investors moves prices to eliminate it.

If equilibrium prices do not permit arbitrage, then the law ofone price holds:

If di¤erent assets produce the same future payo¤s, then thecurrent prices of these assets must be the same.

Not all markets are always in equilibrium: in some casesarbitrage trades may not be possible.




Example: Covered Interest Parity

Covered interest parity links spot and forward foreignexchange markets to foreign and domestic money markets.

Let F0� be the current date 0 forward price for exchanging oneunit of a foreign currency � periods in the future, e.g. thedollar price to be paid � periods in the future for delivery ofone unit of foreign currency � periods in the future.

Let S0 be the spot price of foreign exchange, that is, thecurrent date 0 dollar price of one unit of foreign currency tobe delivered immediately.

Also let Rf be the per-period risk-free (money market) returnfor borrowing or lending in dollars over the period 0 to � , anddenote as R�f the per-period risk-free return for borrowing orlending in the foreign currency over the period 0 to � .




Covered Interest Parity cont�d

Now let�s construct a zero net cost portfolio.

1 Sell forward one unit of foreign currency at price F0� (0 cost).2 Purchase the present value of one unit of foreign currency,1=R�f

� and invest in a foreign bond at R�f . (S0=R�f� cost).

3 Borrow S0=R�f� dollars at the per-period return Rf (�S0=R�f �

cost)

At date � , note that the foreign currency investment yieldsR�f

�=R�f� = 1 unit of the foreign currency, which covers the

short position in the forward foreign exchange contract.For delivering this foreign currency, we receive F0� dollars butwe also owe a sum of R�f S0=R

��f due to our dollar borrowing.

Thus, our net proceeds at date � are

F0� � R�f S0=R��f (20)George Pennacchi University of Illinois



Covered Interest Parity cont�d

These proceeds are nonrandom; they depend only on pricesand riskless rates quoted at date 0.

If this amount is positive (negative), then we should buy (sell)this portfolio as it represents an arbitrage.

Thus, in the absence of arbitrage, it must be that

F0� = S0R�f =R��f (21)

which is the covered interest parity condition: given S0, R�fand Rf , the forward rate is pinned down.

Thus, when applicable, pricing assets or contracts by rulingout arbitrage is attractive in that assumptions regardinginvestor preferences or beliefs are not required.




Example: Arbitrage and CAPM

Suppose a single source of (market) risk determines allrisky-asset returns according to the linear relationship

eRi = ai + bi~f (22)

where eRi is the i th asset�s return and ~f is a single risk factorgenerating all asset returns, where E [~f ] = 0 is assumed.

ai is asset i�s expected return, that is, E [eRi ] = ai .bi is the sensitivity of asset i to the risk factor (asset i�s betacoe¢ cient).

There is also a riskfree asset returning Rf .

Now construct a portfolio of two assets, with a proportion !invested in asset i and (1� !) invested in asset j .




Arbitrage and CAPM cont�d

This portfolio�s return is given byeRp = !ai + (1� !)aj + !bi~f + (1� !)bj~f (23)

= !(ai � aj ) + aj + [!(bi � bj ) + bj ] ~f

If the portfolio weights are chosen such that

!� =bj

bj � bi(24)

then the random component of the portfolio�s return iseliminated: Rp is risk-free.The absence of arbitrage requires Rp = Rf , so that

Rp = !�(ai � aj ) + aj = Rf (25)





Rearranging this condition

bj (ai � aj )bj � bi

+ aj = Rf

bjai � biajbj � bi

= Rf

which can also be written as

ai � Rfbi

=aj � Rfbj

� � (26)

which states that expected excess returns, per unit of risk,must be equal for all assets. � is de�ned as this risk premiumper unit of factor risk.





Equation (26) is a fundamental relationship, and similarlaw-of-one-price conditions hold for virtually all asset pricingmodels.

For example, we can rewrite the CAPM equation (12) as

�Ri � Rf�im�i

=(�Rm � Rf )

�m� Se (27)

so that the ratio of an asset�s expected return premium,�Ri � Rf , to its quantity of market risk, �im�i , is the same forall assets and equals the slope of the capital market line, Se .

What can arbitrage arguments tell us about more generalmodels of risk factor pricing?




Linear Factor Models

The CAPM assumption that all assets can be held by allindividual investors is clearly an oversimpli�cation.

In addition to the risk from returns on a global portfolio ofmarketable assets, individuals are likely to face multiplesources of nondiversi�able risks. This is a motivation for themultifactor Arbitrage Pricing Theory (APT) model.

APT does not make assumptions about investor preferencesbut uses arbitrage pricing to restrict an asset�s risk premium.

Assume that there are k risk factors and n assets in theeconomy, where n > k.

Let biz be the sensitivity of the i th asset to the z th risk factor,where ~fz is the random realization of risk factor z .




Linear Factor Models cont�d

Let e"i be idiosyncratic risk speci�c to asset i , which byde�nition is independent of the k risk factors, ~f1,...,~fk , and thespeci�c risk of any other asset j , e"j .For ai , the expected return on asset i , the linearreturn-generating process is assumed to be

eRi = ai +Pkz=1 biz~fz + e"i (28)

where E [e"i ] = E h~fzi = E he"i~fzi = 0 and E [e"ie"j ] = 0 8 i 6= j .The risk factors are transformed to be mutually independentand normalized to unit variance, E

h~f 2zi= 1.




Linear Factor Models cont�d

Finally, the idiosyncratic variance is assumed �nite:

E�e"2i � � s2i < S2 <1 (29)

Note Cov�eRi ; ~fz� = Cov �biz~fz ; ~fz� = bizCov �~fz ; ~fz� = biz .

In the previous example there was only systematic risk, whichwe could eliminate with a hedge portfolio. Now each asset�sreturn contains an idiosyncratic risk component.

We will use the notion of asymptotic arbitrage to argue thatassets�expected returns will be "close" to the relationshipthat would result if they had no idiosyncratic risk, since it isdiversi�able with a large number of assets.




Arbitrage Pricing Theory

Consider a portfolio of n assets where �ij is the covariancebetween the returns on assets i and j and the portfolio�sinvestment amounts are W n � [W n

1 Wn2 :::W

nn ]0.

Consider a sequence of these portfolios where the number ofassets in the economy is increasing, n = 2; 3; : : : .

De�nition: An asymptotic arbitrage exists if:(A) The portfolio requires zero net investment:Pn

i=1Wni = 0

(B) The portfolio return becomes certain as n gets large:

limn!1

Pni=1Pnj=1W

ni W

nj �ij = 0




Arbitrage Pricing Theory cont�d

(C) The portfolio�s expected return is always bounded above zeroPni=1W

ni ai � � > 0

Theorem: If no asymptotic arbitrage opportunities exist, then theexpected return of asset i , i = 1; :::; n, is described by the followinglinear relation:

ai = �0 +Pkz=1 biz�z + � i (�)

where �0 is a constant, �z is the risk premium for risk factor efz ,z = 1; :::; k, and the expected return deviations, � i , satisfy





Pni=1 � i = 0 (i)Pn

i=1 biz� i = 0; z = 1; :::; k (ii)

limn!1

1nPni=1 �

2i = 0 (iii)

Note that (iii) says that the average squared error (deviation)from the pricing rule (�) goes to zero as n becomes large.Thus, as the number of assets increases relative to the riskfactors, expected returns will, on average, become closelyapproximated by the relation ai = �0 +

Pkz=1 biz�z .





Also note that if there is a risk-free asset (biz = 0, 8 z), itsreturn is approximately �0.

Proof: For a given n > k, think of running a cross-sectionalregression of the ai�s on the biz�s by projecting the dependentvariable vector a = [a1 a2 ::: an]0 on the k explanatory variablevectors bz = [b1z b2z ::: bnz ]0, z = 1; :::; k. De�ne � i as theregression residual for observation i , i = 1; :::; n.

Denote �0 as the regression intercept and �z , z = 1; :::; k, asthe estimated coe¢ cient on explanatory variable z .

The regression estimates and residuals must then satisfy

ai = �0 +Pkz=1 biz�z + � i (30)





By the properties of OLS,Pni=1 � i = 0 and

Pni=1 biz� i = 0,

z = 1; :::; k. Thus, we have shown that (�), (i), and (ii) canbe satis�ed.

The last and most important part of the proof is to show that(iii) must hold in the absence of asymptotic arbitrage.

Construct a zero-net-investment arbitrage portfolio with thefollowing investment amounts:

Wi =� iqPni=1 �

2i n

(31)

so that greater amounts are invested in assets having thegreatest relative expected return deviation.





The arbitrage portfolio return is given by

~Rp =nXi=1

W i eRi (32)

=1qPni=1 �

2i n

"nXi=1

� i eRi#=

1qPni=1 �

2i n

"nXi=1

� i

ai +

kXz=1

biz~fz + e"i!#

SincePni=1 biz� i = 0, z = 1; :::; k, this equals

~Rp =1qPni=1 �

2i n[Pni=1 � i (ai + e"i )] (33)





Calculate this portfolio�s mean and variance. Takingexpectations:

Eh~Rpi=

1qPni=1 �

2i n[Pni=1 � iai ] (34)

since E [e"i ] = 0. Substitute ai = �0 +Pkz=1 biz�z + � i :

Eh~Rpi=

1qPni=1 �

2i n

"�0

nXi=1

� i +kXz=1

�z

nXi=1

� ibiz

!+

nXi=1

�2i

#(35)

and sincePni=1 � i = 0 and

Pni=1 � ibiz = 0, this simpli�es to





Eh~Rpi=

1qPni=1 �

2i n

nXi=1

�2i =

vuut1n

nXi=1

�2i (36)

Calculate the variance. First subtract (34) from (33):

~Rp � Eh~Rpi=

1qPni=1 �

2i n[Pni=1 � ie"i ] (37)

Since E [e"ie"j ] = 0 for i 6= j and E [e"2i ] = s2i :E��~Rp � E

h~Rpi�2�

=

Pni=1 �

2i s2i

nPni=1 �

2i<

Pni=1 �

2i S2

nPni=1 �

2i=S2

n(38)





Thus, as n!1, the variance of the portfolio�s return goes tozero, and the portfolio�s actual return converges to itsexpected return in (36):

limn!1

~Rp = Eh~Rpi=

r1nPni=1 �

2i (39)

and absent asymptotic arbitrage opportunities, this certainreturn must equal zero. This is equivalent to requiring

limn!1

1nPni=1 �

2i = 0 (40)

which is condition (iii).





We see that APT, given by the relation ai = �0 +Pkz=1 biz�z ,

can be interpreted as a multi-beta generalization of CAPM.

However, whereas CAPM says that its single beta should bethe sensitivity of an asset�s return to that of the marketportfolio, APT gives no guidance as to what are theeconomy�s multiple underlying risk factors.

Empirical implementations include Chen, Roll and Ross(1986), Fama and French (1993), Heaton and Lucas (2000),Chen, Novy-Marx and Zhang (2011).

We will later develop another multi-beta asset pricing model,the Intertemporal CAPM (Merton, 1973).




Summary

The CAPM arises naturally from mean-variance analysis.

CAPM and APT can also be derived from arbitragearguments and linear models of returns.

Arbitrage pricing arguments are very useful for pricingcomplex securities, e.g. derivatives.



capm, arbitrage, and linear factor · pdf filecapm, arbitrage, and linear factor models george...

Documents