coups, corporations, and classified information e k s neconweb.umd.edu/~kaplan/coups.pdf · 2012....
TRANSCRIPT
COUPS, CORPORATIONS, AND CLASSIFIEDINFORMATION∗
ARINDRAJIT DUBE
ETHAN KAPLAN
SURESH NAIDU
We estimate the impact of coups and top-secret coup authorizations on as-set prices of partially nationalized multinational companies that stood to ben-efit from U.S.-backed coups. Stock returns of highly exposed firms reacted tocoup authorizations classified as top-secret. The average cumulative abnormalreturn to a coup authorization was 9% over 4 days for a fully nationalized com-pany, rising to more than 13% over 16 days. Precoup authorizations accountedfor a larger share of stock price increases than the actual coup events them-selves. There is no effect in the case of the widely publicized, poorly executedCuban operations, consistent with abnormal returns to coup authorizations re-flectingcredibleprivate information. Wealsointroducetwonewintuitiveandeasyto implement nonparametric tests that do not rely on asymptotic justifications.JEL Codes: F50, G14.
I. INTRODUCTION
Covert operations conducted by intelligence agencies were a keycomponent of superpower foreign policy during the Cold War. Forthe United States, many of these operations had the expressedgoal of replacing “unfriendly” regimes—often ones that had ex-propriated multinational corporate property—and were plannedunder extreme secrecy. Since corporate property was always re-stored after a successful regime change, these operations werepotentially profitable tonationalizedcompanies. If foreknowledgeof these operations was truly secret, then precoup asset pricesshouldnot havereflectedtheexpectedfuturegains. However, thisarticle shows that not only were U.S.-supported coups valuable to
∗We thank Martin Berlin, Remeike Forbes, Nathan Lane, Zihe Liu, EttorePanetti, Andre Shepley, and Laurence Wilse-Samson for excellent research assis-tance. Frans Buelens helped us greatly in obtaining data. Noel Maurer sharedhis list of U.S. multinational expropriations with us. Oliver Boguth provided his-torical data on three Fama-French factors. Marcos Chamon, Stefano DellaVigna,Ray Fisman, Eric Freeman, David Gibbs, Lena Nekby, Torsten Persson, JohnPrados, Gerard Roland, and seminar participants at CEMFI, Hampshire College,LSE, IIES, NBERPolitical Economy Summer Institute, the NewSchool, NYU, theSanta Fe Institute, the Stockholm University Economics Department, the Stock-holm School of Economics, UC Berkeley, UC Riverside, the University of Michiganat Ann Arbor, the University of Oslo, and the University of Warwick all providedhelpful comments.c© The Author(s) 2011. Published by Oxford University Press, on behalf of President and
Fellows of Harvard College. All rights reserved. For Permissions, please email: [email protected] Quarterly Journal of Economics, (2011) 126, 1375–1409. doi:10.1093/qje/qjr030.Advance Access publication on August 11, 2011.
1375
at McK
eldon Library 34 on January 31, 2012
http://qje.oxfordjournals.org/D
ownloaded from
1376 QUARTERLY JOURNAL OF ECONOMICS
partially nationalized multinationals, but in addition, assettraders arbitraged supposedly “top-secret” information concern-ing plans to overthrow foreign governments.
Specifically, we estimate the effect of secret U.S., as well asallied government decisions to overthrow foreign governments onthe stock prices of companies that stood to benefit from regimechange. We consider companies that had a large fraction of theirassets expropriated by a government that was subsequently atarget of a U.S.-sponsored covert operation aimed at overthrow-ing the regime. Using timelines reconstructed from official CIAdocuments, we find statistically and economically significant ef-fects on stock prices both from the regime change itself and from“top-secret” authorizations. Total stock price gains from coup au-thorizations were three times larger in magnitude than pricechanges fromthecoups themselves. Wethus showthat thereweresubstantial economic incentives for firms to lobby for these op-erations. Although we are unable to discern precisely who wastrading, or whether these economic incentives were decisive forU.S. policy makers (versus political ideology or geopolitics), we doshow that regime changes led to significant economic gains forcorporations that stood to benefit from U.S. interventions in de-veloping countries.
Ourfindings complement otherevidenceinempirical politicaleconomy that large, politically connectedfirms benefit from favor-able political regimes (Fisman 2001; Faccio 2006; Knight 2006;Snowberg, Wolfers, and Zeitzwitz 2007). However, we show thatfirms benefit not only from publicly announced events but alsofrom top-secret events, suggesting information flows from covertoperations into markets. Our results are consistent with recentpapers that have used asset price data to show that companiescanprofit fromconflict (GuidolinandLaFerrara2007; DellaVignaand La Ferrara 2010). We also provide evidence that private in-formation generally leaks into asset prices slowly over time. Thisis consistent with both private information theories of asset pricedetermination (Allen, Morris, and Shin 2006) and the empiricalliterature on insider trading (Meulbroek 1992). We differentiateour work from the prior work on insider trading in so far as theprivate information being traded on concerns government policy,and not company decisions or other information generated withinthe company.
The theoretical literature on coups in economics has empha-sized the role of domestic elites (Acemoglu and Robinson 2006).
at McK
eldon Library 34 on January 31, 2012
http://qje.oxfordjournals.org/D
ownloaded from
COUPS, CORPORATIONS, AND CLASSIFIED INFORMATION 1377
However, antidemocratic political transitions have often beeninstigated, planned, and even partially executed from abroad,most notably by the United States and the former Soviet Unionduring the Cold War. Operating under the threat of nuclearwar, direct conflict between the two superpowers was replaced bycovert and proxy operations to install supporting regimes (Chom-sky 1986; Kinzer 2006). According to Easterly et al. (2010), 24countryleaders wereinstalledbytheCIA and16 bytheKGBsincethe end of World War II.
Our article also makes a methodological contribution to hy-pothesis testing in event studies. The structure of our event studyallows us to improve on existing nonparametric tests. Nonpara-metric tests used in event studies do not use exact small sam-ple distributions but tests with faster asymptotic convergence toa normal distribution (Corrado and Zivney 1992; Campbell, Lo,and Mackinlay 1997). We introduce two new small sample testsmotivatedby Fisher’s exact test that are validwithout asymptoticjustifications.
Section II discusses the history of U.S. covert interventions,with background on each of the coups in our sample. Section IIIdescribes the data and our selection of companies and events.Section IV outlines our estimation strategies and Section Vreports ourmainresults alongwithanumberofrobustness checksand small sample tests. Section VI provides an interpretation ofour main results; we decompose coup gains to a multinationalinto public and private information components. We conclude inSection VII.
II. BACKGROUND AND HISTORY
The Central Intelligence Agency was created in 1947 underthe National Security Act. The act allowed for “functions and du-ties related to intelligence affecting the national security,” in ad-dition to intelligence gathering (Weiner 2007). Initially, the scopeof the CIA was relegated tointelligence, though a substantial andvocal group advocated for a more active role for the agency. Thisculminated in National Security Council Directive No. 4, whichordered the CIA toundertake covert actions against communism.In the UnitedStates, covert operations designedtooverthrowfor-eign governments were usually first approved by the director ofthe CIA and then subsequently by the president of the UnitedStates (Weiner 2007).
at McK
eldon Library 34 on January 31, 2012
http://qje.oxfordjournals.org/D
ownloaded from
1378 QUARTERLY JOURNAL OF ECONOMICS
After Eisenhower’s election in 1952, Allen Dulles was ap-pointeddirectorof theagency. UnderDulles, theCIA expandeditsrole to include planning and executing overthrows of foreign gov-ernments using military force. All but eight of the CIA operationslisted in Table I, including four of the five studied in this articlebeganduringDulles’s reignas CIA directorundertheEisenhoweradministration. Allen Dulles was supported by his brother, JohnFoster Dulles, who was the contemporaneous Secretary of State.The Dulles brothers together wielded substantial influence overAmerican foreign policy from 1952 to 1960.
In1974, partlyduetopublicoutcryovertheU.S. involvementin the military coupin Chile, the Hughes-Ryan Act increasedcon-gressional oversight of CIA covert operations. In 1975, the U.S.legislature formed subcommittees to investigate American covert
TABLE I
COUP SELECTION
Planning DocsProject Country Year Declassified Description Coup Exprop.
Ajax Iran 1953 Yes Coup against Mossadeq Yes YesFU/Belt Chile 1970–73 Yes Coup against Allende Yes YesBloodstone Germany 1946 No Recruitment of Nazis No NoBrushfire US 1955 Yes Propaganda at Universities No NoCamelot Chile 1960s No Funded Anthro. Research No NAST/Circus Tibet 1955 No Trained Tibetan Rebels Yes NoDemocracy Nicaragua 1985 No Anti-Sandinista Operations No YesIA/Feature Angola 1975 No Supported Savimbi No YesFiend Albania 1949 No Insurgency Yes NoPM/Forget All over 1950s No Pro-U.S. Media Distortion No NAHaik Indonesia 1956/57 No Military Support for Rebels Yes YesHardNose Vietnam 1965 No Disrupt Viet Cong Supplies No NoMomentum Laos 1959 No Trained Hmong in Laos No NoMongoose Cuba 1961 Yes Post-Bay of Pigs Operations No YesOpera France 1951 No Electoral Manipulations No NoPaper China 1951 No Invasion from Burma No NoPB/Fortune, Guatemala 1952–54 Yes Coup Against Arbenz Yes Yes
PB/SuccessStole N. Korea 1950/51 No Sabotage No NoTiger Syria 1956 Yes Assassination Attempts No NoWashtub Guatemala 1954 Yes Anti-Arbenz Propaganda No YesWizard Congo 1960 Yes Lumumba Assassination Yes YesZapata Cuba 1960–61 Yes Bay of Pigs Yes Yes
Notes. Project is the name of the operation. Country is the target country of the operation. Year isthe year when the operation was carried out. Planning documents records yes if the planning documentsare publicly available. Description is a description of the operation. Coup is recorded as yes if a coup wasplanned as part of the operation and no otherwise, and Exprop. refers to whether the regime nationalized (orexpropriated) property from multinational firms operating within the country.
at McK
eldon Library 34 on January 31, 2012
http://qje.oxfordjournals.org/D
ownloaded from
COUPS, CORPORATIONS, AND CLASSIFIED INFORMATION 1379
action. Afterward, the intensity and scope of U.S. covert actionsfell substantially (Johnson 1989). The height of covert CIA activ-ity lasted slightly more than 20 years, encompassing the periodbetween 1952 and 1974.
Our sample of coups includes five such covert attempts. Thefirst one occurred in Iran in August 1953, when the CIA, jointlywith the United Kingdom’s MI6, engineered a toppling of PrimeMinister Mossadegh. Mossadegh had nationalized the oil fieldsand refinery at Abadan, which were the property of the Anglo-Iranian Oil Company, itself a partially publicly owned companyof the U.K. government. In Guatemala, the CIA overthrow ofJacobo Arbenz Guzman in June 1954 occurred after the Arbenzgovernment had nationalized most of United Fruit’s assets inGuatemala. Next, in 1960 and 1961, both the United States andBelgium engagedin independent operations topolitically neutral-ize the government of Patrice Lumumba in the Congo. Lumumbahad refused to allow Katanga, a copper-rich enclave controlledby the Belgian multinational Union Miniere, to secede and avoidtaxation and potential nationalization. In Cuba, the Castro gov-ernment nationalized all U.S. property in 1960, one year beforethe failed Bay of Pigs coup attempt in April 1961. Finally, theChilean nationalization of copper and other foreign-owned assetsbegan under the Frei government, but proposed compensationwas substantially lower and nationalizations more frequent afterthe Allende government came to power in late 1970. Allende wasinofficeless thanthreeyears beforehewas killedduringa couponSeptember 11, 1973. In Online Appendix A, we provide a more de-tailedsynopsis of each coup, focusing on the nature of the precoupregime, the motivations behind the expropriations, the foreignresponses, and the resolution of the coup.
The qualitative evidence on links between business and coupplanners is substantial. First, much of the early CIA leadershipwas recruited from Wall Street. A 1945 report on the CIA’s pre-cursor by Colonel Richard Park claimed that the “hiring andpromotion of senior officers rested not on merit but on an old boynetwork from Wall Street” (Weiner 2007). Second, there was di-rect contact between the companies that had been nationalizedandtheCIA. Forexample, at thetimeofthecoupplanningagainstArbenz, three high-ranking members of the executive branch ofgovernment had strong connections with the United Fruit Com-pany. AlanDulles, a formermemberof theboardofdirectors of theUnited Fruit Company, was director of the CIA. Thomas Dudley
at McK
eldon Library 34 on January 31, 2012
http://qje.oxfordjournals.org/D
ownloaded from
1380 QUARTERLY JOURNAL OF ECONOMICS
Cabot held at different times the positions of director of Interna-tional Security Affairs in the State Department and CEO of theUnited Fruit Company. His younger brother, John Moore Cabot,was secretary of Inter-American Affairs during much of the coupplanning in 1953 and 1954. Besides the fact that Anglo-Iranianwas a majority state-owned company, the company met with CIAagent (and later historian) Kermit Roosevelt, who alleged in his1954 history that the initial plan for the coupwas proposedby theAnglo-Iranian Oil Company. In Belgium, the royal court and thepowerful bank Societe Generale tied together a social and finan-cial network of colonial officials and businesses. De Witte writesthat “the incontrovertible political conclusion is that the politicalclass, including the [Belgian] court, had a direct material interestin the outcome of the Congocrisis” (De Witte 2001). Most directly,theministerof AfricanAffairs, a keyinstigatorandplannerof Op-erationBarracuda, Haroldd’Aspremont-Lydenwas thenephewofGobert d’Aspremont-Lyden who was an administrator for UnionMiniere. The Senate Church Committee reported that the CIAheld meetings with U.S. multinationals involved in Chile on aregular basis, even tothe point of ITT (whose boardincludedJohnMcCone, a formerdirectorof theCIA) notoriouslyofferingtheCIA$1 million to overthrow Allende’s government (Weiner 2007). Inshort, social links between the government officials responsiblefor the coups and financial interests are well documented. Secretplans for regime change could have easily made it into the ears offinancial actors who, even if not directly connected to the affectedcompanies, could arbitrage this information on the market.
Our results are consistent with the presence of both direct in-formation leakage between political decision makers andthe com-panies that stood to benefit, as well as indirect information flowstothemarket. Weareunabletoproducedefinitiveevidenceontheidentity of the traders or pinpoint the exact source of the informa-tion leakage.
III. DATA
We focused on the set of all CIA coups where (a) the CIAattempted to effect regime change, (b) the relevant planningdocuments have been declassified, and (c) the government hadexpropriated property from a publicly listed multinational. Thedetails of how we obtained a comprehensive list of coups, de-classified documents, and expropriations are described in Online
at McK
eldon Library 34 on January 31, 2012
http://qje.oxfordjournals.org/D
ownloaded from
COUPS, CORPORATIONS, AND CLASSIFIED INFORMATION 1381
Appendix B. We are left with five coups where all three of ourcriteria are satisfied: Iran, Guatemala, Congo, Cuba, and Chile.Online Appendix A provides detailed historical background foreach of these coups.
We first extract all of the coup authorization events from thetimelines based on the declassified planning documents. Theseevents are restricted to those where either a coup was explicitlyapproved by the head of a government or ministry (the presidentof the United States, prime minister of the United Kingdom, orthe Ministry for African Affairs in Belgium), the head of an intel-ligence agency (the CIA or MI6), or where US$1 million or morewas allocatedtotheoverthrowofaforeigngovernment. Inthecaseof Congo, we include the date of the assassination of Lumumba,which happened in secrecy and was not known publicly for closeto one month. Authorization events are coded as “good”(+1) or“bad”(−1) depending on whether they increase or decrease thelikelihood of a coup. Our selection and coding of authorizationevents is presented in Table II.
We also extract public events from the official timelines foruse as controls in some specifications. Publicevents are restricted
TABLE II
AUTHORIZATION EVENT SELECTION
Date Country Description Good Canceled
September 15, 1970 Chile Nixon Authorizes Anti-Allende Plan (Incl.Poss.Coup) Y NJanuary 28, 1971 Chile 40 Committee Appropriates $1.2 Million Y NOctober 26, 1972 Chile 40 Committee Appropriates $1.4 Million Y NAugust 20, 1973 Chile 40 Committee Appropriates $1 Million Y NAugust 18, 1960 Congo Eisenhower Endorses Lumumba’s Elimination Y YSeptember 12, 1960 Congo Belgian Operation Barracuda Begins Y YOctober 11, 1960 Congo Operation Barracuda Canceled N YDecember 5, 1960 Congo CIA Stops Operation N YJanuary 18, 1961 Congo Lumumba Secretly Killed Y NMarch 17, 1960 Cuba Eisenhower Approves Plan to Overthrow Castro Y NAugust 18, 1960 Cuba Eisenhower Approves $13 Million to Overthrow Castro Y NJanuary 30, 1961 Cuba Kennedy Authorizes Continuing Bay of Pigs Op Y NNovember 4, 1961 Cuba Operation Mongoose Planning Authorized Y YFebruary 26, 1962 Cuba Operation Mongoose Scaled Back N YOctober 30, 1962 Cuba Operation Mongoose Cancelled N YAugust 18, 1952 Guatemala DCIA Approves PBFortune (Coup to Overthrow Arbenz) Y YOctober 8, 1952 Guatemala PBFortune Halted N YDecember 9, 1953 Guatemala DCIA Approves PBSuccess (Coup to Overthrow Arbenz) Y NApril 19, 1954 Guatemala Full Approval Given to PBSuccess Y NJune 19, 1953 Iran CIA/MI6 Both Approve Coup Y NJuly 1, 1953 Iran Churchill Approves Coup Y NJuly 11, 1953 Iran Eisenhower Appoves Coup Y N
Notes. Date is the date of the event. Country is the target country of the coup attempt. Description givesa brief description of the event. Good is coded as Y if the event should raise the share value of the companyand N if the event should lower the share value of the company. Canceled is coded as Y if the operation wascancelled and N if it was executed; the 40 Committee was the subgroup of the executive branch NationalSecurity Council responsible for authorizing covert actions after 1964.
at McK
eldon Library 34 on January 31, 2012
http://qje.oxfordjournals.org/D
ownloaded from
1382 QUARTERLY JOURNAL OF ECONOMICS
to dates where company assets are nationalized or regime tran-sitions and consolidations occur. The public events are coded as“good”(+1) or “bad”(−1), where “good” events are those that arelikely to increase the stock price and “bad” events are ones thatare likely to cause a decline in the stock price. The public eventsandtheircodingare listedinOnlineAppendixTableAI. Table VIIlists the dates of the regime changes themselves.
In addition to the data on the events, we also construct adata set of daily stock returns for publicly traded companies thatwere expropriated by the regimes that were then overthrown bythe CIA. Using a variety of sources, also documented in OnlineAppendix A, we obtain the lists of companies expropriatedin eachcountry. For each of these companies, we obtain the amounts ex-propriated from various sources; we obtain daily stock marketdata either from CRSP or from archival sources. We define theexposure of a company to be the value of the assets expropriateddivided by the average market capitalization in the year prior tothe nationalizing regime coming into power. We also use market-level daily Fama-French four factors: excess return of the NYSE,high minus low(book toprice ratio), small minus big (market cap-italization), and momentum. For years prior to1962, we obtainedthe daily HML and SMB factor data series from Oliver Boguth,and we constructed the daily momentum factor ourselves. Post-1962 data on the factors come from Ken French’s website. Addi-tionally, we used a Perl script to generate a daily count of thenumber of New York Times articles mentioning both the name ofthe country and the country’s leader. Summary statistics of themain variables are presented in Table III.
IV. METHODOLOGY
Our main hypothesis is that authorization events result inan increase in the stock price of the affected company over thedays following the event. We consider cumulative abnormal re-turns after the authorization events. In contrast to public events,we expect stock price reactions to top-secret events to potentiallydiffuse slowly. Our benchmark specification estimates a 4-day re-turn starting at the event date, though we consider alternativespecifications ranging from 1 to 21 days. We employ two differentestimation strategies: a regression using the augmented Fama-French four-factor model, and a newset of distribution-free smallsampletests.
at McK
eldon Library 34 on January 31, 2012
http://qje.oxfordjournals.org/D
ownloaded from
COUPS, CORPORATIONS, AND CLASSIFIED INFORMATION 1383
TA
BL
EII
I
SU
MM
AR
YS
TA
TIS
TIC
S
Var
iabl
e
Com
pan
yC
oun
try
N4-
Dig
itS
ICM
ark
etC
apE
xpro
p.V
alu
eE
xpos
ure
Mea
n(R
awR
etu
rn)
SD
(Raw
Ret
urn
)V
olu
me
Dai
lyA
vg.
NY
TS
tori
e s
An
con
da
Co
Ch
ile
2224
3333
4.80
E+
083.
20E
+08
0.66
660.
0000
0.02
3424
298.
610.
5494
Bet
hle
hem
Ste
elC
orp
Ch
ile
2225
3312
9.79
E+
082.
50E
+07
0.02
550.
0002
0.01
7736
475.
60.
5494
Cer
roC
orp
Ch
ile
2224
1031
1.53
E+
081.
41E
+07
0.09
23−
0.00
010.
0231
1185
8.5
0.54
94G
ener
alT
ire
&R
ubr
Co
Ch
ile
2225
3011
3.29
E+
081.
20E
+07
0.03
65−
0.00
020.
0188
1451
4.7
0.54
94In
tern
atio
nal
Tel
&T
eleg
Cor
pC
hil
e22
2336
622.
57E
+09
1.07
E+
080.
0417
0.00
000.
0183
6193
9.7
0.55
01K
enn
ecot
tC
opp
erC
orp
Ch
ile
2225
3331
1.33
E+
092.
17E
+08
0.16
330.
0002
0.01
9431
554.
10.
5494
Un
ion
Min
iere
Con
go11
2410
211.
85E
+11
6.25
E+
100.
3379
−0.
0009
0.02
680.
8823
Am
eric
anS
uga
rR
efn
gC
oC
uba
2085
2061
5.84
E+
075.
52E
+07
0.94
520.
0007
0.01
6770
9.2
2.67
49C
anad
aD
ryC
orp
Cu
ba20
8820
904.
90E
+07
1.11
E+
060.
0227
0.00
030.
0127
1949
.12.
6733
Coc
aC
ola
Co
Cu
ba20
8720
906.
05E
+08
1.87
E+
070.
0310
0.00
050.
0115
2301
.32.
6592
Col
gate
Pal
mol
ive
Co
Cu
ba20
8728
412.
79E
+08
9.88
E+
060.
0354
0.00
060.
0167
3880
.82.
6740
Con
tin
enta
lC
anIn
cC
uba
2089
3411
5.55
E+
086.
07E
+06
0.01
09−
0.00
010.
0165
4590
.72.
6696
Fre
epor
tS
ulp
hu
rC
oC
uba
2089
1477
2.26
E+
086.
02E
+07
0.26
580.
0002
0.01
7127
30.5
2.67
25In
tern
atio
nal
Tel
&T
eleg
Cor
pC
uba
2087
3662
5.40
E+
088.
90E
+07
0.16
490.
0005
0.02
0611
711.
52.
6714
Lon
eS
tar
Cem
ent
Cor
pC
uba
2087
3272
2.52
E+
081.
69E
+07
0.06
720.
0001
0.01
6335
43.9
2.67
16S
wif
t&
Co
Cu
ba20
8820
112.
44E
+08
4.05
E+
060.
0166
0.00
000.
0127
2607
.22.
6738
Un
ited
Fru
itC
oC
uba
2088
2062
3.03
E+
085.
88E
+07
0.19
41−
0.00
020.
0165
7255
.92.
6733
Woo
lwor
thF
WC
oC
uba
2088
5331
5.58
E+
086.
26E
+06
0.01
120.
0002
0.01
0635
37.8
2.66
55U
nit
edF
ruit
Co
Gu
atem
ala
3469
120
5.31
E+
087.
83E
+07
0.14
750.
0001
0.01
1634
12.3
0.21
70A
ngl
o-Ir
ania
nIr
an23
9129
107.
46E
+09
2.31
E+
090.
3103
0.00
060.
0204
0.75
25
Not
es.S
um
mar
yst
atis
tics
byco
un
try
and
com
pan
yar
esh
own
over
the
even
tw
ind
ow.
Ngi
ves
the
nu
mbe
rof
obse
rvat
ion
sfo
rth
em
ajor
ity
ofli
sted
vari
able
sfo
ra
give
nco
mp
any
ina
give
nco
un
try;
inso
me
case
s,p
arti
cula
rva
riab
les
are
mis
sin
gfo
ra
few
day
sfo
ra
give
nco
mp
any/
cou
ntr
y.M
ark
etC
apis
the
aver
age
pri
ceti
mes
the
outs
tan
din
gsh
ares
star
tin
gtw
oye
ars
befo
reth
en
atio
nal
izin
gre
gim
eco
mes
top
ower
and
end
ing
one
year
befo
reth
en
atio
nal
izin
gre
gim
eco
mes
top
ower
.Exp
rop
riat
edV
alu
eis
the
dol
lar
amou
nt
ofth
eas
sets
that
wer
eex
pro
pri
ated
from
the
com
pan
yby
the
cou
pco
un
try
gove
rnm
ent.
Exp
osu
reis
the
rati
oof
nat
ion
aliz
edto
tota
las
sets
for
the
com
pan
y/co
un
try.
Raw
retu
rns
and
volu
me
are
atth
ed
aily
leve
l.D
aily
Ave
rage
NY
TS
tori
esar
ed
aily
cou
nts
ofar
ticl
esin
the
New
Yor
kT
imes
wh
ich
men
tion
both
aco
un
try
and
the
cou
ntr
y’s
lead
erby
nam
e.
at McK
eldon Library 34 on January 31, 2012
http://qje.oxfordjournals.org/D
ownloaded from
1384 QUARTERLY JOURNAL OF ECONOMICS
IV.A. Regression Method
Fortheregressionmethod, weregress acompany’s stockpricereturnonanindicatorforauthorizationevents interactedwiththecompany’s exposure. Wealsocontrol for fourFama-Frenchfactors(excess return of the NYSE, SMB, HML, and momentum):
(1) Rft = Xtβf + γcEft(k) + εft ;
here Rft is the one-day raw stock return for firm f between theclosing price at date t− 1 and the closing price at date t, and Xt isthe vector of factors. Eft(k) is a variable that takes on the value ofa company’s exposure for a k-day period beginning with an autho-rization day, and 0 otherwise. The average daily abnormal returnover the k-day period in country c is γc. The cumulative abnormalreturn is kγc.1 We consider values of k ranging from 1 to21. In ourmultiple country regressions, we report the mean of the country-
specificcoefficients∑
c γc
C , whereC is thetotal numberof countries.Our sample is the time period starting exactly one year before the
nationalizing regime comes to power until the day before the be-ginning of the coup. The standard error for the cumulative abnor-mal return is given by the maximum of robust standard errors,standard errors clustered on date, and standard errors clusteredon company.
IV.B. Small Sample Tests
One problem with the regression method as well as tradi-tional event studies is that the distribution of abnormal returnsis often non-normal, and the number of events is often small. As aresult, use of conventional standard errors may produce an incor-rect test size. We provide two nonparametric small sample testsbased on the sign and rank tests used in the literature. Unlikethe conventional rank and sign tests, however, we use “exact”distributions that do not rely on asymptotic justifications.
The standard rank and sign tests are motivated by the obser-vation that these test statistics converge much faster to a normaldistribution than the mean. Others have noted that the sign testhas an analogue to Fisher’s exact test, which uses the binomialdistribution to calculate a distribution-free test for significance,
1. Note that this is a standard approximation to (1 + γk)k − 1.
at McK
eldon Library 34 on January 31, 2012
http://qje.oxfordjournals.org/D
ownloaded from
COUPS, CORPORATIONS, AND CLASSIFIED INFORMATION 1385
which we also implement. We also note that the percentile, basedupon the rank, has a uniform distribution. This permits an anal-ogous distribution-free test for the average rank.
We begin by estimating a market model with the four factorsinan“estimationwindow”that is priortoanycoup-relatedevents.Ourestimationwindowis twocalendaryears in lengthandbeginsthree years before the nationalizing regime comes to power. Weestimate firm-specific cumulative abnormal returns (CARS) forfour-day windows starting with authorization dates. We weightthese CARs by relative company exposure within country andformcountry-portfoliospecificCARs. Theoverall CARtakes asim-ple average of returns over country-portfolios.
We first generalize the sign test by considering the number ofevents that have a k-day CAR greater than a given percentile p,where p is computed using k-day CARs in the estimation samplefrom country c. When cumulative abnormal returns are indepen-dentlydistributedacross countries andevents, theone-sidedprob-ability of getting mp or more abnormal returns above the pth is:
(2) 1−M∑
i=mp
(Mi
)
pi (1− p)M−i ,
where M is the total number of events. This is the p-value ofthe one-sided binomial sign test. Since the pth percentile returnis estimated based on a finite estimation sample, and multipleevents within the same country use the same estimated pth per-centile cutoff, this will induce a cross-event correlation in themeasured percentiles within countries. Therefore, besides calcu-lating the p-value analytically using Equation 2, we also followthe literature on randomization inference (Andrews 2003; Conleyand Taber 2011) and simulate our test statistic. First we drawTc percentiles from a uniform distribution, where Tc is the sizeof country c’s estimation window. We then draw Mc additionalreturns, where Mc is the number of events, from a uniform dis-tribution.2 We then estimate the pth percentile return from theTc draws. Next, we count the number of Mc draws above the pthpercentile of the Tc draws. We do this for all five countries andthen compute the average number of event returns above the pth
2. Both the binomial and the uniform tests can be shown to be independentof the distribution of the underlying returns for a wide class of continuous distri-bution functions.
at McK
eldon Library 34 on January 31, 2012
http://qje.oxfordjournals.org/D
ownloaded from
1386 QUARTERLY JOURNAL OF ECONOMICS
percentile, andrepeat this procedure 10,000 times toestimate thesimulated counterpart to Equation 2.
Finally, parallel to the binomial test, we construct an ana-logueoftheranktest (Corrado1989; Campbell, Lo, andMackinlay1997) exploiting the independence of events in our country port-foliosample toobtain exact inference. We rank each of our eventsrelative to the distribution of abnormal returns in the estimationwindow. We then convert the rank into a percentile. Noting that,fori.i.d. variables, percentile is uniformlydistributed, wecomputethe CDF for the sum of the percentiles of M independently anduniformly distributed random variables over the interval [0, 1].3
Without loss of generality, we assume that the mean percentilem ≥ 0.5. Given the symmetry of the cumulative distribution func-tion, the one-sided p-value of getting a percentile rank greaterthan m is then:
(3) 1−M∑
j=0
((−1)j (m− j)M 1(m ≥ j)
j!(M − j) !
)
We derive test statistics using the analytical equation fromEquation 3. However, similar to the binomial test, we also sim-ulate the distribution of average ranks. We report the modifiedsign and rank test results by country as well for the successfulcoups and the full sample. Finally, for the purpose of comparison,we also report asymptotic standard errors using the standard de-viations of returns in the estimation window (Campbell, Lo, andMackinlay 1997).
V. RESULTS
V.A. Baseline Results
In Table IV, we report the cumulative abnormal returns forauthorization events interacted with exposure over periods rang-ing from 1 to 16 days in length. We use (0, k − 1) to denote thek-day period beginning with the day of the event. We find clearevidencethat stockprices react positivelytoauthorizationevents.Row 1 of Table IV shows that in the pooled sample of all compa-nies, the average four-day stock price return for an authorization
3. This test was suggested, but not pursued, by Corrado (1989).
at McK
eldon Library 34 on January 31, 2012
http://qje.oxfordjournals.org/D
ownloaded from
COUPS, CORPORATIONS, AND CLASSIFIED INFORMATION 1387
TA
BL
EIV
EF
FE
CT
OF
SE
CR
ET
CO
UP
AU
TH
OR
IZA
TIO
NS
ON
ST
OC
KR
ET
UR
NS. M
AIN
EF
FE
CT
S: C
UM
UL
AT
IVE
AB
NO
RM
AL
RE
TU
RN
S
(0,0
)(0
,3)
(0,6
)(0
,9)
(0,1
2)(0
,15)
All
Cou
ps
0.04
350.
0938
0.09
900.
1055
0.12
040.
1342
(0.0
162)∗∗∗
(0.0
270)∗∗∗
(0.0
345)∗∗∗
(0.0
390)∗∗∗
(0.0
424)∗∗∗
(0.0
522)∗∗
2 215
722
157
2215
722
157
2215
722
157
Su
cces
sfu
lC
oup
s0.
0551
0.12
080.
1274
0.13
090.
1459
0.16
40(0
.020
1)∗∗∗
(0.0
336)∗∗∗
(0.0
425)∗∗∗
(0.0
481)∗∗∗
(0.0
523)∗∗∗
(0.0
647)∗∗
8555
8555
8555
8555
8555
8555
Can
cele
dC
oup
s0.
0729
0.13
410.
1414
0.13
590.
1564
0.19
71(0
.033
7)∗∗
(0.0
546)∗∗
(0.0
681)∗∗
(0.0
730)∗
(0.0
777)∗∗
(0.1
018)∗
1525
715
257
1525
715
257
1525
715
257
Ch
ile
−0.
0095
0.01
720.
0003
0.02
140.
0183
0.01
04(0
.006
6)(0
.027
4)(0
.037
3)(0
.049
1)(0
.051
0)(0
.062
0)60
9160
9160
9160
9160
9160
91C
ongo
0.16
670.
2270
0.20
140.
2429
0.22
830.
2581
(0.0
771)∗∗
(0.1
196)∗
(0.1
335)
(0.1
426)∗
(0.1
546)
(0.1
719)
421
421
421
421
421
421
at McK
eldon Library 34 on January 31, 2012
http://qje.oxfordjournals.org/D
ownloaded from
1388 QUARTERLY JOURNAL OF ECONOMICS
TA
BL
EIV
( CO
NT
INU
ED
)
(0,0
)(0
,3)
(0,6
)(0
,9)
(0,1
2)(0
,15)
Con
go-B
elgi
um
even
ts0.
2730
0.26
320.
3179
0.42
600.
3914
0.46
22(0
.079
4)∗∗∗
(0.1
895)
(0.1
972)
(0.2
029)∗∗
(0.2
182)∗
(0.2
260)∗∗
421
421
421
421
421
421
Cu
ba−
0.00
30−
0.01
41−
0.01
470.
0039
0.01
830.
0147
(0.0
079)
(0.0
125)
(0.0
178)
(0.0
202)
(0.0
222)
(0.0
263)
1360
213
602
1360
213
602
1360
213
602
Gu
atem
ala
0.04
910.
1650
0.20
490.
1365
0.20
110.
1859
(0.0
203)∗∗
(0.0
530)∗∗∗
(0.0
896)∗∗
(0.1
136)
(0.1
274)
(0.1
662)
1234
1234
1234
1234
1234
1234
Iran
0.01
440.
0739
0.10
300.
1229
0.13
590.
2017
(0.0
110)
(0.0
184)∗∗∗
(0.0
428)∗∗
(0.0
385)∗∗∗
(0.0
349)∗∗∗
(0.0
792)∗∗
809
809
809
809
809
809
Not
es.F
orsi
ngl
eco
un
try
regr
essi
ons,
the
rep
orte
dco
effic
ien
tis
onan
ind
icat
orfo
rau
thor
izat
ion
even
tsin
tera
cted
wit
hco
mp
any
exp
osu
re,m
uli
pli
edby
the
len
gth
ofth
ew
ind
ow.
Mu
lti-
cou
ntr
yre
gres
sion
sre
por
tth
em
ean
ofth
eco
un
try
coef
ficie
nts
.A
llre
gres
sion
sco
ntr
olfo
ran
inte
ract
ion
ofa
com
pan
yd
um
my
(or
cou
ntr
y-sp
ecifi
cco
mp
any
du
mm
yfo
rm
ult
i-co
un
try
regr
essi
ons)
wit
hth
efo
ur
Fam
a-F
ren
chfa
ctor
s.A
lld
ates
wh
ere
aco
mp
any
chan
ged
its
nam
eor
chan
ged
its
outs
tan
din
gsh
ares
bym
ore
than
5%w
ere
dro
pp
ed.
On
ed
ayp
rice
chan
ges
grea
ter
than
50%
inm
agn
itu
de
wer
ed
rop
ped
.“S
ucc
essf
ul
cou
ps”
excl
ud
esC
uba
.“C
ance
led
cou
ps”
only
use
sau
thor
izat
ion
san
dd
eau
thor
izat
ion
sof
cou
ps
that
wer
eev
entu
ally
can
cele
d.
Col
um
nn
um
bers
atth
eto
pin
par
enth
eses
den
ote
the
nu
mbe
rof
day
sbe
fore
and
afte
rth
eau
thor
izat
ion
sw
hic
har
ein
clu
ded
asp
art
ofth
ed
um
my
vari
able
for
the
auth
oriz
atio
nev
ent,
e.g.
,(0
,3)
refe
rsto
the
retu
rnbe
twee
nth
eev
ent
dat
ean
dth
ree
day
saf
ter
the
even
td
ate.
Sta
nd
ard
erro
rsre
por
ted
inp
aren
thes
esar
eth
em
axim
um
ofcl
ust
ered
byco
mp
any,
clu
ster
edby
dat
e,an
dro
bust
.Sta
tist
ical
sign
ifica
nce
at10
%,5
%,a
nd
1%le
vels
isd
enot
edby
*,**
,an
d**
*,re
spec
tive
ly.
at McK
eldon Library 34 on January 31, 2012
http://qje.oxfordjournals.org/D
ownloaded from
COUPS, CORPORATIONS, AND CLASSIFIED INFORMATION 1389
event is 9.4% with a standard error of 2.7%. This implies that ahypothetical company that had all its assets expropriated couldbe expected, on average, to experience roughly a 9.4% increase inits stock price within the four days following the secret authoriza-tion of a CIA coup. The cumulative abnormal returns are gener-ally significant at the 1% level for the all-country sample from 4through 13 days after the event. The abnormal returns continuetoincrease between days 4 and16 after the event, consistent withthe hypothesis that private information is incorporated intoassetprices with a delay.
In row 2, we restrict attention to the set of four successfulcoups (i.e., excluding Cuba), and the corresponding estimates areconsistently larger by around 25%–30%. The sample size dropssubstantially due to the large number of expropriated firms inCuba. In row 3, we restrict attention to the events that were au-thorizations (and deauthorizations) of coups that were later can-celed. The mean effect increases somewhat in magnitude (13.4%after 4 days), reaching a maximum of 19.7% at 10% significanceafter 16 days. We interpret the results on the canceled coups toprovideadditional evidencethat thestockpricereactions reflectedchanges in beliefs due to the authorizations themselves, and notthe expected coup or trends leading up to the coup.4
Rows 4–9 show the results separately for each country. ForChile, the effect is positive by the fourth day after the authoriza-tion event, but small andinsignificant. It alsostays small throughthelongerhorizons considered. Inrow5, weconsiderCongo, whichexhibits a large 16.7% effect on the day of an authorization event.The cumulative abnormal return increases to 22.7% after 4 daysand then stabilizes, becoming statistically insignificant after 10days. In row 6, we restrict attention to the events in the Congosample that were decisions made by Belgian officials, as the af-fected company was Belgian and the operation was independentof the United States. Effects in this sample are even larger, withan immediate 27.3% effect after the event, rising to a 5% signifi-cant 46.2% after 16 days.
Row 7 shows the results for Cuba. There are two operationsand thus two sets of events in Cuba. The first is the failed Bay
4. Although not reported in the table, if we further restrict attention to thedeauthorizationevents themselves, thestockpriceof a fullyexposedcompanyfallsby 11.7% within four days of a deauthorization, which further confirms this inter-pretation.
at McK
eldon Library 34 on January 31, 2012
http://qje.oxfordjournals.org/D
ownloaded from
1390 QUARTERLY JOURNAL OF ECONOMICS
of Pigs invasion. The second is Operation Mongoose, which wasstartedaftertheBayofPigs but was ultimatelycanceled. Morede-tails about theCubanoperations areavailableinOnlineAppendixA. There is virtually no effect in the Cuba subsample even after16 days and, though not reported in the tables, for both opera-tions considered individually. The qualitative evidence suggeststwo possible reasons for the absence of an effect in Cuba: (1) Dueto the high degree of public aggression from the United States to-ward Cuba, including numerous bombing missions, the coup wasalready commonly believed to be in planning and thus informa-tion about top-secret authorizations were not considered “news”by financial market actors.5 (2) Traders were pessimistic aboutsuccess, partiallyowingtoacombinationof incompetenceandlackof political commitment towardthe coupby the Kennedy adminis-tration. Thoughwearenot abletoconvincinglyreject eitherexpla-nation, we do provide additional evidence later that some tradersdid believe in the possibility of a successful Bay of Pigs operation.
Rows 8 and 9 show the results for Guatemala and Iran, re-spectively. Guatemala shows an immediate and significant 4.9%increase, which continues to grow to 16.5% after four days and20.5% after seven days, also significant at 5% confidence. Afterthis, the coefficient in the Guatemala subsample is not statisti-cally significant, although the point estimate generally remainslarge. In the Iran subsample, we do not see an immediate reac-tion to the event, but we do see a significant 7.4% effect after 4days, increasing to 10.3% after 7 days and continuing to increaseto 20.2% at 16 days, all significant at the 1% or 5% level. Overall,our country results shows that in the three out of the five coun-tries with statistically significant effects, the results were visibleand clear within four days. However, in all these cases, the ef-fects tended to grow over the following days, consistent with slowdiffusion of private information into asset prices.
The effects reported in Table IV are for a hypothetical com-pany that was fully nationalized. To obtain the average effect forthe sample of companies in a given country, we would need tomultiply the coefficient by the mean exposure for companies inthat country. The average exposure in the sample was 17.9%, sothe second column of Table IV implies that the cumulative returnin the sample companies was 1.6% after four days. As a specific
5. “WhenKennedyreads the[NYT]storyheexclaims that Castrodoesn’t needspies in the United States; all he has to do is read the newspaper” (Wyden 1979).
at McK
eldon Library 34 on January 31, 2012
http://qje.oxfordjournals.org/D
ownloaded from
COUPS, CORPORATIONS, AND CLASSIFIED INFORMATION 1391
example, Union Miniere had 33.8% of its overall assets exposed,which implies that the cumulative abnormal return in the Congosubsample was 7.6% after four days. Similarly, United Fruit had14.8% of its assets exposed, which implies a 2.4% return over fourdays. Finally, Anglo-Iranian had 31.0% of its assets nationalizedin Iran, andsothe impliedcumulative four-day increase followingan authorization event for that company was 2.3%.
Figure I provides graphical evidence, parallel with Table IV,on abnormal returns around an authorization event, with 95%confidence intervals shown. We compute cumulative abnormalreturns CAR(k) using the regression method aggregated acrossevents foreachofthe20 days priortoas well as followinganevent.Forthe20 days priortotheevent, weaggregatebackwardstarting
FIGURE I
Cumulative Abnormal Returns: All Countries
The thicker line represents the average of country-specific coefficients on anindicator for authorization events interacted with company exposure, mulipliedby the length of the window. The horizontal axis labels denote the number of daysbeforeoraftertheauthorizations whichareincludedas part ofthedummyvariablefor the authorization event, e.g., 4 refers tothe return between the event date andfourdays aftertheevent datewhile−4 refers tothereturnbetweenfourdays priorto the event date and the event date. All regressions control for an interaction ofa country-specific company dummy with the four Fama-French factors. All dateswhere a company changed its name or changed its outstanding shares by morethan5% weredropped. Onedaypricechanges greaterthan50% inmagnitudeweredropped. The thinner lines (and square symbols) represent the 95% confidenceintervals using standard errors that are the maximum of clustered by company,clustered by date, and robust.
at McK
eldon Library 34 on January 31, 2012
http://qje.oxfordjournals.org/D
ownloaded from
1392 QUARTERLY JOURNAL OF ECONOMICS
at the event date (date 0), so CAR(−k) is the cumulative abnor-mal return between dates −k and 0. For returns starting priorto date 0, we also include as a control an indicator for a 10-dayperiod after an authorization date, in the case when the eventsare sufficiently close together that cumulative returns prior toone authorization event includes returns that follow another au-thorization event. The only country where the windows overlapis Iran, and none of the other figures look different if we do notaccount for the overlap in CAR(k) and CAR(−k) windows whencumulating over days prior to the event. For our full sample, cu-mulative abnormal returns become significant at a 5% level onthe day of an event and remain significant. The rise over this pe-riod is generally monotonic until day 16 and seems to be perma-nent. Considering returns prior to the event date, however, theCAR(−k)’s shownotrends andare never significant. We concludethat there was no pre-existing trend in the stock price prior toan event, suggesting that the CIA did not authorize coups in re-sponse to proximate drops in the value of connected companiesor pre-existing political trends that would also be priced into thestock return. Figure II shows the CAR graphs separately by coun-try. As expected, individual country time paths are more impre-cise due to sample size limitations, with consistently significantresults only in Congo, Guatemala, and Iran. There is no evidenceof a persistent and significant pretrend in any of the individualcountries. Overall, the evidence on timing shows that authoriza-tion events led to positive asset price movements—usually withsome lag.
V.B. Robustness
Our benchmark specification (second column of Table IV)shows that abnormal returns were positive and significant in thefourdays followinganauthorizationevent. However, this couldbedue to downturns in the broad market, contemporaneous infor-mation about public events, or positive industry-specific shocks.To show that the positive abnormal returns reflect changes incompany-specific returns, we consider a number of robustnesschecks in Table V. All are estimated for the pooled sample, theset of successful coups, and separately by country. We computecumulative abnormal returns over a four-day period followingan authorization event. Except for the first and fifth columns,all specifications include the four Fama-French factors interacted
at McK
eldon Library 34 on January 31, 2012
http://qje.oxfordjournals.org/D
ownloaded from
COUPS, CORPORATIONS, AND CLASSIFIED INFORMATION 1393
FIGURE II
Cumulative Abnormal Returns by Country
The thicker line represents the coefficients on an indicator for authorizationevents interacted with company exposure, muliplied by the length of the window.The horizontal axis labels denote the number of days before or after the autho-rizations which are included as part of the dummy variable for the authorizationevent, e.g., 4 refers to the return between the event date and four days after theevent date while −4 refers to the return between four days prior to the eventdate and the event date. All regressions control for an interaction of a companydummy with the four Fama-French factors. All dates where a company changedits name or changed its outstanding shares by more than 5% were dropped. Oneday price changes greater than 50% in magnitude were dropped. The thinner lines(andsquaresymbols)represent the95% confidenceintervals usingstandarderrorsthat are the maximum of clustered by company, clustered by date, and robust.
at McK
eldon Library 34 on January 31, 2012
http://qje.oxfordjournals.org/D
ownloaded from
1394 QUARTERLY JOURNAL OF ECONOMICS
TA
BL
EV
RO
BU
ST
NE
SS
Raw
Pu
blic
No
NY
TN
ews
Tre
nd
Mar
ket
Mat
ched
Log
Ret
urn
sE
ven
ts/N
YT
Su
bsam
ple
Con
trol
sP
lace
boP
lace
boV
olu
me
All
Cou
ps
0.09
470.
0723
0.12
490.
0989
0.00
020.
0068
19.0
429
(0.0
282)∗∗∗
(0.0
222)∗∗∗
(0.0
137)∗∗∗
(0.0
034)∗∗∗
(0.0
011)
(0.0
216)
(2.2
102)∗∗∗
2215
722
157
7123
2215
722
157
1723
920
895
Su
cces
sfu
l0.
1210
0.09
390.
1153
0.12
59−
0.00
130.
0111
26.4
944
Cou
ps
(0.0
350)∗∗∗
(0.0
275)∗∗∗
(0.0
332)∗∗∗
(0.0
372)∗∗∗
(0.0
082)
(0.0
268)
(3.3
202)∗∗∗
8555
8555
5224
8555
8555
6670
7324
Ch
ile
0.03
650.
0191
0.10
060.
0243
0.01
54−
0.01
4920
.497
0(0
.037
1)(0
.027
9)(0
.076
5)(0
.031
9)(0
.013
6)(0
.031
7)(0
.753
4)∗∗∗
6091
6091
3530
6091
6091
4764
6091
Con
go0.
2274
0.12
02–
0.25
32−
0.00
67−
0.02
45–
(0.1
180)∗
( 0.0
909)
–(0
.128
2)∗∗
( 0.0
133)
(0.0
216)
–42
142
1–
421
421
322
–
at McK
eldon Library 34 on January 31, 2012
http://qje.oxfordjournals.org/D
ownloaded from
COUPS, CORPORATIONS, AND CLASSIFIED INFORMATION 1395
TA
BL
EV
(CO
NT
INU
ED
)
Raw
Pu
blic
No
NY
TN
ews
Tre
nd
Mar
ket
Mat
ched
Log
Ret
urn
sE
ven
ts/N
YT
Su
bsam
ple
Con
trol
sP
lace
boP
lace
boV
olu
me
Cu
ba−
0.01
03−
0.01
540.
0276
−0.
0098
0.00
66−
0.00
854.
1386
(0.0
138)
(0.0
124)
(0.0
365)
(0.0
144)
(0.0
088)
(0.0
145)
(2.5
058)∗
1360
213
602
1899
1360
213
602
1056
913
571
Gu
atem
ala
0.13
940.
1648
0.13
730.
1909
−0.
0311
0.02
5532
.439
1(0
.062
8)∗∗
(0.0
530)∗∗∗
(0.0
603)∗∗
(0.0
621)∗∗∗
(0.0
224)
(0.0
916)
(12.
5956
)∗∗
1234
1234
1068
1234
1234
965
1233
Iran
0.08
060.
0738
0.10
610.
0359
0.01
710.
0528
–(0
.018
9)∗∗∗
(0.0
189)∗∗∗
(0.0
137)∗∗∗
(0.0
305)
(0.0
146)
(0.0
400)
–80
980
939
880
980
961
9–
Not
es.
Est
imat
esar
eon
(0,3
)re
turn
s.F
orsi
ngl
eco
un
try
regr
essi
ons,
the
rep
orte
dco
effic
ien
tis
onan
ind
icat
orfo
rau
thor
izat
ion
even
tsin
tera
cted
wit
hco
mp
any
exp
osu
re,
mu
lip
lied
byth
ele
ngt
hof
the
win
dow
(i.e
.,4)
.Mu
ltic
oun
try
regr
essi
ons
rep
ort
the
mea
nof
the
cou
ntr
yco
effic
ien
ts.E
xcep
tfo
rth
e“R
awre
turn
s”an
d“M
ark
etP
lace
bo”
spec
ifica
tion
s,re
gres
sion
sco
ntr
olfo
ran
inte
ract
ion
ofa
com
pan
yd
um
my
(or
cou
ntr
y-sp
ecifi
cco
mp
any
du
mm
yfo
rm
ult
i-co
un
try
regr
essi
ons)
wit
hth
efo
ur
Fam
a-F
ren
chfa
ctor
s.A
lld
ates
wh
ere
aco
mp
any
chan
ged
its
nam
eor
chan
ged
its
outs
tan
din
gsh
ares
bym
ore
than
5%w
ere
dro
pp
ed.
On
e-d
ayp
rice
chan
ges
grea
ter
than
50%
inm
agn
itu
de
wer
ed
rop
ped
.“S
ucc
essf
ul
cou
ps”
excl
ud
esC
uba
.P
ubl
icin
form
atio
nco
ntr
ols
incl
ud
e(a
)an
exp
osu
re-i
nte
ract
edco
un
try
spec
ific
effe
ctof
the
nu
mbe
rof
New
Yor
kT
imes
arti
cles
men
tion
ing
aco
un
try
and
its
lead
erby
nam
ean
d(b
)co
un
try-
spec
ific
inte
ract
ion
betw
een
pu
blic
even
td
um
mie
san
dex
pos
ure
.No
NY
Tco
lum
nd
rop
sal
lobs
erva
tion
sw
ith
any
New
Yor
kT
imes
arti
cles
men
tion
ing
aco
un
try
and
its
lead
erby
nam
eon
that
dat
e.“T
ren
dC
ontr
ols”
con
trol
for
loca
ltr
end
sby
incl
ud
ing
anad
dit
ion
ald
um
my
ina
20-d
aysy
mm
etri
cw
ind
owar
oun
dea
chau
thor
izat
ion
dat
e.“M
ark
etP
lace
bo”
regr
esse
sth
eN
YS
Ere
turn
onth
eex
pos
ure
-in
tera
cted
even
td
ates
.“M
atch
edP
lace
bo”
rep
lace
sea
chco
mp
any’
sst
ock
retu
rnw
ith
that
ofth
eco
mp
any
wit
hth
ecl
oses
tm
ark
etca
pit
aliz
atio
n,f
acto
rlo
adin
gs,a
nd
mea
nan
dst
and
ard
dev
iati
onof
retu
rns
wit
hin
the
sam
e3-
dig
itS
ICco
de.
“Log
Vol
um
e”ru
ns
the
base
lin
esp
ecifi
cati
onw
ith
the
log
ofvo
lum
eas
the
dep
end
ent
vari
able
.Sta
nd
ard
erro
rsre
por
ted
inp
aren
thes
esar
eth
em
axim
um
ofcl
ust
ered
byco
mp
any,
clu
ster
edby
dat
e,an
dro
bust
.Sta
tist
ical
sign
ifica
nce
at10
%,5
%,a
nd
1%le
vels
isd
enot
edby
*,**
,an
d**
*,re
spec
tive
ly.
at McK
eldon Library 34 on January 31, 2012
http://qje.oxfordjournals.org/D
ownloaded from
1396 QUARTERLY JOURNAL OF ECONOMICS
with a company dummy (or country-specific company dummiesfor multi-country regressions) as controls. As in Table IV, we re-port the coefficient on the authorization dummy interacted withthe company’s exposure, multiplied by the number of days in thewindow(four in this case); multicountry estimates average the co-efficients across the countries.
First, we regress raw returns, unadjusted by any of the mar-ket factors, on our authorization events. We confirm that thecumulative abnormal return effects were due to increases in theaffectedcompany’s stockprices, andnot duetochanges inmarket-level movements. The first column of Table V shows a four-daycumulative abnormal return of 9.5%, virtually identical to ourbenchmark specification.
Top-secret decisions to overthrow foreign governments mayhave coincided with public events in the targeted countries.This could bias our estimates, reflecting the effect of publicnews rather than private information. In the second, we con-trol for the number of articles in the New York Times men-tioning both the country and the country leader by name,as well as other public events; these other events are na-tionalizations of foreign-owned property as well as electoraltransitions and consolidations which are also mentioned inthe timelines. Public events are listed in Online AppendixTable AI. We multiply these measures with company exposureand the country dummies, and include them as controls in ourmain specification. The coefficient in the pooled sample is onlyslightly smaller than the one in the main specification, and stillshows a 7.2% four-day return which is significant at the 1% level.In the third column we drop all dates where the New York Timeshad at least one article mentioning both the country and theleader by name (Meulbroek 1992). Since most days have at leastone political article about the coup countries, we lose over two-thirds of our sample in this specification, making this a strongtest. However, our effect actually becomes stronger despite thecountrywiththelargest baselineeffect, Congo, droppingout of thesample. The mean effect in the pooled country sample is a 12.5%return over four days and still significant at the 1% level. Congois very prominently covered in the news, and hence does not haveany events that are not contemporaneous with some New YorkTimes coverage. Though all the countries lose observations fromthe sample restrictions, the estimates for Chile andIran are actu-ally larger than in the baseline specification, and the coefficients
at McK
eldon Library 34 on January 31, 2012
http://qje.oxfordjournals.org/D
ownloaded from
COUPS, CORPORATIONS, AND CLASSIFIED INFORMATION 1397
for Guatemala and Iran are still significant at least at the 5%confidencelevel. Cuba onlyhas oneauthorizationdatethat has nocontemporaneous New York Times articles about Cuba and Cas-tro, reflecting the extensive leakage of the Bay of Pigs operationas well as general news interest in Cuba over the sample period.The scaling back of the second operation, Mongoose, on Febru-ary 2, 1962, does indeed fall on a news-free day. Although notsignificant, the positive and relatively larger coefficient on thissubsample is consistent with our interpretation that secret deau-thorizations docause decreases in stock prices when they actuallyconstitute “news.”
One potential explanation for our findings is price momen-tum around the authorization dates. This may either reflect pre-existing information flows or trading activities unrelated to coupplanning. We include a control that interacts the exposure mea-sure with a dummy that is equal to 1 in a 20-day window aroundeach authorization event. This specification tests whether the ab-normal returns are higher in the 4 days right after an authoriza-tion than in the average of the 20-day period surrounding eachauthorization event. The fourth column of Table V shows that thefour-day abnormal return is 9.9%, actually slightly higher thanour benchmark, and statistically significant at the 1% level. Pre-existing price trends do not explain our results.
We also consider two placebos. In the fifth column of Table Vwe regress NYSE index returns on our private information vari-able, omittingtheotherthreefactors. Ourpooledestimateis equalto 0.02% and is insignificant. None of the country-specific regres-sions are significant at the 10% level either. In the sixth columnof Table V, we use daily stock returns from a matched company,where the match is constructed by taking the company closestin the Mahalanobis metric (constructed from market capitaliza-tion and market beta) within each three-digit industry code, sub-ject to having data available for all of the authorization dates.The matched companies are listed in Online Appendix Table AII.This placebo is also insignificant in the pooled sample as well asall the subsamples, suggesting that our effects are not driven byindustry-specific shocks.
Finally, we consider the effect of authorizations on the log oftrading volumes for the set of countries for which data is avail-able. In both the pooled samples as well as the individual coun-try regressions, our event variable is positive and significant.This is true even in Chile and Cuba, where the effect on returns
at McK
eldon Library 34 on January 31, 2012
http://qje.oxfordjournals.org/D
ownloaded from
1398 QUARTERLY JOURNAL OF ECONOMICS
was insignificant. The finding of increased trading in the fourdays including and just after authorization days is consistentwith theoretical predictions of heterogeneous belief models (Wang1994) of stock trading as well as prior empirical work on the vol-ume impacts of insider trading (Cornell and Sirri 1992).
V.C. Time-Shifted Placebos
As additional evidence that our effects are not an artifact ofthe data, we reestimate our main specification on a set of placebodates. We take our 4-day cumulative abnormal returns and shiftour authorization events forwards as well as backward by 5, 10,15, 20, 25, 30, 35, and 40 days. For an s day shift, we estimate:
(4) Rft = Xtβf + γc,sEft+s(4)+ εft
As in our baseline specification, we report the mean cumulativefour-day return across countries:
∑c γc,s
C . We exclude all days withother authorizations, public events, or that occur during the coupitself. We graph our estimates against the number of days shiftedin Figure III.
Out ofthe19 time-shiftedregressions, γs is onlysignificant fors = 0, our benchmark specification with cumulative abnormal re-turnof approximately9.38% fora fullyexposedcompany, whichissignificant at the 1% level. None of the 18 other dates have a mag-nitude above 4% and none of them are significant at even the 10%level. The placeboestimates reinforce that our baseline estimatesare not due to local serial correlation in returns. The pattern ofno abnormal returns before a decision, sizable abnormal returnsjust after a decision, and smaller possible abnormal returns inthe medium run after a decision is consistent with our hypothe-sis of secret authorization events causing an increase in the stockprice.
V.D. Small Sample Tests
In Table VI, we present the results from our small sam-ple tests. First, we present the four day CARs of countryportfolios, based on out-of-sample estimates. The CARs here rep-resent the actual (exposure-weighted) change in stock prices foraffected companies in a given sample, while the regression coeffi-cients represent the effects for a hypothetical company that wasfullyexposed. Forcomparability, theregressioncoefficients would
at McK
eldon Library 34 on January 31, 2012
http://qje.oxfordjournals.org/D
ownloaded from
COUPS, CORPORATIONS, AND CLASSIFIED INFORMATION 1399
FIGURE III
Time-Shifted Placebos
The thick line plots the average of country-specific coefficients for a regres-sion of daily stock returns on an indicator for authorization events interactedwithcompany exposure and multiplied by the four day window including and after anauthorizationevent. Thehorizontal axis labels denotethenumberofdays bywhichwe shift the authorization date, e.g., 20 represents the four day return if we shiftthe authorization day forward by 20 days, while−20 represents a four day returnif we shift the authorization date backwards by 20 days. All regressions control foran interaction of a country-specific company dummy with the four Fama-Frenchfactors. All dates where a company changed its name or changed its outstandingshares by more than 5% were dropped. One day price changes greater than 50%in magnitude were dropped. The dashed lines represent the 95% confidence inter-vals using standard errors that are the maximum of standard errors clustered bycompany, clustered by date, and robust.
need to be multiplied by mean exposure levels, although the com-parability is inexact due to how exposures are treated in the twocases. The results are listedin row1 of Table VI. For the full sam-ple, theaveragefour-dayweightedCAR was 2.6%. Theestimateisstatistically significant at the 1% level using asymptoticstandarderrors.
Turning to our small sample tests, we find that 18 out ofthe 22 events in the full country sample have returns greaterthan the median return in the “estimation window” (i.e., the yearprior to any nationalization event), producing a one-sided proba-bility value under the null hypothesis of 0.35%.6 Thirteen of thoseevents have returns above the 80th percentile, which would oc-cur by chance alone with probability less than 0.02%. Eight of theevents have returns greater than the 90th percentile, which has
6. Inthetext, wereport thehigherof theanalytical andsimulatedprobabilityvalues. Both are reported in the table. All reported probability values are one-sided.
at McK
eldon Library 34 on January 31, 2012
http://qje.oxfordjournals.org/D
ownloaded from
1400 QUARTERLY JOURNAL OF ECONOMICS
TA
BL
EV
I
SM
AL
LS
AM
PL
ET
ES
TS
5C
oun
try
Su
cces
sfu
lC
oup
sC
hil
eC
ongo
Cu
baG
uat
emal
aIr
an
4D
ayC
AR
0.02
620.
0393
0.01
890.
0768
−0.
0086
0.02
390.
0243
Asy
mp
toti
cS
tan
dar
dE
rror
(0.0
030)∗∗∗
( 0.0
039)∗∗∗
( 0.0
149)
(0.0
195)∗∗∗
( 0.0
076)
(0.0
093)∗∗
( 0.0
165)
Nu
mbe
rof
2216
45
64
3E
ven
tsB
inom
ial
Sig
nT
est
Nu
mbe
rA
bove
Med
ian
1815
35
34
3M
edia
nP
-Val
ue:
0.00
22∗∗∗
0.00
03∗∗∗
0.31
250.
0313∗∗
0.65
630.
0625∗
0.12
50A
nal
ytic
alP
-Val
ue:
0.00
35∗∗∗
0 .00
06∗∗∗
0 .32
940.
0355∗∗
0 .66
020.
0688∗
0 .13
57S
imu
late
dN
um
ber
Abo
ve13
122
51
32
80th
Per
cen
tile
P-V
alu
e:0.
0001∗∗∗
0 .00
00∗∗∗
0 .18
080.
0003∗∗∗
0 .73
950.
0272∗∗
0 .10
40A
nal
ytic
alP
-Val
ue:
0.00
02∗∗
0.00
00∗∗∗
0.19
210.
0005∗∗∗
0.74
030.
0314∗∗
0.10
33S
imu
late
dN
um
ber
Abo
ve8
82
30
30
90th
Per
cen
tile
at McK
eldon Library 34 on January 31, 2012
http://qje.oxfordjournals.org/D
ownloaded from
COUPS, CORPORATIONS, AND CLASSIFIED INFORMATION 1401
TA
BL
EV
I
( CO
NT
INU
ED
)
5C
oun
try
Su
cces
sfu
lC
oup
sC
hil
eC
ongo
Cu
baG
uat
emal
aIr
an
P-V
alu
e:0.
0009∗∗∗
0 .00
01∗∗∗
0 .05
23∗
0 .00
856∗∗∗
1 .00
000.
0037∗∗∗
1 .00
00A
nal
ytic
alP
-Val
ue:
0.00
11∗∗
0 .00
03∗∗∗
0 .05
02∗
0 .01
26∗∗
1 .00
000.
0059∗∗∗
1 .00
00S
imu
late
dU
nif
orm
Ran
kT
est
Mea
nR
ank
0.74
400.
8195
0.64
170.
9350
0.44
180.
8803
0.82
11P
-Val
ue:
0.00
00∗∗∗
0.00
00∗∗∗
0.17
000.
0000∗∗∗
0.68
520.
0022∗∗∗
0.02
57∗∗
An
alyt
ical
P-V
alu
e:0.
0006∗∗∗
0 .00
00∗∗∗
0 .17
660.
0001∗∗∗
0 .69
520.
0033∗∗∗
0 .02
61∗∗
Sim
ula
ted
Not
es.T
his
tabl
ere
por
ts4
Day
Cu
mu
lati
veA
bnor
mal
Ret
urn
su
sin
g(e
xpos
ure
wei
ghte
d)
com
pan
yp
ortf
olio
sfo
rin
div
idu
alco
un
trie
s.M
ult
icou
ntr
yes
tim
ates
rep
ort
aver
ages
ofco
un
try
por
tofo
lio
retu
rns.
Asy
mp
toti
cst
and
ard
erro
ris
com
pu
ted
usi
ng
stan
dar
dd
evia
tion
sof
retu
rns
inth
ees
tim
atio
nsa
mp
le.
*,**
,an
d**
*d
enot
est
atis
tica
lsi
gnifi
can
ceu
sin
gas
ymp
toti
cin
fere
nce
atth
e10
%,5
%,a
nd
1%le
vels
,res
pec
tive
ly.“
Su
cces
sfu
lC
oup
s”ex
clu
des
Cu
ba.
For
the
Bin
omia
lS
ign
Tes
t,“N
um
ber
abov
eth
em
edia
n”
(an
d80
than
d90
thp
erce
nti
les)
rep
orts
the
nu
mbe
rof
4-d
ayev
ents
abov
eth
em
edia
n(a
nd
80th
and
90th
per
cen
tile
s)of
the
abn
orm
alre
turn
dis
trib
uti
onin
the
esti
mat
ion
win
dow
.“P
-Val
ue:
An
alyt
ical
”re
por
tsth
eas
soci
ated
P-V
alu
eu
sin
gth
eB
inom
ial
Dis
trib
uti
onto
give
the
pro
babi
lity
ofh
avin
gat
leas
tX
nu
mbe
rof
even
tsab
ove
the
cuto
ff(m
edia
nor
80th
or90
thp
erce
nti
le).
“P-V
alu
e:S
imu
late
d”
rep
orts
the
p-v
alu
efo
ra
sim
ula
ted
dis
trib
uti
onof
hav
ing
atle
ast
Xn
um
ber
ofev
ents
abov
eth
ecu
toff
(med
ian
or80
thp
erce
nti
leor
90th
per
cen
tile
)ou
tof
Yto
tal
even
ts,ac
cou
nti
ng
for
the
cuto
ffva
lue
bein
ges
tim
ated
usi
ng
the
actu
aln
um
ber
ofd
ays
inth
ees
tim
atio
nsa
mp
le.
For
the
Un
ifor
mR
ank
Tes
t,“M
ean
ran
k”
isth
eav
erag
ep
erce
nti
lera
nk
ofab
nor
mal
retu
rns
for
even
tsre
lati
veto
the
esti
mat
ion
win
dow
.“P
-Val
ue:
An
alyt
ical
”u
ses
the
un
ifor
md
istr
ibu
tion
toca
lcu
late
the
pro
babi
lity
ofh
avin
gan
aver
age
ran
kof
Kev
ents
grea
ter
than
oreq
ual
toM
.“P
-Val
ue:
Sim
ula
ted
”re
por
tsth
ep
-val
ue
for
asi
mu
late
dd
istr
ibu
tion
ofh
avin
gan
aver
age
ofK
even
tsh
avin
gra
nk
grea
ter
than
oreq
ual
toM
,acc
oun
tin
gfo
rth
era
nk
sbe
ing
esti
mat
edu
sin
gth
eac
tual
nu
mbe
rof
day
sin
the
esti
mat
ion
sam
ple
.
at McK
eldon Library 34 on January 31, 2012
http://qje.oxfordjournals.org/D
ownloaded from
1402 QUARTERLY JOURNAL OF ECONOMICS
an associated probability value of 0.11% under the null. Finally,the average rank of all 22 events is 0.74, which would be obtainedby chance with a probability less than 0.06%. When we con-sider the set of four successful coups, the conclusion is strength-ened. The probability values associated with the uniform ranktest, as well as the binomial sign tests (for 50th, 80th, and 90thpercentiles) are all under 0.1%.
Due to small sample size, the individual country tests havelow power and thus p-values are larger. Congo and Guatemalaconsistently produce probability values under 10% for all thetests, andsmallerformost. Iranproduces probabilityvalues rang-ing between 3% and 14% except for the 90th percentile, whereasChile ranges from 5% to33%. Finally, consistent with our results,Cuba shows no systematic increase in returns following autho-rization events. For example, only three out of the six events showpositive returns, and the rest are negative.
Our results also show heterogeneity across events. Thoughthere does not seem to be a substantial reaction to a few events,most showpositive reactions. Andmany showreactions that werevery strong, as exemplified by the fact that 8 out of 22 events areabove the 90th percentile in returns.
Overall, our modified sign and rank tests provide strong evi-dence that the four-day returns after authorization events are, onaverage, highly statistically significant, and our conclusions arenot driven by the size of our sample and non-normal distributionof returns. Also, they show us that there are reactions to someevents and not to others. However, when there is a reaction, theeffect is strong and unmistakable.
VI. ASSESSING THE GAINS FROM COUPS
We also estimate abnormal returns for the coup attemptsthemselves using our main specification. We do this for two rea-sons. First, we want to test if these companies were affected bythe actual coup attempts, confirming that companies were ben-efiting from the anticipated regime change. Second, we want tocompare the direct effect of the coup itself tothe total net rise dueto precoup authorizations.
We look at two estimates of the effect of the coup: abnormalreturns duringthecoupwindowandabnormal returns onthefirstday of the new regime. We define the coup window as the periodfrom and including the first day of the coup to and including the
at McK
eldon Library 34 on January 31, 2012
http://qje.oxfordjournals.org/D
ownloaded from
COUPS, CORPORATIONS, AND CLASSIFIED INFORMATION 1403
first day of the newregime (the last day of the coupattempt in thecase of Cuba). These dates are listed in Table VII.
Over the duration of the coup, the average cumulative returnacross countries was 12.1%. The result is slightly higher at 13.4%when we restrict attention tothe successful coups. The first day ofthe new government measure is slightly lower for both the full aswell as successful coups samples at 10.0% and11.8%, respectively.
The individual country estimates are also relatively simi-lar across the two measures for most of our sample. Chile’s andCongo’s coups are both one-day events, and so the effect is identi-cal across measures: 6.1% andsignificant at the 5% level for Chileand 8.7% and insignificant at conventional levels for Congo. Theeffects for Cuba are near −5% for both measures. The first dayof the new government effect is significant at the 1% level, rein-forcing that there is belief in the possibility of a successful coup inCuba.7 The coupwindoweffect is larger for Iran than the first dayof the new government. The coup window effect, 18.8%, is signif-icant at the 10% level; the first day of the new government effectis substantially smaller at 7.0%. For Guatemala, the sign actuallyflips. The coupwindoweffect for Guatemala is negative andsome-what sizable. The first day of the newgovernment effect, however,is quite a bit larger and positive in sign. The two numbers are−10.3% and 22.7%, the latter number being statistically signifi-cant at the1% level. Weattributethestockprice fall overthecoupwindowtothe fact that the junta which initially took power whenArbenz resigned did not support the return of assets to UnitedFruit. Further exacerbating the uncertainty around United Fruitassets, the 11-days following Arbenz’s resignation saw four in-terim governments come to power. Finally, the candidate backedby the CIA, Castillo Armas, took power (Gleijeses 1991). De-spite the uncertainty, Armas eventually returned United Fruitassets.
We now compare the magnitudes of the net authorizationevents to the coup event effects. We use the country-specific 13-day CARs to compute the value per authorization for each coun-try. The longer horizon return is used to capture the full assetprice change due to a leaked authorization. The total rise in thestockpriceduetoauthorizations is thenjust oneplus thereturnto
7. In a prior version of the article, we also included an estimate of the returnon the first day of the coup. For Cuba, the estimate was positive and significant,reinforcingtheviewthat sometraders thought that a successful coupwas possible.
at McK
eldon Library 34 on January 31, 2012
http://qje.oxfordjournals.org/D
ownloaded from
1404 QUARTERLY JOURNAL OF ECONOMICS
TA
BL
EV
II
GA
INS
FR
OM
CO
UP
AN
DA
UT
HO
RIZ
AT
ION
EV
EN
TS
Cou
pB
egin
sC
oup
En
ds
Cou
pW
ind
ow
Fir
stD
ayof
New
Gov
ern
-m
ent
12D
ayA
uth
.E
ffec
t
Tot
alG
ain
from
Au
th.
Eve
nts
Tot
alG
ain
from
Au
than
dF
irst
Day
New
Gov
Rel
ativ
eG
ain
from
Au
th.
Eve
nts
All
0.12
110.
1004
0.12
040.
3136
0.44
550.
7575
(0.0
463)∗∗∗
(0.0
259)∗∗∗
2217
322
165
Top
40.
1335
0.11
790.
1459
0.42
480.
5928
0.78
28(0
.060
3)∗∗
(0.0
419)∗∗∗
8 571
8563
Ch
ile
9/11
/197
39/
11/1
973
0.06
130.
0613
0.01
830.
0750
0.14
100.
5503
(0.0
250)∗∗
( 0.0
250)∗∗
6 097
6097
Con
go2/
5/19
612/
5/19
610.
0869
0.08
690.
2283
0.22
830.
3350
0.72
42(0
.094
7)(0
.094
7)42
142
1
at McK
eldon Library 34 on January 31, 2012
http://qje.oxfordjournals.org/D
ownloaded from
COUPS, CORPORATIONS, AND CLASSIFIED INFORMATION 1405
TA
BL
EV
II
( CO
NT
INU
ED
)
Tot
alR
elat
ive
Fir
stD
ay12
Day
Tot
alG
ain
Gai
nfr
omG
ain
Fro
mC
oup
Cou
pC
oup
ofN
ewA
uth
.fr
omA
uth
.A
uth
and
Fir
stD
ayA
uth
.B
egin
sE
nd
sW
ind
owG
over
nm
ent
Eff
ect
Eve
nts
New
Gov
Eve
nts
Cu
ba4/
15/1
961
4/20
/196
1−
0.04
45−
0.05
460.
0183
0.03
70−
0.01
96−
2.10
47(0
.028
3)(0
.014
1)∗∗∗
1360
213
602
Gu
atem
ala
6/19
/195
46/
28/1
954
−0.
1030
0.22
740.
2011
0.44
260.
7706
0.66
06(0
.173
7)(0
.070
4)∗∗∗
1235
Iran
8/15
/195
38/
20/1
953
0.18
750.
0703
0.13
590.
4657
0.56
860.
8689
(0.1
054)∗
(0.0
526)
813
810
Not
es.
For
sin
gle-
cou
ntr
yre
gres
sion
s,th
ere
por
ted
coef
ficie
nt
ison
anin
dic
ator
for
the
rele
van
tco
up
per
iod
inte
ract
edw
ith
com
pan
yex
pos
ure
,m
uli
pli
edby
the
len
gth
ofth
ere
leva
nt
cou
pp
erio
d.M
ult
icou
ntr
yre
gres
sion
sre
por
tth
em
ean
ofth
eco
un
try
coef
ficie
nts
.T
he
cou
pw
ind
owis
defi
ned
asth
efu
llle
ngt
hof
tim
ebe
twee
nth
ebe
gin
nin
gan
dth
een
dof
the
cou
p.
Th
efi
rst
day
ofth
en
ewgo
vern
men
tis
the
firs
td
ayaf
ter
the
cou
p(i
nth
eca
seof
Cu
ba,
this
isth
efi
rst
day
afte
rth
een
dof
the
inva
sion
).A
llre
gres
sion
sco
ntr
olfo
ran
inte
ract
ion
ofa
com
pan
yd
um
my
(or
cou
ntr
y-sp
ecifi
cco
mp
any
du
mm
yfo
rm
ult
icou
ntr
yre
gres
sion
s)w
ith
the
fou
rF
ama-
Fre
nch
fact
ors.
All
dat
esw
her
ea
com
pan
ych
ange
dit
sn
ame
orch
ange
dit
sou
tsta
nd
ing
shar
esby
mor
eth
an5%
wer
ed
rop
ped
.Sin
ceC
uba
’sco
up
was
un
succ
essf
ul,
the
stoc
kp
rice
chan
ges
are
neg
ativ
e.S
tan
dar
der
rors
rep
orte
din
par
enth
eses
are
the
max
imu
mof
thos
ecl
ust
ered
byco
mp
any,
clu
ster
edby
dat
e,an
dro
bust
.Sta
tist
ical
sign
ifica
nce
at10
%,5
%,a
nd
1%le
vels
isd
enot
edby
*,**
,an
d**
*re
spec
tive
ly.P
erev
ent
auth
oriz
atio
nev
ent
gain
isth
ecu
mu
lati
veab
nor
mal
retu
rnov
era
13-d
ayp
erio
dfo
ra
com
pan
yin
aco
un
try
esti
mat
edin
div
idu
ally
.Tot
alga
ins
from
auth
oriz
atio
nev
ents
ison
ep
lus
the
abn
orm
alre
turn
toth
ep
ower
ofth
en
um
ber
ofn
etev
ents
;in
the
case
ofG
uat
emal
a,th
en
um
ber
ofn
etev
ents
is2
out
ofto
tal
4ev
ents
sin
ce1
even
tis
anab
orte
dco
up
and
thu
sco
un
ted
asn
egat
ive;
inth
eca
seof
Con
go,
the
nu
mbe
rof
net
even
tsis
1,be
cau
seou
tof
5ev
ents
,2
are
neg
ativ
e;in
the
case
ofC
uba
,th
en
etev
ents
is2
beca
use
2of
the
6ev
ents
are
neg
ativ
e.T
he
mu
ltic
oun
try
dec
omp
osit
ion
rais
eson
ep
lus
the
esti
mat
edm
ean
mu
ltic
oun
try
effe
ctto
the
pow
erof
the
aver
age
nu
mbe
rof
even
tsac
ross
the
rele
van
tco
un
trie
san
du
ses
the
rele
van
tm
ult
icou
ntr
yfi
rst
day
ofn
ewgo
vern
men
tes
tim
ate
for
the
gain
from
the
cou
pev
ent.
Th
eto
tal
gain
from
auth
oriz
atio
np
lus
cou
pev
ents
isth
ecu
mu
lati
vega
infr
omth
eau
thor
izat
ion
even
tsti
mes
one
plu
sth
ega
infr
omth
efi
rst
day
ofth
en
ewgo
vern
men
t.T
he
rela
tive
gain
from
auth
oriz
atio
nev
ents
isth
esh
are
ofth
eto
tal
gain
from
the
cou
p(i
ncl
ud
ing
stoc
km
ark
etri
ses)
du
eto
auth
oriz
atio
nev
ents
.
at McK
eldon Library 34 on January 31, 2012
http://qje.oxfordjournals.org/D
ownloaded from
1406 QUARTERLY JOURNAL OF ECONOMICS
an authorization raised tothe power of the net number of events8
plus the return over the coup window:
(5) (1 + RC, Auth)N (1 + RCoup
),
where RC, Auth is the 13-day cumulative abnormal return in coun-try C, N is the net number of authorization events, and RCoup isthe cumulative abnormal return in country C on the first day ofthe new regime. We use the return on the first day of the newgovernment because, due to the length of the coup in Guatemalaand the ensuing political instability after the end of the Arbenzregime, there is a net negative change in the stock price over theexact coup window.
The results are listed in Table VII. Though we can combinethe effects of the authorization events and the coup itself in mostof the countries, the failure of the operation in Cuba makes inter-pretation of the resulting comparison difficult. Thus we interpretthe Cuba numbers as the relative magnitude of stock price move-ments from the coup event and the authorization events. The in-clusion of Cuba in our cross-country sample also makes the fullsample decomposition difficult to interpret. Although we reportboth the Cuba decomposition and full sample decomposition, wefocus on the successful coup sample and the other four countries.
If we assume the only source of coup-related asset pricemovements are our events, together with the coup itself, we canestimate the total gains from the coup. The average gain per au-thorization in the all country sample is 12.0%, and the mean re-turn on the first day of the postcoup regime is 10.0%. For the setof successful coups, the gain from authorization events is roughlythree times that from the coup events; 75.5% of the relative gaincomes from authorization events. By country, the total gains fromthe couprangedgreatly. For a fully exposedcompany, the returnsrange from 14.1% in Chile to 77.1% in Guatemala. We also com-pute that the relative percentage benefit of the coup attributableto ex ante authorization events, which amounted to 55.0% inChile, 66.1% in Guatemala, 72.4% in Congo, and 86.9% in Iran.
8. In the case of Guatemala, the number of net events is two out of a totalof four events since one event was an aborted coup and thus counted as negative;in the case of Congo, the number of net events is one, because out of five events,two are negative; in the case of Cuba, the net events is two because two of the sixevents are negative. For the pooled country samples, we use the mean number ofevents across countries as the net events. Thus gives us 2.6 for the full countrysample and 2.4 for the successful coups sample.
at McK
eldon Library 34 on January 31, 2012
http://qje.oxfordjournals.org/D
ownloaded from
COUPS, CORPORATIONS, AND CLASSIFIED INFORMATION 1407
Overall, much of the gains from the coup occurred before the coupitself due to speculation from top-secret information. This sug-gests that estimates of the value of the coup to a company thatonly considered the stock price reaction to the coup itself wouldbe dramatically understated.
VII. CONCLUSION
Covert operations organized and abetted by foreign govern-ments have playeda substantial role in the political andeconomicdevelopment of poorer countries around the world. We look atCIA-backed coups against governments that had nationalized aconsiderable amount of foreign investment. Using an event-studymethodology, we find that private information regarding coup au-thorizations and planning increased the stock prices of expropri-ated multinationals that stood to benefit from the regime change.The presence of these abnormal returns suggests that there wereleaks of classified information to asset traders. Consistent withtheories of asset price determination under private information,this information often took some time to be fully reflected in thestock price.
We find that coup authorizations, on net, contributed sub-stantially more to stock price rises of highly exposed companiesthan the coup events themselves. This suggests that most of thevalueofthecouptotheaffectedcompanies hadalreadybeenantic-ipated and incorporated into the asset prices before the operationwas undertaken.
Our results are robust to a variety of controls for alternatesources of information, including publicevents andnewspaper ar-ticles. They are also robust across countries with the exceptionof Cuba. The anomalous results for Cuba are potentially due topublic information leaks and inadequate organization that sur-rounded that particular coup attempt. Our results are consistentwithevidenceinpolitical sciencethat business interests exert dis-proportionate influence on foreign policy (Jacobs and Page 2005),as well as historical accounts that suggest that protecting for-eigninvestments was amotivationforundertakingregimechange(Kinzer 2006). However, further empirical research is neededto uncover whether economic factors were decisive determi-nants of U.S. government decisions to covertly overthrow foreigngovernments.
at McK
eldon Library 34 on January 31, 2012
http://qje.oxfordjournals.org/D
ownloaded from
1408 QUARTERLY JOURNAL OF ECONOMICS
UNIVERSITY MASSACHUSETTS AMHERST AND IZAUNIVERSITY OF MARYLAND AT COLLEGE PARK AND INSTITUTE FOR
INTERNATIONAL ECONOMIC STUDIES, STOCKHOLM UNIVERSITY
COLUMBIA UNIVERSITY AND NBER
REFERENCES
Acemoglu, Daron, and James Robinson, Economic Origins of Dictatorship andDemocracy (Cambridge, MA: Cambridge University Press, 2006).
Allen, Franklin, Stephen Morris, and Hyun Shin, “Beauty Contests and Bubbles,”Review of Financial Studies, 19 (2006), 719–752.
Andrews, Donald, “End-of-Sample Instability Tests,” Econometrica, 71 (2003),1661–1694.
Campbell, John, Andrew Lo, and Craig Mackinlay, The Econometrics of FinancialMarkets (Princeton, NJ: Princeton University Press, 1997).
Chomsky, Noam, Turning the Tide: U.S. Intervention in Central America and theStruggle for Peace (Cambridge, MA: South End Press, 1986).
Conley, Timothy, and Christopher Taber, “Inference with “Difference in Differ-ences” with a Small Number of Policy Changes,” Review of Economics andStatistics, 93 (2011), 113–125.
Cornell, Bradford, and Erik R. Sirri, “The Reaction of Investors and Stock Pricesto Insider Trading,” Journal of Finance, 47 (1992), 1031–1059.
Corrado, Charles, “A Nonparametric Test for Abnormal Security-Price Per-formance in Event Studies,” Journal of Financial Economics, 23 (1989),385–395.
Corrado, Charles, andTerry Zivney, “The Specification andPower of the Sign Testin Event Study Hypothesis Tests Using Daily Stock Returns,” Journal of Fi-nancial and Quantitative Analysis, 27 (1992), 465–478.
DellaVigna, Stefano, andElianaLaFerrara, “DetectingIllegal Arms Trade,”Amer-ican Economic Journal: Economic Policy, 2 (2010), 26–57.
De Witte, Ludo, The Assassination of Lumumba (New York: Verso, 2001).Easterly, William, Nathan Nunn, Shankar Satyanath, andDan Berger, “Commer-
cial Imperialim? Political Influence and Trade during the Cold War,” NBERWorking Paper No. 15981, May 2010.
Faccio, Mara, “Politically Connected Firms,” American Economic Review, 96(2006), 369–386.
Fisman, Raymond, “EstimatingtheValueof Political Connections,” American Eco-nomic Review, 91 (2001), 1095–1102.
Guidolin, Massimo, andEliana La Ferrara, “Diamonds AreForever, Wars AreNot.Is Conflict Bad for Private Firms?” American Economic Review, 97 (2007),1978–1993.
Gleijeses, Piero, Shattered Hope: The Guatemalan Revolution and the UnitedStates (Princeton, NJ: Princeton University Press, 1991).
Jacobs, Lawrence, and Benjamin I. Page, “Who Influences U.S. Foreign Policy?”American Political Science Review, 99 (2005), 107–124.
Johnson, Loch, “Covert Action and Accountability: Decision-Making for Amer-ica’s Secret Foreign Policy,” International Studies Quarterly, 33 (1989),81–109.
Kinzer, Stephen, Overthrow: America’s Century of Regime Change from Hawaii toIraq (New York: Times Books Press, 2006).
Knight, Brian, “Are Policy Platforms Capitalized into Equity Prices? EvidencefromtheBush/Gore2000 Presidential Election,” Journal of Public Economics,90 (2006), 751–773.
Meulbroek, Lisa, “An Empirical Analysis of Insider Trading,” Journal of Finance,48 (1992), 1661–1699.
Snowberg, Erik, JustinWolfers, andEricZeitzwitz, “PartisanImpacts ontheEcon-omy: Evidence from Prediction Markets and Close Elections,” Quarterly Jour-nal of Economics, 122 (2007), 807–829.
at McK
eldon Library 34 on January 31, 2012
http://qje.oxfordjournals.org/D
ownloaded from
COUPS, CORPORATIONS, AND CLASSIFIED INFORMATION 1409
Wang, Jiang, “A Model of Competitive Stock Trading Volume,” Journal of PoliticalEconomy, 102 (1994), 127–169.
Weiner, Tim, Legacy of Ashes: A History of the CIA (New York: Doubleday Press,2007).
Wyden, Peter, Bay of Pigs: The Untold Story (New York: Simon and Schuster,1979).
at McK
eldon Library 34 on January 31, 2012
http://qje.oxfordjournals.org/D
ownloaded from