CIA a criminal organisation - Finance Fraud
The CIA is a criminal organization that needs to be shut down immediately, maybe by the SEC will suffice
Coups, Corporations, and Classified Information
Arindrajit Dube, Ethan Kaplan and Suresh Naidu
UC Berkeley; IIES, Stockholm University; Harvard University
September 19, 2008
Abstract
We estimate the impact of coups and top-secret coup authorizations on asset prices of partially nationalized US companies that stood to benefit from US backed coups. A small number of highly exposed and well connected firms reacted to coup authorizations classified as top-secret. The average abnormal return to a coup authorization is 1.7% over 4 days, rising to 3.4% over thirteen days.
Pre-coup authorizations account for a larger share of stock price increases than the actual coup events themselves.There is no effect in the case of the widely publicized, poorly executed Bay of Pigs invasion, consistent with abnormal returns to coup authorizations reflecting credible private information. We also introduce two new intuitive and easy to implement nonparametric tests that do not rely on asymptotic sample size approximations.
We would like to thank Zihe Liu and Ettore Panetti for excellent research assistance. Stefano Della Vigna, Robert Gibbs, John Prados, Gerard Roland and seminar participants at NYU, the Santa FE Institute, UC Berkeley, and the University of Warwick all provided helpful comments.
11 Introduction
This paper estimates the effect of secret United States decisions to overthrow foreign governments on the stock market prices of well-connected companies that stood to benefit from regime change.
We look at companies that had a large fraction of their assets expropriated by a government that was subsequently a target of a U.S. sponsored covert operation aimed at overthrowing the regime. We find statistically and economically significant effects on stock prices both from the regime change itself and from “top secret” authorizations.
Using offcial timelines reconstructed from CIA documents, we estimate the impact of key decisions concerning coup planning on stock returns. In particular, we find a strong impact on stock prices in a limited number of companies which were both (1.) highly exposed in the country under consideration and (2.) well connected to the CIA. Our results are evidence of two distinct phenomena. First, we provide indirect evidence of organizational leaks from the CIA and/or other parts of the executive branch to financial markets.
Secondly, we provide evidence that covert interventions provided arbitrage opportunities for traders of companies connected to the CIA, implying that the coups were valuable to those corporations. Our findings complement other evidence in empirical political economy that large, politically connected firms benefitted from favorable political regimes (Faccio, 2006; Fisman, 2001; Jayachandran, 2006; Knight, 2006; Snowberg et al., 2007). However, we show that firms benefft not only from publicly announced events but also from top-secret events, suggesting information flows from covert operations into markets.
Our results are consistent with recent papers that have used asset price data to show that companies can profit from conflict (DellaVigna and La Ferrara, 2007; Guidolin and La Ferrara, 2007). We also provide evidence that private information leaks into asset prices slowly over time. This is consistent with both private information theories of asset price determination (Allen et al., 2006) and the empirical literature on insider trading (Meulbroek, 1992).
Additionally, we interpret our results as providing an estimate of the value of a coup to a potential corporate beneficiary. Net total price rises from coup authorizations are larger in magnitude than price changes from the coups themselves. There exists widespread scholarly disagreement on the motivations behind covert interventions, ranging from ideological motives (Westad, 2005) to protecting the economic interests of powerful lobbies in the intervening governments (Gibbs, 1991).
While we are unable to resolve this debate, we show that regime changes lead to significant economic gains for corporations that stood to benefit from U.S. interventions in developing countries.
Within economics, the literature on anti-democratic political transitions has emphasized the role of domestic elites (Acemoglu and Robinson, 2006). However, coups have often been instigated, planned and even partially executed from abroad, most notably by the U.S. and the Soviet Union during the Cold War.
Operating under the threat of nuclear war, direct conflict between the two superpowers was replaced by covert and proxy operations to install supporting regimes. According to Easterly et al. (2008), 24 country leaders were installed by the CIA and 16 by the KGB since the end of the Second World War.
Our paper also makes an econometric contribution to hypothesis testing in event studies. The structure of our event study allows us to improve on existing nonparametric tests. Nonparametric tests used in event studies do not use exact small sample distributions but rather distributions with faster asymptotic convergence to a normal distribution (Campbell et al., 1997; Guidolin and La Ferrara, 2007). We introduce two new small sample tests that are valid without asymptotic approximations based upon the number of events.
Section II of this paper discusses the history of U.S. covert interventions, with backgrounds on each of the coups in our sample. Section III describes the data and our selection of companies and events. Section IV outlines our estimation strategies and Section V reports our main results along with a number of robustness checks.
In section VI, we present and implement our small sample tests. Section VII provides an interpretation of our main results; we decompose the total value of a coup to a multinational into public and private components. We also calibrate a simple asset pricing equation and back out the implied changes in the stock market’s assessment of the probability of a future coup. We conclude in section VIII.
2 Background and History: The CIA
The Central Intelligence Agency was brought into existence in 1947 under the National Security Act of July 26. The act allowed for “functions and duties related to intelligence affecting the national security”, in addition to intelligence gathering (Weiner, 2007). Initially, the scope of the CIA was relegated to intelligence, though a substantial and vocal group advocated for a more active role for the agency. Most of the CIA’s legal authority derived from National Security Council Directive No. 4, which ordered the CIA to undertake covert actions against communism.
Covert operations designed to overthrow foreign governments necessitated the approval of the director of the CIA in addition to the President of the United States. A 1978 executive order described covert actions as “operations conducted abroad in support of national foreign policy objectives which are designed to further offcial United States programs and policies abroad and which are planned and executed so that the role of the United States government is not apparent or acknowledged publicly” (Johnson, 1989).
After Eisenhower’s election in 1952, Allen Dulles was appointed director of the agency. Under Dulles, the CIA expanded its role to include planning and executing overthrows of foreign governments using military force.
All but 5 of the CIA operations in Table I, including 3 of the 4 studied in this paper, began during Dulles’ reign as CIA director under the Eisenhower administration. Allen Dulles was supported by his brother, John Foster Dulles, who was the contemporaneous Secretary of State. The Dulles brothers together wielded substantial influence over American foreign policy from
1952 to 1960.
The qualitative evidence suggesting links between U.S. finance and the CIA is substantial. Firstly, the Dulles brothers both worked for Sullivan and Cromwell, a prominent Wall Street law firm that included, for example, United Fruit in its clientele.
Secondly, much of the CIA, particularly under Dulles, was extracted from law firms close to the financial sector. For example, Frank Wisner, who was in charge of the covert operations wing of the CIA(called the Offce of Policy Coordination(OPC)), worked for the Wall Street law firm Carter Ledyard prior to joining the OPC. “Wisner, in turn, recruited Barnes and Fitzgerald, both Harvard-trained Wall Street lawyers” (Thomas, 1996).
In addition, CIA leadership would often consult with corporations that had investments in countries of interest to the CIA.We exploit declassified records of these meetings in determining our set of companies with access to the CIA below.
In 1974, partly due to public outcry over the U.S. involvement in the military coup in Chile, the Hughes-Ryan Act increased congressional oversight of CIA covert operations. In 1975, the U.S. legislature formed subcommittees to investigate American covert action. Thus, the intensity and scope of U.S. covert actions fell substantially (Johnson, 1989). The height of covert CIA activity lasted slightly more than twenty years encompassing the period from 1952 to 1974.
Our sample of coups includes 4 such covert attempts. The first one occurred in Iran in August, 1953, when the CIA, assisted by the UK MI6, engineered a toppling of Prime Minister Mossadegh. Mossadegh had nationalized the oil fields and refinery at Abadan, which were the property of the Anglo-Iranian oil company, itself a nationally owned company of the UK government.
In Guatemala, the CIA overthrow of Jacobo Arbenz Guzman in June, 1954 occurred after the Arbenz government had nationalised most of United Fruit’s assets in Guatemala.
In Cuba, the Castro government nationalized all US property in 1960, one year before the failed Bay of Pigs coup attempt in April, 1961. Finally, the Chilean nationalization of copper and other foreign owned assets began under the Frei government but accelerated after the Allende government came to power in late 1970. Allende was in offce less than 3 years before he was killed in a coup on September 11, 1973.
In appendix A, we provide a more detailed synopsis of each coup, focusing on the nature of the pre-coup regime, the motivations behind the expropriations, the American response, and the resolution of the coup.
3 Data
3.1 Coup Selection
We selected our sample of coups on the following basis: (1.) a CIA timeline of events or a secondary timeline based upon an original CIA document existed, (2.) the coup contained secret planning events including at least one covert authorization of a coup attempt by a national intelligence agency and/or a head of state, and (3.) the coup authorization was against a government which nationalized property of at least one sufficiently exposed multinational firm with publicly traded shares. Table I shows a full list of CIA operations from Prados (2006). The highlighted operations are those that met our criteria, which limited us to 4 coup attempts.
Operation Ajax in Iran in 1953 led to the overthrow of Muhammed Mossadegh. Operations PBFortune and PBSuccess in Guatemala in 1952 and 1954 respectively culminated in the overthrow Jacobo Arbenz Guzman. The US unsuccessfully attempted to overthrow the Fidel Castro government in Operation Zapata in 1961. Finally, Operation FU/Belt in Chile, which began in 1970, contributed to the overthrow of Salvador Allende.
3.2 Event Selection
Our primary source of events are timelines reconstructed directly from declassified CIA sources by offcial historians. Operation Ajax in Iran was constructed by the New York Times on the basis of the internal CIA history of the Iran operation written by Wilber (1954) and declassified in 20002
Available at http://www.gwu.edu/~nsarchiv/NSAEBB/NSAEBB28/
Available at http://www.nytimes.com/library/world/mideast/041600iran-cia-index.html
In the case of Guatemala, the CIA itself did an internal timeline of the operation, which we used3
The Bay of Pigs timeline comes from the National Security Archives, housed at George Washington University, which has filed virtually all of the Freedom of Information Act (declassication) requests regarding Cuba and the CIA. For FU/Belt in Chile, we used the timeline constructed by the Church Committee which was a committee set up in 1975 by the US Senate to investigate foreign intelligence operations5
The Church Committee Report, which was recently declassified, created a timeline of events based upon top-secret CIA documents for Chile.
We first extract all of the authorization events from the offcial timelines. These are restricted to those where either the coup was explicitly approved by the head of a government (the President of the United States or the Prime Minister of the United Kingdom) or the head of an intelligence agency (the CIA or MI6) or where US $1 million or more were allocated to the overthrow of a foreign government. Authorization events are coded as “good” or “bad” depending on whether they increase or decrease the likelihood of a coup. Our selection and coding of authorization events is presented in Table III.
We also extract public events from the offcial timelines for use as controls in some specifications. Public events are restricted to dates where company assets are nationalized or regime transitions and consolidations occur. The public events are coded as “good” or “bad”, where “good” events are those which are likely to increase the stock price and ”bad” events are ones which are likely to cause a decline in the stock price. The public events and their coding is listed in Table IVA; Table IVB lists the dates of the regime changes themselves.
Available at http://www.gwu.edu/~nsarchiv/NSAEBB/NSAEBB4/
Available at http://www.gwu.edu/~nsarchiv/bayofpigs/chron.html
Available at http://foia.state.gov/Reports/ChurchReport.asp
73.3 Company Selection
We apply 3 criteria to select our sample of companies. First, a company must be publicly traded, so that we can observe a stock price. Secondly, the company must be “well-connected”, in terms of being linked to the CIA. Finally, the company should be highly exposed to political changes in the affected country, in the sense that a large fraction of a company’s assets are in that country.
We begin with the list of all companies nationalized by a regime prior to the coup, which we obtain from the CIA timelines. In the cases of Chile and Cuba, the nationalized companies nationalized are not mentioned by name in the timelines themselves. We obtain lists of nationalized copmanies in Chilea and Cuba lists of nationalized companies from Congressional testimonies about expropriations of U.S. companies.
First, we require that the company was listed on the NYSE, NASDAQ, or AMEX exchanges, which we determine from listing in the CRSP database. Second, we determine, from declassified sources, whether or not the company had met with the CIA. We first include all companies that were listed as having met with the CIA in the internal histories. Then we do automated searches of the declassified CIA documents in the National Security Archive for CIA memos mentioning the company and the country in the period of the CIA operation.
If a company is listed together with the country in a declassified memo, it is included in the sample. Finally, we calculate the percent of a company’s assets that were in the country, which we call a company’s exposure, for the remaining companies. In our benchmark specification, we include only those companies which had the highest exposure for each country.
For example, ITT met with CIA offcials about Chile. However, ITT’s assets in Chile were only 7.3% of its total assets (Table II) and thus it would be difficult to pick up the impact of even a large change in the probability of a coup. Its stock price does not seem to have reacted to coup authorizations. Alternatively, Anglo-Lautaro Nitrate Ltd. was a small publicly traded company with a majority of its assets in Chile (Table II). However, it did not meet with the CIA and was almost surely not privy to information about coup authorizations.
In our robustness section, we consider two alternative selections of companies. First, we consider the top 10 most exposed, in total assets rather than percentage terms. In Guatemala and Iran, only one company was effected. In Chile and Cuba, only some are publicly traded and thus available on CRSP. In total, this specification adds 9 companies, all of which are listed in Table II. Second, we consider country portfolios of all companies which were listed as having met with the CIA and which had over 15% of their assets. In comparison with our baseline where we consider the most exposed connected company in the nationalizing country, this specification adds only one company, ITT in Cuba.
4 Methodology
Our main hypothesis is that authorization events should result in a slow increase in the stock price of the affected company over the days following the event. There are multiple reasons that prices may react steadily and slowly as opposed to all at once with private information. First, the information may itself slowly take time to diffuse. Second, there may be secondary trading or momentum; traders may update based upon previous price increases. Third, traders may be cautious and wait to see if other investors are trading on the private information (Allen et. al., 2006). For this reason we look at windows of different lengths around the authorization events.
Our benchmark specification is a 4 day window starting at the event date. In this paper, we employ two different estimation strategies. The first, which we call the “regression method”, includes the contemporaneous market return as a control along with dummies for contemporaneous authorization events in a single specification where the dependent variable is the raw stock return. Our second approach is the event-study methodology originally developed by Fama (Campbell et al., 1997; Fama, 1969).
We first estimate abnormal returns using a pre-event sample, where abnormal returns are returns in excess of what would be predicted in a simple linear market model. We then calculate the mean cumulative abnormal returns for a number of days after each event, and test to see if it significantly different from 0. We refer to the second approach as the “out of sample method”, referring to the fact that the abnormal returns are calculated using a sample of stock market returns from before the authorization events.
4.1 Regression Method
For the regression method we estimate the following equation with OLS:
Rft = af + ßfRmt + Dft + ft (1)
Rft is the one day raw stock return for firm f between date t and date t-1, Rmt is the one day New York Stock Exchange index return between date t and date t-1, and Dft is a k-day dummy variable which takes on a value of one on an authorization day and for the k-1 days following an authorization day. The average daily abnormal return over the k days after an event (inclusive) is f.
The cumulative abnormal return is kf, the average abnormal return times the event window length. Our sample is the time period starting exactly one year before the nationalizing regime comes to power until exactly one year after the end of the coup. The standard error for the cumulative abnormal return is given by the standard error on the regression coeffcient multiplied by the length of the window. Except where noted, we report heteroskedasticity-robust standard errors.
4.2 Out of Sample Method
The out of sample method first estimates a market model in an “estimation window” that is prior to any coup-related events. Our estimation window is one calendar year in length and begins 2 years before the nationalizing regime comes to power.
The number of trading days differ in the estimation window. Chile has 235 days; Cuba has 250 days; Guatemala has 282 days; and Iran has 260 days. Guatemala has more trading days because the NYSE was open on Saturdays until September 29, 1952. Also since more trading holidays have each firm, we estimate:
Rft = af + ßfRmt + ft (2)
Using the estimated coefficients from (2), we calculate the abnormal returns around our authorization events as the difference between the actual and predicted returns for a given date:
ˆ ARft = Rft - ˆ af - ˆ ßfRmt (3)
We consider windows around the authorization events of length k, where k is between 0 and 15 days. We take the average abnormal return over the k days as:
Pt0+k
t=t0
ˆ ARft k + 1
The cumulative abnormal return for k + 1 days for firm f, CAR(f, k + 1), is defined as:
CAR(f, k + 1) = t0+k X t=t0 ˆ ARft
The standard error for the average abnormal return for an individual event is the estimated standard deviation from the estimation window multiplied by the square root of the length of the CAR period:
ˆ sARf v k + 1
To compute cumulative abnormal returns for multiple events, we add up the CARs across events. In most event studies, there is one event per firm. In our setup, that is not the case. However, as long as our events are independent over time within firm been added over time, the number of trading days per year has decreased over time.
Lastly, Anglo-Iranian traded on the London Stock Exchange during the period in question, which accounts for the lower number of trading days in comparison with Guatemala. We opted to use exactly one year for the estimation window rather than a fixed number of trading days. However, the choice of the estimation window does not impact our results.
as well as across firms, we can use abuse notation use and f to index events over a set F of events with |F| number of events. In this case, different events may correspond to the same firm. Then, we can compute the CAR for a group of firms:
CAR(F, k + 1) = P|F| f=1 Pt0+k t=t0 ˆ ARft |F|
and we can compute the standard error by:
P|F| f=1 ˆ sARf vk + 1 |F|
5 Results
5.1 Baseline Results
In Table 5, we report the cumulative abnormal returns for authorization events using window lengths ranging 1 to 16 days. We find clear evidence that stock prices react positively to authorization events using both our regression and out of sample methods. In the pooled sample, the average 4 day stock price return for an authorization event is 1.7% with a standard error of 0.7%.
The cumulative abnormal returns are signifficant for the all-country sample from 4 day through 16 day cumulative abnormal returns at a minimum of 10% level of signifficance and often at a 1% level, depending upon the specification. The abnormal returns are largest between 3 and 12 days after the event, consistent with the hypothesis that private information is incorporated into asset prices with a delay.
Figure 1 provides graphical evidence on abnormal returns around an authorization event, with 95% confidence intervals shown. We compute cumulative abnormal returns, aggregated across events, for each of the 22 days following an event and each of the 22 days prior to an event, aggregated backwards in time. Cumulative abnormal returns become significant at a 5% level on the 4th day after an event and remain significant until day 13.
Moreover, the gains seem to be permanent, although not statistically distinguishable from 0 after approximately 2 weeks. Going backwards in time from the event date, however, the cumulative abnormal returns show no trends and are never signifcant.
The effects for Iran and Guatemala are consistently the strongest. In both cases, the average cumulative abnormal return after 4 days is around 2.5% with a standard error of less than 0.9% using the regression method. The out of sample method’s estimates are almost identical for Guatemala and smaller for Iran. The standard errors are consistently smaller using the out of sample method. The Chile estimates are slightly smaller in magnitude. The peak effect is almost 0.5% per day over a 4 day horizon. The abnormal returns for Guatemala and Iran are 0.5-0.6% per day by the day after the event and they remain that high for the first 6 days.
We do not find an effect within the Cuba subsample. There is no detectable change in the stock prices of affected companies following a decision to invade Cuba, whether made by the CIA or the President. This could in part be due to the poor planning and execution of Operation Zapata. Much of the information was leaked to the press ahead of time7
. Additionally, substantial errors in the Bay of Pigs planning and implementation may have made investors rightfully skeptical about the likely success of the operation.
5.2 Robustness
We perform a number of robustness checks. All are estimated both in the pooled sample and by country. We compute cumulative abnormal returns over a 4 day period following an authorization event. All specifications are estimated using the regression method.
Kennedy reads the [NYT] story he exclaims that Castro doesn’t need spies in the United States; all he has to do is read the newspaper” (Wyden, 1979)
135.2.1 Public Events and Media Coverage
Top-secret decisions to overthrow foreign governments may have coincided with public events in the targeted countries. This could bias our estimates, reflecting the effect of public news rather than private information. We control for other events in two different ways. First, we control for the number of articles in the NY Times mentioning the country by name.
Second, we control for other public events; these are nationalizations of foreign owned property as well as electoral transitions and consolidations which are also listed in the declassified timelines. They are listed in Table IVA. Third, we control simultaneously for both public events and NY Times articles.
Lastly, we also try dropping all dates where the NY times had at least one article on the country (Meulbroek, 1992). This is a strong test. Since most days have at least one article mentioning any given one of our countries, we lose most of our sample in this specification.
Table VI reports 4-day cumulative abnormal returns. We find that controlling for public events and New York Times articles does not afect our results. The average aggregate effect for a 4-day period is between 1.7% to 2.4% and significant at the 1% level, depending on the specification.
This is true even when we restrict to days with no New York Times articles about the relevant country. Our results by country are largely similar to those from the baseline specifications. One exception is the estimate for Cuba on the sample restricted to days where the New York Times had no coverage of Cuba.
In this case, the coeficient is 1.7%, and significant at the 10% level. This is consistent with the theory that top-secret news about authorizations is more credible when it remains covert.
5.2.2 Other Robustness Checks
We also consider raw returns, unadjusted by a market return, reassuring us that our cumulative abnormal return effects are due to increases in the treatment company stock prices rather than drops in the market. Column 1 of Table VII shows an 14average 0.49% cumulative abnormal return per day. To control for potential serial correlation in returns, we cluster on month. As can be seen in Column 2 of Table VII, this reduces our standard errors across specifications and does not alter any of our qualitative results.
We control for industry returns by first constructing an equal-weighted basket of returns for all companies in the same 3-digit industry as our treatment companies. We exclude the treatment companies themselves, and otherwise restrict the basket to companies which were listed in CRSP for the entire event window period for the treatment company in question. We then regress the returns of the treatment company on the NYSE index, the authorization events, and the equal weighted industry index. Column 3 of Table VII shows the estimates from this specification, and again the efect is unchanged.
We also consider two placebos, reported in columns 7 and 8 of Table VII. We regress NYSE index returns on our event dummies. We also regress our equal-weighted baskets of industry returns on country-specific NYSE index returns and the authorization event dummies. The 4 day abnormal returns are small and insignificant in all of the samples, both with the NYSE returns as the dependent variable and with the industry returns as the dependent variable.
We consider two other specifications where we look at a broader set of companies, reported in columns 4 and 6 of Table VII. First, we construct an equal-weighted basket of all companies within a country who (1.) met with the CIA and (2.) held 15% or more of their assets in the nationalizing country. Our results in this specification are similar to our baseline results. This is unsurprising since our sample in this specification is the same as in our baseline with the exception of the addition of ITT to the country portfolio in Cuba. In our second specification, we consider all publicly listed companies in the top ten most exposed companies operating within the nationalizing country. Here we use gross asset exposure rather than exposure as a percentage of total assets. This leaves us with 13 companies in total, listed in Table II.
The 4 day abnormal returns for Chile and Cuba are both negative and not significant 15at the 10% level. This is consistent with the hypothesis that only companies which were both large and highly exposed would react to authorization events. One potential explanation for our fndings is pre-existing market momentum. We include a dummy that is equal to 1 in a 20 day window around each authorization event. This specification tests whether the abnormal returns are higher in the 4 days right after an authorization than in the average of the 20 day period surrounding each authorization event. Column 5 of Table VII shows that the average abnormal return per day is approximately 0.52% and is significant at the 1% level. Pre-existing price trends do not explain our results.
5.3 Time-Shifted Placebos
As additional evidence that our effects are not an artifact of the data, we rerun our main specification on placebo dates. We take our 4 day cumulative abnormal returns and shift our authorization events forwards as well as backwards by 5, 10, 15, 20 and 30 days. For a K day shift, we estimate:
Rft = af + ßfRmt + fKDft+K + ft (4)
Out of the 11 time-shifted regressions, fK is only significant for K = 0, our benchmark specification with cumulative abnormal return of approximately 1.7% which is significant at the 1% level. The cumulative abnormal returns 5 days or 10 days before a authorization event are zero to a tenth of a percentage point. The two largest of the remaining ten abnormal returns are the ones for K = 30 and K = 5. Both are between 0.6% and 0.7%. All other abnormal returns are well below 0.5%. The placebo estimates reinforce that our baseline estimates are due to local serial correlation in returns. The pattern of no abnormal returns before a decision, sizeable abnormal returns just after a decision, and smaller possible abnormal returns in the medium run after a decision is consistent with our hypothesis of secret authorization events causing a slow increase in the stock price.
165.4 Coup Effects
We now estimate abnormal returns, using the regression method, for coup attempts. We do this for two reasons. First, we want to show that these companies were affected by the coup attempts themselves, confirming that companies were beneffitting from the regime change. Second, we want to compare the direct effect of the coup itself to the total net rise due to pre-coup authorizations.
We look at 3 specifications: abnormal returns on the first day of the coup, abnormal returns on the first day of the new regime, and abnormal returns during the coup window. We define the coup window as the period from and including the first day of the coup to and including the first day of the new regime. For Cuba, which was unsuccessful, the coup window is the duration of the Bay of Pigs operation, as given in the CIA timeline. These dates are listed in Table IVB.
Since our coup window lengths vary across countries, instead of reporting cumulative abnormal returns, we report the average daily abnormal return during the window. Our results are large and signifficant. On an average day during the coup window, our treatment companies, experienced a stock price rise of 0.8%. The individual company average abnormal returns vary from United Fruit in Guatemala which had zero rise on average during the coup window to Anaconda in Chile which experienced a 4.6% increase in its stock price. Anaconda’s large increase in its stock price was partially due to the fact that the coup happened quickly and was consolidated essentially immediately; this is dififerent from our 3 other coups where it took longer for the overthrow to succeed or fail.
Cuba’s abnormal returns were negative because the coup failed. This suggests that the possibility of a coup against the Castro regime in Cuba had already been priced into American Sugar’s stock. Anglo-Iranian oil had a large increase over the coup window. It was approximately 1.4% per day and significant at the 10% level. The insignificant estimate for Guatemala is perhaps due to the high degree of political uncertainty following the coup. When the Arbenz government finally resigned on June 28, 1954, there was still speculation about whether the coup would be successful. Also, in the 11 days after the fall of the Arbenz regime, 5 separate juntas gained control of the government. Lastly, within two weeks of the end of the Guatemalan coup, the United Fruit Company was hit with a large anti-trust law suit (Gleijeses, 1991).
We consider two other measures of the effect of the coup: the abnormal return on the first trading day of the coup and the abnormal return on the first trading day of the new regime. The average abnormal return across companies on the first day of the coup was approximately 2.3% and signifficant at the 1% level. In both Chile and Cuba, the returns were signifficant at the 10% level or higher. The abnormal returns in Cuba were positive, perhaps indicating that markets expected the coup to be successful on the first day. The abnormal returns on the first day following the coup attempt (whether or not it was successful) are large for all companies, with Anglo-Iranian at 2.0%, American Sugar at -3.3%, United Fruit at 3.7% and Anaconda at 4.6%. The CIA-engineered regime changes had a substantial impact on stock prices for the exposed companies in our treatment sample.
Small Sample Distribution Tests
Conditional distributions of abnormal returns are potentially skewed in small samples, thus distorting test size. In response to this problem, tests have been developed which have better small-sample distribution properties than t-tests of OLS coefficients for event study regressions. The two most common are the sign test and the rank test. Both tests focus on median as opposed to mean returns and thus are more informative in the presence of abnormal return skewness. However, the sign and rank tests are both only asymptotically of correct size.
Guidolin and La Ferrara (2007) provides a more detailed summary of the sign and rank tests. The sign test signs events +1 or -1 depending upon whether the event abnormal returns are above or below median abnormal returns. Thus, for each event, we defne Gft = 1(eft - median(e* f ) > 0), where eft is the the cumulative abnormal return during event t in country f, and the median is taken over the estimation window. 18Under a null hypothesis that event abnormal returns are identically and independently drawn from the same returns distribution as estimation window returns, the average event sign should not be significantly diferent from zero. The sigh test divides the average sign across firms of the event date,
PGft by the standard deviation of the average signs of the firms over the estimation window. This statistic is given by:
Pft Gft0 q1 k Pt0-1 t=t0-k( PfGft)2 (5)
The test is asymptotically distributed according to a unit normal but simulations have shown it to have faster small sample convergence properties to the normal in comparison with standard tests on OLS coefficients.
The rank test assigns a rank to each event abnormal return for a firm relative to its estimation window (a total of k1(f) - k0(f) + 1 daily abnormal returns), which we denote by fift. Under a null hypothesis that mean event rank is not signifficantly above or below the median rank in the estimation window, the rank test divides the mean rank for event abnormal returns by the standard deviation of the mean rank across firms over the estimation window.
Pf (fft0 - k+1 2 ) q1 k Pt0-1 t=t0-k( Pft fift - k+1 2 )2 (6)
Again, this ratio is asymptotically normally distributed with rapid small sample convergence properties. The rapidity of small sample convergence is verified through simulations (Corrado and Zivney, 1992).
While the sign and rank tests are a definite improvement over small sample OLS estimates, they have two main drawbacks in the context of the current paper. First, when testing the impact of a single event on multiple companies, it is sensible to control for intra-day correlation in returns across companies. However, when each event occurs on a different day, this is not necessary. Moreover, attempting to do so in small samples will incorrectly estimate the standard deviation of the returns. Second, the small sample convergence properties of the sign and rank tests are only verified through simulations and thus the speed of convergence may depend upon the distribution of returns. Since all of our events occur on different days, we do not need to take intra-day covariances across companies into account. This allows us to construct actual small sample tests (as opposed to asymptotic tests). We create a test based upon the binomial distribution to supplant the sign test and a test based upon the uniform distribution to supplant the rank test. We report results both for the 4 company sample and the 3 company sample (excluding American Sugar in Cuba).
6.1 Generalized Bernoulli Test
The Bernoulli test is a small sample test corresponding to the sign test which is implementable when events are distributed identically and independetly of one another. In general, we would like to come up with a statistic which computes the probability of observing at least as many abnormal returns above or below the median or, more generally, any given percentile:
P X f Gf ! = m
In the canonical event study, one event occurs on a given day and a group of firms all experience the event. In this case, it makes sense to assume that there is serial correlation in the event returns. Therefore, we can assymptotically approximate the probability by a normal distribution, incorporating the covariance in returns by implementing the sign test. However, if abnormal returns are distributed identically and independently across time and across events, then we know that the events are distributed exactly according to a Bernoulli distribution and the joint probability of
getting m events higher than the pth percentile is given by the cumulative Bernoulli
distribution:
FM(m; p) = 1 - M X f=m M i pi (1 - p) M-i (7)
Without loss of generality, we assume that p = .5. Then, due to the symmetry of
the cumulative Bernoulli distribution, the two-sided probability of getting m or more
abnormal returns above the 1 pth percentile or below the pth percentile is given by
2 PM
read the rest:
Stumble It!
0 Comments:
Post a Comment
<< Home