Master thesis – Business economics: Finance
Insider Trading and Bank Bailouts
Abstract: this thesis examines the trading done by bank insiders in the time period surrounding the recent financial crisis of 2007/2008. Using an event study, the insiders’ cumulative abnormal returns around buy and sell trades can be investigated. For both types of trades, the one-week returns are found to be significant. In addition, two separate proxies are examined: the monthly change in ownership and monthly trading activity. Pre-crisis insiders of TARP recipients decreased their positions, whilst overall insiders still increased their ownership. Looking at trading activity, this increased overall during the bailout period 2008-2009, however insiders of TARP recipients seem to have withheld from trading.JEL classification: G01, G14, G21.
Keywords: Insider trading, banks, crisis, TARP.
Master Business Economics: Finance
Name: Maia ten Kortenaar
Student number: 10459979
Thesis Supervisor: Dr. T. Jochem
Statement of originality:
This document is written by student Maia ten Kortenaar, who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
i
Table of content
1. Introduction ... 1 2. Literature review ... 3 2.1 Insider trading ... 3 2.2 TARP ... 42.3 Too big to fail ... 5
2.4 Efficient markets and event studies ... 6
2.5 Related research ... 6
2.6 Hypotheses ... 8
3. Methodology ... 10
3.1 Cumulative abnormal returns ... 10
3.2 Changes in ownership ... 12 3.3 Trading activity ... 13 4. Data ... 15 4.1 Insider trades ... 15 4.2 Bank sample ... 16 4.3 Other manipulations ... 16 5. Results ... 18 5.1 Descriptive statistics ... 18 5.2 CARs ... 21 5.3 Changes in ownership ... 26 5.4 Trading activity ... 29 6. Robustness Checks ... 31 6.1 CARs ... 31 6.2 Changes in ownership ... 32 6.3 Trading activity ... 35 7. Conclusion ... 38 Appendix ... 41 References ... 45
1
1. Introduction
Insider trading is a much debated subject in the financial industry and is researched extensively with regard to information announcements. Opinions on the possible benefits or harm this type of trading causes are divided. According to Aktas (2008) insider trading can stimulate market efficiency through faster price discovery. Whilst on the other hand, insider trading harms welfare by making profits off the uninformed investor (Aktas, 2008). In any case, most evidence indicates that insiders are able to generate abnormal returns when trading on their monopolistic information (Gosnell, 1992). Even outsiders mimicking trading strategies as filed by corporate insiders can make a profit. This raises an interesting situation during the recent financial crisis, during which the information asymmetry about the financial positions of firms became larger regarding insiders and uninformed investors, especially in the financial sector. Insider trading can be interpreted as controversial because the profit made by the informed party is generated by a loss of an uninformed party. Many financial institutions were facing an unstable and uncertain situation during 2007-2008 when the financial crisis was at its peak. Some of these institutions in the U.S. received governmental help in the form of TARP (troubled asset relieve program). In particular, many banks received this government aid. Approximately 250 billion dollars in total have been used to stabilize the U.S. banks (U.S. Department of the Treasury, 2016). The crisis led research into a rather new field, since some of these banks are so immensely large and interconnected that they are thought of as “too big to fail”. This implies the government will make sure the banks do not go bankrupt to prevent the devastating effects this would have on the economy and society, this phenomenon can trigger moral hazard in risk-taking in banks. A large fraction of the TARP recipients will also be considered TBTF. These two aspects provide ground for interesting research.
If insiders were aware of the unstable situation that was evolving, they might have acted on this knowledge. Some research in this area has already been conducted. Recent findings by Cziraki (2015) indicate bank insiders increased the selling of their stock in the early beginning of the financial crisis, when the housing prices just began to drop. Since not all banks received TARP funding, there is a possibility to research if bank insiders behaved differently if their bank received TARP. Insiders might have behaved differently knowing the bank would be bailed out, taking larger risks a priori. Findings by Jagolinzer et al. (2014) imply that insiders did anticipate the recovery of stock prices after the crisis, but they do not find any evidence that the corporate insiders predicted the financial crisis by analysing their trades. This leaves a contradiction in the current literature.
The importance of this research lies in the knowledge that can be extracted on trading during crisis by insiders in very large (financial) institutions. Regulators can adjust the current law to prevent the exploitation of insider knowledge during difficult financial periods, this is especially important in these financial institutions but can also be extrapolated to insider trading in general. If the institutions
2 that are too big to fail receive help in all situations, this creates a moral hazard problem. In the end this means that taxpayers’ money ends up in the hands of the corporate employees, when there are much more valuable destinations for this money. Another problem would emerge if insiders massively sold off their shares during financial difficult periods, this would drive the stock prices further down. Steps can be taken to prevent banks from getting TBTF, the first movements are already seen in current legislation. Altogether, this thesis adds to the research on crises and insider trading and the intersection of these two platforms.
This thesis examines insider profitability around their trades during the period 2004-2012, and 2006-2011 in particular, which includes the recent financial crisis. Multiple event windows are used for the event study to calculate cumulative abnormal returns around each trade. Regressions are performed using these found CARs as dependent variable, in these models special attention is on TARP and the different time periods. For both purchases and sales done by insiders, significant CARs are found, positive returns for purchases and negative ones for sales. Running a cross-sectional regression including indicators for time periods and TARP recipients shows that insiders of TARP banks made significantly more profitable returns on their selling transactions during the bailout period. For buy transactions the TARP intensity, which is TARP relative to the banks’ value, shows higher abnormal returns. Thus, insiders within banks receiving high relative amounts of TARP made more profitable share purchases. Two more models are tested next to the CAR regressions, monthly changes in ownership and the trading activity in banks. In the pre-crisis period insiders of TARP recipients decreased their ownership more on a monthly basis than those in non-TARP banks. This implies that insiders were actually aware of the situation and this finding is in line with the finding by Cziraki (2015), in contrast to Jagolinzer (2014) who finds no evidence for the prediction of the crisis. Trading activity results in the opposite of what is expected, the higher the TARP-intensity, the less the insiders traded during the bailout period. So, overall the insiders made profitable trades in both groups of banks, with no overall effect on profitability for TARP. Overall, insiders traded more frequently during the bailout period compared to the post-crisis period.
The structure of this thesis is as follows, the first chapter will discuss relevant research on insider trading and the necessary back ground information. This will build the hypotheses to be tested in this paper. Next, chapter 3 defines the methodology used to perform the research and test the hypotheses, these include CARs and cross-sectional regressions. Chapter 4 will explain the dataset and performed manipulations. The results stemming from these regressions will be discussed in chapter 5. Robustness checks will be discussed in chapter 6. Finally, chapter 7 will conclude the thesis with a conclusion and limitations as well as provide points for future research.
3
2. Literature review
This chapter will provide the basis for the research conducted in this thesis. First, insider trading in general will be briefly discussed alongside the already mentioned proxies TARP and TBTF. After this, important papers closely related to the proposed research question are discussed in decreasing order of relevance. Finally, the hypotheses will be discussed and motivated.
2.1 Insider trading
The main subject studied in this thesis is insider trading. When referring to insider trading, one has to specify whether illegal or legal insider trading is addressed. Legal insider trading is in 1934 defined by the Securities and Exchange Commission (SEC), as buying and selling stock by corporate insiders, which can be directors, officers, managers and so forth. A person owning 10% or more of any type of security of the firm is also considered an insider. Reasons to trade can be liquidity needs or general perceptions of the market which can result in profits or avoidance of losses. Each trade by these insiders has to be reported to the SEC using forms 3, 4 and 5. Form 3 is the initial filing of ownership, incentive programs often contain shares to align the insider’s goals with those of the firm. Form 4 covers changes in ownership as the result of transactions, and last all transactions not yet reported should be filed using a form 5. Each corporate insider has to sign the filing document themselves.
On the other hand, illegal insider trading is considered trading securities in breach of duty or trust while possessing non-public information on the security. This can also be the case if someone outside the firm is tipped off by a corporate insider and trades on this information, this can be a relative or friend for example. In this thesis only the legal version of insider trading, which is reported to the SEC, is used to test the hypotheses. Illegal insider trading is hard to pinpoint since this often happens through tipped connections which are unknown (SEC, 2013). Often one can only judge the trader to be in breach of legislation quite some time after reporting or the trade can be not reported at all. Upholding the legislation requires an active position by the SEC.
When violating insider laws the penalties differ significantly, it can be up to three times the profit gained or loss avoided by trading on non-public information, this is SEC rule 10b-5 which is commonly brought forward in lawsuits against fraud. Wilful violation of the more general antifraud rule, SEC rule 10b, can result in penalties ranging from a fee of 5 million US dollars to 20 years of imprisonment (Seitzinger, 2016). Above numbers show that insider trading is something to be careful about.
In 2002 the Sarbanes-Oxley Act (SOX) was implemented which, among other effects, accelerated the reporting deadline for insider trades and ownership (SEC, 2013). For the initial filing (form 3) this means one has to report within ten days of becoming an officer, director or beneficial
4 owner. More relevant for this research is form 4. The SOX act requires changes in ownership to be reported within two business days after the transaction. If the transactions are somehow eligible for deferred reporting, form 5 requires reporting within 45 days after the fiscal year end.
2.2 TARP
As already brought forward in the introduction, this thesis looks at insider trading during the financial crisis in particular and will use the government issued troubled asset relieve program (TARP) as one of the proxies. This subchapter will briefly discuss the recent financial crisis and the response of the U.S. government.
The financial crisis finds its origin in the U.S., the burst of the housing bubble. Preceding the burst of the bubble, rising housing prices lead to an increasing number of mortgages being sold to homeowners. Under normal economic circumstances these homeowners would not have been able to take out large amounts of money with high interest rates. In early 2006 the first signs of the financial crisis begin to show. The housing market starts to collapse, with these dropping prices people can no longer afford their so-called subprime mortgages. Hedge funds and banks with money invested in the asset backed securities see the values of their securities drop, in the end becoming completely worthless (Acharya, 2009). The financial system is so interconnected that the insolvency problems spread around the globe. Liquidity in the financial system is very limited and this has serious consequences for the entire economy. Small and large loan requests can no longer be fulfilled resulting in firms going bankrupt and employees having to be let go.
On October 14th 2008 the U.S. government announces a program to enhance the market
stability and liquidity as well as strengthen the financial institutions (Federal Reserve, 2013). The program was based on three complementary initiatives: a capital development initiative, the capital purchase program and the liquidity guarantee program. Together these programs offered lower-cost capital to financial institutions and the possibility to sell equity interests to the U.S. Treasury (Federal Reserve, 2013). Farruggio et al. (2013) looked into this capital assistance programs, the main reason to implement these programs was to overcome a “loan freeze” and by doing this keep the market as liquid as possible. In their paper they show that announcements of TARP and capital repayments by banks increase shareholder value and decrease systematic risk. This increase in shareholder value on announcement can possibly be captured by insiders.
Bayazitova and Shivdasani found that some banks opted out of participating in the TARP program or rejected offered capital injections. There are multiple reasons for this, first of all the healthier banks were more likely to reject aid. Secondly, the interference of the government in executive compensation was a major issue for banks. If TARP was accepted, the government required
5 limits on executive compensation. This was a reason for some banks not to participate. Finally, a capital infusion program is generally associated with a negative signalling effect. However, no evidence is found by Bayazitova and Shivdasani for this theory (2012).
2.3 Too big to fail
Closely related to TARP funding is the phenomenon of banks becoming too big to fail in light of the recent crisis. The first sign of this is the collapse of Bear Stearns, which was the first financial institution in serious trouble, where the creditors where protected by the Federal Reserve subsidizing a takeover. After this the Fed decided not to help out the Lehman Brothers, this left investors and the market unsure what to expect next. However, the situation the failure of Lehman caused was so problematic that the government became determined to not let this happen to the following institutions (Baker & McArthur, 2009). This is exactly what is implied by “too big to fail”, if large financial institutions would go bankrupt, the cost to society is significantly large leading to inference by the government. It will intervene to prevent the institution from failing at any cost. This system might trigger excessive risk-taking by officers or directors in these banks because the risk of bankruptcy is as good as nihil and thus so are the consequences.
Policy makers are leaning towards a policy to a reform to smaller banks by splitting them up into smaller constituent parts and by decreasing the balance sheet. The latter can be accomplished by not renewing loans and securities after repayment (Stern & Feldman, 2009). In current research if a bank is considered TBTF, an implicit subsidy is addressed to this institution. In most cases large banks can acquire capital more easily and at lower cost relative to small banks. Being large does not mean that the institutions is more efficient necessarily. Davies and Tracey (2014) looked into this observation. They use a rather simple definition to determine banks to be TBTF in their research, if the difference between the “support and “standalone” rating is greater than zero the bank is considered to be TBTF. This difference is called the ratings uplift and is also described by Noss and Sowerbutts (2012) of the Bank of England. The ratings are of course not numerical to begin with, the original ratings are translated to numerical ordinal values. For example, a AAA+ rating can be translated to 21, this is the highest value. A rating of C will be translated to 1, the lowest value. Furthermore, Davies and Tracey find that economies of scale are affected by TBTF factors, which means decreasing bank size would lead to economies of scale foregone. Although, some cost advantages of TBTF are not addressed to economies of scale but are due to TBTF factors.
A bank being TBTF could again lead to excessive risk taking by its employees because there are no consequences attached for the bank employees to the effects of risk-taking. Many of these TBTF institutions will have received TARP funding during the crisis, however some will have opted out for
6 reasons explained in the previous subchapter. For example, the interference of the government in the executive compensation.
2.4 Efficient markets and event studies
This research looks at abnormal trading returns around insiders’ trades, first some information about the efficient market hypothesis needs to be discussed to be able to define these abnormal returns. Market efficiency comes in three forms: weak, semi-strong and strong (Fama, 1969). With weak form efficiency the market only contains historical information on stock prices, a semi-strong market contains obviously publicly available information and finally a strong form market contains all information so no profits can be obtained from monopolistic information. This thesis assumes the stock market to be semi-strong efficient, thus insiders should be able to profit from their trades on monopolistic information. The market is assumed to be efficient enough to incorporate the new information revealed by the trade. This is an important assumption, if the market would not react to new information directly, the insiders would not be able to make any profits.
Event study methodology build on the EMH and is used to examine insider trades during the recent financial crisis of 2008, event studies are commonly used in finance and economic research. It is very useful to determine effects around one specific date or event because it takes out the general market movements of the stock prices (MacKinlay, 1997). There are different ways to calculate the expected stock return, one can use a constant mean return or the market model. The market model shows better results with regard to the variance of the abnormal returns.
2.5 Related research
Two working papers relating insider trading to the financial crisis are published by Cziraki (2015) and Jagolinzer (2014), these are already briefly mentioned in the introduction. The latter is a working paper by Jagolinzer et al. (2014) which examines the extent of insider predictability using top executives in financial institutions in the U.S. banking system. They find no evidence regarding insider anticipation of the financial crisis and the effect of it on the financial institutions. However, insiders do seem to have anticipated the recovery of share prices after the announcement of TARP intervention and traded on this knowledge. This paper examines similar questions as covered in this thesis, however difference is in the measurement of insider trading and additional important variables which will be explained throughout the thesis. Jagolinzer et al. (2014) use a sample of 261 banks that received TARP to test their hypotheses considering TARP, whereas this thesis will use a larger dataset and will make use of a difference-in-difference regression and thus a control group of banks that did not receive TARP funding.
7 The second paper related to this thesis proposal is by Cziraki (2015). His approach focuses on banks and insiders by looking at exposure to the housing market before and during the financial crisis of 2008. Cziraki’s findings indicate that insiders in high-exposed banks were selling their stock more frequently, they sold 20% more of their stock compared to low-exposed banks. Interestingly, these insiders already increased their selling in 2006, when the housing prices just started to drop. This could indicate that insiders were aware of the possible effects and dangers of investing in the crashing housing market at that time, however there is no reason to believe that bankers realized the possible dangers of the housing market before 2004/2005. The findings by Cziraki (2015) contradict the findings by Jagolinzer (2014) on potential awareness of the crashing market.
The recent crisis also provides ground for Baghat and Bolton (2014), they wrote a paper on executive compensation in banks during this period. Their main findings are that managerial compensation incentives lead to excessive risk taking and that the poor performance of banks was not the result of unforeseen risk. They used three different measures for risk-taking: bank z-scores, asset-write downs and loans from the Fed. This finding could indicate that high risk taking top executives, which means high risk banks, have large amounts of stock to trade because they receive this as compensation. This could lead to more insider trading in these high risk banks. On the other hand, Fahlenbrach and Stulz (2011) advocate that poor performance by banks could be seen as a result of unforeseen risk.
Abumustafa and Nusair (2011) looked at general insider trading during the financial crisis, one main finding is that insider trading is profitable in the short run but not in the long run. Another interesting finding was that insider trading increased significantly in the last quarter of 2008 and first quarter of 2009, there seems to be a connection to the financial crisis and the frequency of insider trading. The increase of insider trading in Q4 2008 is interesting for this thesis, it might be related to TARP help.
The papers discussed above all look at the most recent financial crisis of 2007-2008. In 1987 there was a market crash similar to the one in 2007, Seyhun (1990) conducted research on the response of corporate insiders to this crash. The trading behaviour of insiders does not indicate any awareness of the market crash about to happen. When the stock prices were significantly rising preceding the crash, insiders were not heavily buying or selling stock. This indicates that they found the fundamental values in line with the stock prices and thus were not expecting a crash. However, when the prices began to drop, the stocks with greater decreases were purchased more frequently. The same stock shows greater price recovery in the period after the crash. The evidence suggests an overreaction effect in market pricing.
8 In contrast to Seyhun, Marin and Olivier (2008) find that insider selling intensifies preceding a crash, e.g. a drop in stock value. However, there is no evidence that insider purchasing increases preceding a jump in stock prices. This finding implies that insiders do not trade as often on positive information as they do on negative information. Taking this into consideration when looking at the financial crisis, it should be visible that many insiders traded if they knew the bad situation the banking system was in. Especially so if it was known that the bank would not receive any TARP funding.
Madura and Wiant (1995) are one of the first to look at insider trading in banks. In particular, they were interested in the informational content of insider trades in banks, they find size and recent performance to be key characteristics for the information contained in these trades.
2.6 Hypotheses
The main objective of this thesis is to test whether bank insiders traded differently if their bank received government help, so called TARP funding. Previous research suggests that insiders did trade when the situation of the banks worsened, when the housing prices started to collapse. Banks that received TARP were the ones in the most problematic situation and often deemed too important to fail by the government. These banks are commonly called “too-big-to-fail”. Is there a difference in profitability of insider trades among TARP recipients and non-recipients? This raises another question on the risk-taking of the banks’ employees in relation to insider trading, will corporate insiders take on more risk if they know the bank will be bailed out in the end? The following hypotheses will be addressed in this thesis:
Hypothesis 1: There are positive abnormal insider trading returns for banks receiving TARP during the bailout period.
This hypothesis is at the core of the research, to be able to examine abnormal returns for insiders they need to exist. This is motivated by the paper of Jagolinzer et al. (2014) which finds evidence for anticipation of stock price recovery for insiders. This implies that insiders expected stock prices to rise again after the announcement of government interference and possibly made a profit out of this.
There are two important assumptions made here, the abnormal returns are defined by the reaction of the market to the new information and trades. If the market does not react the stock prices will not increase and thus there are no abnormal returns for the insiders. The market will only react if it is efficient enough to incorporate the signal into the prices and if the trades had some informational content.
9 Hypothesis 2: Insiders decreased their position in TARP receiving banks prior to the crisis. To explain this hypothesis, first a distinction needs to be made in the information the corporate insider has. The information can be about the financial situation of the bank itself or information indicating that the bank will receive TARP. If the insider has information on the, for example, deteriorating financial situation of the bank, he or she is better off selling the stock at that time than keeping it and watching the price drop. This is motivated by findings of Cziraki (2015), he finds the portion of insiders decreasing their ownership to be increasing for both high-exposed and low exposed banks. However, in high-exposed banks the fraction of insiders decreasing is 20% higher relative to low-exposed banks. Assuming that high-exposed banks are more likely to receive and accept TARP funding, hypothesis 2 is to be expected.
Hypothesis 3: There is more insider trading present in TARP receiving banks pre-crisis.
This hypothesis is backed by findings of Baghat and Bolton (2014), they find that excessive risk taking increases by managerial compensation incentives. These incentives can consist largely of stock and options in the bank, this leads to more insider trading in general for these banks because larger amounts of stock are owned by insiders. Most trading will be on the selling side as the CEO’s already own the stock. Prior to the crisis these insiders traded more, as they had more shares to trade. One important assumption is that the stock provided by the incentive program is not restricted, furthermore it is important to keep in mind that option grants are not included in this research. The excessive risk taking caused by the incentive program can lead to a more vulnerable bank which requires governmental assistance. The reasoning is as follows, banks receiving TARP are more at risk because extra funding is needed in contrast to a bank that is able to survive on its own. The higher the government fund is relative to the banks’ total value, the riskier the bank.
During the bailout period there is an effect expected in the opposite direction, banks that received TARP will be watched more closely and regulated by the government. Therefore, insiders are less likely to trade their shares.
By addressing both the returns of insiders and the amount of insider trading a broad field regarding insider trading is covered. Hypotheses 2 and 3 seem a little opposing, which makes the research more interesting.
10
3. Methodology
This chapter builds the models used to test the hypotheses discussed above. First, the abnormal returns generated by the insider trades need to be defined. After obtaining these, multiple regressions will test the above mentioned hypotheses on returns, ownership and trade activity.
3.1 Cumulative abnormal returns
The first step in this thesis is to calculated the cumulative abnormal returns for the insiders in all U.S. banks during 2004-2012. After this is completed, the CARs can be used as a dependent variable in the cross-sectional regression. These abnormal returns are calculated using an event study. Details on the event study will follow below.
Insider profitability will be measured by the abnormal returns around each insider trade, each trade will be an event. The profitability is the excess returns the insiders make by trading on their corporate information rather than investing in the market. The formula below defines the abnormal returns:
𝐴𝑅𝑖,𝑡 = 𝑅𝑖,𝑡− (∝̂𝑖+ 𝛽̂ 𝑅𝑚𝑖 𝑡+ 𝜀𝑖,𝑡)
Thus, the left hand side defined by the abnormal returns are is calculated by the actual returns minus the expected returns for each insider trade, this model is called the market model or CAPM. One could also use a constant mean return or the multifactor model. However, research indicates that using the four-factor model does not add significant benefits or accuracy to the calculated returns (MacKinlay, 1997). The expected returns are calculated by regressing the actual returns on a market index for a specified estimation period. The actual returns need to be calculated from the stock prices and are defined as follows:
𝑅𝑖,𝑡 = (𝑃𝑡+1− 𝑃𝑡)/𝑃𝑡
The stock price on the next day minus the stock price today, divided by the stock price today.
Some assumptions need to be made for the use of an event study and defining the interpretation of it. An important assumption that has to be made is market efficiency. The market has to be efficient enough to be able to incorporate the information contained in the trades and thus reflect the information in the stock prices. Only if this is the case, there will be abnormal returns for insiders because these are dependent on the market returns as shown in the formula above. Similarly, the market is only able to react if the trades contain new information. If all information is already publicly known, it is already incorporated into the stock price if the market is efficient (Fama et al. 1969).
Now, the cumulative abnormal returns can be found for the insiders, this is the sum of all abnormal returns over the specified event window. By analysing the CARs, both short term and longer
11 term, it is possible to see if insiders made profits by buying or selling their stock. The formula below shows how the CARs are defined.
𝐶𝐴𝑅𝑖,𝑡(𝑡1, 𝑡2) = ∑ 𝐴𝑅𝑖,𝑡 𝑡2
𝑡=𝑡1
Some important dates need to be discussed before continuing to the calculations. The event date is equal to the transaction date as reported by the insider, this is t=0. This date is generally different from the date the trade becomes known to the public, only after this the market can react. Insiders have two days to report their transaction to the SEC, therefore t2 is equal to t0 plus two days and plus the amount of trading days of interest. Three event windows will be used of 5, 21, and 63 days, which represent one week of trading days, one month and three months respectively. For t1 the day before the event is used, t=-1, the reasoning for this is to prevent biases in the results by the influence of the trade itself on the stock price. These definitions lead to event windows of [-1, 𝑡𝑎𝑛𝑛𝑜𝑢𝑛𝑐𝑒+ 5], [-1,
𝑡𝑎𝑛𝑛𝑜𝑢𝑛𝑐𝑒+ 21] and [-1, 𝑡𝑎𝑛𝑛𝑜𝑢𝑛𝑐𝑒+ 63], where 𝑡𝑎𝑛𝑛𝑜𝑢𝑛𝑐𝑒 is t=0 plus two days. To see if the returns
made by insiders are abnormal, a base case return of the market is needed. The estimation window for this will be preceding the event to make sure the expected returns are not biased because of influence of the event itself. Therefore, an estimation window of [-150, -30], or differently put 120 days, is used, as advised by MacKinlay (1997). Event studies seem to be biased on the long term so therefore no event windows longer than 63 days are examined (MacKinlay, 1997).
To test hypothesis 1, this research will use a difference-in-difference regression, containing variables of interest and important controls.
𝐶𝐴𝑅𝑖,𝑡 = ∝0+∝1𝑃𝑟𝑒𝐶𝑟𝑖𝑠𝑖𝑠𝑡+∝2𝐵𝑎𝑖𝑙𝑜𝑢𝑡𝑡+∝3𝑇𝐴𝑅𝑃𝑖+∝4𝑃𝑟𝑒𝐶𝑟𝑖𝑠𝑖𝑠𝑡∗ 𝑇𝐴𝑅𝑃𝑖+∝5𝐵𝑎𝑖𝑙𝑜𝑢𝑡𝑡∗
𝑇𝐴𝑅𝑃𝑖+∝6𝑇𝐴𝑅𝑃 𝑖𝑛𝑡𝑒𝑛𝑠𝑖𝑡𝑦𝑖+∝𝑛∑ 𝑍𝑖 + 𝜀𝑖𝑡 (1)
The first regression uses cumulative abnormal insider trade returns as dependent variable, for three event windows the abnormal returns are calculated. These windows represent, as mentioned above, one week, one month and three months of trading days.
The right-hand side contains a control for the pre-crisis period, this is included in the regression using a dummy variable. This dummy would have the value 1 if the time period is before the year 2008. The variable will have a value 0 if the trade occurred after this date. The decision to use pre-crisis and not crisis or post-crisis is for the convenience of interpretation. The pre-crisis period is of general interest to see whether the insiders were already aware of some information regarding the bailouts or the financial situation of the banks. The model also contains a dummy variable for the bailout period,
12 indicating that the time period is 2008 or 2009 with a value of 1. Both pre-crisis and bailout period’s coefficient will be relative to the post-crisis period, this period is from 2010 to 2012.
During the bailout period TARP was given out to several banks. Whether a bank received TARP funding during this period is incorporated using a dummy variable, TARP, indicating with 1 that the bank received TARP and 0 otherwise. The coefficient for TARP is expected to be positive according to hypothesis 1. An interaction term between these two dummy variables will indicate if there is a significant relationship between pre-crisis insider trading and banks obtaining a bailout afterwards. If there is, the coefficient ∝3 is expected to be positive. The following interaction term is between TARP
and the bailout period, indicating whether TARP receiving insiders made higher abnormal returns. The variable TARP intensity is the amount of funding received as a ratio of market capitalisation, which is calculated by the number of shares outstanding times the share price. This will control for variation in the amounts of funding received by the banks and the size of the institution, smaller banks are more likely to receive smaller amounts of funding but if for some reason a small bank received a large amount, this variable should control for that. The coefficient is expected to be positive, as the relative amount of TARP increases the stock price will likely rise more compared to lower or no TARP recipients.
Some control variables will be added to the regression to control for bank specific variation, these are shown in the model as the vector 𝑍𝑖. These variables include the logarithm of total book
value of market capitalisation, which will control for the banks’ size. The stock return on the day before the transaction and the average stock return of the previous month. These variables are followed by the market-to-book ratio, which is calculated as the ratio of market capitalisation and total assets. The return on assets (ROA) is defined as net income divided by total assets.
Finally, some variables on the riskiness and creditworthiness of the banks. S&P’s numeric is a variable containing the historical rating for that year as addressed to the bank by Standard and Poor’s, transformed to a numerical value, where 21 is the highest value indicating a AAA+ rating and 1 represents a rating of C. Leverage is calculated by dividing total liabilities by total assets, high leverage indicates the bank to be more risk-taking.
3.2 Changes in ownership
The second hypothesis can be answered using the difference-in-difference regression model below. The left hand side consists of the changes in ownership and the explanatory variables are similar as in the previous regression above.
∆ 𝑂𝑤𝑛𝑒𝑟𝑠ℎ𝑖𝑝𝑖,𝑡 =∝0+∝1𝑃𝑟𝑒𝐶𝑟𝑖𝑠𝑖𝑠𝑡+∝2𝐵𝑎𝑖𝑙𝑜𝑢𝑡𝑡+∝3𝑇𝐴𝑅𝑃𝑖+∝4𝑇𝐴𝑅𝑃 𝑖𝑛𝑡𝑒𝑛𝑠𝑖𝑡𝑦𝑖+
13 The dependent variable is measured by the change in ownership caused by the trades during a one-month period. More specifically, it is defined as the number of shares purchased minus shares sold over a one-month period for each bank. This results in a negative value for net selling shares and positive for net purchasing shares. Explanatory variables are dummy variables for the pre-crisis and bailout period, leaving the post-crisis period as the relative value. TARP and TARP intensity are also included; both are expected to have a negative effect on the change in ownership looking at hypothesis 2. The TARP interaction variables will indicate if there is a significant relationship in increasing or decreasing ownership for TARP recipients during the different periods.
The vector 𝑍𝑖 contains the control variables total shares traded measured over all banks, the
natural logarithm of the number of shares outstanding, the natural logarithm of the banks’ market capitalisation, the market-to-book ratio and the ROA. These variables are calculated in the same manner as for the CARs regression, only this time all values are monthly. For measures of risk the Standard and Poor’s rating is used plus the amount of leverage the bank has on its balance sheet. Interaction variables of these two variables with TARP are also included in the model.
3.3 Trading activity
Finally, the third and last hypothesis can be answered using a similar regression model using a different dependent variable, trading activity is used in this case. To measure the trading activity, the number of transactions are added up over a one-month period for each bank, so both purchases and sales transactions together form the total value in opposition to the change in ownership variable. This number of transactions is then divided by the total transactions calculated over all banks and finally the natural logarithm is taken. This is good for the fit of the OLS model, as the variable is highly skewed to the right, however interpretation wise is it less helpful. The model is as follows:
𝑇𝑟𝑎𝑑𝑖𝑛𝑔 𝑎𝑐𝑡𝑖𝑣𝑖𝑡𝑦𝑖,𝑡 =∝0+∝1𝑃𝑟𝑒𝐶𝑟𝑖𝑠𝑖𝑠𝑡+∝2𝐵𝑎𝑖𝑙𝑜𝑢𝑡𝑡+∝3𝑇𝐴𝑅𝑃𝑖+∝4𝑇𝐴𝑅𝑃 𝑖𝑛𝑡𝑒𝑛𝑠𝑖𝑡𝑦 +
𝑃𝑟𝑒𝐶𝑟𝑖𝑠𝑖𝑠𝑡∗ 𝑇𝐴𝑅𝑃𝑖𝑛𝑡𝑖+∝5𝐵𝑎𝑖𝑙𝑜𝑢𝑡𝑡∗ 𝑇𝐴𝑅𝑃𝑖𝑛𝑡𝑖+∝𝑛 ∑ 𝑍𝑖+ 𝜀𝑖𝑡 (3)
As in the previous regressions, there are dummy variables for the pre-crisis and bailout period as well as TARP. Both interaction variables are also present, only now the TARP variable is the TARP intensity. Higher trading activity is to be expected for banks facing higher risk, TARP intensity is used as a posteriori measure for risk. The higher the amount of TARP received relative to the total market value of the bank, the higher the risks the bank was facing thus needing a larger bailout sum. Positive coefficients are therefore expected for TARP. The coefficient for pre-crisis is expected to be negative, the crisis increased information asymmetry leading to more information advantage for insiders. The
14 bailout period coefficient is expected to be positive, the information asymmetry became larger during this period. The vector 𝑍𝑖 contains again the Standard and Poor’s rating, leverage, the logarithm of
market capitalisation and the ROA. The interaction terms of TARP (dummy) with the rating and leverage are also included. Next to these, the average amount of trades for each bank is added, the amount of transactions in the previous month, as well as the overall number of transactions over all banks.
15
4. Data
To be able to test the above explained regression models and hypotheses, a specific dataset is needed. This chapter will explain in detail how the data set is gathered, constructed and what manipulations are performed on the dataset.
4.1 Insider trades
The main dataset is that of the insider trades, if a bank did not have any reported insider transactions it is not included in the sample. The time period used for the research is January 2004 to December 2011, this surrounds the building up and the peak of the crisis by 2 years roughly. Most studies in the area of insider trading have used older datasets from before 2000. The use of more recent data will contribute to the relevance of this study as there were quite some changes in the legislation on insider trading in the past 10 to 15 years, for example the SOX-act in 2002 which limited the reporting period to two days.
Data on insider transactions is obtained from Thomson Reuters, specifically from table 1. This database contains all filed 3, 4 and 5 forms. For every insider is collected: their position, transaction codes, the number of shares traded, industry and sector codes. Two reported dates are gathered as well, the transaction date and the SEC date, the latter is the date when the trade becomes known to the public. The difference between these two dates can be two at maximum, as already mentioned in chapter 3. To obtain the required data sample the rough set needs to be adjusted accordingly. All sectors outside “01” are dropped as well as all industries outside “04”. This leaves only banks in the sample. Trades done by insiders other than CEOs are deleted from the sample, as trades by CEOs are considered to be most informative (Seyhun, 1990). Furthermore, trades by other insiders in the firm are likely correlated to those of the CEO. Transaction codes other than “P” and “S”, indicating purchases and sales respectively, are dropped from the sample. This takes out option grants, gifts and other non-open market transactions from the sample. By just keeping purchases and sales, only form 4 filings are left, which are used to report transactions to the SEC. A total of 8942 trades are left after these manipulations, each of which will represent an event for the event study.
The next step is to combine the daily stock returns for each bank and a market index with the insider trade sample. The daily trading volume, shares outstanding and closing prices are gathered from CRSP Security Daily. Using the closing prices, a stock return is calculated for each bank as well as the market capitalisation. The formula for calculating stock return can be found in chapter 3, whereas the market capitalisation is found by multiplying the number of shares outstanding by the closing stock price. Finally, a market index is needed to calculate the expected returns, CRSP provides a value-weighted index without dividends, included in this index are NASDAQ, NYSE, AMEX and ARCA. Daily values for this weighted market return are obtained from CRSP Stock / Security Files. The transaction
16 date and CUSIP are used to merge the daily stock prices and index into the trades data. The CUSIP codes from Thomson Reuters are 6-digit and those provided by CRSP are 9-digit, therefore the converter tool provided by CRSP is used to shorten the 9-digit codes to be able to perform the merge. By expanding the dataset in Stata for each transaction, it is possible to calculate the CARs with this dataset. Each transaction has a specific date, so for each of these the estimation window will be different. The CARs are calculated using the method as described in the previous chapter. After these manipulations one has an unbalanced panel data set with the reported transaction dates per bank and for each of these transactions the CARs for one week, one month and three months.
4.2 Bank sample
To be able to run the cross-sectional regression more bank specific variables are needed. This can be done by merging on CUSIP and year, as the balance sheet data is annual. Bank specific data and the daily stock data are both collected from the CCM: CRSP CompuStat Merged data base. This database contains both annual bank data from CompuStat Global and CRSP Security Files data, however it is more easily linked together through CCM compared to using the two separate databases. For each bank total assets, liabilities and net income are collected from CCM Banking. These are all annual values. CompuStat Capital IQ contains ratings by Standards & Poor.
Data on TARP recipients is provided from previous work by the supervisor of this thesis: dr. T. Jochem. The original data source is a quarterly inspection report for congress by the Office of the Special Inspector General (SIGTARP, 2012). The variable TARP is merged in as a dummy variable, a one indicating the bank received TARP. The RSSD codes provided by the FED can be translated to PERMCO identifications using the linking table as provided by the New York Federal Reserve. Having merged the bank specific data with the TARP and insider trades, this leaves a total dataset of 7297 insider trades spread over 2004 to 2012.
4.3 Other manipulations
The time period variables are added manually, pre-crisis is one for 2004-2007, bailout is one for 2008-2009 and this leaves post-crisis to be 2010-2012. TARP intensity is the amount of U.S. dollars received in millions divided by the market capitalisation in millions. Leverage is calculated by dividing total liabilities by total assets. ROA is defined as the net income divided by total assets. The market-to-book ratio is the ratio of the market value of equity, or market capitalisation, and the book value of equity as found in CompuStat. The S&P ratings are transformed from letters to numerical values, AAA+ indicates 21 and C represents 1. To create the control variable market capitalisation, the natural logarithm is taken for a better fit to the OLS model and the dependent variable. The same applies to the number of shares outstanding, both these variables are skewed to the right.
17 To be able to collapse to monthly observations an indicator is needed for each bank-month observation, after this it is possible to collapse the variables. The variables that do not change over a one-month period, such as assets, can be collapsed by using the first observation. The variables that do change are collapsed as a sum, this creates variables as total shares traded and total buy or sell transactions.
18
5. Results
This chapter discusses the results found by using dataset and employing the models as proposed in chapter 3 and 4. First, the descriptive statistics on all important bank-specific variables are examined. The CARs are examined next, and finally the cross sectional regressions to answer the remaining hypotheses addressed in chapter 2.
5.1 Descriptive statistics
To get a first impression of the data sample used in this research some general findings are discussed in this paragraph. Table 1, as found on the next page, presents summary statistics for the variables divided up in TARP receiving banks and banks that did not receive any governmental help. The total sample contains 309 banks of which 128 received TARP and the remaining 181 did not. Examining the balance sheet data of the banks, it can be seen that for TARP recipients the minimum and maximum values have much wider ranges and thus larger standard errors. This indicates that the type of banks receiving funding was not one particular size, but a combination of very large and relatively smaller banks. The dollar value of the market capitalisation shows the same evidence, larger range in values for TARP recipients. The relative amount of TARP money received ranges from 1.5% to 213%, which are quite extreme differences.
Leverage levels are very similar for both groups, with only very small differences in range and median. The TARP recipients received some lower scores by Standard and Poor´s, with 8 as lowest score whereas the lowest value for the non-recipients is 12. ROA is similarly distributed for both groups, as is the market-to-book ratio. The average number of shares traded in one transaction, traded shares, is more spread out for non-TARP banks. The number of shares outstanding, shown in the table in millions, is overall a bit higher for TARP recipients. Both the minimum and maximum amount of shares outstanding are above those of the non-TARP group.
The number of buy transactions does not show strong differences for the two groups, the distributions are very similar. For the number of sell transactions, it can be seen that the number of selling transactions is pulled up by larger amounts of trades for some banks in the TARP group. The median is equal to that of the non-TARP group, thus the mean is increased by the higher values on the right side.
It is important to keep in mind that these trades in the dataset are only the form 4 filings, as filed to the SEC by the insiders themselves. Of course, more insider activity occurred during this time period, incentive based pay often uses option grants for example. These are left out because too little insider knowledge is included in these activities.
19
Table 1 - Summary statistics bank sample
Non-TARP TARP
Variables N mean p50 sd min max N mean p50 sd min max
Assets 181 10,147 1,283 63,792 111.0 695,516 128 42,205 1,915 232,194 340.4 2.119e+06 Equity 181 919.2 107.1 5,848 13.50 65,568 128 3,667 160.8 19,727 16.35 172,339 Liabilities 181 9,191 1,148 57,642 90.72 625,224 128 38,310 1,758 211,722 284.9 1.940e+06 Income 181 100.7 6.384 753.4 -880.5 7,759 128 404.2 7.022 2,369 -1,218 18,801 Amount TARP 181 0 0 0 0 0 128 634.4 40.70 2,572 7 25,000 TARP intensity 181 0 0 0 0 0 128 0.377 0.256 0.380 0.0150 2.139 Leverage 181 0.906 0.908 0.0259 0.812 0.965 128 0.904 0.903 0.0197 0.827 0.954 S&P’s rating 181 20.42 21 1.854 12 21 128 19.69 21 2.812 8 21 ROA 181 0.00412 0.00667 0.0112 -0.0654 0.0266 128 0.00414 0.00614 0.00945 -0.0385 0.0197 Market capitalisation 181 1,668 138.6 10,582 12.26 111,029 128 4,889 180.1 23,432 4.917 222,063 Market-to-Book 181 0.135 0.128 0.0593 0.00677 0.360 128 0.115 0.103 0.0596 0.0108 0.338 Traded shares 181 10,896 1,815 74,557 33.46 1,000,000 128 7,968 2,440 15,672 50 125,083 Shares outstanding 181 44.34 9.727 209.1 1.248 2,106 128 151.0 13.57 586.0 1.506 5,064 Buy 181 9.657 3 16.76 0 116 128 10.41 5 15.82 0 105 Sell 181 8.282 1 23.28 0 205 128 21.23 1 73.14 0 588
Notes: this table provides the summary statistics for the total bank sample used to test the hypotheses, the sample contains 310 banks overall. Assets, Equity, Liabilities and
Net Income are presented in millions of dollars. The Amount TARP is zero for banks which did not receive any funding and a numerical dollar value in millions for banks that
did. Leverage is defined as Liabilities divided by Assets. S&P rating is the rating given out by Standard and Poor’s converted to a numerical value, where AAA is 21 and C is equal to 0. ROA is the return on assets for each bank, calculated as Net Income divided by Assets. Market cap is the number of shares multiplied by the number of shares standing out, presented here in millions of dollars. Market-to-book is market capitalisation divided by the book value of equity. Traded shares is the average amount of shares traded overall insiders per bank. Shares outstanding is the number of shares outstanding on average. Buy and Sell are the number of buy or sell trades respectively per bank. All data is calculated over the time period 2004 to 2012.
20 The second summary table contains the cumulative abnormal returns. For convenience the CARs are presented separately for sell and buy trades. As mentioned before, for insiders to profit from their trade the return should be positive for purchases, this means the stock price increased since the moment they bought the share and would be able to sell for a higher price. For sales the opposite is true, if the stock is sold and the stock price drops afterwards, this does not mean real cash for the insider but it means he or she prevented further losses by selling. These aspects can be seen in table 2 below, the mean and median are negative for sell transactions with the exception of the three-month CAR mean. Maximum values are positive, this can easily be the case if an insider only traded for liquidity reasons and not on behalf of any other motive. For purchases the mean and median are positive, as would be expected. Again, the minimum values are negative which can be caused by liquidity trading. However, it could also present an opportunistic insider trade which did not go as expected. Overall the values for buy transactions are more extreme and for all three CARs the standard deviation is higher than those of sell transactions. The sell transactions are approximately more normally distributed compared to buy transactions. This could be due to buying shares being a stronger signal than selling them, as often the insiders are granted shares as a reward.
Over the total time period of January 2004 to December 2012 there were more sales than purchases as reported to the SEC. At first sight this seems in line with insiders decreasing their positions, however it is important to keep in mind that this can also be due to CEOs selling their shares obtained by payment schemes. They might decide to sell when in need of liquidity or diversification.
Table 2 - Summary statistics CARs
(1) (2) (3) (4) (5) (6)
Sell
Variables N mean p50 sd min max
CAR1 4,217 -0.00439 -0.00456 0.0498 -0.823 0.747 CAR2 4,217 -0.0132 -0.0186 0.0881 -1.739 0.998 CAR3 4,217 -0.00218 -0.0141 0.188 -5.015 1.058 Buy
N mean p50 sd min max
CAR1 3,080 0.0175 0.00906 0.112 -0.889 1.126
CAR2 3,080 0.0309 0.0112 0.317 -3.209 7.478
CAR3 3,080 0.0619 0.0257 0.443 -8.939 7.622
Notes: this table contain the summary statistics for the cumulative abnormal returns of all three event windows, 5, 21, and 63 trading days over the period 2004-2012. The results are split up by type of transaction, buy or sell. The number of trades are presented by N, the mean and median follow in column 2 and 3. Abnormal returns are defined as the actual return of the banks’ stock minus the expected return using the market model.
21
5.2 CARs
In the previous paragraph the summary statistics of the found CARs were discussed, in this subchapter the CARs are tested to be significantly different from zero and the cross-sectional regressions will be discussed. The cumulative abnormal returns, CARs, are calculated as described in the methodology. The table below shows that the CARs for all three event windows are significantly different from zero with the exception of three month CARs for sell transactions. The coefficient is still negative, only no longer significantly different from zero for this window.
Table 3 - CAR significance
(1) (2) (3) (4) (5) (6)
Variables CAR1 CAR2 CAR3 CAR1 CAR2 CAR3
Constant 0.0175*** 0.0309*** 0.0619*** -0.00439** -0.0132*** -0.00218
(3.807) (2.597) (3.389) (-2.352) (-3.171) (-0.0777)
Observations 3,080 3,080 3,080 4,217 4,217 4,217
Cluster Yes Yes Yes Yes Yes Yes
Trade Buy Buy Buy Sell Sell Sell
Robust t-statistics in parentheses *** p<0.01, ** p<0.05, * p<0.1
Notes: This table shows regression results of an OLS for each of the event windows, divided up by buy and sell transactions, the calculated standard errors are robust and clustered at the bank-level. Each CAR is calculated as the sum of abnormal returns over 5, 21 or 63 trading days, representing one week, one month and three months. Abnormal returns are defined as the actual return of the banks’ stock minus the expected return using the market model.
It can be noted from the table above that over the entire research period more insider sales than purchases have taken place. As expected, the buy transactions have positive coefficients and the sell transactions have negative, all coefficients are significant at the 1 per cent level except for the three month sell transaction. This indicates that insiders in banks indeed do earn abnormal returns on both shares purchases and sales, however this dataset implies these are not necessarily present on the long term. Hypotheses 1 can already partly be answered looking at these results, overall the insiders in both groups of banks seem to be making abnormal profits with their trades. The returns on buy transactions are higher relative to returns of sell transactions, a buy order shows a return of 1.75% over 5 trading days, whilst a sell order has a coefficient of -.439%. This could be an indication that buying by insiders contains more information than selling, which is logical because insiders can sell their shares for liquidity reasons and not on market expectations.
22 The CARs have shown to be significantly different from zero, now regression model 1 can be estimated. As introduced is chapter 3, the model uses the CAR as dependent variable and TARP and time period variables to examine significant differences in these groups. Table 3 on page 24 contains the regression results. For each event window three different regressions are examined, an OLS only containing the binary variables for time periods and TARP and their interaction variables. The next two regressions are also OLS’s, however now also containing control variables. Column 2 is calculated using robust standard errors and column 3 has robust standard errors plus it is clustered at the bank level i. Clustering is important as it is very likely that the trades are correlated within a bank. Two tables are provided, one for buy transactions and one for sell transactions. This way the interpretation of the coefficients is more straight forward as a positive sign has a different meaning for a buy or sell order respectively, the same goes for a negative sign.
First, table 4 contains the regressions on sell transactions only, meaning a more negative CAR is more profitable for the insider. Column number 4 is the one of most interest, this is the OLS (ordinary least squares) including control variables and clustered at the bank-level during 2006-2011 for CAR1. The bailout period is the first significant variable, at 5% to be exact. The coefficient is 0.0458, indicating that selling during 2008-2009 was less profitable by 4.58%1 than selling post-crisis. Table 5 shows that
this variable is not significant for buy transactions, in fact none of the time periods are significant. Back to table 4, TARP is significant at 10%. This indicates an overall difference between the CARs of TARP and non-TARP banks. The coefficient is 0.0299, which means that for sell transactions the trades were less profitable for TARP insiders by almost 3%. Again, this variable is not significant for buy transactions in table 5. However, the TARP intensity variable is significant at 5% and positive for buy transactions, thus the higher the TARP amount is relative to the market capitalisation the higher the CARs are. This makes sense because the stock price is likely to rise more if more TARP is received, even though this higher amount is needed to cover the worse financial situation of the bank. For sell transactions there is no overall effect for TARP-intensity, there is however an effect for the interaction variable bailout*TARP. With a coefficient of -0.0441 and significance at 10% this implies that insider from TARP recipients made more profitable trades than those of non-TARP recipients, 4.41% higher returns for a one-week period. This is not the case for buy transactions according to table 5. The final interesting result from column 4 is the significance at 5 % for leverage, the coefficient is 0.165 for sell transactions. This means that for banks with higher leverage, the insiders made less profitable sell transactions. There is no significant result for table 5. Table 4 and 5 both show that for the one month CARs the
1 For buy transactions a positive sign indicates a profitable trade, for sell transactions a negative sign indicates
23 same results apply, even more significant, whilst for the three-month period not all variables remain significant.
Hypothesis 1 expects significant abnormal returns for buy and sell transactions for TARP receiving banks during the bailout period. To conclude on hypothesis 1, the insiders of both groups of banks were able to trade profitably, not only necessarily TARP recipients. Table 3 shows that the values for the CARs are significant and the signs fit the expectations. During the bailout period insiders of TARP recipients were able to generate higher abnormal returns for selling transactions, relative to non-TARP recipients. This finding is not true for buying transactions, however the non-TARP intensity does have a positive effect on the profitability of the purchase.
24
Table 4 - CAR sell transactions
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
Variables CAR1 CAR1 CAR1 CAR1 CAR2 CAR2 CAR2 CAR2 CAR3 CAR3 CAR3 CAR3
Pre-crisis 0.00713 0.0180*** 0.0180 0.0238 0.0399*** 0.0431*** 0.0431** 0.0582* 0.114*** 0.0362 0.0362 0.0600 (1.538) (3.055) (1.645) (1.542) (4.018) (3.907) (2.007) (1.950) (3.952) (1.226) (0.665) (0.782) Bailout period 0.0289*** 0.0386*** 0.0386*** 0.0458** 0.0710*** 0.0709*** 0.0709** 0.0869** 0.126*** 0.0892** 0.0892 0.123 (3.917) (4.802) (2.664) (2.592) (5.352) (4.671) (2.000) (2.108) (3.772) (2.252) (0.927) (1.111) TARP 0.00744 0.0173*** 0.0173 0.0299* 0.0567*** 0.0686*** 0.0686*** 0.0841*** 0.158*** 0.125*** 0.125** 0.154** (1.243) (3.088) (1.351) (1.811) (5.024) (7.299) (3.216) (3.066) (5.153) (5.209) (2.314) (2.032) TARP-intensity -0.00348 -0.0370* -0.0370 -0.0474 -0.00267 -0.0789** -0.0789 -0.0984 -0.190*** -0.0880 -0.0880 -0.0983 (-0.159) (-1.815) (-1.178) (-1.251) (-0.0861) (-2.285) (-1.334) (-1.419) (-3.454) (-1.098) (-0.580) (-0.514) TARP*bailout -0.0350*** -0.0351*** -0.0351 -0.0441* -0.136*** -0.112*** -0.112*** -0.118** -0.0398 -0.0423 -0.0423 -0.0725 (-4.130) (-4.426) (-1.637) (-1.694) (-9.246) (-8.545) (-2.839) (-2.577) (-1.104) (-1.518) (-0.405) (-0.590) TARP*pre-crisis -0.00483 -0.000324 -0.000324 -0.00875 -0.0400*** -0.0165** -0.0165 -0.0141 -0.120*** -0.0757*** -0.0757 -0.0839 (-0.870) (-0.0709) (-0.0259) (-0.525) (-3.624) (-2.079) (-0.609) (-0.358) (-3.941) (-4.223) (-1.327) (-0.991) ln(Market cap) -0.00531*** -0.00531*** -0.00644*** -0.0135*** -0.0135*** -0.0176*** -0.0154*** -0.0154** -0.0190** (-6.796) (-3.660) (-3.389) (-8.800) (-4.044) (-3.924) (-3.938) (-2.129) (-2.044) Return month-1 3.885*** 3.885** 4.094** 4.742*** 4.742** 4.723** 5.080*** 5.080 4.309 (6.674) (2.489) (2.230) (5.670) (2.420) (2.148) (3.657) (0.996) (0.754) Return day-1 0.299** 0.299 0.300 0.266 0.266 0.247 -0.219 -0.219 -0.196 (2.517) (1.589) (1.529) (1.480) (1.001) (0.919) (-0.601) (-0.514) (-0.448) Leverage 0.161*** 0.161** 0.165** 0.755*** 0.755*** 0.823*** 2.218*** 2.218*** 2.350** (4.438) (2.143) (2.027) (11.05) (3.205) (2.841) (13.19) (2.729) (2.345) Market-to-Book -0.0640 -0.0640 -0.0726 -0.122 -0.122 -0.139 0.310 0.310 0.365 (-1.173) (-0.852) (-0.784) (-1.145) (-0.939) (-0.881) (1.005) (0.863) (0.827) S&P's numeric -0.000789* -0.000789 -0.00137 -0.00394*** -0.00394** -0.00459** -0.00472*** -0.00472 -0.00283 (-1.870) (-1.219) (-1.364) (-5.499) (-2.572) (-2.011) (-3.158) (-0.944) (-0.410) ROA 1.041* 1.041 1.119 2.200** 2.200 2.411 4.350 4.350 4.421 (1.821) (1.442) (1.377) (2.023) (1.335) (1.351) (1.413) (0.968) (0.904) Observations 4,217 4,217 4,217 3,146 4,217 4,217 4,217 3,146 4,217 4,217 4,217 3,146 R-squared 0.016 0.132 0.132 0.140 0.082 0.188 0.188 0.207 0.068 0.172 0.172 0.201
Controls No Yes Yes Yes No Yes Yes Yes No Yes Yes Yes
Cluster No No Yes Yes No No Yes Yes No No Yes Yes
Time period 2004-2012 2004-2012 2004-2012 2006-2011 2004-2012 2004-2012 2004-2012 2006-2011 2004-2012 2004-2012 2004-2012 2006-2011 Robust t-statistics in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Notes: The three dependent variables are the cumulative abnormal returns around selling transactions for one trading week, one month and three months. Pre-crisis and bailout are time period dummy variables for 2004-2007 and 2008-2009 respectively. TARP indicates whether the bank received financial aid with a value 1. TARP intensity is the dollar value received divided by the market cap of the bank. As controls the log of market cap, previous month return, previous day return, leverage, market-to-book, S&P rating and the return on assets. The columns alternate clustering and time periods.
25
Table 5 - CAR buy transactions
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
Variables CAR1 CAR1 CAR1 CAR1 CAR2 CAR2 CAR2 CAR2 CAR3 CAR3 CAR3 CAR3
Pre-crisis 0.00534 0.00273 0.00273 0.00628 0.0218* 0.0701*** 0.0701*** 0.0788*** -0.00638 0.0808** 0.0808 0.0705 (0.993) (0.449) (0.267) (0.491) (1.694) (4.883) (3.254) (2.899) (-0.196) (2.248) (1.585) (1.062) Bailout period 0.0333*** 0.00807 0.00807 0.0129 0.113*** 0.0402*** 0.0402 0.0436 0.170*** 0.0981*** 0.0981 0.0835 (3.588) (1.131) (0.752) (1.109) (3.628) (2.805) (1.400) (1.474) (3.664) (2.994) (1.601) (1.226) TARP 0.00685 0.00333 0.00333 -0.00253 -0.0109 -0.0280* -0.0280 -0.0363 -0.00200 -0.0401 -0.0401 -0.0782 (0.750) (0.371) (0.266) (-0.172) (-0.679) (-1.849) (-1.314) (-1.471) (-0.0554) (-1.068) (-0.734) (-1.174) TARP-intensity 0.0451*** 0.0285** 0.0285 0.0493** 0.0953*** 0.0421*** 0.0421* 0.0653** 0.139*** 0.0825*** 0.0825** 0.107** (3.434) (2.295) (1.514) (2.098) (5.368) (3.107) (1.757) (2.388) (5.400) (3.522) (2.247) (2.112) TARP*bailout -0.0424*** -0.00662 -0.00662 -0.0116 -0.101*** 0.00855 0.00855 0.00255 -0.0923* 0.0330 0.0330 0.0492 (-3.309) (-0.618) (-0.407) (-0.692) (-3.025) (0.436) (0.241) (0.0713) (-1.803) (0.826) (0.464) (0.638) TARP*pre-crisis -0.0111 -0.00192 -0.00192 -0.00222 -0.00671 0.0114 0.0114 0.0139 0.0217 0.0464 0.0464 0.0738 (-1.356) (-0.251) (-0.176) (-0.169) (-0.435) (0.794) (0.525) (0.495) (0.608) (1.352) (0.922) (1.202) ln(Market cap) -0.00152 -0.00152 -5.07e-05 0.0167*** 0.0167** 0.0219*** 0.0187** 0.0187 0.0274*
(-0.859) (-0.487) (-0.0138) (4.452) (2.547) (2.898) (2.486) (1.381) (1.727) Return month-1 1.730*** 1.730 1.678 16.94*** 16.94*** 17.09*** 17.33*** 17.33*** 17.50*** (3.213) (1.637) (1.641) (13.23) (6.486) (6.585) (13.37) (6.642) (6.606) Return day-1 0.825*** 0.825*** 0.836*** 0.302** 0.302 0.305 0.761*** 0.761 0.753 (12.73) (8.759) (8.849) (2.466) (1.301) (1.303) (3.412) (1.367) (1.350) Leverage 0.0170 0.0170 -0.0178 -0.475*** -0.475** -0.538* -0.538** -0.538 -0.850 (0.231) (0.105) (-0.0905) (-3.765) (-2.072) (-1.950) (-2.239) (-0.846) (-1.076) Market-to-Book 0.0723* 0.0723 0.0618 -0.678*** -0.678*** -0.789*** -1.326*** -1.326*** -1.709*** (1.653) (0.866) (0.656) (-7.695) (-3.883) (-3.431) (-8.520) (-3.662) (-3.662) S&P's numeric -0.000160 -0.000160 0.000620 9.55e-05 9.55e-05 0.00150 -0.0144*** -0.0144 -0.0135 (-0.159) (-0.0702) (0.206) (0.0476) (0.0202) (0.237) (-3.938) (-1.460) (-1.042)
ROA -0.887*** -0.887 -0.835 -1.567*** -1.567** -1.318 -0.313 -0.313 0.388
(-3.335) (-1.342) (-1.192) (-4.232) (-2.003) (-1.647) (-0.453) (-0.236) (0.299) Observations 3,080 3,080 3,080 2,466 3,080 3,080 3,080 2,466 3,080 3,080 3,080 2,466 R-squared 0.025 0.267 0.267 0.276 0.019 0.690 0.690 0.704 0.031 0.412 0.412 0.436
Controls No Yes Yes Yes No Yes Yes Yes No Yes Yes Yes
Cluster No No Yes Yes No No Yes Yes No No Yes Yes
Time period 2004-2012 2004-2012 2004-2012 2006-2011 2004-2012 2004-2012 2004-2012 2006-2011 2004-2012 2004-2012 2004-2012 2006-2011 Robust t-statistics in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Notes: The three dependent variables are the cumulative abnormal returns around buying transactions for one trading week, one month and three months. Pre-crisis and bailout are time period dummy variables for 2004-2007 and 2008-2009 respectively. TARP indicates whether the bank received financial aid with a value 1. TARP intensity is the dollar value received divided by the market cap of the bank. As controls the log of market cap, previous month return, previous day return, leverage, market-to-book, S&P rating and the return on assets. The columns alternate clustering and time periods.
