The determinants of dividend payouts during the financial crisis of 2008-2009 : the case of US Banks

(1)

Julja Prodani

ID: 10825371

The Determinants of

Dividend Payouts

During the Financial

Crisis of 2008-2009: The

Case of US Banks

Universiteit van Amsterdam

Msc. Business Economics, Finance

Master Thesis

Advisor: Torsten Jochem

(2)

1

(3)

2

Abstract

The determinants of dividend payouts and the effect that agency conflicts have on dividend payouts is a well-researched topic. This paper studies the conflicts of interest that give rise to agency problems in a more comprehensive way, analyzing them not only among shareholders and insiders, but also bondholders and other taxpayers. We conduct a cross-sectional analysis of 340 US banks during the last quarter of 2008, which is the quarter in which TARP was launched. The shock of inflow of TARP funds into the recipient banks and the use of these funds from the part of banks sheds new light into the incentives governing bank dividend payout decisions in times of a financial crisis. Our empirical analysis documents a strong positive relation between TARP participation and dividend payout, which serves as evidence of risk-shifting from shareholders to debtholders and taxpayers. Specifically, our Logit model shows that TARP participation has a positive marginal effect of around 13% on the decision to pay out dividends; however, we do not find conclusive evidence on the effect of TARP participation on the decision to increase dividend payouts. Furthermore, we do not find conclusive evidence of agency cost theories related to private benefits of directors and executives. Lastly, we find that the most prominent determinant of dividend payouts is expected growth. Our findings have important policy implications that stipulate that government-funded capital injection programs either should not be undertaken, or should be accompanied with clear and enforceable requirements, such as the prohibition of dividend payouts.

(4)

3

I. Introduction

During the recent financial crisis banks have been heavily criticized for their risk taking decisions and disregard of social costs. One of the criticisms has focused on the large dividend payouts of stressed banks, almost concurrently to their receipt of government funds or insurance for their troubled assets. The relevance of this topic is reflected in the steps that are being taken by the Federal Reserve Board (2011) and the Basel Committee on Banking Supervision (2011) to increase oversight of the dividend policies of banks. This paper sheds light on the effect that the conflicts of interest among bank insiders, shareholders, debtholders, and other taxpayers have on dividend payouts. Therefore, this paper provides not only an analysis of the “fundamental” determinants of dividend payouts, but also an analysis of the effect of incentive-driven behavior of banks on dividend payouts during the recent financial crisis.

Our research is inspired from the launch of the TARP program, which was an exogenous shock to the usual bank behavior towards dividend payouts. This program has given rise to extensive debate not only because of the high costs borne by taxpayers, but also because of the uncertainty of the manner in which TARP funds were going to be used by banks. Following the bankruptcy of Lehman Brothers in September 2008, the US government had little choice but to “massively” intervene in the financial sector in the hopes of bringing back financial stability and restoring trust. We analyze a cross-section of 340 US banks, 158 (46%) of which have participated in TARP, during the last quarter of 2008.

Anecdotal evidence suggests that we may find that a number of large banks paid out large dividends in a short time frame before or concurrently with their reception of TARP funds. Significance of this evidence would mean that banks clearly favor their shareholders at the cost of society (taxpayers) and debtholders. If we will find this to indeed be the case, then we will also test whether the share holdings of executives and board members, which are the decision-makers on dividend policy, significantly positively affect these payouts. Finding significance would mean that indeed banks serve the needs of its board members at the cost of other affected parties.

(6)

5

Our research will analyze the degree to which TARP participation and insider holdings impacted the decision to pay out dividends. Although there is some previous research done on TARP banks, to the best of our knowledge there is no research that specifically tests the effect of TARP participation on dividend payouts. Our results have important implications not only on the implementation of government aid (capital injection) programs, but also on debt- vs. equity-based compensation packages of bank executive officers.

A. Organization of TARP

The receipt of government guarantees, capital injections, and insurance of banks’ troubled assets will be determined by their participation in TARP. On October 3, 2008, President George W. Bush signed into law the Emergency Economic Stabilization Act of 2008 (EESA), that created the Troubled Asset Relief Program (TARP), which authorized the Treasury Department to purchase or insure up to $700 billion of troubled assets. This amount was later reduced by the Dodd-Frank Wall Street Reform and Consumer Protection Act, but still remained at a significant $475 billion. Among different TARP categories, the Bank Investment Programs (BIPs) category is of the greatest importance to our study. Under the BIP, Treasury invested approximately $245 billion across five distinct bank programs, each with the purpose to achieve a different goal towards the aim of stabilizing the US banking system (TARP Programs, 2015). The following are the five bank programs encompassed by the BIP.

The Asset Guarantee Program (AGP), launched in January 2009, involved government support to financial institutions whose failure would have disrupted the banking system and the economy. The program stipulated a joint assistance from Treasury, the Federal Reserve and FDIC to support the value of certain assets held by qualifying financial institutions (FIs) by agreeing to absorb a portion of losses incurred on those assets.

The Supervisory Capital Assessment Program (SCAP) & Capital Assistance Program (CAP) focused on restarting lending by banks. The SCAP was a "stress test" of the nation's 19 largest bank holding companies, while the CAP was established to provide assistance to the

(7)

6

banks that were found to need but could not raise adequate private capital. The aim of these two programs was to restore confidence that participating banks had sufficient resources to withstand losses in an adverse economic scenario and to make it possible for them to raise private capital. Given that 18 out of the 19 participating FIs had adequate capital, CAP closed on November 9, 2009 without making any investments.

The Capital Purchase Program (CPP) was initiated with the purpose of stabilizing the banking sector and the whole financial system by injecting capital in viable financial institutions of all sizes. The CPP helped strengthen the capital position of FIs and build confidence in the financial system. There were 707 participating FIs, with the last investment made in December 2009. For the purposes of this paper we will focus on this TARP program.

The Community Development Capital Initiative (CDCI) was initiated on February 3, 2010 to help viable Community Development Financial Institutions (CDFIs) and provide financing to communities (especially low‐ and moderate‐income ones) that were underserved by traditional banks and consequently found it hard to obtain credit when the economy began its downturn. The program, which was completed in September 2010, had eighty four recipient institutions and amounted to an investment of approximately $570 million.

The Targeted Investment Program (TIP), launched in December 2008, gave the Treasury the right to provide additional or new funding to systemically important FIs.

The above-mentioned programs crystalize the magnitude and importance of TARP as a tool to stabilize the financial sector, and therefore place banks’ decision-making over the use of received funds at the center of the financial recovery. The first statutory report of the Congressional Budget Office (CBO) on TARP transactions concludes that as of December 31, 2008, the transactions totaled $247 billion. At the time, CBO estimated that the subsidy cost of those transactions (the difference between what the Treasury paid for the investments or lent to the firms and the market value of those transactions) amounted to $64 billion (TARP Report on Transactions 2008). However, the debt held by the public had actually increased by $247 billion, as opposed to only $64billion, given that Treasury had to borrow all the

(8)

7

amount disbursed. Out of this cost, $178 billion was spent in the CPP through purchases of preferred stock and warrants from 214 U.S. FIs.

The preferred shares purchased as part of the CPP had “perpetual life”, i.e. no maturity, and were redeemable at any time after three years from FI’s reception of capital from CPP. They were also accompanied by warrants that allowed the government to purchase common stock for 15% of the amount invested in preferred stock, at a price equal to the average price for the 20 trading days preceding the date of the FI’s sale of the preferred shares. For as long as the shares were not redeemed, participating FIs were required to pay a dividend (on preferred shares) of 5% of the government’s investment in that FI for the first five years, and 9% of that amount thereafter. In cases in which a financial institution did not have shares of publicly traded common stock, the warrants were used to acquire additional shares of preferred stock. Lastly, TARP participation imposed some restrictions on executive compensation, dividend payments to shareholders, and amount of common stock repurchases.

One area that still remains gray is the participation criteria for TARP. Nevertheless, declarations from officials make clear that the program was voluntary and that banks were urged to participate in it for as long as they were relatively “healthy”. However, during the life of TARP many banks that did not have a strong capital position also got funds.

Hereafter, when we mention TARP we specifically mean the BIP category of TARP.

B. Criticisms

The receipt of TARP funds, i.e. taxpayer money, put banks under intense scrutiny. Everyone was concerned with banks’ capital strength, risk-taking decisions, use of funds, and most importantly repayment of TARP money. The public raised many questions and commentators answered with many criticisms towards banks. Suddenly criticisms were proliferating across conferences and news articles. Starting from October 2008, executive bonuses, golden parachutes, and dividend payments of TARP banks almost concurrently to their receipt of TARP funds received a lot of scrutiny and criticism (Liesman 2009). The

(9)

8

arguments doubted the rationality of either banks getting government money if they already had enough money to pay dividends, or of banks that need government money and still pay out large dividends.

As of October 30, 2008, U.S. banks were on their way to pay more than half of the sum they received from TARP for new lending to their shareholders over the next three years (Appelbaum 2008). The banks were prohibited to increase dividend payments without government permission and to repurchase stock, but they were already paying out huge dividends before the crisis and had permission to keep paying that rate. Given that these banks were financially distressed, there was an inherent conflict of interest between shareholders and debtholders because in case of bankruptcy each dollar paid as dividend would lessen the amount that creditors could seize. Furthermore, in this specific case, the conflict of interest extended to taxpayers, as it was their money that financed TARP funds and therefore dividend payouts. In TARP, taxpayer protection was only reflected insofar as the legislation required banks that sold “bad” assets to the government to provide warrants and therefore allow taxpayers to benefit in case of banks’ future growth (US Senate Committee on Banking). However, the Treasury did not have much say on the decision-making of TARP banks; it received warrants only for non-voting shares, or agreed not to vote the stock (EESA 2008).

The question then is why these banks chose to continue paying or even increase dividend payments in the period of crisis buildup while they were incurring losses. Banks explained that the dividends were needed to support their stock price, but economic literature says that healthy firms do not need to pay dividends every quarter, because retained cash can be paid out later or can be used to repurchase the bank’s stock. Therefore, the reason banks maintained large dividend payouts in a time of high perceived risk, difficulty of borrowing, and low capital buffers could be that insiders benefited from these payouts. The directors of some banks were among the leading beneficiaries of the dividend payouts (Scharfstein, Stein, 2008).

The condemnation of taxpayers and economists was not directed only towards banks; the government’s behavior over bank accountability fueled a lot of dissatisfaction. As of February 9, 2009, The Congressional Oversight Panel still did not know what banks were

(10)

9

doing with taxpayer money (Congressional Oversight Panel 2009). Furthermore, the whole purpose of TARP program was being questioned: did TARP actually strengthen banks’ capital position, did participating banks increase lending to boost the real economy, and did the government have the power and willingness to control the behavior of TARP banks? Many empirical papers have investigated into moral hazard and risk-shifting incentives of banks, agency conflicts and dividend payouts. The following is an overview of the related literature that will help put this article into perspective.

II. Literature Review

For simplicity, we divide the related literature into two sub-groups: one comprises conflicts of interest, agency costs, risk-shifting and moral hazard, and the other comprises the “fundamental” determinants of dividend payouts.

A. Literature on conflicts of interest, agency costs, moral hazard,

and risk-shifting incentives

Agency issues are of importance to our topic because of the importance of corporate governance on firm performance and because we are considering the ability of insiders to influence dividend payments in their benefit. Figure I schematizes the main conflicts of interest among insiders, shareholders, debtholders, and taxpayers. The ability of executives (CEO, CFO, COO, and other officers) to influence decision-making, including dividend payout decisions (Grove (2011)) is curbed by the board of directors; if the board-executive officers governance is not separated appropriately, insiders influence decisions (including dividend policy) more easily. Grove (2011) also finds that separating the roles of CEO and chairman of the board can significantly mitigate agency costs.

(11)

10

Studies show that corporate governance mechanisms related to risk management are associated with a better bank performance during the financial crisis of 2007-2008 (Aebi 2012) and less risk-taking. With regard to the effect of insider debt vs equity holdings and bank risk-taking, Bekkum (2014) finds that inside debt compensation held by top officers of U.S. banks (in the end of 2006) is significantly negatively related to risk (as measured by 2007-2009 bank risk exposure in terms of lost stock market value, volatility, tail risk, and the probability of financial distress) and risk-taking. Furthermore, banks whose managers had large inside debt holdings had a higher quality of assets and a more conservative balance sheet management. Their results favor the theory that debt-based executive compensation curbs risk-taking incentives. Similarly, many authors linked the equity-based CEO compensation (stock options and firm equity) to risky bank policies before the financial crisis (DeYoung et al. 2013). Stakeholders of banks are numerous - debtholders, shareholders, the government. Still, even though more than 90% of the banks’ balance sheet is debt, shareholders control the firm. Studies show that both the boards and the compensation package for CEOs represent the shareholders’ preference for increasing risks (Mehran 2011).

especially in case of bankruptcy, each $ paid to shareholders as dividend lessens the amount that debtholders can seize

Figure I. The main conflicts of interest among insiders, shareholders, debtholders, and taxpayers

agency costs; the higher insiders' equity holdings,

moral hazard related with

banks´ too-big-to-fail status distribution of TARP funds the higher the risk-shifting incentives and risk-taking decisions

Taxpayers Shareholders

Shareholders other than

insiders

Insiders that are

predominantly shareholders

(12)

11

Srivastav (2014) also studies the effect of insider debt on banks’ payout policy. He shows that CEOs with higher inside debt relative to equity are associated with more conservative bank payout policies. CEOs paid with more inside debt reduce payouts through a decrease in both dividends and repurchases. Their findings hold also for a subsample of TARP banks, where the risk-shifting consists in diverting to shareholders not only debtholders’ money, but also taxpayers’ money. During the past five years, there have been many policy discussions aimed at turning more bank employees into holders of inside debt (Liikanen Report, 2012). Substantial research done on dividend payments as a tool to reduce information asymmetry and agency costs confirms the use of dividends (larger and smoother) as a compensation for agency costs (Brockman (2014). This research is consistent with previous research from Jensen (1986) that finds that dividends help address the agency costs of free cash flow between insiders and outsiders because they represent an ongoing commitment to pay out cash.

From the above paragraph we conclude that most of the evidence suggests that inside debt helps curb risk-shifting incentives by aligning the interests of CEOs to those of creditors and regulators, and in the case of TARP banks to the interest of taxpayers. These issues have given rise to a wide discussion of managerial compensation contracts and the need for inside debt compensation.

Kanas (2013) studies the relation between bank dividends and bank risk from 1984 to 2011, and confirms the existence of risk-taking and risk-shifting in US commercial banks subject to regulatory regime changes. Though the period under consideration includes TARP, his focus is on the introduction of the PCA regulation in 1992. The paper documents strong evidence of risk-taking and risk-shifting in the post-PCA period. Risk-taking is the causal link from bank dividends to bank risk, through the effect that dividends have on a bank’s ability to build a solid capital buffer and therefore its risk (Onali, 2010). Thus, risk-taking implies a positive link from dividends to risk (Laeven, 2002); evidence of this link in the post-PCA period means that PCA, or regulation in general, is ineffective in controlling risk-taking.

On the other side, risk-shifting is the causality link from risk to dividends. Dividends are a means of transferring wealth from the other stakeholders of the bank to its

(13)

12

shareholders, lessening the impact of the probability of default on shareholders and increasing this negative impact on debtholders. The regulatory regimes can impact risk-shifting incentives through the use of capital requirements, which force banks to internalize some of the negative externalities of default. Higher capital requirements make dividend payouts more costly, as there is less money available to invest in growth (John et al., 2000). If capital requirements are effective in controlling risk-shifting, the benefits of risk-shifting will be offset by the opportunity costs triggered by capital requirements.

Dividend payouts of banks are a type of risk-shifting from other stakeholders to shareholders, basically diverting cash from the former to the latter and leaving the latter with the riskier and more illiquid assets. Therefore, it is important to know how these risk-shifting incentives can be influenced. With respect to the conflict of interest between debtholders and shareholders and resulting dividend policies, Brockman (2009) finds that country-level creditor rights influence dividend policies as a means towards the equilibrium balance of power between debt and equity claims. In the instances when creditor rights are weak, creditors want a more restrictive payout policy and managers agree with it. Especially in troubled firms, CEOs face a trade-off between an increase in current payouts to the benefit of shareholders and a preservation of cash for debtholders in case of default. The dividend cash payments of TARP banks close to the time of receipt of government aid were a clear example of protecting shareholder interests as opposed to debtholder or taxpayer interests. Therefore, inside debt could have been effective in mitigating risk-shifting incentives in TARP banks.

Finally, to shed some more light into further insider interests and bank incentives, we present a series of papers that look into the factors associated with a bank’s early exit from TARP. As it was frequently announced, executive pay restrictions were in many cases the rationale for early TARP repayments. Both Wilson (2012) and Bayazitova (2012) indeed find that high levels of (excess) CEO pay in 2008 were associated with significant early exits and increased the likelihood of early repayment. Nevertheless, Lei Li (2013) also finds that on average TARP banks employed about one-third of their TARP capital to support new loans and kept the rest to strengthen their balance sheets. However, Black (2013) documents risk-shifting incentives as he finds that for TARP banks, relative to non-TARP banks, the risk of loan originations increased at large banks. To further clarify the picture, it is worth

(14)

13

mentioning that while small TARP banks maintained a level of large commercial and industrial (C&I) loans comparable to non-TARP banks, big TARP banks decreased loan volume. Therefore, for large TARP banks the higher risk of loan originations was not caused by increased loan amounts but rather by originating higher-risk, higher-interest loans. This hints to the fact that participation in TARP worsens the risk-shifting incentives that stem from moral hazard related with banks’ too-big-to-fail status.

Furthermore, there are a series of other papers that confirm a disregard of social costs from the part of bank insiders during the financial crisis, although not in light of dividend payouts. These articles that analyze insider behavior find evidence of agency costs. For example, Cziraki (2014) analyzes insider trading patterns of bank insiders and finds that they foresaw the poor performance of their bank and traded during 2006 to earn predicted risk-adjusted stock returns of 40% during the crisis. Furthermore, with the start of the decline in US housing prices, the insiders of high-exposure banks were 20% more likely to sell stock than insiders of comparable firms.

While this section has analyzed agency costs and their effect on dividend policy, the next section will analyze dividend policy in light of bank fundamentals.

B. Literature on the determinants of dividend payouts

Analyzing thirty years of data on US financial companies, Floyd (2014) documents the existence of a growth in payouts over the last 15 years that is fueled by increased repurchases, and especially by an explosive growth in payouts in the years prior to the crisis. At the beginning of the crisis, financial institutions (including TARP banks) demonstrate a reluctance to cut dividends. There are three fundamental theories related to dividend payout decisions. The signaling hypothesis theory states that dividends are used as an indicator of future prospects. The regulatory pressure hypothesis states that undercapitalized banks tend to retain earnings rather than pay dividends. The Fama-French factors hypothesis stipulates that there are three fundamental determinants of bank payouts- size, profitability, and growth-, all of which are positively related to dividend payouts.

(15)

14

The results of empirical research on the signaling hypothesis are controversial. Flibeck and Mullineaux (1993) use an event study and find that unexpected dividend announcements have a direct impact on equity valuation, therefore not rejecting the signaling hypothesis. Collins et al. (1994) use a Generalized Least Squares regression with the dividend-to-earnings ratio as a dependent variable and find that the market-to-book equity ratio is negatively related to dividends, therefore not supporting the signaling hypothesis. Boldin and Leggett (1995) use an ordered probit response model to find that dividend payout ratio positively impacts banks’ quality (the dependent variable measured by banks’ ratings), therefore not rejecting the signaling hypothesis. The findings of a recent article from Floyd (2014) also go in line with the idea that banks use dividends to signal financial strength. Flibeck and Mullineaux (1999) use an OLS regression with dependent variable the abnormal return associated with the announcement of dividend increases and are not able to reject the agency hypothesis. Lastly, Abreu and Gulamhussen (2013) are able to reject the signaling hypothesis during the pre-crisis period, but not able to do so during the crisis period.

Empirical research has not drawn a unified conclusion about the applicability and significance of the Fama-French determinants of dividend payouts either. Theis and Dutta (2009) use an OLS regression in which the dependent variable is dividend-to-share price ratio. They do not find support for the Fama and French (2001) characteristics of dividend payouts of banks (size, profitability, and investment opportunities). On the other hand, Abreu and Gulamhussen (2013) use a Tobit model with the dividend-to-total asset ratio as the dependent variable and find that the Fama-French (2001) determinants of dividend payouts are significant.

Regarding the regulatory pressure hypothesis, there are two main contributing papers with differing results. Theis and Dutta (2009) find a positive impact of capital on dividends, therefore not rejecting the regulatory pressure hypothesis. Abreu and Gulamhussen (2013) test the determinants of dividend payouts for the pre-crisis period (2004-2006) and crisis period (2007-2009) and find evidence of the significance of the regulatory pressure hypothesis only during the crisis period.

(16)

15

C. Hypotheses

There are three natural questions that arise from the related literature. In this financial crisis period, controlling for other determinants of dividend payouts, is TARP participation positively associated with dividend payouts? If yes, this would be evidence of risk-shifting from debtholders and taxpayers to shareholders. The second question that arises, which will be our second hypothesis, is whether banks with higher insider holdings had significantly higher dividend payouts, after controlling for other determinants. A verification of this hypothesis would support the agency cost theory, which stipulates that the agents (management) take decisions in their own interest to the detriment of the principals. The last hypothesis, which in a way combines the first two, is whether dividend payouts of banks that participated in TARP were higher for those participative banks with higher insider holdings. All three hypotheses will be first tested by comparing the “no dividend payment” versus “dividend payment” states in a Logit regression, and then tested in a Tobit regression. The purpose of this two-layer analysis is to see not only whether TARP funds and insider holdings influenced the binary decision of paying vs. not paying dividends (in the Logit regression), but also to account for the variation in dividend payouts by using a Tobit regression.

III. Methodology

We conduct a cross sectional analysis of the determinants of dividend payouts for 340 US banks during the last quarter of 2008. This time period was chosen as the most relevant to our discussion of bank behavior and incentives for the following reasons. The fourth quarter of 2008 had the highest number of bank subscriptions in TARP (one-third of all 869 participating banks). Furthermore, this period was the least affected by legislative constraints - stricter rules that prohibited large dividend payouts and compensations were put in place starting from mid-February 2009- and external forces such as public condemnation; 2009 was characterized by an overwhelming amount of negative publicity by the media, condemnation by the public, and scrutiny by the regulators regarding all incentive-based bank decisions, including dividends. Therefore, we do not extend our

(17)

16

analysis of dividend payouts during 2009 because many banks engaged in early repayment of TARP funds in order to avoid the restrictions on dividend payments and executive compensation imposed by TARP and therefore their dividend payout decisions could be influenced by their desire of early repayment and less by their inherent incentive-based behavior, producing noise and raising endogeneity issues. The analysis of dividend payments does not consider the previous quarters of 2008 either, because TARP was a last-minute announcement and banks couldn’t have known that they would get government money to aid their recapitalization. Therefore, the link between early dividend payment in expectation of TARP funds, which we want to test, could not have been plausibly existent. Given the above-mentioned reasons, we analyze only dividend payouts of the last quarter of 2008 and the increase in dividend payouts from the third to the fourth quarter of 2008.

Our sample, though originally much bigger, is now comprised of 340 banks. The reasons for this reduction in the sample size are first the exclusion of banks that participated in TARP starting from 2009 and second the exclusion of banks with missing data on some of the explanatory variables. From the 340 banks in our sample, 158 (46%) have participated in TARP, while 182 have not. The steps that we will follow in our analysis will be the following. First, we will analyze our sample to see whether there is an increase in dividend payouts from the third quarter of 2008 (Q3 2008) to the fourth quarter of 2008 (Q4 2008). We will check not only for an aggregate increase in dividend payouts, but also for a difference in dividend payouts of TARP banks as compared to non-TARP banks. Then, we will check whether the dividend payout increase is higher for TARP banks compared to non-TARP banks. Second, if we indeed find evidence of an increase in dividend payouts, we will empirically check if TARP participation affected dividend payouts using both Logit and Tobit models. Third, if we find evidence in support of a positive association of TARP with dividend payouts, we will also check whether insider ownership played a role in dividend payouts using again Logit and Tobit models.

The following is the rationalization of our choice of these two econometric models. We will first use the Logit model and present the results in terms of marginal effects. For this probabilistic model, the usual coefficients are not easily interpretable while the marginal effects effectively show the marginal effect that a one-unit increase in the explanatory variable has on the dependent variable. Nevertheless, the dependent variable

(18)

17

of the Logit model is binary and therefore does not enable us to make use the variation on dividend payouts. This is the reason why we also run a Tobit regression, which will allow us to see how TARP participation and insider ownership affect dividends payouts by making use of the continuous distribution of dividend payouts. The Tobit regression is very similar to the simple OLS regression, with the only difference that it does not allow for negative values, i.e. it is censored to the right at zero. It is logical that we use a Tobit regression given that the dividend payout variable, defined as dividends over assets, cannot take negative values.

The main regressions from which we will draw our conclusions are of the following form. Through both Logit and Tobit, we will first test for the effect of TARP participation on dividend payouts and control for fundamental determinants. This regression will have the following form:

Dividend payouti= β0+ β1sizei + β2*profitabilityi + β3* expected growthi + β4* historical

growthi + β5*Tier1 ratioi + β6*TARP Dummyi+ ei

If we indeed find a positive association of TARP participation with dividend payouts, we check whether insider ownership was one of the drivers of these payouts. As mentioned in the literature review, Scharfstein and others argue that payouts were positively related with insider ownership. In the following regression we test the role that insider ownership played in dividend payout decisions, while controlling for all the above-mentioned variables.

Dividend payouti= β0+ β1sizei + β2*profitabilityi + β3* expected growthi + β4* historical

growthi + β5*Tier1 ratioi + β6*TARP Dummyi+ β7*Insider ownershipi + β8*TARP

Dummy*Insider sharesi + ei

The dummy for TARP participation was included to test our first hypothesis of whether TARP participation was positively associated with dividend payouts, i.e. if there is evidence of risk-shifting from shareholders to debtholders and taxpayers. To test our second hypothesis of whether stock holdings of the bank’s board of directors and officers are positively associated with dividend payouts and conclude on the existence of agency costs, we include their share ownership as another independent variable. To test our third hypothesis of whether, from only TARP banks, insider holdings were positively associated

(19)

18

with dividend payouts, we include an interaction term of TARP dummy and insider ownership variables. To find evidence in favor of our hypothesis we would need a positive coefficient for β6, β7, and β8.

We draw upon related literature to include the most common determinants of dividend payouts as control variables and test three other main theories related to dividend payouts: the applicability of Fama-French characteristics of dividend payouts, the signaling hypothesis and the regulatory pressure hypothesis. More importantly, from this regression we can rightly conclude on the significance of conflicts of interest, agency issues and risk-shifting after controlling for all other drivers of dividend payouts and across-entity differences.

First, we include the Fama-French determinants of dividend payouts, namely size, profitability, and growth opportunities. The size of the bank is represented by the natural logarithm of quarterly assets. The logarithm is taken in order to smooth the distribution of asset values and lower the standard deviation of this variable. A positive coefficient would mean that the bigger banks are more likely to pay more dividends. This is expected, given that large banks are more difficult to monitor (more opaque) and more likely to raise capital in the equity market. Profitability, similarly to related literature, is calculated as quarterly net income divided by quarterly total assets. A positive coefficient for profitability would mean that the better the company generates cash, the more it has to pay out as dividends; on the other side, it can be that the more profitable a company is, the less dividends it wants to pay because it can rather reinvest them and be profitable. The effect of growth opportunities, for as long as they are recent past opportunities, is captured by the historical growth variable. This variable is calculated as the rate of growth of assets from 2007 to 2008. We expect a negative coefficient associated with the historical growth variable, which would mean that banks that have grown recently have not paid dividends but rather plowed back their earnings to avoid costly refinancing.

The signaling hypothesis foresees that high dividends not only signal good prospects of the bank, but also increase the potential for further equity financing as equity-holders are more satisfied. This hypothesis should be reflected by a significant positive coefficient of the expected growth variable, defined as the ratio of market value of equity to book value of

(20)

19

equity. The reasoning behind its construction is that the markets are efficient and therefore incorporate expectations in the current market price, while the book value does not incorporate expectations. Nevertheless, we can also expect a negative coefficient of growth opportunities, because the higher the growth opportunities of the bank, the more it will save money and plowback earnings to direct it into these projects and avoid costly refinancing rather than paying it out as dividends.

The regulatory pressure hypothesis foresees lower dividend payments for undercapitalized banks, given that they are expected to retain their earnings to strengthen their capital position rather than distribute them. Given that the regulators have placed great importance on Tier 1 capital as a measure of over/ undercapitalization, we use the risk-adjusted Tier 1 ratio, defined as tier 1 capital over the risk adjusted assets, as the variable that reflects the regulatory pressure. Following our reasoning, we expect a positive coefficient associated with Tier 1 ratio, which would mean that the better-capitalized banks pay out higher dividends.

Our dependent variable, the dividend payout, is defined in two different ways. First, it is defined as “quarterly cash payments on common stock divided by quarterly total assets” in the fourth quarter of 2008 (in both Tobit and Logit regressions). Needless to say that in the Logit regression this is a binary variable that takes the value of 0 if the bank does not pay dividends and the value of 1 otherwise, while in the Tobit regression this variable is continuous. Then, the dependent variable is defined as “the increase in quarterly cash payments on common stock divided by quarterly total assets from Q3 2008 to Q4 2008”. In the Logit regression this variable takes the value of 1 if the bank has a higher dividend payout in Q4 2008 than in Q3 2008 and the value of 0 otherwise, while in the Tobit regression this variable is continuous. The dividends are scaled by total assets, as opposed to earnings or share price, to ensure that the variable is not driven by the stock price or earning volatility that were predominant during the crisis. The dividend payments on preferred shares are excluded as by definition the TARP banks were obliged to pay a 5% dividend on preferred shares to the US Treasury for the first five years of their reception of TARP money. Therefore, including preferred dividends in our dividend payout variable would unduly enforce the positive effect that TARP participation had on dividend payouts without providing any insight on the actual incentives that gave rise to the payout.

(21)

20

IV. Data and Descriptive Statistics

The data needed for this empirical research is fundamental data taken from banks’ financial statements and insider ownership data. Our analysis uses quarterly data of the pre-last and pre-last quarter of 2008 for a sample of 340 US banks.

To answer the first hypothesis of whether TARP banks paid out more dividends than non-TARP banks, we need data from TARP fund distribution. This data is extracted from the Economic Research website of the Federal Reserve Bank of St. Louis, which originally gathered the data from The US Department of the Treasury and the National Information Center. To answer the second hypothesis of whether insider ownership is positively associated with dividend payouts we need data regarding insider ownership, which we hand-collected from the Def 14A (proxy) filings of all banks in our sample. The proxy that we chose was the closest filing prior to the reception of TARP funds, assuming that ownership of these insiders does not change much during a short time given the requirements of reporting all major stock transactions. In line with the information from the Def 14A and with the rationale of dividend-payout decision-making, we defined the insider ownership variable as the total number of shares owned by the board of directors and officers (managers) of the bank; the decision of whether to pay dividends and their magnitude is completely determined and announced by the board of directors. Given that in many banks the chief officers are part of the board or have a means of influencing the board’s decision, we deem that the ownership of officers is also of importance to analyze the effect that stock ownership has on dividend payout decisions. In the Def 14a filings, we find the necessary data under the section “Security ownership of directors, executive officers, and principal shareholders”. In the table that comprises this section, we add only the share ownership of directors and executive officers, as the principal shareholders are outsiders.

The rest of the data, such as dividend payments and control variables, is taken from the Compustat database in quarterly frequency for the last quarter of 2008.

Table I presents the descriptive statistics for the variables used in our regression. All variables are in a quarterly frequency; historical growth is the only variable taken as the percentage growth of annual assets from 2007 to 2008, because this one-year period

(22)

21

ignores seasonal fluctuations and noise. From the table we see that the standard deviation of all variables is small, except the deviation of insider holdings. The mean of almost all variables, as expected, is positive; however, given the poor performance and losses that characterized this period of crisis, the average profitability of banks in our sample is -0.2%. An interesting observation is the average risk adjusted Tier 1 ratio, which is 11.2%; this ratio is relatively high, considering that the fourth quarter of 2008 was in the midst of the crisis.

To gain more insight into the relations between our variables, we present a table with pairwise correlations (Table II). We note that participation in TARP is highly correlated (with a 0.31 correlation coefficient) with the size of the bank as represented by its total assets. This fact does not come as a surprise as we might expect that bigger banks, which are those that cause the most disruption if they fail, are more likely to receive government support in the form of TARP funds. The negative correlation between TARP participation and profitability might reveal the fact that TARP funds were given to banks that were incurring more losses than their counterparties; nevertheless, this correlation coefficient of -0.09 is not very substantial Another interesting fact to notice is the negative relation of TARP participation with the expected growth; this would imply that government was more

Variables # observations Mean Std. Dev. Min Max

Dividend payout Dummy 340 0.79 0.40 0 1

Div. payout increase Dummy 340 0.48 0.50 0 1

Cash dividend 340 16.44 111.12 0 1483

Increase in cash dividend 340 0.95 9.71 -82 104

Dummy TARP 340 0.46 0.50 0 1

Interaction 340 1481786 3367906 0 23600000

Insider shares 340 2889640 5499377 59807 5.24E+07

Insider ownership % 340 0.18 0.76 0 13.84

Ln assets 340 7.58 1.51 4.74 14.59

Profitability 340 0.00 0.01 -0.06 0.01

Expected growth 340 0.93 0.56 0.10 3.09

Risk adjusted Tier 1 ratio 340 11.21 3.61 0.00 24.51

Historical growth 340 0.12 0.28 -0.16 4.47

(23)

22

inclined to give funds to banks with lower foreseen growth, maybe in the hopes of enhancing growth opportunities for these banks and propping up the use of TARP funds for good investments. Lastly, it is important to note that there does not exist a strong correlation between the risk adjusted Tier 1 ratio and TARP participation; this might serve as evidence to counter the claim that a bank´s strength of capital position as represented by the Tier 1 ratio might have been a criteria for TARP participation. However, we should not jump into conclusions without considering that this low correlation coefficient might be so just because the banks in our sample already have a relatively high risk adjusted Tier 1 ratio of around 11%. For this purpose we check the adjusted Tier 1 ratio of all banks in Compustat database and find that the average Tier 1 ratio is the same. Therefore, we can say that our sample is representative enough and that Tier 1 ratio has not been a substantial TARP participation criteria.

With respect to the relation of profitability with other variables, we can note that profitability Is positively related to the Tier 1 ratio; this suggests that banks with a stronger capital position as represented by the Tier 1 ratio have a higher profitability than their counterparts with lower capitalization. This makes economic sense given that we would expect the healthier, better capitalized, banks to be more profitable. This correlation also counters banks’ arguments that higher capital requirements might lower their profitability. Furthermore, we notice that both historical and expected growth are positively related with profitability and among themselves. These positive correlations show that a banks´ growth (past and future) and profitability are positively associated and stable over time. Lastly, we note that insider shares are highly correlated with TARP participation with a correlation coefficient as high as 0.47. From the high positive correlation coefficient of 0.51 we cannot conclude more than the general rational view that a higher size of the bank is usually associated with a larger number of shares and therefore the insiders would proportionally have more shares of a big bank as compared to a smaller one.

(24)

23

V. Results

Analyzing the dividend payouts of all banks in our sample, we indeed discover that banks paid higher dividends in Q4 2008 compared to Q3 2008. Given the bubble that characterized 2007, we deem the dividend payouts not to be seasonal and therefore exclude the payouts during 2007 from our analysis. In Graph I we scale the dividend paid on common stock by assets, so that the higher dividends of big banks will not have bigger weight on our results. From Graph I we observe that dividend payouts of both TARP and non-TARP banks increased from Q3 2008 to Q4 2008. Furthermore, we produce Table III that shows the number of banks that increased or decreased these dividend payouts. Even though our sample is comprised of slightly less TARP banks than non-TARP banks, we notice that there are more TARP banks that increase dividends (82 banks) compared to 81 non-TARP banks that increased dividends in Q4 2008. Furthermore, 50 non-non-TARP banks lowered dividends in Q4 2008 as compared to only 42 TARP banks that lowered these dividends. The remaining number of banks in our sample had a constant dividend payout throughout the two last quarters of 2008.

In order to get more insight into the magnitude of the increase of dividend payouts in TARP as opposed to non-TARP banks, we produce Graph II, which shows the rate of growth of dividend payouts. Though TARP banks are seen to perform very similarly to

non-Div. payout Dummy TARP Dummy Interaction Insider shares Ln assets Profitability Expected gr. Hist. gr. Tier 1 Div. payout Dummy 1

TARP Dummy 0.0952 1 Interaction -0.0334 0.0507 1 Insider shares 0.0488 0.4729 0.4994 1 Ln assets Q 0.2039 0.3064 0.5086 0.548 1 Profitability 0.1617 -0.0899 -0.1187 -0.1843 -0.1468 1 Expected growth 0.3293 -0.247 0.0923 -0.0619 0.1647 0.2795 1 Historical growth 0.0087 0.0087 0.1185 0.012 0.0442 0.1237 0.1096 1 Tier 1 0.0487 -0.0478 -0.095 -0.0031 -0.0428 0.1269 0.0874 0.0053 1

(25)

24

TARP banks, we can notice a slightly higher growth rate in the dividend payouts of TARP banks as compared to non-TARP banks. This finding leaves room for an empirical test of whether TARP participation was positively associated with dividend payouts.

Graph I. Dividend payouts

0.000 0.020 0.040 0.060 0.080 0.100 0.120 0.140 0.160 0.180 0.200 Q2 2008 Q3 2008 Q4 2008

Dividend payouts of non-TARP banks

Dividend payouts of TARP banks

Table III. Number of banks that increased / decreased dividends in Q4 2008 compared to Q3 2008

Increase in Q4 dividends Decrease in Q4 dividends

non-TARP banks 81 50

TARP banks 82 42

Graph II. Rate of growth of dividend payouts

-0.100 -0.050 0.000 0.050 0.100 0.150 0.200 0.250 Q2->Q3 2008 Q3-> Q4 2008

Growth rate of Div payouts of non-TARP banks

Growth rate of Div payouts of TARP banks

(26)

25

A. Empirically testing the significance of TARP participation

Our second step is to empirically test whether TARP participation was positively associated with the decision to pay out dividends and the decision to increase dividend payouts. Table III presents the results of two empirical models, one Logit (regression 1 and 3) and one Tobit (regression 2 and 4), with the dependent variable first defined as “quarterly cash payments on common stock divided by quarterly total assets in Q4 2008” (in regression 1 and 2) and then as “the increase in quarterly cash payments on common stock divided by quarterly total assets from Q3 2008 to Q4 2008” (in regression 3 and 4).

Table III shows the coefficients that help us determine the magnitude of the effect of each independent variable on the dependent variable, the p-values that show us the significance of these individual coefficients, the R2 that shows us the significance of the Tobit model, and the “Prob > chi2” that shows us the significance of the Logit model. It is important to mention that while the significance of the Tobit regression can be deducted from its R2, the significance of the Logit regression is deducted from its “Prob > chi2” statistic, which shows the probability of obtaining the chi-square statistic given that the null hypothesis is true, i.e. if there is in fact no effect of the all explanatory variables together on the dependent variable. Therefore, this is the p-value that we should look at to determine the significance of the Logit model. We observe that all model specifications of Table III are highly significant, even at the 1% level. This means that in aggregate, our explanatory variables contribute to the explanation of dividend payouts. Therefore, we can interpret the meaning of the coefficients associated with the significant explanatory variables and ignore the meaning of the non-significant coefficients associated with other explanatory variables.

Given that the coefficients in a usual Logit regression are not interpretable, or at least not easily interpretable, we first run the logistic model and then compute the marginal effects of the explanatory variables in STATA. Therefore, the coefficients that we have presented in Table III for the Logit regression (columns 1 and 3) are the marginal effects of each explanatory variable on the dependent variable. It is worth noting that the p-values of the coefficients do not change in the typical Logit regression as opposed to the marginal-effects-Logit regression, whereas the coefficient indeed does change. For the Tobit model

(27)

26

we present the usual coefficients in column 3 and 4. Table III below presents the results of our econometric models.

From our analysis in regression (1) and (2) we observe that TARP participation is significantly (at 1% and 5% significance level respectively) and positively associated with the decision to pay out dividends, both in the Logit and Tobit regressions. This means that, after controlling for other variables, TARP banks were more likely to pay out dividends than non-TARP banks. The Logit regression tells us that non-TARP participation increases the probability of

Explanatory Variables (1) (2) (3) (4) Dummy TARP 0.125*** 0.0002** 0.089* 0.0002

(0.002) (0.044) (0.081) (0.309) Expected growth 0.364*** 0.0009*** 0.345*** 0.0004***

(0.000) (0.000) (0.000) (0.000)

Ln assets 2.86E-06 -5.75E-12 5.71E-07 -1.70E-10

(0.130) (0.987) (0.548) (0.672) Profitability 2.634 0.009 5.141 -0.002 (0.230) (0.160) (0.151) (0.821) Historical growth -0.027 -0.0001 .361* -0.00002

(0.635) (0.418) (0.053) (0.922) Risk adjusted Tier1 ratio 0.002 0.00003** 0.007 0.00002

(0.648) (0.014) (0.333) (0.156) Intercept -0.0004** -0.0005** (0.033) (0.024) Number of observations 340 340 340 340 Log likelihood -132.823 1854.38 -193.28 1821.273 LR chi2 71.92 104.64 75.87 16.97

Prob > chi2 (Significance) 0.000 0.000 0.000 0.009

Pseudo R2 0.213 -0.029 0.164 -0.005

Table III. Logit and Probit regressions

The Logit regressions ((1) & (3)) and the Tobit regressions ((2) & (4)) have their dependent variables first defined as “quarterly cash payments on common stock over quarterly total assets in Q4 2008” in regression (1) and (2) and then as “the increase in quarterly cash payments on common stock over quarterly total assets from Q3 2008 to Q4 2008” in regression (3) and (4). P‐values are reported in parentheses. *, **, and *** indicate significance at 10%, 5%, and 1%, respectively.

(28)

27

dividend payouts by 12.5%, whereas the Tobit regression tells us that TARP participation only has a positive 0.02% effect on the decision to pay dividends. However, TARP participation is less significantly associated with an increase in dividend payout. Whereas in the Logit regression in column (3) this significance goes down to 10% level of significance, in the Tobit regression in column (4) TARP participation is no longer significant. From this analysis, we can conclude that TARP participation is indeed positively significantly associated with the decision to pay out dividends, but not so with the decision to increase dividend payouts.

An important observation is that expected growth is the most significant explanatory variable in all regressions. While Logit regressions in column 1 and 3 show that an increase by 1 unit in expected growth rate has a positive marginal effect of around 35% on both the decision to pay out dividends and the decision to increase the dividend payout, Tobit regressions show that an increase by 1 unit in growth rate positively affects the decision to pay out dividend and the decision to increase dividend payouts by only 0.0009 or 0.0004 units respectively. The coefficients of the Logit model allow us to conclude that the economic significance of the expected growth on dividend payouts is substantial. Nevertheless, this positive significance of expected growth provides support for the theory that dividends are used as a signal of good prospects or expected growth.

The rest of the explanatory variables, such as size (represented by the natural logarithm of assets) and profitability do not significantly affect the decision on dividend payouts. Therefore, we can reject these two Fama-French factors as determinants of dividend payouts. Historical growth comes out as significant only in one instance, at 5% level, with a substantial marginal effect of 36.1% on the decision to increase dividend payouts. We also cannot strongly confirm the regulatory pressure hypothesis, given that the Tier1 ratio is significant only in 1 of the four regressions. Furthermore, the magnitude of the coefficient is small, which does not give it high economic significance.

(29)

28

B. Empirically testing the significance of insider holdings

Given that in almost all the above regressions TARP participation is significantly and positively related to dividend payouts, we have established enough ground to check whether there are any further reasons that might have given rise to the decisions to pay out dividends and even increase dividend payouts. As mentioned in the introduction, there was substantial negative publicity on banks’ decisions to pay out dividends during the financial crisis. This negative publicity culminated when the media highlighted that the directors of some banks were among the leading beneficiaries of dividend payouts. Therefore, in this section we are going to empirically test how widespread was this phenomenon and if indeed insider ownership is positively related with the decision to pay out dividends and with the decision to increase dividend payouts. For this purpose, we include insider ownership (number of shares held by directors and officers) as an explanatory variable. Furthermore, we include an interaction term of TARP participation with insider ownership to check whether insider ownership differently affects dividend payouts depending on whether the bank has participated in TARP or not. The results of our analysis are presented in Table IV.

(30)

29

What is directly obvious from Table IV is that the addition of the variables related to insider ownership do not significantly change the results obtained in Table III. This means that the previously-found results are robust. Specifically, once more we find TARP participation to be a significant determinant of the decision to pay out dividends, but not so on the decision to increase dividend payout. Similarly, the results for expected growth are very similar to those of Table III; expected growth always remains a significant determinant

Explanatory Variables (1) (2) (3) (4) Dummy TARP 0.127*** 0.186*** 0.074 0.2 (0.004) (0.003) (0.208) (0.104) Expected growth 0.385*** 0.325*** 0.338*** 0.651*** (0.000) (0.000) (0.000) (0.000) Ln assets 7.20e-06** 0.000000176 0.000000367 0.000000282 (0.027) (0.356) (0.621) (0.422) Profitability 2.123 4.496 5.745 15.712** (0.343) (0.170) (0.121) (0.035) Historical growth -0.025 -0.03 .377** 0.339** (0.653) (0.733) (0.043) (0.044)

Risk adjusted Tier1 ratio 0.0001 0.006 0.007 0.016

(0.988) (0.437) (0.329) (0.283)

Shares held -1.24e-08*** -5.82e-09 2.34e-09 4.47e-09

(0.01) (0.319) (0.715) (0.687)

Interaction -1.94E-09 1.46e-09 4.86e-09 3.74e-09

(0.811) (0.892) (0.667) (0.861) Intercept 0.321*** -0.804*** (0.003) (0.000) Number of observations 340 340 340 340 Log likelihood -129.321 -248.291 -192.937 -311.019 LR chi2 78.93 58.03 76.56 70.89

Prob > chi2 (Significance) 0.000 0.000 0.000 0.000

Pseudo R2 0.234 0.105 0.166 0.102

reported in parentheses. *, **, and *** indicate significance at 10%, 5%, and 1%, respectively. The Logit regressions ((1) & (3)) and the Tobit regressions ((2) & (4)) have their dependent variables first defined as “quarterly cash payments on common stock divided by quarterly total assets in Q4 2008” in regression (1) and (2) and then as “the increase in quarterly cash payments on common stock divided by quarterly total assets from Q3 2008 to Q4 2008” in regression (3) and (4). P‐values are

(31)

30

of dividend payouts, and even it becomes more economically significant (an effect of 0.338 up to 0.651 points) in regression 3 and 4 of Table IV.

Size comes up as significant at 5% only in the first specification (with a negligible coefficient), while it remains insignificant in the other specifications. Historical growth, in contrast to the results in Table III, comes out as significant at 5% in the specifications whose dependent variable is an increase in dividend payouts. The Logit regression shows a marginal effect of 37.7% increase in the dependent variable for a 1 unit increase in historical growth, which constitutes an economically significant effect. Profitability, similarly to the results in Table III, does not prove to be an important determinant of dividend payouts; it only comes significant once in the Tobit regression with the increase in dividend payouts as the dependent variable. Tier 1 ratio is not significant in any instance, allowing us to reject the regulatory pressure hypothesis.

With regards to the agency cost hypothesis in light of conflicts of interest between insiders and the rest of interested parties, we do not find a conclusive result. While the interaction is always insignificant, which means that insider ownership similarly affects TARP banks and non-TARP banks, the insider ownership variable is significant in one specification. From the Logit regression with the dummy of dividend payouts as a dependent variable, we see that insider ownership has a negative marginal effect (as opposed to a positive one that would support the agency cost hypothesis) that is significant even at 1% level. Nevertheless, the marginal effect is economically small even after considering the magnitude (mean) of the insider shares variable. In the other three specifications insider ownership is insignificant.

In conclusion to our analysis we can say that in general our results support a conflict-of-interest theory that results is risk shifting from shareholders to taxpayers and the rest of society. Specifically, the positive effect that TARP participation has on dividend payouts means that taxpayer money that was injected in banks in the hopes of strengthening their capital position was misused by the board of the banks, which paid out (part of) the money to satisfy shareholders. Nevertheless, the insignificance of the interaction term and the insignificance and even negative marginal effect of insider ownership on dividend payout lead us to reject the agency cost theory that was based on the conflict of interest between

(32)

31

insiders and outsiders. Regarding the rest of the variables, we can conclude that expected growth is the most significant determinant of dividend payouts in all instances; as expected, it is positive related, leading to support for the signaling hypothesis. Lastly, we believe that it is worrisome that we do not find support for the regulatory pressure hypothesis; there is no relation between a strong capital position and a higher dividend payout, as we would suppose it should be in healthy firms.

VI. Robustness Analysis

As part of our robustness analysis, we try to conduct the same analysis varying the definition of some of the explanatory variables and adding another one. This will help us detect if any significant relationship was only due to our specification of an explanatory variable or if we missed an explanatory variable. As a base model to compare to our robustness analysis we choose regression 1 of table IV. A similar analysis has been performed for all other regressions of Table IV, but the results remain the same and therefore we decide not to present all results here. However, the results can be made available upon request. The changes that we incorporate are described below and illustrated in Table IV.

First and most importantly, we want to make sure that we control for the banks’ policy of paying out dividends before the last quarter of 2008. Given that banks that paid dividends in Q3 2008 are more likely to continue doing so than banks that did not pay dividend in the Q3 2008, we decide to include a dummy variable that takes the value of 1 if there was any dividend paid in Q3 2008 and the value of 0 otherwise. The inclusion of this dummy variable does not produce the slightest effect on our model, thereby adding to our belief of the robustness of our model.

Then, doubtful of the validity of only the Tier 1 capital ratio as a representative of the regulatory pressure hypothesis, we substitute this explanatory variable with the aggregate Tier 1 and Tier 2 capital ratio. This is done in the belief that the pressure from the regulators considers the whole capital position of the bank, which includes not only Tier 1, but also Tier 2 capital. We find that our previous results were not driven by the narrow specification of

(33)

32

this variable. The inclusion of Tier 1 and 2 as a capital measure does not change the results of regression 1 in Table IV.

As a third robustness test we considered scaling the quarterly dividend payments by net income. However, given the crisis period and the high volatility in net income associated with it, we believed that this specification would produce distorted results.

Overall, we can say that our model is robust and that the results of our model were not driven by certain specifications of variables or by an omitted variable bias.

(34)

33

VII. Conclusion

We have used several Logit and Tobit models to analyze the determinants of dividend payouts, with a focus on bank behavior and its effect on dividend payouts, in a

Explanatory Variables (1) (2)

Dummy TARP 0.127*** 0.128***

(0.004) (0.004)

Expected growth 0.385*** 0.383***

(0.000) (0.000)

Ln assets 7.20e-06** 7.14e-06**

(0.027) (0.029)

Profitability 2.123 2.066

(0.343) (0.357)

Historical growth -0.025 -0.025

(0.653) (0.654)

Risk adjusted Tier1 ratio 0.0001 0.001

(0.989) (0.797)

Shares held -1.24e-08*** -1.23e-08***

(0.01) (0.01)

Interaction -1.94E-09 -2.05e-09

(0.811) (0.801) Dummy Q3 2008 dividend 0.001 (0.984) Number of observations 340 340 Log likelihood -129.321 -129.288 LR chi2 78.93 78.99

Prob > chi2 (Significance) 0.000 0.000

Pseudo R2 0.234 0.234

Table V. Robustness tests on Logit regression

Regression (1) adds the Dummy for Q3 dividends to regression (1) in Table IV. Regression (2) uses the aggregate Tier 1 & 2 capital

measure instead of the previously-used Tier 1 ratio in regression (1) in Table IV. P-values are reported in parentheses. *, **, and ***

(35)

34

cross-section of 340 US banks during the last quarter of 2008. This period was unique as it included an unprecedented inflow of government funds into banks. Our research has enhanced the understanding of bank behavior and incentives in times of crisis and especially in the presence of external capital injections.

Our main results are the following. First, TARP participation increases by around 13% the probability of the decision to pay out dividends. Therefore, our results document a risk shifting behavior that consists in shifting risk from the shareholders of the bank to the rest of society, such as taxpayers, government, and debtholders of the banks. The related policy implication requires either the cancellation of all such capital injections or at least an accompaniment of these injections with strong clear rules that specifically require a stop of dividend payout and the like. From our results we can conclude that TARP did not fulfill its purpose, or at least not in the best possible way, as a substantial amount of the capital that was injected to strengthen banks’ capital position just flew out as dividends as opposed to being kept as a loss absorbing buffer. The TARP prohibitions of dividend payouts were only suggestions during the last quarter of 2008 and strongly enforced only after February 2009. These requirements were not enough to control bank incentives. However, we do not find conclusive evidence of a positive relation between TARP participation and the decision to increase dividend payouts.

Furthermore, we do not find conclusive evidence of an effect of insider ownership on dividend payout decisions. Moreover, the all-time determinant of dividend payouts is expected growth, with a marginal effect of a positive 38% on the probability of paying out dividends and increasing dividend payouts for a one-unit increase in expected growth. This result is therefore economically significant and provides support for the signaling theory, which suggests that banks pay out dividends to signal positive future prospects.

We do not find significant evidence on the Fama-French determinants of dividend payouts, such as size, profitability, and historical growth. Each of these determinants came out only once as significant in all our specifications. Lastly, we find no evidence of the regulatory pressure hypothesis. This means that banks with higher Tier 1 ratios did not pay more dividends and banks with lower Tier 1 ratios did not pay lower dividends. The

The determinants of dividend payouts during the financial crisis of 2008-2009 : the case of US Banks

Julja Prodani

ID: 10825371