Banks executive compensation and risk-taking an analysis of the U.S. banking industry between

Banks executive compensation and risk-taking an analysis of the U.S. banking industry between 2007-2015 by D.C.M. (Dennis) van der Heijden U1259449 ANR: 597290 Email: Academic year: 2016 2017 Tilburg School of Economics and Management (TiSEM) Master Thesis Finance Supervisor: F. Castiglionesi Word count: 16.789 August 11 th, 2017

Abstract In this research, the relationship between equity-based compensation and risk-taking in the banking industry from 2007 to 2015 is empirically investigated. In addition, the effect of several CEO- and bankspecific characteristics are included in this analysis. Empirical results are obtained with the help of Ordinary Least Square (OLS) regressions models. Using a sample of 158 different CEOs, divided over 109 U.S. banks, I find (1) no evidence for a relationship between equity-based compensation and risktaking for the total sample from 2007 to 2015, (2) partial evidence for a negative relationship between equity-based compensation and risk-taking for banks with relatively high debt ratios and for relatively large banks in terms of total assets, (3) partial evidence for a negative relationship between equity-based compensation and risk-taking in the year 2014, and (4) partial evidence for a negative relationship between option-based compensation and risk-taking. Findings may be appealing for shareholders and regulators for the optimization of CEO compensation packages in the banking industry. 2

Table of contents 1 Introduction 4 2 Literature review 9 2.1 The objective of compensation packages 9 2.2 Executive compensation and risk-taking 10 2.3 Executive compensation and risk-taking in the banking industry 12 2.4 Equity-based compensation and risk averse behavior 13 2.5 The measurement of risk 14 3 Hypotheses 17 4 Methodology 20 4.1 Multivariate regression analysis 20 4.2 The measurement of equity-based compensation 21 4.3 The measurement of risk-taking 21 4.4 Measuring the relationship between equity-based compensation and risk-taking 22 4.5 Option-based/stock-based compensation and risk-taking 23 4.6 The comparison of risk levels over the years 24 5 Data 25 5.1 Sample construction 25 5.2 Descriptive statistics 26 6 Results 30 6.1 Parametric analysis 30 6.2 Multivariate regression analysis: equity-based compensation and risk 31 6.3 Multivariate regression analysis: subsamples and cross-section 34 6.4 Multivariate regression analysis: option-based/stock-based compensation and risk 40 7 Conclusion and limitations 43 7.1 Conclusion 43 7.2 Limitations and recommendations 44 8 References 45 9 Appendix 48 3

1. Introduction Banks fulfill an important role in the economic system. The banking industry is responsible for the providence of numerous services, which makes this industry indispensable to the economy. For example, they create liquidity by financing illiquid assets with more liquid funds, such as deposits. Banks form the link between savers and investors in that way, also by fulfilling the role of the monitor. Furthermore, banks are known for the financing of for instance loans, mortgages, and businesses. As Cai, Cherny, and Milbroun (2010) mentioned, the banking sector in the United States is responsible for up to 35 percent of the domestic profits. In that context, it is not surprising that executives in this banking industry are highly compensated. However, due to the latest financial crisis, for which the financial sector is blamed at least partially, the general public and also bank regulators have become (even more) interested in the amounts and structures of the compensation packages of banking executives. One main question which arises in the debates about bank executives' compensation, is whether these compensation packages give banking executives incentives for risk-taking. In particular, the equity-based compensation part, which includes stock option grants and restricted stock grants, is often linked to this risk-taking. This link arises from the classical principal-agent theory, in which the shareholders (the owners of the firm) represent the role of the principal, while the Chief Executive Officer (CEO) (who controls the firm) represents the role of the agent. The shareholders main objective is the maximization of the share price, whereas the CEO also cares about other objectives such as personal wealth and prestige. To align the incentives of the CEO with those of the shareholders, CEOs are often rewarded partially with equity-based components. It is argued that this mismatch in incentives could be solved when the CEO s personal wealth is tied to the firm. Although the alignment of incentives may be beneficial for shareholders, it may harm other stakeholders of the firm. In fact, equity-based compensation is linked numerous times to risk-taking in the existing literature. John, Saunders and Senbet (2000) already showed that managerial compensation has its influence on firms investment choices. However, in the existing academic literature, there is no overall consensus regarding the strength and the direction of the relationship between equity-based compensation and risk-taking. Thereby, it has to be mentioned that the relationship between these two concepts is already examined numerous times in different industries. However, as Chen, Steiner & Whyte (2006) noticed rightly, there seems to be a gap in the financial literature on this topic, since this relationship is not widely investigated for the banking industry. That being said, it is actually important to investigate the relationship between equity-based compensation and risk-taking in the banking industry specifically, as the findings on this relationship in other industries are not necessarily generalizable, because of some 4

specific characteristics in the banking industry. One such characteristic is for example that most banks have fairly high debt ratios compared to other industries. Therefore, banks are even more sensitive to a possible clash between debt holders and shareholders. Also, it is often argued that banking executives are not monitored appropriately compared to executives in other industries due to deposit insurance (Deyoung, Peng & Yan, 2010). Depositors do not have strong incentives to monitor the CEO, as they know that they are secured up a certain dollar amount in times of financial distress. In addition, because banks, and especially large banks, are indispensable to the economic system, they are considered to be too-big-to-fail. This means that a collapse of these large banks will result in a domino effect: it induces a collapse of the whole economic system. Acknowledging this, it can affect the behavior of the banking executives, knowing that the banks will be saved by the government in case of financial failure (Houston & James, 1995). This thesis will focus on the following research question: To what extent, and in which direction, is CEO equity-based compensation related to risk-taking in the banking industry? It is hard to predict the direction of a possible relationship at this point, as there is no consensus about this direction in the existing literature. Thereby, most studies were executed before the latest financial crisis, but it can be fairly assumed that this crisis has had its influence on the relationship between equity-based compensation and risk-taking. This relationship will be examined for the period from 2007 to 2015. This means that differences in the relationship during the crisis and post-crisis will also be taken into account. In addition, before looking at the relationship between equity-based compensation and risk-taking, both differences in compensation structures as well as differences in risk levels in the banking industry over the years will be investigated. The relationship between equity-based compensation and risk-taking will be measured with the help of Ordinary Least Square (OLS) regression models. Next to these regression models, a parametric test will be executed in order to test significant differences in risk levels over the sample period of this study. However this is not the main focus of this thesis, it shows that it is important to include year dummies and it also shows the relevance of executing cross-section analysis. Regarding the regression models, equitybased compensation is build up as the sum of stock option grants and restricted stock grants, and will be measured as a proportion of total compensation. Risk however, is more complex to measure, as there exist many types of risk. Based on other financial papers, and especially the one of Chen et al. (2006), this study will make use of four market-based risk measures: (1) total risk, (2) idiosyncratic risk, (3) systematic risk, and (4) interest rate risk. By using these measures, I am able to investigate the effect of equity-based compensation on different risk aspects banks are facing. There are two main differences between this study and the study of Chen et al. (2006). Firstly, a different time frame is used. Chen et al. 5

(2006) investigates the period from 1992 to 2000, whereas this thesis focuses on the period from 2007 to 2015. And secondly, Chen et al. (2006) focused on the relationship between stock option grants and risktaking, whereas in this thesis, the relationship between total equity-based compensation and risk-taking is also considered, including restricted stock grants. In the regression models, the risk measures will serve as the dependent variable, whereas equity-based compensation is the independent variable. Besides the regression equations that investigate relationship between equity-based compensation and risk-taking, a second and third model will be build to examine the relationship between stock option grants and risktaking and between restricted stock grants and risk-taking. This is particularly interesting, because it is expected that stock option grants have a stronger relationship with risk-taking than the restricted stock grants have. That is because the differences in characteristics of these equity-based components can result in different relationships with risk-taking. For example, the magnitude for potential gains is larger for stock option grants compared to restricted stock grants. The value of restricted stock grants is more steady over time, whereas small changes in the underlying stock price can result in significant changes in the value of the stock option grants. In addition, restricted stock grants can result in a wealth loss when the share price decreases, while stock option grants do not bear the potential downside risk, as the CEOs will simply not exercise their options when they are out of the money. So, the value of stock option grants cannot become negative, but they have an unlimited potential upside when the underlying stock price increases (Mehran,1995). On the other hand, restricted stock grants do always have at least have some value, but this value can also become negative as the underlying stock prices decreases. Next to these main variables of interest, a number of other variables is included in the regression models. These control variables are included, as they may affect the relationship between equity-based compensation and risk-taking. These variables control for differences in (1) bank characteristics, (2) CEO characteristics, and (3) year-to-year differences. Data is gathered via Wharton Research Data Services (WRDS). The database Execucomp provides me with all CEO specific information, such as compensation, age, and tenure. After obtaining this information, Compustat is used to obtain all bank specific data, like total debt outstanding and total assets. These databases contain mainly U.S. data, so the constructed sample consists of U.S. banks and - CEOs only. Data is gathered for the period from 2007 to 2015. I do not find evidence for a significant relationship between equity-based compensation and risk-taking for the period from 2007 to 2015, based on the complete regression model. This also means that I am not able to draw conclusions about the direction of this relationship. Two possible explanations are discussed that could explain the insignificant results. Firstly, due to poorer hedging opportunities, possible incentives that arise from equity-based compensation regarding risk-taking, could be offset. This is a 6

result of the decrease in cash compensation in combination with an increase in equity-based compensation during the investigated time period. A decrease in cash compensation, which is untied to the bank, along with an increase in equity-based compensation, which is tied to the bank, results in a poorer diversified portfolio for the CEO. This is in line with the hedging theory of Smith and Stulz (1985). Secondly, an explanation for the insignificant results is found with the help of the changed compensation package structures. During the period from 2007 to 2015, I find an impressive increase in restricted stock grants in combination with a decrease in stock option plans. This could explain the insignificancy, as it is argued by many researchers that the relationship between stock option grants and risk-taking is significantly stronger than the relationship of restricted stock grants on this risk-taking, due to differences in characteristics mentioned above. Therefore, I executed two other regression models which examines the relationship between option-based compensation and risk-taking and between stock-based compensation and risk-taking. With the help of the first regression model, I find that there exists a negative and significant relationship between option-based compensation and risk-taking, at least for three of the four market-based risk measures. These empirical findings are in line with the findings of Low (2006), Parrino,Poteshman, and Weisbach (2005), Rogers (2002), and Smith and Stulz (1985). On the other hand, in line with the expectations based on the differences in characteristics between the two equity-based components, no evidence is found for a relationship between restricted stock grants and risk-taking. Dividing the dataset into subsamples based on debt ratio, I find partial evidence for a negative relationship between equity-based compensation and risk-taking in for banks with a low debt ratio (below 0.87714). This is in line with the studies of D Hulster (2009) and Hamada (1972). Dividing the dataset into subsamples based on the natural logarithm of total assets, a measure of size, I also find partial evidence for a negative relationship between equity-based compensation and risk-taking for large banks (above 9.31128). Also, with the help of cross-section analysis, I find partial evidence for a negative relationship between equity-based compensation and risk-taking in the year 2014. Again, these findings are in line with the findings of Low (2006), Parrino Poteshman, and Weisbach (2005), Rogers (2002), and Smith and Stulz (1985), who all found evidence for a negative relationship between equity-based compensation and risk-taking. In this study, equity-based compensation is used as an independent variable, whereas risk-taking is used as the dependent variable. Therefore, only one direction of causality is taken into account. However, it can be fairly argued that equity-based compensation not only influences risk-taking, but also vice versa. This means that the causation of the relationship between equity-based compensation and risk-taking runs in both directions. The obtained coefficients of the OLS regressions may therefore be biased due to 7

endogeneity, which is a result of reverse causality. Therefore, reverse causality has to be considered as a limitation of this study. The remainder of this thesis is organized as follows. In chapter 2, an extensive body of existing literature will be reviewed. From this literature review, hypotheses will be formulated in chapter 3. Chapter 4 will deal with the research methods that are used in this study. Thereafter, in chapter 5, it will be described how data is gathered and the dataset is constructed. In chapter 6, the empirically obtained results of this thesis will be discussed in combination with the reviewed literature. Finally, in chapter 7, conclusions will be drawn and limitations of this research will be discussed. 8

2. Literature review The principles of executive compensation have been widely researched in the academic literature. Executive compensation has been recognized not only for its basic rewarding principles, but also as an important influential mechanism which has its effect on multiple major business dimensions. Executive compensation is often linked to the classical principal-agent theory (Guay, 1999). In a perfect world, in which executives could be monitored optimally, shareholders should be able to enforce all managerial activities flawlessly. That means, executives will make the right decision for the business in every possible situation. Since we simply do not live in a perfect world, the compensation packages of executives should be designed in such a way that it aligns the interests of the executive with those of the shareholders of the firm. If these compensation packages are structured optimally, executives should select and take on investments which maximize the wealth of the shareholder (John et al., (2000); Chen et al., (2006); Bolton, Mehran & Shapiro, (2015). In order to align the interests, executives have to become a shareholder of their own business. Therefore, equity-based instruments such as stock option grants and restricted stock grants are included into the executives compensation packages. However, numerous researchers suggest that this equity compensation has its effect on the risk-taking behavior of the executive, and therefore in turn, its effect on the risk of the business as a whole. 2.1 The objective of compensation packages The total compensation package of most executives, not only in the banking sector, consists of different layers. These packages often include salary, bonuses, stock option grants, restricted stock grants, and other compensation components. Stock option grants and restricted stock grants are part of the equitybased compensation. The rationale of this equity-based pay is that it should align the interests of policy makers of the bank (the executives) with those of the residual owners of the bank (the shareholders) (Bolton et al., 2015). Without alignment of interests, managers may try to maximize their personal interests, such as influence and power in the company. In other words, they may gamble with shareholders invested money on behalf of their own interests instead of maximizing shareholder value. This mismatch in interests is often referred to as the principal-agent problem, in which shareholders represent the principal, and executives represent the agent. To solve this problem, non-cash payments that represent ownership in the firm is included in the compensation packages of executives. The reason for this compensation structure is that executives should be more willing to maximize shareholder value as the executives wealth itself also increases because of that (Jensen & Murphy, 1990). 9

2.2 Equity-based compensation and risk-taking The alignment of interests between executives and shareholders is not necessarily favorable for all stakeholders of a firm. The emphasis on equity-based compensation may lead to higher levels of risktaking among executives, as incentives are created to maximize stock prices. This, in turn, may harm the firm s stability. Other stakeholders, such as debtholders and depositors, may encounter the negative byproduct of the equity-based compensation. This problem arises because of different payoffs between debtholders and shareholders in different situations. These differences have been described extensively in the existing literature about executive compensation and risk-taking, not only in the banking industry. Agrawal and Mandelker (1987) already found that owning equity in a firm induces executives to select investments with more variance. Also, Defusco, Johnson, and Zorn (1990) reported that stock return volatility increases as a result of stock option plans for executives. These findings cannot necessarily be generalized to the banking industry though, for various reasons. These will be discussed in the next paragraph. However, when a firm is not able to repay the debtholders, it implicitly means that shareholders will not receive any money since they are the residual claimant. In other words, shareholders only receive any money that is leftover after all debtholders are paid back. On the other hand, after repaying the debts, shareholders receive every dollar that is left over from the realized value. This means that shareholders cannot lose more than their invested amount of money, but they can gain an unlimited amount of money when the bank is doing well. In fact, debtholders are bearing the potential losses from risky investments (investments with high payoffs and high likelihood of failure), whereas shareholders profit from the potential gains from these investments (Cai et al., 2010). Thus, the more risk the bank takes, the greater the potential for the shareholders. This may trigger executives to take risks to maximize the value of shareholders, and therewith also their own wealth through equity-based compensation. To be able to understand why executives are incentivized to take more risk due to the equity-based compensation, it is necessary to explain the characteristics of the two different forms of equity-based compensation: stock option grants and restricted stock grants. Other compensation types, such as bonuses, are often used for short-term successes, while equity-based compensation items represent long-term incentives (Brander & Poitevin, 1992). It is often assumed that executive stock ownership through stock grants has the same influence on risk incentives as stock option grants have. However, due to asymmetric payoff structures and risk properties, this is not necessarily true. Stock option grants give executives the right to purchase stocks of the bank they are working for, at a preset price. Typically, this price is determined as the price in the market at the moment the stock option grant is offered. Executives are usually allowed to purchase, or exercise as it is often referred to, these stock options after a predetermined period. For example, at the first of January 2017, stock options are 10

granted at the current market price of $50. The executive is allowed to exercise the stock option in two years. This implies that, when the stock price rises to for example $100, the executive is able to purchase the stocks at a price of $50, and sell those stocks directly in the market for $100. Thus in that way, executives are incentivized to boost the stock price (Chauvin & Shenoy, 2001). On the one hand, this aligns the interests of the executive with the interests of the shareholders. On the other hand, this may create incentives for the executive to take risks to maximize his/her own wealth. This in turn, as mentioned before, may seriously harm other stakeholders of the firm and even the firm s own stability. Restricted stock grants are a second type of equity-based compensation. Executives of a bank are rewarded with stocks of the bank which they are working for. These stocks are restricted for the executive. The executive becomes the legal owner of the stocks, but he/she is not allowed to sell those until they are vested. This could be for example after a certain time period, or when a business goal is accomplished. Although both stock option grants and restricted stock grants serve the same purpose, there are some differences between the two types of equity-based compensation (Irving, Landsman & Lindsey, 2011). The magnitude for potential gains is larger for stock option grants compared to the restricted stock grants. This is because the options can be exercised at a fixed price, while the underlying stock price can have risen significantly during the time period from the grant offer until the exercise date. At the other hand, the stock option grant is worth nothing when the underlying stock price does not increase, or even decreases, during this period. Small adjustments in price can significantly change the value of these stock option grants. The value of restricted stock grants, however, is steadier over time. Thereby, the stock grant still has at least some value when the stock price decreases, unless the bank goes bankrupt (Dodonova & Khoroshilov, 2006). Another important difference between the two equity instruments is that only stock ownership, obtained from the restricted stock grants, can result in a wealth loss for the executive when the share price decreases. Stock option grants at the other hand, do not bear any potential risk, because executives will just not exercise their options when these are out of the money. So, executives holding stock option grants have an unlimited potential upside, but they do not bear the potential downside risk (Mehran, 1995). In other words, for wealth purposes as a stock option holder, it does not matter if the share price is far below or just slightly below the exercise price. While on the other hand, these same stock option holders, gain unlimited upside potential. This concept is relatable to the clash arising from different interests between shareholders and debtholder. Figure 1 below visualizes the payoff differences between the stock option grants (on the left) and the restricted stock grants (on the right). As shown, the value of the stock option grant is zero (but never below zero, or negative) as long as the underlying share price is lower than the exercise price, whereas the line steeply increases as the option becomes in-the-money. From the line regarding the restricted stock 11

option grants it can concluded that it moves steadier (less steep) compared to the stock option grant, and the value can even become negative when the underlying share price decreases. It moves along with the stock price changes. Stock option grants Restricted stock grants 0 0 Figure 1: Payoff differences between stock option grants and restricted stock grants 2.3 Executive compensation and risk-taking in the banking industry Although the abovementioned problem exists in most industries, this thesis will focus especially on the banking industry. Houston and James (1995) found that there exist significant differences between industrial firms and banks in the way compensation packages are designed. These differences exist both in terms of the absolute amount of total compensation, as well as the relative importance of each component in these compensation packages. The abovementioned problem can therefore be different, and possibly even worse, in the banking industry due to three main reasons. Firstly, because banks have in general high debt ratios. They hold large proportions of leverage compared to equity to finance their operations. This makes banks more sensitive to the clash between shareholders and debtholders. Therefore, the compensation structures leading to risk incentives could be even worse in the banking industry (Bebchuk & Spamann, 2009). Secondly, most depositors are secured up to a certain dollar amount in times of financial distress at banks. This phenomenon is called deposit insurance. Due to this insurance, depositors do not have strong incentives to monitor the executives. In addition, executives are not afraid of bank runs because of the insurance, which allows them to make riskier decisions (DeYoung, Peng & Yan, 2010). Thirdly, especially large banks, are considered to be too-big-to-fail. A collapse of these large banks will result in collapse of the economic system as a whole, since almost all businesses are dependent on these banks. This makes these banks indispensable to the economy. Acknowledging this fact, executives may be less scared of possible negative outcomes of their risky behavior, knowing that the bank will be saved by the government in case of failure (Houston & James, 1995). 12

Similar to what studies found for non-banking industries, as mentioned earlier in paragraph 2.3, a positive relationship between equity-based compensation and risk-taking has also been found for the banking industry. For example, Chen et al. (2006) show that as a result of deregulation in the banking industry, stock option-based compensation has increased significantly during the period from 1992 to 2000. They suggest that the compensation structures, which originated from this trend, induce risk-taking. Their results therefore support an increased risk hypothesis over a risk aversion hypothesis. Mehran and Rosenberg (2008) support these findings. They also found that equity-based compensation leads executives to undertake riskier projects. In addition, Coles, Daniel, and Naveen (2006) provide empirical evidence of a positive relationship between compensation structures and investment policy, debt policy, and firm risk. They find that, controlling for pay-performance sensitivity (delta), CEOs whose wealth is more sensitive to stock volatility (vega), implement riskier policy choices. This results in (1) more research and development investments, (2) less investment in plant, property, and equipment, (3) more focus on particular components of the business, and (4) higher levels of leverage. In contrast to that study, different risk measures are used in this study. These risk measures do not only include risk based on stock return volatility, as vega and delta do, but will also consider other risks banks are facing, such as interest rate risk. Note that this thesis will particularly focus on the compensation packages and its possible relationship with risk, of the Chief Executive Officer (CEO). Firstly, because the data available in the Execucomp database mainly contains CEO information. Secondly, because it can be reasonably assumed that if there exists any relationship between equity-based compensation and risk-taking, it should be at the CEO level due to the fact that a CEO has the most influence in the decision making process within the company. 2.4 Equity-based compensation and risk averse behavior While numerous studies suggest that equity-based compensation packages result in an increase in risk taken by executives in the decision making process, there are also studies that suggest that equity-based compensation stimulates risk-adversity. Smith and Stulz (1985) found that stock grants and stock option grants are useful tools for reducing the conflicting interests between shareholders and executives by linking executive wealth to the firm s share price, as a first principle. However, they also suggest that over time, executives may become risk-averse because of high levels of equity-based compensation. This concern is created due to the fact that, after several years, these executives hold relatively high proportions of shares in their own business. This, in turn, results in badly diversified executives who may then become risk-averse and therefore neglect positive net present value (NPV) projects which carry high risk levels. Parrino et al. (2005) and Low (2006) support these findings. Low (2006) found a five percent decrease in firm risk, which is concentrated among businesses with low managerial equity compensation 13

incentives. However, it is worth mentioning that her sample consists of a wide variety of firms in various industries, and more importantly, financial firms were excluded. John and John (1993) disagree on the conclusions drawn by Smith and Stulz (1985). They state that stockholding executives can benefit together with the other equity financiers if they are able to shift the risks to debtholders. Thereby, the easier this risk shifting for equity holders to debtholders, the greater the incentive for high equity-based compensated executives to take more risk. Although compensation packages could result in more risk taking, the compensation policies are not necessarily designed to encourage executives taking risks. Lewellen (2006) found in contrary to other studies that stock option compensation can in some cases result in more risk-averse behavior. When the stock options are in-the-money for example, executives may take less risks to ensure themselves to end up with exercisable options. Whereas additional risk may be taken when the stock options are rewarded, as they are often out-of-the-money in first instance. Also, Houston and James (1995) do not find evidence for the hypothesis that banking executives are incentivized to invest in riskier projects. According to them, the relationship between equity-based compensation and risk-taking in the banking industry varies heavily with firm specific characteristics, but also with the conditions of the compensations packages. However, their research focusses on comparing the compensation packages of the banking industry to the compensation packages of other industries, rather than investigating the effect of the compensation structures on risk across the banking industry. Finally, Rogers (2002) also found negative coefficients in his models, implicating a negative relationship between CEO risk-taking and stock and option holdings. He found that this effect only weakly appears in one-stage models. However, if modeled in a system including simultaneous equations, the negative relationship seems to exist strongly. Rogers (2002) concludes that this risk-averse behavior arises because of hedging in the CEO s personal portfolio. 2.5 The measurement of risk There is an extensive body of research conducted in the field of compensation, its structures, and equitybased compensation in particular. Besides its relationship with risk, researchers also focused on the relationship between compensation and firm performance. However, both Mayers and Smith (1992) and Smith and Watts (1992) found that these concepts are not as significantly related in regulated industries as they are in industries with less regulation. Therefore, incentives which possibly arise from the compensation packages, may not be that strong and may be different compared to incentives created in other industries. However, rather than the relationship with performance, abovementioned studies and arguments show that the relationship between compensation and risk is an interesting and important relationship to investigate for the banking industry. 14

Looking at the Oxford English Dictionary (2016), risk is defined as the following: A situation involving exposure to danger. It has to be said that risk on itself does not have that much of a meaning as it is an abstract concept which is, depending on its context, interpretable in different ways. In other words, there is not one universal explanation for the word risk, which fits all types and forms of risk as used in the academic literature. Reviewing the extensive body of literature about equity compensation and risk, I find that there are two main ways in which risk is measured in prior studies. The first method which is used in several studies regarding compensation and risk, is using the proxies vega and delta. These proxies are for example used by Coles et al. (2006), DeYoung et al. (2013), Belkhir and Chazi (2010), Knopf, Nam, and Thornton (2002), and Hagendorff and Vallascas (2011). Vega represents pay-risk sensitivity. It explains the changes in wealth of a CEO in relation with changes in the volatility of returns of the underlying stock. This volatility is measured in standard deviations. In terms of options, it measures to what extent the option is sensitive to changes in the volatility of the underlying stock. Delta on the other hand, represents pay-performance sensitivity. It explains the elasticity of the wealth of a CEO in relation with percentage changes in the stock price. Again, in terms of options, it compares changes in the price of an option in respect to price changes in the underlying stock. These proxies can easily be used and are helpful in understanding the relationship between CEO compensation and risk. However, these proxies are only tied to risk that has something to do with stock returns. Thereby, it can be fairly assumed that there are numerous other factors which affect the stock return volatility. Several other researchers use a different method for measuring risk. This second method is also used widely in the existing literature. For example, Chen et al. (2006) and DeYoung et al. (2013) used this method. It examines the relationship between compensation structures and several market-based risk measures, using a regression. Usually, the following four risk measures are included in this method: (1) total risk, which is measured through the standard deviation of the stock returns; (2) idiosyncratic risk, which is measured by the standard deviation of the residuals of the Ordinary Least Squares (OLS) regression model; (3) systematic risk, which is measured by the coefficient of the regression between the (daily) stock return and the (daily) return on the equally weighted index; (4) interest rate risk, which is measured by the coefficient of the regression between the (daily) stock return and the (daily) three-month treasury bill yield. These market-based risk measures are thereafter used to run a second regression model in which the relationship between the risk measure and the compensation structures is actually measured. It has to be mentioned that it is not necessarily the case that one of the two methods is better than the other one. These are just two different methods for measuring risk. The difference is that the first method, which looks at vega and delta, focuses specifically on the relationship between CEO wealth and the 15

volatility in stock returns, which is just one aspect of risk banks have are facing. Whereas the second method, which includes four market-based risk measures, does not focus specifically on the risk which arises from stock return volatility. It rather looks at the effect of the compensation structures on several risks, which banks have to deal with. 16

3. Hypotheses In this chapter, hypotheses will be formulated regarding the relationship between equity-based compensation and banks risk-taking. At the end of this thesis, these hypotheses will be either accepted or rejected, based on upcoming results in combination with the relevant literature. As discussed in the literature review above, prior studies suggest that shareholders are well willing to take on high levels of firm risk, because they benefit from the possible unlimited profits arising from this risktaking (e.g. Chen et al. (2006), John et al. (2000), Bebchuk & Spamann (2009)). Equity-based compensation is often included into the compensation packages of CEOs. The main reason for this is that it should align the interests of the CEO with those of the shareholders, as these shareholders are not able to perfectly monitor the executive s decision making. However, this alignment of interests makes that the CEO becomes in fact a shareholder as well. This completes the circle, meaning that the CEO also may increase bank risk to maximize his/her own wealth along with those of the shareholders, at the expense of debtholders. As mentioned in paragraph 2.4, this effect may be even stronger in the banking industry because of (1) high debt ratios, (2) deposit insurance, and (3) too-big-to-fail. From this, the following hypothesis can be derived: H1: Risk-taking is positively related to equity-based compensation in the banking industry. Thereby, it has to be recognized that stock option grants and restricted stock grants are two different types of equity-based compensation, both with their own characteristics. As mentioned in paragraph 2.3, the magnitude for potential gains is larger for the stock option grants than for restricted stock grants. Potential risk-taking incentives which arise from equity-based compensation are therefore expected to be stronger for the stock option grants compared to the restricted stock grants. Therefore, it is also interesting to execute a model which only measures the relationship between stock option compensation and risk. More details about this relationship and the corresponding models will be discussed in upcoming chapters. In contrast to this first hypothesis, it could also be the case that equity-based compensation has a negative relationship with bank risk. This thesis focuses on the period from 2007 to 2015. This implies that both crisis as well as post-crisis years are included. Due to the enormous impact of this financial crisis, executives may have become more careful in their decision-making, being afraid of the consequences of high risk-taking policies. Thereby, after the financial crisis, regulation policies have become stricter, making it harder for executives to take risk at the cost of the bank. Also, the influence of (the announcement of) Basel III plays a key role in this (Blundell-Wignall & Atkinson (2010), Cosimano & Hakura (2011). So, although there are studies that support the risk-taking hypothesis, it is worth considering a more conservative, risk-averse hypothesis, which is in line with findings of Smith and Stulz 17

(1985), Low (2006), and Rogers (2002). Because when equity-based compensation increases, it implies that the diversification in the CEO s own portfolio decreases, as the executive simply holds a larger proportion of the same stock compared to his/her other investments. Thereby, the stock option grants, which are part of the equity-based compensation, are issued usually at-the-money. As Smith and Stulz (1985) mentioned, this could lead to risk-averse behavior, as the value of these stock option grants become zero when the underlying stock price is lower than the price at issuance, at the time the options may be exercised. Therefore, in line with these theories, the second hypothesis can be derived as follows: H2: Risk-taking is negatively related to equity-based compensation in the banking industry. This thesis will also take a look at the differences in risk-taking during the crisis, and post-crisis. Although this is not the main focus of this thesis, it is worth looking at these differences, since the financial crisis logically had an enormous impact on the banking system. Thereby, investigating these differences could also help in providing answers to the main research question of this study. It is expected that risk levels where higher during the crisis than after the crisis. After the crisis, it is likely that CEOs become more conservative in their investing behavior. Thereby, stricter regulation after the latest financial crisis, such as higher mandatory capital requirements, should result in safer banks. These assumptions are strengthened by the findings of Guiso, Sapienza, and Zingales (2013). They found evidence for more risk-aversion after the financial crisis. This phenomenon is also found for other financial crises in the past. Schwert (1989) for instance, found significant higher stock return volatility during the Great Depression from 1929-1939 In other words, the banking industry should face less risks after the crisis (Blundell-Wignall & Atkinson (2010), Cosimano & Hakura (2011). The third hypothesis can therefore be formulated as follows: H3: The banking industry faces higher levels of risk during the crisis period compared to the post-crisis period. Lastly, based on the differences in characteristics between stock option grants and restricted stock grants, which are extensively explained in paragraph 2.2 along with figure 1, a hypotheses can be formulated regarding the relationship of between both equity-based components and risk-taking. In line with the findings of Irving, Landsman, and Lindsey (2011), Dodonova and Khorsoshilov (2006), and Mehran (1995), it is expected that a possible relationship between equity-based compensation and risk-taking will be stronger for the stock option grants compared to restricted stock grants. Therefore, the last hypothesis can be formulated as follows: H4: The relationship between stock option grants and risk-taking is stronger than the relationship between restricted stock grants and risk-taking in the banking industry. 18

In the upcoming chapter, the research methods will be described, which will be used in order to obtain an insight in the relationship between equity-based compensation and risk-taking. With the obtained results, answers regarding the above formulated hypotheses will be provided. 19

4. Methodology This chapter will provide an overview of the methods that are used throughout this research. Firstly, it will be described why this research will make use of a multivariate regression analysis. Secondly, I will take a look at the measurement of equity-based compensation. A first regression model will be introduced, which will provide us with four market-based risk measures. Thirdly, a second regression model will be created, which brings all variables together and provides an insight into the relationship between equitybased compensation and risk-taking. Fourthly, two last regression models will be constructed to measure the relationship between option-based compensation and risk-taking and between stock-based compensation and risk-taking in particular. And lastly, a parametric analysis will be explained, which will be used for the comparison of risk levels over the years. 4.1 Multivariate regression analysis To test the above formulated hypotheses, a model has to be created in order to translate the relevant constructs into measurable variables. To provide useful answers on the hypotheses, it has to be examined whether there exists a relationship between the main variables, being equity-based compensation and risktaking. A common method to examine these types of relationships which is used widely in the financial literature, is by running a regression model. Although this study focusses mainly on the relationship between these two variables, a multivariate regression model will be formed, rather than a more simplistic univariate model. Using a univariate model would imply that a one-on-one analysis between equity-based compensation and bank risk will be performed, excluding any other variable. However, it can be reasonably assumed that there are other factors that have a potential influence on this relationship. For example, firm fixed effects like the size of a bank in terms of total assets can have an impact on this relationship. Also, the economic circumstances at the time of measurement can have influence on the outcomes. Therefore, a multivariate model will be used, which includes such variables. Again, this multivariate model will not be used particularly because of the interest in the relationship between the dependent variable, equity-based compensation, and these added independent variables, but rather because it helps explaining the relationship between the two main variables of this study. Besides that, the outcomes of univariate regression models are more likely to be biased (as a result of omitted variables) compared to multivariate models, as it just examines the relationship between two variables without controlling for potential factors which may affect this relationship. 20

4.2 The measurement of equity-based compensation As mentioned, to test the hypotheses, the variables have to be transformed into measurable constructs. The most important independent variable in the regression model is equity-based compensation. Equitybased compensation will be measured as a proportion of the total compensation package of an executive, in order to make this construct both measurable and comparable. Using proportions instead of dollar amounts, executives compensation packages can be compared in a more relevant way. For example, CEO A gets one million dollar equity-based compensation and nine million dollar other compensation, whereas CEO B gets one million dollar equity-based compensation and three million dollar other compensation. If I compare equity-based compensation between executives A and B in dollar amounts, I will not find any differences. However, if I take a look at proportional amounts, I find that executive A only receives ten percent of its compensation in the form of equity, whereas executive B receives twentyfive percent equity-based compensation, a noteworthy difference. 4.3 The measurement of risk-taking Risk is a more complex variable in terms of measurement. As discussed in paragraph 2.5, there are different methods to measure risk. Bank risk cannot be measured with just one variable, as there exist different types of risk. For the purpose of this thesis, it is relevant to select risk measures which are likely to be influenced by the behavior of the CEO. As mentioned earlier, the constructs vega and delta are commonly used as risk measures in the academic literature regarding risk and CEO compensation structures. However, researchers who used these constructs to examine that relationship mainly focused on the risk that is derived from stock return volatilities. This study on the other hand, also wants to take a look at the potential influence of the compensation structures on other risks, which banks are facing. Therefore, a more extensive model, that is also well known and broadly used in the financial literature, will be used to obtain four market-based risk measures. These measures include (1) total risk, (2) idiosyncratic risk, (3) systematic risk, and (4) interest rate risk. All of these four risk measures will be obtained using an Ordinary Least Squares (OLS) regression model. The following model, which is also used by Chen et al. (2006), will be used to estimate the coefficients for systematic risk ( ) and interest rate risk ( ) for every bank in the sample (which will be extensively described in chapter 5), separately: (1) 21

Where; = the daily bank stock return, = the constant term, = the coefficient for systematic risk = the daily return on the equally-weighted index, = the coefficient for interest rate risk = the daily three-month T-bill yield from the Federal Reserve Bank of St. Louis, = the error term of the model. Then, from this equitation, I am able to estimate the two additional risk measures: = the standard deviation of the stock returns, which represents the total risk, and = the standard deviation of the residuals, which represents the idiosyncratic risk. Note that similar to the delta and vega method, this model also takes the volatility in stock returns into account. And thus, in addition, it looks at other risk aspects that banks face. 4.4 Measuring the relationship between equity-based compensation and risk-taking After running this first OLS regression, I am able to model each of the four risk measures as a function of the proportion of equity-based compensation. In order to do this, a second OLS regression model has to be constructed. This model should, logically, include both the risk measures as well as the proportions of equity-based compensation. However, in addition to these two variables, I have to include control variables that may affect this relationship as well. Adding these control variables eliminate biasness caused by omitted variables. Thereby, they act as additional explanatory variables, which increases the The firm specific variables which will be included in the regression model are debt ratio and total assets. In addition, control variables regarding CEO specific characteristics are incorporated, being age, tenure, and gender. Lastly, year dummies are included, which control for time effects. Brining all the variables together, the following OLS regression model is created to estimate the relationship between the risk measures and equity-based compensation: (2) 22

Where; = one of the four risk measures derived from model (1):,, or, = the proportion of equity-based compensation over total compensation, = total liabilities divided by the total assets of the bank, = the natural logarithm of the total assets of the bank, which measures bank size, = the CEO s in years, = the CEO s tenure in years, = a dummy variable which is 1 if the executive is a male, 0 for female, and = a dummy variable which is 1 or 0 for each of the years from 2007 to 2015. This model will also be used for regressions based on subsamples and cross-section analysis. 4.5 Option-based/stock-based compensation and risk-taking With the help of the two provided OLS regression models, I will be able to get insight into the relationship between equity-based compensation and risk-taking. However, it is expected that the influence of stock option grants on risk-taking is stronger than the relationship between restricted stock grants and risk-taking, as is discussed in prior chapters. Therefore, this research will also take a look particularly at the effect of stock options grants on risk-taking and restricted stock grants on risk-taking. To investigate these relationship, the same models will be used as described above. The only change will be that I will use the variable and, which respectively measure the proportion of stock option-based compensation and the proportion of stock-based compensation over total compensation. These variables will replace the variable from equation (2). The models will then look as follows: (3) (4) Where all variables remain the same as in equation (2), expect for; = the proportion of option-based compensation over total compensation, and = the proportion of stock-based compensation over total compensation. 23

4.6 The comparison of risk levels over the years Besides the regression models, one other technique will be used in this thesis: a parametric analysis. This analysis will be used to compare risk levels in the banking industry over the years. In particular, this test will be executed to investigate the differences in risk during the crisis (2007-2009), and post-crisis (2010-2015). By doing so, it should be able to test the third hypothesis of this study. The parametric analysis tests whether the differences in the risk measures alter significantly from zero. If that is the case, it can be concluded that there are significant differences in the bank risk during the crisis compared to post-crisis. This in turn, will lead to the acceptation of the third hypothesis. 24

5. Data 5.1 Sample construction For the construction of the dataset, several steps had to be taken. Not all relevant variables could be obtained with one single inquiry. However, it has to be said that I was able to gather all the data from one source, being Wharton Research Data Services (WRDS). Firstly, I started by gathering the data about the CEOs. With the help of the Execucomp database, I obtained information about executives in the U.S. banking industry for the period 2007-2015. Since this study is particularly focused on CEOs, all data about non-ceo executives has been dropped. Thereafter, all observations that contain missing values regarding the CEO s compensation package, were dropped. The variables age and gender were readily available. The variable equitycomp has been created by dividing the sum of stock option grants and restricted stock grants by total compensation. Execucomp includes the following compensation items into the total compensation variable, which is denoted by TDC1: salary, bonus, other annual, restricted stock grants, stock option grants, all other and, LTIP payouts. All variables are measured in thousands of dollars. Logically, optioncomp has been created in a similar way, by dividing only the value of the stock option grants by total compensation. I was also able to create the variable tenure, since the variables date became CEO and date left as CEO are also included in the Execucomp database. After completing all these steps, 870 observations were left. Thereafter, with the help of Compustat, annual fundamentals of the banks were gathered. As a result, I was able to create the debtratio variable, which is computed by total liabilities divided by total assets. Also, the natural logarithm of total assets, which serves as a measure of bank size, has been formed out of the Compustat database. For the purpose of this study, it more meaningful to take the natural logarithm of total assets, rather than looking at absolute values. There are two main reasons for this. Firstly, OLS regressions assume that standard errors, which are measured by the residuals, are normally distributed. However, it is possible that there exists some form of skewness in this variable. Secondly, the interpretation of changes in this variable are more useful, as the logarithm adjusts for relative changes. For example, the differences between total assets of $15.000 and $20.000, and between $300.000 and $305.000 are both $5000 in absolute terms. However, percentage-wise these differences are enormous. The natural logarithm puts these differences in perspective. After merging and dropping the missing values, the dataset that includes all independent variables consisted of 854 observations. Then, a dataset including all variables of the first regression model had to be created, in order to generate the four market-based risk measures. Firstly, I obtained the daily bank stock returns and the daily returns on the equally-weighted index using the Center for Research in Security Prices (CRSP). Secondly, I 25

gathered the daily three-month T-bill yield from the Federal Reserve Bank of St. Louis with the help of their website: https://fred.stlouisfed.org/series/tb3ms. The development of this yield over the years is shown in a graph, which can be found in Appendix I. Lastly, I merged the data, having all required data in one database. With the help of STATA, I was then able to create all the four market-based risk measures. These are measured for each bank, and for each year separately. The variable that represents total risk (, is obtained by taking the standard deviations of the bank stock returns. Subsequently, both beta coefficients, representing systematic risk ( ) and interest rate risk ( ), are created by running regression model (1). This model regresses the daily bank stock return on both the daily return on the equally-weighted index and the daily three-month T-bill yield from the Federal Reserve Bank of St. Louis. Out of this model, the standard deviations of the residuals are saved. These standard deviations represent idiosyncratic risk (, and therewith, all market-based risk measures are successfully obtained. Also, year dummies are created to control for potential year-to-year differences caused by unobservable or missing variables. Finally, the dataset containing the risk measures was merged with the main dataset, which includes all relevant variables for regression model (2). Again, missing and unreliable values were dropped. Ultimately, the final sample consists of 842 unique, complete, and reliable observations. Furthermore, the sample contains 158 different CEOs, who are divided over 109 U.S. banks. A complete list containing all these banks can be found in Appendix II. 5.2 Descriptive statistics In this section, descriptive statistics are provided and discussed. Firstly, in table 1 below, summary statistics about the both the dependent - as well as the independent variables are presented. All noteworthy numbers will be discussed. Thereafter, with the help of table 2, the development over the years in the banking industry of total compensation and the equity-based compensation items stock option grants and restricted stock grants, will be visualized. Out of table 1 it can be concluded that on average, CEOs in the sample got 32.06% of their total compensation in the form of equity-based compensation. This means that nearly one-third of the total compensation is paid in the form of stock option grants or restricted stock grants, during the period 2007-2015. However, as can be derived from the standard deviation, the minimum, and the maximum, the differences in proportion of equity-based compensation are quite large. However, looking at the small difference between the mean and the median in combination with the skewness, it can be said that EQUITYCOMP is normally distributed throughout the sample. Thereby, which also can be said for the 26

variable OPTIONCOMP, there were also CEOs who did not get rewarded with any form of equity-based compensation. Looking at OPTIONCOMP, I find that an average of 9.07% of the total compensation is reward in the form of stock option grants. This variable however, is skewed positively. This indicates that there are some CEOs in the sample who got very large proportions of option-based compensation. It is worth mentioning that the median of this variable is zero, which means that at least fifty percent of the CEOs in the sample did not have any stock option grants included in their compensation packages. The year-to-year differences in the compensation structures will be further discussed in table 2. Table 1: Descriptive statistics Variable N Mean Median Standard deviation Minimum Maximum Skewness EQUITYCOMP 842 32.06% 32.55% 24.95% 0.00% 96.80% 0.25 OPTIONCOMP 842 9.07% 0.00% 15.67% 0.00% 87.61% 2.13 Total risk 842 0.0264 0.0195 0.0172 0.0082 0.1006 1.76 Idiosyncratic risk 842 0.0203 0.0146 0.0142 0.0066 0.9304 2.11 Systematic risk 842 1.3998 1.3436 0.3837 0.0205 3.2583 1.02 Interest rate risk 842-0.0383-0.0169 0.3556-2.5301 1.8037-0.93 DEBTRATIO 842 0.8915 0.8916 0.0244 0.7802 0.9941-0.10 LN(TA) 842 9.6351 9.3113 1.4978 7.1712 14.7606 1.31 AGE 842 56.3670 56.0 6.7956 33.0 80.0 0.24 TENURE 842 8.8717 6.0 8.0940 0.0 36.0 1.26 GENDER 842 0.9727 1.0 0.1631 0.0 1.0-5.80 In this table, descriptive statistics of the sample are presented. EQUITYCOMP is the proportional amount of equitybased compensation relative to total compensation (in %); OPTIONCOMP is the proportional amount of optionbased compensation relative to total compensation (in %); measures total risk; measures idiosyncratic risk; measures systematic risk; measures interest rate risk; DEBTRATIO is the ratio of total liabilities over total assets; LN(TA) is the natural logarithm of the total assets (in thousands of dollars); AGE represents the age of the CEO (in years); TENURE represents the CEO s tenure (in years); and GENDER is a dummy variable which takes the value of 1 if the CEO is a male, 0 for female. Furthermore, I find that total risk has a mean value of 0.0264, whereas the average value for idiosyncratic risk is 0.0203. Systematic risk has a mean value of 1.3998, while the beta covering interest rate risk has an average of -0.0383. Looking at the first control variable DEBTRATIO, I find a mean value of 0.8915. This means that the average debt over total assets ratio is 0.8915. In the banking industry, this ratio is typically significantly higher compared to other industries. However not included in table 1, the average size of the banks in the sample in terms of assets is $93.16 million, with a standard deviation of $34.15 27

million. Further, I see that the average age of the CEOs is 56. The youngest CEO is John Mariner Kemper, who is the CEO of UMB Financial Corporation, being only 33 years old. On the other hand, the oldest one is 80 years old, being Bill G. Hartley from Southside Bancshares Inc. The age variable is relatively normally distributed among the sample. The average CEO s tenure is 8.87 years, with a minimum of 0 years, and a maximum of 36 years. Finally, I find an average of 0.97 for the gender dummy variable. This implies that roughly 97% of the sample consists of males, whereas only 3% of the sample consists of female CEOs. The structure of CEO s compensation packages changes over the years. Table 2 below visualizes the development over the years of compensation structures in the banking industry. Again, noteworthy values will be discussed below this table. Table 2: The structure of compensation packages over the years in the banking industry Year N AVG TOTAL COMP AVG OPTION COMP AVG STOCK COMP AVG OPTION % AVG STOCK % AVG EQUITY % 2007 100 3395.73 836.75 985.13 18.30 15.25 33.56 2008 97 2745.52 880.30 753.94 18.35 14.23 32.59 2009 91 2303.85 451.39 755.14 9.94 17.83 27.77 2010 93 3104.67 347.48 1096.94 5.13 21.81 26.94 2011 89 3476.69 384.75 1348.79 6.25 24.89 31.10 2012 91 3485.08 319.00 1342.33 5.65 25.70 31.35 2013 93 3627.43 227.48 1470.89 6.06 25.19 34.24 2014 95 4097.98 215.32 1823.32 4.99 29.62 34.61 2015 93 4043.63 216.79 1740.84 5.82 30.22 36.04 In this table, statistics regarding compensation structures are presented over the period 2007-2015. AVG TOTAL COMP is the average total compensation (in thousands of dollars); AVG OPTION COMP is the average optionbased compensation (in thousands of dollars); AVG STOCK COMP is the average stock-based compensation; AVG OPTION % is the average proportion option-based compensation over total compensation; AVG STOCK % is the average proportion stock-based compensation over total compensation; and AVG EQUITY % is the average proportion of equity-based compensation over total compensation. This is the sum of AVG OPTION % and AVG STOCK %. Looking at table 2, I see that the average total compensation dropped impressively by roughly one million dollar from 2007 to 2009. These decreases in 2008 and 2009 are most likely caused by the financial crisis that hit the banking industry severely. After this period however, the average total compensation increases steeply to over four million dollar. Thereby it has to be mentioned, although not included in table 2, that the standard deviation of the average total compensation is quite large. This means that the total 28

compensation differs significantly among the CEOs in the banking industry. Further, I see an even more impressive decrease in both the absolute values of the average option-based compensation as well as the proportional value of this variable. Over the period from 2007 to 2015, the average option-based compensation dropped by roughly 75%. It is mentioned numerous times in the existing financial literature that option-based compensation likely induces risk-taking. Many economists argue that this risk-taking behavior by CEOs is an important factor which has contributed to the financial crisis. After the crisis, regulation has become stricter. Option-based compensation could have decreased because of the potential risk that comes with it. Meanwhile, the average stock-based compensation increased steeply over the years. However, there seems to be a similar pattern as for the average total compensation. In other words, I see a decrease during the crisis, but a significant increase afterwards. In terms of percentages, logically, similar conclusions can be drawn as drawn for the absolute values. Lastly, summing up option- and stock-based compensation, it can be said that the proportion of equitybased compensation remains relatively steady over the years. On average, about one-third of total compensation is rewarded in equity-based components. However, although the overall equity-based compensation may seem to be steady over the years, the structure of this type of compensation has changed enormously. Post-crisis, the largest part of equity-based compensation comes from restricted stock grants, whereas during the crisis years (2007-2009), stock option grants were more popular. Chen et al. (2006) found that option-based compensation was actually increasing impressively from 1992 to 2000. They argue that this increase is mainly the result of the deregulation at that time. For the time-period used in this study, I find opposite results. As mentioned, it is likely that stricter regulation post-crisis along with more cautiousness among CEOs has lead to this decrease in option-based compensation. More arguments for this effect will be provided, along with the results, in the next chapter. 29

6. Results In this section, the results of the comprehensively discussed models of chapter 4 are presented. These results will be discussed in combination with the prior academic literature. It is investigated whether there exists a relationship equity-based compensation and risk-taking. Once again, the four market-based risk measures, as extensively discussed in chapter 4, will be used as the dependent variables. According to first hypothesis, it is expected that equity-based compensation is positively related to risk-taking. A significant positive sign for this variable in the regression model will support this hypothesis. On the contrary, based on the second hypothesis, a negative relationship between equity-based compensation and risk-taking is expected. This hypothesis will be supported by a negative sign on the equity-based compensation variable in the regression model. The third hypothesis will be tested with a parametric analysis, as discussed in paragraph 4.6. With this test, the differences between risk levels, which banks are facing during the crisis period and during the post-crisis period, will be investigated. The fourth and last hypothesis about option-based and stock-based compensation will be tested with the help of regression models based on equations (3) and (4). The findings of the parametric analysis will be provided in paragraph 6.1. Thereafter, the results of the multivariate regression models will be provided in sections 6.2, 6.3, and 6.4. 6.1 Parametric analysis Table 3 below presents the results of the parametric analysis. The mean values for the four-market based risk measures, (1) total risk, (2) idiosyncratic risk, (3) systematic risk, and (4) interest rate risk, are calculated for the periods 2007-2009 and 2010-2015. In order to do so, the total sample is divided into two subsamples. The first period represents years during the financial crisis, whereas the latter represents the post-crisis period of this study. Statistical significance is measured with the T-statistic, with the help of a T-test. From these results, it can be concluded that the mean values for all four market-based risk measures are impressively higher during the time period 2007-2009, compared to the time period 2010-2015. This indicates that U.S. banks face higher risk levels during the crisis, compared to post-crisis. For example, looking at total risk, which measured by the standard deviation of the daily stock returns, I find an average value of 0.0408 over the years 2007-2009, whereas the mean value of this standard deviation was only 0.0190 over the years 2010-2015. Looking at the obtained T-statistics from the T-tests, I find that these results are statistically significant on a 99% confidence level. These results are in line with the findings of Guiso, Sapienza, and Zingales (2013) about time-varying risk. Furthermore, it has to be said 30

that similar results have been found for other financial crises in the past. Schwert (1989) for example, found that stock return variability was also extremely high during the Great Depression of 1929-1939. Table 3: The comparison between risk levels during the crisis period and the post-crisis period Mean value of: For the time period 2007-2009 (N=288) For the time period 2010-2015 (N=554) T-statistic for the difference in means Total risk, 0.0408 0.0190-57.70*** Idiosyncratic risk, 0.0314 0.0145-50.11*** Systematic risk, 1.6023 1.2945-24.60*** Interest rate risk, 0.2826-0.7301-4.34*** In this table, the results of the comparison between risk levels during the crisis period (2007-2009) and the postcrisis period (2010-2015) are shown. The risk-measures (1) total risk, (2) idiosyncratic risk, (3) systematic risk, and (4) interest rate risk are taken into consideration. Statistical significant difference at the 99% level is denoted by ***. Although these results may not seem very striking, as it seems intuitively logical that a financial crisis is associated with higher levels of risk, it is important to note these differences in risk over the years. Because observing these differences is just a first step, the next step should be to look at the origins of these significant differences. That is what will be examined with the help of the upcoming multivariate regression analyses. Furthermore, these results demonstrate why it is necessary to include year dummies into the regression models. If the models are not controlled for year-to-year differences, outcomes may become biased. This, in turn, will lead to unreliable conclusions regarding the relationship between equity-based compensation and risk-taking. 6.2 Multivariate regression analysis: equity-based compensation and risk-taking The outcomes of the regression model based on equation (2), which is constructed in section 4.4, will be presented in this section. In table 4 below, the results of four OLS regression models are shown. They all investigate the effect of equity-based compensation on each of the risk variables. The four market-based risk measures are the dependent variables, whereas equity-based compensation and the other control variables act as independent variables. For all variables, the coefficients are given together with their corresponding T-statistics (in parentheses). Furthermore, the number of observations and the adjusted R- 31

squared are both given. The adjusted R-squared shows the percentage of variation which is explained by the independent variables that affect the dependent variable, adjusted for the number of variables. Table 4: Outcomes of the Ordinary Least Square (OLS) regression model measuring the relationship between the four market-based risk measures and equity-based compensation. Dependent variables: Independent variables: -0.00028-0.00083-0.01252 0.12225 (-0.18) (-0.55) (-0.23) (0.24) 0.08933*** 0.10613*** -0.47511 0.06248 (6.45) (8.13) (-1.00) (1.39) -0.00005-0.00043** -0.03479*** -0.25100*** (-0.18) (-1.69) (3.77) (2.90) 0.00003 0.00005-0.00296-0.01585 (0.47) (1.02) (-1.59) (-0.90) -0.00006-0.00008* -0.00021-0.00799 (-1.39) (-1.85) (-0.13) (-0.54) -0.00609*** -0.00583*** -0.13139* -0.29765 (-3.03) (-3.07) (-1.90) (-0.46) Constant -0.04572*** -0.06843*** 1.85945*** -8.06632* (-3.54) (-5.62) (4.18) (-1.93) Year dummies YES YES YES YES N 842 842 842 842 Adjusted R² 69.63% 60.50% 28.18% 26.11% In this table, the results of four OLS regression models are shown. Both the coefficients as well as the T-statistics are given (in parentheses). The risk measures at the top of the table are the dependent variables, whereas equitybased compensation, together with the control variables on the left side, act as the independent variables. measures total risk; measures idiosyncratic risk; measures systematic risk; measures interest rate risk; EQUITYCOMP is the proportional amount of equity-based compensation relative to total compensation; DEBTRATIO is the ratio of total liabilities over total assets; LN(TA) is the natural logarithm of the total assets; AGE represents the age of the CEO; TENURE represents the CEO s tenure; and GENDER is a dummy variable which takes the value of 1 if the CEO is a male, 0 for female. ***, **, * show statistical significance at the 99%, 95%, and 90% level respectively. Looking at the coefficients of equity-based compensation, I find that the signs are negative for three risk measures, being total (1) total risk, (2) idiosyncratic risk, and (3) systematic risk. The coefficient for interest rate risk shows a positive sign. However, neither of these outcomes are significant. This indicates that there does not exist a strong enough relationship between equity-based compensation and risk-taking for the total sample, according to this regression model. Therefore, it can be concluded that these results do neither support the first - nor the second hypothesis. 32

A possible explanation for this can be found in the differences in characteristics of stock options grants and restricted stock grants, as analyzed in chapter 2 (Irving, Landsman & Lindsey, (2011); Dodonova & Khoroshilov, (2006); Mehran (1995). In this chapter, it is concluded with the help of the existing financial literature that the incentives which arise from equity-based compensation are stronger for option-based compensation than for stock-based compensation. This is in line with the findings of Guay (1999), who concluded that only stock options lead significantly affect risk-taking. Thereby, it is shown in table 2 that the largest part of equity-based compensation consists of restricted stock grants. This is especially true for the period from 2010 to 2015, where stock-based compensation was about five larger than option-based compensation. Therefore, it is likely that a possible impact of stock option grants on risk is mitigated or outweighed by the restricted stock grants. Although not shown in the table, I find that the signs of the coefficients remain unchanged, but the T-statistics increase, when I execute the regression models only for the years in which option-based compensation was larger than stock-based compensation. This is in line with the expectation that the influence of stock option grants on risk is greater than the impact of restricted stock grants. It is also possible to explain the insignificant results with the help of the hedging theory of Smith and Stulz (1985). In contrast to equity-based compensation, cash compensation is untied to the bank. Compensation in cash provides the CEO with the opportunity to diversify his/her own portfolio. Restricted stock grants on the other hand, are tied to the bank. These may not be sold in the market. Therefore, it is wise for CEOs to invest in other instruments to diversify their personal portfolios, as the stock-based compensation is greatly undiversified. CEOs can invest in other securities, to mitigate the firm specific risk of their portfolios. By doing so, the CEOs improve the risk-return tradeoff of their investment portfolio. This process of diversifying their portfolios can be described as hedging. Smith and Stulz (1985) concluded that CEOs who have greater hedging possibilities are more likely to execute risky policies. This implies that the impact of equity-based compensation on risk-taking will decrease when the ability to hedge decreases. Looking at the development of cash compensation from 2007 to 2015, I find several noteworthy things that could explain the decrease in the ability to hedge among banking CEOs. Firstly, the cash compensation was relatively low in the years 2007 and 2008, probably due to the financial crisis. Thereafter, this salary rises to roughly $9 million in 2010. However, after 2010 it decreases for the next three years, to a rough $8 million in 2013. The development of cash compensation over the years can be found in appendix III. Secondly, as can be seen in table 2, the restricted stock grants kept increasing during the whole period from 2007 to 2015. This increase in stock-based compensation in combination with the decrease in cash compensation, could have resulted in poorer hedging opportunities. And this, in turn, could explain (at least partially) an insignificant relationship between equity-based compensation and risk-taking. 33

Looking at the control variables, I firstly find two significant and positive coefficients for DEBTRATIO. These coefficients support the expectation that debt is positively related with risk (D Hulster, (2009); Hamada, (1972). Secondly, I find significant and negative coefficients for LN(TA) in all four models, suggesting that larger banks face lower risk levels. This can be explained by the ability for larger banks to diversify, which results in lower levels of risk. Thirdly, the insignificant coefficients for age en tenure suggest that there is no relationship between age and risk-taking and between tenure and risk-taking. Lastly, I find significant and negative coefficients for GENDER in three of the four models. This would indicate that female CEOs tend to take more risks compared to male CEOs. This is not in line with the overall findings in the academic literature. For example, Brynes, Miller, and Schafer (1999) find evidence for greater risk-taking behavior among males. These findings are supported by Nicholson et al. (2005). A possible explanation for the findings in this study could be that only 23 out of the 842 observations are female CEOs. These 23 observations include only five different female CEOs. Because of this very small number, the findings for this gender variable may be distorted and are therefore not generalizable. The adjusted R² is fairly high for all four models. In particular, this number is impressively high in the first two models, including total risk and idiosyncratic risk. For total risk, this means that about 70% of the variation in this model is explained by the independent variables. For idiosyncratic risk, this is more than 60%. This also indicates that it is not likely that there exists endogeneity due to an omitted variable, which explains a lot of the variation in the dependent variable. 6.3 Multivariate regression analysis: subsamples and cross-section The results in section 6.2 suggest that there does not exist a strong enough relationship between equitybased compensation and risk-taking for the total sample from 2007 to 2015. However, it is still possible that there is a exists a relationship between these two components when the sample is split into subsamples, based on one of the control variables. In addition, it is useful to look if there exists a relationship between equity-based compensation and risk-taking in a certain year, or a certain period, rather than over the whole period from 2007 to 2015. To test this, a cross-section analysis will be executed as well. I started with splitting the dataset in quartiles on the basis of debt ratio. I find that the 25% percentile has a value of 0.87714, and the 75% percentile has a value of 0.90584. This means that the 25% of banks with the lowest debt ratios in the dataset have a debt ratio of 0.87714 or lower, while the 25% of banks with the highest debt ratio have a ratio of 0.90584 or higher. The outcomes of running the same regression model as in section 6.2, based on these subsamples, are presented in tables 5 and 6. 34

Table 5: Outcomes of the Ordinary Least Square (OLS) regression model measuring the relationship between the four market-based risk measures and equity-based compensation, for banks with a debt ratio of 0.87714 or lower. Dependent variables: Independent variables: -0.00265-0.00363-0.007852 0.07376 (-0.68) (-0.99) (-0.63) (0.06) 0.05064*** 0.05523*** 0.16456 0.22799 (8.14) (9.37) (0.59) (1.18) -0.00000-0.00041-0.03847* 0.32678* (-0.01) (-0.70) (1.95) (1.70) -0.00010-0.00009-0.00253-0.015997 (-0.88) (-0.80) (-0.70) (-0.45) 0.00009 0.00011-0.00343-0.02762 (0.78) (0.97) (-0.94) (-0.77) -0.00189-0.00071-0.16260-0.11963 (-0.33) (-0.13) (-0.88) (-0.07) Constant -0.44040*** -0.48328*** 0.05684-2.36498 (-7.34) (-8.50) (0.03) (-1.27) Year dummies YES YES YES YES N 211 211 211 211 Adjusted R² 67.91% 63.29% 26.32% 12.37% In this table, the results of four OLS regression models are shown. Both the coefficients as well as the T-statistics are given (in parentheses). The risk measures at the top of the table are the dependent variables, whereas equitybased compensation, together with the control variables on the left side, act as the independent variables. measures total risk; measures idiosyncratic risk; measures systematic risk; measures interest rate risk; EQUITYCOMP is the proportional amount of equity-based compensation relative to total compensation; DEBTRATIO is the ratio of total liabilities over total assets; LN(TA) is the natural logarithm of the total assets; AGE represents the age of the CEO; TENURE represents the CEO s tenure; and GENDER is a dummy variable which takes the value of 1 if the CEO is a male, 0 for female. ***, **, * show statistical significance at the 99%, 95%, and 90% level respectively. From these two tables, it can be concluded that there seems to exist a negative relationship between equity-based compensation and risk-taking for banks with relatively high debt ratios, at least for three of the four market-based risk measures. On the other hand, no evidence is found for this relationship for banks with relatively low debt ratios. A possible explanation for these outcomes can be found with the help of the agency theory. The agency costs of the free cash flows are reduced because of the creation of debt, as it reduces the available money to invest for CEOs. If CEOs use the free cash flows for investing in risky projects, chances are that the bank is not able to repay all the debt when these risky investments do not payoff (Kochhar, 1996). This, in turn, has also negative effects on the value of the equity-based 35

compensation for the CEO, as it will logically decrease when the bank is doing poorly. The higher the debt ratio of the bank, the stronger this effect will be. This could explain the negative relationship between equity-based compensation and risk-taking for banks with relatively high debt ratios, as CEOs of banks with high debt ratios may be more conservative than CEOs of banks with lower levels of debt. Table 6: Outcomes of the Ordinary Least Square (OLS) regression model measuring the relationship between the four market-based risk measures and equity-based compensation, for banks with a debt ratio of 0.90584 or higher. Dependent variables: Independent variables: -0.00472** -0.00394** -0.01660* 0.06773 (2.26) (2.09) (1.77) (0.07) 0.04905* 0.44040* 0.37654*** 0.79936 (1.67) (1.66) (2.96) (0.60) 0.00015-0.00003 0.01373 0.27001 (0.35) (-0.07) (0.71) (1.39) -0.00007-0.00006-0.00465 0.01311 (-0.81) (-0.79) (-1.22) (0.34) 0.00006 0.00000 0.01067*** -0.01552 (0.96) (0.17) (3.60) (-0.52) -0.00396* -0.00221-0.25841** 0.91434 (-1.67) (-1.03) (-2.43) (0.85) Constant 0.01107 0.00450-1.60336-1.10928 (0.44) (0.20) (-1.42) (-0.98) Year dummies YES YES YES YES N 211 211 211 211 Adjusted R² 74.15% 64.52% 26.64% 42.15% In this table, the results of four OLS regression models are shown. Both the coefficients as well as the T-statistics are given (in parentheses). The risk measures at the top of the table are the dependent variables, whereas equitybased compensation, together with the control variables on the left side, act as the independent variables. measures total risk; measures idiosyncratic risk; measures systematic risk; measures interest rate risk; EQUITYCOMP is the proportional amount of equity-based compensation relative to total compensation; DEBTRATIO is the ratio of total liabilities over total assets; LN(TA) is the natural logarithm of the total assets; AGE represents the age of the CEO; TENURE represents the CEO s tenure; and GENDER is a dummy variable which takes the value of 1 if the CEO is a male, 0 for female. ***, **, * show statistical significance at the 99%, 95%, and 90% level respectively. After splitting the dataset on the basis of debt ratio, I split the dataset on the basis of size. The variable used in this study to measure bank size is the natural logarithm of total assets. I split the dataset on the median, which has a value of 9.311283. Tables 7 and 8 show the outcomes of the OLS regression models, based on the two subsamples on the basis of bank size in terms of assets. 36

Table 7: Outcomes of the Ordinary Least Square (OLS) regression model measuring the relationship between the four market-based risk measures and equity-based compensation, for banks with a natural logarithm of total assets of 9.311283 or lower. Dependent variables: Independent variables: 0.00176 0.00228-0.02191 0.06820 (0.75) (0.54) (-0.28) (0.83) 0.14911*** 0.16494*** 0.84708 0.73148 (8.25) (9.39) (1.38) (0.12) -0.00080-0.00235** -0.17542*** -0.60972 (-0.75) (-2.24) (4.81) (-1.62) 0.00008 0.00008 0.00038-0.04057 (1.00) (1.11) (0.14) (-1.49) -0.00010-0.00009-0.00309-0.00669 (-1.59) (-1.52) (-1.42) (-0.30) -0.00846*** -0.00758*** -0.02355*** 0.44336 (-3.66) (-3.37) (-3.01) (0.55) Constant -0.10605*** -0.11072*** -0.79066 1.84159 (-5.14) (-5.53) (-1.13) (0.26) Year dummies YES YES YES YES N 421 421 421 421 Adjusted R² 67.43% 58.79% 26.69% 25.58% In this table, the results of four OLS regression models are shown. Both the coefficients as well as the T-statistics are given (in parentheses). The risk measures at the top of the table are the dependent variables, whereas equitybased compensation, together with the control variables on the left side, act as the independent variables. measures total risk; measures idiosyncratic risk; measures systematic risk; measures interest rate risk; EQUITYCOMP is the proportional amount of equity-based compensation relative to total compensation; DEBTRATIO is the ratio of total liabilities over total assets; LN(TA) is the natural logarithm of the total assets; AGE represents the age of the CEO; TENURE represents the CEO s tenure; and GENDER is a dummy variable which takes the value of 1 if the CEO is a male, 0 for female. ***, **, * show statistical significance at the 99%, 95%, and 90% level respectively. From tables 7 and 8, it can be concluded that there seems to exist a negative relationship between equitybased compensation and risk-taking for the, in terms of size, relatively large banks, at least for two market-based risk measures (total risk and idiosyncratic risk). However, it cannot be concluded that this relationship also exists for the relatively smaller banks. It is actually difficult to rationalize these outcomes, as there is not much written about the relationship between equity-based compensation and risk-taking in combination with differences in bank size. As mentioned before, in the existing literature it is shown multiple times that larger banks face lower levels of risk as a result of diversification opportunities, holding all other factors constant. Yet, this does not necessarily explain a negative 37

relationship between equity-based compensation and risk for these larger banks. However, what can be concluded is that these results are (at least partially) in line with the findings of Low (2006), Parrino et al. (2005), Rogers (2002), and Smith and Stulz (1985), who all found evidence for a negative relationship between equity-based compensation and risk-taking. Table 8: Outcomes of the Ordinary Least Square (OLS) regression model measuring the relationship between the four market-based risk measures and equity-based compensation, for banks with a natural logarithm of total assets of 9.311283 or higher. Dependent variables: Independent variables: -0.00409** -0.00436** -0.05166-0.23918 (-1.98) (-2.33) (-0.70) (-0.37) -0.00764-0.00215-0.10164 0.11750* (-0.36) (-0.11) (-1.34) (1.78) 0.00056 0.00026-0.02551* -0.03326*** (1.52) (0.79) (1.93) (1.70) -0.000143** -0.00011* -0.00603** 0.00082 (-2.00) (-1.69) (-2.36) (0.04) -0.00002-0.00005 0.00128-0.00375 (-0.32) (-0.86) (0.55) (-0.19) -0.00008 0.00029-0.08272-0.15664 (-0.02) (0.08) (-0.59) (-1.28) Constant 0.02970 0.02167 0.24758*** -1.25738 (1.47) (1.18) (3.41) (-1.99) Year dummies YES YES YES YES N 421 421 421 421 Adjusted R² 76.37% 69.12% 38.77% 32.17% In this table, the results of four OLS regression models are shown. Both the coefficients as well as the T-statistics are given (in parentheses). The risk measures at the top of the table are the dependent variables, whereas equitybased compensation, together with the control variables on the left side, act as the independent variables. measures total risk; measures idiosyncratic risk; measures systematic risk; measures interest rate risk; EQUITYCOMP is the proportional amount of equity-based compensation relative to total compensation; DEBTRATIO is the ratio of total liabilities over total assets; LN(TA) is the natural logarithm of the total assets; AGE represents the age of the CEO; TENURE represents the CEO s tenure; and GENDER is a dummy variable which takes the value of 1 if the CEO is a male, 0 for female. ***, **, * show statistical significance at the 99%, 95%, and 90% level respectively. Besides creating subsamples on the basis of debt ratio and total assets, I executed regression models based on subsamples on gender, tenure, and age. However, same as for the total regression model outcomes in table 4, no significant outcomes or differences were found in these regressions. 38

Although no significant relationship between equity-based compensation and risk-taking is found based on the total sample from 2007 to 2015, it is still possible that there is exists a relationship between these components in a certain year in this period. That is why it is useful to execute a cross-section analysis. The same regression model based on equation (2) is used, except the year dummies. In other words, there is taken a look at the same relationship, with the same control variables, but this time for each year from 2007 to 2015 separately. I do only find evidence for a relationship between equity-based compensation and risk-taking in the year 2014, at least for two market-based risk measures. In all other years, no significant results were obtained regarding this relationship. The outcomes of the regression model for the year 2014 are presented in table 8 below. Table 9: Outcomes of the Ordinary Least Square (OLS) regression model measuring the relationship between the four market-based risk measures and equity-based compensation, for the year 2014. Dependent variables: Independent variables: -0.00271** -0.00356*** -0.11369-0.17587 (-2.03) (-2.58) (0.94) (-0.77) -0.01529*** -0.01357-0.87139 0.26965 (-1.37) (-1.17) (-0.86) (1.41) -0.00050** -0.00036* -0.05443*** 0.13095*** (-2.45) (-1.69) (-2.93) (3.72) -0.00001 0.00002-0.00719* 0.03020 (-0.27) (0.43) (-1.80) (0.40) 0.00000-0.00003 0.00881*** -.11147** (0.17) (-1.00) (3.00) (-2.00) 0.00021 0.00075-0.1206-0.11212 (0.12) (0.43) (-0.78) (-0.38) Constant 0.03366*** 0.02620** 2.91040*** -3.81736** (3.31) (2.49) (3.16) (-2.19) N 95 95 95 95 Adjusted R² 19.32% 16.82% 14.81% 17.73% In this table, the results of four OLS regression models are shown. Both the coefficients as well as the T-statistics are given (in parentheses). The risk measures at the top of the table are the dependent variables, whereas equitybased compensation, together with the control variables on the left side, act as the independent variables. measures total risk; measures idiosyncratic risk; measures systematic risk; measures interest rate risk; EQUITYCOMP is the proportional amount of equity-based compensation relative to total compensation; DEBTRATIO is the ratio of total liabilities over total assets; LN(TA) is the natural logarithm of the total assets; AGE represents the age of the CEO; TENURE represents the CEO s tenure; and GENDER is a dummy variable which takes the value of 1 if the CEO is a male, 0 for female. ***, **, * show statistical significance at the 99%, 95%, and 90% level respectively. 39

It is difficult to explain why this relationship exists specifically in 2014. However, as argued before, due to the impact of the latest financial crisis, CEOs have become more conservative in their investing behavior. This is at least partially a result of stricter regulation, mainly due to the introduction of Basel III (Blundell-Wignall & Atkinson (2010), Cosimano & Hakura (2011). That could at least explain why this relationship is found in a post-crisis year, rather than in a crisis year. 6.4 Multivariate regression analysis: option-based/stock-based compensation and risk After obtaining insignificant results on the relationship between equity-based compensation and risktaking for the total dataset, I started looking for possible explanations for this insignificancy. As mentioned in section 5.2, and being extensively discussed in section 2.2, a possible explanation can be found in the differences in the characteristics of the equity-based compensation components. It is suggested that the possible influence of stock option grants on risk-taking is stronger compared to the influence of restricted stock grants. Thereby, the proportion of option-based compensation decreased, whereas the proportion of stock-based compensation heavily increased, over the period from 2007 to 2015. It is therefore possible that the influence of stock option grants on risk is unnoticeable, or offset, due to the restricted stock grants. Chen et al. 2006 also mention the importance of investigating the relationship between stock option grants and risk-taking in particular: although both managerial stock ownership and option-based compensation are equity ownership, the former represents current ownership and the latter future ownership. While the current ownership may increase or decrease in value, the future ownership (stock options) can experience more dramatic outcomes with exercise values that may reasonably fluctuate from zero to several million dollars due to the leverage effect. This possibly makes stock options a more powerful variable for investigating risk related principal-agent problems in banking (Chen et al., 2006, p. 917). The results of the regression models based on equations (3) and (4), which are constructed in section 4.5, will be given. In table 10 and 11 below, the results of four OLS regression models are shown. These equations are different from the equations in table 4, as they investigate the relationship between optionbased compensation and risk and stock-based compensation and risk in particular, rather than the combined relationship with risk. All other variables remain the same in this model. Looking at the coefficients for the variable OPTIONCOMP in table 10, which measure the proportional amount of option-based compensation, I find three negative coefficients. And more importantly, these coefficients are, in contrast to the coefficients for equity-based compensation, significant. This implies that, in contrast to equity-based compensation, there seems to exist a negative relationship between option-based compensation and risk-taking in the banking industry. These findings are in line with the 40

findings of Low (2006), Parrino et al. (2005), Rogers (2002), and Smith and Stulz (1985). At the other hand, these results are not in line with the findings of for example John et al. (2000), and Chen et al. (2006). At least for the latter, the differences in the findings can be explained due to differences in both the structure of the compensation packages and the time period, which are extensively discussed in section 5.2. These findings support the risk aversion hypothesis (H2) over the risk-taking hypothesis (H1). All other variables in the regression models do not change significantly compared to coefficients in table 4. This means that all conclusions formulated in section 6.2 hold regarding these variables. Table 10: Outcomes of the Ordinary Least Square (OLS) regression model measuring the relationship between the four market-based risk measures and option-based compensation. Dependent variables: Independent variables: -0.00481** -0.00416* -0.25300*** 0.79228 (-2.10) (-1.93) (-3.23) (1.07) 0.09084*** 0.10741*** -0.39593 0.60031 (6.57) (8.23) (-0.83) (1.34) -0.00004-0.00040* -0.03976*** -0.02431*** (0.18) (-1.87) (5.06) (3.27) 0.00001 0.00004-0.00352* -0.01424 (0.28) (0.86) (-1.89) (-0.81) -0.00006-0.00008* -0.00007-0.00864 (-1.34) (-1.77) (-0.05) (-0.59) -0.00612*** -0.00585*** -0.13311* -0.29357 (-3.05) (-3.09) (-1.93) (-0.45) Constant -0.04717*** -0.06934*** 1.78464*** -7.87299* (-3.66) (-5.71) (4.04) (-1.89) Year dummies YES YES YES YES N 842 842 842 842 Adjusted R² 69.79% 60.67% 29.07% 26.21% In this table, the results of four OLS regression models are shown. Both the coefficients as well as the T-statistics are given (in parentheses). The risk measures at the top of the table are the dependent variables, whereas optionbased compensation, together with the control variables on the left side, act as the independent variables. measures total risk; measures idiosyncratic risk; measures systematic risk; measures interest rate risk; OPTIONCOMP is the proportional amount of option-based compensation relative to total compensation; DEBTRATIO is the ratio of total liabilities over total assets; LN(TA) is the natural logarithm of the total assets; AGE represents the age of the CEO; TENURE represents the CEO s tenure; and GENDER is a dummy variable which takes the value of 1 if the CEO is a male, 0 for female. ***, **, * show statistical significance at the 99%, 95%, and 90% level respectively. 41

Table 11: Outcomes of the Ordinary Least Square (OLS) regression model measuring the relationship between the four market-based risk measures and stock-based compensation. Dependent variables: Independent variables: -0.0058-0.00454-0.13187-0.07137 (-0.94) (-1.07) (-0.99) (-1.32) 0.06801 0.09611*** 0.00251-0.20852 (1.53) (2.59) (0.00) (-0.48) -0.00258*** -0.00080** -0.08538*** 0.38419*** (2.68) (2.03) (3.99) (4.14) 0.00007 0.00007-0.00032-0.00408 (0.36) (0.44) (-0.08) (-0.23) 0.00005 0.00004-0.00216-0.01992 (0.32) (0.31) (-0.56) (-1.15) -0.01511** -0.01268** -0.30573* -0.37959 (-0.33) (-2.11) (-1.90) (-0.54) Constant -0.03234-0.06081* 1.16434-0.51412 (-0.77) (-8.50) (1.25) (-0.13) N 288 288 288 288 Adjusted R² 2.48% 3.18% 4.77% 5.27% In this table, the results of four OLS regression models are shown. Both the coefficients as well as the T-statistics are given (in parentheses). The risk measures at the top of the table are the dependent variables, whereas stock-based compensation, together with the control variables on the left side, act as the independent variables. measures total risk; measures idiosyncratic risk; measures systematic risk; measures interest rate risk; STOCKCOMP is the proportional amount of option-based compensation relative to total compensation; DEBTRATIO is the ratio of total liabilities over total assets; LN(TA) is the natural logarithm of the total assets; AGE represents the age of the CEO; TENURE represents the CEO s tenure; and GENDER is a dummy variable which takes the value of 1 if the CEO is a male, 0 for female. ***, **, * show statistical significance at the 99%, 95%, and 90% level respectively. From this table, it can be concluded that no evidence is found for a relationship between stock-based compensation and risk-taking, as the coefficients for the stock-based compensation variable are all not significant. This also means that the fourth hypothesis can be accepted, as I did find evidence for a relationship between option-based compensation and risk-taking. Regarding this hypothesis, it was expected that the relationship between option grants and risk-taking is stronger than the relationship between stock grants and risk-taking. This is in line with the expectations based on the differences in characteristics between stock option grants and restricted stock grants, as extensively discussed in both the introduction and section 2.2. 42

7. Conclusion and limitations In this study, it is tried to find empirical evidence for the relationship between equity-based compensation and risk-taking in the banking industry. To test this relationship, the compensation packages of 158 CEOs divided over 109 banks from 2007 to 2015 were investigated together with multiple CEO- and bankspecific variables. The main goal of this study was to find an answer on the following research question: To what extent, and in which direction, is CEO equity-based compensation related to risk-taking in the banking industry? This study contributes to existing literature by using more recent data, and by looking at both equity-based compensation components together, being stock option grants and restricted stock grants, rather than examining only one of these components. 7.1 Main findings With the help of OLS regression models, I was able to investigate the relationship between equity-based compensation and risk-taking. Based on the outcomes on the first empirical model based on equation (2), I showed that there is not found enough evidence for a relationship between equity-based compensation and risk-taking for the total sample for the period from 2007 to 2015. Two possible explanations are discussed that could explain these findings. Firstly, due to a decrease in cash compensation in combination with an increase in equity-based compensation during the period from 2007 to 2015, CEOs had less opportunities to diversify their personal portfolios. Possible incentives that arise from equitybased compensation regarding risk-taking could have become smaller, or even offset, because of this. This explanation is in line with the well-known hedging theory of Smith and Stulz (1985). Secondly, the changes of the compensation package structures over the investigated period could explain the insignificant results. Restricted stock grants are increased heavily from 2007 to 2015, while stock option grants drastically decreased in this period. In line with the findings of Dodonova and Khoroshilov (2006), and Irving, Landsman and Lindsey (2011), it was expected that the relationship between stock option grants and risk-taking is significantly stronger than the relationship between restricted stock grants and risk-taking. The changes in these two components could therefore explain why no relationship between equity-based compensation and risk-taking is found. I also empirically tested these differences between stock option grants and restricted stock grants, and their relationships with risk-taking, with the help of regression models built on the basis of equations (3) and (4) respectively. A negative relationship is found between stock option grants and risk-taking, whereas no evidence is found for a relationship between restricted stock grants and risk-taking. These findings are in line with the expectations and therefore hypothesis 4 can be accepted. 43

Also, after obtaining insignificant results in the first regression model, several subsamples have been created on the basis of debt ratio, size, age, tenure, and gender. From regressions based on these subsamples, two significant relationships were found: (1) a negative relationship between equity-based compensation and risk-taking for banks with a relatively high debt ratio, and (2) a negative relationship between equity-based compensation and risk-taking for relatively large banks in terms of total assets. In addition, a cross-section analysis has been performed to test the relationship between equity-based compensation and risk-taking in a specific year. With the help of this cross-section analysis, a third negative relationship is found between equity-based compensation and risk-taking in the year 2014. It has to be mentioned that the coefficients for the four market-based risk measures in the regression models for these three findings were not all significant, meaning that only partial evidence is found. The findings of negative relationships between equity-based compensation and risk-taking are in line with the findings of Low (2006), Parrino et al. (2005), Rogers (2002), and Smith and Stulz (1985). This means, as no positive relationships between these two variables are obtained, hypothesis 1 has to be rejected. On the other hand, hypothesis 2 has to be accepted partially. Although no empirical evidence is found in the complete regression model based on equation (2), some evidence is found for a negative relationship between equity-based compensation and risk-taking with the help of subsamples and cross-section analysis. Besides these main findings, a brief look has been taken at differences in risk levels in the banking industry over the years. Evidence is found for significant higher risk levels during the crisis years (2007-2009) compared to the post-crisis years (2010-2015). This leads to the acceptation of hypothesis 3. 7.2 Limitations and recommendations In this study, equity-based compensation is used as an independent variable, whereas risk-taking is used as the dependent variable. Therefore, only one direction of causality is taken into account. However, it can be fairly argued that equity-based compensation not only influences risk-taking, but also vice versa. This means that the causation of the relationship between equity-based compensation and risk-taking runs in both directions. The obtained coefficients of the OLS regressions may therefore be biased due to endogeneity, which is a result of reverse causality. Therefore, reverse causality has to be considered as a major limitation of this study. For further research, it is recommended to use lagged values for the risk variables and equity-based compensation or to run regressions in a simultaneous model, instead of using the Ordinary Least Squares approach, in order to reduce or even eliminate this limitation. 44

8. References: Agrawal, A., & Mandelker, G. N. (1987). Managerial incentives and corporate investment and financing decisions. The Journal of Finance, 42(4), 823-837. Bebchuk, L. A., & Spamann, H. (2009). Regulating bankers' pay. Cambridge: Harvard Law School. Belkhir, M., & Chazi, A. (2010). Compensation Vega, Deregulation, and Risk Taking: Lessons from the US Banking Industry. Journal of Business Finance & Accounting, 37(9 10), 1218-1247. Blundell-Wignall, A., & Atkinson, P. (2010). Thinking beyond basel III. OECD Journal: Financial Market Trends, 2010(1), 9-33. Bolton, P., Mehran, H., & Shapiro, J. (2015). Executive compensation and risk taking. Review of Finance, 19, 2139-2181. Brander, J. A., & Poitevin, M. (1992). Managerial compensation and the agency costs of debt finance. Managerial and Decision Economics, 13(1), 55-64. Byrnes, J. P., Miller, D. C., & Schafer, W. D. (1999). Gender differences in risk taking: A meta-analysis. Cai, J., Cherny, K., & Milbourn, T. (2010). Compensation and risk incentives in banking and finance. Economic Commentary, 13(1), 1-6. Chauvin, K. W., & Shenoy, C. (2001). Stock price decreases prior to executive stock option grants. Journal of Corporate Finance, 7(1), 53-76. Chen, C. R., Steiner, T. L., & Whyte, A. M. (2006). Does stock option-based executive compensation induce risk-taking? An analysis of the banking industry. Journal of Banking & Finance, 30(3), 915-945. Coles, J. L., Daniel, N. D., & Naveen, L. (2006). Managerial incentives and risk-taking. Journal of financial Economics, 79(2), 431-468 Cosimano, T. F., & Hakura, D. (2011). Bank behavior in response to Basel III: A cross-country analysis. D'Hulster, K. (2009). The leverage ratio: A new binding limit on banks. DeFusco, R. A., Johnson, R. R., & Zorn, T. S. (1990). The effect of executive stock option plans on stockholders and bondholders. The Journal of Finance, 45(2), 617-627. DeYoung, R., Peng, E. Y., & Yan, M. (2010). Executive compensation and business policy choices at US commercial banks. Kansas: The Federal Reserve Bank of Kansas City. 45

Dictionary, O. E. (2016). Oxford English Dictionary. The Library. Dodonova, A., & Khoroshilov, Y. (2006). Optimal Incentive Contracts for Loss Averse Managers: Stock Options versus Restricted Stock Grants. Financial Review, 41(4), 451-482. Guay, W. R. (1999). The sensitivity of CEO wealth to equity risk: an analysis of the magnitude and determinants. Journal of Financial Economics, 53(1), 43-71. Guiso, L., Sapienza, P., & Zingales, L. (2013). Time varying risk aversion (No. w19284). National Bureau of Economic Research. Hagendorff, J., & Vallascas, F. (2011). CEO pay incentives and risk-taking: Evidence from bank acquisitions. Journal of Corporate Finance, 17(4), 1078-1095. Hamada, R. S. (1972). The effect of the firm's capital structure on the systematic risk of common stocks. The Journal of Finance, 27(2), 435-452. Houston, J. F., & James, C. (1995). CEO compensation and bank risk Is compensation in banking structured to promote risk taking? Journal of Monetary Economics, 36(2), 405-431. Hubbard, R. G., & Palia, D. (1995). Executive pay and performance evidence from the US banking industry. Journal of financial economics, 39(1), 105-130. Irving, J. H., Landsman, W. R., & Lindsey, B. P. (2011). The valuation differences between stock option and restricted stock grants for US firms. Journal of Business Finance & Accounting, 38(3 4), 395-412. Jensen, M. C., & Murphy, K. J. (1990). Performance pay and top-management incentives. Journal of political economy, 98(2), 225-264. John, T. A., & John, K. (1993). Top management compensation and capital structure. The Journal of Finance, 48(3), 949-974. John, K., Saunders, A., & Senbet, L. W. (2000). A theory of bank regulation and management compensation. Review of Financial Studies, 13, 95-125. Kochhar, R. (1996). Explaining firm capital structure: The role of agency theory vs. transaction cost economics. Strategic Management Journal, 713-728. Knopf, J. D., Nam, J., & Thornton Jr, J. H. (2002). The volatility and price sensitivities of managerial stock option portfolios and corporate hedging. The Journal of Finance, 57(2), 801-813. 46

Lewellen, K. (2006). Financing decisions when managers are risk averse. Journal of Financial Economics, 82(3), 551-589. Liu, Y., & Mauer, D. C. (2011). Corporate cash holdings and CEO compensation incentives. Journal of Financial Economics, 102(1), 183-198. Low, A. (2009). Managerial risk-taking behavior and equity-based compensation. Journal of Financial Economics, 92(3), 470-490. Magnan, M. L., & St-Onge, S. (1997). Bank performance and executive compensation: A managerial discretion perspective. Strategic Management Journal, 573-581. Markowitz, H. (1952). Portfolio selection. The journal of finance, 7(1), 77-91. Mayers, D., & Smith Jr, C. W. (1992). Executive compensation in the life insurance industry. Journal of Business, 51-74. Mehran, H. (1995). Executive compensation structure, ownership, and firm performance. Journal of financial economics, 38(2), 163-184. Mehran, H., & Rosenberg, J. (2008). The effect of CEO stock options on bank investment choice, borrowing, and capital. Federal Reserve Bank of New York Working Paper. Parrino, R., Poteshman, A. M., & Weisbach, M. S. (2005). Measuring investment distortions when risk averse managers decide whether to undertake risky projects. Financial Management, 34(1), 21-60. Rogers, D. A. (2002). Does executive portfolio structure affect risk management? CEO risk-taking incentives and corporate derivatives usage. Journal of Banking & Finance, 26(2), 271-295. Schwert, G. W. (1989). Why does stock market volatility change over time?. The journal of finance, 44(5), 1115-1153. Smith, C. W., & Stulz, R. M. (1985). The determinants of firms' hedging policies. Journal of financial and quantitative analysis, 20(04), 391-405. Smith, C. W., & Watts, R. L. (1992). The investment opportunity set and corporate financing, dividend, and compensation policies. Journal of financial Economics, 32(3), 263-292. 47

9. Appendix: Appendix I: The daily three-month T-bill yield from the Federal Reserve Bank of St. Louis from 2007 to2015. Appendix II: A list of all banks that are included in the dataset Company name: 1 AMERIS BANCORP 2 ASSOCIATED BANC-CORP 3 BANCORPSOUTH INC 4 BANK OF AMERICA CORP 5 BANK OF HAWAII CORP 6 BANK OF NEW YORK MELLON CORP 7 BANK OF THE OZARKS INC 8 BANNER CORP 9 BB&T CORP 10 BOSTON PRIVATE FINL HOLDINGS 11 CARDINAL FINANCIAL CORP 12 CASCADE BANCORP 13 CATHAY GENERAL BANCORP 14 CENTRAL PACIFIC FINANCIAL CP 15 CHEMICAL FINANCIAL CORP 16 CITIZENS FINANCIAL GROUP INC 17 CITY HOLDING CO 18 CITY NATIONAL CORP 19 COLONIAL BANCGROUP 48