CEO Compensation, Firm Risk and the Effect of CEO Characteristics:

Transcription

1 CEO Compensation, Firm Risk and the Effect of CEO Characteristics: Evidence from the U.S. Financial Industry Master Thesis in Finance Name: T. C. Janssen Administration number: s Date: December 2, 2013 Faculty: TiSEM, Tilburg University Supervisor: Dr. F. Feriozzi Second reader: Dr. O. G. Spalt

2 Abstract This research empirically investigates the relationship between CEO compensation and firm risk during the period of in the U.S. financial industry. Furthermore, the effect of several individual characteristics of CEO s, such as age, tenure and gender, are included in the analysis. First, this study finds that there is substantial cross-sectional heterogeneity in compensation practices across different financial sub-industries. Next, residual compensation is correlated with price-based measures of firm risk, such as market beta and return volatility, and cumulative excess returns. No significant relationship between CEO compensation and book leverage is identified. Finally, this study did not find conclusive evidence that the relationship between the total amount of CEO compensation and firm risk is influenced by CEO characteristics as age, tenure and gender. These findings show that while pay practices differ among subindustries of the financial industry, residual compensation affects the level of firm risk for the entire industry. The evidence that this relationship is influenced by individual attributes of CEO s is very weak. 2

3 Table of Contents 1. Introduction Literature Review Executive Compensation Deregulation in Banking Executive Compensation and Risk in the Financial Industry CEO Characteristics, Executive Compensation and Risk-Taking Behavior Tenure and Executive Compensation Age, Tenure and Risk-Taking Behavior Gender and Risk-Taking Behavior Data and Measures Data Methodology Results Sub-Industry Compensation Differences CEO Compensation and Risk-Taking Behavior Price-Based Risk Measures Return Outcomes Book-Based Risk Measures Sub-Industry Differences in Risk-Taking The Influence of CEO Characteristics Age Tenure Gender Adding Multiple CEO Characteristics Simultaneously Sub-Industry Differences in the influence of CEO Characteristics Conclusion Limitations Recommendations References Tables and Figures Appendix I: Included Firms in the Sample

4 1. Introduction With the current financial crisis hitting hard, the debate over executive compensation is very much alive again. Public opinion is that these bonuses are an outrage and are not acceptable in the financial climate that we are facing. In January 2009, U.S. President Obama branded Wall Street bankers shameful for giving themselves nearly $20 billion in bonuses, while the economy was deteriorating and the government was spending billions to bail out a lot of financial institutions. 1 In February 2012, the European Union even reached a preliminary deal to severely restrict bank executives bonuses, forbidding bonuses to exceed a bankers fixed salary and allowing flexible pay to increase to twice fixed salary only with explicit shareholder approval. 2 The current discussion again raises the question what the function of executive compensation is in corporations, and in the financial industry in particular. By empirically testing some of the questions that are raised by the public debate and discussion in the media this research tries to find an answer to the following question: What is the influence of Chief Executive Officer compensation on risk-taking behavior in the financial industry, and how is this relationship affected by individual CEO characteristics? To establish a complete view on the problem, several other questions are used to increase the understanding of the complex relationship between CEO compensation and risk-taking behavior in the financial industry. What is the relationship between CEO compensation and risk-taking behavior for firms in the financial industry? What is the difference in CEO compensation and its relationship with risk-taking in different subcategories of the financial industry? What is the effect of individual CEO characteristics on the relationship between executive compensation and risk-taking behavior? These questions are translated into a number of hypothesis, that form the motivation for the empirical analysis. The first hypothesis expects that substantial heterogeneity exists in levels of compensation and their relationship with firm size between different sub-industries of the financial industry. Next, increasing the total compensation of CEO s is expected to increase firm risk, through policy choices as investments decisions and leverage choices. In the third hypothesis several individual CEO characteristics, such as age, tenure and gender, are expected to have an influence on the risk-taking behavior of firms and the relationship between CEO compensation and firm risk. 1 Obama Calls Wall Street Bonuses Shameful, New York Times January 29, 2009, 2 EU Reaches Deal to Curb Bank Bonuses, The Wall Street Journal February 28, 2013, EFTTopStories 4

5 Based on panel data of the U.S. financial industry on executive compensation, risk-taking, and CEO characteristics from , this research investigates how cross-sectional variation in compensation practices is related to heterogeneity in subsequent risk-taking. Since an endogeneity problem in the form of reverse causality can be present in the relationship between CEO compensation and firm risk, the design of this study is not causal. The measure used to determine the effect of compensation and shorttermism is the residual of total annual compensation of the CEO, controlling for firm size and sub-industry classifications. To control for sub-industry classifications, the financial industry is divided into three categories: primary dealers, banks/lenders, and insurers. The sample is split up into two periods, an early period that covers data from , and a late period that is defined as The first three years of each period are used to obtain an average level of CEO compensation. Then the residual compensation measure is obtained by regressing the log of average CEO compensation of on the firm s market capitalization of 1999, allowing for sub-industry heterogeneity. A similar procedure is followed with data from to get the residual compensation for the late period. Next, using data from and , various measures of firm risk are computed for the early and late period. The first measures are price-based risk measures, which contain market beta, return volatility and a price-based risk measure. Other measures are excess returns and book leverage. The baseline analysis of this research is the regression of these different measures of firm risk on residual compensation for both periods. Finally, individual CEO characteristics are included in these regressions to determine their effect on the outcomes. The following findings are established in this research. First, it is confirmed that there is cross-sectional heterogeneity in pay practices between several sub-industries of the financial industry. On average primary dealers and insures offer higher amounts of total compensation than banks, lenders and bankholding companies. In addition, compensation tends to increase less with firm size for CEO s of insurance companies, compared to their equivalents of primary dealers and banks. The residual compensation measure, which is obtained from this regression for both the early and late period, is highly correlated across these two sub-samples. Therefore, heterogeneity can be interpreted as the effect of permanent differences between the different firms. Second, this study finds some correlation between residual compensation and measures of subsequent risk-taking. Firms with a higher total amount of CEO compensation have increased risk when looking at price-based risk measures. This effect is persistent across both periods, with the exception of return volatility in the late period. In addition, higher residual compensation is related with more extreme return outcomes. In the late period, which is characterized by the housing boom in the U.S., increased CEO compensation is associated with negative cumulative excess returns. In the early period this effect if positive, though statistically insignificant. No significant relationship between residual compensation and book leverage is identified. These findings suggest that substantial heterogeneity among financial firms exists, where high levels of both CEO compensation and risk-taking are related. Excluding several subindustries from the analysis had no major impact on the results and only increased the significance of several coefficients in the late period. 5

6 Finally, individual CEO characteristics age, tenure and gender are included in the analysis to determine their impact on the relationship between compensation and firm risk. No conclusive evidence is found that these characteristics play an important role in this relationship. The results on age and tenure differ in direction and significance across the different measures of firm risk and timeframes. The effect of gender on risk is statistically significant most of the time, but varies in direction in unexplainable ways. This is likely to be caused by the very low number of female CEO s in the sample, which makes it impossible to interpret the influence of gender on the relationship between CEO compensation and risk-taking behavior. In general, this study finds substantial differences in pay practices within the U.S. financial industries. However, for all sub-industries and the financial industry as a whole, a positive relationship between the total amount of CEO compensation and risk-taking seems present, though differing in strength. The evidence that this relationship is influenced by individual characteristics of CEO s as age, tenure and gender is very weak. 6

7 2. Literature Review Executive compensation has been widely discussed in academic literature and is recognized as an important mechanism of corporate governance. The problem of optimal management compensation is related to the separation between ownership and control. The conflict of interest between shareholders of a publicly owned corporation and the corporation s Chief Executive Officer (CEO) is a classic example of a principal-agent problem (Jensen & Meckling, 1976). In addition, Jensen and Murphy (1990) argue that if shareholders are completely able to monitor the CEO s activities and the firm s investment opportunities, they could design a contract that would perfectly enforce the managerial actions required in every state of the world. However, since this is not the case, agency theory predicts that compensation policy will be designed to provide executives with incentives to select projects that increase shareholder wealth. Therefore, Jensen and Murphy (1990) consider it appropriate to take shareholder wealth as a basis for executive pay, with mechanism enhancing value increasing incentives, such as performance-based bonuses and salary revisions, stock options, and performance-based dismissal decisions. This view is supported by Smith and Watts (1992), who find that contracting theories play an important role in explaining cross-sectional variation in observed financial compensation. 2.1 Executive Compensation The available research on executive compensation was summarized by Murphy (1999) over a decade ago and more recently by Frydman and Jenter (2010). One of the first aspects that can be noticed is the enormous increase in executive compensation at the end of the 20 th century. With data from Frydman and Saks (2010) a J-shaped relationship between time and the median of total compensation is displayed in Figure 1 (Frydman and Jenter, 2010). Figure 1: Median Executive Compensation (Frydman and Jenter, 2010, p. 37) 7

8 Despite substantial pay heterogeneity in companies, five important components of executive compensation can be identified: salary, annual bonus, benefits from long-term incentive plans, restricted options grants, and restricted stock grants (Frydman and Jenter, 2010; Murphy, 1999). As shown in Figure 2, the relative importance of these components has shifted considerably over the past years. Especially during the last two decades stock options have become a very important component of executive pay. This is a development which is highlighted in Figure 2b (Frydman and Jenter, 2010). Figure 2a: The Structure of CEO Compensation (Frydman and Jenter, 2010, p. 38) Figure 2b: The Structure of CEO Compensation (Frydman and Jenter, 2010, p.39) 8

9 Murphy (1999) shows that executive compensation practices vary with company size, industry and country. By examining the existing literature on the aspects, he finds that (1) larger firms experience higher average levels of pay and lower pay-performance sensitivity; (2) both levels of pay and payperformance sensitivities are lower in regulated utilities than in industrial firms; (3) compared to other countries, both levels of pay and pay-performance sensitivities are higher in the U.S. In addition, the payperformance sensitivities are mainly driven by stock options and stock ownership, compared to other forms of compensation (Jensen and Murphy, 1990; Hall and Liebman, 1998) This is consistent with Frydman and Jenter (2010), who state that much of the effect of performance on CEO wealth goes through option and stock revaluations, rather than through changes in annual pay. The finding that larger firms have higher levels of pay is supported by analysis from for instance Rosen (1982) and Kostiuk (1990), who show that larger firms are more likely to employ better-qualified and better-paid managers. The inverse relationship between firm size and pay-performance sensitivity can be explained by the fact that risk averse and wealth constrained CEO s of large firms can feasibly own a very small fraction of the company through stock, options, and incentive compensation (Murphy, 1990). The industry differences in both level of pay and pay-performance sensitivity depend on the level of regulation present in an industry. Highly regulated industries have lower levels of pay and less pay-performance sensitivity. This finding is consistent with developments in the U.S. banking industry, as shown by Barro and Barro (1990), Crawford, Ezell and Miles (1995), and Hubbard and Palia (1995). The effects of the deregulation period in the 1980 s will be explained in more detail in the next section. Finally, Abowd and Bognanno (1995) state that U.S. firms have a different compensation structure, with more emphasis on options and less on salary, which leads to higher levels of pay and increased pay-performance sensitivity. Based on the literature described in the section above, the following relationships are predicted: H.1a: Substantial heterogeneity in CEO compensation exists between sub-industries in the financial industry. H.1b: The relationship between CEO compensation and firm s risk-taking behavior differs in strength for sub-industries in the financial industry. Despite the differences existing between firms, industries and countries, literature clearly displays a positive relationship between executive compensation and firm performance. This relationship has for instance been examined empirically by Barro and Barro (1990), Jensen and Murphy (1990), Crawford et al. (1995), Hubbard and Palia (1995), and Mehran (1995), who all find a positive relationship between pay and performance. However, Frydman and Jenter (2010) add that because compensation packages are endogenous and correlated with many unobservables, establishing a causal relationship between executive compensation and firm performance is extremely difficult. 9

10 2.2 Deregulation in Banking As mentioned previously, the period of deregulation in the 1980 s was of major influence on the compensation structure in the financial industry. For instance, Crawford et al. (1995) show a dramatic increase in the relationship between CEO pay and firm performance after 1981, which is not the case prior to deregulation. In addition they find a significant increase in the pay-performance sensitivity of both CEO salary and bonus, and CEO option holdings. Hubbard and Palia (1995) experience similar results, with both higher levels of CEO compensation and a more striking compensation-performance relationship when interstate banking is permitted. The effect of deregulation on compensation packages is also noted by Cuñat and Guadalupe (2007), who find that the variable components of pay increased with payperformance sensitivities and, at the same time, the fixed component of pay declined. They attribute this higher reliance on performance related pay to an increase in competition in the banking industry, following the period of deregulation. The increase in pay-performance sensitivity after deregulation is consistent with economic literature about adverse selection and moral hazard. Pay-performance sensitivity is likely to increase as the input of a CEO becomes more important to the success of a firm or as managerial discretion is increased (Crawford et al., 1995). As a result of the change in executive pay following the period of deregulation in the U.S. financial industry, the percentage of stock option compensation has increased. Chen, Steiner and Whyte (2006) argue that consequently this structure causes risk-taking and the stock of option-based wealth also induces risktaking. DeYoung, Peng and Yan (2010) add that deregulation and technological developments have expanded the capacities of banks for risk-taking. They find evidence that bank boards structured CEO compensation to encourage activities to exploit the new growth opportunities, which leads to contractual risk-taking incentives. Furthermore, Houston and James (1995) find that bank shareholders face greater risk-taking incentives than shareholders of other levered firms, since the incentives are structured to enhance risky behavior of executives to pursue shareholders interests. However they also find that, on average, bank CEO s receive less cash compensation, are less likely to participate in a stock option plan, hold fewer stock options and receive a smaller percentage of their total compensation in the form of options and stock, compared to their equivalents in other industries. In addition, Houston and James (1995) show a positive and significant relation between equity-based incentives and the value of a bank s charter. This finding seems to contradict the expectation in literature that compensation policies in the financial industry promote risk taking. These results are not only meaningful to shareholders of financial companies, but also have important implications for regulators monitoring the risk level of banks (Chen et al., 2006). 2.3 Executive Compensation and Risk in the Financial Industry As mentioned in the previous section, the period of deregulation in the U.S. banking industry had a significant impact on the compensation structure of executives. Frydman and Jenter (2010) argue that a sizable literature has studied the relationship between executive incentives and corporate risk taking. Results suggest that placing more emphasis on equity incentives is associated with less risk taking. On the other hand, convexity in the portfolios of executive due to options is likely to lead to more risk taking. 10

11 John, Saunders and Senbet (1995) note that the compensation packages of CEO s have an influence on the investment choices made by firms, and that these effects are more severe when moral hazard and managerial discretion are present. In addition, Coles, Daniel and Naveen (2006) empirically find a strong causal relation between executive compensation and investment policy, debt policy, and firm risk. They show that higher sensitivity of CEO wealth to stock volatility induces riskier policy choices, including relatively more investment in research and development (R&D), less investment in property, plant and equipment (PPE), a more focused strategy, and higher leverage. On the other hand, riskier policy choices generally lead to compensation structures with a higher sensitivity of CEO wealth to stock volatility and lower pay-performance sensitivity. As a consequence of the previously mentioned deregulation, bank shareholders face greater risk-taking incentives than shareholders of other levered firms (Houston and James, 1995). The effectiveness of incentives provided to bank managers determines to a large extent the risk-taking behavior in this case, which might be constructed to promote risk taking to pursue shareholders interests. John, Saunders and Senbet (2000) support this view and state that shareholders of banks choose an optimal management compensation structure that induces first-best value-maximizing investment choices by the bank s executives. However, this does not mean that compensation policies in the financial industry are designed to encourage excessive risk taking. Houston and James (1995) find little evidence that CEO s of banks are provided with incentives to engage in risky activities, which could imply that the moral hazard problem in banking is not that severe. Lewellen (2006) adds that contrary to the intuition in previous research, options can significantly decrease the preference for risk and debt of an executive, especially when these options are in-the-money. An implication of this result is that executives could favor equity more after option grants or stock price increases. The volatility effect, which reflects the divergence between the incentives of managers and shareholders, varies strongly with firm characteristics such as firm volatility or leverage, and with the specifications of compensation contracts. While the structure of executive compensation in banking might not be designed to encourage risk-taking behavior, the possibility of this effect is still present. Chen et al. (2006) argue that the structure of executive compensation in the banking industry, with a higher percentage of stock option compensation, induces risk taking. Mehran and Rosenberg (2008) also find empirical evidence that suggests that stock option grants do lead to higher risk taking by banking executives, especially through project choice. Guay (1999) attributes this risk-taking behavior to the convexity in incentive schemes. He argues that the convexity generates a positive relation between a manager s wealth and firm risk. Compensation packages with convex payoffs, such as stock options and common stock, can induce risk-averse managers to invest in valuable risk-increasing projects that they might otherwise pass and thus influence their investing behavior. This effect of increasing the convexity of the relationship between managerial compensation and stock price is much stronger for stock options than for common stock. In addition, consistent with executives making investment and financing decisions in accordance with their risk-taking incentives, firms equity risk is positively related to the convexity provided to CEO s (Guay, 1999). 11

12 Since the optimal contracts of executives depend on the preferences of shareholders, it is interesting to describe the relationship between the owners of banks and the respective risk-taking behavior of managers. Hakenes and Schnabel (2013) show that compensation structures arise endogenously as a response to agency problems within banks. In the case of risk-shifting problems, the shareholder designs a bonus scheme that induces bank executives to take excessive risk. Alternatively, bonuses can be used to provide incentives for managers to take effort. However, since the shareholders do not completely internalize the benefits of higher effort, an underinvestment problem exists. Cheng, Hong and Scheinkman (2010) argue that investors aiming at short-term gains invest in firms with more short-term incentives and higher risk-taking. This attitude will consequently be reflected in the pay structures of the executives of such firms. Generally, their findings suggest that investors with heterogeneous short-termist preferences invest in different firms and provide incentives to the firms executives to take different levels of risk Thus, CEO compensation, and the incentives it generates, affects firm risk through policy choices. Nam, Ottoo and Thornton (2003) find that managerial incentives to bear risk play an important role in project choice and leverage decisions. When the sensitivity of the stock options of a CEO to stock return volatility increases, firms tend to make decisions that increase firm risk. In particular, firms choose higher debt ratios and invest more in research and development. This is also acknowledged by DeYoung et al. (2010), who state that a higher pay-risk sensitivity encourages riskier policy choices, while a higher payperformance sensitivity encourages less risky policy choices. In addition, Murphy (1995) argues that executive wealth is explicitly related to stock-price performance through changes in stock holdings, restricted stock, and stock options. Cohen et al. (2000) argue, in line with the previous, that there are two ways to increase price volatility, and hence stock options value. The first is to take on riskier projects and the second to increase firm leverage. In addition, Guay (1999) finds a positive relationship between volatility and leverage, since more leveraged firms tend to have higher volatility. CEO s have the ability to affect the volatility of their firm s equity. Risk-averse CEO s, who are compensated in traditional ways (salary, bonus and stock) have incentive to keep the volatility low when they hold a large fraction of their human capital and financial wealth in the firm (Cohen et al., 2000). Coles et al. (2006), state that ultimately all decisions about investment composition, firm focus and leverage should have an effect on stock prices and their volatility. Adams (2012) acknowledges the difficulty of structuring the optimal contract. She argues that performance-based pay in the form of equity is necessary to align incentives of CEO and shareholders and that total compensation should increase as risk increases. However, equity incentives could lead to excessive risk taking. Regulatory restrictions on executives bonuses have the ability to mitigate the riskshifting problem. However, this comes at the cost of reducing the incentives on effort. Finally, a strong alignment of interests between shareholders and executives in the financial industry can destabilize firms, if shareholders have strong risk-taking incentives. Laeven and Levine (2009) find that banks with more powerful owners tend to take greater risk, since they have stronger incentives to take risk than nonshareholding managers and debt holders. These findings show that the same regulation can have different effects on bank risk, depending on the corporate governance structure of a bank. 12

13 Though evidence from literature is not consistent on the relationship between CEO compensation and the level of risk that is taken in the companies in the financial industry, this research empirically tests the following hypothesis: H.2: Increasing the total amount of CEO compensation leads to higher risk-taking behavior in financial firms. 2.4 CEO Characteristics, Executive Compensation and Risk-Taking Behavior While in literature the framework between executive compensation and risk-taking behavior has been firmly established, the effect of specific CEO characteristics on this relationship is less clear. However, several characteristics of CEO s have been identified in literature that are associated with executive compensation and their risk-taking behavior. The most important of these attributes are tenure, age and gender. Hence, hypothesis 2, that addresses the effect of individual CEO characteristics on the relationship between CEO Compensation and risk-taking behavior, is divided into three different hypotheses Tenure and Executive Compensation Murphy (1986) argues that CEO s are more likely to receive stock options in their tenure at a firm. This view is supported by Houston and James (1995). In addition they find that the tenure is positively related to the percentage of firm stock held by the CEO. These findings are consistent with existing theory which suggests that CEO s with a higher tenure, of whom a more extensive track record of their value added has been established, have a reduced need for incentive based compensation Age, Tenure and Risk-Taking Behavior In early work, Wallach and Kogan (1961) find that older persons are significantly more conservative in making financial decisions. This negative relationship between age and risk is also found by Vroom and Pahl (1971), and Nicholson, Soane, Fenton-O Creevy and Willman (2005). The findings are consistent when applied in the business context of CEO s by Cohen, Hall and Viceira (2000). They show that CEO age is negatively related to volatility. Therefore, the following hypothesis is established: H.3a: Older CEO s are less likely to engage in large risk-taking behavior. Next, literature suggests that the relationship between CEO tenure and risk-taking is negative. Hambrick and Finkelstein (1987) focus on entrepreneurial practices of CEO s and observe that longer CEO tenures are related to compliance with industry norms and practices, rather than upsetting the status quo. Zahra (2005) also finds that CEO tenure is negatively associated with entrepreneurial risk taking, such as innovation and new market entry. A negative relationship between CEO tenure and firm risk is also identified by Coles et al. (2006). Berger, Ofek and Yermack (1997) add that CEO s with longer tenures and higher cash compensation are more likely to be entrenched and therefore try to avoid risk. MacCrimmon and Wehrung (1990) use maturity, a combined variable of age, seniority and dependents, to investigate the relationship with risk-taking behavior. They find that maturity is negatively related to several risk measures, such as use of personal debt, willingness to take business risk and aggregated measures. Thus, 13

14 the most risk-averse managers were in general older, had longer seniority in the firm and fewer dependents. Consequently, the hypothesized relationship is: H.3b: CEO s with greater tenure are less likely to engage in large risk-taking behavior Gender and Risk-Taking Behavior Risk is often associated in literature as a male phenomenon. This is consistent with Powell and Ansic (1997), Byrnes, Miller and Schafer (1999) and Nicholson et al. (2005). In addition, Powell and Ansic (1997) find that males and females adopt different strategies in financial decision environments. However, the difference in these strategies has no significant impact on their ability to perform. This leads to the final hypothesis: H.3c: Male CEO s are more likely to engage in large risk-taking behavior. 14

15 3. Data and Measures 3.1 Data The data collection and creation of the sample is based on the method by Cheng et al. (2010) and explained in further detail below. However, this research focuses on a somewhat different timeframe to examine the effect of the last available fifteen fiscal years. Starting point to create a sample and classify financial firms is the CRSP-COMPUSTAT Merger Annual Fundamentals file, In this file firms with SIC codes 60, 61, 6211, 6282, 6712, 6331, and 6351 are included. Cheng et al. (2010) indicate that SIC codes obtained from CRSP do not always exactly match the SIC classification, particularly for bank holding companies. Therefore, firms from the more expansive three-digit SIC codes of 670, 671, 633, and 635 are checked with the EDGAR database of the U.S. Securities and Exchange Commission. No manual changes had to be made based on this check. In addition, Fannie Mae has been removed from the sample, which is in line with Cheng et al. (2010). The first group that is constructed consists of primary dealers, which is done by hand-matching a list of 21 primary dealers of the Federal Reserve Bank in New York. 3 When a primary dealer is a subsidiary of a holding company in CRSP, the bank holding company is grouped together with the primary dealer. This group includes 6020 commercial banks and 6211 security brokers, dealers and flotation companies. Data on four of these companies is not available. A dummy variable is constructed to indicate if a financial firm is a primary dealer or not. Next, SIC codes obtained from the current classification list on the OSHA website is used to classify firms into a second group of banks, lenders and bank holding companies that have no primary dealer subsidiaries. This part of the sample consists of SIC 60 depository institutions, SIC 61 nondepository credit institutions and SIC 6712 bank holding companies. Finally, a third group of financial companies is formed by insurers from SIC 6331 fire, marine and casualty insurance and SIC 6351 surety insurance. After that, all companies that did not match the criteria described above were removed from the sample. After the sample is constructed and financial firms have been classified, the CRSP monthly returns of these companies are linked to their accounting data. This data is retrieved from the CRSP-COMPUSTAT Quarterly Fundamentals file. Next, the merged database is linked with the ExecuComp Annual Compensation database. This database contains information on CEO compensation and CEO characteristics on the top five executives of over 3,000 companies from 1992 forward. Data about non-ceo s will be removed from this file, since this research focuses solely on the role of the CEO. To be included in the baseline sample of financial firms, data on a company has to be available from all three databases. A complete list of all companies that are included in the sample can be found in Appendix I. 3 Based on the list available on the website of the Federal Reserve Bank in New York on September 10, Retrieved from, 15

16 3.2 Methodology The baseline analysis to test the relationship between CEO compensation and the level of risk that is taken in the firms they manage is based on the technique used by Cheng et al. (2010), who regress several measures of risk on residual CEO compensation controlling for several firm specific characteristics for companies in the U.S. financial industry. Where Cheng et al. (2010) average compensation across the top five executives in each firm, this research will focus solely on the role of the CEO. Residual compensation is defined as the residual of total annual CEO compensation, controlling for firm size and sub-industry classifications. Cheng et al. (2010) argue that residual compensation, which is based on an aggregate of all forms of direct compensation, as a measure of short-term pay practices contains less noise than looking at specific components of the compensation packages. This measure is different from more traditional measures of executive incentives, such as insider ownership. In recent work, Fahlenbrach and Stulz (2009) find that insider ownership has little predictive power for risk-taking and that it is very common for CEO s in the financial industry to have high values of ownership. In addition Cheng et al. (2010) state that residual compensation is better capable of picking up implicit incentives that are not captured by inside ownership and consequently have more explanatory power for the risk-taking of firms. Top executives, such as CEO s, even if they have high ownership stakes, face very strong incentives related to market pressure from investors that have a short-term perspective. Hence, implicit incentives with respect to the threat of being fired also matter. Nonetheless, a higher level of total annual compensation for a CEO might pick up a firm culture for high-powered incentives, whether they are bonuses or the fear of getting fired based on shortterm performance. Therefore, residual compensation is a proper measure of both explicit incentives of the firm and implicit incentives of short-term investors (Cheng et al., 2010). To obtain the residual compensation measure, total compensation (including salary, bonus, equity and options grants, and other direct annual compensation) is regressed on two control variables. The first of the control variables is firm size, since larger firms are more likely to attract better qualified executives (Gabaix and Landier, 2008; Kostiuk; 1990; Murphy, 1999; Rosen, 1982). The second is heterogeneity in sub-industry classifications among financial firms. The financial industry is divided into three categories: (1) primary dealers, (2) banks, lenders and bank-holding companies, and (3) insurance companies. This separation is made to control for different compensation practices within industries, since Murphy (1999) finds that substantial heterogeneity exists in the relationship between compensation and size across nonfinancial industries. These three groups are a rough split among firms that engage in investment banking and trading, commercial banks and lenders, and financial insurers (Cheng et al., 2010). The sample is split up into two different periods, covering the last fifteen years of available data on executive compensation, managerial attributes and risk-taking measures. The early period, defined as 1997 up to 2005 contains the end of the dot-com era. The late period, which uses data from , includes the housing crisis and potential policy changes following this economic shock. Cheng et al. (2010) state that splitting up the sample into two different periods is admittedly ad-hoc and other macroeconomic factors are likely to have an influence on the available data. However, this set-up also has several benefits compared to panel estimation. As shown in a later section, residual pay levels in the two cross-sections are highly correlated. Consequently, it can be argued that permanent effects are captured 16

17 in the different regressions. Furthermore, this set-up clarifies that residual pay in the cross-sections is very similar and leads to a clear-cut framework to measure the effects of executive compensation on risktaking behavior. In addition, the design best captures cumulative returns over longer time horizons. Finally, working with a pooled panel set-up and clustering standard errors by firm in the robustness sections display similar results (Cheng et al., 2010). Total direct compensation TDC1 (Salary + Bonus + Value of Option Grants + Other Annual Compensation + Restricted Stock Grants + Long-term Incentive Payouts + All Other Compensation) from the ExecuComp database is used as baseline measure of executive compensation. Size is measured by Market Capitalization, which is defined in a year as common shares outstanding times price on December 31 of a year. As explained in the literature section, CEO compensation has an effect on firm risk through policy choices, especially focused on project choice and leverage decisions. Consequently, to determine the risk-taking behavior of CEO s, four measures of firm risk are used: (1) the beta of the firm s stock, (2) the firm s stock return volatility, (3) the cumulative return to the firm s stock, and (4) book leverage. This is a simplification of the risk measures that are used by Cheng et al. (2010), who include several more measures. The firm s Market Beta and Return Volatility are calculated using the CRSP Daily Return File. In this computation, the market return equals the CRSP Value-Weighted Index return (including dividends). Data on the risk-free return rate is collected from the Fama-French Portfolios and Factors database, available via Wharton Research Data Services. In order to be included in the analysis of Market Beta en Return Volatility, at least one year s worth of observations (252 trading days) is required. Market beta (Cheng et al., 2010; Lewellen, 2005) and stock return volatility (Cheng et al., 2010; Cohen et al., 2000; Coles et al., 2006; Fahlenbrach and Stulz, 2009; Guay, 1999; Houston and James, 1995; Laeven and Levine, 2009; Lewellen, 2005; Nam et al. 2003; Saunders et al., 1990 ) are price-based risk measures that have been used before in literature in relation with executive compensation. To compute each firm s Cumulative Excess Return, each firm s cumulative compounded return in a given period is calculated and subtracted from the cumulative compounded return of the market in that period. Similar to the price-based risk measures, return outcomes have been used in previous research in relationship with executive compensation (Cheng et al., 2010; Cohen et al., 2000). Book Leverage is included following Adrian and Shin (2009), who analyze the leverage characteristics of investment banks by measuring Book Leverage as the ratio of book assets to book equity. As explained before, one way for firms to increase risk is to increase leverage. Book leverage is used to measure leverage, since Welch (2004) argue that market leverage can change passively because of changes in stock price performance, and may not be an active managerial choice. However, market leverage might be able to capture CEO s incentives better through stock price volatility. Coles et al. (2006) argue that there is a trade-off between the measure of interest of the CEO (market leverage) and the less noisy measure of CEO decision making (book leverage). Since return volatility is already used as a price-based risk measure, here the focus will be on book leverage. Again, book leverage has also been used many times to find a relationship between executive compensation and firms risk (Cheng et al., 2010; Cohen et al., 2000; Coles et al., 2006; Guay, 1999; Lewellen, 2005; Rajgopal and Chevlin, 2002). 17

18 In addition, an important average price-based risk measure is computed. This measure is an equalweighted average of the standardized z-scores of market beta and return volatility. The risk measures can contain a lot of noise, so averaging them provides a cleaner measure of firm risk-taking (Cheng et al., 2010). Therefore the price-based risk score is the main dependent variable of interest. The aim of this research is to empirically test if cross-sectional variation in CEO compensation is related to heterogeneity in the financial industry and subsequent risk-taking. Thus, the study does not predict a causal relationship between executive compensation and the following risk-taking behavior by firms. Indeed, an endogeneity problem exists when examining the relationship between compensation packages and risk, since both variables are likely to have an influence on each other. Chen et al. (2006) argue that firm risk may impact the contractual design of a CEO, which is also acknowledged by Coles et al. (2006) and Low (2009). For example, a risk-averse CEO of a more risky firm would prefer a larger amount of salary-based compensation rather than equity based compensation, or would require a larger amount of total pay to compensate for the more volatile performances. While previous research about the effect of CEO compensation on risk is more evolved, the combination with CEO characteristics is less well-known. However, Houston and James (1995) did include several individual characteristics of CEO s in their research. The attributes that are included in the empirical tests are based on available literature on the subject and the research by Houston and James (1995). To get an understanding of the effect of CEO characteristics on the relationship between executive compensation and firm risk, four characteristics of CEO s are included in a more extensive regression: (1) age, (2) years of CEO within firm, (3) years of CEO as CEO of firm, and (4) gender. As mentioned earlier in the theoretical section of this thesis, these characteristics are all likely to influence the relationship between CEO compensation and risk. All variables are constructed using the available data in the ExecuComp database. Gender is a dummy variable, which is one if the CEO is male and zero otherwise. 18

19 4. Results The goal of this thesis is to relate differences in risk-taking behavior across firms in the U.S. financial industry to cross-sectional difference in their CEO compensation. To reach this, the sample consisting of data on U.S. financial firms is split up into two periods, as explained earlier. An early period, which starts at 1997 and stops at the end of 2005, and a late period, covering the years 2003 until Again, it is important to control for two aspects when testing relationships involving executive compensation. The first is firm size, since there is clear evidence that better personnel in general work for bigger firms (Gabaix and Landier, 2008; Kostiuk; 1990; Murphy, 1999; Rosen, 1982). The second is heterogeneity in sub-industry classifications among financial firms as described in the methodology section. Consequently, a residual compensation measure is used. This measure uses the residuals from a cross-sectional regression of log executive compensation on log market capitalization and sub-industry classifications. In the set-up of the residual compensation measure, it would be ideal to control for heterogeneity by allowing both slopes and intercepts across all three sub-industries. However, the limited number of primary dealers makes it almost impossible to form reliable estimates of the slope and intercept within that specific group. Ignoring this fact and running a regression that allows for slopes and intercepts to vary across all sub-industries yield a large standard error on the slope for primary dealers. This fact is also acknowledged by Cheng et al. (2010), who use similar regressions. Therefore, the log of average executive compensation in ( ) is regressed on the log of market capitalization in 1999 (2005), allowing intercepts to vary by sub-industry and where insurers are allowed to have a different slope than primary dealers and banks. Next, the risk-taking behavior of the firms following the years of the calculated average executive compensation is reviewed. This is done in the period of for the early period, and for the late period. There risk-taking measures include market beta, daily return volatility, cumulative excess returns, and average book leverage. In addition, a price-based risk measure is constructed by taking an equal-weighted average of the standardized z-scores of market beta and return volatility. These measures of risk-taking are regressed on log residual executive compensation and several characteristics of CEO s. These characteristics include age, tenure and gender, since literature argues that these are the ones most likely to influence the risk-taking relationship. The final dataset is built upon two cross-sections. Panel A has data on CEO compensation for 134 firms (5 primary dealers, 101 banks, and 28 insurers) in and their following risk-taking behavior in Panel B contains data on CEO compensation for 146 firms (6 primary dealers, 100 banks, and 40 insurers) in and their risk-taking behavior in Within the dataset, 94 companies are reported in both the early and late period. Table 1 reports summary statistics for the most important variables that are used in the analysis. The numbers are comparable to a previous study by Cheng et al. (2010), who use very similar methods. Since both executive compensation and market capitalization do not scale linearly, the log of these variables is 19

20 used in the analysis. One issue that can be noticed, is the highly skewed data for the gender dummy variable. The very limited number of female CEO s complicates empirical analysis on this subject. The sample covers a broad-cross-section of finance, including top investment banks, commercial banks and insurers, as well as smaller firms acting in the U.S. financial industry. 4.1 Sub-Industry Compensation Differences Figure 3 displays the relationship between log executive compensation and log market capitalization divided in the three different sub-categories of financial firms for both the early and late period. Panel A plots the log of market capitalization in 1999 against the average log of CEO compensation in The three different lines represent the linear overlay for the three different sub-industries: primary dealers, banks, and insurance companies. The slope of the lines suggests a clear and positive relationship between firm size and executive compensation, which was expected based on literature. On average primary dealers seem to have a higher level of pay compared to banks/lenders and insurers. Panel B, which plots log market capitalization against log executive compensation for the late period displays similar results. However, the slope of the primary dealers is somewhat different compared to the early period and as expected. This might be the consequence of the very limited numbers of firms in this category that are present in the sample. As mentioned earlier, in the actual regression that is used to compute residual compensation, the slope of primary dealers and banks, lenders and bank-holding companies are grouped to mitigate this problem. This regression that leads to the residual compensation measure, regresses the log of executive compensation on log market capitalization. A insurer specific slope and different intercepts for primary dealers, bank/lenders and bank-holding companies, and insurers are allowed in the regression to control for sub-industry classifications. Results of this regression are reported in Panel A of Table 2. Cheng et al. (2010) argue that using their methodology, heteroskedasticity is an a priori major concern, since substantial heterogeneity among primary dealers, banks, and insurers can be expected. Therefore, robust standard errors are reported in all tables which are better applicable to small samples (MacKinnon and White, 1985; Long and Ervin, 2000). The coefficient for log market capitalization in both the early and late period, and respectively, is positive and highly significant. Hence, providing empirical evidence for the expectations in theory and the intuitive feeling of Figure 3. The coefficient for the insurers specific slope is negative for both periods ( and ). However, this result is not significant in the early period and only significant at a 10% level in the late period, showing weak proof that CEO compensation in insurance companies increases less quickly with firm size compared to primary dealers and banks. It is not only the pace in which compensation grows as firms become bigger, that shows subindustry differences. The average level of CEO compensation also differs significantly between the three groups, where on average CEO s of primary dealers earn the highest pay. The results for the early and late period have a lot of similarities. Nevertheless, both the intercept and slope for insurance companies are not significant in the early period. The R-squared for the models is and for the early and late period respectively. 20