The influence of CEO compensation composition on bank risk.
Empirical evidence from Western-European banks
Master Thesis 2 August 2012
University of Groningen Faculty of Economics and Business MSc Business Administration Finance
Name: T.C. Veenis Supervisor: Dr. L. Dam
Student number: 1484656
1 The influence of CEO compensation composition on bank risk
Abstract
This thesis examines the influence of CEO compensation components on bank risk by using a panel dataset of 53 listed Western-European bank holding companies from 2005 to 2009. The focus is on market risk and default risk viewed from a capital market perspective. CEO compensation is decomposed into fixed salary, variable cash bonuses and variable equity based compensation. In order to measure risk, equity return volatility is used as a proxy for market risk while the distance to default is applied to proxy default risk. This thesis provides evidence that CEO compensation is well able in explaining bank differences in market risk, whereas it only has modest predictability in cross-bank differences in default risk. The results indicate a strong positive influence of variable equity based compensation on bank risk supporting the public statement that bonuses have promoted bank risk taking during the financial crisis. However, the supposed positive influence of variable cash bonuses on bank risk taking is not supported by the results. Therefore, it is crucial to distinguish between the influence of cash bonuses and equity based bonuses on bank risk.
Key words: Bank Risk, Moral Hazard, Compensation Structure
2
1.
Introduction
Recently, the financial crisis has highlighted the importance of risk taking incentives in financial firms. Especially compensation structures are commonly mentioned as a driver of excessive risk taking within banks during the financial crisis (Acrey et al., 2011; Bebchuck & Spamann, 2009). In the Netherlands, the political debate about possible explanations for the financial crisis tends to result in an “emotional” discussion about the bonus culture within banks. The most widespread argument would be that banks have “perverse” incentive schemes fuelling excessive risk taking by the bank executives to secure their short-term bonuses. Bank executives are accused of opportunistic short-term behaviour and for abusing the safety net of the government when things went wrong1.
In this thesis the influence of CEO compensation structures on bank risk is examined by using a panel dataset of 53 listed Western-European bank holding companies from 2005 to 2009. CEO compensation is decomposed into fixed salary, variable cash bonuses and variable equity based compensation. The focus is on market and default risk. The specified models control for bank and CEO characteristics, the presence of a block holder, macroeconomic influences and the influence of CEO holdings in the bank. In particular, this thesis hypothesizes whether variable (fixed) compensation increases (decreases) bank risk.
Agency theory provides two primary perspectives for analysing the relationship between compensation structures and bank risk (Balachandran et al., 2010; Mülbert, 2009). First, the compensation structures are used for incentive alignment between the risk-averse management and the risk-neutral shareholders (Balachandran et al., 2010). Second, there exist a fundamental problem between the willingness to take risks between shareholders and the other stakeholders of the bank. Shareholders might have an incentive to take excessive risk resulting from the leveraged structure of banks and the limited liability of shareholders (Bebchuck & Spamann, 2009). Hence, the alignment of the incentives of the CEO and shareholders can induce risks that are socially excessive, but optimal from a shareholder perspective (Bebchuck & Spamann, 2009).
Since most of the prior literature is based on U.S. financial institutions, this thesis contributes by using a sample of 53 listed Western-European bank holding companies. Furthermore, this thesis contributes to the public debate about the relationship between compensation policy and bank risk. In this context empirical evidence is provided about the relationship between different CEO compen-sation components and bank risk within Western-European countries.
This thesis provides evidence that CEO compensation is better in explaining cross-bank differ-ences in market risk than cross-bank differdiffer-ences in default risk. The results indicate a strong positive influence of variable equity based compensation on bank market risk supporting the public statement that bonuses have promoted bank risk taking during the financial crisis. This relationship is to a large
1
3 extent robust for country specific effects and for the inclusion of two-year-lagged compensation variables. Furthermore, this thesis also provides modest evidence that equity based compensation increases default risk. In general, the results support the statement that risk-neutral shareholders can effectively use variable equity based compensation to induce risk taking by the risk-averse CEO. However, the supposed positive influence of variable cash bonuses on bank risk taking is not supported by the results. Therefore, it is crucial to distinguish between the influence of cash bonuses and equity based bonuses on bank risk.
Overall, the results are predominantly in line with prior research of Balachandran et al. (2010) and suggest that a restriction of equity based compensation is an effective way to discourage bank risk taking. An explanation for the conflicting influence between cash and equity based compensation on bank risk might be that the CEO only receives a cash bonus when the bank is in the “state of solvency” (Vallascas & Hagendorff, 2012). Therefore, cash bonus schemes provide the CEO with an incentive to avoid institutional failure (Vallascas & Hagendorff, 2012). Thereby, cash bonuses are often tied to accounting performance measures, which are more depending on historical results (Murphy, 1999). These accounting performance measures can be considered to be more precise and less risky in comparison with the more sensitive and forward looking share prices (Balachandran et al., 2010; Scott, 2009; Tirole, 2006).
The results, regarding to equity based compensation, fit well in the agency literature stating that risk-neutral shareholders can effectively use compensation structures as a mechanism to induce risk taking by the risk-averse management. By exploiting the leveraged structure of banks, in combination with creditor dispersion, deposit insurance and potential state support, shareholders can use variable equity compensation to induce risk taking at the expense of other stakeholders including the tax paying society (Bebchuck & Spamann, 2009; Mülbert, 2009).
4
2. Theoretical Background
A. Moral hazard problems within banks
A.1. Moral hazard problems and compensation contracts
The moral hazard information problem originates from the agency theory and describes information problems between the shareholder (principal) and manager (agent) due to the separation of ownership and control (Jensen & Meckling, 1976). According to the agency theory, the problems of asymmetric information between principals and agents can be reduced by developing effective com-pensation contracts (Scott, 2009). The risk-averse manager must bear some amount of risk in order to align the incentives of the management with the interests of the risk-neutral shareholders (Scott, 2009). High exposure to risk provides the risk-averse manager with an incentive to underinvest in risky projects. However, a low degree of risk might lead to shirking of the management (Scott, 2009).
The manager is assumed to be more risk-averse than the shareholders due to his over-investment of human capital in the firm and his undiversified over-investment portfolio when he holds a stake in the bank (Acrey et al., 2011; Murphy, 1999). The agency theory states that the fixed compen-sation element should satisfy the reservation wage, which can be considered as the utility obtained by the manager from the next best employment opportunity and depends on labour market efficiency (Acrey et al., 2011; Scott, 2009). The variable compensation element should satisfy the incentive compatibility constraint (Acrey et al., 2011; Tirole, 2006). The value of the variable bonus must exceed the private benefit that the CEO experiences from shirking. Therefore, the purpose of the variable incentive schemes is to guarantee that the CEO “behaves” (Tirole, 2006). The variable element is normally related to a mix of difference performance measures. This mix can broadly be divided into short-term measures related to “precise” accounting information and long-term measures related to “sensitive” share price information (Scott, 2009; Banker& Datar, 1989).
5 A.2. Bank specific moral hazard problems
It is widely acknowledged in the literature that the moral hazard problem plays a more severe role in financial institutions (See for example: Balachandran et al., 2010; Mülbert, 2009; Bebchuck & Spamann, 2009; Heremans, 2007). This problem primary relates to the leveraged structure of banks, the probability of receiving state support, deposit insurance systems, weak monitoring through creditors and the opaqueness of the balance sheet of banks (Mülbert, 2009).
According to Balachandran et al. (2010), high leverage is inherent to the banking business, since the primary right to exist for banks relates to the transformation of deposits into loans. In case the incentives of the executives are aligned with the incentives of the shareholders, the leveraged structure of the banks can encourage risk taking. The value of the stock in the leveraged banks can be seen as a call option, whereby an increase in the volatility, or riskiness, of the bank’s assets increases the value of the option (Bolton et al., 2010; Bebchuck & Spamann, 2009). Higher volatility increases the upside potential for the shareholders without increasing the downside risk due to the limited liability of the shareholders (Bolton et al., 2010; Bebchuck & Spamann, 2009). Bebchuck & Spamann (2009) argue that, when the interests of the executives are aligned with the interests of the shareholders, executives will underweight the downside of the risky strategies. Therefore, more risk is imposed to the preferred shareholders, debt holders, depositors and the government, while the gains of the risky behaviour are captured by the shareholders (Bebchuck & Spamann, 2009).
The extent of the moral hazard problem in banks also depends on the probability that banks in financial distress receive government support (Dam & Koetter, 2012; Mülbert, 2009). It can be argued that an increasing probability of receiving state support leads to less monitoring by creditors and a higher risk appetite of shareholders who might benefit from the higher upside potential (Bebchuck & Spamann, 2009; Mülbert, 2009). Therefore, it is apparent that a higher explicit or implicit probability of receiving government support in case of default increases bank risk taking (Dam & Koetter, 2012)2. Furthermore, the existence of a deposit insurance system in the economy might also lead to less monitoring by the depositors. The deposit insurance system makes the small-depositors at least partly protected for occurred losses resulting from risky behaviour, which provides them with less incentive to examine the risk strategy of the bank (Mülbert, 2009). In addition, the wide dispersion of depositors implies that depositors do not have the resources nor the knowledge to monitor the banks (Bebchuck & Spamann, 2009).
The moral hazard problem between debt holders and shareholders is also strengthened by the wide dispersion of debt holders. Resulting from the free riding problem, the individual debt holders have less incentive to monitor the bank (Mülbert, 2009). This lack of monitoring might increase with the likelihood of government support in case of bankruptcy. Furthermore, shareholders might benefit
2
6 from “ex-post opportunism” at the expense of debt holders, since banks can easily change the riskiness of their portfolio of financial assets after debt financing has been received (Mülbert, 2009).
Another bank specific aspect increasing the moral hazard problem, in comparison with non-financial firms, is the opaqueness of the balance sheets (Mülbert, 2009; Heremans, 2007). Mülbert (2009) states that it is more difficult for third parties to assess the quality of outstanding loans than the more physical assets of industrial firms. The assessment of assets becomes even more difficult when banks invest in more exotic securities like asset backed securities or credit default swaps (Mülbert, 2009). Banks can quickly change their risk strategies resulting from their portfolio of financials assets or by using securitization methods, which makes monitoring of the bank more difficult (Mülbert, 2009; Resti & Sironi, 2009). This might decrease the effectiveness of compensation contracts for incentive alignment. It can be argued that it is difficult for the monitoring board to determine whether the management has met their performance targets or that it just results from a change to a riskier strategy (Mülbert, 2009).
B. Prior empirical research
Resulting from the financial crisis, several researchers started to examine the relation between compensation and bank risk (See for example: Hagendorff & Vallascas, 2011; Acrey et al., 2011; Balachandran et al., 2010 , Cheng et al., 2010). The literature focuses mainly on default or market risk. Market risk is mostly measured by equity return volatility or systematic risk (Guo et al., 2011; Cheng et al., 2010; Laeven & Levine, 2009). Researchers focusing on default risk often apply a variant of the Merton distance to default principle (Hagendorff & Vallascas, 201; Acrey et al., 2011; Balachandran et al., 2010). Besides these market based measures, researchers also use accounting measures, like the z-score, loans past due or leverage to proxy default risk (Dam & Koetter, 2012; Guo et al., 2011; Acrey et al., 2011; Laeven & Levine, 2009).
The literature also provides different research designs to examine the relationship between compensation structures and bank risk. Several researchers regress the compensation components on different risk measures (Guo et al., 2011; Acrey et al., 2011; Balachandran et al., 2010). Cheng et al. (2010) regress residual pay on the risk measures, while Hagendorff & Vallascas (2011) examine the relationship between variable compensation and default risk by considering the take-over strategy of the CEO. Others focus on the influence of vega, which measures the sensitivity of CEO wealth to risk, on bank risk (Hagendorff & Vallascas, 2011; Belkhir & Chazi, 2010).
7 or negatively correlated with bank risk. Although, Acrey et al. (2011) do report a positive influence when the level of trading assets or securitization income is used as risk measure.
Balachandran et al. (2010) examine the relationship between compensation structures and bank risk by using a sample of 177 U.S. financial firms over the period from 1995 to 2008. Balachan-dran et al. (2010) report evidence that the proportion of equity based compensation increases the probability of default. Balachandran et al. (2010) also report a negative influence of cash bonuses on default risk. They conclude that cash bonuses, which are normally tied to accounting performance measures, are less risky since they reflect historical performance. In contrast, “share prices are considered to be more forward looking and subject to changes in the market expectations about the bank’s performance” (Balachandran et al., 2010). This indicated negative relationship between cash bonuses and bank risk by Balachandran et al. (2010) is in line with the work of Vallascas & Hagendorff (2012). Vallascas & Hagendorff (2012) provide evidence for a negative influence of variable cash bonuses on default risk when they focus on the influence of cash bonuses on default risk by using a sample of 41 European banks and 76 U.S. banks over the period from 2000 to 2008. Vallascas & Hagendorff (2012) conclude that cash bonuses can be linked to financial stability and that regulators should be cautious by the regulation of variable cash bonuses.
The documented positive relationship between variable equity based compensation and bank risk by Balachandran et al. (2010) is also supported by other empirical evidence. Chen et al. (2006) indicate that stock option based compensation induces market risk taking when using a sample 68 U.S. banks over the period from 1992 to 2000. Other researchers, focusing on the influence of vega on bank risk, also provide evidence that pay risk sensitivity increases bank risk (Hagendorff & Vallascas, 2011; Belkhir & Chazi, 2010). Hagendorff & Vallascas (2011) analyse the relationship between CEO com-pensation and CEO take-over strategy by using a sample of 172 bank acquisitions over the period from 1992 to 2007. They report evidence that CEO’s, whose pay is more sensitive to risk, engage in riskier take-overs thereby reducing the bank’s distance to default. The research of Belkhir & Chazi (2010) confirms the positive relationship between vega and bank risk by using a sample of 156 bank holding companies over the period from 1993 to 2006. Though, Belkhir & Chazi (2010) indicate a positive influence of vega on market risk instead of default risk.
8 Overall, the different research designs and risk measures do not provide unambiguous evidence about the relationship between compensation and bank risk. Acrey et al., (2010) report even mixed results when using different risk measures. While most researchers hypothesize and indicate a positive influence of equity compensation on bank risk, the hypothesized direction of cash bonuses on bank is more mixed. Although most results indicate a negative influence of cash bonuses on bank risk. The variety of methods to measure risk indicate the importance of a proper specification of bank risk. The total risk inherent to banking activities can be broadly divided into: interest rate risk, market risk, default risk and operational risk (Resti & Sironi, 2009). Referring to Dam & Koetter (2012), these different types of risk may out weight each other and the influence of these elements on total risk might be dependent on the composition of the bank’s assets and liabilities. Following this reasoning banks can have, for example, a high exposure to market risk in case a bank holds a undiversified portfolio of financial instruments for trading purposes (Resti & Sironi, 2009). However, in case the total value of these instruments is low in comparison to the bank’s total assets, it only marginally influences the overall risk of the bank (Dam & Koetter, 2012). Therefore, the focus in this thesis is on overall risk. Total equity return volatility of the bank primary reflects the overall market risk suffered by the shareholders of the bank. The distance to default proxies the probability that the bank goes bankrupt and ceases to exist. An advantage of the distance to default is that it explicitly controls for the financial risk of the bank, since banks can use financial leverage to turn low risk bank assets into more risky investments for their shareholders (Crosbie & Bohn, 2003). Therefore the distance to default measure reflects the ultimate risk of the bank, which is relevant for all stakeholders.
C. Hypothesis development
9 The agency theory predicts a negative relation between fixed salary and the willingness of the CEO to engage in risky behaviour since the CEO face penalties in case of insufficient performance (Acrey et al., 2011; Houston & James, 1995). However, the CEO does not gain much by “a highly positive outcome”, which provides the CEO with an incentive to avoid the risky investments to secure the continuation of employment (Acrey et al., 2011; Houston & James, 1995). It is expected that the relative proportion of fixed salary in the CEO compensation structure has a negative influence on bank risk. Following Acrey et al. (2011) I expect a positive influence of the short-term cash bonuses on bank risk resulting from the “explicit short-term performance focus” of this compensation component. Furthermore, the balance sheet opaqueness leads to less effective monitoring by the board and enables the CEO to quickly change to a riskier strategy for meeting short-term performance targets (Mülbert, 2009). It can be argued that the CEO experiences a stronger incentive to take risk when the relative weight of this component increases. Therefore, it is expected that the relative proportion of cash bonuses in the CEO compensation structure has a positive influence on bank risk. The third hypothesis simply follows from the agency theory, which prescribes that the management must bear risk in order to align the incentives of the management with the interest of the shareholders. Equity based compensation is normally included due to the sensitivity of share prices to CEO effort and the ability to motivate the CEO over several years (Scott, 2009; Mülbert, 2009). Therefore, it is expected that the combined proportion of option and share based compensation in the CEO compensation package has a positive influence on bank risk.
10
3. Methodology
A. Specification of risk measures A.1. Measuring market risk
The applied measure of market risk is the volatility of equity returns, which is among others used by Guo et al., (2011), Cheng et al. (2010) and Laeven & Levine (2009). The volatility of equity returns is calculated as the annualized volatility of weekly equity returns over the period from 2006 to 2010. Consistent with Laeven & Levine (2009), I use the total return index in Datastream to collect the share price data. The total return index is a price series adjusted for dividend payments by assuming that all dividends are reinvested in the bank. The returns are retrieved from the share prices by taking the natural logarithm of the share price at the end of the period divided by the share price at the begin of the period (Resti & Sironi, 2009). The first measure of bank risk for bank i in year t is defined as:
= (1)
A.2. Measuring default risk
To derive the probability of default from market data, the literature distinguishes between reduced and structural models (Resti & Sironi, 2009). The structural models include both financial and business risk. Financial risk is related to financial leverage and business risk is related to the volatility of the bank’s assets. In contrast, the reduced models, which normally derive the default probability from bond spreads, ignore the traits that drive default (Resti & Sironi, 2009).
In this thesis the distance to default, based on the structural model of Moody’s KMV company, is applied for measuring default risk (Crosbie & Bohn, 2003). This default measure is derived from the contingent claim analysis of Merton (1974) and measures the number of standard deviations in asset value that the bank is away from an assumed default point (Crosbie & Bohn, 2003). The main reason for using a structural model is that Grobb et al. (2006) demonstrate that this default measure outperforms subordinated bond spreads as an indicator of bank fragility when using a sample of 103 European banks. Furthermore, the application of a structural model based on option pricing intuition suits well to the literature, which states that the leveraged structure of banks in combination with equity based incentives encourage risk taking (Bebchuck & Spamann, 2009). The limited liability of shareholders provides them with the “option of defaulting and giving the company to the creditors rather than repaying the debt” (Resti & Sironi, 2009).
11 defaults when the market value of the assets is not sufficient to repay the book debt (Crosbie & Bohn, 2003). However, based on empirical evidence, Crosbie & Bohn (2003) conclude that firms in reality do not automatically default when the asset value approaches the book value of debt due to long-term elements in the debt of firms. Crosbie & Bohn (2003) state that the relevant net worth of the firm is the value of the firm’s assets minus an assumed default point. In case of default the banks implied asset value is below the assumed default point. The applied distance to default measure (DD) can be defined as:
=
(2)
The distance to default is the market value of the bank’s total assets ( ) minus the assumed default point (DP), divided by the market value of the assets ( ) times the volatility (σ ) in asset value. The distance to default decreases with higher asset volatility, lower market value of assets or a higher book value of debt (Crosbie & Bohn, 2003). In line with Crosbie & Bohn (2003), the default point (DP) is assumed to be equal to the bank’s short term book debt plus half of the long term book debt. The value of the assets ( ) and the volatility of the assets (σ ) can be derived from the Black & Scholes (1973) option pricing formula since the book debt (X), market capitalisation ( ) and equity volatility ( ) are known for listed banks. Resti & Sironi (2009) demonstrate that the asset value ( ) and asset volatility (σ ) can be obtained by solving the following non-linear system of two equations with two unknowns. = !"1) − %&'( !"2) (3)
=
* !"1)σ (4)
The term represents the cumulative normal distribution function. The Black & Scholes (1973) option pricing methodology requires assumptions with respect to the risk free interest rate (r) and time period (T)3. Consistent with Vallascas & Hagendorff (2012), the risk free interest rate is assumed to be equal to the 12-month Euribor rate and the required time period is assumed to be one year. The values of and σ , given the values of and , can now be solved by iteration (Resti & Sironi, 2009). In this thesis the solver function in excel is applied to solve the values of the asset market value and the asset volatility that satisfies the conditions of the equity value and volatility. For estimation purposes the natural logarithm of the distance to default is taken to smooth the distribution of the calculated distance to default of the banks in the sample (Laeven & Levine, 2009).
3
In the Black and Scholes (1973) option pricing formula: "1 =+,
-
./0&/1223'
12 B. Specification of compensation variables
The different compensation components serve as the main explanatory variables in the estimation models. Each compensation component is examined in combination with the other compensation components, since focusing on one separate component can lead to misleading inferences due to the interconnection of the compensation components (Houston & James, 1995). Hence, the relative value rather than the absolute value of the compensation elements is important. To be able to examine the influence of the relative weight of each compensation element, the individual elements are in line with Balachandran et al. (2010) scaled by the total annual compensation. This result in three main compensation components based on the base salary, the cash bonus and the value of the combined equity based incentive4. The three components are scaled by the total compensation, which is equal to the sum of the three components plus other annual compensation or benefits. Compensation related to pension schemes of the CEO is not included, since the banks in the sample do not commonly report these amounts.
The influence of option and share based compensation on risk taking is, besides the reported combined amount of long-term compensation, measured by two dummy variables. The value of the dummy variable is only equal to one in case the long-term program contains a condition that restricts the CEO to sell the shares or options within the period of one or two years dependent on the number of lagged years.
C. Specification of control variables
To control for insider ownership by the CEO, the following three variables are applied (Fahlenbrach & Stulz, 2011). The first variable measures the percentage of shares held by the CEO in relation to the total outstanding shares (Acrey et al., 2011). The second variable measures the wealth of the CEO invested in shares. This value is calculated as the reported amount of shares held by the CEO times the end of year share price divided by the total annual compensation (Acrey et al., 2011). As a third alternative measure, I use the natural logarithm of the total value of shares held by the CEO. Several other control variables are also included in the estimation models. First bank size is a potential determinant of bank risk taking. Large banks are normally more diversified decreasing the overall risk of the bank (Balachandran et al., 2010). However, banks which are “too big to fail” might have an incentive to increase risk, since the government turns up for part of the costs in case of default (Acrey et al., 2011). Second, to control for the influence of the available investment set and growth options, I use the market to book value as a proxy (Houston & James, 1995). The inclusion of the market to book ratio might also be relevant since low chartered banks, indicated by low market to book equity, can have incentives to increase risk (Benston & Evan, 2006). With respect to the CEO
4
13 specific characteristics I control, consistent with Acrey et al. (2011), for the age of the CEO, gender and for a change of the CEO during the period. At last, I control for the existence of a block holder. Laeven and Levine (2009) report evidence that the ownership structure is important for the degree of risk taking by banks. They find evidence that banks with a large diversified and powerful block holder enhance more risk taking. Large and powerful shareholders are better able to induce risk taking by the more risk-averse management. Therefore, I control for the presence of a block holder holding at least 5% of the total shares in the first model and at least 10% of the total shares in the second model. The specification of all variables is presented in Table I (p.15).
D. Construction of estimation models
The data in the collected dataset can be classified as panel data, since the dataset contains both time and cross-section elements (Brooks, 2008). Hence, for every bank is information collected for more than one year. Widely used panel techniques are the pooled OLS model, the Fixed Effects (FE) model and the Random Effects (RE) model (Brooks, 2008). The output of the Redundant Fixed Effects Likelihood Ratio (RFELR) test, reported in Table AI in the appendix (p.40), indicates the existence of both time and cross-section fixed effects in the sample (Brooks, 2008). Therefore, the pooled OLS assumption of an equal intercept for every bank and year in the sample might be too restrictive (Brooks, 2008). Furthermore, the results of the Hausman test, also reported in table AI, indicate that the latent variable significantly correlates with one or more independent variables in all models. Therefore, estimation with the RE-method is not suitable, given the characteristics of the dataset. The application of the RE-models leads to biased and inconsistent estimation of the regression coefficients (Brooks, 2008).
The reported test statistics of the RFELR and Hausman test suggest that the use of a FE-model, which controls for entity and time specific heterogeneity, is the most appropriate. However, the main problem of the FE-method is that it only works with sufficient variation in the variables over time within the entities. The change of the compensation variables over time within a specific bank is used to explain the change of risk measures over time in this bank (Brooks, 2008). A brief examination of the one-year-lagged correlation of each of the compensation variables indicates a positive correlation for all compensation variables. The positive one-year-lagged correlation ranges from 0.405 to 0.645 with a confidence level of 99%. Therefore, it can be assumed that the compensation structures do not differ much over time within the banks. The relative time-invariance of the different compensation components leads to the conclusion that the FE-estimator is unlikely to be a good predictor of the relationship between compensation and bank risk (Brooks, 2008).
14 (Wooldridge, 2010). Therefore, the results are estimated by using “White period standard errors” (bank clustering). These standard deviations are heteroskedasticity robust and control for within-bank autocorrelation in the residuals (Petersen, 2009). Moreover, time dummies are included to capture the time specific effects. Finally, the models are also estimated by including country dummies to control for fixed country effects. It might be the case that the relation between compensation and bank risk reflects cross-country differences rather than cross-bank variation in CEO compensation structures (Laeven & Levine, 2009).
Another potential problem when examining the influence of CEO compensation on bank risk could be the presence of an endogenous relation between risk and compensation. It might be the case that CEO compensation is also reversely related to the degree of risk. For example, Oxelheim and Randoy (2006) state, based on agency theory, that the risk-averse CEO demands a risk premium for increased risk of dismissal. To mitigate the potential endogeneity problems in the relationship between bank risk and compensation I use, consistent with Acrey et al. (2011), lagged compensation variables. One-year-lagged compensation variables will be used for the examination of the influence of the CEO compensation components on bank risk. In the subsection “further robustness checks” I will also control whether the results are confirmed with two-year-lagged compensation variables. The two estimation models with one-year-lagged compensation variables can be defined as follows:
= 6789% : ;+ 67 = + !> ) + ?+ @ t=1,2,...,T and i=1,2,…,N (1) = 6789% : ;+ 67 = + !> ) + ?+ @ t=1,2,...,T and i=1,2,…,N (2)
The dependent variables and represent the observed value of the relevant risk measure for bank i in year t. This term is defined as the volatility of share returns in model 1 and as the bank’s distance to default in model 2. The term 789% : ; represents the one-year-lagged explanatory compensation variables of bank i in year t-1. The term 7 = reflects the value of the control variables for bank i at time t. The risk measures and the terms 789% : ; and 7 = do all vary over entity and time. The term > captures the country fixed effects which influences the risk measures in the cross-country dimension, but do not vary in the time dimension. The term ? captures the unobserved time-specific effects, which influences the risk measures in the time dimension, but do not vary cross-sectionally. At last, the term @ is the idiosyncratic error term which measures the residual disturbance and varies per entity i and per bank year t (Brooks, 2008).
as-15 sumption of the error terms is violated. However, the fact that the residuals are not normally distributed has only a minor effect on the estimated coefficients if all others assumptions are fulfilled and the sample large enough is (Brooks, 2008). Based on the sample size of 177 observations and the application of White period standard deviations, which are heteroskedasticity robust and assume within-bank autocorrelation, I do not expect severe issues with regards to inference resulting from the non-normal distribution.
Table I
Specification of variables
Variable Description Group Expected
Sign
Model 1
1 Equity Volatility Annualized equity volatility Dependent/ risk
2 SalaryF; Salary scaled by total compensation Compensation -
3 Cash bonusF; Cash bonus scaled by total compensation Compensation +
4 Equity payF; Equity bonus scaled by total compensation Compensation +
5 Option planF; Existence option plan effective year + lag Compensation +
6 Share planF; Existence share plan effective year + lag Compensation +
7 % shares CEO Percentage of shares hold by CEO CEO stake -
8 Value CEO stake Natural logarithm of value shares hold by CEO
CEO stake -
9 Bank Size Natural logarithm of book assets Control: Bank
specific
+
10 Investment Set Market to book ratio Control: Bank
specific
+/-
11 Capital Capital ratio Control: Bank
specific
+/- 12 Block5% Block holder owning at least 5% of the
shares
Control: Bank specific
+
13 GDP/ capita Natural logarithm of the GDP per capita Control: Macro +/-
14 CEO age Natural logarithm of the age of the CEO Control: CEO specific +/- 15 CEO change Dummy equal to 1 in year of CEO change Control: CEO specific +/- 16 CEO gender Dummy variable to control for gender Control: CEO specific +/-
Model 2
1 Distance to default Number of std. dev. to default point Dependent/ risk
2 SalaryF; Salary scaled by total compensation Compensation -
3 Cash bonusF; Cash bonus scaled by total compensation Compensation +
4 Equity payF; Equity bonus scaled by total compensation Compensation +
5 Option planF; Existence option plan effective year + lag Compensation +
6 Share planF; Existence share plan effective year + lag Compensation +
7 REL Stake CEO Value Shares CEO divided by compensation CEO stake -
8 Bank Size Natural logarithm of book assets Control: Bank
specific
+
9 Investment Set Market to book ratio Control: Bank
specific
+/-
10 Capital Capital ratio Control: Bank
specific
+/- 11 Block10% Block holder owning at least 10% of the
shares
Control: Bank specific
+
12 GDP/capita Natural logarithm of the GDP per capita Control: Macro +/-
16
4. Data
A. Sample selection
The collected panel dataset contains information about 53 publicly listed banks from 12 different Western-European countries. The compensation information is hand collected from the published annual reports over the period from 2005 to 2008. The CEO specific information is collected from the annual reports over the period from 2006 to 2010. The financial information of the banks is obtained from the databases “Bankscope” and “Datastream”. The GDP data comes from the World Bank. Since I use lagged variables the risk and control variables are collected over the period from 2006 to 2010.
The procedure for the inclusion of the banks in the sample is as follows. First, I have selected all listed banks in Bankscope in the year 2007 from the following Western-European and Scandi-navian countries: Austria, Belgium, Denmark, France, Germany, Great-Britain, Ireland, Liechtenstein, Netherlands, Norway, Sweden and Switzerland. The Southern-European countries Greece, Italy, Por-tugal and Spain are excluded from the sample, since the data collection learned that most of these banks do not report the required CEO compensation information. Including only the minority of banks reporting the CEO remuneration may thus lead to a reporting bias. For a similar reason I have also excluded Finland from the country selection5. When discussing the results, I will also examine to what extent the results are influenced by constitutional differences or country specific outliers (Laeven & Levine, 2009).
After the country selection, I have selected all publicly listed banks in Bankscope leading to a list of 244 banks. From this list I have selected all banks with total book assets above or equal to one billion euro in the year 2008. Thereafter, I removed the subsidiaries to remain a list of bank holding companies. The focus lies on the holding level of the banks since the incentives of executives might influence the risk taking behaviour in the subsidiaries (Bebchuck & Spamann, 2009). In line with Beltratti & Stulz (2010) I impose further conditions to the loan to assets ratio and the deposits to assets ratio to exclude the non-depository banks. All banks included in the sample have a loan to assets and a deposits to assets ratio above 5%. From this selection I have removed the remaining investment and brokerage firms that were not excluded by the previous restrictions and banks that do not report in English or German. At last I have excluded the banks that did not report sufficient compensation infor-mation. For banks reporting in foreign currency is the end of the year exchange rate applied to convert the foreign currency to euros. Moreover, banks with a broken book year are assigned to the calendar year in which most of the months passed by. The above procedure, which is summarized in Table II (p.17), results in a sample of 53 listed bank holding companies from 12 different countries. The 53 selected banks should in principle have led to a total sample of 212 years. However, mainly as a result
5
17 of insufficient compensation information in the years 2005 and 2006 and the non-listening of banks in some years, the total dataset consists of 177 years.
Table II
Summary of sample selection
Procedure Criterion Remaining sample
1. Begin selection Publicly listed banks in selected countries 244
2. Size Total book assets ≥ EUR 1.000.000.000 151
3. Bank holding company Removal of subsidiaries 120
4. Depository banks Deposit/ loan requirement of 5% of total assets 93 5. Investment fund/ broker Classification investment fund/ broker/ other 79 6. Compensation Information Reporting sufficient compensation information 53
7. Sample 53
B. Descriptive statistics
B.1. Dependent risk variables
The volatility of equity returns strongly varies over the period from 2006 to 2009, which is indicated by a standard deviation of 33.5% and a range from 5.8% to 202.9% in Table III (p.18). The average volatility over the period is 48.2%, but varies from an annual average volatility of 21.4% in the year 2006 to 67.5% in the year 2009. After 2007 the average volatility of the equity returns increases substantially from 26.6% to 64.1% in 2008 and increase further to 67.5% in 2009, reflecting the financial crisis. In 2010 the equity return volatility becomes more stable again. Table III also indicates that the strong increase in equity return volatility, during the financial crisis, is combined with a strong decrease in the average market capitalization of the banks. The average market capi-talization drops from EUR 24.4 billion at the end of 2006 to EUR 7.3 billion in the year 2008. The strong differences in volatility over different years support the results of the RFELR test indicating the existence of time specific effects in the sample.
18 Table III
Descriptive statistics of dependent risk variables
2010 2009 2008 2007 2006 Period 2006-2009 Equity volatility Mean (%) 34.7 67.5 64.1 26.6 21.4 48.2 Median (%) 31.9 63.5 65.2 27.7 21.0 34.8 Standard deviation (%) 17.8 36.6 31.6 10.0 4.9 33.5 Maximum (%) 114.1 178.7 202.9 77.9 38.0 202.9 Minimum (%) 9.5 11.9 5.8 12.6 10.2 5.8
Market cap (EUR 1.000.000) 14,076 13,023 7,314 19,222 24,410 15,609
Distance to default Mean 23.6 13.7 19.97 22.2 18.8 18.4 Median 17.5 6.9 14.3 17.9 15.4 14.0 Standard deviation 22.5 26.6 21.4 24.5 18.5 22.4 Maximum 94.2 165.5 122.1 155.5 125.4 165.5 Minimum 3.9 1.4 0.5 1.5 3.0 0.5 Banks 53 52 50 41 34 53 Observations 53 52 50 41 34 177
Figure 1 in the appendix (p. 44) provides an insight in the different components of the distance to default measure. The figures 1B and 1C demonstrate that both the equity value and volatility vary substantially stronger than the implied asset value and volatility. Furthermore, figure 1D shows that the market value of the equity ranges between 4.9% and 10.9% of the implied market asset value. This is consistent with the general statement that the inherent nature of the bank’s assets is stable, but that leverage is used to amplify the equity risk of banks (Crosbie & Bohn, 2003). The descriptive charac-teristics support the statement that leverage is used to turn low risk bank assets into more risky investments for shareholders (Crosbie & Bohn, 2003).
B.2. Independent compensation variables
19 Table IV (p.20) shows that the typical annual CEO compensation structure consists of a rela-tive base salary of 49.4%, a relarela-tive cash bonus of 28.7% and a relarela-tive equity based bonus of 19.2%6. The data points out that the CEO always receives a base salary with a minimum of 5.9% of the annual compensation. In at least one year the compensation of the CEO consists entirely of fixed salary. The most extreme compensation structures, regarding to variable compensation, consist of 90.8% of cash bonuses and 87.5% of equity bonuses. Table AII in the appendix (p. 43) provides the characteristics of CEO compensation structures across different countries and suggest that CEO compensation varies per country. The average of the total variable element in the compensation structures ranges from 16.6% by Norwegian banks to 72.6% by banks from Great-Britain. Therefore, the data in Table AII suggest that the variable element of CEO compensation is higher in the Anglo-Saxon countries.
B.3. Control variables
The descriptive statistics of all control variables are also presented in Table IV. The value of the variable “Bank Size” ranges from 1.1 billion to 2586.7 billion with a standard deviation of 612.7 billion and does not indicate irregularities. The average market to book value of the banks is 1.24, but strongly varies with a standard deviation of 0.91. In the maximum the market equity is 6.45 times the book equity, while in the minimum the market equity is only 4% of the book equity. These characteristics suggest that the growth options of the banks strongly differ as judged by the capital markets. The data in Table IV shows that in 60% of the years a shareholder with at least 10% of the cash flows rights is present7. In case the threshold lowered to 5%, the presence of a block holder rises to 82%. The capital ratio is defined as the tier 1 plus tier 2 capital divided by the risk weighted assets of the bank and is on average 13.8%. Table IV also shows that the variable LN GDP per capital varies from 10.0 to 11.7 indicating only a small dispersion of the data. With respect to the CEO specific variables the data indicates that the average CEO is a male of 53 years old (LN 53 ≈ 4). In 10% of the years a CEO is changed.
Table IV shows that the average CEO in the sample has 140.1% of his annual compensation invested in the bank and holds on average 0.2% of the total outstanding shares. The data shows a highly dispersed distribution of the equity stake of the CEO’s ranging from 0 to 34 times the CEO’s annual compensation. The large difference between the average and median might also indicate that the average is influenced by an outlier. To control for the influence of outliers, I also run the model after “winsorizing” the distribution of the variable “equity stake” to the average plus three standard deviations (Balachandran et al., 2010; Kennedy et al., 1992). This procedure, however, does not change the direction nor substantially change the significance of the estimated coefficient.
6
Note that this amount is not exactly equal to 100%, since other annual compensation is also included when the compensation components are scaled by total compensation.
7
20 Table IV
Descriptive statistics independent variables
This table gives an overview of the characteristics of the compensation and control variables. The provided descriptive statistics are based on the sample with one-year-lagged compensation variables.
Period 2006-2009 Mean Median Maximum Minimum Standard dev. N
Compensation SalaryF; (%) 49.4 45.5 100 5.8 28.70 177 Cash bonusF; (%) 28.7 27.8 0.91 0.0 22.97 177 Equity payF; (%) 19.2 11.1 86.9 0.0 21.95 177 Option planF; 0.52 1.0 1.0 0.0 0.501 177 Share planF; 0.54 1.0 1.0 0.0 0.500 177 Ownership Shares (%) 0.2 0.001 6.0 0.0 0.8 177 LN CEO Stake 11.3 13.0 19.3 0.0 5.4 177
Rel. Stake CEO 1.40 0.4 33.7 0.0 4.1 177
Bank-specific
Bank Size (bn EUR) 455.6 184.9 2586.7 1.1 612.7 177
Investment Set 1.24 1.13 5.70 0.04 0.91 177 Capital 13.3 11.9 45.5 1.1 5.4 177 Block5% 0.82 1.0 1.0 0.0 0.39 177 Block 10% 0.59 1.000 1.0 0.0 0.49 177 Macro GDP/ capita 10.5 10.4 11.3 10.0 0.27 177 CEO-specific CEO age 53 53 70 35 6.30 177 CEO change 0.17 0.0 1.0 0.0 0.38 177 Gender 0.977 1.0 1.0 0.0 0.15 177 C. Multicollinearity
21 both cash bonus (0.10) and equity based bonus (0.30 in the compensation structures. These results suggest that larger banks make more use of variable compensation when rewarding the CEO. Based on the presented correlation matrices and the described actions, I do not expect that the regression results are substantially influenced by the presence of multicollinearity problems.
22
5. Results
A. Explanatory power of the models
To examine the added predictability in bank risk by the CEO compensation and ownership variables, I firstly run the regression models with only the control variables. The results of this OLS analysis are presented in Table AV in the appendix (p.43). The adjusted R square of the one-year-lagged framework without compensation and ownership variables is 0.508 in model 1 and 0.331 in model 2. The inclusion of the compensation and ownership variables increases the adjusted R square of model 1 to 0.568 and to 0.391 in model 2. Therefore, I conclude that the added compensation and ownership variables increase the power of the model in explaining the variation in bank risk. Moreover, the F-statistic of all models is significant at the 99% confidence level leading to the rejection of the null hypothesis, which states that all coefficients are equal to zero (Brooks, 2008). The combination of the increasing adjusted R squares and the significant F-statistics implies that the models are suitable for testing the hypotheses.
Before examining the influence of the compensation components, I will briefly discuss the influence of the control variables, which are significant at the 5% significance level. First, the results in Table AV in the appendix (p.43) indicate a negative influence of a bank’s growth options on the degree of market risk. Furthermore, the results indicate that a change of CEO has a positive effect on the degree on market risk. The results also provide some evidence for the existence of a negative influ-ence of a bank’s capital ratio on the distance to default. This result implies that banks with higher core capital are more willing to take risks. However, this relationship might be interconnected with the bank’s ownership structure as reported by Laeven & Levine (2009).
23 B. Baseline results
The main results indicating the influence of one-year-lagged compensation variables on equity volatility and the distance to default are respectively reported in Table VI and VII (p.24, 25). The results which are controlled for country fixed effects are respectively reported in Table VIII and IX (p. 30, 31). Further robustness checks are reported in subsection C. Due to the supposed multicollinearity between the individual compensation components, the focus is on the models whereby the influence of the compensations variables is separately estimated.
The results in Table VI indicate a significant influence of the variables “Cash bonus”, “Equity pay”, “Option plan” and “Share plan” on equity volatility. The variable “Equity pay” has the strongest influence on equity volatility in comparison with the other relative compensation components. The average positive influence of this component is 6.39%8. The average value of the relative equity pay explains 13.3% in the average value of equity volatility. The variable “cash bonus” has the weakest negative influence of the relative components. The average weight of this component decreases equity volatility with 5.37%. The average weight of the relative cash bonus explains 11.1% in the average equity volatility value. The dummy variable “Share plan” has a weaker influence on equity volatility than the dummy variable “option plan”. The application of a share plan increases equity volatility with 9.7% and explains, on average, 10.9% in equity return volatility. The application of an option plan increases equity volatility with 13.2% and explains, on average, 14.2% in equity return volatility.
The influence of the compensation variables on the distance to default measure is overall consistent with the influence of the compensation variables on equity volatility. Note that for the distance to default, higher values indicate lower risk. The main difference is, however, that the influence of the variable “salary” is now significant whereas the influence of the variable “cash bonus” is not. The component “equity pay” has a weaker influence than the fixed salary component. The average weight of this component decreases the distance to default with 0.19 standard deviations and explains only 6.5% in the average distance to default value9. The average relative weight of the fixed salary increases the distance to default with 0.31 standard deviations and explains 10.7% in the average distance to default value. The application of an option or share plan decreases the LN distance to default with respectively 0.46 and 0.34 standard deviations, which is respectively 8.2% and 6.4% of the average LN distance to default value. Overall, the coefficients have besides statistical relevance also economic relevance in explaining both market and default risk. However, the compensation com-ponents have, on average, a substantial stronger influence on equity volatility than on the distance to default of banks. Therefore, CEO compensation structures are supposed to be a more effective mechanism to determine the degree of market risk than default risk.
8
Calculation: 0.192 (average variable “Equity pay”) × 0.333 (coefficient) ≈ 6.39%. The average value of relative “Equity pay” explains 6.39%/48.2% ≈ 13.3% in the average value of equity return volatility.
9
24 Table VI
Market Risk and CEO Compensation
This table presents the one-year-lagged OLS results of model 1 including the compensation variables, time fixed effects and control variables. The annualized equity volatility is used as dependent variable. The reported coefficients are estimated by using white period standard deviations, which allow for heteroskedasticity and autocorrelation between the residuals of annual observations pertaining to the same entity (bank clustering). These standard errors are reported within the parentheses. The compensation components are estimated separately and in combination with the other compensation components.
Dependent risk measure: Annualized equity volatility
2006-2009 All Salary Cash bonus Equity pay Option plan Share plan
SalaryF; -1.247* -0.092 (0.744) (0.114) Cash bonusF; -1.336* -0.187** (0.733) (0.091) Equity payF; -0.993 0.333** (0.712) (0.151) Option planF; 0.080 0.132*** (0.051) (0.051) Share planF; 0.085* 0.097** (0.048) (0.047) % Shares CEOF; 2.643 1.392 2.432* 0.254 1.397 3.048* (1.784) (1.938) (1.345) (1.729) (2.768) (1.654) Stake CEOF; -0.013*** -0.003 -0.003 -0.007 -0.007 -0.006 (0.005) (0.004) (0.005) (0.004) (0.004) (0.004) Size 0.021 0.031*** 0.034*** 0.024** 0.028** 0.033*** (0.013) (0.012) (0.011) (0.012) (0.014) 0.010 Growth -0.084** -0.067* -0.074** -0.076** -0.074** -0.071** (0.037) (0.035) (0.037) (0.036) (0.028) (0.035) Capital Ratio -0.003 -0.001 0.001 -0.001 -0.001 -0.001 (0.004) (0.005) (0.004) (0.004) (0.004) (0.004) Block 5% 0.085* 0.017 0.012 0.044 0.027 0.042 (0.050) (0.055) (0.060) (0.058) (0.049) (0.051) GDP per capita -0.004 -0.060 -0.132 -0.015 -0.003 -0.055 (0.098) (0.087) (0.080) (0.081) (0.098) (0.075) CEO age -0.097 -0.172 -0.173 -0.099 -0.162 -0.161 (0.165) (0.175) (0.177) (0.177) (0.176) (0.169) CEO change 0.184*** 0.180*** 0.188*** 0.187*** 0.169*** 0.182*** (0.059) (0.065) (0.064) (0.064) (0.049) (0.063) Gender (1=Female) 0.082 -0.061 -0.042 -0.059 0.005 -0.012 (0.063) (0.047) (0.052) (0.046) (0.132) (0.052) Constant 1.403 0.979 1.653 0.353 0.288 0.699 (1.350) (1.052) (1.051) (1.053) (1.494) (0.969)
Time Dummies Yes yes yes yes yes yes
Banks 53 53 53 53 53 53
Observations 177 177 177 177 177 177
Adjusted RZ 0.568 0.505 0.514 0.532 0.527 0.516
F-statistic 13.9*** 13.8*** 14.3*** 15.3*** 15.0*** 14.4***
25 Table VII
Default Risk and CEO Compensation
This table presents the one-year-lagged OLS results of model 2 including the compensation variables, time fixed effects and control variables. The LN distance to default is used as dependent variable. The reported coefficients are estimated by using white period standard deviations, which allow for heteroskedasticity and autocorrelation between the residuals of annual observations pertaining to the same entity (bank clustering). These standard errors are reported within the parentheses. The compensation components are estimated separately and in combination with the other compensation components.
Dependent risk measure: Distance to Default
2006-2009 ALL Salary Cash bonus Equity pay Option plan Share plan
SalaryF; -0.351 0.618* (1.700) (0.354) Cash bonusF; -0.583 -0.062 (1.666) (0.409) Equity payF; -0.964 -0.981** (1.598) (0.432) Option planF; -0.335 -0.460** (0.236) (0.207) Share planF; -0.163 -0.344* (0.187) (0.178) Rel Stake CEOF; -0.025 -0.034 -0.042* -0.028 -0.030 -0.043* (0.021) (0.022) (0.024) (0.021) (0.022) (0.024) Size 0.149** 0.119* 0.098* 0.133** 0.132** 0.113* (0.059) (0.062) (0.058) (0.061) (0.058) (0.059) Growth options -0.190 -0.242 -0.251 -0.205 -0.211 -0.221 (0.166) (0.175) (0.163) (0.171) (0.167) (0.170) Capital Ratio -0.021 -0.028 -0.036** -0.029* -0.029* -0.027 (0.017) (0.017) (0.017) (0.017) (0.017) (0.017) Block 10% -0.332* -0.239 -0.163 -0.297 -0.235 -0.239 (0.182) (0.196) (0.186) (0.188) (0.184) (0.178) GDP per capita 0.384 0.590 0.798** 0.606 0.540 0.721** (0.411) (0.426) (0.382) (0.374) (0.382) (0.363) CEO age 0.653 0.858 0.980* 0.670 0.845 0.853 (0.557) (0.598) (0.552) (0.577) (0.555) (0.546) CEO change -0.316* -0.297 -0.284 -0.338* -0.278 -0.317* (0.180) (0.197) (0.180) (0.188) (0.178) (0.180) Gender -0.217 0.042 0.054 0.0001 -0.205 -0.135 (1=Female) (0.211) (0.186) (0.175) (0.182) (0.205) (0.200) Constant -5.809 -9.174 -10.917* -8.530 -8.212 -9.785 (6.778) (6.733) 6.043 (6.060) (5.951) (6.199)
Time Dummies yes yes yes yes yes yes
Banks 53 53 53 53 53 53
Observations 177 177 177 177 177 177
Adjusted RZ 0.391 0.368 0.350 0.379 0.386 0.373
F-statistic 7.6*** 8.9*** 8.3*** 9.3*** 9.5*** 9.1***
26 As mentioned before, the indicated relationship between the compensation components and bank risk in Table VI and VII (p.24, 25) might reflect cross-country differences instead of cross-bank differences in CEO compensation structures (Laeven & Levine, 2009). Following the reasoning of Laeven & Levine (2009), it might be the case that good countries adopt more effective laws, corporate governance codes or regulation, which induces the banks to behave more “prudently” and to apply more “conservative” compensation packages at the same time. Therefore, the results provided in Table VIII and IX (p.30, 31) are controlled for country specific influences through the inclusion of country fixed effects.
Table AVI in the appendix (p.45) provides an overview of the country fixed effects indicating the existence of country specific influences in the sample10. The results indicate that Austrian and Irish banks have both higher market and higher default risk in comparison to the reference group of German banks. Furthermore, banks from Denmark, France and Liechtenstein display only lower market risk, while the British banks display only higher default risk. Banks from Switzerland have higher default risk, while the market risk of these banks is lower in comparison with the German banks. These results justify the inclusion of country fixed effects. The explanation of the country differences is, however, beyond the scope of this thesis.
The robustness of the relationship between bank risk and CEO compensation structures to country specific effects is to a large extent ambiguous. The controlled results for fixed country effects, presented in Table VIII (p.30), confirm all the earlier indicated results concerning the relationship between CEO compensation and equity volatility. Therefore, I conclude that the indicated relationship between the different compensation components and market risk is not caused by constitutional differences. In contrast, the results concerning the relationship between CEO compensation and the distance to default, presented in Table IX (p.31), are not robust to country specific influences. None of the significant results hold when the country dummies are added. Therefore, the relationship between the compensation components and default risk is not robust to constitutional differences.
B.1. Fixed salary and bank risk
The results provide only modest evidence for the hypothesis of a negative influence of relative fixed salary on bank risk, although all significant results are in the predicted direction. First, the pooled cross-section framework indicates a positive influence of fixed salary on the distance default with 90% confidence. This result indicates that a higher relative base salary in a bank’s compensation structure can be associated with lower default risk. However, the relative fixed salary only modestly affects the distance to default. Consistent with all results, related to the distance to default, this relationship is not significant when the country fixed effects are included.
27 Remarkably, as presented in the robustness check section, the strongest evidence for a positive influence of fixed salary on bank risk can be found in the two-year-lagged framework. Again the relative amount of fixed salary increases the distance to default in the pooled cross-section model. However, the results estimated with two-year-lagged compensation variables also indicate a significant negative influence on market risk, which is also robust for constitutional differences. Therefore, these results support the hypothesis stating that the relative amount of fixed salary decreases bank risk. From a contracting perspective, these results can be explained by the statement that the risk-averse CEO is less willing to take risks to continue employment. The CEO does not gain much by a highly risky positive outcome, since his compensation is largely fixed (Scott, 2009). Moreover, the CEO might suffer penalties in case of poor performance, which is more likely to occur by more risky behaviour (Acrey et al., 2011; Houston and James, 1995).
B.2. Cash bonuses and bank risk
The provided evidence is inconsistent with the hypothesis that variable cash compensation increases bank risk. The results estimated with one-year-lagged compensation variables indicate a negative influence of variable cash compensation on equity volatility at the 5% significance level. As presented in Table VIII and IX (p.30, 31), this relationship is robust for constitutional differences. Therefore, the relative weigh of cash bonuses in compensation structures decreases market risk. Hence, the hypothesis is rejected. Furthermore, the results in Table VII do not provide evidence for a significant influence of relative cash bonuses on the distance to default with 90% confidence.
28 B.2. Equity based compensation
The results provide strong evidence for a positive influence of equity based compensation on bank risk. All significant results are in the predicted direction and in almost all cases is a positive significant influence reported. First, the pooled cross-section results indicate that the value of the annual equity based compensation has a significant positive influence on both market and default risk at the 5% significance level. Hence, this type of compensation increases equity volatility, while it decreases the distance to default. Second, the results indicate a significant influence of the existence of an option plan on the bank risk measures in the predicted direction. The existence of an option plan leads, on average, to an increase in equity volatility of 13.2% at the 1% significance level and to an average decrease of 0.46 standard deviations in the LN distance to default measure at the 5% significance level. Therefore, the results support the hypothesis that the application of a CEO option plan increases bank risk. Third, the results show that the influence of an applied CEO share plan is also in line with the results of the relative equity based component and the existence of an option plan. As expected, the reported effect of a share plan on bank risk is less strong than the effect of an option plan. Moreover, as presented in Table VIII and Table IX, the relationship between equity based compensation is also robust to constitutional differences. Similar to the other compensation variables, the influence of equity based compensation on default risk does not hold when the country fixed are included. Therefore, this relationship is not robust to constitutional differences.
The combined results provide evidence for a strong positive influence of equity based compensation on bank risk. It can be concluded that long term equity based compensation can be an effective mechanism for shareholders to induce risk taking by the management. The introduction of an option plan will provide the CEO with more incentive to take risk in comparison with a share plan. However, the results also indicate that equity based compensation is a more effective mechanism to influence the degree of market risk than the degree of default risk. As mentioned before, the coefficients have a stronger influence on market risk than on default risk and only the relationship between equity compensation and market risk is robust for constitutional differences.
29 The strong positive relationship between variable equity based compensation and bank risk can be interpreted, based on agency theory, that banks paying higher equity compensation have better incentive alignment between shareholders and managers. Hence, variable equity based compensation can be considered as an effective mechanism for the risk-neutral shareholders to induce risk taking by the risk-averse management (Scott, 2009). This relationship fits also in the theoretical implications of the strengthened moral hazard problem between shareholders and other stakeholders in banks. The inherent leveraged structure of banks provide the shareholders with incentives to take excessive risk in comparison to other stakeholder groups, when shareholders equity is considered as a call option on the bank’s assets (Balachandran et al., 2010; Bebchuck & Spamann; 2010; Mülbert, 2009)11. Shareholders have, resulting from their limited liability, an incentive to increase the volatility of the bank’s assets to maximize the value of the shareholders equity (Bolton et al., 2010; Bebchuck & Spamann, 2010). This incentive of shareholders might be strengthened by a lack of monitoring by creditors, balance sheet opaqueness, the existence of deposit insurance and/ or state support programs (Bebchuck & Spamann, 2010; Mülbert, 2009). The results support the statement that shareholders can effectively use variable equity based compensation to align the incentives of the CEO with their own interests to induce excessive risk taking at the expense of the other stakeholders.
B.3. CEO stake
The CEO ownership variables provide mixed evidence for the assumed risk-aversion of the CEO. First, the results show only in two of the five cases a positive influence of the percentage of shares held by the CEO on the amount of equity volatility at the 10% significance level. Second, the influence of the variable “Value CEO stake” is only significant when all other compensation variables are included in model 1. Although, the negative influence is in the predicted direction and highly significant, this standalone result provides only weak evidence. Third, the variable measuring the influence of the stake of the CEO on bank risk, in relation to his total annual compensation, is only significant in two of the five cases at the 10% significance level. Overall, the mixed results provide little evidence for the statement that the amount of equity holdings by the CEO decreases the amount of bank risk. Therefore, the results do not consistently confirm the assumed risk-aversion of the CEO. Furthermore, the results indicate that the compensation variables are stronger and more consistent in predicting bank risk than the ownership variables. Referring to the evidence of Cheng et al. (2010), the influence of the compensation variable might offset the influence of the CEO ownership variables.
11
