Bank executive compensation structure, risk taking and the financial crisis

Abstract

This paper investigates (1) how the composition of executive compensation is related to a bank’s incentive to take excessive risk, (2) whether executive compensation in larger banks, especially the too-big-to-fail (TBTF) banks, induces more severe moral hazard behavior, and (3) how the relation between bank executive compensation and risk taking changes before and during the recent financial crisis. We find that bank risk measured by the Z-score and the volatility of stock returns increases with both the percentages of short-term and long-term incentive compensation. However, greater proportion of incentive pay decreases the likelihood for a bank to become a problem or failed institution. This result holds for the periods before and during the recent financial crisis. The distress-mitigating effects of incentive compensation are further confirmed by our finding that both the proportions of bonus and long-term incentives are positively related to bank valuation and performance. Interestingly, we find that TBTF banks experience greater risk taking (lower Z-score) and are more likely to be in financial distress than smaller banks. However, greater incentive compensation in TBTF banks helps reduce their insolvency risk relative to smaller institutions.

1 Introduction

The financial crisis since 2007 has put the compensation structure of the banking firms at the forefront of many policy debates on the root causes of the banking crisis. US regulation on executive compensation has become more restrictive following the onset of the recent banking crisis. Despite the emphasis on compensation policies in the past and present banking regulations, the role of executive compensation in bank excessive risk-taking is not well understood (Murphy 2009). Public outcry against Wall Street compensation culture suggests that it is common for the general public to view executive pay as a major contributory factor in the banking industry’s meltdown. However, others argue that it is not all that clear that executive compensation is the right suspect to be held responsible for the financial crisis.¹

This paper seeks to study these controversial issues by examining whether the composition of executive compensation plays a role in promoting excessive risk taking in the banking industry. While recognizing that executive compensation is multifaceted, this paper focuses on investigating (1) how the composition of executive compensation is related to a bank’s incentive to take excessive risk, (2) whether executive compensation in larger banks, especially the too-big-to-fail (TBTF) banks, induces more severe moral hazard behavior, and (3) how the relation between bank executive compensation and risk taking changes before and during the recent financial crisis. Our findings could be useful for regulators and creditors to effectively monitor banking firms, and are helpful for shareholders and the board of directors to set the optimal compensation policy for their organizations.²

This paper contributes to the literature in the following ways. First, we distinguish normal risk taking from excessive risk taking when examining the relation between executive compensation structure and bank moral hazard. This distinction is crucial because shareholders benefit from reasonable risk taking that increases firm value, but excessive risk taking results in moral hazard that reduces firm value and transfers wealth from creditors and deposit insurers to bank shareholders (Kane 1995). Nonetheless, there is a lack of clear understanding on how executive compensation affects “excessive risk taking” while political pressures to reform the compensation structure are mounting in the financial-services industry (Murphy 2009). This paper tackles this challenge from multiple angles. We examine whether the effects of incentive compensation on a bank’s stock return volatility and Z-score differ from those on the likelihood of the bank being in financial distress. We classify a bank as being in financial distress if it is either a problem bank or a failed bank. Similar to Ashcraft (2008), we define a bank as a problem institution if its non-performing loans to equity ratio is over the 85th percentile of all US banks on an annual basis.³ We also use FDIC’s list of failed banks since the financial crisis to construct an alternative definition of problem institutions. These distressed banks are expected to increase the loss exposure of the FDIC. It is reasonable to expect and widely reported in the literature that in general, a bank becomes financially distressed as a result of taking excessive risk.

To put our work in the context of previous literature, we start our empirical analysis by examining the determinants of executive compensation structure. We find that riskier banks and TBTF banks compensate their executives with greater percentages of incentive pay. Although previous studies often treat results of this kind as evidence of moral hazard, we caution that this finding itself is not sufficient to affirm the incentive-pay induced moral-hazard. Managers receiving long-term equity-based incentive compensation may avoid myopic investments that are excessively risky (Fahlenbrach and Stulz 2011; Carpenter 2000; Ross 2004).

To confirm whether compensation-induced moral hazard is present, we then compare the effects of incentive compensation on traditional risk measures⁴ with those on the likelihood of being in financial distress. We find that bank risk increases with both the percentages of short-term and long-term incentive compensation. Fortin et al. (2010) suggest that a positive relationship between BHC risk and executive compensation may raise red flags for regulators. Our paper goes a step further and tries to ascertain if this elevated risk indeed leads to what regulators should really be concerned about—distress and failure among banks. Interestingly, we find that offering greater percentage of incentive compensation decreases the probability for a BHC to be in financial distress. These results are robust when we address the potential endogeneity of the compensation variables in the analysis, and when we include delta (CEO pay-performance sensitivity) and vega (sensitivity of CEO wealth to stock volatility) as additional control variables. The contrast between the volatility-increasing effect and the distress-mitigating effect of incentive pay could be partly explained by Carpenter (2000) and Ross (2004) who show that a risk-averse manager paid with his firm’s stocks or stock options may not strictly prefer to increase the firm’s asset risk. Core and Guay (2010) also suggest that greater equity-based compensation can potentially either increase or decrease risk taking, depending on whether the executive had the right amount of equity incentives to start with. The distress-mitigating effects of incentive compensation are further confirmed by our finding that both the proportions of bonus and long-term incentives are positively related to BHCs’ valuation and performance. This suggests that in general, managerial incentives induced by greater proportions of incentive compensation enhance firm valuation and performance, and this in turn reduces the likelihood for banks to be in financial distress.

The second contribution of this paper is our finding on the relation among executive compensation, risk taking and moral hazard around the recent financial crisis.⁵ As Faulkender et al. (2010) point out, whether flawed compensation incentives caused the recent financial crisis remains an unanswered empirical question. We find that the relation between incentive compensation on the standard deviation of stock returns and the Z-scores did not experience significant change before and during the crisis. This suggests that although both bonus and long-term incentives can serve as mechanisms to increase risk taking incentives of bank managers, this risk-increasing effect did not appear to escalate during the current financial crisis. Moreover, greater proportions of bonus and long-term incentive compensation of the TBTF banks helped increase the Z-scores and reduce the likelihood that these banks become financially distressed during the pre-crisis period of 2003–2006.

Finally, this paper provides significant evidence on the pitfalls of allowing banking firms to become too big to fail. Results indicate that TBTF banks are more likely to be in financial distress and have lower Z-scores. This suggests that TBTF banks tend to experience greater risk shifting that increases the loss exposure of the creditors and the depositor insurer. However, our findings suggest that the moral hazard problems of the TBTF banks are not due to greater proportions of incentive compensation paid to their CEOs, but rather are magnified by the size of the banks.

The remainder of this paper is organized as follows. Section 2 discusses the testable hypotheses and related literature. Section 3 describes the data and sample. Section 4 presents the empirical results. We conclude our paper with Section 5.

2 Testable hypotheses and related literature

2.1 The contracting hypothesis and the moral-hazard hypothesis

A number of studies employ contracting theories to design a firm’s optimal managerial compensation structure. Smith and Watts (1992) suggest that it is more difficult for shareholders to observe the actions of a manager in a firm that has higher growth opportunities. As a result, firms with greater growth options are more likely to use stock options to tie the manager’s compensation to firm value. On the other hand, Holmstrom and Milgrom (1987) emphasize a trade-off between inducing the optimal amount of unobservable effort by the manager and minimizing the amount of risk she is required to bear, and predict that the pay-for-performance sensitivity (PPS)⁶ of managerial compensation is decreasing in the variance of the firm’s performance. It is worth noting that John and John (1993) are the first to examine the interaction between the agency cost of equity and the agency cost of debt in setting the optimal managerial compensation. They show that shareholders also gain from reduced agency cost of debt, i.e., from reduced risk-shifting incentives of managers. The optimal managerial compensation derived from their model suggests that the PPS of a firm’s managerial compensation should be negatively related to its leverage ratio. This is because as leverage increases, the PPS of managerial compensation should decrease to offset the greater risk-shifting incentives of the managers.

Summarizing the predictions of these contracting theories, our contracting hypothesis states that banks with greater growth options use more equity-based incentive compensation, and the PPS of executive compensation is positively related to the bank’s capital ratio and negatively related to the variance of the bank’s stock returns.

On the other hand, banks’ executive compensation may encourage excessive risk taking. This argument emphasizes another layer of agency conflicts, i.e., the conflicts of interests between shareholders and the deposit insurer. Bank depositors may allow shareholders and/or managers of an insolvent bank to stay in place only because their deposits are federally guaranteed (Kane 1995). With the protections offered by the federal safety net and the potentially imperfect risk pricing of the deposit insurance, bank shareholders/managers may take excessive risk that is not deemed safe and sound. This kind of moral-hazard behavior increases the loss exposure of the insurance fund. As suggested by Houston and James (1995) and others, the moral-hazard problem could be particularly severe for banks that are troubled and/or too big to fail (TBTF). Shareholders of troubled banks have less to lose in the event of bank closure, and TBTF institutions receive implicit government subsidies for taking greater risk. Although bank moral hazard may be associated with a number of factors, our paper examines whether a bank’s incentive to take excessive risk is induced by its CEO’s incentive pay. Our moral hazard hypothesis states that banks use incentive compensation to promote excessive risk taking at the expense of the deposit insurer, and this risk-shifting incentive is particularly severe for troubled and TBTF banks.

2.2 Relation to the existing literature

A number of studies have examined executive compensation in the banking industry. Here we mainly review empirical papers that study the relation among executive compensation, bank risk, and performance, and discuss our paper’s relation to this body of literature.⁷

2.2.1 Determinants of CEO compensation

Several papers offer empirical support for contracting theories. Hubbard and Palia (1995) and Cuñat and Guadalupe (2009) find that the level and the PPS of bank CEO compensation tend to increase following deregulations in the industry. Their finding supports the conjecture that less strict interstate regulation leads to a higher level of competition that requires a more capable CEO and thus greater PPS and higher level of executive pay. Harjoto and Mullineaux (2003) find that CEO compensation increases with BHCs’ growth options, financial leverage and the standard deviation of stock returns, largely in support of the contracting theories of Smith and Watts (1992), Holmstrom and Milgrom (1987) and John and John (1993). Hermalin and Wallace (2001) find evidence consistent with the agency theory that managerial compensation increases with firm risk to compensate managers for bearing that risk.⁸ John and Qian (2003) find that the PPS of CEO compensation is lower for banks relative to manufacturing firms, consistent with the contracting theory of John and John (1993). John et al. (2010) extend the model of John and John (1993) and find evidence supporting their argument that outside monitoring that can discipline shareholders’ risk-shifting incentive allows for higher PPS in CEO pay to better align the interests between shareholders and managers.

In addition to testing the contracting theories, Houston and James (1995) and Crawford et al. (1995) examine the moral-hazard hypothesis as well. Houston and James (1995) find that bank CEOs held less equity-based compensation than nonbanking firms and that troubled and TBTF banks did not have greater stock options. They interpret these results as inconsistent with the moral-hazard hypothesis but in line with the contracting theory based on Smith and Watts (1992). Crawford et al. (1995) suggest that their findings are consistent with the contracting hypothesis that greater growth opportunities after the deregulation in 1981–1982 were associated with stronger pay-performance relations. They also find that although the increase in PPS was not significantly different between high-capitalization and low-capitalization banks,⁹ the low-cap banks had larger increases in the number of options issued and in the value of these options after deregulation than high-cap banks. They interpret these results as providing mixed evidence for the hypothesis that banks use executive compensation to induce risk shifting to the FDIC.

In summary, the existing literature on the determinants of CEO compensation provides supporting evidence for various contracting theories. However, there are mixed results on whether executive compensation is used to shift risk from shareholders to creditors and the deposit insurer.

2.2.2 Relations between executive compensation and risk-taking and performance

However, even when researchers find that lower capitalized, troubled or TBTF banks offer greater equity-based incentive compensation to their executives, this is necessary but not sufficient to ascertain that greater incentive compensation is used to induce excessive risk taking. It is crucial to understand how executive compensation affects bank risk as well. Moreover, examining how the compensation structure affects the valuation and performance of the banking firms further helps determine whether the risk-taking incentives induced by executive compensation is detrimental or beneficial to firm value. Nonetheless, only a small number of studies (e.g. Palia and Porter 2004; Benston and Evan 2006; Chen et al. 2006; Fortin et al. 2010; Cheng et al. 2010) empirically examine the effect of executive compensation on the riskiness of the banking firms, and even fewer studies (Palia and Porter 2004; Sierra et al. 2006; Fahlenbrach and Stulz 2011) empirically examine the effect of managerial compensation on bank valuation and performance.

Palia and Porter (2004) find that the level of salary and bonus of CEO compensation is negatively related to bank risk, consistent with the theory of John et al. (2000) that bank risk (measured by the standard deviation of stock returns) decreases when managers’ salary and bonus increase. Moreover, their results indicate that the value of the CEO’s stock holding is positively related to bank risk, corroborating the findings of Saunders et al. (1990). They also find that banks’ Tobin’s q increases with the value of stock options held by the CEOs.

Chen et al. (2006) find that banks increased the option-based executive compensation after deregulations in the banking industry, and the increased option-based compensation induces greater risk taking in the period of 1992–2000. The risk measures in Chen et al. (2006) include the standard deviation of stock returns, systematic, idiosyncratic and interest rate risks. In addition, Benston and Evan (2006) examine the relation between CEO incentive compensation and bank risk taking around the enactment of the 1991 Federal Deposit Insurance Corporation Improvement Act (FDICIA). They find that banks with low charter value and greater bonus payment have greater stock return volatility, less return on assets (ROA) and greater likelihood to fail in the pre-FDICIA period of 1988–1991, and this kind of risk-shifting outcome disappeared in the post-FDICIA period of 1992–1994. Moreover, banks with greater long-term incentive compensation and high charter value are less likely to fail in both periods. For a sample of BHCs during 1992–1997, Sierra et al. (2006) use a simultaneous equations system to investigate the relation among CEO total compensation, firm performance (i.e., ROA) and board strength. They find that while CEO compensation increases with the standard deviation of ROA, greater CEO compensation also resulted in higher ROA.

Some recent studies have begun to examine the role of executive compensation in the most recent financial crisis. Using firm-level data in 2005 to explain BHCs’ risk taking in 2006, Fortin et al. (2010) find that BHCs whose CEOs were granted more stock options and higher bonuses in 2005 exhibited greater risk in 2006, while BHCs whose CEOs received higher base salaries experienced less risk. Cheng et al. (2010) find during the period of 1992–2008, financial firms with greater residual compensation (defined as average total compensation of top-five executives controlling for firm size and finance sub-industry classifications) tend to have higher beta, higher return volatility and greater exposure to the ABX subprime index subsequently. They suggest these results support the conjecture that heterogeneous short-term investors invest in different firms and incentivize them to take different levels of risks, but are inconsistent with the view that executive compensation leads to mis-governance or management entrenchment.

Fahlenbrach and Stulz (2011) examine whether bank performance from 2007 and 2008 (i.e., ROA, ROE and buy-and-hold returns) is related to the CEO’s cash bonus to salary ratio, the CEO’s equity incentives and equity risk in 2006. They find that greater CEO equity incentives in 2006 were either negatively or insignificantly related to the performance measures in 2007–2008. In addition, banks with higher cash bonuses to salary ratios or greater equity risk incentives did not perform worse in 2006. Their findings that bank CEOs did not reduce their holdings of shares before the crisis is inconsistent with the view that CEOs knowingly focus on the short-term performance of their firms.

Overall, the existing literature indicates that the empirical findings on the relation among executive compensation, risk taking, valuation and performance in the banking industry vary with the data and sample selection, regulatory environment, methodologies used and compensation measurements. There is a lack of study that distinguishes normal risk taking from excessive risk taking and uses longer sample period to examine how the relations among executive compensation, risk taking, valuation and performance change before and during the financial crisis. No study has examined how executive compensation structure affects a BHC’s incentive to extract subsidies from the deposit insurance after 1994. Given the changing regulatory and financial environment after the mid 1990s, it is critical to understand how executive compensation structure has evolved to affect bank risk taking in more recent time periods. Our paper addresses these critical issues and provides implications on the role of executive competition for bank stakeholders and regulators.

3 Data and sample

The data on the compensation of the CEOs in bank holding companies are obtained from the Execucomp database and the data on the financial statements are from the Bankscope database. Execucomp data start from 1992 due to the availability of consistent disclosure of option portfolios beginning that year. Consequently, our sample period is between 1992 and 2008. We limit our sample to only publicly-traded bank holding companies with available consolidated statements from Bankscope. The Execucomp database provides compensation data, the age and tenure for mostly the top five executives in a year. It provides data for banks that are part of the S&P 1500.

We calculate the Total Compensation variable by taking the maximum of variable TDC1 (defined as the sum of the following items reported in Execucomp: salary, bonus, other annual, restricted stock grants, long-term incentive plan payouts, all other, and the value of option grants) and TDC2 (defined as the sum of salary, bonus, other annual, restricted stock grants, long-term incentive plan payouts, all other, and the value of options exercised).¹⁰ The difference between TDC1 and TDC2 is how the stock options are treated in the calculation of the total compensation. The former uses the value of options granted while the latter uses the value of options exercised. The variables Salary and Bonus are the dollar value of the base salary and bonus earned by the named executive officer during the fiscal year, respectively. Because the bonus paid to the executives is short-term in nature, the variable Bonus is a short-term compensation measure in our paper. Finally, we calculate the variable Long-Term Incentives by subtracting salary and bonus from Total Compensation.¹¹ This variable captures mainly equity-based incentive compensation and other incentives that are longer-term in nature relative to salary and bonus.¹² All compensation numbers have been adjusted for inflation using the Consumer Price Index obtained from the Bureau of Labor Statistics (with the CPI equals 100 in 2008). Furthermore, to avoid compensation figures for part of the year, we exclude data points when the year of the reported compensation is the same as the year of hire. We also winsorize extreme (1st and 99th) percentiles of the compensation data to alleviate the effect of possibly spurious outliers.

To calculate the tenure of the executives, we take the year of the financial report and subtract the year of joining the company. If the year of joining the company is not available, we use the year of becoming CEO instead. We extract the percentage of total shares outstanding held by the executive (Shareholding) from Execucomp. We also calculate the CEO ownership by including not only the shares owned by the CEO, but also options that are exercisable or will become exercisable within 60 days. Our regression results of using this variable are similar to those using Shareholding. However, the number of observations is much smaller with the more inclusive ownership variable. Therefore, we only report results using Shareholding in the tables.

We collect data on various measures of bank risk. As in Laeven and Levine (2009), Pathan (2009) and Hermalin and Wallace (2001), we measure bank risk using the standard deviation of the bank’s stock returns (σ). In particular, we estimate σ using monthly stock returns obtained from the CRSP database during the 36-month period prior to the end of the year. In unreported tables, we have also calculated 60-month and weekly return volatility measures for robustness checks. The results are robust to these alternative measures. Similar to Laeven and Levine (2009) and Esty (1997), we use the standard deviation of the return on assets σ _ROA as an alternative measure of bank risk. Using this variable produces similar results to those using the standard deviation of stock returns. We use the Z-score as another measure of bank risk. As in Laeven and Levine (2009), we calculate the Z-score as (ROA + Capital)/σ _ROA, where Capital is the bank holding company total equity to assets ratio. The Z-score measures the distance from bank insolvency.

Similar to Ashcraft (2008), we classify a bank as a problem institution if its problem loans to equity ratio is over the 85th percentile of all US banks on an annual basis. We use the non-performing loans to proxy for the problem loans.¹³ Ashcraft finds that this classification of problem banks closely approximates a CAMEL rating of 3, 4 or 5. A bank with a CAMEL rating of 3, 4 or 5 is viewed as a problem bank by the regulators. It is reasonable to expect that a bank becomes a problem bank as a result of taking excessive risk. It is important to point out that during a banking crisis, it is possible that significantly more that 15 % of the banks are distressed. However, the measure proposed by Ashcraft (2008) is not an absolute, but a relative measure. It intends to capture the most vulnerable of the banks in a year.

To make sure that our results are robust to other definitions of problem banks, we use the FDIC’s list of banks that have failed since the financial crisis to construct our second definition of problem institutions.¹⁴ Results from this alternative measure of financial distress confirm the major findings from using the first definition of problem banks. We thus only report the results from the first definition of problem banks in the table.

Similar to Houston and James (1995), we use Tobin’s q to proxy for a bank’s growth opportunities. Tobin’s q is estimated as the bank’s market value of equity plus the book value of liabilities, divided by the book value of assets. This variable is also used as a valuation measure in many previous studies (Laeven and Levine 2009). We use a dummy variable, TBTF, to indicate whether a bank is too big to fail. In September 1984, the Comptroller of the Currency testified before Congress that some banks were “too big to fail,” and for those banks total deposit insurance would be provided (O’Hara and Shaw 1990). The Comptroller stated that this policy would apply to the eleven largest banks. However, the FDICIA of 1991 mandates that the resolution of failed banks at the lowest cost to the FDIC and reduces bank regulatory agencies’ incentives to follow a too-big-to-fail policy (Wall 2010). Given the changing nature of the TBTF policy, we use three alternative cutoffs (i.e., top ten, top eleven or top five) based on the asset size of the BHCs to categorize TBTF institutions. In particular, we take the whole Bankscope sample of US BHCs and rank them according to their asset size. The variable TBTF takes a value of one if the institution is one of the ten largest BHCs by size in a particular year, and zero otherwise. We also use top eleven and top five largest BHCs respectively to construct the TBTF dummy. Results are consistent among the three definitions of TBTF variable. Results are robust even if we treat the top ten banks at the beginning of our sample period as TBTF banks and assume they stay as too-big-to-fail throughout our sample period. We thus only report in the tables with TBTF institutions defined as the top ten largest BHCs in a particular year. It is reasonable to assume that these largest institutions are more likely to be bailed out by the US government given their failure’s potential impact on systematic risk. In addition, we obtain a BHC’s annual stock return (Annualret) from the CRSP database.

The data on Gindex is from Risk-Metrics. This index, devised by Gompers et al. (2003), is a measure of the quality of corporate governance. Gindex has a possible range from 1 to 24 for every firm. Higher Gindex is associated with less shareholder rights and greater power of management. Although this measure is not available for all firm years, it tends to be very stable over time. As a result, we use linear imputation to calculate the values for years when this index values are not available.

In this paper, we also examine whether the relation between executive compensation and bank risk taking varied before or during the financial crisis which started in 2007.¹⁵ The previous recession took place in 2000–2001. By year 2003, the economy was mostly out of the recession and on a path to substantial growth. As a result, we take the years between 2003 and 2006 as the ‘before crisis’ period, and the years between 2007 and 2008 as the ‘during crisis’ period.

Table 1 provides the definitions of the variables used in this paper. Table 2 presents the descriptive statistics on CEO compensation and CEO characteristics for the firm years of our sample. The average age of the executives is approximately 64 years. The total compensation, salary, bonus and long-term incentive compensation for the CEOs in our sample averaged $10.157 million, $0.852 million, $1.701 million and $7.340 million in 2008 dollars, respectively. The average tenure of the CEOs is 11.93 years. The percentage shareholding of an executive ranges from 0 to 36.96 %, with an average of 1.72 %. Table 3 presents the summary statistics of the characteristics of the BHCs for the firm years of our sample.

Table 1 Variable description

Table 2 Summary statistics of the variables on executive compensation and executive characteristics for 134 bank holding companies during the period of 1992–2008

Table 3 Summary statistics of the variables on the characteristics of 134 bank holding companies during the period of 1992–2008

As a robustness check, we use the spread on banks’ credit default swaps (CDS) as an alternative measure of bank default risk. We extract daily bid and ask information on the CDS spreads for all the banks from Bloomberg. Because reliable data on the CDS are available from 2002, we obtain the CDS spreads for the period of 2002–2008. As in Völz and Wedow (2011), we use the spreads of 5-year CDS because trading liquidity is highest for this maturity bucket. To calculate the prices of CDS, we follow Chen, Fabozzi, and Sverdlove (2010). As they explain, since the illiquidity of CDS market always reduces the price, CDS settlement price will be lower than the perfectly liquid CDS premium. The perfectly liquid premiums are unobservable, and they cannot be more than the ask quotes. As a result, we use two different methods for calculating CDS price—the asking price and the mid-point of the bid and the ask prices. However, only 5 banks from our executive compensation sample have valid CDS data. We confirm the accuracy of this small sample size by comparing our sample with the list of sample institutions reported by Völz and Wedow (2011) who obtain the CDS data for all the banking firms from Bloomberg and Datastream. Since compensation data are issued annually, we apply two different methodologies to merge the CDS data to Execucomp data—taking the year-end value and taking the average value over the year. Regardless of the two methodologies, they are highly correlated (more than 90 % correlation). However, the number of observations with CDS spread is too small for us to run comparable regressions as those in Tables 5, 6, 7, 8, 9, 10. Nonetheless, the correlation matrix for CDS ask price, CDS mean (bid/ask average), σ and Z in Table 4 shows that the correlation coefficients between CDS_ask and σ and between CDS_ask and the Z-score are 0.6247 and −0.2780, respectively, both statistically significant at the 5 percent level. These strong correlations suggest that despite σ and Z being historical measures of bank risk, they are highly correlated with the forward looking risk measure, the CDS spread. This suggests that risk measures σ and Z are good proxies for banks’ expected risk.

Table 4 Correlation matrix

4 Regression results

4.1 The determinants of executive compensation

To test the contracting hypothesis and the moral-hazard hypothesis, we first estimate a set of compensation regressions as specified in Eq. (1)

$$\begin{aligned} Comp_{\text{it}} & = a_{0} + a_{1} Tobin_{{{\text{it}} - 1}} + a_{2} \sigma_{{{\text{it}} - 1}} + a_{3} Capital + \left( {a_{4} + a_{5} \sigma_{{{\text{it}} - 1}} + a_{6} Capital_{{{\text{it}} - 1}} } \right)Ret_{\text{it}} \\ \quad + a_{7} TBTF_{{{\text{it}} - 1}} + a_{8} ProblemBank_{it - 1} + a_{9}^{{\prime }} X1_{it - 1} + a_{10}^{{\prime }} YearDummies + \varepsilon_{it} \\ \end{aligned}$$

(1)

where the dependent variable Comp _it is one of the following four compensation measures of the CEO of firm i in year t: (1) total compensation, (2) salary, (3) bonus (short-term incentive), and (4) long-term incentives. The explanatory variables include 1-year lagged values of Tobin’s q (Tobin), the standard deviation of the previous 36-month stock returns (σ), interactive terms between the dollar return to shareholders (Ret) and one of the two variables, σ, and the bank’s capital to assets ratio (Capital), the TBTF dummy, and the ProblemBank dummy. Eq. (1) also includes a vector of control variables (X1), year dummies, and a random error term ε _it. The control variables (X1) include Ln(Assets), Age, Tenure, and Gindex.

As in Aggarwal and Samwick (1999) and John et al. (2010), the dependent variable in Eq. (1) is measured in dollars.¹⁶ The specification in Eq. (1) allows us to examine not only the determinants of the level of executive compensation, but also the determinants of the PPS. The PPS of executive compensation can be estimated as the sum in the parenthesis in Eq. (1), i.e., a ₄ + a ₅ σ _it−1 + a ₆ Capital _it−1. We use 1-year lagged values of the explanatory variables so that they are not jointly determined with the contemporaneous dependent variable. John et al. (2010) show that using lagged values of explanatory variables in their compensation regressions yield similar results to those of the simultaneous equations that are used to address potential endogeneity problems. Positive a ₁, a ₄, and a ₆ and negative a ₅ in Eq. (1) are consistent with the contracting hypothesis. On the other hand, if banks use the short-term and/or long-term incentives to promote excessive risk taking, as the moral-hazard hypothesis suggests, we expect that riskier banks, problem banks and TBTF banks pay greater incentive compensation to their CEOs to encourage aggressive risk taking. As in Hermalin and Wallace (2001), we use σ and Capital to proxy for an institution’s overall risk and insolvency risk, respectively. Under the moral hazard hypothesis, we expect the coefficients a ₂, a ₇, and a ₈ to be positive and coefficient a ₃ to be negative for the regressions with bonus and long-term incentives. However, positive a ₂, a ₇, and a ₈ and negative a ₃ are necessary but not sufficient to support the moral-hazard hypothesis. This is because riskier banks may pay more to risk-averse executives (i.e., a positive a ₂ and a negative a ₃) to compensate for their greater employment risk even though their risk taking is not deemed to be excessive. Positive a ₇ and a ₈ of the long-term incentives regression are also consistent with the scenario that TBTF or problem banks want to use long-term incentive compensation to avoid short-term managerial opportunism. Moreover, as Houston and James (1995) point out, firms with greater growth options have greater franchise value.¹⁷ Higher franchise value reduces banks’ incentive to take excessive risk to shift losses to the FDIC. If moral-hazard incentive is greater for banks with lower franchise value, and banks with lower franchise value use greater incentive compensation to promote risk taking, we expect to find a negative coefficient on Tobin (a ₁) in the regressions of the bonus and/or long-term incentives in Eq. (1) under the moral hazard hypothesis.

We include Ln(Assets), Age, and Tenure as control variables because previous studies (Barro and Barro 1990; Murphy 1999, and others) indicate that firm size, the age and the experience of the executives might be important factors to explain executive compensation. John et al. (2008) suggest that the quality of shareholder protection may affect corporate risk taking. If BHCs intend to use managerial compensation to influence risk taking, then the quality of shareholder protection may also affect the structure of executive compensation. As in John et al. (2008), we also include the corporate governance index of Gompers, Ishii, and Metrick (Gindex) to measure the quality of shareholder protections. Because the CEO may have different compensation structure than other top executives (Angbazo and Narayanan 1997; Demsetz and Saidenberg 1999), we also estimate Eq. (1) using a sample of the top five executives of the BHCs and include a CEO dummy as an additional explanatory variable. The results from these regressions confirm our CEO-only analysis. To conserve space, we only report the results from the CEO-only sample in the tables.

Table 5 reports the results from the pooled panel data regressions of the total compensation, salary, bonus and long-term incentives of the CEOs of the bank holding companies for the period of 1992–2008. The error terms are clustered at the firm level. We adjust our standard errors using Newey–West methodology to address the potential heteroskedasticity problem.

Table 5 Determinants of CEO compensation for the period of 1992–2008

Results indicate that consistent with the contracting theory of Smith and Watts (1992), the coefficient on Tobin’s q is significantly positive in the regression of the long-term incentive compensation.

In addition, the coefficient on Ret is significantly positive for the regression of the CEO bonus, corroborating the finding of Harjoto and Mullineaux (2003) that the CEO bonus increases with the return to shareholders. The cross product of σ and Ret is significantly negative for the regressions of the long-term incentive compensation and the total compensation, indicating that the pay for performance sensitivity of managerial compensation is negatively related to the volatility of the firm’s stock returns. This finding lends support to the agency model of Holmstrom and Milgrom (1987). Overall, these findings support the contracting hypothesis, and indicate that on average, executive compensation of the banking firms is designed to align the interests between shareholders and managers.

Table 5 also indicates that the coefficient on σ is significantly positive for all four regressions. This suggests that riskier banks pay more to their CEOs. The coefficient of Capital is negative but only statistically significant at the 10 % level in the bonus regression. The significantly positive coefficient on TBTF in the bonus regression indicates that TBTF banks pay greater short-term incentives to their CEOs. However, whether greater executive compensation of riskier banks and TBTF banks is to encourage managers to take excessive risk or to reward managers for bearing greater employment risk and for making greater efforts to manage a more complex institution cannot be concluded from these compensation regressions alone.¹⁸ Moreover, Table 5 shows that the coefficient on ProblemBank is not significant in any of the four regressions. This result is inconsistent with the view that problem banks pay more to their executives to promote greater risk taking.

The results on the control variables are also interesting. Corroborating the findings of Barro and Barro (1990) and others, we find that banks with greater asset size pay significantly higher compensation to their executives. The salary of CEOs also increases with their tenure at the bank, and all the components of compensation increase with the age of the executives. We also find that the coefficient on Gindex is significantly negative for the bonus and the long-term incentives regressions. Given that firms with less shareholder rights have higher Gindex, this result suggests that firms with better shareholder protection pay greater incentive compensation to their CEOs.

4.2 The effect of the financial crisis on executive compensation

To examine the effect of the recent financial crisis on the determinants of executive compensation, we add the interactions between the Crisis dummy and other explanatory variables in the regressions to explain the amount of the components of executive compensation. The Crisis dummy equals 1 if the observation is from the crisis period of 2007–2008, and 0 if it is from the period of 2003–2006.¹⁹ The interaction terms are included in the reported compensation regressions in Table 6 only if they are significant in at least one of the four compensation regressions. The regressions in Table 6 are pooled panel data regressions of the total compensation, salary, bonus and long-term incentives of the CEOs of the bank holding companies for the period of 2003–2008. We use 1-year lagged values of the explanatory variables so that they are not jointly determined with the contemporaneous dependent variable.

Table 6 The determinants of CEO compensation around the crisis: 2003–2008

Table 6 shows that the coefficient on the cross product of ProblemBank and Crisis is significantly negative at the 10 % level for the salary regression only, and this coefficient is negative but insignificant for the other three regressions. Because of its low significance level for this coefficient, we do not intend to make strong inference from this result. In addition, the coefficient on the cross product of Ln(Assets) and Crisis is significantly negative in the total compensation regression, indicating that larger institutions had greater reduction in total compensation to their CEOs during the crisis. Although the results suggest that the compensation practices in larger banks behaved in the way that we would expect during an economic downturn, the change in salary for larger institutions during the crisis is not economically significant. For example, a one standard deviation change in Ln(ASSETS) yields only $4,800 (i.e., 1.62 × $3,000) change in compensation, but the mean compensation is around $10 million.

Table 6 also indicates that the relation between growth opportunities and long-term incentives remain positive and statistically significant, consistent with the contracting theory of Smith and Watts (1992). Although the coefficients of the cross product of σ × Ret have the same signs as in Table 5, they are no longer statistically significant. The results on most of the control variables are similar to Table 5, except that the coefficients on Tenure and Gindex are no longer statistically significant.

In summary, our results on the determinants of the level of executive compensation are consistent with the predictions of various contracting theories. Our finding that problem institutions did not pay greater incentive compensation during either the whole sample period or the sub-period of 2003–2008 suggests that on average, bank in financial distress did not increase executive pay to encourage greater risk taking. However, we find that TBTF banks and banks with greater stock return volatility did have greater level of executive compensation. Nonetheless, this result is only necessary but not sufficient to show that greater incentive pay is used to promote excessive risk taking for riskier and TBTF banks. We next examine how the incentive compensation of CEOs affects bank risk taking.

4.3 Executive compensation, risk taking and the effect of financial crisis

As suggested by Ang et al. (2002), pay structure is at least as significant as pay levels. We characterize compensation structure by the percentage of each compensation component in total pay similar to Ang et al. (2002). We examine whether greater proportions of bonuses and long-term incentives promote risk taking by estimating the following regressions:

$$Risk_{\text{it}} = b_{0} + b_{1} Pct\_Bonus_{{{\text{it}} - 3}} + b_{2} Pct\_LTIncentives_{{{\text{it}} - 3}} + b_{3}^{{\prime }} X2_{{{\text{it}} - 3}} + b_{4}^{{\prime }} YearDummies + \varepsilon_{\text{it}}$$

(2)

We use various measures of bank risk (Risk) discussed in Sect. 3 as dependent variables to estimate the regressions in Eq. (2). Table 7 reports the results from the pooled panel data regressions of (1) the natural logarithm of the standard deviation of a bank’s monthly stock returns during the previous 36 months, lnσ, and (2) the Z-score, Z. We also use other risk measures, such as the natural logarithm of the standard deviation of a bank’s monthly stock returns during the previous 60 months and the natural logarithm of σ _ROA as the regressand (not reported in the table), and the results are qualitatively similar to the regression using lnσ as the dependent variable. It is useful to note that since σ cannot be negative, we use natural logarithm transformation of the dependent variable σ (i.e., Box–Cox transformation) to alleviate potential issues of heteroscedasticity associated with the distributions of the variable to measure risk. This is not a significant issue in our regressions analysis, since our regression estimates with raw σ produces conclusions that are qualitatively similar.

Table 7 Executive compensation and bank risk

The key explanatory variables of interest, Pct_Bonus and Pct_LTIncentives, are bonus and long-term incentives as percentages of the CEO’s total compensation, respectively, defined similar to Ang et al. (2002) and Benston and Evan (2006). As robustness checks, we also use bonus and long-term incentives scaled by the log of total assets as in Fortin et al. (2010), or the log of bonus and long-term incentives as alternative explanatory variables in the regressions. Results from using these alternative measures (not reported in the tables) are qualitatively similar to the ones from using variables Pct_Bonus and Pct_LTIncentives. Because the first risk measure lnσ is calculated with 36 months of stock returns, we use 3-year lagged values of the explanatory variables in the regressions of lnσ so that they are not jointly determined with the contemporaneous dependent variable. We use 1-year lagged values of the explanatory variables in the regressions of Z.

Columns (1) and (3) of Table 7 report the regressions for the whole period of 1992–2008, while Columns (2) and (4) examine the effects of the financial crisis and a bank’s TBTF status on the relation between executive compensation and risk taking using the period of 2003–2008. We include the cross products of the incentive compensation variables and the Crisis dummy, and the cross products of the incentive compensation variables and the TBTF dummy as additional explanatory variables in Columns (2) and (4). The control variables in X2 include the percentage ownership of the CEO (Shareholding and Shareholding ²), TBTF, Ln(Assets), return on equity (ROE),²⁰ Age, Tenure, Gindex and year dummies.²¹ We also use the percentage of shares and options²² owned by the CEO relative to the total number of shares outstanding as an alternative measure CEO ownership. Results from using this alternative measure (not reported in tables) are similar to those using Shareholding.

If greater proportions of bonus and long-term incentives promote risk taking, we expect b ₁ and b ₂ in Eq. (2) to be positive in the regression of lnσ. Because higher Z-scores indicate greater distance from insolvency, for the regression of Z, b ₁ and b ₂ should be negative if greater proportions of incentive compensation increase insolvency risk. However, it is important to keep in mind that finding a positive relation between incentive compensation and bank risk from regressions in Eq. (2) is only necessary but not sufficient to conclude that banks use their compensation structure to promote excessive risk taking.

Table 7 shows that the coefficients on Pct_Bonus and Pct_LTIncentives are both significantly positive for the regressions of lnσ in Columns (1) and (2). As expected, the coefficients on the two variables are all significantly negative for the regression of Z in Columns (3) and (4). These results suggest that bank risk increases with both the percentages of short-term and long-term incentive compensation. These findings suggest that incentive compensation can serve as a mechanism to increase the risk taking incentives of bank managers. While the coefficient estimates of Ln(Assets) and TBTF are mostly insignificant, the coefficients of TBTF in Columns (3) and (4) are negative and statistically significant. This implies that TBTF banks have higher risks as measured by the Z-score.

Columns (2) and (4) show that none of the interactive terms between the incentive compensation variables and the Crisis dummy is significant for the lnσ or Z-score. This suggests that although both bonus and long-term incentives can serve as mechanisms to increase the risk taking incentives of bank managers, this risk-increasing effect did not escalate during the current financial crisis. The interactive terms between the incentive compensation variables and the TBTF dummy are insignificant in the regression of lnσ. However, they are significantly positive in Column (4). We perform an F-test on whether the sum of the coefficients on Pct_Bonus and Pct_Bonus × TBTF and the sum of the coefficients on Pct_LTIncentives and Pct_LTIncentives × TBTF are significantly different from zero. We find the former sum is insignificant while the latter one is significantly positive at the 5 % level. This suggests that for a TBTF bank, the percentage of bonus in CEO pay does not affect its Z-score, while greater long-term incentive compensation increases its Z-score (i.e., lowers bank risk). In addition, the significantly negative coefficient on the TBTF dummy suggests that TBTF banks experience greater risk than smaller banks. These findings indicate that despite TBTF banks’ greater risk as shown by their lower Z-score, greater incentive compensation in TBTF banks helps reduce their risk relative to smaller institutions.

4.4 Executive compensation and financial distress

The moral-hazard hypothesis states that banks use incentive compensation to promote excessive risk taking at the expense of the deposit insurer, and this risk-shifting incentive is particularly severe for troubled and TBTF banks. We perform a logit regression to ascertain whether incentive compensation increases the likelihood for banks to be in financial distress, and whether TBTF banks are more likely to use incentive compensation to promote excessive risk taking. We then compare the findings of this logit model with those of the determinants of bank risk (i.e., σ and Z-score) to determine whether CEO incentive compensation has different effects on the likelihood of bank distress and on bank risk taking. This analysis allows us to distinguish normal risk taking from excessive risk taking, and is a unique contribution of our study to the banking literature.

We define a bank as being in financial distress and thus, being excessively risky, if it is classified as a problem bank. As discussed in Sect. 3, we classify a bank as a problem institution if its problem loans to equity ratio is over the 85th percentile of all US banks. Falling into the problem-bank category should be caused by the bank’s excessive risk taking behavior ex ante. This definition of problem banks allows us to distinguish normal risk taking from excessive risk taking and gives us a dynamic (i.e., annual) look at the health of the banks, rather than a rigid outcome like failure.

To examine whether a bank’s financial distress can be attributed to the compensation structure of its CEO, we estimate a logit regression as specified in Eq. (3).

$$\begin{aligned} Prob(ProblemBank = 1) & = F(c_{0} + c_{1} Pct\_Bonus_{{{\text{it}} - 1}} + c_{2} Pct\_LTIncentives_{{{\text{it}} - 1}} + c_{3} Pct\_Bonus_{{{\text{it}} - 1}} \times TBTF \\ \quad + c_{4} Pct\_LTIncentives_{{{\text{it}} - 1}} \times TBTF_{{}} + c_{5}^{{\prime }} X3_{{{\text{it}} - 1}} ), \\ \end{aligned}$$

(3)

where variable ProblemBank equals 1 if a bank is a problem bank, and 0 otherwise. X3 is a set of control variables that include 1-year lagged values of Shareholding, Shareholding ², TBTF, Ln(Assets), ROE, Age, Tenure and Gindex. Eq. (3) indicates that the probability of a bank being a problem institution is a function of the term in the parenthesis. If a bank uses the bonus and long-term incentive compensation to promote excessive risk taking, then c ₁ and c ₂ should be positive. In addition, because TBTF banks have greater incentives to take excessive risk because of the implicit government guarantee, we expect c ₃ and c ₄ to be positive under the moral hazard hypothesis.²³

The reason we need to perform both regressions in Eqs. (1) and (3) is that a sufficient condition to support the moral hazard hypothesis requires positive a ₂, a ₇, and a ₈, and negative a ₃ in Eq. (1), and positive c ₁ and c ₂ in Eq. (3). Performing regressions in Eqs. (2) and (3) allows us to show the difference between normal risk taking and excessive risk taking.

Table 8 reports the results of the logit regressions. Columns (1)–(3) present the regression estimates with the maximum likelihood and Columns (4)–(6) present the estimates using a two-stage estimation method. This is mainly done to show that our results are robust to concerns of endogeneity. Coles, Daniel and Naveen (2006) provide a detailed guidance on how to test for endogeneity issues in executive compensation. We followed their advice and have reproduced our results using multiple empirical tests as suggested in Coles, Daniel and Naveen (2006). For the sake of brevity, we only mention the two-stage estimates using lagged values of the variables as instruments in this paper. To examine whether the determinants of financial distress changed after the start of the financial crisis, Columns (1), (2) and (3) report the logit regressions for the whole sample period (1992–2008), before-the-crisis period (2003–2006), and during-the-crisis period (2007–2008) respectively. A similar format applies to Columns (4), (5) and (6).

Table 8 Executive compensation and bank distress

Table 8 shows that neither the coefficients on Pct_Bonus and Pct_LTIncentives nor the coefficients on the interactive terms are significantly positive. The coefficients on Pct_Bonus in Column (5), on the cross product of TBTF and Pct_Bonus in Columns (1), (2), (4) and (5), on Pct_LTIncentives in Columns (1), (3) and (5) and on the cross product of TBTF and Pct_LTIncentives in Columns (1), (2), (4) and (5) are significantly negative. The significantly positive coefficients on the TBTF dummy in Columns (1), (2), (4) and (5) suggests that TBTF banks are more likely to be in financial distress. Taken together with the findings in Table 7, these results indicate that, despite the risk-increasing effect of incentive compensation, greater proportions of bonus and long-term incentives do not result in excessive risk taking. The theoretical model of Ross (2004) may offer some explanations to these findings. Ross (2004) distinguishes three effects of incentive compensation: the convexity effect, the translation effect, and the magnification effect. The convexity effect refers to the risk-increasing effect of stock options in the compensation schedule of the executive. The translation effect occurs when the compensation schedule moves the evaluation of a risk taking activity to a different part of the executive’s utility function with either greater or lesser risk aversion. The magnification effect depends on whether the value of incentive pay increases faster or slower than the value of the underlying shares. Ross (2004) suggests that if the increase is faster (slower), then the magnification effect will make the executive more (less) risk averse. He argues that the sum of the three effects determines whether the executive becomes more or less risk averse when he/she is paid with more stock options. Carpenter (2000) also shows that under certain condition, an executive may reduce firm risk when the number of call options in executive pay increases. The changing nature of executives’ risk aversion along different compensation schedules may explain our finding that on average, the risk-increasing effect of incentive compensation does not increase bank risk to an excessive level.

Moreover, Table 8 shows that despite TBTF banks’ higher likelihood to be in financial distress, greater incentive compensation in these large banks actually helps reduce their insolvency risk. This finding is in line with the result in Table 7 that greater incentive compensation in TBTF banks helps reduce their risk relative to smaller institutions.

As Sundaram and Yermack (2007) suggest, executive pension payment may be viewed as a form of inside debt holdings by executives. To examine whether including this debt-like component in Pct_LTIncentives affects our results, we include the proportion of pension payment in the CEO’s total compensation as an additional explanatory variable in the logit regressions. Correspondingly, we exclude the CEO pension payment from Pct_LTIncentives in the logit regressions. Results from this alternative specification confirm the signs of the coefficients on Pct_Bonus and Pct_LTIncentives in Table 8.²⁴ Finally, it is important to note that in our first definition of problem institutions, we use the measure of problem loans to equity ratio being over the 85th percentile to construct a dummy dependent variable for our logit regressions. This is mainly to follow Ashcraft (2008) who finds that this definition of problem banks closely approximates a CAMEL-based classification of problem banks used by the regulators. However, one could debate this cut-off point. To remedy such concern, we replicate our results using problem loans to equity ratio as a continuous variable. Our results hold. We also replicate the equations with various loan-loss measures and find qualitatively similar results.

4.5 Executive compensation, valuation, performance, and the financial crisis

To determine whether managerial incentives induced by various components of compensation are beneficial or detrimental to the banking firms, we perform a set of regressions to examine whether greater incentive compensation enhances firm valuation and performance. We estimate the following regressions of the valuation and performance measures.

$$\begin{aligned} Firm\,Valuation\,or\,Performance & = d_{0} + d_{1} Pct\_Bonus_{{{\text{it}} - 1}} + d_{2} Pct\_LTIncentives_{{{\text{it}} - 1}} + d_{3} Pct\_Bonus_{{{\text{it}} - 1}} \times TBTF \\ \quad + d_{4} Pct\_LTIncentives_{{{\text{it}} - 1}} \times TBTF + d_{5}^{{\prime }} X4_{{{\text{it}} - 1}} + d_{6}^{{\prime }} YearDummies + \varepsilon_{\text{it}} \\ \end{aligned}$$

(4)

where the valuation measure is Tobin’s q, the performance measures are the Sharpe ratio, return on assets (ROA) and return on equity (ROE). X4 is a vector of control variables which include 1-year lagged values of TBTF, Capital, Shareholding, Shareholding ², ProblemBank, Ln(Assets), Age, Tenure, Gindex and year dummies.

Table 9 reports the results on the panel data regressions of CEO compensation on these valuation and performance measures for the period of 1992–2008. To examine whether the recent financial crisis changed the relation between executive compensation and firm valuation and performance, we also report in Table 10 regressions with additional interactive terms between the Crisis dummy and the proportions of bonus and long-term incentives for the period of 2003–2008. The error terms of the regressions are clustered at the firm level.

Table 9 Executive compensation, valuation and firm performance for the period of 1992–2008

Table 10 Executive compensation, valuation and firm performance around the financial crisis: 2003–2008

We find that the coefficients on Pct_Bonus, Pct_LTIncentives are significantly positive in almost all the regressions of the valuation and performance measures for both the whole sample period and the subperiod of 2003–2008. The exception is the coefficient of Pct_LTIncentives in the regression of the Sharpe ratio in Table 10. It is positive but insignificant. These findings suggest that overall, both short-term and long-term incentives are positively related to bank valuation and performance. The positive coefficients on Pct_LTIncentives are consistent with the findings of Mehran (1995) that greater equity-based executive compensation increases firm value and firm performance for a sample of manufacturing firms. Frye (2004) also finds that for a sample of firms in non-regulated industries, their Tobin’s q is positively related to the percentage of equity based employee compensation (for both executives and non-executives).²⁵ Together with the findings in Table 8, our results support the view that greater proportions of incentive compensation do not encourage excessive risk taking. Instead, they increase firm valuation, improve the risk-return tradeoff (higher Sharpe ratio), and enhance firm performance as measured by the ROA and ROE.

Moreover, Tables 9 and 10 show that the interactive terms Pct_Bonus × TBTF and Pct_LTIncentives × TBTF are insignificant for the regression of the Sharpe ratio in Panel (1). This suggests that the effect of incentive compensation on the risk-adjusted returns for TBTF banks is similar to that for other banks. The coefficients on these interactive terms in Panels (2), (3) and (4) are significantly negative (albeit only significant at the 10 % level for the regressions of Tobin’s q and ROE) for the whole-period sample (Table 9). These negative coefficients of the interactive terms are no longer significant in the regression of Tobin’q for the subperiod of 2003–2008 (Table 10). The contrast in the significance level of the coefficients on these interactive variables between the regressions in the first two panels and those in the last two panels could be due to the fact that the Sharpe ratio and Tobin’s q are based on market information, while ROA and ROE are accounting measures. Out of the four performance and valuation measures, the Sharpe ratio is the only one that is adjusted for risk. These findings suggest that all else equal, TBTF banks with greater percentages of incentive compensation have lower ROA and ROE than other banks. However, TBTF banks do not experience worse risk-adjusted stock returns with greater incentive compensation.

Tables 9 and 10 also show that the coefficient on TBTF is insignificant for the regressions of the Sharpe ratio and Tobin’s q, but is significantly positive for the whole sample period and the sub-period of 2003–2008 only for the ROA regressions. This suggests that TBTF banks do not have significantly different market-based performance and valuation measures (i.e., the Sharpe ratio and Tobin’s q) from other banks while experiencing higher ROA than non-TBTF banks. These findings suggest the sources of difference in the performance and valuation measures (especially market-based measures of Sharpe ratio and Tobin’s q) between TBTF banks and other banks have been mostly included in the control variables. Moreover, results indicate that higher capitalized banks have greater Tobin’s q, ROA and ROE. As expected, problem banks have lower valuation and poorer stock and accounting performance. Moreover, the insignificance of the interactive crisis terms in Table 10 indicates that the recent financial crisis did not change the effect of incentive compensation on bank valuation and performance.

4.5.1 Further robustness checks

Because we use panel data methodology, the variables included in our regression estimates, both dependent and independent, are mostly time-dependent. To ensure that we control for some of the biases associated with such data, we have undertaken steps commonly utilized in the finance literature—such as, taking at least a 1-year lag (if not mentioned otherwise) for the explanatory variables and including time dummies wherever possible. To ensure that our results are robust, we have replicated the tables using two additional robustness methods—first, we take a 2-year lag of the explanatory variables and second, we take the long-term averages of the both dependent and explanatory variables. Using both of these additional methodologies, we find that our results are qualitatively similar.

In this study, we define the compensation variables as Salary/Total Compensation, Bonus/Total Compensation and Long-term Incentive/Total Compensation. It is possible to define these variables in different ways, such as using Firm Size as the denominator (e.g. Salary/Firm Size) or taking the natural logarithm of the variables [e.g. Ln(Salary)]. We have replicated our regressions with these alternative definitions. The conclusions are qualitatively similar.

To further check the robustness of the results to endogeneity issues, we also devise a method of logit regression estimations. The first step involves turning the dependent variables into indicator (dummy) variables. For example, to replicate the regressions in Table 7 column (1), we create an indicator variable which equals 1 if the level of volatility is greater than the median volatility of the banks in that particular year and 0 otherwise. Then we estimate the regression equation with the same explanatory variables and robust error clustering. The results are similar to our simple OLS regressions. The benefit of this methodology is that, while alleviating the correlation between the error term and the explanatory variables, we do not need to find any instruments as is necessary in the instrumental variable (IV) regressions. As a result, it may not suffer from problems of over-identification or weak instruments common to IV models. We replicate this process for Table 7 (risk measures of volatility and Z-score), Table 9 (Bank Valuation) and Table 10 (Bank Valuation). All the results hold. Therefore, it does not appear likely that our results are fully driven by dependent and explanatory variables that are jointly determined or have some time-dependent nature.

Moreover, we conduct robustness tests to determine if our results hold when we include delta and vega in our regressions as additional control variables. Following Guay (1999), Core and Guay (2002) and Coles et al. (2006), delta is the dollar change in the CEO wealth for a one percent change in the price of the stock, while vega is the sensitivity of the stock options held by the CEO to a 0.01 change in the annualized stock return volatility. While controlling for delta and vega, the coefficients on the proportions of bonus (Pct_Bonus) and long-term incentives (Pct_LTIncentives) confirm the findings of the logit models in Table 8. We also include delta and vega as additional control variables in the regressions of the performance measures as in Table 9. The coefficients on Pct_Bonus and Pct_LTincentives remain significantly positive as in Table 9.

5 Conclusions

This paper examines how the composition of executive compensation is related to a bank’s incentive to take excessive risk, and whether executive incentive compensation in larger banks, especially the too-big-to-fail (TBTF) banks induces more severe moral hazard behavior. Our sample, covering the period of 1992–2008, allows us to examine the relation between bank executive compensation and risk taking before and during the recent financial crisis.

We find that bank risk measured by the Z-score and the volatility of stock returns increases with both the percentages of short-term and long-term incentive compensation. However, greater proportion of incentive pay does not increase the likelihood for a bank to be a problem or failed institution. This result is corroborated by our finding that greater proportion of incentive pay is positively related to banks’ valuation and performance. Overall, these findings are consistent with our results on the determinants of the executive compensation structure. We find that the composition of bank executive compensation is in line with the prediction of the contracting theories. Our paper highlights the importance of distinguishing normal risk taking from excessive risk taking in examining the relation between executive compensation structure and bank moral hazard. Interestingly, we find that TBTF banks experience greater risk taking (lower Z-score) and are more likely to be in financial distress than smaller banks. However, greater incentive compensation in TBTF banks helps reduce their insolvency risk relative to smaller institutions.