Downside Risk and Agency Problems in the U.S. Financial Sector:
Examining the Eect of Risk Incentives from 2007 to 2010*
Sjoerd Van Bekkum
†This Version:
November 22, 2010
Abstract
I consider risk incentives within nancial institutions in the presence of two types of potential agency problems. Risk incentives are associated with non-negative to positive returns in subsequent years, in line with shareholder objectives. However, risk incentives also increase nancial institutions' risk tolerance, magnitude of extreme losses, and exposure to loss spillovers in subsequent years. This is not in line with societal objectives. These ndings indicate a risk-shifting problem between shareholders and society, rather than the standard manager-shareholder agency problem addressed by recent regulation.
In particular, risk incentives increase downside risk over and above leverage eects, and induce managers to implement signicantly more aggressive leverage policy.
Keywords: Executive Compensation, Risk Management, Financial Institutions JEL Classications: G21, G28, G30, G34, J33
* Financial support by Erasmus University, the Niels Stensen Foundation, and the Tinbergen Institute is gratefully acknowledged. I thank my adviser, David Yermack, as well as Viral Acharya, Giuseppe Corvasce, Ingolf Dittmann, Anjolein Schmeits, and Dan Zhang for valuable comments and suggestions.
†New York University, Stern School of Business, 44 West Fourth Street, New York, NY 10012-1126; e-mail:
[email protected]; phone: (+1) 212-998-0317.
1 Introduction
Since the near collapse of the nancial system in 2007-2008, a widespread assumption gained ground that the risk of nancial institutions will be more eectively monitored once more power is assigned to shareholders.
For instance, specically targeting the nancial sector, the Corporate and Financial Institution Compensation Fairness Act of 2009 expanded the rights of shareholders in approving compensation practices, and appointing directors on compensation committees. The same assumption underlies the Dodd-Frank Wall Street Reform and Consumer Protection Act of 2010, which mandates shareholders to vote on executive compensation and empowers shareholders to design their own compensation proposals. However, as rst pointed out by Guay (1999), it is well known that shareholders worry about executives taking too little risk. For this reason, equity-based compensation depends on share performance (price incentives) and return volatility (risk incentives) that should increase shareholder value and encourage risk-taking.
Considering the tension between shareholders' desire for value-increasing risk and legislators' purpose to limit value-destroying risk, I investigate this assumption by examining the role that shareholder-based incentives play in achieving shareholder objectives and societal objectives. Additionally, since a growing literature documents the importance of leverage policy in the 2007-2008 nancial crisis, I specically consider leverage eects in the relation between risk incentives and societal objectives.
The evidence presented in this article suggests that, when it comes to nancial institutions, shareholder objectives are not identical to societal objectives. I nd that risk-taking incentives are positively related to returns if recent (post-crisis) data are included, indicating that risk incentives induce executives to enhance nancial performance. This is in line with shareholder objectives. At the same time, I nd that managers of nancial rms with stronger risk incentives are willing to absorb larger losses, take signicant risks leading to higher losses during bad times, and increase the risk of loss spillovers from other nancial institutions.
Therefore, risk incentives have a destabilizing eect on nancial institutions after 2006. This is not in line with societal objectives. It follows that the problem is not so much the standard managerial agency problem between managers and shareholders as it is the risk-shifting problem between shareholders and society.
This article makes three contributions to the literature. First, I analyze the extent to which risk incentives are aligned with shareholder interests between December 2006 and June 2010. This is done by a compre- hensive update on more timely work that investigates executive compensation within nancial institutions during the crisis. The leading example of this research is Fahlenbrach and Stulz (2011), who nd little to no signicant eect of 2006 risk incentives on shareholder returns from mid-2007 to December 2008. Since CEOs had substantial wealth invested in their banks and suered substantial losses during the crisis, they conclude that CEOs focused on the interests of their shareholders (...) and took actions that they believed
the market would welcome. I extend their analysis using data regarding incentives and shareholder returns up to mid-2010, permitting me to consider the full extent of the crisis.
Specically, I nd that risk incentives do not signicantly increase 2007-2008 returns after including non- banking nancial institutions, calculating risk incentives and price incentives on a dollar-per-share basis, and controlling for industry and leverage eects. This result corresponds closely to Fahlenbrach and Stulz (2011).
However, when the sample period is extended to include data up to mid-2010, the results reveal a positive and signicant relation between risk incentives and subsequent buy-and-hold returns. This indicates that compensation mattered between 2006 and 2010, and is consistent with theories that shareholders implement compensation policies to enhance rm performance. Since managers appropriately serve shareholder interests by enhancing returns in subsequent years, the results support the conclusion of Fahlenbrach and Stulzthat there is no conict of interest between managers and shareholders.
Second, I analyze the extent to which managerial risk incentives are aligned with societal objectives. While shareholders are willing to take risks for commensurate returns, society needs nancial intermediaries to provide an uninterrupted supply of credit and capital. Failing to do so has far-reaching economic consequences with long recovery times, so that society is primarily concerned with nancial stability. Therefore, societal objectives can be distinguished from shareholder objectives by separating downside risk from upside risk, respectively.
Downside risk is calculated by realized value-at-risk, expected shortfall, and marginal expected shortfall statistics. Each of these statistics incorporates skewed and fat-tailed return distributions that characterize crisis periods, and includes the risk of o-balance sheet items that have played a key role in the crisis.
Value-at-risk (VaR) (Guldimann et al., 1994) represents the loss that could occur over a given period of time with a given probability. Since it is the main statistic used in external reports, internal audit ratings, and regulatory reporting, it is an ex post assessment of an institution's willingness to absorb losses. Expected shortfall (ES) (Artzner et al., 1999) represents the average capital loss when losses exceed the VaR threshold.
Therefore, it summarizes the risk of loss in case of excessive, rare events like the mortgage crisis starting in 2007. Marginal expected shortfall (MES) (Tasche, 2002; Yamai and Yoshiba, 2005) represents the expected loss when losses of industry peers exceed VaR, measuring exposure to severe industry problems. This feature was important during the credit and liquidity freezes in 2008.
Risk incentives are associated with increased VaR, ES, and MES in subsequent years. A one standard deviation increase in risk incentives is associated with a subsequent annualized increase of 4.8% in VaR, 5.8%
in ES, and 5.4% in MES. While this eect is partially oset by price incentives inducing managers to avoid risk (Smith and Stulz, 1985; Guay, 1999), the impact from risk incentives is more signicant, explains more variation in downside risk, and is 1.7% to 2.3% greater than the impact from price incentives. Therefore,
equity compensation had a net destabilizing eect on nancial institutions after 2006. This is not in line with societal objectives. Additional evidence is provided by linking incentives to alternative enterprise-wide measures of future downside risk. Stronger risk incentives also increase nancial rms' implied put volatility in subsequent years, as well as the future probability of nancial distress. Furthermore, I demonstrate that the positive link between incentives and downside risk is robust to a range of alternative specications. Given that risk and price incentives eectively address the conict of interest between managers and shareholders, these results indicate the presence of a risk-shifting problem between shareholders and society. Interestingly, they also indicate that the strengthening of shareholder governance, as promoted by both Acts, increases the risks that the Acts seek to prevent.
Finally, leverage plays a prominent role in the link between stock-based incentives and downside risk.
To begin with, prior research has documented the important role of leverage policy in the crisis (Kashyap et al., 2008; Acharya and Schabl, 2009), and leverage is an important control variable. In addition, many studies have researched the eect of leverage on managerial equity compensation (John and John, 1993 and studies cited therein; Guay, 1999; John et al., 2007), treating leverage as exogenously given. I re-examine this relationship for nancial institutions, who adjust leverage in response to changes in asset prices (Adrian and Shin, 2008; Brunnermeier, 2009; He et al., 2010). Consequently, leverage should be considered an endogenous managerial choice variable for these rms. If leverage is endogenous and managerial incentives are a signicant determinant of downside risk, I expect risk incentives to increase downside risk directly, as well as indirectly through leverage policy. To investigate this, I estimate the direct and indirect eect simultaneously, allowing both eects to vary together.
The ndings below demonstrate that the losses incurred from risk incentives followed over and above the losses incurred from leverage. It follows that compensation made a distinctly important contribution to the nancial crisis. The simultaneous estimates identify leverage as a signicant instrument for managers to increase risk, which is positively aected by risk incentives and negatively by price incentives. Estimates of the indirect eect reveal that risk incentives increased downside risk by inducing managers to implement more aggressive leverage policy.
The remainder of this paper is organized as follows. Section 2 describes the data collection procedure and the calculation of variables that are used in the analysis. Section 3 describes the association between incentives, nancial leverage and downside risk, presents the empirical framework quantifying this relation, and addresses measurement issues that arise when doing so. Section 4 motivates several testable expecta- tions about the interface between incentives, leverage, and downside risk. It discusses the empirical results regarding risk incentives and shareholder objectives in Section 4.1, and risk incentives and public objectives in Section 4.2. I calculate the total economic eect of risk incentives and price incentives in Section 4.3
to determine whether the results on risk incentives extend to equity compensation in general. Section 4.4 introduces the role of endogenous leverage eects in the relationship between risk incentives and downside risk. Section 4.5 presents additional evidence regarding risk incentives and downside risk, and discusses a range of robustness checks. I conclude in Section 5.
2 Data and variables
The sample covers a period from the end of 2006 to mid-2010, as it is from 2006 that the Securities and Exchange Commission (SEC) requires disclosure of detailed information regarding current and previous option and equity grants. Since specic concerns exist about the incentives of non-CEO executives in nancial rms (e.g., the Federal Reserve Bank's Press Release of October 22, 2009: Fahlenbrach and Stulz, 2011), I collect information on all available executives of nancial rms (i.e. rms with a primary SIC code from 6000 to 6999) for which a valid COMPUSTAT/Bloomberg/ExecuComp match exists. The nal sample covers a total of 1,640 individual executives within 229 nancial institutions over four years. Not all executives have been employed for the full sample period, others have switched employers, and some institutions have disappeared from the sample, so that the sample includes 4,213 valid executive-year observations.
As discussed in Section 2.1 below, I construct value-at-risk, expected shortfall, and marginal expected shortfall using pricing data from Bloomberg. In Section 2.2 below, I calculate risk incentives and perfor- mance incentives using the volatility of Bloomberg daily stock prices, zero-coupon interest rates from Ivy DB/Option Metrics, and several data items from COMPUSTAT and ExecuComp. To ensure that all stock price information includes relevant information from nancial statements, price information for scal year t is based on data from July of year t to June of year t+ 1 (choosing dierent months does not change the results). Since more than 95% of all scal years end in December, this conforms to the ExecuComp scal year denition and excludes potential price reactions to the publication of compensation reports and nancial statements. Thus, ExecuComp data on stock and option series cover the period from December 31, 2006 to December 31, 2009, and Bloomberg price data cover the period from July 1, 2007 to June 30, 2010.
2.1 Downside risk
To investigate how shareholders' desire to increase upside risk relates to legislators' purpose to limit downside risk, an appropriate risk measure should distinguish between upside and downside risk or, equivalently, gains and losses. In addition, since the empirical distribution of stock returns from 2007 to 2010 is skewed and has fat tails, it is desirable that a risk measure doesn't assume normality and can be estimated non-parametrically.
Furthermore, the measure should minimize the role of managerial discretion and account for o-balance
sheet items. To the extent that nancial companies use o-balance sheet items, many important nancial performance measures are likely to be distorted (Altman, 2000). These items include the structured nance instruments that played a key role during the 2007-2008 crisis including asset-backed securities, mortgage- backed securities, and many credit derivative products.
These requirements limit the use of previously employed measures that analyze executive compensation, such as stock price volatility as in Saunders et al. (1990), DeFusco et al. (1990), and Cohen et al. (2000) (that doesn't distinguish between upside risk and downside risk), distance-to-default as in Barth and Levine (2001), Sundaram and Yermack (2007), and Beltratti et al. (2010) (that assumes normality) and the Roy (1952) z-score as in Laeven and Levine (2009) (that relies on balance sheet items).
Instead, I represent the existence and signicance of a risk-return trade-o by an institution's value- at-risk (VaR), expected shortfall (ES), and marginal expected shortfall (MES). Examining the left tail of the return distribution, these techniques have been specically designed for measuring nancial market risk.
Given a probability levelα, each statistic describes a dierent aspect of downside risk. VaR resembles the maximum loss in the majority of events, ES the average loss for extremely negative events, and MES the average loss during extremely negative events in the nancial sector as a whole.
The value that is at risk can be interpreted as a threshold value, such that the probability of the mark- to-market loss exceeding this value within a given time frame is α (Jorion, 2007). For example, if a bank has a one-day 95% VaR of 0.08, there is anα= 0.05probability that the bank's equity will fall in value by more than 8% over a one day period. VaR (Guldimann et al., 1994) is dened as the maximum (rm-wide) loss in100(1−α)% of the time:
VaR1−αit (Rit) =−sup{z|Pr [Rit< z]< α},
in whichRit is rmi's return at timet, andz is a percentile corresponding to the pre-specied parameter α. Because I am calculating risk ex post, it is straightforward to obtain100(1−α)%daily VaR by selecting the lowest100α% of daily observations for each rm in a given scal year. Value-at-risk is then the largest (i.e., least negative) of these observations.1
To get a better impression of the return distribution when losses exceedVaR1−α(Rt), I also measure the expected loss for the worst100α%of the cases, as expected shortfall (ES)(Artzner et al., 1999). While VaR focuses on the maximum loss near the center of the return distribution, ES provides information about losses
1Previous studies nd very similar results between the non-parametric denition of VaR above and more elaborate, parametric denitions (Bali et al., 2009).
given a rare event by describing the mean of the left tail of the return distribution:
ESαit(Rit) =−E
Rit|Rit≤VaR1−αit (Rit) .
This denition describes the mean return from the 5% of observations that are excluded to calculate 95%
daily VaR, and can be interpreted as the average loss suered in the worst100α%of the time.
To get a sense of how nancial institutions are exposed to problems at their industry peers, I calculate marginal expected shortfall (Tasche, 2002; Yamai and Yoshiba, 2005) by dening R∗t as the return on the MSCI USA/FINANCE index (MXUS0FN), obtained from Bloomberg. I then follow Acharya et al. (2010) and calculate marginal expected shortfall as follows:
MESαit(Rit) =−E
Rit|R∗t ≤VaR1−αt (R∗t) .
In words, marginal expected shortfall measures the sensitivity of overall exposure to losses from other nancial rms by calculating the expected shortfall of a stock, given that the industry return is below its 100α-th percentile. Since marginal expected shortfall captures the exposure to negative externalities from other nancial institutions, it captures risk spillovers and an institution's vulnerability to the spreading of losses.
In their Revisions to the Basel II market risk framework, the Basel Committee on Banking Supervision requires that In calculating value-at-risk, an instantaneous price shock equivalent to a 10-day movement in prices is to be used, i.e. the minimum holding period will be ten trading days. Banks may use value-at-risk numbers calculated according to shorter holding periods scaled up to ten days by the square root of time (Basel Committee on Banking Supervision, 2009). I follow the same standards and measure VaR, ES, and MES using daily returns. To be compared with Basel regulations, the VaR statistic should be multiplied by
√10so that estimates for the 10-day horizon are obtained. Clearly, incentive eects on VaR are proportional to changes in the buy-and-hold horizon (N), amplifying VaR, ES, and MES coecients by a factor √
N. Finally, since all returns in the left tail are negative, I multiply VaR, ES, and MES with minus one in the equations above. This facilitates interpretation with a positive coecient indicating a positive eect on risk, leading to the central result of a positive intertemporal eect of risk incentives on downside risk.
2.2 Risk incentives
The main explanatory variable in this article represents risk incentives and is calculated in four steps.
First, vega equals the Black and Scholes (1973) derivative of option value with respect to volatility (vega), calculated as in Guay (1999). Second, for an executive in a given year, vega is expressed in dollar terms by
multiplying by the number of options awarded, equal to the total number of shares that underlie exercisable, unexercisable, or unearned options. Awarded stock is assumed to have a vega of zero, and equals the number of (unearned or unvested) shares, plus those that are owned or have been awarded through an equity incentive plan. No adjustments are made for vesting, illiquidity, etc. when calculating risk incentives. Restricted stock, as well as unexercised and unearned options, are treated as if owned unconditionally. Third, for each executive in a given year, total risk incentives are calculated by summing dollar-expressed vega, weighted by the (stock or option) award's relative contribution to the executive's total equity compensation. Finally, risk incentives are scaled by the number of common shares outstanding, so that incentives are expressed on a dollar-per-share basis (Jensen and Murphy, 1990; Yermack, 1995).2
I obtain information for the last three steps directly from ExecuComp and COMPUSTAT. To complete the rst step and calculate vega, several additional data items are needed to estimate the dividend yield, rm volatility, and the risk-free rate. The dividend yield is calculated from COMPUSTAT data. Annualized daily volatility would lead to erratic option values due to feverish sentiment in the nancial sector during 2008, with 10% of observations having a volatility of 90% or more and some volatilities as high as 420%.
This would overstate my results, so that estimates for volatility are based on weekly historical volatility, obtained from Bloomberg pricing data. I obtain estimates for the risk-free rate by an approximation of the continuously compounded zero-coupon interest rate, obtained through Option Metrics. This interest rate is calculated from a collection of continuously compounded zero-coupon interest rates at various maturities, derived from LIBOR rates and settlement prices of CME Eurodollar futures. Risk-free rates are averaged over three-month periods, and option awards with particular option estimation/expiration quarters (e.g., 2007Q42014Q3) are matched to the average rate for corresponding interest estimation/maturity quarters.
2.3 Control variables
Since balance sheets of nancial institutions are continuously marked to market, I use market leverage (rather than book leverage) as the measure for leverage policy. Market leverage equals the sum of current debt (COMPUSTAT data item code: DLC) and long term debt (item code: DLTT), divided by the quasi- market value of assets. The quasi-market value of assets equals total assets (item code: AT) plus equity market value, minus equity book value (item code: CEQ).
It is well known that large value stocks bear lower risk than small growth stocks. Market-to-book is equity market value (item code PRCC_F times item code CSHO), divided by equity book value (item code:
2Similar results are found for the relation between downside risk and other common denitions of risk incentives, such as vega (e.g., Guay, 1999) and total dollar-valued risk incentives (e.g., Fahlenbrach and Stulz, 2011). While it could also be argued that short-term compensation increases downside risk by inducing managers to make myopic policy choices, I do not nd evidence for this proposition.
CEQ). I proxy for size using the log of market value. Other determinants that contribute positively to rm risk are leverage, annual returns, and operating performance. These controls are included in each model below, and calculated conventionally using COMPUSTAT items. Operating performance is calculated by net income before extraordinary items and discontinued operations (item code: NI), divided by total assets (item code: AT). Yearly returns are log-dierenced closing prices at scal year end.
I include xed eects in the risk and leverage equations to prevent unobserved heterogeneity that can jointly aect leverage and risk incentives. One specic source of such heterogeneity is return variance since it is likely that risk incentives and downside risk are both aected by uncertainty. For instance, a change in variance will also cause a change in downside risk, as both describe the same return distribution. At the same time, prices usually fall when volatility increases, which entails lower vega. Therefore, lagged return variance is included.
Finally, both risk incentives and price incentives aect risk taking behavior, as the sensitivity of executive wealth to share prices may change through the eect of risk-taking on value. Therefore, price incentives need to be controlled for. I calculate price incentives following the exact same procedure used to calculate risk incentives, outlined in the previous subsection. However, price incentives are based on stock and option delta (not vega). Delta equals the Black and Scholes (1973) derivative of option value with respect to share price, and is assumed to be equal to one for awarded stock.
2.4 Preliminary results
All the variables that are used in the analyses below are summarized in Table 1. From the compensation measures, it can be seen that incentives for nancial executives are stronger when compared to non-nancial executives in previous work. Total deltas are substantially higher than the average stock delta of 0.6, and price incentives are larger than as in Yermack (1995). Generally speaking, executives in the sample receive more equity-based compensation than in comparable studies of non-nancial rms. This is in line with the recent ndings of Core and Guay (2010) who nd that, compared to non-nancial rms, CEO risk-taking incentives are larger for the 24 largest U.S. banks.
Anecdotally, the sample contains some very large values for vega. Eleven executives in the sample have a vega larger than one. Ten of these executives work at CME Group, Inc., a derivatives market place, and one at Goldman Sachs Group, Inc. The top 23 positions are all held by executives from both institutions.
However, this pattern disappears once incentives are measured on a dollar-per-share basis.
The explanatory variables provide an impression of the problems in the sector during the crisis. Annual buy-and-hold returns over the sample period are negative and widely dispersed. However, net income (as
a fraction of total assets) is positive, on average, and much less volatile hinting at much of the action (i.e., losses) occurring o-balance sheet. The amount of debt as a fraction of asset value ranges from zero (for 2.75% of observations) to more than 90% (for 1.92% of observations), suggesting that some nancial institutions have solvency problems. Finally, leverage growth ranges from -46% to +62% indicating that substantial nancial restructuring takes place from one year to another.
[Insert Table 1 about here]
Some of the key ndings can be illustrated by the data displayed in Figure 1, where I plot rm-wide risk incentives at the end of 2006 against the time-series average of expected shortfall over the 2007-2009 period.
As can be seen from the upward-sloping tted regression line, the relationship is clearly positive suggesting that risk incentives increased losses in subsequent years. The graph also indicates that most delisted rms (identied as such by the Center for Research in Security Prices and marked by a +) had risk incentives above the sample median. While the analysis below goes beyond the bivariate relations in the graph, it provides a rst indication that nancial rms with stronger risk incentives suered greater losses during the crisis.
[Insert Figure 1 about here]
3 Empirical strategy
In this study, I argue that risk incentives increase downside risk in subsequent years, and that the increase partially follows from risk incentives inducing managers to implement more aggressive leverage policy. To x ideas, consider a nancial institutioniwhose shareholders reward manager j with a compensation contract in year t that involves the rm's equity. Equity compensation includes executive stock options, restricted stock, or both. The rm decides how much risk incentives (mjt) will appropriately incentify the manager.
Faced with this contract, managers choose to implement a particular leverage policy (yit). Bothyitandmjt
aect share price dynamics that determine a nancial institution's exposure to downside risk (rit). Ignoring subscripts, I am interested in the following relation:
dr dm = ∂r
∂m+∂r
∂y
∂y
∂m. (1)
The eects of leverage and compensation in Eq. (1) have previously been investigated in isolation, thereby requiring all other factors to remain constant as each eect varies. However, compensation aects rm performance primarily through the actions of managers (including leverage) that can vary together if they are jointly determined. Therefore, I determine the equilibrium response to a change in incentives, as well as the total change in downside risk as a function of incentives and leverage policy.
The rst right-hand side term in Eq. (1) is interpreted as the direct eect of risk incentives on downside risk, and can be estimated using the following model:
rit=β0+β1mjt−1+β2yit−1+β3Xit,t−1+β41crisis+β5Fi+ε1jt, (2)
in which βk, k = 0, ...,5 are coecient scalars or vectors; rit represents one of the downside risk statistics;
mjt−1represents risk incentives;yit−1 represents leverage;Xit,t−1 includes the control variables (last year's price incentives, current size, current market-to-book, last year's share price performance, last year's net income, and last year's return variance);1crisisrepresents a dummy variable equal to one if the current year equals the crisis years 2007 or 2008, and zero otherwise; andFi captures rm-level xed eects controlling for unobserved heterogeneity. I address causality issues by regressing the downside risk statistics on lagged values of compensation and leverage policy, and several lagged control variables.
The model in Eq. (2) is used to investigate ∂r/∂min isolation. To estimate both right-hand side terms in Eq. (1), two adjustments are made to this model. First, I focus on CEOs, or CFOs and CFOs, since only the very top executives decide on leverage policy. In addition, I include leverage growth (β6∆yit−1) as an additional variable in Eq. (2) to evaluate endogenous leverage decisions. I then re-estimate the model in conjunction with an additional equation that captures the eect of compensation on leverage growth,
∆yit−1=γ0+γ1mjt−1+γ2Dit−1+γ31crisis+γ4Fi+ε2jt−1, (3)
where ∆yit−1 represents leverage in rst dierences, γk, k = 0, ...,5 are coecient scalars or vectors,mjt−1
represents risk incentives,1crisis represents the crisis dummy, andFi captures rm-level xed eects ensur- ing that changes in leverage depend on within-variation (i.e., on year-to-year changes in the explanatory variables). The set of control variablesDit−1 is based on Adrian and Shin (2008) who examine endogenous
leverage policy in a long time series of ve investment banks, and includes total assets, lagged leverage (yit−2), and lagged downside risk (rit−2).
The rst right-hand side term in Eq. (1) is estimated by βˆ1 in the augmented Eq. (2). The second right-hand side term in Eq. (1) is estimated by a combination of simultaneously estimated coecients, βˆ6·ˆγ1. This term can be interpreted as the indirect eect of risk incentives on downside risk through the rm's leverage policy.3 The total changedr/dmis thus obtained through two xed-eects regressions whose covariance matrix is estimated simultaneously. Covariances between the error termsε1andε2are potentially non-zero allowing the coecients to be correlated within equations and across equations. This allows the two eects in Eq. (1) to vary together.
Some institutions are more heavily weighed than others since the number of executives diers between nancial institutions, and heterogeneous observed characteristics may lead to within-rm dependence of standard errors. While this will continue to give unbiased coecient estimates, the resulting covariance matrix is no longer consistent. I address this by computing Froot (1989) clustered standard errors in both single equation and multiple equation models, with clusters at the rm level.
Signicance of the product βˆ6·ˆγ1 is based on three dierent critical regions. First, standard errors are obtained from the distribution of a rst-order Taylor series approximation (Greene, 2003) using a result known as the delta method. The standard errors obtained through the delta method control for heterogeneity by rm clustering, and are correct under the assumption of normally distributed estimates of the indirect eect.
Second, since this assumption might be violated, I also report bootstrapped standard errors obtained from sampling observations from the data (with replacement). The bootstrap calculates standard errors from the distribution of βˆ6·γˆ1 after re-estimating the system of equations 1,000 times. Finally, since a product of coecients could be skewed and kurtotic, condence intervals may be asymmetric. Therefore, I also report bias-corrected condence intervals ofβˆ6·γˆ1as described in Efron and Tibshirani (1993). If the lower bound of the interval is above zero, one can consider the condence interval statistically dierent from zero (at the specied signicance level).
4 Risk incentives and the U.S. nancial sector
In this section, I motivate the choice of leverage and risk incentives as two important, interconnected mech- anisms aecting the 2007-2008 turmoil in the nancial sector. Essentially, I argue that managers in nancial
3Since Eq. (3) includes compensation and leverage, both lagged one year relative to downside risk, it could be argued that compensation is determined by leverage policy instead of the reverse. This is not controlled for, because a contemporaneous leverage eect on compensation will bias results against the hypotheses. For instance, leverage increases costs of nancial distress for shareholders so that relatively few risk incentives are optimal in more levered rms (John and John, 1993). Empirically, Guay (1999) reports a negative eect of leverage on vega and nds that growth rms (which have lower debt) provide CEOs with higher vega.
institutions take on risk by the direct eect of risk incentives, as well as through the indirect eect of risk incentives on leverage policy. Such risk-taking serves the interests of shareholders, but not those of society.
Section 4.1 demonstrates that risk incentives are benecial from a shareholder's point of view. Therefore, no evidence is found regarding the standard agency problem between managers and shareholders. However, Section 4.2 demonstrates that risk incentives have a destabilizing eect on the nancial sector. This indicates the presence of a risk-shifting problem between shareholders and society. In Section 4.3, I nd that the risk- seeking eect from risk incentives is larger and more signicant than the risk-avoiding eect from price incentives, leading to a net positive association between stock-based compensation and downside risk in subsequent years. Section 4.4 demonstrates that the incentives eect is distinctly important from leverage eects, and that leverage is endogenous. Risk incentives jointly determine leverage and downside risk, and risk incentives signicantly increase downside risk by inducing managers to implement more aggressive leverage policy. Section 4.5 provides additional evidence that risk incentives increase downside risk in subsequent years, and presents a range of robustness checks.
4.1 Risk incentives and shareholder objectives
In the nancial sector, plotting the industry's evolution of mean annual returns from 2006 to 2009 would result in a pronounced V -shape centered around 2008. This is exemplied by the course of events at Goldman Sachs, which sold part of its equity in October 2008 to the U.S. Treasury for $10 billion as part of the Troubled Asset Relief Program (TARP). However, around mid-2009, Goldman Sachs had repurchased the shares, announced the highest quarterly prot in the bank's 140-year history, and reserved $11.4 billion toward employee compensation that year.
Therefore, a natural point of departure is an update of the study by Fahlenbrach and Stulz (2011). For CEOs of 98 banks, Fahlenbrach and Stulz (2011) link several compensation measures, measured in 2006, to one and a half-year buy-and-hold returns from July 2007 to December 2008. They conclude that returns during the crisis are unrelated to equity compensation, with weak evidence regarding risk incentives that disappears once they include control variables.
I repeat their analysis using more controls, more rms and executives, and more recent data by including the year 2009 and the rst half of 2010. Table 2 presents the results after regressing one-year, two-year, and three-year buy-and-hold returns on 2006 incentives. I also regress one-year and two-year buy-and-hold returns on 2007 incentives, and one-year buy-and-hold returns on 2008 incentives. Explanatory variables include price incentives, stock returns, market-to-book, and market value, similar to the model in Fahlenbrach and Stulz (2011). For consistency with the analysis on downside risk below, I also control for leverage, net
income, return variance, rm xed eects, and a 2007-2008 crisis dummy.
[Insert Table 2 about here]
In Table 2, the results are very similar to Fahlenbrach and Stulz (2011) when covering their sample period.
One-year and two-year buy-and-hold returns for 2006 incentives are insignicant, and so are one-year buy- and-hold returns for 2007 incentives. However, a dierent picture emerges when post-2008 data is included, and risk incentives are positively related to stock returns held until June 2010. Not surprisingly, this result becomes stronger as the crisis progresses. If returns are held until June 2010, stronger risk incentives lead to higher returns for stocks purchased in mid-2007, mid-2008, and mid-2009 at the 6.3%, 5.4%, and 3.2%
signicance level, respectively.
Table 2 reveals that executives focused on the interests of their shareholders before and during the crisis. This is evidence against the standard manager-shareholder agency problem that is addressed by current legislation, and supports the view of Fahlenbrach and Stulz (2011) that managers took actions that they believed the market would welcome. This view is consistent with theories that shareholders choose compensation contracts in order to implement second-best policies that increase equity returns. This view is further reinforced by Carpenter et al. (2011), who observe that shareholders of nancial institutions have suered most from the crisis, but permitted executives to pay back TARP money that imposed regulatory constraints on compensation.
4.2 Risk incentives and public objectives
Maximizing returns does not necessarily concur with the objectives of society, which needs nancial institu- tions to provide an uninterrupted supply of credit and capital. As a consequence, societal concerns are about limiting (or minimizing) downside risk in the nancial sector. This subsection looks at risk and contagion aspects by examining the eect of risk incentives on value-at-risk, expected shortfall, and marginal expected shortfall. Each statistic describes a dierent aspect of downside risk. VaR resembles the maximum loss in the majority of events, ES the average loss when extremely negative events occur, and MES the average loss when extremely negative events occur at industry peers. The results are presented in Table 3.
[Insert Table 3 about here]
Most risk control variables in Table 3 have an expected eect on VaR, ES, and MES. The sign of market-to- book is always negative and signicant. This makes sense since a decrease in market value is associated with both a lower ratio and more risk. Price incentives have a negative impact on downside risk indicating that price incentives induce managers to avoid risk by exposing them to more idiosyncratic uncertainty. The size coecient is negative and signicant indicating that larger nancial institutions have more stable cash ows and less risk exposure. Past return variance has a signicantly positive eect on downside risk in subsequent years. Last year's return is not signicantly dierent from zero implying that good recent stock market performance does not indicate higher or lower losses. Interestingly, higher net income leads to higher losses in subsequent years suggesting that large accounting prots are potentially misleading. It can also be seen that theR2statistics are quite high.
Not surprisingly, the 2007-2008 crisis years are associated with higher losses. When all controls are included, the crisis dummy increases downside risk with 7% to 10%. In fact, a simple regression of 95%
daily VaR, expected shortfall and marginal expected shortfall on a constant and the crisis dummy explains about 20% of the sample variation between the rms in Table 3 indicating that nancial rms experienced an extremely strong aggregate shock.
4.2.1 Risk incentives and risk tolerance
Much literature exists regarding the ex ante eect of compensation on risk-taking (Haugen and Senbet, 1981;
Tufano, 1996; Guay, 1999; Knopf et al., 2002; Rajgopal and Shevlin, 2002; Coles et al., 2006; Williams and Rao, 2006; Sundaram and Yermack, 2007; Edmans and Gabaix, 2010). For instance, Knopf et al. (2002) argue that vega should increase a manager's appetite for risk, and show that managers with higher vegas tend to hedge less. In the oil and gas industry, Rajgopal and Shevlin (2002) nd that higher vega is related to greater exploration risk and less risk hedging. For non-nancial rms, Coles et al. (2006) nd a positive causal link between CEO compensation and a wide range of investment decisions. Translating these ndings to the nancial industry, I expect that risk incentives increase an institution's willingness to absorb losses, represented by higher value-at-risk: ∂rVaRt /∂mt−1>0 .
Because VaR focuses on the maximum loss in 95% of the cases, or the largest likely loss, it is a useful quantity for corporate control. Consequently, VaR is disclosed by nancial institutions in external reports, the main statistic employed as an internal control standard for audit ratings or self-assessment, and required by law in regulatory reporting. Therefore, while nancial rms have some discretion in calculating VaR and
use ex ante calculations of expected VaR, realized VaR is an ex post measure of a nancial institution's willingness to absorb losses.
From the rst four data columns in Table 3, I observe that if risk incentives increase, VaR is signicantly higher in subsequent years. Considering the denition of VaR, these results suggest that risk incentives have increased nancial institutions' realized loss threshold. To the extent that the ex post VaR statistic concurs with ex ante VaR (this number is not reported publicly for all rms and could be subject to dierences in estimation methodology), increased VaR is consistent with more lenient internal, external and regulatory risk governance. This evidence suggests that risk incentives lead to more risky policy choices and increased the amount of losses that nancial institutions have been willing to absorb.
4.2.2 Risk incentives and extreme shareholder loss
Consistent with a large literature on the delta (Jensen and Murphy, 1990) and vega (Guay, 1999) of executive compensation contracts, it is generally assumed that shareholders implement compensation policies that have a positive eect on rm performance and rm risk. Focusing on the latter, Guay (1999), Cohen et al.
(2000), Hanlon et al. (2004), and Coles et al. (2006) all nd a positive link between vega and rm volatility.
However, while it is often assumed that increased vega leads to policies that maximize shareholder value (good volatility leading to upside risk), increased volatility may also indicate negative price movements (bad volatility leading to downside risk). Either way, CEO wealth increases. It follows that nancial rms' recent drop in market value may also be positively linked to vega, and I expect risk-inducing compensation to have exacerbated losses during the nancial crisis: ∂rESt /∂mt−1>0.
Actual losses are not well measured by VaR, which represents the largest likely loss. Therefore, it is fully uninformative about the size of the actual loss in case an extreme, unlikely event occurs. For this reason, I estimate the eect of risk incentives on expected shortfall (ES), which measures the expected loss for the worst 5% of the cases (i.e., when losses exceed VaR).
Data Columns 5 to 8 of Table 3 report the impact of leverage and risk incentives on next year's ES. As with VaR, ES for the worst 5% of the scal year is signicantly higher for nancial institutions with stronger risk incentives. Similar to VaR, I nd that compensation and leverage are distinctly important in their eect on ES. While these losses may be compensated for by the upper 95% of returns (and shareholders may still be willing to accept these losses), they clearly conict with the public interest. Financial institutions with executives receiving more risk incentives suered larger losses once the rm is doing poorly.
4.2.3 Risk incentives and loss spillovers
ES represents losses when a nancial rm itself is in a bad condition. However, an institution may also be increasingly exposed to distressed peers, or stress within the nancial sector as a whole. In an important paper, Shleifer and Vishny (1992) demonstrate that losses might also occur because other institutions face similar constraints at the same time. This is relevant to the current crisis, as many banks could no longer roll over their short-term debt in 2008. This was a direct consequence of sector-wide increased margin and collateral requirements, and a general tightening of lending (Brunnermeier and Pedersen, 2008). Given the interconnectedness of the nancial sector, I expect risk incentives to exacerbate losses beyond the boundaries of the rm and to precipitate loss spillovers to peer institutions: ∂rMESt /∂mt−1>0.
While expected shortfall measures the expected loss when a rm's return distribution is in its left tail, marginal expected shortfall resembles the expected loss when the value-weighted return distribution of the aggregate nancial sector is in its left tail. Thus, MES measures the exposure to negative externalities from other nancial institutions, and captures risk spillovers and an institution's vulnerability to the spreading of losses.
The rightmost four data columns of Table 3 present results on MES that concur with those on VaR and ES. Stronger risk incentives are associated with larger MES in subsequent years, and risk incentives increase MES over and above the increased risk from leverage. This result complements previous work describing leverage as an important cause of the crisis (Adrian and Shin, 2008; Kashyap et al., 2008; Brunnermeier, 2009) by showing that risk incentives made a distinctly important contribution to the nancial crisis. Since MES measures the average loss of a nancial rm when other institutions are under distress, these results suggest that risk incentives lead to increased risk of loss spillovers in the next year, and that risk incentives have increased nancial institutions' exposure to trouble elsewhere in the nancial sector.
The results on VaR, ES, and MES indicate that downside risk of nancial rms is higher when risk incentives are better aligned with shareholder interests. This does not necessarily conict with shareholder interests, who are not concerned with risk per se but with the trade-o between risk and return. Societal objectives, however, are more about nancial stability (i.e., low risk) than wealth accumulation (i.e., positive returns). Having established that managers are well aligned with shareholders, the results in Table 3 indicate the presence of a conict of interest between shareholders and society.
4.3 The economic eect of equity compensation
The analysis above indicates that risk incentives induced managers to take risk during the nancial crisis, and lead to more risk tolerance, higher losses, and greater negative spillovers. However, economic theory
predicts an opposite eect for managers' sensitivity of wealth to share price performance (price incentives).
Managers with price incentives share gains and losses with shareholders, encouraging them to work harder and more eectively. At the same time, price incentives expose managers to more idiosyncratic risk, inducing them to avoid risk (Smith and Stulz, 1985; Guay, 1999).
The negative eect of price incentives on downside risk is conrmed in Table 3, so that the impact of risk incentives on total equity compensation (i.e., risk incentives and price incentives) is not immediately clear. Therefore, I calculate the net economic impact of the two counteracting forces by examining how a one standard deviation increase in risk incentives (σrisk = 0.06%) and price incentives (σprice = 0.43%) aects downside risk and returns in subsequent years.
From the fourth data column in Table 3, one can see that a one standard deviation increase in risk incentives is associated with an annualized increase in value-at-risk of √
250(0.0006×5.080) = 4.77% in subsequent years. A one standard deviation increase in price incentives is associated with an annualized decrease in value-at-risk of √
250(0.0043×0.467) = 3.12%. It follows that equity compensation has a positive net eect on value-at-risk of 1.7% per annum. Repeating the calculations for Columns 8 and 12 in Table 3, I nd that equity compensation also had a positive net eect on expected shortfall and marginal expected shortfall of 2.3% and 2.1%, respectively.
The eect of risk incentives accounts for 7.5% of the sample-wide standard deviation of value-at-risk:
(0.0006×5.080)/0.040 = 0.075σV aR. Price incentives, on the other hand, explain about 5% of the sample- wide standard deviation of value-at-risk. Similarly, risk incentives explain 7% of the variation in ES and 9% of the variation in MES, while price incentives explain only 4% of the variation in ES and 5% of the variation in MES. This demonstrates that variation in risk incentives explains a larger portion of downside risk than variation in price incentives. I also note that, compared to price incentives that are signicant at the 1% signicance level, the eect of price incentives on ES and MES is less strong, with signicance at the 5% level.
Ideally, one would also calculate the existence and signicance of a risk-return trade-o. Performing similar calculations on Table 2, a one standard deviation increase in risk incentives is associated with an annualized increased return of 3.3%, 1.3%, or 4.4% if the stock was bought in mid-2007, mid-2008, or mid- 2009, respectively, and held until mid-2010. while the eect of price incentives is not distinguishable from zero for any interval. However, I should be careful in quantitatively comparing the eects of risk incentives on returns in Table 2 with the eects on downside risk in Table 3. Downside risk regressions include rm-level xed eects and are calculated using annualized daily returns, while return regressions include SIC-level xed eects and are calculated using multi-year buy-and-hold returns. Nevertheless, it is clear that the aggravated losses from risk incentives in Table 3 are at least partially compensated for by positive returns in Table 2.
The risk-avoiding eect of price incentives is smaller, has less explanatory power, and is less signicant than the risk-seeking eect of risk incentives, leading to a net positive intertemporal link between equity compensation and downside risk. Therefore, the evidence on destabilizing risk incentives, described in Sections 4.1 and 4.2, extends to equity compensation in general. Considering these results, it can be argued that shareholder supervision is not the appropriate tool to monitor the risk of nancial institutions, and that aligning the interests of executives in nancial institutions with those of their shareholders has been counterproductive for society.
4.4 Risk incentives, downside risk, and endogenous leverage policy
Having established that managers seek risks for the benet of shareholders and at the cost of nancial stabil- ity, I turn to the question of how these managers take these risks. One obvious way for nancial institutions to increase risk is by increasing leverage. This can be easily seen in Table 3, where high leverage leads to bigger downside risks and loss spillovers (resembled by increased VaR, expected shortfall and marginal ex- pected shortfall) in subsequent years. This result is consistent with conventional corporate nancing theory and conrms that increased leverage has led to increased downside risk: ∂rt/∂yt−1>0.
However, for nancial institutions, leverage decisions may be endogenous and causation reversed. This has been well-documented for the 2007-8 crisis (Fostel and Geanakoplos, 2008; Kashyap et al., 2008; Brunner- meier, 2009; He et al., 2010). Financial institutions have highly levered balance sheets and generally borrow short-term, using their assets as collateral. When these institutions' asset value eroded in 2007 and 2008, leverage ratio requirements were no longer met and rms needed to de-leverage their positions by selling part of their assets. These forced sales occurred when the prices of these assets were low, depressing prices further and leading to an even lower value of their assets (Brunnermeier, 2009). Consequently, this raised concerns about the solvency and liquidity of the banking system, and margin and collateral requirements were increased (Brunnermeier and Pedersen, 2008). In turn, due to these tightened lending standards, nancial institutions could no longer roll over their short-term debt, leading to further assets sales and de-leveraging.
I expect that if leverage is endogenous, as suggested by this series of events, managers implement a more (less) aggressive leverage policy when asset prices rise (decline) and downside risks are low (high):
∂yt/∂rt−1<0. Adrian and Shin (2008) document empirical evidence of this idea for ve investment banks.
They nd that leverage growth is positively related to changes in repurchase agreements and to total asset growth, and negatively related to changes in lagged value-at-risk. Their results demonstrate that leverage is indeed endogenous, as well as strongly pro-cyclical.
Importantly, the endogenous relationship between leverage and downside risk can be directly linked to
the compensation literature above. Compensation is the second-best solution to a principal-agent problem, and any signicant eect of compensation on capital losses results from managers implementing certain investment and nancial policies. If leverage is endogenous and increases risk, managers may implement a more aggressive leverage policy as their compensation contracts induce them to do so. Therefore, for nancial rms, I expect that leverage is positively aected by risk incentives: ∂yt/∂mt >0. I also expect that the positive link between risk incentives and downside risk can at least partially be mediated by leverage, and that stronger risk incentives lead to increased downside risk through leverage growth: ∂rt/∂yt−1·∂yt−1/∂mt−1>
0.
[Insert Table 4 about here]
To investigate the extent to which leverage is endogenous and aected by incentives, Table 4 presents estimation results of the system formed by Eqs. (2) and (3). Since not all executives exert inuence on the leverage policy of a rm, all non-CEO and non-CFO executives are purged from the sample. The Breusch and Pagan (1980)χ2-tests reject the null of no signicant correlation between the residuals of the equations, so that cross-model covariances are signicant.4 This supports the assumption that the direct eect and indirect eect in Eq. (1) vary together and are jointly determined, so that simultaneous estimates are necessary.
The results in Table 4 are consistent with the expectations regarding incentives, nancial policy, and downside risk. First, estimates of the downside risk model in Eq. (2) are similar to those in Table 3, with variables having the same sign and similar signicance levels. As previously, rms with stronger risk incentives are exposed to more downside risk in subsequent years (∂rt/∂mt−1>0). Compared to previous results, Table 4 reports estimates that also include leverage in rst dierences, which is more appropriate for an endogenous choice variable. Leverage and leverage growth are both positive and signicantly related to downside risk in subsequent years (∂rt/∂yt−1>0).
Additionally, rm xed eect estimates of the leverage model in Eq. (3) indicate that year-to-year changes in the three downside risk measures have a negative eect on leverage growth in subsequent years. This is consistent with the de-leveraging idea where leverage increases (decreases) when downside risk is low (high)
4Firm-level xed eects control for dierences in the rm's contracting environment and ensure that changes in leverage depend on within-variation (i.e., on year-to-year changes in the explanatory variables). Standard errors are slightly smaller when compared to Table 3, since a generalized least squares-type, full covariance matrix is estimated. In the downside risk model, I include the level of current leverage (not the previous year's leverage) to avoid collinearity issues.
in previous years (∂yt/∂rt−1 <0). Leverage changes most signicantly when problems exist elsewhere in the nancial sector (i.e., for MES). This supports the idea of leverage being pro-cyclical as in Adrian and Shin (2008). Leverage growth is positively related to year-to-year changes in lagged leverage suggesting that rms with high leverage tend to have more aggressive leverage policy.
More importantly, the results in Table 4 indicate that price incentives are associated with negative leverage changes, and risk incentives are associated with positive leverage changes (∂yt/∂mt>0). Signs of the incentives coecients are similar in both equations. Since leverage growth positively aects downside risk, this is a rst hint that leverage is an endogenous managerial choice variable aected by managerial compensation.
[Insert Table 5 about here]
In Table 5, this hypothesis is formally tested. Panel (A) presents estimates of the cross-derivative in Eq. (1) after estimating Eqs. (2) and (3) for CEOs and CFOs, as presented in Table 4. Panel (B) presents results for the subset of CEO executives. The coecients in Table 5 can be calculated using Table 4, by multiplying the eect of leverage growth on downside risk with the eect of risk incentives on leverage growth. Signicance is based on standard errors by the delta method, on normal-based bootstrapped standard errors, and on bias-adjusted bootstrap condence intervals.
Table 5 shows that leverage policy is a signicant channel through which risk incentives increase downside risk. Irrespective of how signicance is determined, for both the CEO and the CEO/CFO subsample, the positive eect of risk incentives on VaR and ES carries signicantly through leverage at the 95% condence level. This conrms that risk incentives signicantly increase downside risk by inducing managers to im- plement more aggressive leverage policy. The mediating leverage eect on MES is only signicant when bias-adjusted condence intervals determine signicance, and only for the CEO/CFO subsample. Therefore, I am cautious not to read too much into the results on loss spillovers in Table 5.
I should note that the indirect eect of risk incentives in Table 5 is small compared to the direct eect of risk incentives in Table 4, with the indirect coecient about 1/10 of the size of the direct coecient. This leaves much of the correlation between risk incentives and downside risk unexplained. Nevertheless, it does not change the result that risk incentives induce executives to signicantly increase VaR and ES through active balance sheet management, measured by leverage growth. Put dierently, active leverage policy is a signicant channel for risk-induced managers to increase rm risk.
4.5 Additional evidence on risk incentives and downside risk
In the analysis above, the VaR, ES, and MES statistics provide information about the left tail of the empirical return distribution. However, since the statistics are solely based on daily equity returns, no information is provided on the total return to equity and debt. For many of these distressed institutions, it can be expected that the market value of debt deviates strongly from its book value, and there may be day-to-day eects on the value of the debt. This complicates any direct debt proxies such as debt book value. However, if risk incentives are an important determinant of enterprise-wide downside risk, I expect a strong link between risk incentives and risk proxies that are not solely based on equity returns.
One such example would be the relation between risk incentives and nancial distress. To investigate this, I estimate a probit model comparable to Table 3. The binary dependent variable equals one if (compared to 2006 year-end market value) nancial institutions lose at least 85% of market value in 2008 or 2009, or if rms stopped trading publicly in 2008 or 2009 due to bankruptcy (in which case losses have accumulated to 100%).5 The variable equals zero otherwise. This cross-sectional variable measures downside risk in terms of a (near) bankruptcy during the crisis period, which is an event relevant to both debt holders and equity holders. The key explanatory variable is a nancial institution's risk incentives. The control variables are similar to those in the models for VaR, ES, and MES: leverage, price incentives, market-to-book, rm size, annual returns, net income, and variance. Because institutions have very limited year-to-year variation in the delisting variable, a panel approach is no longer appropriate. Therefore, rm-level xed eects are replaced by 4-digit SIC industry-level xed eects, and the main explanatory variables are measured at the end of 2006.
Table 6 presents coecient estimates with Froot (1989) clustered standard errors. Consistent with the ndings on VaR, ES, and MES in Table 3, risk incentives in 2006 have a positive and signicant eect on the probability of nancial distress in 2008 or 2009. Risk incentives increase the probability of distress over and above the increased probability of nancial distress that can be attributed to high leverage. The coecient of 2006 price incentives is still negative, but no longer signicant. In contrast, the coecient on risk incentives is positive and signicant, indicating the presence of the risk shifting problem between shareholders and society.
[Insert Table 6 about here]
5For this purpose, the binary variable is set to one if a nancial institution is delisted from its exchange with a delisting code between 500 and 599, obtained from the Center for Research in Security Prices (CRSP) stock data le. In the sample, delisting occurred because of insucient capital, surplus, and/or equity (delisting code 560), at the company's request (code 570), because of bankruptcy (code 574), because the rm does not meet exchange's nancial guidelines for continued listing (code 584), or to protect investors and the public interest (code 585).
If risk incentives contribute to downside risk, I also expect a positive relationship between risk incentives and the implied volatility of put options written on a nancial institution's underlying stock. Implied put volatilities reect enterprise-wide downside risk because put options provide insurance to bondholders and shareholders alike. As concerns about nancial institutions accumulate, a stronger demand for put options increases the price, and hence the implied volatility (see, for instance, Bates (1991)). To investigate this, I regress the mean implied put volatility written on a nancial institution's stock against the explanatory variables in the models for VaR, ES, and MES. The econometric specication is identical to the one used for the models for VaR, ES, and MES in Table 3. Implied put volatilities are obtained from Option Metrics.
Table 7 presents the xed-eects estimates with Froot (1989) clustered standard errors. Compared to Table 3, the sample in Table 7 is smaller as it only includes nancial institutions with tradable put options.
Again, the negative and signicant coecients on risk incentives and leverage evidence are supportive of the main result that risk incentives have a signicantly positive eect on downside risk, over and above leverage eects. This is additional evidence on the shareholder-society risk shifting problem.
[Insert Table 7 about here]
Finally, a range of other specication checks are presented in Table 8, addressing three concerns related to the downside risk statistics. First, in practice, VaR is often reported at a 100(1−α) = 99% (not 95%) condence level. Second, while rm xed eects control for a dierent potential for moral hazard and risk between rms (Himmelberg et al., 1999), additional heterogeneity could stem from unobserved dierences between executives within the same nancial institution. For instance, compensation and an executive's eect on downside risk during the crisis may also depend on individual talent and responsibilities. Finally, it could be argued that CEOs are key decision makers within most rms, and that the incentives of most other executives are irrelevant.
Table 8 presents supportive evidence that takes these issues into account after re-estimating the models for VaR, ES, and MES from Table 3. In the leftmost three data columns, the probability level in the denitions of VaR, ES, and MES is adjusted to α = 0.01. The coecients on risk incentives are larger than in Table 3. However, because a scal year has about 250 trading days, setting α= 0.01leaves only two to three observations per rm to calculate the downside risk statistics. Therefore, the coecients for
α= 0.01could be strongly biased and should be interpreted with caution. In the middle three data columns, xed eects on the executive level are included to account for unobserved heterogeneity in individual talent and responsibilities. The results also support the view of risk incentives increasing downside risk, and these coecients are larger than in Table 3. In addition, the coecients on price incentives are no longer statistically dierent from zero suggesting that the negative eect of price incentives in Table 3 is driven by within-rm variation between executives. In the rightmost three data columns, non-CEO observations are excluded altogether. Again, the intertemporal relation between risk incentives and downside risk becomes stronger with larger coecients.
[Insert Table 8 about here]
5 Conclusions
This article studies the sensitivity of managerial wealth to price volatility (risk incentives) in the presence of two types of potential agency problems: the standard manager-shareholder agency problem and the risk-shifting problem between shareholders and society. To this end, I rst update more timely research evaluating the eect of risk incentives on shareholder returns during the nancial crisis. Next, I explore the intertemporal relation between risk incentives and three downside risk statistics. Value-at-risk (VaR) resembles a nancial institution's tolerance to losses, expected shortfall (ES) resembles losses when the rm does poorly, and marginal expected shortfall (MES) resembles an institution's exposure to loss spillovers from its peers. I also calculate the net economic eect of stock-based compensation (i.e., risk incentives and price incentives) after accounting for the osetting eect of price incentives. Since previous literature suggests that leverage plays a crucial role in the crisis, specic attention is given to the endogenous role of leverage as an instrument to increase rm risk.
Between 2006 and 2010, I document a signicant and positive association between risk incentives and consecutive buy-and-hold returns, in line with shareholder objectives. However, risk incentives also increase downside risk in consecutive years. This is not in line with societal objectives. It follows that the problem is not so much the standard managerial agency problem between managers and shareholders as the risk- shifting problem between shareholders and society. The increase in downside risk from risk incentives exceeds the decrease from price incentives, so that equity compensation has a net destabilizing eect on nancial
institutions after 2006. The increased exposure to downside risk from risk incentives follows over and above the increased exposure from leverage policy. In fact, risk incentives signicantly increase downside risk by inducing managers to implement more aggressive leverage policy, manifested by high leverage growth.
Additional evidence regarding the intertemporal link between risk incentives and downside risk is provided by demonstrating that end-of-2006 risk incentives are positively correlated to the probability of distress in 2008 or 2009, and to implied put volatility in subsequent years. The ndings are robust for executives, CEOs and CFOs, and CEOs; for dierent probability levels in the VaR, ES, and MES calculations; and for rm-level and executive-level xed eects.
While signicant, the indirect eect of risk incentives on downside risk through leverage is small compared to the direct eect of risk incentives on downside risk. This could indicate that, to a large extent, much of the variation in the link between incentives and risk is idiosyncratic. Indeed, many elements that play a role in this relation are unobservable, such as managerial ability and eort, risk culture, securitization strategies, and regulatory capital calculations including (but not limited to) expected, ex ante VaR. However, it could also indicate that the total eect of risk incentives pervades other organizational features that are at the discretion of management. Hence, analyzing other management policies could be an interesting alley for future research. For instance, Ellul and Yerramilli (2010) show that nancial institutions are less exposed to risk when having stronger risk governance in place. The present study corroborates their ndings, since less governed rms typically install more generous compensation packages.
In the wake of the nancial crisis, legislators have passed the Corporate and Financial Institution Com- pensation Fairness Act 2009 and the Dodd-Frank Wall Street Reform and Consumer Protection Act 2010.
The Acts have amended the Securities Exchange Act of 1934 to expand the rights of shareholders in approving compensation, appointing directors on compensation committees, and proposing compensation plans. These requirements address moral hazard between managers and shareholders, assuming that the monitoring of risk will be more eective once more power is assigned to shareholders. However, I nd that risk incentives enhance shareholder returns between 2006 and 2010, providing evidence against the presence of this agency problem. Therefore, this result implies that a shift toward more shareholder-based compensation is likely to be ineective in the nancial sector.
Instead, I nd evidence consistent with the presence of a dierent agency problem. The positive link between risk incentives and downside risk indicates that shareholder-based compensation increased the very losses that the Acts seek to prevent. Considering that the interests of managers are well-aligned with share- holders, this hints at moral hazard between shareholders and society, similar to Carpenter et al. (2011).
Therefore, this result implies that a shift towards more shareholder-based compensation may be counterpro- ductive in the nancial sector, as it may exacerbate the conict of interest between shareholders and society.