CEO compensation and risk-taking at financial firms: Evidence from U.S. federal loan assistance

Journal of Corporate Finance 47 (2017) 131–150

Contents lists available at ScienceDirect

Journal of Corporate Finance journal homepage: www.elsevier.com/locate/jcorpfin

CEO compensation and risk-taking at financial firms: Evidence from U.S. federal loan assistance☆ Amar Gande a, Swaminathan Kalpathy b,⁎ a b

Edwin L. Cox School of Business, Southern Methodist University, Dallas, TX 75275-0333, United States Neeley School of Business, Texas Christian University, Fort Worth, TX 76109, United States

a r t i c l e

i n f o

Article history: Received 6 September 2016 Received in revised form 7 July 2017 Accepted 1 September 2017 Available online 6 September 2017 JEL classification: G01 G21 G32 J33 M12 M52

a b s t r a c t We examine whether risk-taking among the largest financial firms in the U.S. is related to CEO equity incentives before the 2008 financial crisis. Using data on U.S. Federal Reserve emergency loans provided to these firms, we find that the amount of emergency loans and total days the loans are outstanding are increasing in pre-crisis CEO risk-taking incentives – “vega”. Our results are robust to accounting for endogeneity in CEO equity incentives and selection of financial firms into emergency loan programs. We also rule out the possibility that our results are driven by a bank's funding base, bank complexity, CEO overconfidence, or matching of CEOs to select banks. We conclude that equity incentives (vega) embedded in CEO compensation contracts were positively associated with risk-taking in financial firms which resulted in potential solvency problems. We also find some evidence, although somewhat weaker, that higher incentive alignment (“delta”) mitigated such problems in those financial firms. © 2017 Elsevier B.V. All rights reserved.

Keywords: CEO compensation CEO incentives Financial crisis Financial deregulation Federal emergency loans

1. Introduction Executive compensation at financial firms has received considerable regulatory scrutiny based on a view that compensation contracts incentivized managers at financial firms to undertake excessive risks during the financial crisis of 2008.1 The Dodd-

☆ We thank seminar participants at the European Finance Association (EFA) meetings at the University of Cambridge, U.K., the CBS Conference on Executive Compensation after the Financial Crisis at Copenhagen, the Australian National University (ANU)-Financial Research Network (FIRN) Banking and Financial Stability Meeting, the Financial Institutions, Regulation and Corporate Governance (FIRCG) Conference at the Melbourne Business School, the Federal Reserve Bank of Dallas, Indian School of Business (ISB), Southern Methodist University (SMU), Texas Christian University (TCU), and the University of Texas at Arlington (UTA) for their valuable comments. We thank Jeffry Netter (Editor) and an anonymous referee for helpful comments and suggestions. We are grateful to Lucian Bebchuk, Jeffrey Coles, Rudiger Fahlenbrach, Neal Galpin, Lisa Goh, Huseyin Gulen, Yian Liu, Meijun Qian, Lemma Senbet, Rex Thompson, Kumar Venkataraman and Andrew Winton for many helpful suggestions. We thank Vinay Katyal and Abhi Khawarey for excellent research assistance. ⁎ Corresponding author. E-mail addresses: [email protected] (A. Gande), [email protected] (S. Kalpathy). 1 A detailed report prepared by the Financial Crisis Inquiry Commission (FCIC) points to “widespread failures in financial regulation; dramatic breakdowns in corporate governance; excessive borrowing and risk-taking by households and Wall Street.” See http://www.fcic.gov/report for a full description of the Committee's report.

http://dx.doi.org/10.1016/j.jcorpfin.2017.09.001 0929-1199/© 2017 Elsevier B.V. All rights reserved.

132

A. Gande, S. Kalpathy / Journal of Corporate Finance 47 (2017) 131–150

Frank Wall Street Reform and Consumer Protection Act of 2010 called for specific proposals limiting incentive compensation for executive officers of financial firms with assets greater than $1 billion.2 Equity-based compensation, especially stock options, can mitigate agency problems by inducing risk-averse managers to undertake value-enhancing risky investments (see Haugen and Senbet, 1981, and Smith and Stulz, 1985). There is empirical evidence that the sensitivity of firm-related equity wealth to stock return volatility encourages CEOs to implement riskier investment and financial policies (see, for example, Coles et al., 2006 and Low, 2009).3 Chen et al. (2006) show that following deregulation of the 1990s, banks have increasingly employed stock option-based compensation, and as a result, the structure of executive compensation induces risk-taking. In fact, some scholars (e.g., Bebchuk and Spamann, 2010) have expressed the concern that executive compensation contracts can intensify risk-taking at financial firms due to implicit risk-shifting incentives resulting from high leverage. We analyze the role of equity incentives embedded in CEO compensation contracts (i.e., CEO pay-performance sensitivity (“Delta”) and risk-taking incentives (“Vega”)) in determining bank performance. We use a novel dataset of emergency loans provided by the Federal Reserve during the financial crisis, and measure bank performance by the extent to which a bank received external government support, specifically the amount of U.S. Federal Reserve emergency loan assistance received by a financial firm during the financial crisis. We contribute to this literature by providing an alternate informative variable to measure bank performance especially during crisis times. We argue that bank performance during a crisis depends on a first order basis on whether or not a financial institution will survive the crisis, which in turn depends on the extent of federal financial assistance received by the bank. We provide an analysis of the information content of our variable relative to that of traditional measures of bank performance in Section 4.8. Our study also complements existing literature on federal assistance to financial firms in terms of equity financing (e.g., Troubled Asset Relief Program (TARP) as analyzed in Bayazitova and Shivdasani (2012) – see Section 4.10 for details) by presenting evidence based on loan (i.e., non-TARP) assistance to financial firms. One potential shortcoming of analyzing TARP is that a number of healthy firms were included in TARP because regulators were concerned that inclusion of weak firms alone could send an adverse signal to market participants about firms selected for TARP assistance. In contrast, there were no such concerns with the loan programs we study here since the identity of recipients was revealed well after the financial crisis had subsided. We identify sixty-nine large financial institutions (for which we also have compensation data in Execucomp) that received emergency loan assistance from the U.S. government during 2007–2010 through a series of loan programs (see Panel A of Table 1 for details). Data on federal emergency loans were first made publicly available on December 1, 2010. Subsequently, on January 9, 2012, some additional data including details on the Discount Window program were made available to Bloomberg pursuant to its request under the Freedom of Information Act (see Section 3 for details). We find evidence that the federal loan assistance is strongly related to pre-crisis CEO vega, after controlling for the characteristics of financial firms (firm size, leverage, market-to-book ratio of assets, and pre-crisis stock returns), other attributes of CEO compensation (e.g., CEO Delta), and including fixed effects for the type and timing of Federal program. The associated economic effects are non-trivial. For example, when we consider all programs together, a 10% increase in pre-crisis CEO vega is associated with a 1.67% increase in federal loan assistance. The above finding suggests that equity incentives embedded in CEO compensation contracts are strongly associated with bank performance. This brings up a related question. What was the source of the poor performance of banks that resulted in them requiring governmental loan assistance? We start with the explanation that bank CEO risk-taking incentives had an adverse impact on bank solvency. Extant literature has largely argued that compensation contracts create incentives for CEOs to take risks. For example, equity incentives could lead bank CEOs into making riskier acquisitions (see Hagendorff and Vallascas, 2011) and to undertake riskier investment choices (see DeYoung et al., 2013 and Cheng et al., 2015). While there is general agreement in the literature that compensation contracts create incentives for CEOs to take risks (as described above), there is no consensus on whether such risk-taking manifests itself in bank performance. On one hand, Bai and Elyasiani (2013) find that higher CEO vegas lead to greater bank instability (measured by Z-score), whereas on the other hand other studies such as Fahlenbrach and Stulz (2011) and Kolasinski and Yang (2017) find an insignificant relation between pre-crisis CEO vegas and crisis-period bank performance measures, such as buy and hold returns and return on assets, and risk exposures. This lack of consensus is also closely related to the debate in the literature on liquidity versus solvency. One view is that banks did not take on a higher level of risk but simply faced a liquidity shock, and responded by using the emergency loan assistance facilities (which were specifically designed and set up to solve such liquidity problems). Proponents of the liquidity view (e.g., Fahlenbrach and Stulz, 2011) would argue that the observed positive relation between pre-crisis CEO vega and emergency loan assistance was a result of the liquidity shock that banks were exposed to during the financial crisis rather than solvency problems stemming from risk-enhancing activities. An alternative view is that firms took on a higher level of risk which resulted in potential solvency problems which left these firms with no other option but to use the emergency financial assistance provided by the Federal Reserve. Proponents of the solvency view would suggest that if liquidity-based explanations and other reasons that are

2 The relevant portion of the regulation is in Section 956 entitled “Enhanced Compensation Structure Reporting.” See http://www.gpo.gov/fdsys/pkg/BILLS111hr4173enr/pdf/BILLS-111hr4173enr.pdf for details. 3 Other studies on the relation between executive risk incentives and firm-level risk-taking include – Haugen and Senbet (1981), Smith and Stulz (1985), Gilson and Vetsuypens (1993), Houston and James (1995), Guay (1999), Carpenter (2000), Cohen et al. (2000), Knopf et al. (2002), Rajgopal and Shevlin (2002), Ross (2004), Kalpathy (2009), Shue and Townsend (2017).

A. Gande, S. Kalpathy / Journal of Corporate Finance 47 (2017) 131–150

133

Table 1 Descriptive statistics of the U.S. Federal Reserve loan facilities initiated during the financial crisis and firms receiving assistance. This table provides summary statistics of loan assistance provided by the U.S. Federal Reserve Board during the financial crisis and the firms receiving assistance. Panel A provides information on the loan programs. The Federal Reserve credit facilities start in August of 2007 and end in April of 2010. “All Federal Reserve Programs” include Asset-Backed Commercial Paper Money Market Mutual Fund Liquidity Facility (AMLF), Commercial Paper Funding Facility (CPFF), Primary Dealer Credit Facility (PDCF), Term Auction Facility (TAF), Term Securities Lending Facility (TSLF), Single-tranche Open Market Operations (ST OMO) and Discount Window (DW). Data is aggregated for each firm at the program level for each year. Panel B provides summary statistics relating to characteristics of 69 financial firms receiving credit assistance from the U.S. Federal Reserve Board during the financial crisis. Market capitalization is stock price multiplied by total shares outstanding. Total assets is the book value of assets. Market-to-book ratio is defined as the sum of market value of equity and book value of total liabilities divided by the book value of assets. Stock return is annual (raw) stock returns. Book leverage is total liabilities divided by book value of assets. Volatility is annualized standard deviation of monthly stock returns using five years of data. CEO tenure is length of time (in years) an individual has been CEO. Option awards is the Black-Scholes value of stock options awards during a year. Stock awards is the sum of restricted stock awards granted during a year and any stock awards granted under long-term incentive plans. Delta is the dollar change in the value of the portfolio of stock and option holdings for a 1% change in stock price. Vega is the dollar change in the value of the portfolio of option holdings for a 0.01 change in volatility of stock returns. All variables are measured at the end of fiscal year 2006. All dollar amounts are expressed in constant 2007 units. Panel A N

Mean

Std. dev.

Min

25th Pctile

Median

75th Pctile

Max

31,947 13,408 3,681

0.03 5 0.03

150 175 20

750 980 193

4,368 5,000 777

170,949 62,167 33,910

Loan amount (in $ million) All Federal Reserve programs Term Auction Facility (TAF) Discount Window (DW)

149 94 92

11,692 6,286 1,023

Panel B Mean

Std. dev.

25th Pctile

Median

75th Pctile

Characteristics of financial firms (N = 69) Market capitalization (in $ million) Total assets (in $ million) Market-to-book Stock return Book leverage Volatility

25,322 102,999 1.09 0.15 0.90 0.20

36,590 111,884 0.07 0.14 0.04 0.06

1,480 10,234 1.05 0.07 0.89 0.14

9,161 48,910 1.08 0.14 0.90 0.18

28,433 185,869 1.12 0.22 0.92 0.23

Characteristics of CEOs (N = 69) CEO tenure (years) Salary + bonus (in $ ‘000) Option and stock awards (in $ ‘000) Total compensation (in $ ‘000) Option and stock awards/total compensation Delta of portfolio of stock and options (in $ ‘000) Vega of portfolio of stock and options (in $ ‘000)

6.49 3,878 7,268 11,147 0.45 1,204 284

5.41 4,069 10,576 12,798 0.29 1,643 365

2.00 1,093 365 1,984 0.22 174 16

5.00 1,872 2,541 5,836 0.49 655 127

9.00 5,404 11,023 17,692 0.68 1,550 378

not directly related to risk-taking cannot fully explain the evidence, then perhaps one cannot out entirely rule out a solvencybased explanation. Our paper makes a significant contribution to the liquidity versus solvency debate. We posit that if firms were to face “solvency” problems (e.g., as a result of taking on a higher level of risk) they would require financial assistance over a prolonged period of time. In contrast, firms that face “liquidity” problems are likely to need financial assistance only for a shorter period of time. In other words, even if the original maturity of a lender of last resort (LOLR) financing is relatively short, if a firm continues to refinance that facility over an extended period of time, such a firm is likely to have more than a liquidity problem. We find that those firms where CEOs had the highest risk-taking incentives, as measured by high levels of vega, had emergency loans outstanding for a longer period of time with the Federal Reserve. This result will not allow us to rule out a potential solvencybased explanation. We next examine other explanations, i.e., solvency problems unrelated to risk-taking incentives which could potentially explain bank performance (and hence our results), such as CEO overconfidence, funding base, matching of CEOs based on education with select banks, firm complexity etc. For example, Ho et al. (2016) find evidence that bank CEO overconfidence is related to weakened lending standards and increased leverage prior to the financial crisis which made those banks more vulnerable to financial shocks. Their study however does not explicitly control for CEO equity incentives. We augment our model by including a measure of CEO overconfidence and find that CEO overconfidence is positively related to federal emergency loan assistance although the statistical significance is generally weak. Importantly, our measure of CEO risk-taking incentives continues to remain statistically significant in all specifications. Another possible explanation is the funding base of a borrowing firm. That is, whether a financial firm with a limited access to deposit financing is more likely to utilize federal emergency loans and whether this explains our evidence. When we augment our regressions with deposits-to-liabilities ratio, a proxy for a firm's funding base, we find that a firm that relies more on deposits is less likely to use federal emergency loans. Nevertheless, we find that our measure of CEO risk-taking incentives continues to remain statistically significant suggesting that CEO incentives is an important channel that explains bank risk-taking even after we control for the funding base of the financial firm.

134

A. Gande, S. Kalpathy / Journal of Corporate Finance 47 (2017) 131–150

We examine if our results are driven by CEOs with certain attributes (education) self-selecting into large banks and whether this is correlated with bank policy choices, and hence bank performance. Accordingly, we examine if our results continue to hold whilst explicitly controlling for endogenous firm-manager matching (King et al., 2016; Pan, 2017). Our results on the positive association between the extent on federal emergency loans and CEO risk-taking incentives continue to hold after accounting for CEO-bank matching. We next address endogeneity in CEO incentives by employing an instrumental variables approach and accounting for selection bias through Tobit analysis (where the probability of seeking emergency financial assistance and the size of emergency loans are modeled simultaneously) and continue to find that CEO vega is positively and significantly related to federal emergency loan assistance. In summary, we make several important contributions to the literature on equity incentives and bank performance. First, we contribute to the liquidity versus solvency debate and show that equity incentives embedded in CEO compensation contracts are positively associated with risk-taking in financial firms, which resulted in potential solvency problems that left these firms with no other option but to use the emergency financial assistance provided by the Federal Reserve. Second, we provide an alternate variable measuring bank performance, especially during crisis times, namely, the extent of governmental assistance to financial firms that is more informative than traditional measures used in the literature. Finally, we show that other explanations (e.g., CEO overconfidence, funding base, firm complexity, CEO-bank matching) do not drive our results, and hence provide clarity on the first order importance of equity incentives affecting bank performance through the risk-taking channel. Overall, our results on managerial incentives for risk-taking (e.g., CEO vega) among banks complement theoretical work on banking with policy recommendations. One such approach, articulated in John et al. (2000), is incorporating the incentive features of CEO compensation in pricing FDIC insurance through covenant-like conditions. Another approach is basing compensation for bank CEOs with debt-like features as a way to curb excessive risk-taking (Bolton et al., 2015). In contrast, regulation mandating explicit restrictions on the level of managerial compensation has unintended consequences (Perry and Zenner, 2001; Murphy, 2012), and is not supported by the empirical evidence till date which is inconclusive about whether average CEO pay is too much or too little (Faulkender et al., 2010). Our paper is organized as follows: Section 2 discusses the related literature and our testable hypotheses. Section 3 describes our data. Section 4 analyzes federal loan assistance during the crisis and its association with pre-crisis CEO incentives. Section 5 concludes.

2. Related literature and hypotheses development In this section, we discuss the literature on bank performance, and the effect of CEO incentives on bank performance. We relate this literature to the development of our hypotheses. Studies of bank performance typically measure performance in terms of specific outcomes (e.g., type of acquisitions, merger performance etc.) or metrics (e.g., holding period returns, return on assets etc.) and try to explain such performance through underlying structures, such as compensation, incentive structure, and governance mechanisms. Early studies of equity incentives and bank performance include Houston and James (1995), Adams and Mehran (2003), and John and Qian (2003) who find evidence of a lower reliance on stock option pay for CEOs of bank holding companies (BHC) compared to manufacturing firms.4 Adams and Mehran (2003) also find evidence of a lower equity ownership for CEOs of bank holding companies (BHC) compared to manufacturing firms.5 While the above early studies suggested that financial firm CEOs' compensation contracts are associated with weaker incentive strength, evidence in these papers predated significant deregulation in the 1990s in the financial services industry that culminated with the passage of the Gramm-Leach-Bliley Act of 1999. Gande et al. (1999) show a considerable increase in competition for securities underwriting leading up to 1999. The deregulation was accompanied by the re-emergence of “universal” banks6 involved in commercial banking, investment banking, securities trading,7 and insurance operations. Considering the effects of financial deregulation on a financial firm's menu of expanded investment opportunities and industry competition,8 it is conceivable that a financial firm will have enhanced incentives to provide its CEO, equity compensation with 4 Deregulation of interstate commercial banking in the 1980s is accompanied by an increase in pay-performance sensitivity for bank CEOs (see Crawford et al., 1995 and Hubbard and Palia, 1995). 5 Financial firms, especially banks are considered “special” or unique and hence differ from non-financial (e.g., manufacturing) firms for reasons, such as their enhanced ability and incentives to screen and monitor borrowers, their unique role in deposit taking and the associated implications of deposit insurance, their higher levels of leverage and fragility of capital structures as compared to non-financial firms. Also, see Winton (1995) for the role of seniority and Rajan and Winton (1995) for the role of covenants and collateral as contractual devices that influence a bank's incentive to monitor a borrower. For comprehensive reviews of why banks are considered “special”, see James and Smith (2000), Gorton and Winton (2003), Saunders and Cornett (2010), and Gande and Saunders (2012). 6 Universal banking was also prevalent prior to the passage of the Glass-Steagall Act in 1933. See Puri (1996) and Gande et al. (1997) for details. 7 It appears that deregulation has also resulted in a significant increase in “securitized” banking and introduction of newer financial instruments for securities trading in recent years. In contrast to traditional banking where banks originate and hold loans in their balance sheets, securitized banking involves packaging and reselling of loans funded mainly by repo agreements. Gorton and Metrick (2010) document a dramatic increase in securitization of assets starting in 2000. Boyd et al. (2010) and Gorton and Metrick (2010) argue that the increase in securitization of assets has contributed significantly to the relative importance of “shadow” banks, that is financial institutions not regulated as banks but involved in traditional banking. 8 DeYoung et al. (2013)) show that contractual risk-taking incentives for chief executive officers (CEOs) increased at large U.S. commercial banks around 2000, when industry deregulation expanded these banks' growth opportunities.

A. Gande, S. Kalpathy / Journal of Corporate Finance 47 (2017) 131–150

135

higher pay-for-performance sensitivity. Minnick et al. (2011) find that banks whose CEOs have higher pay-for-performance sensitivity (PPS) have significantly better abnormal stock returns around the time of the acquisition announcements. Another strand of the banking literature examines the impact of corporate governance mechanisms on bank performance. Aebi et al. (2012) find evidence that risk management-related corporate governance mechanism, specifically the reporting of a chief risk officer (CRO) directly to the board of directors rather than to the CEO is associated with better bank performance during the financial crisis of 2007–2008. Adams and Mehran (2012) find that board size is positively related to bank performance, suggesting that additions of directors may add value as bank holding company (BHC) complexity increases. Beltratti and Stulz (2012) find that banks with more shareholder-friendly boards surprisingly performed significantly worse during the crisis than other banks, were not less risky before the crisis, and reduced loans more during the crisis. Erkens et al. (2012) use a unique dataset of 296 financial firms from 30 countries that were at the center of the crisis and find that firms with more independent boards and higher institutional ownership experienced worse stock returns during the crisis period. Other recent studies have looked at the impact of CEO compensation structure on bank risk taking. For example, equity incentives could lead bank CEOs into making riskier acquisitions (see Hagendorff and Vallascas, 2011) and to undertake riskier investment choices (see Bhagat and Bolton, 2014; Chesney et al., 2016; DeYoung et al., 2013 and Cheng et al., 2015). In contrast, Bennett et al. (2015) analyze inside debt (i.e., pension benefits and deferred compensation) and find that higher pre-crisis holdings of inside debt relative to inside equity by a CEO after controlling for firm leverage is associated with lower default risk and better performance during the crisis period. Given the overall evidence in prior literature of a positive relation between CEO risk-taking incentives and investment and financial policy choices, and the implications of an expanded menu of opportunities and heightened competition resulting from the deregulation of the 1990s, we hypothesize the following (in alternative form): H1. The amount of federal loan assistance to financial firms during the crisis is positively related to CEO risk-taking incentives or “vega” in the pre-crisis period. The literature does not have a clear prediction for the effect of CEO pay-performance sensitivity (PPS) on firm risk.9 On one hand, PPS might incentivize a manager to take on more risk (positive effect) through increased incentive alignment, but on the other hand increasing PPS might increase managerial risk-aversion and cause managers to decrease risk (negative effect). Which of these two effects dominate would depend on the manager's tolerance for risk and its effect on her utility which are both unobservable to a researcher. The hypothesized positive association between the federal loan assistance and CEO risk-taking incentives is not entirely inconsistent with the idea that CEOs respond optimally to incentives by way of their corporate policy choices which are exacerbated expost during the financial crisis. This brings up an important debate relating to liquidity versus solvency. The liquidity view suggests that firms did not take on a higher level of risk but simply faced a liquidity shock, and responded by utilizing the federal emergency loan assistance facilities (which were specifically designed and set up to solve such liquidity problems). The solvency view would suggest that firms took on a higher level of risk which resulted in potential solvency problems that left these firms with no other option but to use the emergency financial assistance provided by the Federal Reserve (irrespective of stated purpose of these facilities). In order to determine whether CEOs respond to equity incentives by taking on excessive risks, we argue that firms that take on “excessive” risks are likely to have “solvency” problems which require federal financial assistance over a longer period of time. In contrast, firms that face ex-post “liquidity” problems (when their CEOs responded optimally ex-ante to incentives) are likely to receive federal financial assistance only for a shorter period of time. Therefore, we hypothesize the following (in alternative form): H2. The duration of federal loan assistance to financial firms during the crisis is positively related to the CEO risk-taking incentives or “vega” in the pre-crisis period. We measure the duration of federal loan assistance in terms of the total number of days that a financial firm has net debt outstanding with the Federal Reserve. Our paper makes an important contribution to the liquidity versus solvency debate. The evidence from the existing literature is mixed. For example, the evidence from Fahlenbrach and Stulz (2011) study is consistent with a liquidity interpretation. They find no evidence that pre-crisis CEO vega is associated with a firm's accounting and stock price performance during the financial crisis.10 The emergency loan assistance data became available subsequent to their study. Our study highlights the differential nature

9 John and John (1993) derive a negative relation between pay-performance sensitivity and leverage due to risk-shifting incentives for equity holders in a levered firm. Demsetz and Lehn (1985) point out that regulation substitutes for the need for explicit incentives by constraining the discretion of managers for project selection. 10 In a related study, Bhagat and Bolton (2014) study the executive compensation structure in fourteen of the largest U.S. (too big to fail) financial institutions during 2000–2008 and find results generally not supportive of the conclusions of Fahlenbrach and Stulz (2011) that the poor performance of banks during the crisis was the result of unforeseen risk. Chesney et al. (2016) find evidence suggesting that incentives of CEOs of U.S. financial institutions to take asset risk (incentives to increase firm value) in years prior to the recent financial crisis were significantly positively (negatively) associated with write-downs during the crisis. Cheng et al. (2015) show that riskier firms may offer higher total pay as compensation for the extra risk in equity stakes borne by risk-averse managers.

136

A. Gande, S. Kalpathy / Journal of Corporate Finance 47 (2017) 131–150

of information11 contained in emergency loans vis-à-vis firm performance measured during the financial crisis.12 See Section 4.8 for additional details on the information content of the federal loan assistance. One might also argue that lender of last resort (LOLR) financing, such as the federal emergency loans studied here should simply reflect liquidity risk. For example, federal emergency assistance programs as per the description of these programs when they were first announced suggest that the Federal Reserve considered the viability of the financial firm when evaluating its funding request. However, evidence in Acharya et al. (2017) indicates that central bank LOLR facilities, such as the Term Securities Lending Facility (TSLF), which is one of the facilities we include in our study, elicit greater and more aggressive participation from less capitalized financial firms, suggesting that dealers with worse financial conditions “crowded out” dealers with better financial conditions in the TSLF auctions. Furthermore, not all LOLR facilities are alike. For example, Armantier et al. (2015) show that banks were willing to pay an average premium of at least 44 basis points to borrow from the TAF (a LOLR auction-facility that we examine in our study) rather than borrow from the traditional LOLR facility, namely the Discount Window (DW) to avoid the stigma associated with borrowing from the Federal Reserve, i.e., a fear of strong repercussions should the market discover that they had borrowed from the Federal Reserve.13 This evidence from their study suggests that the market could have interpreted a bank's participation in certain LOLR programs as being reflective of solvency problems rather than liquidity problems. By examining the extent of borrowing under federal emergency loan assistance as well as the duration of these borrowings, our study is able to provide a clearer picture regarding the liquidity versus solvency debate. 3. Data and sample construction We obtain data on seven different U.S. Federal Reserve emergency credit facilities from Bloomberg website.14 These are: AssetBacked Commercial Paper Money Market Mutual Fund Liquidity Facility (AMLF), Commercial Paper Funding Facility (CPFF), Primary Dealer Credit Facility (PDCF), Term Auction Facility (TAF), Term Securities Lending Facility (TSLF), Single-tranche Open Market Operations (ST OMO) and Discount Window (DW). In several cases, particularly with TAF, the recipient entity is a regional bank. We look up the details of the corresponding Bank Holding Company (BHC) from the websites managed by the U.S. Federal Reserve.15 We hand match the financial firms with CRSP to obtain the PERMNO of 165 BHCs. We merge all firms for which we are able to obtain PERMNO with the CEO compensation data in Execucomp and in the process end up with 69 financial firms. Since Execucomp database covers large firms (i.e. S&P 1500), we are unable to match all firms and therefore end up with the largest financial firms for which we have availability of executive compensation data. We collect supplementary data from CRSP, Compustat, Compustat (Bank) and Compustat (Segment) databases. Table 1 (Panel A) presents descriptive statistics of all the programs for the 69 financial firms that we examine in our study. Federal loan assistance under all programs captures assistance to both depository and non-depository financial institutions. Out of these programs, there are some programs that are directed at depository institutions. For example, DW which has been available to depository institutions all along is typically used by the Federal Reserve to infuse capital into banks that have no other means of raising capital (“lender of last resort”). We provide descriptive statistics for DW separately. Similarly, the TAF program, again directed at depository institutions, was put in place by the Federal Reserve during the financial crisis to address the stigma arising from the use of DW. We also provide the descriptive statistics for TAF separately. The mean (median) annual loan assistance received by a financial firm under the TAF program is $6.3 billion ($980 million) while the mean (median) loan assistance received under the DW program is $1 billion ($193 million). These numbers suggest that the loan programs we examine in this study are economically quite large. We ensure that the loan assistance received by a firm is measured such that only new loans are counted, i.e., if a loan is renewed, the renewed loan is not counted as an additional loan. Table 1 (Panel B) provides a summary of the characteristics of the 69 financial firms that received emergency loans from the Federal Reserve during the financial crisis and their CEOs. All variables in dollar terms are expressed in constant 2007 units. The typical financial firm that receives financial assistance has a (mean) market capitalization of $25.32 billion, total book value of assets of $102.99 billion, and book leverage, defined as total liabilities divided by book value of assets, of 0.90. More than 25% of the financial firms in our sample have leverage in excess of 0.92. For comparison, the typical financial firm, defined as a firm with SIC 11 There are numerous reasons to believe that the Federal Reserve loan programs reflect private information known only to the federal regulators and unavailable to market participants contemporaneously. For instance, during the crisis, the U.S. Treasury Department called for detailed information regarding a bank's financial health that was otherwise not available in publicly disclosed financial statements to conduct “stress tests.” Moreover, there were emergency private meetings between federal regulators and CEOs of large banks. For example, a 2008 New York Times article states “… in the first of a series of emergency meetings at the Federal Reserve building in Lower Manhattan. The meeting was called by Fed officials, with Treasury Secretary Henry M. Paulson Jr. in attendance, and it included top bankers.” The same article goes on to state “Outside the public eye, Fed officials had acquired much more information since March about the interconnections and cross-exposure to risk among Wall Street investment banks, hedge funds and traders in the vast market for credit-default swaps and other derivatives.” 12 We replicated the tests of firm performance in Fahlenbrach and Stulz (2011) for our sample using the Return on Equity and Buy-and-hold returns as dependent variables on the set of explanatory variables in our regressions in Table 2. Our coefficients on Log (Vega) are similar to those in their study and suggests that there is no evidence that CEO risk-taking incentives are related to metrics of a firm's financial health. In contrast, when we run the regression using our dependent variable, namely log of emergency loan assistance, we find a positive and statistically significant coefficient on Log (Vega) suggesting the differential nature of information contained in emergency loans vis-à-vis firm performance measured during the financial crisis. In additional tests, we find the correlation between abnormal stock returns and loan amounts during event month to be statistically insignificant at any meaningful level of significance suggesting that stock markets do not contemporaneously impound information in the federal loan assistance. These results are available from the authors upon request. 13 Also, see Furfine (2001) for some related evidence on stigma of borrowing from the federal discount window, Berger et al. (2016) for evidence on substitutability and complementarity of the discount window and Term Auction Facility. 14 Source: http://www.bloomberg.com. 15 Sources: http://www.ffiec.gov/nicpubweb/nicweb/NicHome.aspx and http://cassidi.stlouisfed.org/institutions/.

A. Gande, S. Kalpathy / Journal of Corporate Finance 47 (2017) 131–150

137

Table 2 Loan assistance during financial crisis and CEO equity incentives in recipient financial firms: OLS estimates. This table provides OLS estimates from regressions of the log of the loan assistance to financial firms during the financial crisis. The Federal Reserve credit facilities start in August of 2007 and end in April of 2010. “All Programs” include sum of credit assistance received under Asset-Backed Commercial Paper Money Market Mutual Fund Liquidity Facility (AMLF), Commercial Paper Funding Facility (CPFF), Primary Dealer Credit Facility (PDCF), Term Auction Facility (TAF), Term Securities Lending Facility (TSLF), Single-tranche Open Market Operations (ST OMO) and Discount Window (DW). Data is aggregated for each firm at the program level for each year. Delta is the dollar change in the value of the portfolio of stock and option holdings for a 1% change in stock price. Vega is the dollar change in the value of the portfolio of option holdings for a 0.01 change in volatility of stock returns. Leverage is total liabilities divided by book value of assets. Market-to-book ratio is defined as the sum of market value of equity and book value of total liabilities divided by the book value of assets. Stock return is annual (raw) stock returns. Model (3) includes program fixed effects. All models include program year fixed effects. Standard errors are corrected for heteroskedasticity and clustering at the firm level. t-statistics are reported in parentheses. ***, **, and * denote significance at less than 1%, 5%, and 10% levels, two-tailed tests, respectively. Program:

All programs

TAF and DW

TAF or DW

Model:

(1)

(2)

(3)

Intercept

−3.642 (−0.58) 0.167** (2.14) −0.184** (−2.17) 0.646*** (6.15) −1.296 (−0.20) 3.220 (0.63) −2.297*** (−4.13) 0.437 137

10.723 (1.10) 0.216*** (3.25) 0.058 (0.64) 0.265* (1.95) −22.099** (−2.16) 7.057 (0.86) −2.536*** (−2.87) 0.383 124

4.515 (0.53) 0.197*** (5.17) −0.012 (−0.17) 0.247** (2.04) −15.854* (−1.98) 9.256 (1.47) −1.866** (−2.53) 0.360 174

Log(Vega) Log (Delta) Log (market cap) Leverage Market-to-book Stock return Adjusted R-square N

codes between 6000 and 6999, in Execucomp database has a (mean) market capitalization of $10.73 billion, total book value of assets of $36.17 billion, and book leverage of 0.74. This suggests that a firm receiving financial assistance from the Federal Reserve is roughly three times in size compared to a financial firm in Execucomp database, and has a higher leverage.16 The typical CEO of a financial firm has been in office for 6.49 years; has a (mean) annual cash compensation, defined as the sum of salary and bonus, of $3.88 million; annual equity compensation, defined as the sum of Black-Scholes value of all option awards and value of all stock awards, of $7.27 million; annual total compensation, defined as sum of cash and equity compensation, of $11.15 million. We normalize the annual equity compensation by dividing it by total compensation during the year. This variable indicates the proportion of total compensation coming from equity-based awards. Roughly half of all the annual compensation for the typical CEO of a financial firm is driven by grants of stock options and stock. An important variable in all our analysis is CEO equity incentives. The equity incentives we examine are the pay-performance sensitivity and risk-taking incentives. We follow the method in Core and Guay (2002) and define “Delta” and “Vega”. Delta captures the pay-performance sensitivity of CEO compensation, and is defined as the dollar change in the value of stock and option holdings for a 1% change in stock price. Vega captures the risk-taking incentives for a CEO, and is defined as the dollar change in the value of option holdings for a 0.01 change in stock return volatility. Under the new reporting rules for compensation that took effect in 2006, firms are required to disclose details of the complete portfolio of exercised and unexercised stock options in the outstanding equity tables. As a result we are able to accurately quantify CEO portfolio delta and vega without having to resort to the one year approximation (OA) method in Core and Guay (2002). The typical CEO has a (mean) change in firm-related wealth of $1.20 million for a 1% change in stock price (or delta) and $284,000 change in firm-related wealth for a 0.01 change in volatility (or vega). One concern with the economic incentive effects that we document above is that they appear to be too small to create large effects in terms of influencing a CEO to gamble and take on excessive risk. We make the following observations to allay this concern. First, our estimates are fairly close to the mean delta and mean vega of $1.1 million and $189,000 reported in Fahlenbrach and Stulz (2011), and the minor differences are likely due to sample differences. Second, as discussed in the introduction, implicit incentives to take on risk are very strong for financial firms due to their high levels of leverage (see, Bebchuk and Spamann, 2010). Consequently, even fairly small levels of explicit risk-taking incentives can combine with the high levels of implicit incentives for the financial firms to result in large amplifying effects on managerial behavior. Third, our calculation of vega assumes no additional convexity coming from common stock. Since common stock can be thought of as a call option on the firm value with an exercise price being the amount of debt, it follows that the convexity delivered by common stock holdings is likely low as long as leverage is not very high. Financial firms carry high levels of leverage so our vega calculations understate the true vega that might include the additional source of convexity arising from a CEO's common stock holdings. Finally, the economic incentive effects we documented above are average effects, and the actual effects can be significantly larger for firms most affected during the financial crisis. For example, Angelo Mozilo, the CEO of Countrywide Financial, a firm that which witnessed huge losses during the 16 Our comparison is conservative because Execucomp database contains the largest firms in the U.S. economy. Comparison with financial firms in Compustat would make differences in firm size significantly larger.

138

A. Gande, S. Kalpathy / Journal of Corporate Finance 47 (2017) 131–150

Table 3 Loan assistance during financial crisis and CEO equity incentives: exposure to real estate market. This table provides OLS estimates from regressions of the log of the loan assistance to financial firms during the financial crisis. The Federal Reserve credit facilities start in August of 2007 and end in April of 2010. “All Programs” include sum of credit assistance received under Asset-Backed Commercial Paper Money Market Mutual Fund Liquidity Facility (AMLF), Commercial Paper Funding Facility (CPFF), Primary Dealer Credit Facility (PDCF), Term Auction Facility (TAF), Term Securities Lending Facility (TSLF), Single-tranche Open Market Operations (ST OMO) and Discount Window (DW). Data is aggregated for each firm at the program level for each year. Delta is the dollar change in the value of the portfolio of stock and option holdings for a 1% change in stock price. Vega is the dollar change in the value of the portfolio of option holdings for a 0.01 change in volatility of stock returns. Leverage is total liabilities divided by book value of assets. Market-to-book ratio is defined as the sum of market value of equity and book value of total liabilities divided by the book value of assets. Stock return is annual (raw) stock returns. Real estate exposure is the sensitivity of monthly stock returns to the Case-Shiller 20-city home price index return. A time-series regression of monthly stock returns during the period 2000 to 2006 is run on Fama-French three factors (namely, size, book-to-market, and market), momentum factor, and the Case-Shiller 20-city home price index return. Model (3) includes program fixed effects. All models include program year fixed effects. Standard errors are corrected for heteroskedasticity and clustering at the firm level. t-statistics are reported in parentheses. ***, **, and * denote significance at less than 1%, 5%, and 10% levels, two-tailed tests, respectively. Program:

All programs

TAF and DW

TAF or DW

Model:

(1)

(2)

(3)

Intercept

−3.263 (−0.54) 0.162** (2.30) −0.216*** (−2.81) 0.575*** (5.06) −2.428 (−0.39) 3.258 (0.67) −2.233*** (−3.96) 0.097* (1.90) 0.448 137

12.551 (1.28) 0.207*** (3.45) 0.023 (0.24) 0.180 (1.15) −22.577** (−2.26) 6.746 (0.85) −2.388*** (−2.77) 0.111* (1.78) 0.398 124

5.694 (0.68) 0.194*** (5.55) −0.051 (−0.65) 0.170 (1.27) −16.119** (−2.06) 9.315 (1.50) −1.723** (−2.37) 0.090** (2.33) 0.372 174

Log (Vega) Log (Delta) Log (market cap) Leverage Market-to-book Stock return Real estate exposure Adjusted R-square N

financial crisis, had a portfolio vega of approx. $1.3 million which is 2.8 standard deviations above the mean portfolio vega of $284,000 as of 2006. 4. Federal Reserve assistance during the crisis and pre-crisis CEO equity incentives In this section, we examine the link between CEO incentives and risk-taking in the context of the financial crisis. To test H1, we examine whether loan assistance provided by the Federal Reserve to financial firms during the crisis is positively related to CEO equity incentives in the pre-crisis period. That is, we regress the log of loan amount granted to a financial firm during the crisis on pre-crisis incentives (i.e., log of delta and log of vega as of 2006). The results are shown in Table 2. In column (1), we aggregate loans across all programs on a yearly basis. In addition, we control for the characteristics of financial firms (e.g., firm size, leverage, market-to-book ratio of assets, and stock return performance) and include program year fixed effects. Consistent with H1, we find evidence that the loan amount received during the financial crisis is strongly related (both economically and statistically) to CEO risk-taking equity incentives. Specifically, a 10% increase in CEO vega is associated with a 1.67% increase in federal loan assistance when we consider all programs. One standard deviation increase in CEO vega (i.e. $370,000), when holding all other explanatory variables constant at their means, is associated with an approximate increase in federal loan assistance of $138 million. This evidence shows that CEOs with higher risk-taking incentives – vega – receive higher federal loan assistance. One possible interpretation is that CEOs of the largest financial firms acted upon these incentives, and engaged in a higher level of risk-taking (e.g., exposure to real estate assets) that culminated in higher federal loan assistance as compared to CEOs with lower risk-taking incentives. We provide some evidence consistent with such an interpretation in Section 4.1. The literature does not have a clear prediction for the effect of CEO pay-performance sensitivity (PPS) on firm risk. On one hand, increased delta could provide managers greater incentives to take on projects with higher risk (as long as they are also higher NPV). However, on the other hand, higher delta could make the CEO more risk-averse and consequently lead to implementation of projects with lower risk. Coles et al. (2006) provide empirical evidence that higher CEO delta is associated with less risky investment and financial policies (i.e. lower R&D expenditures and lower book leverage) in a sample of non-financial firms. Our results show that CEOs of financial firms with higher pay-performance sensitivity (PPS) – delta – receive lower federal loan assistance, i.e., a 10% increase in CEO delta is associated with a 1.84% reduction in federal loan assistance.17 Overall, we find that 17 Delta and vega, as has been shown in the literature, exhibit high correlation. In our data, the correlation is 0.60. In order to assess the importance of multicollinearity on our observed results, we introduce log of delta and log of vega separately, and together, before we include all other explanatory results for all our models. In unreported tests, we find evidence that the coefficient on vega is positive and significant in all the specifications.

A. Gande, S. Kalpathy / Journal of Corporate Finance 47 (2017) 131–150

139

Table 4 Loan assistance during financial crisis and CEO equity incentives in financial firms: firm-level OLS estimates. This table provides OLS estimates from regressions of the log of the loan assistance to financial firms during the financial crisis. The Federal Reserve credit facilities start in August of 2007 and end in April of 2010. “All Programs” include sum of credit assistance received under Asset-Backed Commercial Paper Money Market Mutual Fund Liquidity Facility (AMLF), Commercial Paper Funding Facility (CPFF), Primary Dealer Credit Facility (PDCF), Term Auction Facility (TAF), Term Securities Lending Facility (TSLF), Single-tranche Open Market Operations (ST OMO) and Discount Window (DW). Data is aggregated for each firm at the program level across all years of financial assistance. Delta is the dollar change in the value of the portfolio of stock and option holdings for a 1% change in stock price. Vega is the dollar change in the value of the portfolio of option holdings for a 0.01 change in volatility of stock returns. Leverage is total liabilities divided by book value of assets. Market-to-book ratio is defined as the sum of market value of equity and book value of total liabilities divided by the book value of assets. Stock return is annual (raw) stock returns. All independent variables are as of 2006. Model (3) includes program fixed effects. Standard errors are corrected for heteroskedasticity and clustering at the firm level. t-statistics are reported in parentheses. ***, **, and * denote significance at less than 1%, 5%, and 10% levels, two-tailed tests, respectively. Program:

All programs

TAF and DW

TAF or DW

Model:

(1)

(2)

(3)

Intercept

−3.806 (−0.93) 0.092* (1.89) −0.064 (−0.83) 0.911*** (7.49) 3.361 (1.02) −1.620 (−0.64) 2.992** (2.09) 0.687 69

33.484*** (3.34) 0.101** (2.15) 0.116 (1.22) 0.457*** (2.73) −32.475*** (−3.25) −2.637 (−0.60) −0.308 (−0.16) 0.302 60

16.276* (1.88) 0.179** (2.57) −0.035 (−0.32) 0.413*** (2.75) −16.221* (−1.77) 0.729 (0.19) 0.081 (0.04) 0.293 103

Log (Vega) Log (Delta) Log (market cap) Leverage Market-to-book Stock return Adjusted R-square N

greater risk-taking incentives (vega) and lower levels of incentive alignment (delta) are associated with higher federal loan assistance. In column (2) we conduct analysis similar to column (1) using information only from TAF and DW programs. That is, we aggregate loans across TAF and DW programs on a yearly basis, and run the regression in column (1). In column (3), we include information separately from the TAF and DW programs. That is, instead of aggregating loans across TAF and DW programs on a yearly basis, we include the annual amounts of loans under these two programs separately. As a result, our unit of analysis in this regression is a firm-program year, and we include a program fixed effect in the regression in column (3). In aggregating loans, we only take into account the amount of new loans received during a particular time period. If a loan is renewed during this time period, the renewed loan is not counted as an additional loan. The results in columns (2) and (3) are qualitatively unchanged. Specifically, an increase in CEO vega is associated with a statistically significant increase in federal loan assistance.18 4.1. Federal loan assistance and risk-taking channel One potential concern with our analysis so far is that we have not identified the specific channel of excessive risk-taking, and hence our evidence may be construed as only indirect evidence of excessive risk-taking. Given the non-observability of a firm's investment policy choices, this is clearly a challenging task. Nevertheless, in this section, we attempt to provide some direct evidence of excessive risk-taking. The channel we focus on is exposure to real estate markets. We examine a firm's exposure to real estate markets by its beta coefficient, where the index we use is the Case-Shiller 20-city home price index. Specifically, we run firm-by-firm regression of monthly stock returns during the period 2000 to 2006 on Fama-French three factors (namely, size, book-to-market, and market), momentum factor, and the Case-Shiller 20-city home price index return. The coefficient of the Case-Shiller index is positive and statistically significant in these regressions. We use this coefficient as the measure of sensitivity of a firm to the real estate market. We then augment our regressions in Table 2 with our real estate sensitivity measure. As we can see from Table 3, the coefficient of Log (Vega) continues to be positive and statistically significant in all three models. Moreover, the coefficient of Real Estate Exposure is also positive and statistically significant, suggesting that firms that received larger amounts of federal financial assistance indeed had higher levels of exposure to real estate markets. 4.2. Firm-level analysis A possible concern with our research design is that a firm may appear multiple times in a regression (e.g., Table 2 Column (1)) if it received federal financial assistance in different years. We have accounted for this issue through firm-level clustering of 18 As a robustness check, we also run the regression in column (1) of Table 2 for TAF and for DW separately. We continue to find that the Log (Vega) coefficient is positive and statistically significant in both cases.

140

A. Gande, S. Kalpathy / Journal of Corporate Finance 47 (2017) 131–150

standard errors in Table 2. An alternative way to address this issue is to conduct firm-level analysis where federal financial assistance is aggregated across all years for a given firm. As a result, each firm is represented only once in the regression equation. We re-run our regressions in Table 2 using this firm-level approach. The results, presented in Table 4 using the firm-level approach, are qualitatively unchanged. That is, we continue to find that the loan amount (aggregated across all years) is positively related to the pre-crisis CEO vega, and the coefficient is statistically significant at the 10% level or better in all three regressions in Table 4. 4.3. Endogeneity in CEO equity incentives and selection issues Our analyses in Tables 2 and 4 implicitly assume that CEO equity incentives, namely Vega and Delta are exogenous. However, in reality, both of these are likely to be endogenous which could result in our OLS estimates being biased. Moreover, certain financial firms are likely to select into the emergency programs and any omitted factors that relate to this selection could be correlated with equity incentives. We address these two issues now. 4.3.1. Instrumental variables The source of endogeneity bias that we are most concerned about is any omitted variable that could be correlated with CEO incentives as well as bank policy choices. To the extent that some of these omitted variables are observable we can control for them in our loan amount regressions (for example market-to-book). Nonetheless, there are likely to be omitted variables that are unobservable to researchers. To account for this, we rely on an instrumental variables approach.

Table 5 Loan assistance during financial crisis and CEO equity incentives in recipient financial firms: 2SLS estimates. This table provides instrumental variables (2SLS) regressions of the log of the loan assistance to financial firms and CEO equity incentives. The Federal Reserve credit facilities start in August of 2007 and end in April of 2010. “All Programs” include sum of credit assistance received under Asset-Backed Commercial Paper Money Market Mutual Fund Liquidity Facility (AMLF), Commercial Paper Funding Facility (CPFF), Primary Dealer Credit Facility (PDCF), Term Auction Facility (TAF), Term Securities Lending Facility (TSLF), Single-tranche Open Market Operations (ST OMO) and Discount Window (DW). Data is aggregated for each firm at the program level for each year. Delta is the dollar change in the value of the portfolio of stock and option holdings for a 1% change in stock price. Vega is the dollar change in the value of the portfolio of option holdings for a 0.01 change in volatility of stock returns. Leverage is total liabilities divided by book value of assets. Market-to-book ratio is defined as the sum of market value of equity and book value of total liabilities divided by the book value of assets. Stock return is annual (raw) stock returns. Large CompConsult is an indicator variable equal to one if the bank employs a compensation consultant with high market share, and zero otherwise. Delta, Vega, and Large CompConsult are as of 2006. All other independent variables are contemporaneous with the program year. All models include program year fixed effects. Models (7), (8), and (9) include program fixed effects. Standard errors are corrected for heteroskedasticity and clustering at the firm level. t-statistics are reported in parentheses. ***, **, and * denote significance at less than 1%, 5%, and 10% levels, two-tailed tests, respectively. Program:

All programs

TAF and DW

TAF or DW

Log Loan Amount

Log Delta

Log Vega

Log Loan Amount

Log Delta

Log Vega

Log Loan Amount

Log Delta

Log Vega

Model:

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Intercept

7.046 (0.99) 0.442** (2.03) −0.128 (−0.50) 0.261 (1.59) −3.617 (−0.52) −2.995 (−0.62) −1.597* (−1.80)

−1.698 (−0.43)

1.630 (0.36)

−5.874 (−1.40)

−2.982 (−0.74)

−2.885 (−0.81)

0.205 (0.92) 7.585* (1.87) −3.636 (−0.75) −0.350 (−0.34) −0.285 (−1.01) 0.603*** (2.65) 1.577* (1.88)

−0.183 (−1.04) 7.390 (1.43) 2.792 (0.48) −0.794 (−0.60) 0.496** (2.18) 0.214** (2.17) 1.897** (2.03) 13.43

−0.041 (−0.20) 5.794 (1.07) 1.153 (0.25) −0.539 (−0.48) −0.426* (−1.78) 1.045*** (15.18) 1.473* (1.79)

12.473* (1.73) 0.327** (2.55) −0.230 (−1.00) 0.119 (0.92) −16.741** (−2.41) 7.390 (1.59) −1.882*** (−2.93)

−5.794 (−1.64)

−0.132 (−0.80) 5.319 (1.49) 0.672 (0.15) −0.647 (−0.58) 0.565*** (2.79) 0.092 (1.31) 1.923** (2.10) 13.43

24.177*** (2.99) 0.359** (2.03) 0.107 (0.42) −0.024 (−0.19) −25.236*** (−2.89) −0.592 (−0.10) −1.785** (−2.15)

−0.100 (−0.82) 7.153 (1.56) 2.263 (0.38) −0.826 (−0.67) 0.547*** (3.01) 0.209** (2.29) 1.309* (1.93) 13.43

0.090 (0.55) 5.366 (1.09) 1.734 (0.36) −0.639 (−0.62) −0.446** (−2.18) 1.006*** (17.69) 0.868 (1.39)

Log (Vega) Log (Delta) Log (market cap) Leverage Market-to-book Stock return Lagged Log (Delta) Lagged Log (Vega) Large CompConsult Stock and Yogo (2005) 10% critical value Kleibergen-Paap F-statistic F-stat for excluded instruments Hansen J-statistic p-Value N

126

13.01 3.71**

14.51 14.12***

27.73 25.42***

0.137 0.711 126

0.005 0.942 113

0.056 0.812 157

126

113

113

157

157

A. Gande, S. Kalpathy / Journal of Corporate Finance 47 (2017) 131–150

141

Any instrumental variable needs to satisfy two conditions for it to be a valid instrument. First, it must be correlated with the endogenous regressor (“instrument relevance”). Second, it must not be correlated with the error term (“instrument exogeneity”). We select three instruments – lagged value of vega, lagged value of delta, and large market share compensation consultant. The lagged values of vega and delta are guided by the idea that most firms use a fixed number or fixed value plan in their stock option granting practices. Companies use multi-year stock option plans based on fixed number or fixed value (Hall, 1999 and Shue and Townsend, 2017). This generates stickiness in the grant of stock options and hence the overall level of CEO incentives. As a result, following Coles et al. (2006) and Mehran and Rosenberg (2008), we use the previous year's value of CEO delta (vega) as instruments for current year delta (vega). Prior literature (e.g., Bettis et al., 2016) suggests that large market share compensation consultants play a very important role assisting the compensation committee in designing incentive contracts, so they should directly impact the design of high powered incentives. Accordingly, we use large market share compensation consultant (defined as one if the bank employs a compensation consultant with high market share, and zero otherwise) as an additional instrument. The top seven compensation consultants in our sample that we consider as large market share compensation consultants are: F.W. Cook, Hay Group, Hewitt, Mercer, Pearl Meyer, Towers Perrin and Watson Wyatt. We report our results from the instrumental variables (IV) approach in Table 5. We note that in general our instruments are correlated with delta and vega. More formally, “weak identification” arises when the excluded instruments are correlated with the endogenous regressors, but only weakly. Estimators can perform poorly when instruments are weak, and different estimators are more robust to weak instruments (e.g., LIML) than others (e.g., IV); see, e.g., Stock and Yogo (2005) for further discussion. Stock and Yogo (2005) have compiled critical values for several different estimators (IV, LIML, Fuller-LIML), several different definitions of “perform poorly” (based on bias and test size), and a range of con (up to 100 excluded instruments and up to 2 or 3 endogenous regressors, depending on the estimator). The Kleibergen-Paap F-statistic we report in Table 5 is higher than the 10% critical value of 13.43 specified in Stock and Yogo (2005), with the exception of the model involving “All Programs.” That is, we find weak instruments in model 1 with All Programs. For the other two cases in Table 5, we are able to easily reject the null that we have weak instruments. The overall F-statistics for excluded instruments are significant at 5% or better in all three models, further confirming that our three instruments are highly correlated with delta and vega. Since we have more instruments than endogenous regressors, we are able to test for instrument exogeneity using the Sargan test for overidentifying restrictions, i.e. the Hansen J-statistic. The joint null hypothesis is that the instruments are uncorrelated with the error term and that the excluded instruments are correctly excluded from the estimated model. The p-values for the Hansen J-statistic show that we are unable to reject the null hypothesis associated with the test for overidentifying restrictions which indicates that our instruments are exogenous. We find that similar to Tables 2 and 4, the loan amount is positively related to the pre-crisis CEO vega, and the coefficient is statistically significant at the 5% level or better in columns (1), (4), and (7) in Table 5. 4.3.2. Tobit analysis Results in the previous section suggest that addressing endogeneity in equity incentives does not affect our main results relating to the effects of risk-taking incentives on loan assistance. But there remains the possibility that our results are affected by a bias in the selection of certain financial firms into the loan programs. To address selection concerns, we assume that the financial institution chooses emergency loan assistance, and conditional on this decision decides on the magnitude of emergency loans. Under Heckman (1979), these two decisions are simultaneous, consisting of a probit model for the loan decision, and a linear regression for loan size.


ð1Þ

ln ð1 þ loan sizeÞ ¼ x2 β2 þ ε2

ð2Þ

L ¼ x1 β1 þ ε1

If a bank chooses loan assistance and 0 otherwise. The financial institution approaches the Federal Reserve for emergency financial assistance when the latent variable L∗ measuring the net benefits of loan assistance is greater than zero in Eq. (1). Contingent on the decision to seek loan assistance, which occurs with a probability of less than one, we assume that the size of the loan is based on the model shown in Eq. (2). When x1 =x2 and β1 is restricted to equal β2 in the Heckman model, the Tobit model results. The principal advantage of the Tobit model is that it combines Eqs. (1) and (2) into a single regression to reflect the idea that the factors that drive the emergency loan assistance decision also affect loan size. We perform firm-level analysis where loans are aggregated across all loan years, and zero loans are all financial firms not receiving assistance, and include SIC code fixed effects. The Tobit results are reported in Table 6. We find that the emergency financial assistance grant levels are strongly related to bank size, and to pre-crisis leverage, market-to-book, and stock returns. Moreover, we find strong evidence that loan assistance is negatively related to CEO delta and positively related to CEO vega. The coefficient on CEO vega is positive and statistically significant in all three specifications. The evidence based on instrumental variables and the Tobit analysis suggests that our earlier OLS results are robust to endogeneity and selection bias concerns thereby indicating the causal effect of risk-taking incentives on bank policy choices that result in the bank seeking higher emergency loan assistance from the U.S. Federal Reserve.

142

A. Gande, S. Kalpathy / Journal of Corporate Finance 47 (2017) 131–150

Table 6 Tobit estimation of loan assistance and CEO equity incentives: firm-level estimates. This table provides Tobit estimates from regressions of capital assistance to financial firms during the financial crisis. The Federal Reserve credit facilities start in August of 2007 and end in April of 2010. “All Programs” include sum of credit assistance received under Asset-Backed Commercial Paper Money Market Mutual Fund Liquidity Facility (AMLF), Commercial Paper Funding Facility (CPFF), Primary Dealer Credit Facility (PDCF), Term Auction Facility (TAF), Term Securities Lending Facility (TSLF), Single-tranche Open Market Operations (ST OMO) and Discount Window (DW). Data is aggregated for each firm at the program level across all years of financial assistance. Firms included in the Tobit analysis include firms that receive loan assistance and all finance firms that do not receive any loan assistance. Delta is the dollar change in the value of the portfolio of stock and option holdings for a 1% change in stock price. Vega is the dollar change in the value of the portfolio of option holdings for a 0.01 change in volatility of stock returns. Leverage is total liabilities divided by book value of assets. Market-to-book ratio is defined as the sum of market value of equity and book value of total liabilities divided by the book value of assets. Stock return is annual (raw) stock returns. All independent variables are as of 2006. All models include SIC code fixed effects. Standard errors are corrected for heteroskedasticity and clustering at the firm level. t-statistics are reported in parentheses. ***, **, and * denote significance at less than 1%, 5%, and 10% levels, two-tailed tests, respectively. Program:

All programs

TAF and DW

TAF or DW

Model:

(1)

(2)

(3)

Intercept

−37.827*** (−96.65) 0.145*** (4.44) −0.248*** (−8.33) 2.386*** (53.62) 23.695*** (54.55) −23.527*** (−66.47) 5.961*** (5.60) 0.283 305

−152.579*** (−361.35) 0.091** (2.57) −0.367*** (−11.32) 2.177*** (45.08) 73.629*** (158.58) −28.153*** (−73.70) 7.939*** (7.06) 0.300 305

−41.487*** (−137.52) 0.115*** (4.54) −0.396*** (−16.97) 1.631*** (47.31) 55.802*** (167.98) −20.056*** (−73.45) 7.074*** (8.62) 0.303 348

Log (Vega) Log (Delta) Log (market cap) Leverage Market-to-book Stock return Pseudo R-square N

4.4. Federal loan assistance and duration of loan assistance In this section, we examine whether CEOs of financial firms responded to pre-crisis incentives in their compensation contracts, and took on excessive risks that culminated in them receiving federal loan assistance. We hypothesized in Section 2 that firms that take on “excessive” risks are likely to have “solvency” problems which require financial assistance over a longer period of time. In contrast, firms that face “liquidity” problems are likely to receive financial assistance only for a shorter period of time. We measure the duration of federal financial assistance in terms of the total days in debt (i.e., as a result of federal financial assistance, how many days is the firm in debt under all Federal Reserve programs). We test H2 and present the results in Table 7. We find that the coefficient of Log (Vega) is positive and statistically significant. In other words, CEOs that had the highest incentive in taking risk, as measured by high levels of vega received federal financing for a longer period of time, consistent with an interpretation that they may have acted upon incentives to take on excessive risk.

4.5. Riskiness of assets and loan collateral requirements An additional way to examine the riskiness of assets prior to the crisis would be to investigate whether the collateral requirement for the emergency loans is increasing in the proportion of risky assets (e.g., trading assets, non-agency mortgage-backed securities etc.) prior to the crisis.19 All else equal, riskier loans would require higher levels of collateral. In addressing this question, we use information on collateral for banks that received TAF assistance during the financial crisis. This data was made available publicly by the U.S. Federal Reserve. We focus on TAF because there is information on the various assets that are placed as collateral at the loan level. We find evidence that the ratio of collateral-to-loan is positively related to both the ratio of trading assets-to-total assets as well as the ratio of non-agency mortgage-backed securities-to-total assets held by the bank prior to the financial crisis. Pearson correlation coefficient between collateral-to-loan ratio and trading assetsto-total assets is 0.105 (p-value = 0.041) and between collateral-to-loan ratio and non-agency mortgage-backed securities-tototal assets is 0.162 (p-value = 0.002). Spearman correlation coefficient between trading assets-to-total assets and collateralto-loan ratio is 0.261 (p-value = 0.000) and between non-agency mortgage-backed securities-to-total assets and collateral-toloan ratio is 0.123 (p-value = 0.017). This evidence is consistent with the notion that banks were required to post higher collateral as a percentage of loan amount if they were perceived to be riskier.

19 Berger and Udell (1990) document that collateral pledged in commercial loans plays an important role in explaining riskiness of loans and lenders. Drechsler et al. (2016) show that banks that borrowed heavily from LOLR facilities during the recent financial crisis also used risky collateral.

A. Gande, S. Kalpathy / Journal of Corporate Finance 47 (2017) 131–150

143

Table 7 OLS Estimation of Days-in-debt and CEO Equity Incentives: FirmLevel Estimates. This table provides OLS regressions of total daysin-debt (under All Fed Programs) and CEO equity incentives. The Federal Reserve credit facilities start in August of 2007 and end in April of 2010. “All Programs” include sum of credit assistance received under Asset-Backed Commercial Paper Money Market Mutual Fund Liquidity Facility (AMLF), Commercial Paper Funding Facility (CPFF), Primary Dealer Credit Facility (PDCF), Term Auction Facility (TAF), Term Securities Lending Facility (TSLF), Single-tranche Open Market Operations (ST OMO) and Discount Window (DW). Data is aggregated for each firm at the program level across all years of financial assistance. Delta is the dollar change in the value of the portfolio of stock and option holdings for a 1% change in stock price. Vega is the dollar change in the value of the portfolio of option holdings for a 0.01 change in volatility of stock returns. Leverage is total liabilities divided by book value of assets. Market-to-book ratio is defined as the sum of market value of equity and book value of total liabilities divided by the book value of assets. Stock return is annual (raw) stock returns. All independent variables are as of 2006. Standard errors are corrected for heteroskedasticity and clustering at the firm level. t-statistics are reported in parentheses. ***, **, and * denote significance at less than 1%, 5%, and 10% levels, two-tailed tests, respectively. Model:

(1)

Intercept

105.167 (0.19) 14.994** (2.00) −27.195*** (−2.90) 29.478* (1.99) −235.884 (−0.74) 142.343 (0.39) 114.305 (0.89) 0.206 69

Log (Vega) Log (Delta) Log (market cap) Leverage Market-to-book Stock return Adjusted R-square N

4.6. CEO overconfidence An alternative explanation is that bank risk taking (evidenced in the form of a larger bailout for a financial institution as shown in our study) may have resulted from overconfident CEOs rather than equity incentives. For instance, Ho et al. (2016) show that bank CEO overconfidence is related to higher risk-taking prior to the financial crisis and as evidence they demonstrate that banks with overconfident CEOs are more likely to loosen lending standards and increase leverage compared to other banks. However, their analysis does not control for explicit CEO incentives. We augment our main tables (i.e., Tables 2 and 4) with a CEO overconfidence measure. We follow Ho et al. (2016) and define overconfident CEO as an indicator variable that takes the value one if the CEO is overconfident, and is zero otherwise. The empirical measure of CEO overconfidence employed in Ho et al. (2016) is similar to the definition in Campbell et al. (2011) whereby CEOs are defined as overconfident if they postpone exercising stock options that are more than 100% out-of-themoney at least twice during their tenure. CEOs are flagged as being overconfident from the first time they postpone stock option exercise. We present the results in columns (1) thru (6) of Table 8. We find that the overconfidence measure is significant in just one specification at the 10% level. More importantly, the coefficient on vega is positive and significant in all six specifications. As an additional robustness test, we also specify the model by defining loan assistance as the size of the loans scaled by a bank's total assets. These results are shown in columns (7) thru (9) of Table 8. The coefficient on CEO overconfidence measure is positive and significant at 5% in model (7) and at 10% in model (9). As before, the coefficient on vega continues to be positive and significant in all three specifications. Overall, from the nine regression specifications in Table 8 (loan-year, firm-level, and loan-to-assets), we conclude that risk-taking incentives play an important role in explaining emergency loan assistance even after we explicitly control for CEO overconfidence.

144

A. Gande, S. Kalpathy / Journal of Corporate Finance 47 (2017) 131–150

Table 8 Loan assistance during financial crisis, CEO equity incentives in recipient financial firms, and CEO overconfidence: OLS estimates. This table provides OLS estimates from regressions of loan assistance to financial firms during the financial crisis. The Federal Reserve credit facilities start in August of 2007 and end in April of 2010. “All Programs” include sum of credit assistance received under Asset-Backed Commercial Paper Money Market Mutual Fund Liquidity Facility (AMLF), Commercial Paper Funding Facility (CPFF), Primary Dealer Credit Facility (PDCF), Term Auction Facility (TAF), Term Securities Lending Facility (TSLF), Single-tranche Open Market Operations (ST OMO) and Discount Window (DW). Data is aggregated for each firm at the program level for each year. Delta is the dollar change in the value of the portfolio of stock and option holdings for a 1% change in stock price. Vega is the dollar change in the value of the portfolio of option holdings for a 0.01 change in volatility of stock returns. Leverage is total liabilities divided by book value of assets. Market-to-book ratio is defined as the sum of market value of equity and book value of total liabilities divided by the book value of assets. Stock return is annual (raw) stock returns. Overconfident CEO is an indicator variable that takes the value one if the CEO is overconfident and zero otherwise. CEO overconfidence measures follows definition in Campbell et al. (2011). Standard errors are corrected for heteroskedasticity and clustering at the firm level. t-statistics are reported in parentheses. ***, **, and * denote significance at less than 1%, 5%, and 10% levels, two-tailed tests, respectively. Program:

All programs

TAF and DW

TAF or DW

All programs

TAF and DW

TAF or DW

Log loan assistance: firm-level

Log loan assistance: loan-years

All programs

TAF and DW

TAF or DW

Loan-to-assets: loan-years

Model:

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Intercept

−4.854 (−0.79) 0.170** (2.19) −0.225** (−2.39) 0.665*** (6.25) −0.026 (−0.00) 3.062 (0.62) −2.367*** (−4.27) 0.700 (1.65) 0.450 137

11.285 (1.11) 0.217*** (3.22) 0.075 (0.85) 0.259* (1.91) −22.849** (−2.14) 7.299 (0.90) −2.528*** (−2.92) −0.299 (−0.56) 0.381 124

3.834 (0.43) 0.198*** (5.26) −0.031 (−0.43) 0.251** (2.04) −14.872* (−1.73) 8.949 (1.42) −1.853** (−2.45) 0.308 (0.72) 0.360 174

−3.545 (−0.87) 0.098* (1.70) −0.109 (−1.29) 0.933*** (7.68) 3.588 (1.08) −1.881 (−0.75) 2.323 (1.62) 0.508* (1.69) 0.693 69

33.630*** (3.30) 0.102** (2.14) 0.107 (1.07) 0.460*** (2.71) −32.524*** (−3.21) −2.695 (−0.61) −0.445 (−0.22) 0.110 (0.23) 0.289 60

17.048** (2.04) 0.184** (2.62) −0.072 (−0.64) 0.425*** (2.87) −16.443* (−1.83) 0.338 (0.09) −0.472 (−0.25) 0.487 (1.34) 0.295 103

−76.049* (−1.80) 0.613** (2.22) −1.218** (−2.21) −0.559 (−1.07) 4.859 (0.26) 78.323* (1.79) −9.064*** (−2.68) 4.983** (2.15) 0.137 137

−14.750 (−0.52) 0.458** (2.14) −0.641* (−1.90) −1.413*** (−3.66) −11.367 (−0.35) 38.965 (1.65) −7.224** (−2.35) 2.210 (1.11) 0.167 124

−35.895 (−1.01) 0.560* (1.86) −1.101 (−1.51) −0.933** (−2.56) −33.026 (−1.10) 77.051 (1.53) −5.748** (−2.10) 3.266* (1.72) 0.134 174

Log (Vega) Log (Delta) Log (market cap) Leverage Market-to-book Stock return Overconfident CEO Adjusted R-square N

4.7. Funding base An alternative explanation is the funding base of a borrowing firm. That is, whether a financial firm with limited access to deposit financing is more likely to access the federal emergency loans and whether this explains our evidence. In Table 9 we augment our regressions in Table 6 with deposits-to-liabilities ratio, a proxy for a firm's funding base. We specify a Tobit model given that the issue that we are most concerned about is whether certain financial firms select into the federal emergency loan programs given their funding base. In columns (1) through (3) we include leverage. But leverage and deposits-to-liabilities are highly correlated (correlation coefficient in our sample is 0.55) so in columns (4) through (6) we drop the leverage variable. We find that as expected a firm that relies more on deposits is less likely to use federal emergency loans. Nevertheless, we find that our measure of CEO risk-taking incentives continues to remain statistically significant suggesting that CEO incentives is an important channel that explains bank risk-taking even after we control for a financial firm's funding base. 4.8. Information content of federal loan assistance We conduct a principal component analysis (PCA) to determine the information content of the federal loan assistance. Specifically, we start with a set of variables that are potentially associated with risk-taking preceding the financial crisis, namely: sensitivity to real estate exposure, non-agency mortgage backed securities, commercial and industrial loans, leverage, CEO overconfidence, and funding base measured by the ratio of deposits to total liabilities. Panel A of Table 10 provides the loadings on the six principal components. We focus on the principal components20 whose Eigenvalues are greater than 1, which in our case were the first three principal components. These three principal components cumulatively explain about 70% of the variance of the variables associated with bank risk-taking. As can be seen from the loadings in Panel A, the first component is related to a bank's commercial and industrial loans, and funding base. The second component is related mainly to a bank's exposure to real estate, its holdings of non-agency mortgage backed securities, and its CEO's overconfidence. The third principal component is associated with a bank's leverage. 20 The objective of PCA is to find unit-length linear combinations of the variables with the greatest variance. The first principal component has maximal overall variance. The second principal component has maximal variance among all unit length linear combinations that are uncorrelated to the first principal component, etc. The last principal component has the smallest variance among all unit length linear combinations of the variables. All principal components combined contain the same information as the original variables, but the important information is partitioned over the components in a particular way: the components are orthogonal, and earlier components contain more information than later components. PCA thus is just a linear transformation of the data. It does not assume that the data satisfy a specific statistical model, though it does require that the data be interval-level data—otherwise taking linear combinations is meaningless.

A. Gande, S. Kalpathy / Journal of Corporate Finance 47 (2017) 131–150

145

Table 9 Loan assistance during financial crisis and bank funding base. This table provides Tobit estimates from regressions of capital assistance to financial firms during the financial crisis. The Federal Reserve credit facilities start in August of 2007 and end in April of 2010. “All Programs” include sum of credit assistance received under AssetBacked Commercial Paper Money Market Mutual Fund Liquidity Facility (AMLF), Commercial Paper Funding Facility (CPFF), Primary Dealer Credit Facility (PDCF), Term Auction Facility (TAF), Term Securities Lending Facility (TSLF), Single-tranche Open Market Operations (ST OMO) and Discount Window (DW). Data is aggregated for each firm at the program level across all years of financial assistance. Firms included in the Tobit analysis include firms that receive loan assistance and all finance firms that do not receive any loan assistance. Delta is the dollar change in the value of the portfolio of stock and option holdings for a 1% change in stock price. Vega is the dollar change in the value of the portfolio of option holdings for a 0.01 change in volatility of stock returns. Leverage is total liabilities divided by book value of assets. Marketto-book ratio is defined as the sum of market value of equity and book value of total liabilities divided by the book value of assets. Stock return is annual (raw) stock returns. Deposits-to-Liabilities is the ratio of total deposits to total liabilities. All independent variables are as of 2006. All models include SIC code fixed effects. Standard errors are corrected for heteroskedasticity and clustering at the firm level. t-statistics are reported in parentheses. ***, **, and * denote significance at less than 1%, 5%, and 10% levels, two-tailed tests, respectively. Program:

All programs

TAF and DW

TAF or DW

All programs

TAF and DW

TAF or DW

Model:

(1)

(2)

(3)

(4)

(5)

(6)

−53.186*** (−121.38) 0.091** (2.50) −0.368*** (−11.02) 2.162*** (43.80) 73.065*** (151.95) −27.962*** (−70.43) 7.931*** (6.99) −0.516 (−0.98) 0.300 305

−41.924*** (−133.93) 0.116*** (4.46) −0.395*** (−16.45) 1.671*** (47.41) 57.313*** (166.56) −20.538*** (−72.35) 7.096*** (8.57) 1.280*** (3.40) 0.303 348

−24.336*** (−62.46) 0.123*** (3.81) −0.223*** (−7.55) 2.383*** (54.36)

−17.613*** (−41.17) 0.039 (1.08) −0.203*** (−6.19) 1.983*** (40.94)

−16.063*** (−52.30) 0.079*** (3.10) −0.249*** (−10.51) 1.480*** (42.56)

−24.616*** (−69.62) 5.947*** (5.75) −5.159*** (−10.79) 0.279 305

−27.745*** (−71.51) 7.365*** (6.72) −5.644*** (−10.82) 0.286 305

−20.267*** (−72.74) 6.615*** (8.24) −3.019*** (−8.11) 0.290 348

−37.202*** (−92.36) Log (Vega) 0.143*** (4.30) Log (Delta) −0.266*** (−8.73) Log (market cap) 2.273*** (50.32) Leverage 23.100*** (51.66) Market-to-book −22.240*** (−60.96) Stock return 6.023*** (5.62) Deposits-to-liabilities −4.782*** (−9.86) Pseudo R-square 0.284 N 305

Intercept

We examine how these three principal components relate to the crisis period emergency loan assistance and the bank's performance and report the correlation coefficients in Panel B. We follow the methodology in Fahlenbrach and Stulz (2011) to define a bank's crisis period performance reflected in the buy-and-hold stock returns as well as the return-on-equity. Results in Panel B indicate that the federal emergency loan assistance is correlated with the first two of these three principal components at 5% level or better in terms of statistical significance. In contrast, the stock performance variable and the accounting measure of performance is correlated with only one of the three principal components at 5% level or better in terms of statistical significance. This evidence confirms that the fed loan programs contain highly relevant information relating to a bank's risk-taking activities prior to the financial crisis, consistent with our interpretation in footnotes 11 and 12. This leads to the question as to why we use the federal loan assistance as the primary measure rather than the principal components themselves to evaluate the role of risk-taking incentives. We report the results of the regression of the three principal components on CEO vega and other determinants in Panel C. Results suggest that there is no reliable relation between the principal components considered individually one-at-a time and CEO vega. Overall, the principal component analysis indicates that the federal emergency loan measures capture some of the salient components of a bank's risk-related activities prior to the crisis. Importantly, though, inferences relating to the effect of a bank CEO's risk-taking incentives are only reliably inferred from the federal loan assistance. 4.9. CEO education and bank matching An alternative explanation that could drive our results is whether CEOs with certain background (education) self-select into certain banks and whether this is correlated with bank policy choices, and hence bank performance. Accordingly, we examine whether our results continue to hold whilst explicitly controlling for endogenous firm-manager matching. For a detailed overview of the executive-firm match literature and why controlling for endogenous executive-firm matching would be beneficial in our context, see Pan (2017). We follow King et al. (2016) and collect data on bank CEO education for all financial firms in Execucomp (a subset of which enter our sample) where available. We manually search for CEO educational information from the company's proxy statements, executive biographies from the internet, and LinkedIn profiles. We augment our loan regressions (i.e., Table 2) with CEO educational variables as in King et al. (2016).21 That is, we take the predicted probability from logit in the first stage for each observation (using the covariates on education) and weight the 21 We cannot estimate the top 20 PhD variable since we cannot identify any CEO in our sample that obtained a PhD from a top 20 US institution, according to 2006 US News and World Report university rankings.

146

A. Gande, S. Kalpathy / Journal of Corporate Finance 47 (2017) 131–150

Table 10 Information content of loans: principal component analysis. This table provides results of the principal component analysis (PCA). The sample is described in Table 1. Panel A reports details of the six principal components with the Eigenvalues. Panel B reports the correlation of the three main principal components with federal loan assistance and firm performance measures. Panel C provides the results of OLS regressions of the three principal components on CEO equity incentives. All three models in Panel C include program fixed effects. t-statistics are reported in parentheses. ***, **, and * denote significance at less than 1%, 5%, and 10% levels, two-tailed tests, respectively. Panel A: Principal components Component 1

Component 2

Component 3

Component 4

Component 5

Component 6

Real estate exposure Non-agency MBS/assets C&I loans/assets Leverage Overconfident CEO Deposits-to-liabilities

−0.237 0.0112 0.6592 −0.1954 −0.133 0.6733

0.6513 0.4978 0.1584 0.1901 0.4701 0.214

0.0787 −0.4479 0.2393 0.8538 −0.076 0.0336

0.0296 0.5983 −0.062 0.2618 −0.7541 −0.0119

−0.6944 0.3838 −0.2171 0.3575 0.4168 0.1478

−0.1744 0.215 0.6574 −0.018 0.1143 −0.6912

Eigenvalues Cumulative variance

1.754 0.292

1.452 0.534

1.020 0.704

0.952 0.863

0.561 0.957

0.261 1.000

Panel B: Correlation of principal components with loan assistance and firm performance during the financial crisis

Component 1 Component 2 Component 3

Federal emergency loans

Crisis period buy-and-hold stock returns

Crisis period return on equity

−0.307** 0.397*** 0.151

0.457*** −0.058 −0.161

0.268** 0.089 −0.115

Panel C: OLS regressions of principal components on CEO equity incentives Component 1

Component 2

Component 3

Model:

(1)

(2)

(3)

Intercept

4.500 (1.22) −0.004 (−0.20) −0.052* (−1.77) −0.258*** (−4.13) −6.129** (−2.30) 2.564 (1.38) −0.658 (−0.89) 69

−9.578*** (−3.57) −0.001 (−0.04) 0.188*** (3.68) 0.060 (0.62) 4.430** (2.10) 1.357 (0.77) 2.213** (2.21) 69

−17.809*** (−17.35) 0.000 (0.02) −0.027 (−1.58) −0.035 (−1.12) 20.074*** (25.51) 0.193 (0.29) −0.568 (−1.54) 69

Log (Vega) Log (Delta) Log (market cap) Leverage Market-to-book Stock return N

observations by the inverse of that probability in our second stage (e.g., loan regressions) and additionally include the education variables in the loan regressions. The first-stage logit models whether the bank is in the highest decile of size (i.e. book value of assets). The results are presented in Table 11. We find that the coefficient of Log (Vega) continues to be positive and statistically significant in all the regression even after we account for endogenous firm-CEO education matching. However, the CEO education variables are not statistically significant in Table 11. As additional robustness, we also run bank performance regressions using our sample similar to the analysis in Fahlenbrach and Stulz (2011). That is, we model buy and hold stock returns, return on equity and return on assets as the dependent variables, and follow King et al. (2016) by including CEO educational variables as described above. The results are shown in Table 12. Similar to Fahlenbrach and Stulz (2011), we find that the coefficient on Log (Vega) is not statistically significant whereas the coefficient on Log (Delta) is negative and statistically significant. We find that having a CEO with a MBA from a top 20 school is beneficial to a bank in improving the buy and hold returns (see column 1 of Table 12). When bank performance is measured in terms of return on assets, having a CEO with a UG degree from a top 20 school is beneficial to a bank in improving its return on assets (see column 3 of Table 12). Thus, even though we did not find CEO educational variables significant in the loan regressions, the fact that they are significant when using traditional measures of bank performance suggests that the CEO education variables are constructed correctly and have a relation with bank performance consistent with prior literature.

4.10. Additional robustness tests We conduct additional robustness tests to investigate whether our results are attributable to firm complexity, other components of compensation, the effect of outliers, and whether there is a mechanical relation between loan assistance and CEO vega due to option moneyness.

A. Gande, S. Kalpathy / Journal of Corporate Finance 47 (2017) 131–150

147

Table 11 Loan assistance during financial crisis and CEO equity incentives in recipient financial firms: correcting for CEO education and firm matching. This table provides OLS estimates from regressions of the log of the loan assistance to financial firms during the financial crisis. The Federal Reserve credit facilities start in August of 2007 and end in April of 2010. “All Programs” include sum of credit assistance received under Asset-Backed Commercial Paper Money Market Mutual Fund Liquidity Facility (AMLF), Commercial Paper Funding Facility (CPFF), Primary Dealer Credit Facility (PDCF), Term Auction Facility (TAF), Term Securities Lending Facility (TSLF), Singletranche Open Market Operations (ST OMO) and Discount Window (DW). Data is aggregated for each firm at the program level for each year. Delta is the dollar change in the value of the portfolio of stock and option holdings for a 1% change in stock price. Vega is the dollar change in the value of the portfolio of option holdings for a 0.01 change in volatility of stock returns. Leverage is total liabilities divided by book value of assets. Market-to-book ratio is defined as the sum of market value of equity and book value of total liabilities divided by the book value of assets. Stock return is annual (raw) stock returns. UG (MBA) is an indicator variable if the CEO obtained an undergraduate (MBA) degree from a university ranked in the top 20 of the 2006 US News and World Report university rankings. The regressions are weighted by the inverse of the predicted probability from the first-stage logit regression. The first-stage logit regression models whether a financial firm is in the top decile of total assets where the covariates are UG and MBA. The Model (3) includes program fixed effects. All models include program year fixed effects. Standard errors are corrected for heteroskedasticity and clustering at the firm level. t-statistics are reported in parentheses. ***, **, and * denote significance at less than 1%, 5%, and 10% levels, twotailed tests, respectively. Program:

All programs

TAF and DW

TAF or DW

Model:

(1)

(2)

(3)

Intercept

−3.579 (−0.56) 0.204** (2.54) −0.240** (−2.59) 0.583*** (5.25) −0.928 (−0.14) 3.543 (0.73) −2.249*** (−4.00) −0.060 (−0.10) 0.769 (1.25) 0.438 137

10.795 (1.01) 0.249** (2.64) 0.007 (0.05) 0.203 (1.27) −21.530** (−2.01) 7.232 (0.93) −2.470*** (−2.95) −0.016 (−0.02) 0.673 (0.74) 0.381 124

6.630 (0.66) 0.241*** (3.92) −0.085 (−0.89) 0.167 (1.36) −16.186* (−1.88) 9.549 (1.62) −1.798** (−2.59) 0.189 (0.22) 0.852 (1.25) 0.365 174

Log (Vega) Log (Delta) Log (market cap) Leverage Market-to-book Stock return UG MBA Adjusted R-square N

4.10.1. Firm complexity Large financial firms which received a bulk of federal financial assistance could be more complex to manage, and our results may be picking up firm-complexity. For example, large financial firms on Wall Street have a substantial proprietary trading business, and these firms place greater emphasis on proprietary trading of securities. Consequently, their executives might need to be incentivized to undertake these risky activities. Alternatively, firms with greater complexity of operations might be more likely to receive larger federal financial assistance during the crisis because they impose greater systemic risks due to their complexity and this omitted complexity could potentially be driving both CEO vega as well as loan assistance. We note that our regressions include a commonly used proxy for firm-size, namely log of total assets. Nevertheless, to account for an omitted component of firm-complexity, we augment our regressions in Table 2 with an additional explanatory variable that captures the extent of proprietary trading. Specifically, we augment the explanatory variables in Table 2 with log (trading assets) and redo our analysis. We define trading assets as the disclosed value of assets from trading, measured in millions of U.S. dollars. We are constrained by data limitations in measuring trading assets. We use the following procedure: First, we examine if a financial firm reports this information in Compustat business segment data. For example, if a financial firm, such as Goldman Sachs or Lehman Brothers reports multiple segments with the segment SIC code 6211 (“Security Brokers, Dealers and Flotation Companies”), we use that information. Since financial firms do not consistently use a standardized segment name, we pick the total assets of the segment with the largest value in terms of segment assets when a financial firm reports multiple segments under SIC code 6211. We verified that this process reliably picks the segment for a financial firm that best captures where trading assets are likely to be reported. Second, if a financial firm does not report this information in Compustat segment data, for example, depository institutions, such as J.P. Morgan Chase or Bank of America, we use Bank Compustat data. Specifically we sum the values of IST (“Investment Securities – Total”) and TDST (“Trading/ Dealing Account Securities – Total”) and use it as our estimate of proprietary trading assets. In unreported tests (available from us upon request) we find evidence that the coefficient of Log (Vega) continues to be positive and statistically significant at the 5% level (or better) in all three models after we account for firm-complexity based on proprietary trading. 4.10.2. Other components of compensation The focus of our analysis is on managerial incentive variables, namely CEO vega and delta. In order to develop a comprehensive understanding about the effects of other components of compensation, we include the ratio of CEO bonus-to-salary and CEO

148

A. Gande, S. Kalpathy / Journal of Corporate Finance 47 (2017) 131–150

Table 12 Financial firm performance during financial crisis and CEO equity incentives: correcting for CEO education and firm matching. This table provides OLS estimates from regressions of firm performance during the financial crisis on CEO equity incentives. The financial firms are those that received emergency loan assistance during the financial crisis. The Federal Reserve credit facilities start in August of 2007 and end in April of 2010. Data is aggregated for each firm at the program level across all years of financial assistance. Buy-and-hold returns for firms is calculated as the compounded stock returns from July 2007 to December 2008. Return on equity is defined as the cumulative quarterly net income from 2007Q3 to 2008Q3 divided by the book value of common equity at the end of 2007Q2. Return on assets is defined as the cumulative quarterly net income from 2007Q3 to 2008Q3 divided by the book value of total assets at the end of 2007Q2. Delta is the dollar change in the value of the portfolio of stock and option holdings for a 1% change in stock price. Vega is the dollar change in the value of the portfolio of option holdings for a 0.01 change in volatility of stock returns. Leverage is total liabilities divided by book value of assets. Market-to-book ratio is defined as the sum of market value of equity and book value of total liabilities divided by the book value of assets. Stock return is annual (raw) stock returns. All independent variables are as of 2006. UG (MBA) is an indicator variable if the CEO obtained an undergraduate (MBA) degree from a university ranked in the top 20 of the 2006 US News and World Report university rankings. The regressions are weighted by the inverse of the predicted probability from the first-stage logit regression. The first-stage logit regression models whether a financial firm is in the top decile of total assets where the covariates are UG and MBA. Standard errors are corrected for heteroskedasticity and clustering at the firm level. t-statistics are reported in parentheses. ***, **, and * denote significance at less than 1%, 5%, and 10% levels, two-tailed tests, respectively. Buy-and-hold returns

Return on equity

Return on assets

Model:

(1)

(2)

(3)

Intercept

0.525 (0.52) 0.019 (1.11) −0.065*** (−2.82) −0.061** (−2.01) −1.190* (−1.74) 1.130** (2.11) −0.546** (−2.14) 0.184 (1.42) 0.210** (2.25) 0.296 69

−0.611 (−1.18) 0.007 (0.97) −0.033*** (−4.11) −0.010 (−0.58) −0.120 (−0.28) 1.063*** (4.27) −0.014 (−0.06) 0.096* (1.76) 0.035 (0.44) 0.182 69

−0.099 (−1.55) 0.001 (1.23) −0.002*** (−2.96) −0.002 (−1.04) 0.028 (0.43) 0.098*** (5.49) 0.001 (0.10) 0.007* (1.76) 0.006 (1.29) 0.174 69

Log (Vega) Log (Delta) Log (market cap) Leverage Market-to-book Stock return UG MBA Adjusted R-square N

ownership as additional independent variables. Such an approach can also be found in Fahlenbrach and Stulz (2011). Specifically, we re-run our regressions in Table 2 after including CEO bonus to salary and CEO ownership as additional independent variables. In unreported tests (available from us upon request), we find that the coefficient on vega continues to be positive and statistically significant at the 5% level in all the regression specifications. Moreover, there is no association between emergency loan assistance and the ratio of CEO bonus-to-salary and CEO ownership suggesting that when one looks at a comprehensive set of CEO compensation variables, CEO risk-taking incentives alone explains the variation in emergency loan assistance. Based on this evidence we conclude that risk-enhancing investment and financial policies among financial firms prior to the financial crisis were largely driven by pre-crisis managerial risk-taking incentives. 4.10.3. Outliers Our primary tests employ the logarithmic transformation of federal loan assistance and managerial incentive measures to address problems arising from skewness in these measures. For robustness purposes, we re-run analysis (Table 2) by taking the untransformed loan and compensation measures and using median regressions to reduce the impact of any outliers. Our results (not reported separately in a table) are qualitatively similar. The coefficient on vega is positive and statistically significant. 4.10.4. Option out-of-the-moneyness An important concern is that vega may be mechanically, rather than causally, related to federal loan assistance. For a given current stock price and holding other option pricing inputs constant, vega will be higher when the options are out-of-the money, i.e., the strike price exceeds the stock price. Banks that have performed poorly over several years prior to the onset of the financial crisis are then likely to have higher CEO vega than banks that have performed well. If poorly performing banks are also those most likely to be adversely affected by the crisis (because their business model is poor), the mechanical result will obtain. To address this concern, we analyzed the distribution of moneyness (defined as stock price divided by exercise price, or S/X) for our financial firms in 2006. Since there are enhanced disclosures (starting in 2006) on the full portfolio of options, i.e., for each tranche of options the CEO holds, we know precisely the exercise price of that trance of stock options, we calculate S/X for that tranche and then take a weighted average of S/X across all tranches of stock option holdings. We find that except for one case, the options are at or in-the-money. That is, the banks in our sample have been performing well, and this should alleviate the above concern.

A. Gande, S. Kalpathy / Journal of Corporate Finance 47 (2017) 131–150

149

4.10.5. TARP assistance Several financial firms received assistance under Troubled Asset Relief Program (TARP) during the financial crisis. Our study excludes firms receiving TARP assistance for the following reasons: TARP was equity-financing as opposed to the other programs that we study here which are debt-financing; and a number of healthy firms were included in TARP since regulators were concerned that inclusion of weak firms alone could send an adverse signal to market participants about firms selected for TARP assistance. In contrast, there were no such concerns with the loan programs we study here since the identity of recipients was revealed well after the financial crisis had subsided. See Bayazitova and Shivdasani (2012) for additional details on TARP, and for evidence on the stronger banks exiting the TARP program at the earliest opportunity they had. See also Veronesi and Zingales (2010) for a discussion of how major banks were forced to participate in TARP by the U.S. Treasury. Nonetheless, we repeat our main analysis in Table 2 by replacing log of loan assistance with log of TARP assistance received by banks in 2008 and 2009 and find evidence (not reported in the paper) that the coefficient on Log (Vega) is 0.044 and is significant at the 10% level. The weak evidence relating to TARP is consistent with the idea that healthier financial institutions were included in the TARP program by regulators to avoid sending adverse signals to the market about the types of financial firms selected for capital infusion. 5. Conclusion The financial crisis of 2008 has generated a debate on the role played by CEO compensation in financial firms. We use a novel dataset of emergency loans provided by the Federal Reserve during the financial crisis, and find evidence that an increase in CEO risk-taking incentives in the pre-crisis period is associated with a significant increase in financial assistance to firms during the crisis. Our evidence of a positive association between loan assistance from the Federal Reserve and CEO risk-taking incentives call into question the role of high powered incentives for systemically important financial firms. Current efforts on regulating compensation at financial firms tended to lean towards a one-size fits all strategy (e.g., curbs on the level of compensation). Our evidence suggests that future policy work should focus on incorporating attributes of pay and structure of compensation of CEOs at financial firms in curbing excessive risk-taking, such as incorporating the incentive features of CEO compensation (e.g., CEO vega) into the pricing of FDIC insurance (see John et al., 2000). References Acharya, V., Fleming, M., Hrung, W., Sarkar, A., 2017. Dealer financial conditions and lender-of-last resort facilities. J. Financ. Econ. 123, 81–107. Adams, R., Mehran, H., 2003. Is corporate governance different for bank holding companies? Fed. Reserv. Bank N. Y. Econ. Policy Rev. 9, 123–142. Adams, R., Mehran, H., 2012. Bank board structure and performance: evidence for large bank holding companies. J. Financ. Intermed. 21 (2), 243–267. Aebi, V., Sabato, G., Schmid, M., 2012. Risk management, corporate governance, and bank performance in the financial crisis. J. Bank. Financ. 36, 3213–3226. Armantier, O., Ghysels, E., Sarkar, A., Shrader, J., 2015. Discount window stigma during the 2007–2008 financial crisis. J. Financ. Econ. 118, 317–335. Bai, G., Elyasiani, E., 2013. Bank stability and managerial compensation. J. Bank. Financ. 37, 799–813. Bayazitova, D., Shivdasani, A., 2012. Assessing TARP. Rev. Financ. Stud. 25, 377–407. Bebchuk, L., Spamann, H., 2010. Regulating Bankers' Pay. Georgetown Law J. 98, 247–287. Beltratti, A., Stulz, R.M., 2012. The credit crisis around the globe: why did some banks perform better? J. Financ. Econ. 105, 1–17. Bennett, R., Guntay, L., Unal, H., 2015. Inside debt, bank default risk, and performance during the crisis. J. Financ. Intermed. 24, 487–513. Berger, A.N., Udell, G.F., 1990. Collateral, loan quality and bank risk. J. Monet. Econ. 25, 21–42. Berger, A.N., Black, L.K., Bouwman, C.H.S., Dlugosz, J., 2016. The Federal Reserve's Discount Window and TAF Programs: ‘Pushing on a String?’. Unpublished working paper. University of South Carolina. Bettis, C., Bizjak, J., Coles, J., Kalpathy, S., 2016. Performance-vesting Provisions in Executive Compensation. Unpublished working paper. Texas Christian University. Bhagat, S., Bolton, B., 2014. Financial crisis and bank executive incentive compensation. J. Corp. Finan. 25, 313–341. Bolton, P., Mehran, H., Shapiro, J., 2015. Executive compensation and risk taking. Rev. Financ. 19, 2139–2181. Boyd, J., Jagannathan, R., Kwak, S., 2010. What caused the current financial mess and what can we do about it? J. Invest. Manag. 7, 4–20. Campbell, T., Gallmeyer, M., Johnson, S., Rutherford, J., Stanley, B., 2011. CEO optimism and forced turnover. J. Financ. Econ. 101, 695–712. Carpenter, J., 2000. Does option compensation increase managerial risk appetite? J. Financ. 55, 2311–2331. Chen, C., Steiner, T., Whyte, A., 2006. Does stock-based executive compensation induce risk-taking? An analysis of the banking industry. J. Bank. Financ. 30, 915–945. Cheng, I., Hong, H., Scheinkman, J., 2015. Yesterday's heroes: compensation and risk at financial firms. J. Financ. 70, 839–879. Chesney, M., Stromberg, J., Wagner, A., 2016. Managerial Incentives to Take Asset Risk. Unpublished working paper. University of Zurich. Cohen, R., Hall, B., Viceira, L., 2000. Do Executive Stock Options Encourage Risk-Taking? Unpublished working paper. Harvard University Coles, J., Daniel, N., Naveen, L., 2006. Managerial incentives and risk-taking. J. Financ. Econ. 79, 431–468. Core, J., Guay, W., 2002. Estimating the value of employee stock option portfolios and their sensitivities to price and volatility. J. Account. Res. 40, 613–630. Crawford, A., Ezzell, J., Miles, J., 1995. Bank CEO pay-performance relations and the effects of deregulation. J. Bus. 68, 231–256. Demsetz, H., Lehn, K., 1985. The structure of corporate ownership: causes and consequences. J. Polit. Econ. 93, 1155–1177. DeYoung, R., Peng, E., Yan, M., 2013. Executive compensation and business policy choices at U.S. commercial banks. J. Financ. Quant. Anal. 48, 165–196. Drechsler, I., Drechsel, T., Marques-Ibanez, D., Schnabl, P., 2016. Who borrows from the lender of last resort? J. Financ. 71, 1933–1974. Erkens, D., Hung, M., Matos, P., 2012. Corporate governance in the 2007–2008 financial crisis: evidence from financial institutions worldwide. J. Corp. Finan. 18, 389–411. Fahlenbrach, R., Stulz, R., 2011. Bank CEO incentives and the credit crisis. J. Financ. Econ. 99, 11–26. Faulkender, M., Kadyrzhanova, D., Prabhala, N., Senbet, L.W., 2010. Executive compensation: an overview of research on corporate practices and proposed reforms. J. Appl. Corp. Financ. 22, 107–118. Furfine, C., 2001. The reluctance to borrow from the fed. Econ. Lett. 72, 209–213. Gande, A., Saunders, A., 2012. Are banks still special when there is a secondary market for loans? J. Financ. 67, 1649–1684. Gande, A., Puri, M., Saunders, A., Walter, I., 1997. Bank underwriting of debt securities: modern evidence. Rev. Financ. Stud. 10, 1175–1202. Gande, A., Puri, M., Saunders, A., 1999. Bank entry, competition, and the market for corporate securities underwriting. J. Financ. Econ. 54, 165–195. Gilson, S., Vetsuypens, M., 1993. CEO compensation in financially distressed firms: an empirical analysis. J. Financ. 48, 425–458. Gorton, G., Metrick, A., 2010. Regulating the shadow banking system. Brook. Pap. Econ. Act. 41, 261–312. Gorton, G., Winton, A., 2003. Financial intermediation. In: Constantinides, G., Harris, M., Stulz, R. (Eds.), Handbook of the Economics of Finance (Volume 1A: Corporate Finance). Elsevier, North Holland, pp. 431–552.

150

A. Gande, S. Kalpathy / Journal of Corporate Finance 47 (2017) 131–150

Guay, W., 1999. The sensitivity of CEO wealth to equity risk: an analysis of the magnitude and determinants. J. Financ. Econ. 53, 43–71. Hagendorff, J., Vallascas, F., 2011. CEO pay incentives and risk-taking: evidence from bank acquisitions. J. Corp. Finan. 17, 1078–1095. Hall, B., 1999. The design of multi-year stock option plans. J. Appl. Corp. Financ. 12, 97–106. Haugen, R., Senbet, L., 1981. Resolving the agency problems of external capital through options. J. Financ. 36, 629–647. Heckman, J., 1979. Sample selection bias as a specification error. Econometrica 47, 153–161. Ho, Po-Hsin, Huang, Chia-Wei, Lin, Chih-Yung, Yen, Ju-Fang, 2016. CEO overconfidence and financial crisis: evidence from bank lending and leverage. J. Financ. Econ. 120, 194–209. Houston, J., James, C., 1995. CEO compensation and bank risk: is compensation in banking structured to promote risk taking? J. Monet. Econ. 36, 405–431. Hubbard, R., Palia, D., 1995. Executive pay and performance: evidence from the US banking industry. J. Financ. Econ. 39, 105–130. James, C., Smith, D., 2000. Are banks still special? New evidence on their role in the capital-raising process. J. Appl. Corp. Financ. 12, 8–19. John, T., John, K., 1993. Top-management compensation and capital structure. J. Financ. 48, 949–974. John, K., Qian, Y., 2003. Incentive features in CEO compensation in the banking industry. Econ. Policy Rev. 9, 109–121. John, K., Saunders, A., Senbet, L., 2000. A theory of banking regulation and management compensation. Rev. Financ. Stud. 13, 95–125. Kalpathy, S., 2009. Stock option repricing and its alternatives: an empirical examination. J. Financ. Quant. Anal. 44, 1459–1487. King, T., Srivastav, A., Williams, J., 2016. What's in an education? Implications of CEO education for bank performance. J. Corp. Finan. 37, 287–308. Knopf, J., Nam, J., Thornton, J., 2002. The volatility and price sensitivities of managerial stock option portfolios and corporate hedging. J. Financ. 57, 801–813. Kolasinski, A., Yang, N., 2017. Managerial myopia and the mortgage meltdown. J. Financ. Econ. (Forthcoming). Low, A., 2009. Managerial risk-taking behavior and equity-based compensation. J. Financ. Econ. 92, 470–490. Mehran, H., Rosenberg, J., 2008. The effect of employee stock options on bank investment choice, borrowing, and capital. Federal Reserve Bank of New York Staff Reports, Staff Report No. 305. Minnick, K., Unal, H., Yang, L., 2011. Pay for performance? CEO compensation and acquirer returns in BHCs. Rev. Financ. Stud. 24, 439–472. Murphy, K., 2012. The politics of pay: a legislative history of executive compensation. In: Hill, Jennifer, Thomas, Randall (Eds.), Research Handbook on Executive Pay. Edward Elgar Publishers. Pan, Y., 2017. The determinants and impact of executive-firm matches. Manag. Sci. 63, 185–200. Perry, T., Zenner, M., 2001. Pay for performance? Government regulation and the structure of compensation contracts. J. Financ. Econ. 62, 453–488. Puri, M., 1996. Commercial banks in investment banking: conflict of interest or certification role? J. Financ. Econ. 40, 373–401. Rajan, R.G., Winton, A., 1995. Covenants and collateral as incentives to monitor. J. Financ. 50, 1113–1146. Rajgopal, S., Shevlin, T., 2002. Empirical evidence on the relation between stock option compensation and risk taking. J. Account. Econ. 33, 145–171. Ross, S., 2004. Compensation, incentives, and the duality of risk-aversion and riskiness. J. Financ. 59, 207–225. Saunders, A., Cornett, M.M., 2010. Financial Institutions Management: A Risk Management Approach. McGraw-Hill/Irwin, New York. Shue, K., Townsend, R., 2017. How do quasi-random option grants affect CEO risk-taking? J. Financ. (Forthcoming). Smith, C., Stulz, R., 1985. The determinants of firms' hedging policies. J. Financ. Quant. Anal. 20, 391–405. Stock, J., Yogo, M., 2005. Testing for weak instruments in IV regression. In: Andrews, D.W.K., Stock, J.H. (Eds.), Identification and Inference for Econometric Models: Essays in Honor of Thomas Rothenberg, pp. 80–108. Veronesi, P., Zingales, L., 2010. Paulson's Gift. J. Financ. Econ. 97, 339–368. Winton, Andrew, 1995. Costly state verification and multiple investors: the role of seniority. Rev. Financ. Stud. 8, 91–123.