The dark side of bank CEO risk taking incentives evidence from bank lending decisions

Managerial risktaking behavior in both financial and nonfinancial firms has been an attractive focus for the lens of many researchers (Hubbard and Palia, 1995; Houston and James, 1995; Knopf, Nam and Thornton, 2002; Coles, Daniel, and Naveen, 2006; Chen, Steiner and Whyte, 2006; Acharya and Naqvi, 2012). Excessive CEO risktaking in the financial sector especially has been blamed for playing a crucial role in the build up to the 20082009 financial crisis. Acharya and Naqvi (2012) develop a theoretical model to show that bank overlending may result from managers’ desire to receive higher compensation in the presence of an agency problem between a bank manager and shareholders.1 Other studies have revealed a positive correlation between option compensation and risktaking incentives, thus increasing bank risk taking and bankspecific default risk (Jeitschko and Jeung, 2005; Mehran and Rosenberg, 2007; Balachandran, Kogut and Harnal, 2010; Bebchuk, Cohen and Spamann, 2010; Fahlenbrach and Stulz, 2011; Hagendorff and Vallascas, 2011). For example, Coles, Daniel and Naveen (2006) suggest that the higher Vega gives executives incentive to implement more aggressive debt policy and invest more in riskier assets (e.g. RD). Similarly, DeYoung, Peng, and Yan (2013) show that banks in which CEOs have high risktaking incentives (highVega banks) exhibit substantially larger amounts of both systematic and idiosyncratic risk.2 To some extent, risktaking is good and that is why CEOs are given ESOPs (employee stock ownership plans) and equity stake to converge their interest with those of the shareholders. The problem is when CEOs go overboard and take “excessive” risk which is higher than the optimal level. Although above studies have confirmed that CEO risktaking

元 智 大 學

管理學院商學博士班 (財務金融學程)

博 士 論 文

銀行經理人風險承擔動機的黑暗面：從銀行放款決策分析之 The Dark Side of Bank CEO Risk-taking Incentives:

Evidence from Bank Lending Decisions

林智勇

中華民國 一百零九 年 六月

The Dark Side of Bank CEO Risk-taking Incentives:

Evidence from Bank Lending Decisions

指 導 教 授: 駱建陵 Advisors: Prof CHIEN-LING LO

元 智 大 學 管理學院博士班（財務金融學程）

博 士 論 文

A Dissertation

Submitted to Doctor of Philosophy Program

College of Management Yuan Ze University

in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy in Finance

June 2020 Chungli, Taiwan, Republic of China

中華民國 一百零九 年 六月

摘要

本研究探討銀行首席執行官的冒險動機（vega）如何影響銀行貸款決策 該研究的實證結果表明 vega 與貸款公告周圍的累積異常收益（CARs）顯著負相關, 這證明了 vega 對銀行貸款市場具有真正的影響 另外, 根據現有的 CEO 激勵文獻, 我們發現具有較高冒險精神的 CEO 傾向於放寬銀行貸款合約中的貸款標準，以尋求更高的報酬 有證據表明, 維加係數較高的銀行傾向於收取較低的貸款利差, 要求較少的貸款契約, 並且尋求抵押品的可能性較低 結果將會變弱, 這可支持以下觀點：高 CEO 冒險行為可能會在銀行經理和股東之間造成代理問題

經濟文學雜誌: G21, G32, G34

關鍵字：CEO激勵, 銀行貸款合約, 累積超額收益, 公司治理, 代理問題

The Dark Side of Bank CEO Risk-taking Incentives:

Evidence from Bank Lending Decisions

Student: Tran Thi Thuy Linh Advisors: Chien-Ling Lo

Chih-Yung Lin

Doctor of Philosophy Program

Major of Finance College of Management Yuan Ze University

Abstract

This paper investigates how bank CEO risk-taking incentives(vega) influence bank lending decisions Empirical finding of the study reveals that vega is significantly negatively related to cumulative abnormal returns(CAR) around loan announcements, confirming that vega has a real effect on the bank loan market In addition, consistent with the existing CEO incentive literature, we find that CEOs with higher risk-taking incentives tend to relax their lending standards in bank loan contracts to pursue higher compensation Evidence shows that banks with high vega tend to charge a significantly lower loan spread, demand fewer loan covenants, and have lower probability to seek collateral Results become weaker when banks have strong corporate governance mechanisms, supporting the proposition that high CEO risk-taking incentives may create

an agency problem between a bank manager and shareholders

JEL: G21, G32, G34

Keywords: CEO incentives, bank loan contracts, cumulative abnormal returns, corporate governance, agency problem

Acknowledgements

First and foremost, I would like to express my deep and sincere gratitude to my advisor, Professor Chih-Yung Lin for giving me the opportunity to do research and providing invaluable guidance throughout this research His dynamism, vision, sincerity and motivation have deeply inspired me He has taught me the methodology to carry out the research and to present the research works as clearly as possible It was a great privilege and honor to work and study under his guidance Without his persistent help, the goal of my dissertation would not have been realized I am extremely grateful for what he has offered me

I would also like to express my sincere thanks to Professor Chien-Ling Lo and Professor Po-Hsin Ho, who support and help me a lot in the journey towards this degree

My appreciation also extends to all Professors at College of Management who have given me a great deal of knowledge in the last four years

My thanks go to all my classmates and friends in Yuan Ze university for their valuable shares

Last but not least, I would like to show my endless love to my parents, my husband and my son Thank for all coming to my life, always accompanying and staying by my side

Table of Contents

摘要 iii

Abstract iv

Acknowledgements v

Table of Contents vi

List of Tables vii

Chapter 1 Introduction 1

Chapter 2 Hypothesis development 7

Chapter 3 Data and methodology 11

3.1 Data and other variables 11

3.2 Summary Statistics 12

3.3 Methodology 14

Chapter 4 Empirical Results 17

4.1 Bank’s CEO risk-taking incentives and cumulative abnormal returns(CARs) 17

4.2 Bank’s CEO risk-taking incentives and bank loan spread 17

4.3 Bank’s CEO risk-taking incentives and non-price terms 19

4.4 Bank’s CEO risk taking behavior: A Difference-in-Difference analysis 19

4.5 CEO risk-taking-incentive effect: Corporate-governance channels 21

4.6 Robustness checks 24

4.6.1 Control for CEO characteristics and other compensation schemes 24

4.6.2 Bank size analysis 24

4.6.3 Change regression 25

4.6.4 Bank’s CEO risk-taking incentives and cumulative abnormal returns (CARs) 26

Chapter 5 Conclusion 28

References 29

Appendix A: Variable definition 52

Appendix B: Sample banks 54

Appendix C: Vega measure 55

List of Tables

Table 1 Descriptive Statistics 34

Table 2 Loan, borrower, and lender characteristics for banks with high and low Vega 35

Table 3 Correlation matrix 36

Table 4 Bank’s CEO risk-taking incentives and cumulative abnormal returns (CARs) 37

Table 5 Bank’s CEO risk-taking incentives and bank loan spread 39

Table 6 Bank’s CEO risk-taking incentives: Non-price terms 41

Table 7 CEO risk-taking incentives effect: Difference-in-Difference analysis 43

Table 8 CEO risk-taking incentives and bank loan spread: Evidence from bank governance channels 45

Table 9 Robustness check (I): Control for CEO characteristics and other CEO compensation 46

Table 10 Robustness check (II): Bank size analysis and the change regression 48

Table 11 Robustness check(III): Bank’s CEO risk-taking incentives and cumulative abnormal returns 50

Chapter 1 Introduction

Managerial risk-taking behavior in both financial and non-financial firms has been an

attractive focus for the lens of many researchers (Hubbard and Palia, 1995; Houston and James,

1995; Knopf, Nam and Thornton, 2002; Coles, Daniel, and Naveen, 2006; Chen, Steiner and

Whyte, 2006; Acharya and Naqvi, 2012) Excessive CEO risk-taking in the financial sector

especially has been blamed for playing a crucial role in the build up to the 2008-2009 financial

crisis Acharya and Naqvi (2012) develop a theoretical model to show that bank over-lending may result from managers’ desire to receive higher compensation in the presence of an agency problem between a bank manager and shareholders.1

Other studies have revealed a positive correlation between option compensation and

risk-taking incentives, thus increasing bank risk risk-taking and bank-specific default risk (Jeitschko and

Jeung, 2005; Mehran and Rosenberg, 2007; Balachandran, Kogut and Harnal, 2010; Bebchuk,

Cohen and Spamann, 2010; Fahlenbrach and Stulz, 2011; Hagendorff and Vallascas, 2011)

For example, Coles, Daniel and Naveen (2006) suggest that the higher Vega gives executives

incentive to implement more aggressive debt policy and invest more in riskier assets (e.g

R&D) Similarly, DeYoung, Peng, and Yan (2013) show that banks in which CEOs have high

risk-taking incentives (high-Vega banks) exhibit substantially larger amounts of both systematic

and idiosyncratic risk.2 To some extent, risk-taking is good and that is why CEOs are given

ESOPs (employee stock ownership plans) and equity stake to converge their interest with those

of the shareholders The problem is when CEOs go overboard and take “excessive” risk which

is higher than the optimal level Although above studies have confirmed that CEO risk-taking

1 Acharya and Naqvi (2016) show that, if a bank is awash with deposits from investors, its manager will be more likely to undertake high-risk projects to pursue his/her own self-interest and to sanction excessive loans by lowering lending rates and loosening lending standards (underprice the risk of projects), leading to asset-price bubbles and sowing seeds of future bank failure

2 Gande and Kalpathy (2017) indicate that equity incentives (Vega) embedded in CEO compensation contracts are

positively associated with risk taking in financial firms and result in potential solvency problems

incentives increase bank risk exposure, how such exposure affects bank lending decisions has

not to date been examined Specifically, in this paper we investigate the effects of bank’s CEO

risk-taking incentives (Vega) on bank loan contracting

In lending relationships, cumulative abnormal returns (CARs) in bank loan announcement

studies is helpful in order to evaluate the firm performance(James 1987; Lummer and

McConnell 1989; Dahiya et al., 2003; Billett et al., 1995; Billeett et al., 2006; Kang and Liu

2008) Various authors in their research show that positive announcement returns are observed

in firms having low information asymmetry (Mikkelson and Partch 1986; James 1987; Lummer

and McConnell 1989; Slovin et al., 1992; and Ross 2010) For example, Mikkelson and Partch

(1986) and James (1987) argue that information embedded in the bank loan decisions reflect

the health of firm to capital market by examining the positive excess returns associated with

bank loan announcements Best et al., (1993) indicated that a positive CARs around the time

of bank loan announcements can be considered as the signaling for banks’ valuable monitoring function Consistent with this idea, in this paper we evaluate the bank’s over-lending effect

caused by CEO risk-taking incentive (Vega) is a good or bad signal by paying attention to the

market response to bank loan announcement

We attempt to answer the following five questions regarding the Vega effects on bank loan

contracts: (i) Do banks with higher Vega earn lower cumulative abnormal returns around bank

loan announcement date?; (ii) Do banks with higher Vega charge lower interest rates on loans?;

(iii) Do Vega effects on bank loan contracts also exist in non-price terms(general covenants,

financial covernants, collateral)?; (iv) Are Vega effects weaker by strong corporate governance

mechanisms?; and (v) Do Vega effects still hold after adjusting for other CEO compensation

schemes and CEO characteristics?

We evaluate these questions by using a sample of 20,502 loans to 5,102 U.S firms between

1992 and 2014 We obtain all accounting variables and stock prices from the Compustat database

and the Center for Research in Security Prices (CRSP) Corporate-governance and

CEO-compensation-related variables come from RiskMetrics and ExecuComp We collect bank loan

data from the DealScan database

First, our empirical results show that the relationship between high Vega and cumulative

abnormal returns is significantly negative, i.e., banks with higher Vega can earn lower CARs

around bank loan announcement dates Second, we find that CEO risk-taking incentives in

term of Vega are significantly and negatively correlated with bank loan spread That is, with

high CEO Vega incentives, banks charge lower interest rates on loans after controlling for

lender characteristics, borrower characteristics and other loan conditions The evidence

remains strong when we control for CEO fixed effect and year fixed effect, loan purpose and

loan type, as well as for CEO characteristics For example, the coefficient of Vega is -0.0266,

which shows that, on average, one-standard-deviation increase in vega (i.e., 1.5334 in Table

1) leads to a reduction in Spread by about -0.0408, where -0.0408= -0.0266 × 1.5334 Given

that the average loan spread in our sample is 142.49 basis points (e4.9593), a

one-standard-deviation increase in vega in the data reduce loan spreads by about 5.70 basis points (-5.70 =

142.49 × e−0.0408 – 142.49) Furthermore, the lower 5.70 basis-point loan spread is around

131.64 percent of the estimated coefficients for the effect of corporate tax avoidance and for

the effect of social capital in Hasan, Hoi, Wu, and Zhang (2014, 2017) Accordingly, the

effect of bank CEO incentives on loan spread is not only statistically significant but is also

economically important

In order to resolve the potential endogeneity issue, we adopt a difference-in-difference

(DiD) analysis to perform a robustness check We apply the event of U.S government bailouts

for those bank which need help after Financial Crisis in 2008 Further, Special Master Feinberg,

as TARP (The Troubled Asset Relief Program) was implemented following the crisis,

prohibited option awards to executives of banks receiving relief from TARP This seemed to

put pressure on non-TARP banks to reduce option usage If the bank is in the list, government

would give them funding to help, but they also have to accept some restrictions from

government, therefore, the bank CEO may more carefully consider the spread We denote the

banks that got bailed out as treatment unit, the others are considered as control units The results

of the DiD regression sanction the agency cost view that treatment firms are charged higher

loan spreads in their next loan contract when they experienced the bailout support, this further

validate the agency cost effect throughout the governance channel

We further find evidence that bank managers with high Vega tend to impose fewer

covenants on loans (especially general covenants) and are less likely to require collateral All

our main results are statistically and economically significant and are robust to specifications

with different sets of explanatory variables (namely, lender, borrower and loan characteristics;

industry fixed effect and year fixed effect, loan purpose and loan type)

In further testing, we examine in greater detail whether the effect of CEO Vega derives

from the agency cost channel We use several different variables to measure corporate

governance power and run regressions including the interaction term of Vega and a set of

dummy variables for good corporate governance We argue that, if CEO vega is viewed as a

representation of agency cost, then its effect would be tendered when bank corporate

governance quality is good Our results provide supportive evidence for this governance

channel, confirming that our main findings of a vega incentive effect are driven primarily by

banks with weak corporate governance Hence, our results suggest that the influence of CEO

vega derives from an agency problem between bank managers and shareholders

As a robustness check of our findings, we consider alternative measures of compensation

and the effect of CEO characteristics on bank loan contracts and find similar results In

addition, we use a change regression following Chava et al., (2009) to control for the

omitted-variable bias and the results remain supportive of our hypothesis Finally, to check the

sensitivity of the window selection, we redo the vega analysis with CARs around bank loan

announcement dates but employ different window and note similar results

Our work contributes to the literature in two ways First, we add to the literature on

executive-compensation contracting and bank-lending behaviors A few studies have recently

focused on the CEO risk-taking incentives inherent in bank compensation that affect bank

performance, on riskier bank investment choices and the business policies of U.S banking

companies in the short period before and during the financial crisis of 2008 (Acharya and

Naqvi, 2012; DeYoung, Peng, and Yan, 2013; Bhagat and Bolton, 2014; Cheng, Hong and

Scheinkman, 2015; Chesney, Stromberg and Wagner, 2016; Gande and Kalpathy, 2017)

Different from previous studies, we provide evidence of the relationship between Vega and

loan contracting from a lender’s point of view, i.e., we follow the studies of Coles, Daniel, and

Naveen (2006, 2013) to calculate CEO risk-taking incentives (Vega) and find that banks with

high Vega compensation may reduce lending interest rates and relax lending standards in the

interests of their own wealth Results become weaker, however, when banks have strong

corporate governance mechanisms, supporting the view that high CEO risk-taking incentives

may create an agency problem between a bank manager and shareholders Thus, our results

provide a reference for scholars, policy makers, and market investors for assessing the

significance of the influence of CEO risk-taking incentive on bank lending decisions

Second, this paper provides an explanation of the determinants of bank loan contracts from

a new perspective: the supply side The determinants of bank loan contracts from the demand

side have already been studied extensively (Strahan, 1999; Sufi, 2007; Bharath, Sunder and

Sunder 2008; Graham, Li and Qiu, 2008; Lin et al., 2011), and recently include cultural

differences (Giannetti and Yafeh, 2012), political connections (Houston, Jiang, Lin, and Ma,

2014), tax avoidance (Hasan, Hoi, Wu, and Zhang, 2014), corporate diversification (Aivazian

et al., 2015), social capital (Hasan, Hoi, Wu, and Zhang, 2017), customer concentration

(Campello and Gao, 2017), private information (Carrizosa and Ryan, 2017), private benefit of

control (Lin et al., 2018), and managerial ability (Bui et al., 2018) Different from these

previous studies, we propose that a key factor in bank loan contracts are CEO risk-taking

incentives from the supply side Based on this view, this paper also adds to the literature on bank

loan contracts

Chapter 2 Hypothesis development

Cumulative abnormal returns (CARs) in bank loan announcement studies is helpful in

order to evaluate the firm performance (James 1987; Lummer and McConnell 1989; Dahiya

et al., 2003; Billett et al., 1995; Billeett et al 2006; Kang and Liu 2008;) Various authors in

their research show that positive announcement returns are observed in firms having low

information asymmetry (Mikkelson and Partch 1986; James 1987; Lummer and McConnell

1989; Slovin et al., 1992; and Ross 2010) For example, Mikkelson and Partch 1986 and James

(1987) argue that information embedded in the bank loan decisions reflect the health of firm to

capital market by examining the positive excess returns associated with bank loan

announcements Best et al., (1993) indicated that a positive CARs around the time of bank loan

announcements can be considered as the signaling for banks’ valuable monitoring function

James (1987) argues that loan announcements reflect private information concerning the value

of a firm’s growth and loan spread as well as its response to abnormal returns performance Shockley and Thakor (1997) point out that loan announcement contracts have several essential

features, such as (1) a lower loan spread is a condition of a firm’s credit risk premium; (2) a loan announcement is a flexible commitment that can be customized to meet the firm’s needs and includes restrictive covenants; and (3) it provides a mechanism for banks to decide fee

structures and identify borrowers Slovin et al., (1992) find that renewal of bank credit

agreements depend on favorable stock prices, especially for small firms, because the latter is

less abundant in public information and banks must provide external monitoring Moreover,

loan announcement contracts may be associated with higher returns Maskara and Mullineaux

(2011) state that smaller firms are more likely to announce their loans than are larger firms,

because of the former face more serious information asymmetry problems and rely on such

loan announcements to obtain abnormal returns

Conflict of interest between shareholders and managers is a key agency problem for firms

(Jensen and Meckling, 1976; Jensen, 1986) Firms can use equity-based compensation schemes

to motivate CEOs’ efforts to take on risky projects, thus ameliorating the principal-agent conflict (Jensen and Meckling, 1976; Myers, 1977; Guay, 1999; Smith and Stulz, 1985; Bolton,

Mehran, and Shapiro, 2015) During the financial crisis of 2008, the U.S economy experienced

its worst recession since the Great Depression of the 1930s with the collapse of many of its

largest banking and financial institutions.3 As a result, bank executives faced widespread

criticism for driving the excessive risk taking that ignited the disastrous financial crisis A

growing number of academic researchers discuss the impacts of equity-based compensation

contracts and whether government should do more to limit executive compensation

(Fahlenbrach and Stulz, 2011; Balachandran, Kogut and Harnal 2010; Hagendorff and

Vallascas, 2011; DeYoung, Peng and Yan, 2013; Chesney, Stromberg and Wagner, 2016).4

In early studies of deregulation of the banking industry, many authors have made efforts

to examine the problem of equity-based compensation and bank performance For example,

Houston and James (1995) and Hubbard and Palia (1995) suggest that compensation contracts

create incentives for CEOs to take risks Equity incentives embedded in CEO compensation

contracts are strongly associated with bank performance and this relationship is stronger in

deregulated interstate banking markets Moreover, the deregulation of the banking industry

during the 1990s provides a natural experiment for investigating how firms adjust their

executive compensation contracts as the environment in which they operate becomes relatively

more competitive (Brewer, Hunter and Jackson, 2004) They show evidence that riskier banks

had significantly higher levels of equity-based compensation in the post-deregulation period,

as did banks with greater investment opportunities

Other recent studies examine the impact of CEO compensation structure on bank risk

taking Fahlenbrach and Stulz (2011) use US banking data to uncover the existence of a

relationship between bank performance, stock returns and returns on assets with equity

incentives They find that banks led by CEOs who have better manager-shareholder alignment

was associated with significantly worse performance during the crisis In other words, financial

strategies such as stock holding or option compensation may encourage CEOs to engage in

more risky projects to increase their future equity value Some empirical studies find that the

sensitivity of CEO equity and option holdings to stock volatility has an impact on bank risk

taking (e.g., Chen et al., 2006; Mehran and Rosenberg, 2007; Hagendorff and Vallascas, 2011;

Boyallian and Ruiz-Verdú, 2015).5

Acharya and Naqvi (2012) develop a theoretical model to show that bank over-lending

may result from bank managers’ desire to receive higher compensations Acharya and Naqvi

(2016) further show that, if a bank is awash with deposits from investors, the manager will be

more likely to undertake high-risk projects to pursue his/her own self-interest and may sanction

excessive loans by lowering lending standards, which can lead to asset price bubbles and sow

seeds for future bank failure Both of these papers report that over-lending may result from

bank managers’ desire to receive higher compensation, which is an agency problem between

managers and shareholders

In this paper, if the bank’s over-lending effect caused by CEO risk-taking incentive (Vega)

is a bad signal, so we should expect that market response to bank loan announcement is

negative Thus, our first hypothesis is as follows:

H1: The effect of CEO Vega on cumulative abnormal returns is negative around a loan

announcement date

5 To the contrary, Kolasinski and Yang (2018) do not find any connection between CEO vega and firm risk taking during the crisis period

To the best of our knowledge, no empirical research has to date investigated the specific

relationship between Vega and risk taking in bank loan contracts Our research contributes to

the literature in that CEOs with higher Vega tend to take greater risks by setting up mechanisms

to decrease the price of loans and relax lending standards on bank loan contracts Therefore,

we propose the second hypotheses as follows:

H2: Banks with higher CEO Vega are more likely to charge lower loan spreads

In addition to the price of a loan, other non-price terms can also be imposed if banks

consider control benefits to be a harmful agency problem Some of these terms are found in

Graham, Li, and Qiu (2008) and Rahaman and Zaman (2013) with covenants and collateral

requirements Therefore, our third hypothesis is as follows:

H3: Banks with higher CEO Vega tend to impose more favorable non-price terms on loans

Chapter 3 Data and methodology

3.1 Data and other variables

We obtain our bank loan data from DealScan database, which include loan spread,

maturity, size, and performance; collateral; general and financial covenants; and purpose and

type of loan We need to calculate the rate of change in our main variables at year t based on

the data at year t – 1, our sample is further restricted to include firms with at least two years of

data Thus, the first year of the sample is 1991 and the effective results period begins with

1992 Our final sample includes 20,502 bank-year observations to 5,102 U.S firms between

1992 and 2014

The main variable in this study is Spread, which is natural log of loan spread Loan spread

is measured as the all-in spread drawn in the DealScan database All-in spread drawn is defined

as the amount the borrower pays in terms of basis points over LIBOR or the LIBOR equivalent

for each dollar drawn down Other variables of loan characteristics included the natural

logarithm of loan maturity in months (Maturity), the natural logarithm of amount of loan in

US$ million (Loan Size), a dummy variable that equals one if the loan facility uses performance

pricing (Performance) and otherwise is zero, a dummy variable that takes a value of 1 if a loan

is secured (Collateral), and 0 otherwise), number of financial covenants (FinCov), number of

general covenants( GenCov), and the natural logarithm of the number of general plus financial

covenants (TotalCov)

The firm characteristics are from Compustat: MB ((total assets – book value of equity +

price × common shares outstanding) / total assets), Profitability (earnings before interest, taxes,

depreciation, and amortization (EBITDA) divided by the total assets), Tangibility (net property,

plant, and equipment divided by total assets), CF-volatility (standard deviation of quarterly cash

flows from operations over the four fiscal years prior to the loan initiation year scaled by the total

debt), and Z-score (modified Altman’s Z-score, 1.2 × working capital + 1.4 × retained earnings

+ 3.3 × EBIT+ 0.999 × sales)/ total assets)

Lender characteristics are from Compustat Bank: L_Asset (log of total assets in billions of

dollars), L_ Leverage (ratio of assets to book value of equity), and L_loandep (ratio of average

balance of loans to average balance of deposits)

CEO compensation and characteristics from ExecuComp, which include: Delta (Dollar

change in wealth associated with a one-percent change in the firm’s stock price in US$ million),

CEO inside debt ratio (the value of inside debt divided by the total value of shares and options

owned) We define a CEO’s inside debt as the sum of the balance in the CEO’s pension fund and

nonqualified deferred compensation), TDC1 (this variable includes salary, bonus, stock awards,

option awards, long-term incentive plans, and other annual compensation such as perquisites and

other personal benefits in $million), CEO age (the CEO’s age when the company signs the bank

loan contract), and CEO tenure (the number of years the CEO held his/her position in the

company before signing the bank loan contract)

Bank governance variables from Risk Metrics, which include: Independent directors

(Percentage of outside directors), Institutional ownership (Percentage of shares held by

institutional investors in the firm), Board size (Number of board directors), Female (The

proportion of female directors in the board), Academic (A dummy variable that equals one if the

board includes at least one director from academia, and zero otherwise.) Appendix A

summarizes all variable definitions

3.2 Summary Statistics

Descriptive statistics of CARs, Spread, Vega, and other control variables are shown in

Table 1 The mean, median values of bank loan spread are 4.9593 and 5.1648 The average

spread is 4.9593 (e4.9593 = 142.4940 basis points), with a standard deviation of 0.8248 The

value of Vega from 1992 to 2014 is positive in all firms under consideration, which is

reasonable and consistent with the previous literature on quantifying methods of Vega (Coles,

Daniel and Naveen, 2006; DeYoung, Peng and Yan, 2013) The mean value of Vega is 5.5677

with the standard deviation of 1.5334 respectively

The average of L_Asset, L_ Leverage, and L_loandep are 12.3910, 12.5954, and 0.8357,

respectively Regarding loan characteristics, the average of Maturity, Loan size, Performance,

Collateral, FinCov, GenCov are 3.5975, 4.7611, 0.3887, 0.4658, 0.5862 and 1.0187,

respectively In other words, the loan contracts under consideration have a mean maturity of

37 months (exponential of 3.5975) The mean value of loan size is US$ 117 million (4.7611 in

natural logarithm) A loan contract in the sample has, on average, 2.7695 general covenants

(maximum of 10 covenants) and 1.7971 financial covenants (maximum of 7 covenants) for the

borrowers For the characteristics of borrowing firms, the average Tangibility, Leverage,

Profitability, MB, Z-score and CF volatility values are 0.3000, 0.3494, 0.1139, 0.6991, 1.4804

and 1.8469, respectively

[Insert Table 1 here]

In Table 2, we divide the sample into high- and low-Vega banks based on the median of

Vega We use a t-test to determine the significance of the difference in the means of all variables

between the two groups and report the results for this preliminary univariate test Most of the

mean comparisons show a significant difference at the one-percent level Bank with higher

vega tend to earn lower cumulative abnormal returns around bank loan announcement dates

Banks with higher Vega should charge lower spreads in order to engage in more risk-taking

Banks in the high-Vega group also ask for less collateral as well as impose a lower number of

covenants (both general and financial) compared to those in the lower-vega group

[Insert Table 2 here]

We also report the Pearson’s correlation coefficient matrix in Table 3 As expected, we find

a negative and significant correlation between Spread and CARs, Spread and Vega The

correlation coefficients between bank compensation regarding Vega and other loan terms such

as general and financial covenants are also negative and significant at the one-percent level The

correlation coefficient between Spread and other control variables such as CEO age are

significantly negative, indicating that banks with older CEOs are also associated with a

significantly lower spread in their loans

[Insert Table 3 here]

3.3 Methodology

We first examine the effects of Vega on cumulative abnormal returns using loan

announcements Following Brown and Warner (1985), to examine the effects of Vega in CARs,

we use the sensitivity of earnings announcement Following the studies of Coles, Daniel,

Naveen (2006), DeYoung, Peng, and Yan (2013) and Hasan, Hoi, Wu, and Zhang (2014, 2017),

we use an ordinary least squares (OLS) regression to investigate the effects of CEO risk-taking

incentives in terms of Vega on bank loan spread

𝐶𝐴𝑅[−5,5]𝑖,𝑡 = 𝛼0+ 𝛼1𝑉𝑒𝑔𝑎𝑖,𝑡−1+ 𝛽′𝐹𝑖,𝑡−1+ 𝜃′𝑍𝑖,𝑡+ 𝛾𝑖+ 𝜇𝑡+ 𝜀𝑖𝑡 (1)

𝑆𝑝𝑟𝑒𝑎𝑑𝑖,𝑡 = 𝛼0+ 𝛼1𝑉𝑒𝑔𝑎𝑖,𝑡−1+ 𝛽′𝐹𝑖,𝑡−1+ 𝜃′𝑍𝑖,𝑡+ 𝛾𝑖+ 𝜇𝑡+ 𝜀𝑖,𝑡 (2)

where 𝐶𝐴𝑅[−5; 5]𝑖𝑡 is the cumulative abnormal return (from the Fama-French four factor

model) in the window [-5;5] for firm i in year t 𝑆𝑝𝑟𝑒𝑎𝑑𝑖,𝑡 is the natural logarithm of the bank loan

spread for loan i in year t; 𝑉𝑒𝑔𝑎𝑖,𝑡−1 represents CEO risk-taking incentives, captures the change in

the dollar value of CEO wealth for a 0.01-unit change in stock return volatility for bank i in year

t-1 (Coles, Daniel and Naveen, 2006, 20t-13) 𝐹𝑖,𝑡−1 is a vector of control variables for lenders and

borrowers i in year t-1, including lender and borrower characteristics 𝑍𝑖,𝑡 is the vector of the

control variables for loan i in year t 𝛾𝑖 and 𝜇𝑡 represent the fixed effects of CEO and year,

respectively, and 𝜀𝑖,𝑡 represents the error term of the regression We also control for loan primary

purpose and loan type In all models, the t-statistics reported are based on heteroscedasticity and

sample clustering at firm-level robust standard errors (White, 1980 and Petersen, 2009)

To resolve the endogeneity issue, we adopt a difference-in-difference (DiD) analysis We

denote the banks that got bailed out as treatment unit (Treatment𝑖,𝑡−1), the rest of firms are

considered as control units Also, we construct a dummy variable (Post𝑖,𝑡) that equals one if a

treatment firm’s fiscal year, t, falls after the year of Treatment𝑖,𝑡−1, and zero otherwise The interaction term of variable (Treatment𝑖,𝑡−1) and (Post𝑖,𝑡) is denoted by “Treat_Post”

variable The regression equation is as follows:

Spread𝑖,𝑡 = 𝛼1+ 𝛼2Treat_Post𝑖,𝑡−1𝑖,𝑡+ 𝛽′𝐹𝑖,𝑡−1+ 𝜃′𝑍𝑖,𝑡+ 𝜈𝑖+ 𝜇𝑡+ 𝜀𝑖,𝑡 (3)

𝐶𝐴𝑅[−5,5]𝑖,𝑡= 𝛼1+ 𝛼2Treat_Post𝑖,𝑡−1𝑖,𝑡+ 𝛽′𝐹𝑖,𝑡−1+ 𝜃′𝑍𝑖,𝑡 + 𝜈𝑖+ 𝜇𝑡+ 𝜀𝑖,𝑡 (4)

where Spread𝑖,𝑡 represents the log of bank loan spread for firm i in year t; 𝐶𝐴𝑅[−5; 5]𝑖𝑡 is the

cumulative abnormal return (from the Fama-French four factor model) in the window [-5;5]

for firm i in year t; 𝐹𝑖,𝑡−1 is a vector of control variables for lenders and borrowers i in year

t-1, including lender and borrower characteristics 𝑍𝑖,𝑡 is the vector of the control variables for

loan i in year t; 𝜈𝑖 and 𝜇𝑡 capture the CEO and year fixed effects, respectively; and 𝜀𝑖,𝑡 is the

random error Variable definitions are provided in Appendix A

We use Poisson regression to test the effect of Vega on total number of covenants and on

general and financial covenants The equation is as follows:

𝐿𝑛 (𝐸(𝑌|𝑉𝑒𝑔𝑎𝑖,𝑡−1, 𝐹𝑖,𝑡−1, 𝑍𝑖,𝑡))

(5)

In this formula, Y i,t are different non-price terms of a bank loan contract for firm i in year

t, include: TotalCov (total number of covenants in a loan contract); GenCov( number of general

covenants); and FinCov (number of financial covenants) The Poisson regression method is

used in analyzing the effect of Vega because the number of covenants in a loan contract are

countable data This methodology is previously used in the bank-loan literature, for example

Graham, Li, and Qiu (2008) and Hasan, Hoi, Wu, and Zhang (2014)

In addition, The Probit regression is used to test the influence of Vega on Collateral (a

dummy variable that equals one if a bank requires collateral on a loan to borrower, and zero

otherwise) The equation is as follows:

Pr (𝐶𝑜𝑙𝑙𝑎𝑡𝑒𝑟𝑎𝑙𝑖,𝑡 = 1|𝑉𝑒𝑔𝑎𝑖,𝑡−1, 𝑍𝑖,𝑡) = 𝛷(𝛼0+ 𝛼1𝑉𝑒𝑔𝑎𝑖,𝑡−1+ 𝛽′𝐹𝑖,𝑡−1+ 𝜃′𝑍𝑖,𝑡 + 𝛾𝑖 + 𝜇𝑡)

(6)

Chapter 4 Empirical Results

4.1 Bank’s CEO risk-taking incentives and cumulative abnormal returns(CARs)

In this section, we examine the influence of high Vega in the sensitivity of earnings

announcement to cumulative abnormal returns from the Fama-French four factors model in the

window [-5;5] The Model (1) to (3) of Table 4 shows the regression results based on Equation

(1) Our focus is the coefficient of Vega on cumulative abnormal return We run CEO fixed

effect regressions for all model to control the influence of unobservable omitted variables For

the robustness of our results, we conduct three specifications in the regression setting The first

controls for CEO fixed effect and year fixed effects; the second adds controls for loan purpose;

the third control for all CEO fixed effects, year fixed effects, loan purpose and loan type

Consistent with our expectation, in all specifications, the results are negative and

significant and the coefficient between -0.0038 to - 0.0040, indicating that an increase in CEO

risk taking incentive is associated with a decrease in the sensitivity of earnings announcement

to cumulative abnormal returns Thus, the results support our hypothesis 1 that banks with

higher CEO vega would be able to earn less cumulative abnormal returns

[Insert Table 4 here]

4.2 Bank’s CEO risk-taking incentives and bank loan spread

In this section, we examine the relationship between bank’s CEO risk-taking incentives

and loan spread The Model (1) to (3) of Table 5 shows the regression results based on Equation

(2) Our focus is the coefficient of Vega on bank loan spread We run CEO fixed effect

regressions for all model to control the influence of unobservable omitted variables For the

robustness of our results, we conduct three specifications in the regression setting The first

controls for CEO fixed effect and year fixed effects; the second adds controls for loan purpose;

the third control for all CEO fixed effects, year fixed effects, loan purpose and loan type

Consistent with our expectation, negative and significant coefficients are observed in all

specifications, even though we have controlled for all potential factors, indicating that banks

with higher vega compensation schemes will charge lower spreads for bank loans Thus, the

results support our hypothesis 2 that banks with higher CEO vega are more likely to charge

lower loan spreads

Specifically, the coefficients of Vega are significant at the onepercent level from a

-0.0253 to -0.0266 decrease in log loan spread For example, in Model (3), the coefficient of

Vega is -0.0266, which shows that, on average, one-standard-deviation increase in vega (i.e.,

1.5334 in Table 1) leads to a reduction in Spread by about -0.0408, where -0.0408= -0.0266 ×

1.5334 Based on the estimate in Model (3) and given that the average loan spread in our sample

is 142.49 basis points (e4.9593), a one-standard-deviation increase in vega in the data reduce

loan spreads by about 5.70 basis points (-5.70 = 142.49 × e−0.0408 – 142.49)

Furthermore, the lower 5.70 basis-point loan spread is around 131.64 percent of the

estimated coefficients for the effect of corporate tax avoidance and for the effect of social

capital in Hasan, Hoi, Wu, and Zhang (2014, 2017) Accordingly, the effect of bank CEO

incentives on loan spread is not only statistically significant but is also economically important

With regard to borrower characteristics, firms with higher Profitability, MB, and lower

Leverage, are charged lower loan spreads, consistent with the findings of Graham, Li, and Qiu

(2008), Chava and Roberts (2008) and Hasan, Hoi, Wu, and Zhang (2014) For loan

characteristic variables, loan size and performance pricing are significantly negative affect loan

spread We also find a positive coefficient on financial and general covenants in all models

Overall, these results are consistent with previous studies.6

[Insert Table 5 here]

4.3 Bank’s CEO risk-taking incentives and non-price terms

Following Graham, Li, and Qiu (2008) and Hasan, Hoi, Wu, and Zhang (2014), in term

of dependent variable in Table 6, we use different non-price terms to examine whether CEO

Vega compensation drives banks to relax their lending standards The Model (1) to (3) shows

the regression results based on Equation (5) Model (4) shows the regression results based on

Equation (6)

In Model (1), Vega is negatively correlated with the total covenants at a one-percent

statistically significance However, when we take both general and financial covenants into

consideration in Model (2) and (3), we find that Vega affects only the number of general

covenants The coefficient of Vega for financial covenants in Model (3) is negative, but not

significant Finally, in Model (4), the probit regression reveals that high-Vega banks are less

likely to seek collateral for their loans The results in Table 6 are consistent with our hypothesis

3 that banks with higher CEO Vega tend to impose more favorable non-price terms on loans

[Insert Table 6 here]

4.4 Bank’s CEO risk taking behavior: A Difference-in-Difference analysis

One concern surrounding most empirical studies in the finance field is the endogeneity

problem For example, the CEO Vega may endogenously be determined by the firm’s

compensation committee If some unobservable variables which affect both the loan spread

and CEO Vega are not incorporated in the regression, the estimators could be biased

6 The fit of the models seems to be high In specification (1), Vega helps explain the 48.39 percent change in bank loan spread When control variables are added in the next specification, the adjusted R 2 is somewhat higher, with the last models increasing to 59.42

To resolve the endogeneity issue, we adopt a difference-in-difference (DiD) analysis We

apply the event of U.S government bailouts for those bank which need help after Financial

Crisis in 2008 I choose this event because TARP (The Troubled Asset Relief Program) was

designed to improve the safety and soundness of the banking system, which included specific

provisions aimed at reducing the “excessive risk taking” that was believed to have contributed

to the financial crisis Therefore, TARP might have affected risk-taking incentives relative to

changes in bank lending Further, Special Master Feinberg, as TARP was implemented

following the crisis, prohibited option awards to executives of banks receiving relief from

TARP This seemed to put pressure on TARP banks to reduce option usage

We denote the banks that got bailed out as treatment unit (Treatment𝑖,𝑡−1), the rest of

firms are considered as control units Also, we construct a dummy variable (Post𝑖,𝑡) that equals

one if a treatment firm’s fiscal year, t, falls after the year of Treatment𝑖,𝑡−1, and zero otherwise The interaction term of variable (Treatment𝑖,𝑡−1 ) and ( Post𝑖,𝑡 ) is denoted by

“Treat×Post” variable The regression equation is as Equation (3) and (4)

As far as the results of Difference-in-Difference regression are concerned, our focus is on

the coefficients of Treat×Post For robustness checks on CEO risk-taking incentives and

spread, we expect a positive combined coefficient to support the proposition that after treatment

banks experience an event whereby they are Bailout Recipients, they charged higher loan

spreads as a consequence

Table 7 Panel A, presents the results of the DiD regression In Model 1 and 2, we test with

explanatory variable is Spread, in Model (3) and (4), we run with explanatory variable is

cumulative abnormal returns, controlling two and four fixed effects respectively As we

expected, in all models, the coefficients of Treat × Post become positive and significant

suggesting that the Vega effect in bank’s CEO risk taking behavior of treatment firms changes

after the event The findings in this section help us to isolate the Vega effect on spread, more

support our hypothesis 1 that banks where CEOs have higher risk-taking incentives (Vega)

charge a significantly lower loan spread More importantly, the DiD regressions allow us to

resolve any potential endogeneity concerns such as omitted variables

[Insert Table 7 here]

For support the validity of DiD test, Panel B of Table 7 shows parallel trend assumption

for three years before and after the policy shocks B_1, B_2 and B_3 stand for one, two and

three_year before the event; A_1, A_2 and A_3 stand for one, two and three_year after the

event; C stand for current year of event In four models, we control the year fixed effect and

expect different CEO behaviors after the exogenous shock The results actually come beyond

our expectation with the coefficient of Post is positively statistically significant in first 3

models (indicating that there is significant difference between pre and post period),

furthermore, nearly all coefficient of B_1, B_2 and B_3 is not statistically significant but a

great deal of A_1, A_2, A_3 is significant at the 1% level This evaluation strongly makes our

DiD tests more reliable, therefore powerfully support out main hypothesis of CEO risk taking

incentives in bank lending decisions

4.5 CEO risk-taking-incentive effect: Corporate-governance channels

The literature on corporate governance suggests that a less effective governance

mechanism leads firms to greater agency problems and CEOs at firms with greater agency

problems receive greater compensation (Core, Holthausen, and Larcker, 1999) Therefore, we

can undoubtedly perceive the vital role of a good governance channel in reducing agency risk

and further enhance firm value Concerning mechanism of board quality, it cannot be denied

that independent directors, institutional ownership, board size, female and academic are all

important indicators (Coles, Daniel, and Naveen 2008; Linck, Netter, and Yang 2008) For

example, higher number of independent directors and greater institutional ownership are

associated with lower bond yields and higher credit ratings (Bhojraj and Sengupta, 2003);

outside directors have greater incentives to monitor management because of reputation

considerations (Fama and Jensen, 1983; Weisbach, 1988) and the probability of financial

statement fraud is lower for firms with independent boards (Beasley, 1996) When it comes to

board size, there are two schools of thoughts On the one hand, small boards of directors are

more effective (Jensen 1993, Yermack 1996) On the other hand, Adams and Mehran (2012)

argue for a positive relation between bank performance and board size According to Hillman,

Cannella, and Paetzold (2000), larger board tend to be more effective than smaller board in

terms of information, skills, and legitimacy To some extent, larger board is considered to be

good, however, when the board become too large, it will be less efficient (Eisenberg et all,

1998; Cheng, 2008) So too small or too large is also not a good governance mechanism

Regarding the gender issue, Adams and Ferreira (2009) suggest that female directors have

better attendance records than male directors, and gender-diverse boards allocate greater effort

to monitoring More recent studies find certain relations that companies with directors from

academia are associated with higher performance (Francis, Hasan, and Wu 2012, 2015)

This study proposes that the effect of CEO risk-taking incentives derives from the corporate

governance channel The agency problem that stems from compensation could be the reason why

banks charge lower loan spreads or provide more favorable non-price terms to borrowers

Therefore, in banks with diverse corporate governance, the Vega incentive may have a different

effect on bank loan contracts Based on prior studies, we suggest that the effect of risk-taking

behavior (Vega) would be weaker when banks have strong governance practices Thus, in this

section, we further examine this issue by including the interaction terms and the full sample to

examine how corporate governance influence the effect of risk taking behavior (Vega)

Following the study of Hoechle, Schmid, Walter, and Yermack (2012), we use several

different variables to measure corporate governance structure and run OLS regressions with

the interaction term of Vega and a set of dummy variables for weak corporate governance We

adopt the following proxies of corporate governance: Independent directors (ID), Instirutional

Ownership (IO), board size (BS), female (Fem), and academic (Aca) All the corporate

governance measures are defined in Appendix A If CEO risk-taking incentives are viewed as

an agency problem, then we would expect the effect of Vega to be moderate when banks are

associated with strong corporate governance quality

Data on corporate governance indicators are collected from the RiskMetrics Governance

and Director Database and Thompson Reuters The dummies for good corporate governance

proxies are defined as follows: (1) banks with number of independent directors more than the

median (ID_H); (2) banks with institutional ownership more than the median (IO_H); (3) banks

with the number of directors on the board that is in the middle group (31% to 70%, BS_M); (4)

banks with the number of female directors more than the median (Fem_H); and (5) bank with

at least one board member is from academia (Aca_H)

Table 8 displays the results of corporate governance testing The coefficients of vega in

almost all specifications remain negative and significant, thus confirming our first hypothesis

The interaction terms of Vega and ID_H, IO_H, BS_M, Fem_H, and Aca_H are positive and

significant The evidence seems to be consistent with our expectations, the overall results

confirm that the CEO risk taking incentive is moderated in banks with higher quality of

corporate governance Accordingly, Vega can be viewed as a source of agency problem

between bank manager and shareholders

[Insert Table 8 here]

4.6 Robustness checks

4.6.1 Control for CEO characteristics and other compensation schemes

CEO characteristics and other forms of CEO compensation may also affect considerations

of loan spread when banks negotiate contracts with creditors (Donelson, Jennings and McInnis,

2014; Kabir, Li, and Veld-Merkoulova, 2013) In this study, we claim that CEO Vega

represents a form of executive compensation, although not a compulsory regular form such as

salary, bonus or stock options However, the economic intuition of Vega, in terms of either

agency cost proxy or the complementary incentive-alignment effect, may be more meaningful

in comparison with other forms of compensation Therefore, to confirm that our results are not

biased by other CEO characteristics and compensation, such as CEO inside debt ratio, TDC1,

Delta and CEO tenure we additionally include the latter in our regressions

Table 8 presents the regression results of bank CEO risk-taking incentives on bank loan

spread by considering CEO characteristics and other CEO compensation Across all

specifications, the coefficients of Vega are significantly negatively related to loan spread and

cumulative abnormal returns even when we control for all CEO fixed effect, year fixed effect,

loan purpose, loan type and other compensation For example, in Model (4), the coefficient of

Vega is -0.0244, which is statistically significant to loan spread with t-statistics of -2.16 Thus,

the evidence still supports our hypothesis 1

[Insert Table 9]

4.6.2 Bank size analysis

Bank size can be also considered an important factor in the determination of bank lending

decisions (Berger and Bouwman, 2009 and 2013) To some extent, bigger banks are associated

with strong corporate governance, more market power, good reputation and trust, since they can

easily mobilize deposits (even at lower rates) and attract higher loan demands (even at higher rates)