How Do Managerial Styles Differentiate Across Bank Reactions to Systemic Crises? *

Transcription

1 How Do Managerial Styles Differentiate Across Bank Reactions to Systemic Crises? * Yu Shan April 29th, 2018 Abstract In the face of systemic crises, some bank managers reduce credit risk in their loan portfolios, whereas others exploit potential government bailouts and increase their internal credit risk. I use a connectedness sample method to differentiate risk taking across managerial styles. I find that asset innovators most aggressively reduce within-bank credit risk during financial crises, whereas liability innovators increase internal bank credit risk. However, during non-crisis periods, a bank s credit risk is positively related to its systemic risk exposure, indicating a baseline risk-taking proclivity. Results are robust with regard to within-loan, dynamic generalized method of moments, and lead-lag analysis. JEL Classifications: G20, G21, G24, G28. Keywords: Systemic Risk, Credit Risk, Public Guarantee, Syndicated Bank Loans, Manager Style, Risk-taking, Bank Governance. * I m grateful to Linda Allen for her invaluable support and guidance over the course of this project. I would like to also thank Ayan Bhattacharya, Gayle Delong, Sonali Hazarika, Armen Hovakimian, Michael LaCour-Little, Rajarishi Nahata, Lin Peng, Yao Shen, Nancy Wallace, Jun Wang, Ayako Yasuda, Yildiray Yildirim, Dexin Zhou, and seminar/conference participants at the Baruch College Brownbag Seminar and Financial Management Association 2017 Annual Meeting for their insightful suggestions and comments. All errors are my own. Ph.D. candidate in finance, Bert W. Wasserman Department of Economics and Finance, Zicklin School of Business, Baruch College, New York, NY 10010, yu.shan@baruch.cuny.edu.

2 1. Introduction What are the reactions of financial firms to potential systemic crises or changes in their systemic importance? What determines their reactions? These topics form a part of the broader discussion of management and regulation of systemically important financial institutions. When a systemically important bank fails, there would be significant disruption to financial markets and the overall economy, thereby justifying implicit or explicit government guarantees and subsidies. Previous research has discussed the effect of government guarantees on bank risk-taking behavior. Several researchers have argued that bailout expectations may increase banks risk-taking behavior. This, in turn, may imply a decline in market discipline (Flannery, 1998; Sironi, 2003; Gropp, Vesala, and Vulps, 2006; Acharya, Anginer, and Warburton, 2016) and increase in moral hazard (Kane, 1989; Demirguc-Kunt and Detragiache, 2002; Diamond and Rajan, 2009; Farhi and Tirole, 2012). Other researchers have challenged these conclusions and have proposed a risk-reducing effect driven by higher charter values (Cordella and Levy-Yeyati, 2003), lower undiversifiable contagion risk across banks (Freixas, Parigi, and Rochet, 2000; Allen and Gale, 2001; Diamond and Rajan, 2005; Dell Ariccia and Ratnovski, 2013; Choi, 2014), and clawback provisions (Allen and Li, 2011). Although extant research focuses on establishing a connection between implicit government guarantees and banks risk-taking behavior, these studies do not explore the direct association between risks taken by individual financial institutions and their systemic risk contribution. Current systemic bank regulations, which include countercyclical capital buffers and total loss-absorbing capacity requirements, are considered to be add-ons and do not fully integrate established individual bank risk taking measures (such as risk-based capital regulations) into newer systemic risk regulations. Bank regulations would lead to suboptimal outcomes if an existing association between systemic risk and disaggregated risk exposure on an individual bank basis is ignored. For example, if banks consistently reduce their individual risk exposure during a rise of systemic risk, then systemic bank regulations can be milder due to this automatic stabilizer. Alternatively, if banks regularly aggravate their own risk exposure by exploiting potential moral hazard advantages, systemic risk regulation would not be enough to address system-wide crises. This paper aims to complement extant literature by determining a direct relation between systemic risk and each individual bank s risktaking behavior. In particular, the paper examines whether and how banks internally adjust the credit risk in their syndicated bank loan portfolios as their measure of systemic risk changes. This paper focuses on credit risk because of its dominance in the internal risk exposure of banks (estimated to be approximately 60% of a bank s total risk). Furthermore, this paper defines how banks adjust their credit risk exposure to systemic risk changes as the banks credit-risk sensitivity to systemic risk. 1

3 In this study, I develop and test an empirical measure of each bank s credit-risk sensitivity to systemic risk. I specifically focus on banks credit risk-taking behavior concerning the newly originated syndicated bank loans. Credit risk is a major source of idiosyncratic risk in bank portfolios, and syndicated loans expose syndicate members to a choice of credit risk levels by deciding on participation and the amount of allocation share to contribute in a loan. I use data on syndicated bank loans to examine whether financial institutions adjust their credit risk-taking in response to levels of systemic risks. I measure the credit risk using borrower distance-to-default during loan origination while controlling for several loan characteristics. During periods of no recession or low macro-level systemic risk, bank systemic risk is positively related to credit risk-taking. That is, the more the bank contributes to overall systemic risk, the higher the credit risk in the syndicated bank loan portfolio. This indicates that banks may develop a tendency to seek abnormal returns from risk-enhancing activities. The study provides a robust result that applies to several different specifications, which control for loan, borrower, and bank characteristics, lending relationships, macroeconomic conditions, in addition to a series of fixed effects and endogeneity controls. The study findings reveal that during economic recession or a heightened risk of systemic crisis, there is a weakening in the positive relation between systemic risk and credit risk-taking. Moreover, the banks with the highest systemic risk lower the credit risk in their loan portfolios the most. This implies that during times of economic volatility, the increased risk of insolvency with the potential loss of bank charter value induces banks to reduce their insolvency risk by reducing their credit risk exposure. This acts as an automatic risk-taking stabilizer in the banking system, linking systemic risk to credit risk on an individual bank basis. The study results indicate that during times of volatility, systemically important banks maintain high levels of systemic risk to benefit from government bailouts even while minimizing their credit risk exposure. This may alleviate some of the uncertainty about whether the federal government will support a particular distressed bank. That is, increased systemic risk and reduced loan portfolio credit risk together may improve the chances of government support. These results are consistent with the Baker, Bloom, and Davis (2016) policy uncertainty index, which hit a peak after Lehman Brothers failed, as well as the subsequent reforms codified in the Dodd Frank Act and the Consumer Protection Act, which explicitly aim to reduce expectations of government shield. However, there is much heterogeneity across different banks in their adjustment of credit-risk sensitivity to systemic risk during systemic crises. The most risk-averse banks (as measured by the lower ex-ante credit risk in their syndicated bank loan portfolios) usually retract their internal risk taking when systemic crises are imminent, thereby reducing their credit risk exposure much more than less risk-averse (higher credit risk) banks. On a portfolio basis, the overall sign of the measures of the credit-risk sensitivity to systemic risk flips from negative to positive. In other words, the higher the bank s systemic risk, the lower the bank s 2

4 credit risk in its syndicated bank loan portfolio. This suggests a bank-specific idiosyncratic component. In this paper, I empirically identify this heterogeneity in risk-taking proclivity across banks as managerial style. Extant banking literature mainly assesses the impact of systemic importance and implicit government guarantees on banks risk-taking from an organizational perspective. It ignores the influence of human factors in shaping a bank s responses to potential systemic crises and changes in their bank s systemic importance. Top managers are the main decision makers in a wide range of corporate decisions (Bertrand and Schoar, 2003). Banking studies have suggested that executive compensation, board characteristics, and ownership structure can significantly impact bank risk-taking and bank performance (Beltratti and Stulz, 2012; DeYoung et al., 2013; Erkens et al., 2012; Berger et al., 2014; Minton et al., 2014). However, literature on how idiosyncratic executive-specific effects help explain bank risk-taking behavior is scarce. This study examines the role of managerial styles in determining the variation in banks internal risk-taking responses to changes in systemic importance or potential systemic crisis. It complements the literature in that it investigates how executive attributes or styles, which are independent of observable compensation incentives (e.g., Delta, Vega, and Bonus) and biographical characteristics (age and gender), help explain a large variation in bank s credit risk-taking sensitivity to systemic risk. The paper further examines managerial attitudes toward risk-taking (i.e., managerial style) and finds that differences among different managers, not banks, explain the heterogeneous response of credit risk to heightened systemic risk exposure. This result is consistent with that of Hagendorff et al. (2017). They argued that extreme risk-taking and other unsustainable business models in banking could ultimately be a people problem rooted in the idiosyncratic preferences of individuals and are not easily restrained by regulators and investors. This paper analyzes how idiosyncratic executive-specific effects explain the variation in credit-risk sensitivity to bank-level systemic risk in order to identify the implications of managerial style for risk-taking. I find that banks credit-risk sensitivity to systemic risk is significantly affected by few idiosyncratic manager preferences, that is, managerial styles. Focusing on executives such as CEOs, CFOs, COOs, and executive directors and using the connectedness sample method of Abowd, Kramarz, and Margolis (1999) (AKM, thereafter), I run a series of three-way fixed effects regressions (bank, executives, and year) to estimate the bank and executive fixed effects. Furthermore, I conduct factor analyses for the executive (bank) fixed effects to find dominant factors in explaining patterns across the styles of individual executives (banks). The two dominant factors show two dimensions of innovations that executives and banks show as preferences: asset side and liability side. Executives asset- and liability-side innovation preferences significantly affect a bank s credit-risk sensitivity to systemic risk. 3

5 Based on the loadings of individual executives on the two main factors, I assign each executive into one of four types: (1) Asset innovator but liability traditionalist; (2) Asset and liability innovator; (3) Asset and liability traditionalist; (4) Asset traditionalist and liability innovator. Intuitively, asset innovators are more likely to diversify into mortgage-backed securities and derivatives, whereas traditionalists focus more on lending. Unlike liability traditionalists, liability innovators are more likely to access non-deposit sources of funds. Banks managed by different types of executives exhibit different credit risk-taking sensitivities to systemic risk. In particular, asset innovators restrain and reduce the bank s credit-risk sensitivity to systemic risk during financial crises, whereas liability innovators increase the bank s credit-risk sensitivity to systemic risk during crises. This effect is unique to a manager, rather than a bank, indicating that human attributes, not institutional reasons, may be the leading cause for the often-cited too-big-to-fail problem. In robustness checks, the result is not replicated for a placebo test. The endogenous lender borrower matching is a concern in relation to the baseline model of creditrisk sensitivity to systemic risk. Thus, the results are subject to possible borrower self-selection despite the use of controls and fixed effects in my analysis. A few unknown characteristics of borrowers might induce borrowing firms to choose certain banks, thereby introducing selection bias into the analysis. Borrowers may also respond to bank risk-taking. For example, risky borrowers may depend more on stable refinancing sources during economic downturns and systemic crises. Therefore, risky borrowers may request loans from systemically important banks protected by public guarantees; therefore, they are more likely to survive during economic turmoil. Thus, distinguishing the supply-side from demand-side effects is important. I adopt two methods in this study. First, I conduct a two-stage least-square analysis and employ the geographic distance and number of banks in the state of the borrower as instruments for observed lending relationships. Second, I employ a series of within-loan analyses to empirically control for the demand effects of lending at systemically important banks. I use a within-loan regression analysis that controls for borrower self-selection of lenders to test my findings of a direct relation between systemic risk and credit risk (following the methodology of Chu, Zhang and Zhao, 2017). Based on the fact that a syndicated loan often has multiple lenders, I examine how the systemic risks of banks that fund the same loan affect their contributions to the loan, that is, within-loan estimation, which eliminates any fluctuation on the demand side. I find that systemically risky banks contribute a larger (smaller) portion to risky (safe) loans. The results are more significant for loans with excessive risks compared with other loans originated in the same year. This finding suggests that the direct relationship between systemic risk 4

6 and internal bank credit risk taking during non-crisis periods is driven by supply factors, thereby indicating underlying bank risk-taking proclivities. Endogeneity may also arise from reverse causality such that prior credit risk-taking may affect the current values of bank systemic risk. In fact, current values of systemic risk may not be independent of the credit risk-taking on previous loans. Thus, ignoring the dynamic nature of the independent and dependent variable relation may yield biased and inconsistent estimates (Wintoki, Linck, and Netter, 2012). Two methods are adopted to alleviate the reverse causality concern. First, I use a set of leadlag regressions and find that the causality runs from systemic risk to credit risk and not vice versa. Second, the method by Wintoki, Linck, and Netter (2012) is adopted using a dynamic panel generalized method of moments (GMM) estimation, which confirms the baseline analysis. The rest of the paper is structured as follows. Section 2 introduces the model development. Section 3 reports the variable construction. Section 4 discusses the data and sample construction. Section 5 presents the empirical results, including the baseline results, a two-stage least-squares analysis controlling for borrower self-selection, a lead-lag analysis, and within-loan estimations. Section 6 provides robustness checks, including a different version of within-loan regression and a dynamic panel GMM analysis. Section 7 focuses on how idiosyncratic executive-specific styles explain the variation in credit risk-taking sensitivity to bank-level systemic risk. Section 8 presents the conclusion. 2. Model Development 2.1. The Credit-Risk Sensitivity Model This paper is based on the broad literature on bank bailout and its effects on bank risk-taking. Banks systemic risk, or systemic importance, is directly related to the likelihood of government bailout when the banking firm is in distress, and government credit guarantees can reduce investors risk exposures. Accordingly, clear evidence of market discipline may be difficult to find (Flannery, 1998). Empirical literature has found supporting evidence that the implicit safety nets weaken the predictive power of bank equity and debt prices on bank fragility (Gropp, Vesala, and Vulps, 2006). Furthermore, debt spreads are insensitive to bank risk for debts issued by public sector banks (Sironi, 2003) or largest financial institutions (Acharya, Anginer, and Warburton, 2016). Therefore, banks find it profitable to adopt a risky balance sheet and correlate their risk exposures (Farhi and Tirole, 2012). Some studies have claimed that safety nets in the banking industry lead to additional risk-taking by protected banks (Dam and Koetter, 2012) as well as their competitors (Gropp, Hakenes, and Schnabel, 2011). At the macroeconomic level, Demirguc-Kunt and Detragiache (2002) found that explicit government guarantees increase overall bank 5

7 fragility. In particular, with regard to the Troubled Asset Relief Program (TARP) recipients, Black and Hazelwood (2013) found that large TARP banks initiate riskier loans, which suggests moral hazard due to government support. In contrast to this argument on moral hazard, some studies have proposed a charter value effect through which banks may decrease risk-taking in response to expected public guarantees (Cordella and Levy-Yeyati, 2003). Hakenes and Schnabel (2010) indicated that protected banks may take lower risks when transparency in the banking sector is low and the deposit supply is sufficiently elastic. This paper complements the extant research by investigating the direct relation between financial institutions risk-taking in isolation and their systemic risk contribution. It examines how banks endogenously change their credit risk in their loan portfolio in response to changes in the systemic risks, termed as banks credit risk-taking sensitivity to systemic risks. Two complementary measures of systemic risk are used in this paper. CATFIN (Allen, Bali and Tang, 2012), which is a cross-sectional measure that identifies the overall level of systemic risk in the financial system at each point in time, is used to measure the macro-level aggregate systemic risk. This helps examine whether high levels of aggregate systemic risk that are likely to lead to government bailouts or other interventions induce banks to adjust the credit risk in their loan portfolios upward or downward. ΔCoVaR is used to measure the microlevel systemic risk contribution of each bank (Adrian and Brunnermeier, 2016) to determine the impact of an individual bank s insolvency on systemic risk in the overall financial system. The greater the contribution of an individual bank s insolvency to market-wide systemic risk, the greater is its imposition of systemic risk onto the macroeconomy. I use borrower distance-to-default as the key dependent variable to investigate whether the bank s systemic risk exposure (CATFIN and ΔCoVaR) is related to its decision to lend to borrowers of differing risk levels. I employ an array of control variables such as loan, borrower, and bank characteristics, lending relationships, macroeconomic conditions, as well as a series of fixed effects. A possible concern about the baseline regressions is the endogenous lender borrower matching such that the tests are subject to borrower self-selection. Some unobservable borrower characteristics may induce borrowing firms to choose certain banks, thereby introducing selection bias into the analysis. Borrowers may also respond to bank risk-taking. For example, risky borrowers may depend more on stable refinancing sources, especially during economic downturns. Therefore, risky borrowers may request loans from systemically important banks, which may be protected by public guarantees; thus, they are more likely to survive during periods of economic turmoil. The focus of this study is to examine whether banks control the credit risk in their loan portfolios by adjusting their credit and underwriting standards. Therefore, it is important to separate the supply-side from demand-side effects. This is done using two methods: first, I 6

8 conduct a two-stage least-square analysis and employ the geographic distance and number of banks in the state of the borrower as instruments for observed lending relationships. In the first stage, I regress the lending relationship variable on geographic distance and number of banks and other independent variables. In the second stage, I rerun the baseline regression using fitted value of lending relationship in all specifications. This method allows controlling for the probability of originating a loan between a firm i and a bank j that is independent of the channel of systemically risky banks selecting borrower with certain default risks. Second, I use a series of within-loan analysis to empirically control for the demand effects of lending at systemically important banks. In other words, I use a within-loan regression analysis that controls for borrower self-selection of lenders to test the findings of a direct relation between systemic risk and credit risk (Chu, Zhang and Zhao, 2017). Exploiting the unique feature that a syndicated loan often has multiple lenders, I examine how the systemic risks of banks that fund the same loan affect their contributions to the loan, that is, within-loan estimation, which eliminates any fluctuation on the demand side. Endogeneity may also arise from reverse causality such that current values of bank systemic risk can be affected by credit risk-taking in previous periods. In fact, current values of systemic risk may not be independent of the credit risk-taking on previous loans, and ignoring the dynamic nature of the independent and dependent variable relation may yield biased and inconsistent estimates (Wintoki, Linck, and Netter, 2012). I employ two methods to alleviate the reverse causality concern. First, using contemporaneous borrower s distance-to-default (or lender s ΔCoVaR) as the dependent variables, I use a set of lead-lag regressions and test whether they are associated with lagged lender s ΔCoVaR (or lagged borrower s distance-to-default). Second, the method of Wintoki, Linck, and Netter (2012) is adopted using a dynamic panel GMM estimation. There are several advantages of using dynamic panel GMM method. First, it allows current explanatory variables to be influenced by previous realizations of dependent variable. Second, it eliminates unobservable heterogeneity by first differencing all dependent and independent variables. Third, it uses the combination of variables from the history as valid instruments to account for simultaneity. The dynamic panel GMM estimation confirms the earlier findings Managerial Style Model Extant literature on banking mainly assesses the impact of systemic importance and implicit government guarantees on banks risk-taking from an organizational perspective. It ignores the influence of human factors in shaping a bank s responses to potential systemic crises and changes in their bank s systemic importance. Top managers have a significant influence on a wide range of corporate decisions (Bertrand and Schoar, 2003). Recent banking studies have suggested that 7

9 executive demographic characteristics and compensation, board characteristics, and ownership structure can significantly impact bank risk-taking and bank performance (Beltratti and Stulz, 2012; DeYoung et al., 2013; Erkens et al., 2012; Berger et al., 2014; Minton et al., 2014; Nguyen, Hagendorff and Eshraghi, 2017). In particular, Berger, Kick and Schaeck (2014) indicated that demographic characteristics of executive teams influence corporate governance in banking. Faccio, Marchica, and Mura (2016) documented CEO gender differences in risk-taking and capital allocation efficiency. However, these studies have not examined the idiosyncratic executive-specific effects that explain the variation in banks responses to changes in systemic importance or potential systemic crisis. The study focuses on a time-invariant measure of each individual manager s attitude toward risktaking, termed as the manager s style, instead of focusing on time-varying observable characteristics such as compensation and demographic characteristics. It adopts the method proposed by Hagendorff et al. (2017). They suggested that compensation and various other observable executive characteristics can only describe a small amount of the variation in banks business models and risktaking preferences. They found that much of the variation in bank business policy can be explained by time-invariant executive factors ( styles ), which explain the risk-taking culture in some banks. Since bank credit risk-taking sensitivity to systemic risk is mostly related to bank business models and how banks are managed (aggressively or conservatively) by executives, its variation may also be explained by the unobservable, time-invariant executive fixed effects. This paper complements the literature by investigating how executive attributes or styles, which are independent of observable compensation incentives (delta, vega, and bonus) and biographical characteristics (age and gender), explain a large variation in bank s credit risk-taking sensitivity to systemic risk. 3. Variable Constructions 3.1. Systemic Risks Two complementary measures of systemic risk are used in this paper. CATFIN is used to measure the macro-level aggregate systemic risk (Allen, Bali and Tang, 2012). It is a cross-sectional measure that identifies the overall level of systemic risk in the financial system at each point in time. It examines whether high levels of aggregate systemic risk that are likely to lead to government bailouts or other interventions induce banks to adjust the credit risk in their loan portfolios upward or downward. ΔCoVaR is used to measure the micro-level systemic risk (Adrian and Brunnermeier, 2016). It determines the impact of an individual bank s insolvency on systemic risk in the overall financial system. The greater the contribution of an individual bank s insolvency to market-wide systemic risk, the greater the individual bank s 8

10 imposition of systemic risk on the macroeconomy. This measure determines whether the bank s systemic risk exposure impacts its decision to lend to borrowers of differing risk levels. To construct the macro-level systemic risk, CATFIN 1, the method proposed by Allen, Bali, and Tang (2012), is used and VaR at the 99% confidence level is estimated using three different methodologies: the generalized Pareto distribution (GPD), the skewed generalized error distribution (SGED), and the nonparametric estimation method based on the left tail of the actual empirical distribution without any assumptions about the underlying return distribution. CATFIN is defined as the arithmetic average of the GPD, SGED, and nonparametric VaR measures. Allen, Bali, and Tang (2012) suggested that the risk of macroeconomic slump increases when CATFIN is above some early warning level, which is determined using the Chicago Fed National Activity Index (CFNAI) as a benchmark. CFNAI is an index of 85 existing monthly economic indicators. The Federal Reserve Bank of Chicago denotes the three-month moving average of CFNAI (CFNAI-MA3) value of 0.7 as a turning point indicating economic contraction. Allen, Bali, and Tang (2012) demonstrated that when CATFIN is above some early warning level, it can significantly predict lower economic activity (CFNAI-MA3) one to eight months in advance of the slump. Therefore, CATFIN offers an early warning to alert regulators regarding the risk of economic recessions. This paper also tests whether banks credit risk-taking is affected by whether CATFIN breaches the early warning level. Following Allen, Bali, and Tang (2012), I construct an early warning dummy, which is equal to 1 if CATFIN is above the early warning level and 0 otherwise. For each quarter t, the early warning level is calculated as the median CATFIN using all observations up to quarter t in which CFNAI-MA3 falls below 0.7. To address possible reverse causality, both contemporaneous and lagged measures of systemic risk are used to examine the following quarter s credit risk in the bank s loan portfolio. A historical monthly CATFIN, CFNAI-MA3, Early Warning Level, and Warning dummy are presented in the Online Appendix. The micro-level systemic risk is measured by following the methodology used by Adrian and Brunnermeier (2016) to generate time-varying ΔCoVaR 2. First, I run the following quantile regressions in the weekly data (where j is a financial institution): X j t = α j q + γ j q M t 1 + ε j q,t, (4) 1 I thank Linda Allen and Yi Tang for providing the data on CATFIN. 2 I thank Tobias Adrian and Markus Brunnermeier for sharing their code for computing CoVaR. 9

11 X system j t = α system j q M t 1 + β system j q M t 1 + β system j q X j t + ε system j q,t, (5) where X j system j t denotes the weekly return of institution j in week t, X t denotes the financial sector return in week t, and M t 1 denotes a vector of seven systematic state variables in week t, including three-month yield change, term spread change, TED spread, credit spread change, market return, real estate excess return, and equity volatility. Thereafter, I generate the predicted values from these regressions to obtain j VaR q,t = α q j + γ q j M t 1, (6) j CoVaR q,t = α q system j + r q system j M t 1 + β q system j VaR j q,t, (7) j Finally, I compute the ΔCoVaR q,t for each institution: j ΔCoVaR q,t j = CoVaR q,t j CoVaR 50,t = β system j j (VaR q,t j VaR 50,t ) (8) These regressions result in a panel of weekly ΔCoVaR j j q,t. A quarterly time series of ΔCoVaR q,t is obtained by averaging the weekly risk measures within each quarter. Throughout the paper, q equals 99%, but the results are robust to the 95% level Credit Risk-Taking This paper uses the Merton distance-to-default as a measure of borrower default risk. Since credit risk can also be affected by loan characteristics such as loan amount, maturity, and being secured or not, I control for an array of loan characteristics. The method followed by Bharath and Bharath and Shumway (2008) and Crosbie and Bohn (2003) is adopted in calculating Merton s distance-to-default. The market equity value of a company is modeled as a call option on the company s assets: V E = V A e dt N(d 1 ) Xe rt N(d 2 ) + (1 e dt )V A (1) d 1 = log(v A X )+(r+s A 2 2 )T s A ; d 2 = d 1 s A T (2) where V E denotes the market value of a firm s equity, which is calculated from the CRSP database as the product of share price at the end of the quarter and the number of shares outstanding. X denotes the face value of debt maturing at time T, which is calculated as debt in current liabilities (COMPUSTAT data item 45) plus one half of long-term debt (COMPUSTAT data item 51). V A denotes the value of the firm s assets; r denotes the risk-free rate, which is defined as the 1-year Treasury Constant Maturity Rate obtained from 10

12 the Board of Governors of the Federal Reserve System. s A denotes the volatility of the value of assets. I simultaneously solve the above two equations to find the values of V A and s A. Quarterly Merton s distanceto-default is finally computed as Distance to Default = log (VA ) + (m s 2 A )T X 2 s A T (3) I also calculate quarterly distance-to-default as a robustness check by first calculating the monthly distanceto-default and then taking quarterly averages. Both measures generate similar results in the baseline regressions Bank Business Policy Variables Banks follow different business policies characterized by the selection of different profit-earning assets and funding sources. Based on bank balance sheet data, I follow Hagendorff et al. (2017) and utilize eight bank policy variables that reflect strategic management choices. First, I include noninterest income over the sum of interest and noninterest income (Noninterest income) as a bank s choice between lending-based and feebased activities. Similarly, I use the ratio of loans over total assets (Loans) to measure a bank s focus on interest-generating assets. On the liability side, I measure banks funding structures using loans over total deposits (Loans/Deposits) and the proportion of liabilities that are not financed via deposits (Nondeposit funding). To identify banks with more focus on the trading book and on the banking book, two variables are used: the ratio of mortgage-backed securities over total assets (MBS) and the logarithm of the ratio of the notional amount of derivative contracts held for trading to total assets (Derivatives). To describe banks efforts to reduce diversifiable risk, the term lending diversification is used, which is defined as 1 minus the Herfindahl Index of the shares of real estate, commercial and industrial, consumer, and other loans over total assets. Finally, a bank s funding liquidity risk-taking is measured using the difference between liabilities repricing or maturing within 12 months and assets repricing or maturing with 12 months, scaled by total assets. Detailed definitions for the bank business policy variables are defined in Appendix I Compensation Incentives Literature has shown that banks risk-taking behavior can be influenced by manager compensation incentives (Fahlenbrach and Stulz, 2011; DeYoung et al., 2013; Berger et al., 2014; Nguyen et al., 2017). Executive compensation influences managerial risk preferences through executives portfolio sensitivities to changes in stock prices (Delta) and stock return volatility (Vega). Delta is the sensitivity of manager wealth to bank performance. It is calculated as the dollar changes in manager wealth to stock price 11

13 performance. Vega measures the sensitivity of manager wealth to bank risk, and it is calculated as the dollar change in wealth associated with a 1% increase in stock return volatility. If riskier bank policy increases stock price and volatility, then managers with high Delta or Vega have more incentives to engage in riskier bank policies. Both Delta and Vega are scaled by cash compensation and transformed to natural logarithm. I also control for the log of cash bonus (Bonus) since it is distinct from equity-based compensation and might impact a manager s risk preferences Control Variables I use a set of control variables for loan, borrower, and lender characteristics. To ensure that outliers do not strongly influence statistical results, all observations are set higher than the 99th percentile of each variable to that value; all values lower than the 1 st percentile of each variable are similarly winsorized. All variables are defined in Appendix I. The first set of control variables include bank characteristics such as Bank Total Assets, Bank Capital Ratio, Return on Equity, Liquidity, Loan Charge-offs, Loan Loss Allowance, and Risk-Weighted Assets. Since ΔCoVaR is constructed using market data and most public banks are bank holding companies, bank financial information is measured at the bank holding company level. ln(bank Total Assets) is defined as the natural logarithm of bank total assets (in billions); Bank Capital Ratio is defined as the bank s total capital over bank s total assets; Bank ROE is defined as bank net income over book equity; Bank Liquidity is defined as the sum of cash and available-for-sale securities divided by bank total assets; Loan Chargeoffs is defined as the total charge-offs on loans and leases divided by bank total assets; Loan Loss Allowance is defined as the total allowance for loan and lease losses divided by bank total assets; Risk-Weighted Assets is defined as total risk weighted assets divided by bank total assets. Although I include both lead lenders and participants in all the regressions in the study, I add a lead bank dummy in the within-loan regressions to account for possible unobservable differences between lead and nonlead banks. A bank is defined as a lead lender if its lend arranger credit variable is Yes in Dealscan. The second set of control variables comprises borrower characteristics. ln(borrower Total Assets) is defined as the natural logarithm of total assets (in billions); Tangibility is defined as total property, plant, and equipment divided by total assets; Leverage is defined as the total debt divided by total assets. I also include a lending relationship measure, as in the study by Bharath et al. (2007), since the intensity of past lending relationships can influence the matching between borrowers and lenders. The lending relationship between borrower i and bank j is defined as the dollar amount of loans to borrower i by bank j in the past five years over the total dollar amount of loans by borrower i in the past five years. 12

14 Thereafter, I control for an array of loan characteristics. ln(package Amount) is defined as the natural log of the package amount, where package amount is measured in millions. ln(package Maturity) is defined as the natural log of the maturity of the deal in months. Package maturity is calculated as the value-weighted average of the facility maturities. ln(no. of Lead Banks) is the natural log of the number of lead lenders in the deal syndicate. Finally, I control for macroeconomic conditions. Quarterly GDP per capita growth rate is used to measure the macroeconomic performance. In addition, a recession dummy (Recession) is generated to test the differential effect of bank systemic risk on credit risk-taking in expansion and recession periods. The recession dummy equals 1 if the month of loan origination is designated to be a contraction month by NBER and 0 if it is designated to be an expansion month. 4. Data 4.1. Sample Construction The sample period of this study spans Q to Q Banks quarterly financial data are collected from the Consolidated Financial Statements for Holding Companies (FR Y-9C) available on the Federal Reserve Bank of Chicago website. Market data are obtained from CRSP. Borrowing firms quarterly financial data are collected from Compustat. Syndicated loan data are obtained from Loan Pricing Corporation s (LPC) Dealscan loan database. The Dealscan database contains historical information on the terms and conditions of deals in the global syndicated bank loan market. Borrower financial data are linked to Dealscan using the Dealscan-Compustat linking data provided by Chava and Roberts (2008, updated in August 2012). To construct the sample, I first start with a sample of 219,023 deal packages newly originated between January 1995 and December 2013 from the LPC Dealscan database. Since I conduct my analysis at the bank holding company level, I need to first identify the lenders in my sample and their ultimate parent companies. I utilize the information provided by the Federal Reserve System via its National Information Center (NIC) database to identify financial institutions acting as lenders in the study sample. No identity variables are used for lender and ultimate owner in Dealscan because Dealscan overwrites the ultimate owner of the lenders after mergers and acquisitions, that is, the ultimate owner in Dealscan is the ultimate owner at the end of the merger chain. In this analysis, the ultimate bank holding company or lending parent at the time of the issuance of each loan is identified. The NIC database provides detailed information about financial institutions, including types of institutions, establishment time, ownership information, address changes, name changes, and merger and acquisition history. NIC also provides each financial institution s 13

15 RSSD ID, a unique identifier assigned to each financial institution by the Federal Reserve System. Based on the lender information provided by Dealscan, including name, location, and lending history, each lender s RSSD ID is searched manually. Using their RSSD ID (Item RSSD9001 in Call Report and Y-9C), the lender s ultimate owner at the time of loan origination is determined by cross-checking the information contained in Call Report items RSSD ID of Regulatory High Holder 1 (RSSD9348), Financial Higher Holder ID (RSSD9364), and Financial High Holder Percent of Equity (RSSD9365). The three items provide the RSSD ID of a lender s ultimate bank holding company when a lender has an RSSD ID. For a bank that was acquired by another bank and lost its RSSD ID but kept its lending activity, the acquirer s RSSD ID was applied to the acquired bank as its new RSSD ID, and the new ultimate owner of the acquirer can then be identified. The full bank and bank holding companies merger and acquisition history is obtained from the Federal Bank of Chicago. Using the RSSD ID, Dealscan is linked to the Y-9C to obtain bank financial data. Dealscan is also linked to CRSP to collect market data through the PERMCO-RSSD link table provided by the Federal Reserve Bank of New York. The bank characteristics data at the bank holding company level is collected using Y-9C reports. Using the PERMCO-RSSD link, bank holding companies are merged with their systemic risk measure, ΔCoVaR. This process reduces the study sample to 80,193 packages (140,609 package-lender pairs). Next, borrower characteristics from Compustat are merged with the information on corporate loans in Dealscan using the linking table provided by Chava and Roberts (2008, updated in Aug. 2012). Furthermore, the loans originated between September 2012 and December 2013 are linked manually. Owing to differences in capital structures and financing strategies between financial and nonfinancial firms, loans to financial companies (SIC between 6000 and 6999) are excluded from the sample. The utility firms (SIC code falls between 4900 and 4999) are also excluded because they may have different operating and reporting environments. This leads to a sample of 10,915 packages (22,595 package-lender pairs), which include 3,603 unique borrowers and 214 unique banks owned by 66 unique publicly traded bank holding companies with ΔCoVaR data. Since systemic risk ΔCoVaR is mostly measured at the bank holding company level, in the subsequent analysis, all regressions are run at the loan-bank holding company level Summary Statistics for Baseline Regressions Table 1 presents the summary statistics. There are 22,595 lender-package observations in the baseline regressions. Only package-level data are used in the baseline regression. The key dependent variable is borrower s quarterly distance-to-default, which has a mean of 6.772, a median of 6.002, and a standard deviation of The key independent variables are ΔCoVaR, which has a mean of 5.216, a median of 14

16 4.351, and a standard deviation of 2.550, and CATFIN, which has a mean of 2.393, a median of 2.286, and a standard deviation of The average deal amount is US$878 million, with average maturity of 49 months. On average, there are lead lenders in each package. The Lending Relationship between borrower i and lender j, which is defined as the dollar amount of loans to borrower i by bank j in the past five years over the total dollar amount of loans by borrower i in the past five years, has a mean of This indicates that on average, each bank engaged in 47% of the total amount a typical firm borrowed in the five years preceding the loan origination. Borrowers have an average size of $6.752 billion, with a mean tangibility of and leverage of Banks have a mean size of $779 billion, with mean capital ratio of 8.5% and return on equity of 8.3%. Table 2 presents the Spearman correlation matrix for the variables included in the baseline regressions. As shown in Table 2, borrower distance-to-default is negatively correlated with both ΔCoVaR (the bank-specific measure of systemic risk) and CATFIN (the aggregate measure of systemic risk), with correlation coefficients of and 0.289, respectively. This provides preliminary evidence that banks with higher levels of systemic risk lend to borrowers with higher credit risk during periods of high aggregate systemic risk Sample Construction for Within-Loan Regressions To investigate whether the relation between systemic risks and borrower default risk is driven by the demand rather than supply effects, within-loan estimations methodology by Chu, Zhang and Zhao (2017) are applied. By adding package (facility) fixed effects, the impact of the demand-side factors from the supply-side factors can be eliminated. The sample of within-loan regressions is constructed differently from that used in the baseline regressions. First, since borrower, loan, and macroeconomic characteristics drop out in a package or facility fixed effects regression, only lender characteristics are relevant to the within-loan regressions. Second, to control for an array of lender characteristics, a broader set of lender variables is used, which includes bank size, capital ratio, return on equity, liquidity, loan charge-offs, loan loss allowance, and risk-weighted assets, to control for lender characteristics and avoid omitted variable bias. Third, since the bank allocation share variable in Dealscan is employed to measure bank lending at the individual facility/package level, a series of requirements on this variable is added to increase its validity. The sample is constructed as follows. The samples at both package level and facility level are constructed separately because the data are processed differently between the two levels. For the facility-level sample, an initial sample of 15

17 246,260 loan facilities originated between January 1995 and December 2013 is considered. Following Chu, Zhang, and Zhao (2017), I focus on 182,745 facilities that involve credit lines, term loans, or both in this analysis since they are the dominant types of bank loans borrowed by nonfinancial firms. I further require the facility to have at least two banks as lenders because the allocation share for a sole-lender loan is always 100%. This reduces the sample to 152,549 facilities. Based on lenders name, city, state, and dates of their earliest and latest lending activities, their RSSD ID is searched manually through the NIC, and all lenders who do not have an identifiable RSSD ID are excluded. Next, I identify the ultimate holding companies of those banks from the Consolidated Financial Statements for Holding Companies (FR Y-9C), following the same methodologies introduced in Section 4.2, and exclude lenders who do not have an identifiable ultimate holding company in quarter t-1, where t denotes the quarter of loan origination. This reduces the sample size to 80,041 facilities. DealScan reports a bank s allocation share in a facility for approximately 31.78% of all lender-facility pairs. For each bank in each facility, Dealscan reports the allocation share in percentages. Since a bank s allocation share is used to measure bank lending at the individual facility level, loans without bank allocation share information or with allocation share greater than 100% are excluded, which is apparently erroneous. Since ΔCoVaR is estimated at the bank holding company level, I also calculate the allocation share at the bank holding company level for consistency. For example, if both bank A and bank B belong to a same bank holding company C and if bank A and bank B participate in a same facility with allocation shares of 10% and 20%, respectively, then it is treated as bank holding company C contributing 30% to the facility. If the allocation share for A or B is missing or greater than 100%, then the observations of A, B, and C on this loan are excluded because they lead to erroneous allocation share at the level of bank holding company C. In addition, all facilities in which the sum of all lender shares exceeds 110% are excluded. I choose 110% to account for rounding and minor errors. The above treatments lead to a sample of 15,850 facilities. Thereafter, the sample is restricted to loans from banks with nonmissing market equity data, systemic risk data, and Y-9C financial data in quarter t and t-1. This leads to a sample of 10,800 facilities, which involve 148 unique banks. These 148 banks belong to 68 unique bank holding companies. I then identify the borrowing firms from Compustat using the DealScan-Compustat linking table provided by Chava and Roberts (2009, updated in August 2012). For loans originated after August 2012, the changes in the link are manually adjusted by comparing borrower companies names in the two databases. This requires the borrowing firms to have an SIC code and distance-to-default in quarter t-1 and t to be included in the sample. I exclude utility firms and financial firms (two-digit 16

18 SIC code equal to 49 or between 60 and 69). These procedures lead to a sample of 3,797 facilities, which involve 123 unique banks that correspond to 66 unique bank holding companies. The package-level samples are constructed following similar procedures. First, only the observations of a bank holding company j in a package k are retained if the bank allocation share data are nonmissing for all its subsidiaries in all facilities under this package. Otherwise, the bank holding company s observation in package k is entirely removed from the sample. For example, bank A and bank B belong to a same bank holding company C. A contributes 30% only to facility F1, and B precipitates only in facility F2; however, its contribution share data are missing. F1 and F2 are under the same package. In this case, the observation for C is removed entirely from this package because its total contribution will be biased downward due to the missing data. Second, the allocation share for bank holding company j in package k is defined as the total allocation shares of all banks that belong to j and participate in the package k. A bank s allocation share in a package is its allocation share multiplied by facility amount and then divided by package amount. Third, the lead lender dummy for bank holding company j in package k is 1 if at least one of its subsidiaries acts as a lead lender in at least one facility under the package k. The final package level sample includes 3,081 packages, which involve 123 unique banks that correspond to 66 bank holding companies. For both levels, the lending relationship measure (Bharath et al., 2007) is included to control for the intensity of past lending relationship. Observations in the within-loan regressions are by facility-bank holding company pairs, while lending relationship is a bank-borrower level measure. In many cases, two or more banks that belong to the same bank holding company may participate in a same loan. In these cases, their bank level lending relationship measure is added up to create a bank holding company level measure. The rationale behind is that the proprietary information obtained from past lending activities are shared among banks that belong to a same bank holding company. To account for possible unobservable differences between lead and nonlead banks, a lead bank dummy is added in the regressions, which is equal to 1 if a bank is the lead bank in the package/facility and 0 otherwise. A bank is defined as a lead lender if its lead lender credit variable is Yes in Dealscan. The lead lender credit variable is by facility-bank pairs, and two or more banks that belong to a same bank holding company may participate in a same facility; some of them may be lead lenders, while the others are not. To create a bank holding company level lead lender dummy, the dummy is assigned a value of 1 if at least one of its banks acted as a lead lender in the facility. 17