How Does Capital Affect Bank Performance During Financial Crises?

Transcription

1 How Does Capital Affect Bank Performance During Financial Crises? Allen N. Berger University of South Carolina, Wharton Financial Institutions Center, and CentER Tilburg University Christa H.S. Bouwman Case Western Reserve University and Wharton Financial Institutions Center March 2011 The recent financial crisis has raised important issues regarding bank capital. Various reform proposals involve requiring to hold more capital. But assessing these proposals requires an understanding of how capital affects bank performance. Existing theories produce conflicting predictions regarding the effect of capital on bank performance during normal times and have little to say about the effect during financial crises. This paper addresses these issues empirically by formulating and testing hypotheses regarding the effect of capital on three dimensions of bank performance survival, market share, and profitability during financial crises and normal times. We distinguish between two banking crises and three market crises that occurred in the U.S. over the past quarter century. We have two main results. First, capital helps of all sizes during banking crises. Higher capital helps these increase their probability of survival, market share, and profitability during such crises. Second, higher capital improves the performance of small in all three dimensions during market crises and normal times as well, but the effect on medium and large during these periods is less pronounced. Overall, our results suggest that capital is important for small at all times and is important for medium and large primarily during banking crises. Contact details: Moore School of Business, University of South Carolina, 1705 College Street, Columbia, SC Tel: Fax: Contact details: Weatherhead School of Management, Case Western Reserve University, Euclid Avenue, 362 PBL, Cleveland, OH Tel.: Fax: Keywords: Financial Crises, Survival, Market Share, Profitability, Liquidity Creation, and Banking. JEL Classification: G01, G28, and G21. This is a significantly expanded version of the second part of an earlier paper, Financial Crises and Bank Liquidity Creation. A previous version of this paper was entitled: Bank Capital, Survival and Performance Around Financial Crises. The authors thank Nittai Bergman, Lamont Black, Rebel Cole, Bob DeYoung, Mark Flannery, Paolo Fulghieri, Steven Ongena, Bruno Parigi, Peter Ritchken, Katherine Samolyk, Asani Sarkar, James Thomson, Greg Udell, Todd Vermilyae, and participants at presentations at the American Finance Association Meeting, the Bank for International Settlements, the University of Venice, the European School of Management and Technology in Berlin, the CREI / JFI / CEPR Conference on Financial Crises at Pompeu Fabra, the Boston Federal Reserve, the Philadelphia Federal Reserve, the San Francisco Federal Reserve, the Cleveland Federal Reserve, the International Monetary Fund, the Summer Research Conference in Finance at the ISB in Hyderabad, the Unicredit Conference on Banking and Finance, the University of Kansas Southwind Finance Conference, Erasmus University, and Tilburg University for useful comments.

2 1. Introduction The recent financial crisis has raised fundamental issues about the role of bank equity capital. Various proposals have been put forth which argue that should hold more capital (e.g., Kashyap, Rajan, and Stein 2009, Hart and Zingales 2009, Acharya, Mehran, and Thakor 2010, Basel III (2010)). 1 An underlying premise in all of these proposals is that there are externalities due to the safety net provided to and thus social efficiency can be improved by requiring to operate with more capital, especially during financial crises. Bankers, however, have typically argued that being forced to hold more capital would jeopardize their performance, especially profitability, and the argument that higher capital need not be beneficial has found some support in the academic literature as well (e.g., Calomiris and Kahn 1991). The issue of what effects capital has on bank performance, and how these effects might differ between crises and normal times, thus boils down to an empirical question, and one that we confront in this paper. In particular, the goal of this paper is to empirically examine the effects of bank capital on three dimensions of bank performance probability of survival, market share, and profitability during different types of financial crises and normal times. Theories predict that the effect of bank capital on any of these three dimensions of performance could be positive or negative. As a prelude to a more extensive discussion of this in the next section, here we briefly present the main arguments. Consider survival probability first. Holding fixed the bank s asset and liability portfolios, higher capital mechanically implies a higher survival probability (also suggested by monitoringbased theories like Holmstrom and Tirole 1997), but theories on the disciplining role of demandable debt suggest that a bank with more demandable debt and less capital may make better loans that are less likely to default and cause bank failure (e.g., Calomiris and Kahn 1991). Next, consider the issue of market share. While recent banking theories suggest a positive relationship between capital and market share (e.g., Allen, Carletti, and Marquez forthcoming, Mehran and Thakor forthcoming), the literature on the interaction between a nonfinancial firm s leverage and its product-market dynamics argues that the relationship is negative (e.g., Brander and Lewis 1986). Finally, consider profitability. If capital improves monitoring incentives as in Holmstrom and Tirole (1997), higher capital could enhance return on equity, whereas the literature on the disciplining role of debt suggests the opposite (Calomiris and Kahn 1991). 1 Hoshi and Kashyap (2010) draw lessons for the U.S. from Japan s efforts to recapitalize its banking sector. 1

3 This discussion suggests that although a substantial body of theoretical research is available, these theories focus on different forces and hence produce conflicting implications, pointing strongly to the need for empirical mediation. Understanding the effects of capital is interesting in its own right, but is particularly compelling during times of stress, such as financial crises. While the papers discussed above focus on the role capital plays during normal times, we examine both crises and normal times. We therefore need to extrapolate these theoretical results to the crisis context in a manner which is still consistent with the intuition of the theories, which we do in the next section. For each of the three performance dimensions, we take our cue from the theories and formulate hypotheses that allow us to assess whether capital helps or hurts bank performance. These hypotheses are then tested using data on virtually every U.S. bank from 1984:Q1 until 2009:Q4. We test our survival hypotheses using logit regressions. We regress the log odds ratio of the probability of survival on the bank s pre-crisis capital ratio interacted with a banking crisis dummy, a market crisis dummy, and a normal times dummy, plus a set of control variables. As discussed below, banking crises are those that originated in the banking sector, and market crises are those that originated outside banking in the financial markets. The interaction terms capture the effect of capital on survival during banking crises, market crises, and normal times, respectively. Moreover, we examine small (gross total assets or GTA up to $1 billion), medium (GTA exceeding $1 billion and up to $3 billion), and large (GTA exceeding $3 billion) as three separate groups, since the effect of capital likely differs by bank size (e.g., Berger and Bouwman 2009). 2 Our results support the hypothesis that capital enhances the survival probability for of all sizes during banking crises and, in the case of small, also during market crises and normal times. We test our market share hypotheses by defining market share in terms of the bank s share of aggregate bank liquidity creation. We prefer this measure over the bank s market share of assets because liquidity creation is a better representation of bank output than assets alone, since liquidity creation takes into account all on- and off-balance sheet activities. 3 We regress the percentage change in market share during the crisis on the bank s average pre-crisis capital ratio (interacted with the banking crisis, market crisis, and 2 Gross Total Assets or GTA equals total assets plus the allowance for loan and lease losses and the allocated transfer risk reserve (a reserve for certain foreign loans). Total assets on Call Reports deduct these two reserves, which are held to cover potential credit losses. We add these reserves back to measure the full value of the loans financed. Section 5.4 shows results based on alternative cutoffs between medium and large. 3 Robustness checks yield comparable results based on gross total assets market share (see Section 5.1). 2

4 normal times dummies mentioned above) and a set of control variables. Our results support the hypothesis that capital helps to increase market shares for of all sizes during banking crises and normal times, and for small also during market crises. To test our profitability hypotheses, we run regressions that are similar to the market share regressions except that we use the change in return on equity (ROE) during a crisis as the dependent variable. ROE is an appropriate profitability measure since both net income (the numerator) and equity (the denominator) reflect all of the bank s on- and off-balance sheet activities. 4 Our results support the hypothesis that capital improves profitability for small at all times, for medium during market crises, and for large during banking and market crises. We perform a variety of robustness checks. First, to check the sensitivity of the results to our definitions of various performance measures, we use alternative definitions of survival, market share, and profitability. Second, we use regulatory capital ratios instead of the equity-to-assets ratio to define capital. Third, we drop that are Too-Big-To-Fail from the large-bank sample to see if our large-bank conclusions are driven by the dominance of a few very large. Fourth, we use alternative cutoffs to separate medium and large to examine the sensitivity of our results to the manner in which are classified by size. Fifth, we measure pre-crisis capital ratios averaged over the four quarters before the crisis or one quarter before the crisis starts rather than averaging them over the eight quarters before the crisis. The purpose is to determine if the time period over which pre-crisis capital is defined affects our results. Sixth, while the theories suggest a causal relationship from capital to performance, we recognize that in practice both may be jointly determined. Although our main regression analyses use lagged capital to mitigate this potential endogeneity problem, we also address it more directly using an instrumental variable approach. While most of our analyses focus on the effect of book capital on the performance of individual, we also perform analyses for listed entities (listed and listed bank holding companies). Specifically, we examine how book and market capital ratios affect the performance of these organizations during crises and normal times. We can broadly summarize our findings as follows. First, capital is especially beneficial during 4 A robustness check based on return on assets (ROA), which divides net income by total assets, a measure of the bank s on-balance sheet activities, shows similar results for small and weaker results for medium and large (see Section 5.1). 3

5 banking crises. It helps of all sizes during such crises higher pre-crisis capital increases the odds of survival and enhances market share for of all sizes and it increases profitability for all but medium-sized during such crises. Second, small benefit in all respects from higher capital during market crises and normal times as well. For large and medium, higher capital improves only profitability during market crises and only market share during normal times. While the survival and market share results are highly robust, the profitability results are less so. The remainder of this paper is organized as follows. Section 2 develops the empirical hypotheses. Section 3 explains our empirical approach, describes the financial crises and normal times, describes all the variables and the sample, and provides summary statistics. Section 4 discusses the results of our empirical tests. Section 5 includes the robustness tests. Section 6 contains the additional analysis of listed entities. Section 7 concludes. 2. Development of the empirical hypotheses In this section, we review existing theories to formulate our empirical hypotheses about the effects of bank capital on the survival probability, market share, and profitability of during crises and normal times Survival Hypothesis 1a: Capital enhances the bank s survival probability during crises and normal times. Hypothesis 1b: Capital diminishes the bank s survival probability during crises and normal times. There are several reasons, grounded in existing theories, to believe that capital improves a bank s survival probability. First, there is a set of theories that emphasizes the role of capital as a buffer to absorb shocks to earnings (e.g., Repullo 2004, Von Thadden 2004). If the bank s portfolio, screening, and monitoring choices are held fixed, then this buffer role immediately implies that higher capital increases the probability of the bank s survival. 5 We can view this as the mechanical effect of higher capital. Second, there is another set of theories that focuses on the incentive effects of capital. This set includes theories based on monitoring, screening, and asset-substitution-moral-hazard. The monitoring-based papers include Holmstrom and Tirole (1997), Allen, Carletti, and Marquez (forthcoming), and Mehran and Thakor (forthcoming). A key result in 5 We know from various theories, however, that all of these choices are influenced by the bank s capital structure. 4

6 these papers is that higher bank capital induces higher levels of borrower monitoring by the bank, thereby reducing the probability of default or otherwise improving the bank s survival odds indirectly by increasing the surplus generated by the bank-borrower relationship. 6 In their screening-based theory of banking with behaviorally-biased agents, Coval and Thakor (2005) show that a minimum amount of capital may be essential to the very viability of the bank. The asset-substitution-moral-hazard theories argue that government guarantees cause shareholders to prefer low capital and excessive risk to increase the value of the deposit insurance put option (Merton 1977), and that with more capital will optimally choose less risky portfolios (see the overview of this literature in Freixas and Rochet 2008). A different strand of the theoretical literature suggests that with higher capital may experience lower survival odds. Calomiris and Kahn (1991) show that a capital structure with sufficiently high demand deposits (and by implication lower equity) leads to more effective monitoring of bank managers by informed depositors and hence a smaller likelihood of bad investment decisions. This suggests that a bank with higher capital (and consequently lower deposits) may face a higher probability of bad loans and hence loan default, which may result in a lower survival probability. This paper has spawned a sizeable literature on the market discipline role of bank leverage (see Freixas and Rochet (2008) for an overview). Thus, some theories predict that higher bank capital should lead to a higher survival probability for the bank, whereas others suggests that higher capital may worsen the portfolio choices and liquidity of and hence lead to a lower survival likelihood. 8 These papers do not focus on financial crises per se. However, since a bank s likelihood of survival is lower during a crisis, it follows that the effect of capital on survival may be more positive during a crisis, particularly in light of regulatory discretion in closing, forcing them into assisted mergers, or otherwise 7 6 In Mehran and Thakor (forthcoming), the mechanical effect of capital also shows up higher-capital are more likely to survive not just because they monitor more, but also because they have a greater capital cushion to absorb losses and avoid interim closure. 7 For example, Diamond and Rajan (2001) argue that replacing demandable deposits with capital weakens the bank s incentive to collect repayments from borrowers, thereby reducing loan liquidity. In fact, in an all-equity bank, with no demand deposits to discipline the bank the loan would be completely illiquid because the bank cannot credibly precommit to collect repayment on behalf of others, so the loan could not be sold to another bank, and the bank could additionally credibly threaten to withhold its repayment collection technology. 8 The empirical literature on bank failure finds that capital reduces the probability of failure (e.g., Wheelock and Wilson 2000), but this literature does not focus on crises. 5

7 resolving problem institutions based on their capital ratios Market share Hypothesis 2a: Capital enhances the bank s market share during crises and normal times. Hypothesis 2b: Capital diminishes the bank s market share during crises and normal times. The theories on the effect of capital on market share also produce opposing predictions. In the banking models of Holmstrom and Tirole (1997), Allen and Gale (2004), Boot and Marinc (2008), Allen, Carletti, and Marquez (forthcoming), and Mehran and Thakor (forthcoming), derive a competitive advantage from higher capital. These papers imply that higher-capital will end up with higher market shares. In contrast, there is also a literature that focuses on the relationship between leverage and market share for nonfinancial firms (e.g., Brander and Lewis 1986, Lyandres 2006). This literature shows that more highlylevered firms will be more aggressive in their product-market-expansion strategies and hence suggests that capital and market share will be negatively correlated. The literature discussed above does not focus on crises, so the predictions of these papers should be viewed as applying during normal times. However, it is plausible to argue that the competitive advantage of capital is more pronounced during crises, particularly during banking crises. There are several reasons for this. First, the bank s customers may be more cognizant of the bank s capital during a crisis, making it easier for better-capitalized to take customers away from lesser-capitalized peers. Second, with more capital may have greater flexibility to make certain types of loans, offer credit lines and otherwise create liquidity in various ways that may be unavailable to lower-capital that may feel particularly constrained, possibly by regulators, during crises. Third, banking crises are generally associated with numerous bank failures and near failures. Failing and near-failing tend to be bought by competitors and such acquisitions have to be approved by bank regulators. Since regulatory approval depends in part on the acquiring bank s capital, with higher capital ratios are better positioned to buy their troubled brethren, and hence improve their market share. 9 The benefits of higher capital may be weaker if regulators delay closure of troubled because of regulatory career concerns (Boot and Thakor 1993) or because regulators perceive a high value associated with avoiding bankruptcy as, for example, happened during the recent subprime lending crisis (Veronesi and Zingales 2010). 6

8 2.3. Profitability Hypothesis 3a: Capital enhances the profitability of the bank during crises and normal times. Hypothesis 3b: Capital diminishes the profitability of the bank during crises and normal times. The theoretical literature also offers conflicting predictions about how capital should affect bank profitability. One strand of theory predicts that higher capital enhances profitability. As pointed out above, Holmstrom and Tirole (1997), Allen, Carletti, and Marquez (forthcoming), and Mehran and Thakor (forthcoming) show that high bank capital increases the total surplus generated in the bank-borrower relationship. Assuming that keep a large enough portion of the surplus, higher capital will lead to higher bank profitability. Moreover, if the ratio of the surplus generated by high- versus low-capital is higher during crises, it follows that highcapital will be able to improve their profitability during crises relative to low-capital. 10 In contrast, another strand of theory predicts that higher capital should lead to reduced profitability. The most obvious argument here goes back to Modigliani and Miller (1963), who show that higher capital mechanically leads to lower ROE. Moreover, the literature on the disciplining role of debt (e.g., Calomiris and Kahn 1991) suggests that higher leverage improves the bank s asset choice and hence its profitability. Thus, with higher capital have lower-quality assets. If these assets deteriorate in value more during a crisis than do higher-quality assets, then with higher capital may suffer bigger declines in profitability during crises than their lower-capital counterparts Empirical implications of the theories It is clear from the discussions above that the theories have opposing views on the effect of capital on performance. Our empirical analyses attempt to assess the impact of capital on performance to determine which theories have the strongest empirical support. Not finding support for a particular theory does not constitute a rejection of that theory, but simply implies that the theory for which we do find support is more 10 A recent survey paper by Campello, Giambona, Graham, and Harvey (2009) sheds further light on why with high capital may improve their profitability during crises relative to with low capital. It shows that during the subprime lending crisis, renegotiated in their own favor the terms for lines of credit with borrowers, possibly by threatening to invoke the material-adverse-change clause. It may be that with more capital could do this more easily because their stronger reputation in the loan commitment market gives them greater bargaining power with their borrowers (see Boot, Greenbaum, and Thakor 1993), which would result in increased profitability for high-capital (relative to lowcapital ) during crises. Berger (1995) finds that the relationship between capital and earnings can be positive or negative, but does not differentiate between financial crises and normal times. 7

9 dominant empirically. This is because what we may be picking up in the data is the net effect of opposing forces identified by different theories. 3. Methodology This section first explains our empirical approach and describes the financial crises and normal times. It then explains the performance measures. Next, it discusses the key exogenous variables and the control variables. Finally, it describes the sample and provides summary statistics. In our empirical approach, we examine the effect of capital (and other bank conditions) measured prior to a crisis on bank performance during a crisis. We measure capital before a crisis for two reasons. First, since it is not known a priori when a crisis will strike, the interesting question is whether that have higher capital going into a crisis benefit from these higher capital ratios during a crisis. That is, we want to know whether higher pre-crisis capital ratios result in better performance during a crisis. Second, it mitigates endogeneity concerns because lagged capital and current performance are less likely to be jointly determined Empirical approach and description of financial crises and normal times Our analyses focus on crises that occurred between 1984:Q1 and 2009:Q4. They include two banking crises (crises that originated in the banking sector) and three market crises (crises that originated outside banking in the financial markets). The banking crises are the credit crunch of the early 1990s (1990:Q1 1992:Q4) and the recent subprime lending crisis (2007:Q3 2009:Q4). The market crises are the 1987 stock market crash (1987:Q4); the Russian debt crisis plus Long-Term Capital Management (LTCM) bailout of 1998 (1998:Q3 1998:Q4); and the bursting of the dot.com bubble and the September 11 terrorist attacks of the early 2000s (2000:Q2 2002:Q3). Appendix I describes these crises in detail. Our hypotheses focus on the effect of a bank s (pre-crisis) capital on its performance (survival, market share, and profitability) during a crisis. A key issue is how to measure pre-crisis capital. In all of our main analyses, we average each bank s capital ratio over the eight quarters before the crisis to reduce the impact of outliers. In robustness checks, we alternatively define the pre-crisis period as the four quarters before a crisis or the quarter before a crisis (see Section 5.5). Our survival analyses then link this average pre-crisis capital to whether a bank survived a crisis (it was in the sample one quarter before the crisis and is still in the sample one 8

10 quarter after the crisis; see Section 3.2). Our market share analyses link pre-crisis capital to the bank s percentage change in market share (see Section 3.3), defined as the bank s average market share during a crisis minus its average share over the eight quarters before the crisis, normalized by its average pre-crisis market share and multiplied by 100. Similarly, our profitability analyses link pre-crisis capital to the bank s change in profitability (see Section 3.4), defined as the bank s average ROE during a crisis minus its ROE over the eight quarters before the crisis. We expect the effect of capital to be more positive during financial crises than during normal times. While we highlight above how we examine the effect of (average) pre-crisis capital on bank performance during a crisis, we still have to address how we measure normal times. A naïve approach would be to simply view all non-crisis quarters as such. However, if so, it is not clear then how to examine the effect of capital on bank performance during normal times. To ensure that we analyze actual crises and normal times in a comparable way, we create fake crises to represent normal times. In essence, these fake crises act as a control group. To construct these fake crises, we use the two longest time periods between actual financial crises over our entire sample period. These periods are between the credit crunch and the Russian debt crisis, and between the bursting of the dot.com bubble and the subprime lending crisis. In each case, we take the entire period between the crises, designate the first eight quarters as pre-crisis and the last eight quarters as post-crisis and the remaining quarters in the middle as the fake crisis. This treatment of the first eight quarters as pre-crisis is consistent with our analysis of the banking and market crises. We thus end up with a six-quarter fake crisis period between the credit crunch and the Russian debt crisis (from 1995:Q1 to 1996:Q2) and a three-quarter fake crisis period between the dot.com bubble and the subprime lending crisis (from 2004:Q4 to 2005:Q2). 11 To analyze how capital affects ability to survive crisis and normal times, we pool the data of the banking crises, the market crises, and the normal times. 12 Since we have two banking crises, three market crises, and two normal time periods (the fake crises), we have up to seven observations per bank, i.e.,. Using these data, we run the following logit regressions: 11 Results are qualitatively similar if we instead use five- or four-quarter crisis periods for the first fake crisis, and twoor one-quarter crisis periods for the second fake crisis. 12 Our regressions group the two banking crises together and the three market crises together. This approach allows us to contrast the effect of capital during banking and market crises. To check the robustness of our findings, we also run our main regressions using individual crisis dummies instead, and obtain broadly consistent results (not shown for brevity). 9

11 (1) where measures whether bank i survived crisis or normal time period t (see Section 3.2). is the bank s average capital ratio over the eight quarters before crisis or normal time period t (see Section 3.5).,, and are dummy variables that equal 1 if t is a banking crisis, market crisis, or normal time period, respectively, and 0 otherwise. is a set of control variables measured over the pre-crisis period (see Section 3.6). To examine the impact of capital on a bank s market share and profitability during banking crises, market crises, and normal times, we use the following regression specifications: (2) (3) where is the percentage change in bank i s aggregate market share (see Section 3.3) and is the change in bank i s profitability (see Section 3.4). To mitigate the influence of outliers, both variables are winsorized at the 3% level. 13 variables over the pre-crisis period (see Section 3.6). is as defined above. is a set of control Since each bank enters up to seven times in these regressions, all regressions are estimated with robust standard errors, clustered by bank, to control for heteroskedasticity as well as possible correlation between observations of the same bank in different years. The regressions also include individual crisis and normal times dummies which act as time fixed effects. The literature documents differences by bank size in terms of portfolio composition (e.g., Kashyap, Rajan, and Stein 2002, Berger, Miller, Petersen, Rajan, and Stein 2005) and the effect of capital on liquidity creation which we use to construct our main market share variable (Berger and Bouwman 2009). We therefore 13 Unwinsorized data yielded changes in market share (profitability) that were roughly 54,600 (6,600) times the average change. Winsorization at the 1% level reduced this to 6.3 (23.6) times the average change, which still seemed too high for the change in profitability. Winsorization at the 3% level brought these numbers down to 2.0 and 7.2, respectively. Winsorization at the 1% level yields market share and profitability results that are very similar to the ones presented here, except that the market share results for medium and the profitability results for small are both weaker during banking crises. 10

12 split the sample into small (gross total assets (GTA) up to $1 billion), medium (GTA exceeding $1 billion and up to $3 billion), and large (GTA exceeding $3 billion) and run all regressions separately for these three sets of. All dollar values are in 2009:Q4 terms. Our definition of small conforms to the usual notion of community. The $3 billion cutoff for GTA divides the remaining observations roughly in half Definition of survival To measure whether a bank survived a crisis, we require that the bank was not acquired and did not fail during the crisis. Specifically, we use SURV, a dummy that equals 1 if the bank is in the sample one quarter before such a crisis started and is still in the sample one quarter after the crisis, and 0 otherwise. 15, Definition of market share While most of the literature focuses on the role of capital in traditional that engage only in onbalance-sheet activities, several papers highlight the importance of off-balance sheet activities (Boot, Greenbaum, and Thakor 1993, Holmstrom and Tirole 1997, Kashyap, Rajan, and Stein 2002). We therefore measure a bank s competitive position as the bank s market share of overall bank liquidity creation. Liquidity creation is a superior measure of bank output since it is the only measure based on all the bank s on- and offbalance sheet activities. 17 We calculate the dollar amount of liquidity created by each bank using Berger and Bouwman s (2009) preferred liquidity creation measure. The three-step procedure used to construct this measure is explained in Appendix II. A bank s liquidity creation market share is the dollar amount of liquidity creation by the bank divided by the dollar amount of liquidity created by the industry. 18 To establish whether improve their competitive positions during banking crises, market crises, 14 Berger and Bouwman (2009) also use these cutoffs. As shown in Section 5.4, broadly similar results are obtained when $5 billion and $10 billion cutoffs between medium and large are used instead. 15 Banks that were merged within a bank holding company are not classified as non-survivors since it unclear whether these consolidations occur because the bank is troubled or not. 16 Robustness checks using a slightly longer time window yield similar results (see Section 5.1). 17 As a robustness check, we rerun our main regressions using the market share of gross total assets, and obtain similar results (see Section 5.1). 18 Market shares are not merger adjusted because acquisitions are a key way for to increase their market shares, particularly during banking crises. In our analysis, a bank s market share rises if it acquires another bank. Mergeradjusting market shares would take out this effect. 11

13 and normal times, we define each bank s percentage change in liquidity creation market share, %ΔLCSHARE, as the bank s average market share during a crisis minus its average market share over the eight quarters before the crisis, normalized by its average pre-crisis market share and multiplied by Definition of profitability We measure a bank s profitability using the bank s return on equity (ROE), i.e. net income divided by stockholders equity. 19 This is a comprehensive profitability measure, since may have substantial offbalance sheet portfolios. Banks must allocate capital against every off-balance sheet activity in which they engage. Hence, net income and equity both reflect the bank s on- and off-balance sheet activities. 20 To examine whether a bank improves its profitability during banking crises, market crises, and normal times, we focus on the change in ROE (ΔROE), defined as the bank s average profitability during these crises minus the bank s average ROE over the eight quarters before these crises Key exogenous variables The key exogenous variables are three bank capital ratio-crisis interaction terms: EQRAT * BNKCRIS, EQRAT * MKTCRIS, and EQRAT * NORMALTIME. EQRAT is the ratio of equity capital to gross total assets, averaged over the eight quarters before the crisis. 21 BNKCRIS, MKTCRIS, and NORMALTIME are as defined in Section Control variables The survival regressions contain X, a set of control variables which include: bank credit risk, bank size, bank holding company membership, local market power, and profitability. The market share and profitability regressions contain a slightly smaller set of controls Z, which excludes profitability from X. We discuss these variables in turn. Each control variable is averaged over the eight-quarter pre-crisis period, except when noted 19 If a bank s capital to GTA ratio is less than one percent, we calculate ROE as net income divided by one percent of GTA. For observations for which equity is between 0% and 1% of GTA, dividing by equity would result in extraordinarily high values. For observations for which equity is negative, the conventionally-defined ROE would not make economic sense. We considered the alternative of dropping negative-equity observations, but rejected it because these are the that are most likely to be informative of ability to survive crises. 20 Section 5.1 includes a robustness check in which we use return on assets (ROA) instead. 21 Similar results are obtained in robustness checks which use a bank s regulatory capital ratio (see Section 5.2) or measure capital at alternative points before the crisis starts (see Section 5.5). 12

14 otherwise. Credit risk, defined as the bank s Basel I risk-weighted assets divided by gross total assets, is used as a measure of bank risk taking (e.g., Logan 2001). Risk-weighted assets, a weighted average of the bank s assets and off-balance-sheet activities designed to measure credit risk, is the denominator in the Basel I risk-based capital requirements. Since these requirements only became effective in December 1990 and, hence, were only reported in Call Reports from that moment onward, we use a Federal Reserve Board program to construct risk-weighted assets from the beginning of our sample period. Banks with riskier portfolios (i.e. higher riskweighted assets relative to gross total assets) may be less likely to survive crises. They may also find it harder to improve their market shares and profitability during crises. We therefore interact the credit risk variable with the three crisis dummies, and expect the coefficients on the banking and market crisis interaction terms to be negative in all regressions. Bank size is controlled for by including lnlc, the log of liquidity creation, in all regressions. 22 In addition, we run regressions separately for small, medium, and large. Bank size is expected to have a positive effect on the probability of survival, since it is well-known that larger have higher survival odds than smaller. In contrast, the coefficient on bank size is expected to be negative and significant for all size classes in the market share regressions, since the law of diminishing marginal returns suggests that it will be more difficult for bigger (that already have larger market shares) to improve their market shares. It is unclear whether bank size will have a significant effect on ability to improve their profitability (measured as a bank s ROE) during crises. To control for bank holding company status, we include D-BHC, a dummy variable that equals 1 if the bank was part of a bank holding company (BHC) at any time in the eight quarters preceding the crisis. BHC membership is expected to help a bank survive and strengthen its competitive position because the holding company is required to act as a source of strength to all the it owns, and may also inject equity voluntarily when needed. In addition, other in the holding company provide cross-guarantees. Houston, James, and Marcus (1997) find that bank loan growth depends on BHC membership. Thus, the coefficient on bank holding company status is expected to be positive and significant in the survival and market-share regressions. BHC membership is also expected to reduce the cost of financing and enhance profitability 22 Robustness checks which instead include lngta, the log of gross total assets, yield similar results. These are not shown for brevity. 13

15 because it provides greater access to internal capital markets and provides potential protection against the vagaries of market financing. We control for local market power by including HHI, the bank-level Herfindahl-Hirschman index of deposit concentration for the local markets in which the bank is present. 23 From , we define the local market as the Metropolitan Statistical Area (MSA) or non-msa county in which the offices are located. 24 After 2004, we use the new local market definitions based on Core Based Statistical Area (CBSA) and non- CBSA county. 25 The larger is HHI, the greater is a bank s market power. Since more market power should help a bank survive, the coefficient on HHI is expected to be positive in the survival regressions for of all sizes. More market power likely makes it harder to improve market share since regulatory approval for acquisitions will be more difficult to obtain. The coefficient on HHI is therefore likely to be negative in the market share regressions. Since higher market power increases the profitability of local loans and deposits, more market power should make it easier to improve profitability. Thus, the coefficient on HHI in the profitability regressions is expected to be positive. The survival regressions also include a measure of profitability because that are more profitable before the crisis may be more likely to survive crises. We use a bank s return on assets, ROA, for this purpose. The reason to use ROA instead of ROE is that we want to measure the effect of capital on ability to survive crises and avoid using control variables that include capital Sample and summary statistics For every bank in the U.S., we obtain quarterly Call Report data from 1984:Q1 to 2009:Q4. We keep a bank in the sample if it: 1) has commercial real estate or commercial and industrial loans outstanding; 2) has deposits; and 3) has gross total assets or GTA exceeding $25 million. As indicated above, we split these into three size categories. Analyses that focus on the effect 23 While our focus is on the change in competitive positions measured in terms of their aggregate liquidity creation market shares, we control for local market competition measured as the bank-level Herfindahl-Hirschman index based on local market deposit shares. This is a standard measure of competition used in antitrust analysis in the U.S. Deposits are used for this purpose because deposits are the only bank output variable for which location is known. 24 When appropriate, we use New England County Metropolitan Areas (NECMAs) instead of MSAs, but refer to these as MSAs. 25 The term CBSA collectively refers to Metropolitan Statistical Areas and newly-created Micropolitan Statistical Areas. Areas based on these new standards were announced in June For recent years, the Summary of Deposits data needed to construct HHI is available on the FDIC s website only based on the new definition. It is not possible to use the new definition for our entire sample period. 14

16 of capital on survival have 46,107 small-bank, 1,599 medium-bank, and 1,194 large-bank observations. Analyses that focus on the effect of capital on market share and profitability have 52,107 small-bank, 1,911 medium-bank, and 1,382 large-bank observations. 26 Table 1 contains summary statistics on the regression variables. The sample statistics are shown for banking crises, market crises, and normal times. All financial values are put into real 2009:Q4 dollars (using the implicit GDP price deflator) before size classes are constructed. 4. Main regression results In this section, we discuss the main empirical results Does capital affect the bank s ability to survive during crises and normal times? Table 2 Panel A presents the survival findings for small, medium, and large. Two main results are apparent. First, higher capital helps in all size classes to improve the probability of surviving banking crises. Second, higher capital also helps small improve their odds of survival during market crises and normal times. These results generally support the hypothesis that capital helps survive. These findings have sensible economic interpretations. Capital is the main line of defense against negative shocks for small, since they have limited (and relatively costly) access to the financial market in the event of unanticipated needs. Hence, higher capital enhances the probability of survival for such at all times. This finding is consistent with the earlier-discussed theoretical papers that predict that higher capital increases a bank s survival probability. Medium and large can rely on financial market access, and correspondent and other interbank relationships as risk-mitigation sources in addition to their on-balancesheet capital to survive negative shocks. So, capital is less pivotal for survival during normal times for these than it is for small. However, banking crises create stresses for all, and financial market access and interbank relationships may offer inadequate protection against negative shocks for all but the very largest. Hence, capital may be important for survival for medium and large during banking crises. The market crisis results for medium and large suggest that such crises do not pose much of a survival threat for these. Given the access that medium and large have to the interbank lending market 26 The survival regressions have fewer observations because we do not have data on the first quarter after the end of the subprime lending crisis and because we drop within-bank-holding-company consolidations from these regressions. 15

17 (e.g., federal funds) that itself is unlikely to be adversely impacted by a market crisis, capital may not be critical for these to survive market crises. To judge the economic significance of our findings, Table 2 Panels B, C, and D show the predicted probabilities of non-survival. The top part of each panel shows the average capital ratio and the average capital ratio plus or minus one standard deviation of small, medium, and large over the eight quarters before banking crises, market crises, and normal times. The bottom part of each panel shows the predicted probability of not surviving banking crises, market crises, and normal times at these capital ratios. A small, medium, or large bank with an average capital ratio (10.00%, 8.76%, and 8.30%, respectively) had a probability of not surviving banking crises of 0.61%, 0.55%, and 0.12%, respectively. Reducing capital by one standard deviation more than doubles these probabilities of non-survival. Similarly, increasing capital by one standard deviation reduces the probabilities of non-survival by more than one half for all size classes. The corresponding probabilities for market crises and normal times are generally higher. 27 Turning to the control variables, we find that the coefficients generally have the predicted signs. As expected, being part of a bank holding company, greater market power, and higher profitability helps survive. Banks that held more credit risk before banking and market crises are less likely to survive (significant for small ; for medium only before banking crises). Bank size has no significant effect on medium and large ability to survive. Among small, larger are less likely to survive, which is surprising Does capital affect the market shares of during crises and normal times? Table 3 summarizes the results from regressing the percentage change in a bank s market share (%ΔLCSHARE) during crises on the bank s pre-crisis capital ratio plus control variables. As before, results are shown separately for small, medium, and large. The t-statistics are based on robust standard errors clustered by bank. We find two main results. First, higher capital helps of all sizes improve their market shares during banking crises and normal times. Second, higher capital also helps small to improve their market 27 The survival probabilities of medium actually decrease slightly when capital is higher during market crises and normal times. This is consistent with the negative but insignificant coefficients on EQRAT for these shown in Table 2 Panel A. 16

18 shares during market crises. 28 Capital does not appear to produce such benefits during market crises for medium and large. Somewhat surprisingly, medium with higher capital ratios seem to lose market share during market crises. These results generally support the hypothesis that capital helps to improve their market shares. We can again interpret these results in the context of the theories discussed earlier. Note that capital is essential to survival for small, as discussed earlier. Moreover, these engage largely in relationship lending, and long-lasting bank-borrower relationships are crucial for relationship banking to create value. 29 This means that relationship borrowers will tend to gravitate toward high-capital, since higher capital leads to a higher survival probability for small at all times (see Section 4.1). 30 Thus, it is not surprising that higher capital benefits small in terms of gaining market share at all times. During banking crises and normal times, capital also helps improve the market shares of medium and large. In contrast, during market crises, capital seems to hurt the market shares of medium (no obvious interpretation) and does not significantly affect large, possibly because it is relatively easy for these to turn to the discount window and the interbank market both of which may not experience stress during market crises to ensure unruptured relationships with their borrowers during such crises. This means that while capital may offer large a benefit during market crises, this benefit may be no greater than that before such crises. To judge the economic significance of these results, focus first on the effect of capital during banking crises. The coefficients on the EQRAT * BNKCRIS interaction terms imply that a one standard deviation increase in EQRAT would lead to a 0.373, 0.114, and standard deviation change in for small, medium, and large, respectively, during such crises (not shown for brevity). The corresponding figures for EQRAT * MKTCRIS and EQRAT * NORMALTIME are (0.310, , and ) and (0.231, 0.185, and 0.250), respectively. These results seem economically significant for of all sizes during banking crises and for small also during market crises and normal times. 28 The positive effect of capital on large market shares during normal times disappears when we use a $5 billion or $10 billion cutoff between medium and large, suggesting that our main result is driven by smaller large (see Section 5.4). 29 See, e.g., Ongena and Smith (2001) Bae, Kang, and Lim (2002), and Bharath, Dahiya, Saunders, and Srinivasan (2007) for empirical evidence. 30 This argument is consistent with the intuition in Song and Thakor (2007) who show that that make relationship loans will prefer to finance with stable funds because this increases the likelihood that a relationship loan will not need to be terminated prematurely. 17