Loan diversification, market concentration and bank stability

Loan diversification, market concentration and bank stability January 11, 2018 Jeungbo Shim Assistant Professor Finance and Risk Management University of Colorado-Denver 1475 Lawrence Street Denver, CO 80202 Tel: 303-586-1224 Fax:303-315-8155 Jeungbo.shim@ucdenver.edu 1

Loan diversification, market concentration and bank stability ABSTRACT This paper examines how the choice of loan diversification and exogenous market conditions are associated with a bank s financial stability using the data of U.S. commercial banks. We employ two-stage least squares estimation with instrumental variables to address potential endogeneity concerns. We find that increased loan diversification has a positive impact on the bank s financial strength. We show that market concentration is positively associated with a bank s financial stability, consistent with the concentration-stability view. The results are robust to estimating Heckman s self-selection model, accounting for mergers and acquisitions, or examining subsamples. Keywords: Loan diversification; Market concentration; Banking; Financial stability JEL classification: G21; L22; L25; G33 1. Introduction The number of bank closures has sharply risen during a period of recent economic downturn in the United States. The Federal Deposit Insurance Corporation (FDIC) reports that 465 insured U.S. commercial banks failed between January 2008 and December 2012, while only 27 banks closed from October 2000 to December 2007. These developments have severely strained the resources of the FDIC. 1 The burden could eventually fall on the general taxpayers in the form of higher taxes as noted by Kane (1985) and exemplified by the recent bailouts for the demise of large financial institutions. The high failure of commercial banks is a concern for banking supervisors because bank safety and soundness is a major regulatory responsibility. A number of studies have investigated the factors influencing bank failures. Earlier studies rely on the standard proxies for the CAMEL ratings 2 and show that these CAMEL-type variables 1 From 2008 through 2011, bank failures cost the deposit insurance fund an estimated $88 billion, and with failures accelerating, the fund s balance turned negative in 2009. 2 The acronym "CAMELS" refers to the six components of a bank's financial condition that are assessed: Capital adequacy, Asset quality, Management, Earnings, Liquidity, and Sensitivity to market risk. The Federal Reserve, FDIC, 2

are useful to explain the likelihood of bank failures (e.g., Sinkey, 1975; Cole and Gunther, 1995). Reinhart and Rogoff (2011) argue that external debt surges are a recurrent antecedent to banking crises. DeYoung (2003) finds that aggressive lending, expensive deposit funding and cost inefficiencies are significant predictors of bank failure. Cole and White (2012) and Government Accountability Office (2013) report that the concentration on commercial real estate loans is among the contributing factors that led to an increased likelihood of recent bank closures across all states. The latest banking failure reports and collapse of large financial institutions (such as Lehman Brothers, Washington Mutual and Bear Stearns) during the financial crisis raise the following questions: Would a recent wave of bank failures have been avoided if banks had diversified more in their loan portfolios? Do banks reduce or increase financial fragility from diversification of their loan portfolios? Are banks in highly concentrated market structures more financially stable than those operating in less concentrated market structures or vice versa? What are the important bank and economic characteristics that significantly influence a bank s financial stability? In this paper, we address these unanswered questions by undertaking an empirical investigation using the samples of U.S. commercial banks over the period 2002:Q1-2013:Q3. The corporate finance literature documenting the dark side of diversification suggests that firms should concentrate their activities on specialized area to take greatest advantage of management expertise and reduce agency problems (Jensen, 1986; Berger and Ofek, 1995; Servaes, 1996). On the other hand, previous research indicates that diversification could improve firm value by achieving economies of scale (Gertner et al., 1994) or creating internal capital markets (Houston et al., 1997). Given the conflicting views in the literature, the issue of net effects of diversification and other financial supervisory agencies employ this rating system to provide a summary of bank conditions at the time of an exam. 3

on bank performance and financial stability still draws attention to the need for additional investigation. Unlike most of the prior studies that focus on the link between income-based measure of diversification and financial performance of banks, this paper intends to provide new evidence on whether banks loan portfolio diversification influences their financial stability. The literature suggests that the nature of bank market structure has significant implications for the likelihood that a banking failure will occur. Theoretical models derive contrasting predictions on the relationship between market concentration and bank stability. 3 The concentration-stability view suggests that a highly concentrated banking system is more stable than a less concentrated bank structure because banks in less competitive environments have better profit opportunities, high capital buffers and thereby less incentives to take excessive risks (e.g., Keeley, 1990; Hellman et al., 2000; Allen and Gale, 2000, 2004). In contrast, the concentrationfragility view predicts a positive relationship between market concentration and bank instability. Larger banks in concentrated market structure are likely to receive a greater subsidy through implicit too big to fail policies when they face financial difficulty. This protective policy may create moral hazard problems and boost the bank s risk-taking incentives, leading to banking system fragility (e.g., Boyd and Runkle, 1993; Mishkin, 1999; Gan, 2004). Although theoretical literature on this topic has had some significant effect on bank regulators and policymakers, empirical evidence regarding indicators of banking market concentration and financial stability is very limited, with no clear consensus. This paper attempts to provide further evidence on whether market concentration is positively (negatively) associated with a bank s financial stability. 3 See Boyd and De Nicoló (2005) and Beck et al. (2006) for an overview of the literature examining the relationship between concentration and competition in banking markets and banking system stability. 4

We show that loan diversification is positively associated with a bank s financial stability, implying that commercial banks may be able to diminish financial fragility from the diversification of their loan portfolios. We find that market concentration is positively associated with a bank s financial stability, consistent with the concentration-stability view. We additionally show that the effect of income diversification on bank stability differs systematically depending on the diversification measurement of banking activities. We find that there is a diversification discount when income diversification is measured by two broad categories (net interest income versus noninterest income). However, our analysis measuring the income diversification based on eleven breakdowns of total operating income presents a diversification benefit, which is contrary to the findings of prior diversification-performance literature. Our study adds to the literature in several ways. First, prior empirical research on the relationship between market concentration and bank stability has used the samples of a broader set of heterogeneous countries around the globe (Beck et al., 2006; Uhde and Heimeshoff, 2009). Berger et al. (2004) assert that it should be cautious in drawing inferences from the studies involving international comparisons because there is a possibility that cross-country differences in economic conditions, institutional structure, regulatory and supervisory policies may not be perfectly controlled. We use data obtained from one country to reduce potential biases due to data constraints associated with the common employment of cross-country data. Unlike previous studies that typically use national level market concentration, we measure market concentration based on Core Based Statistical Area (CBSA) because competition occurs locally and market concentration varies significantly by CBSA. 4 This is, to our knowledge, the first study that examines particularly the link between CBSA-level market concentration, loan portfolio 4 The measure of market concentration based on CBSA is discussed in Section 3.3. 5

diversification and bank insolvency risk using more homogeneous data by exclusively focusing on the samples of U.S. commercial banks. The empirical evidence can provide important implications for regulators who are concerned about the safety and financial soundness of commercial banks and for policymakers who involve in policy decisions such as mergers & acquisitions (M & As) guidelines and market competitiveness. In addition, bank managers can obtain a better understanding of the potential opportunities and strategies to improve the bank s financial stability through the outcomes of this research. Second, the extant banking diversification-performance literature typically measures the diversification of banking activities by the breakdown of net interest income versus non-interest income or lending versus non-lending activities (e.g., Stiroh and Rumble, 2006; Laeven and Levine, 2007). Laeven and Levine (2007) measure asset diversification by accounting for variation in the breakdown of total earning assets into two broad categories: net loans and other earnings assets. Their income diversification is measured by dividing total operating income into two wide-ranging categories: net interest income and other operating income. This simple dichotomous distinction may have potential measurement error problems and fail to account for the extent to which banks actually engage in various types of loan making activities or an array of fee-generating activities. To overcome these problems, we identify the detailed composition of a bank s income sources and take into account a range of banking activities in constructing diversification measures. Because we are particularly interested in investigating the impact of loan portfolio diversification, we identify the level of loan diversification by partitioning loan scope into six primary categories. Laeven and Levine (2007) state that asset- and income-based diversification measures are complementary. In the extended analysis, we compute and analyze an income-based measure of diversification by combining seven different types of interest income and four components of non- 6

interest income. 5 The use of abundant U.S. commercial bank data allows us to precisely capture the degree to which each bank provides a broad array of financial services that generate interest and fee income, trading revenue and other types of non-interest income. Finally, we perform numerous robustness checks. It is well documented in the literature that the firm s decision to diversify is endogenous (e.g., Campa and Kedia, 2002; Villalonga, 2004a; Laeven and Levine, 2007). To mitigate potential endogeneity concerns, our preliminary regression analyses use fixed-effect models. We also address this potential endogeneity problem by employing two-stage least squares (2SLS) estimation method with instrumental variables. Additionally, we use Heckman s (1979) two-step procedure to control for likely self-selection bias induced by banks choosing to diversify following the literature (Laeven and Levine, 2007). Graham et al. (2002) argue that conglomerates yields the diversification discount because acquirers tend to purchase already discounted target firms. Our analysis accounts for the impact of M & As. Although the existing theories do not distinguish between banks of different size classes, the effect of loan diversification likely differs by bank size. Thus, we divide the sample into small banks and large banks and run regressions separately for these sets of banks. We perform regressions by dividing our sample into publicly held and privately owned banks to investigate the impact of the choice of bank ownership structure. The paper is structured as follows. Section 2 reviews the literature and formulates main hypotheses. Section 3 discusses variables used in the analysis. The regression methodology is described in Section 4. Section 5 presents the data and analyzes empirical results. Several additional tests are also conducted in Section 5. Section 6 concludes. 5 The detailed components of loan portfolio (interest income and non-interest income) used in our analysis are introduced in Section 3.2 (5.3.5). 7

2. Formulation of main hypotheses We review relevant literature and develop empirical hypotheses regarding the effects of loan diversification and market structure on bank insolvency risk in this section. 2.1. Loan portfolio diversification and a bank s financial stability The basic notion of the portfolio theory is that total volatility of a portfolio decreases as its components have low or negative correlations. Banks may choose to diversify loan-making activities across different sectors to reduce their riskiness. Sinkey and Nash (1993) find that commercial banks specializing in credit-card loans have higher probabilities of insolvency than commercial banks with the mixture of traditional products. Hughes and Mester (1998) consider the level of financial capital as a signal of risk and suggest that diversification lowers the cost of signaling. They argue that as banks grow larger, their loan portfolio and deposit base become more diversified and they are better able to economize on the use of financial capital. However, diversification benefits will be limited if diversified banks lend more of their assets to risky borrowers and operate with greater financial leverage. Froot et al. (1993) and Froot and Stein (1998) present that banks that engage in active risk management hold less capital and invest more aggressively in risky and illiquid loans. Cebenoyan and Strahan (2004) show that banks that manage their credit risk through loan sales and purchases hold more risky loans (commercial real estate loans). Diversification may aggravate bank performance if banks diversify into new lines of business where management does not have expertise and experience (Stiroh, 2006). The costly bank failure provides incentives for the bank to minimize its chance of insolvency by monitoring loan contracts. Some theoretic models suggest that diversification helps financial institutions attain credibility in their role as monitors of borrowers. Diamond (1984) in 8

his delegated monitoring model shows that diversification serves to reduce the financial intermediary s delegation costs and financial intermediaries (such as banks) can lower the probability of their default by adding more independent risks. Similar results are obtained by the related work of Ramakrishnan and Thakor (1984), and Boyd and Prescott (1986). However, this argument is challenged by a theoretical framework that incorporates financial intermediary s monitoring incentives. Winton (1999) argues that credit risk on most bank loans is endogenously affected by the intensity and efficacy of the bank s loan monitoring. Banks can influence the credit risk of a loan investment by improving monitoring quality and screening expertise. Effective monitoring enables the bank to identify troubled loans before they worsen too far, improving loan returns. Acharya et al. (2006) imply that weakened monitoring incentives and a poorer-quality loan portfolio result in diseconomies of scope when a risky bank expands loan activity into new industries and sectors. The theoretical work done by Dell'Ariccia et al. (1999), Dell'Ariccia (2001) and Marquez (2002) suggests that banks suffer an adverse-selection effect when they enter new sectors where incumbent banks have an informational advantage over new entrants by virtue of their established relationships with borrowers (the winner s curse ). This puts entrants in a worse position than the incumbents and may make diversification into new sectors more likely to increase the bank s likelihood of failure and less likely to enhance the bank s monitoring incentives (Winton, 1999). This paper aims to shed light on these contradicting implications by analyzing the loan portfolio diversification of U.S. commercial banks. We expect the loan diversification to be negatively associated with a bank s financial stability if banks diversifying their loan portfolios into new sectors face the winner s curse problem. If diversification involves in expanding into sectors where monitoring expertise is lacking, then the worse returns in new sectors may reduce 9

the bank s average loan returns and increase the risk of bank insolvency. On the other hand, we predict that the loan diversification is positively related to a bank s financial stability if engaging in various loan activities that have low or negative correlations reduces the chance of costly financial distress and increases risk-adjusted returns. Therefore, our null and alternative hypotheses are as follows: Hypothesis 1a. Loan diversification is negatively associated with a bank s financial stability. Hypothesis 1b. Loan diversification is positively associated with a bank s financial stability. 2.2. Market concentration and a bank s financial stability We also investigate the potential impact of market concentration on the bank s financial stability. The economic theory provides conflicting predictions about the relationship between the market structure of the banking industry and bank fragility. The concentration-stability view predicts that a concentrated banking system characterized by a few large banks is more stable since banks in high concentrated markets may be more profitable, better diversified (Diamond, 1984), and easier to monitor (Allen and Gale, 2000), and hence can endure shocks without collapsing. High profits facilitate building up capital buffer to provide protection against adverse financial shocks and increase the franchise value of the bank. Higher franchise values lower incentives to take excessive risk, reducing the moral-hazard problems (Keeley, 1990; Hellmann et al., 2000). A bank with a higher franchise value is likely to preserve these values by limiting its risk exposure. Theoretical arguments indicate that in less concentrated and more competitive banking systems, banks earn lower informational rents from their relationship with borrowers. The pressure on profits may provide fewer incentives for bank managers to properly screen borrowers, increasing the risk of bank fragility (Allen and Gale, 2000, 2004). Boyd et al. (2004) present that the likelihood of a costly banking crisis is lower under monopoly than in competitive market. Beck et 10

al. (2006) find using data from 69 countries that systemic banking crises are less likely in more concentrated banking systems, consistent with the concentration-stability view. On the other hand, the concentration-fragility view suggests that a more concentrated banking structure with a few large institutions is more prone to financial fragility than a less concentrated banking sector with many banks (Boyd and De Nicoló, 2005). Concentrated banking systems generally have larger banks. Large banks in concentrated banking systems tend to receive a greater net subsidy through implicit too important to fail policies. The recent research presents that the presence of deposit insurance and other government interventions lead to moral hazard problems, which may distort banks risk taking incentives. Thus, the potential subsidy for large banks may increase bank s risk taking incentives, heightening the fragility of concentrated banking systems (e.g., Boyd and Runkle, 1993; Mishkin, 1999). Boyd and De Nicoló (2005) argue that banks in more concentrated markets use their market power to earn more rents in their loan markets by charging higher loan rates. When confronted with higher loan rates charged by banks, borrowers may seek more risk to make up profit shortage, ultimately increasing their bankruptcy risk. As a consequence, theoretical analyses done by Boyd and De Nicoló (2005) indicate a negative relationship between market concentration and bank stability. Using consolidated balance sheet data across the 25 Member States of the European Union (EU-25), Uhde and Heimeshoff (2009) provide evidence that national banking market concentration has a negative impact on European banks financial soundness, supporting the concentration-fragility view. We expect to observe a positive relationship between market concentration and a bank s financial stability if more concentrated banking systems where a few large banks dominate the market enhance the bank s financial strength. Alternatively, we predict that market concentration is negatively associated with a bank s financial stability if banks can endure shocks better without 11

collapsing in less concentrated market in which the market is spread among many institutions. Hence, our null and alternative hypotheses are as follows: Hypothesis 2a. A more concentrated banking structure enhances the bank s financial stability. Hypothesis 2b. A more concentrated banking system enhances the bank s financial fragility. 3. Variable estimation 3.1. Measurement of a bank s financial stability Following the literature, we utilize a Z-score as a proxy measure of a bank s financial stability (Hannan and Hanweck, 1988; Stiroh and Rumble, 2006; Laeven and Levine, 2009; Houston et al., 2010; Shim, 2013). The Z-score of each bank is measured by the return on assets (ROA) plus the capital to asset ratio divided by the standard deviation of ROA. The standard deviation of ROA is calculated by using rolling period data over the preceding twelve quarters. The Z-score is considered as a measure of the bank s distance-to-default since it presents the number of standard deviations that profits should fall to push a bank into insolvency. The Z-score is inversely related to the probability of insolvency. Therefore, a higher Z-score indicates a lower probability of bank default. 3.2. Loan portfolio diversification We employ a Herfindahl-Hirschman index (HHI) to construct a loan-based measure of diversification for each bank. Similar to Berger and Bouwman (2013), we classify loan scope of the commercial bank into six major sectors: commercial real estate loans (REA), construction and industrial loans (IND), residential real estate loans (RES), loans to consumers (CON), agricultural 12

loans (AGR), and all other loans (OTH). 6 Loan HHI is calculated by the sum of the squared loan portfolio shares across six types of loans: Loan HHI 2 2 2 2 2 2 REA IND RES CON AGR OTH = + + + + + TOL TOL TOL TOL TOL TOL (1) where TOL denotes total loans and is equal to the sum of the values of REA, IND, RES, CON, AGR, and OTH. The loan portfolio diversification is then calculated by one minus Loan HHI. A lower value of this diversification index suggests that the bank has a specialized loan-making, while the higher value indicates that the bank engages in a combination of various loan-making activities. Loan HHI takes a value of one if all loans are made to a single sector. Alternatively, we also define diversified banks in terms of loan activities using a dummy variable which is equal to one if the loan portfolio diversification index is greater than 75th percentile of diversification index distribution and zero otherwise. 3.3. Market concentration While previous studies typically use the Metropolitan Statistical Areas (MSAs) to define geographical banking markets (e.g., Cetorelli and Strahan, 2006; Dick, 2006), we apply the new local market delineations based on Core Based Statistical Area (CBSA) and non-cbsa county. 7 As a direct measure of local market concentration, we use the deposit Herfindahl-Hirschman index 6 Because a breakdown of the U.S. commercial banks lending into specific industries is not publicly available, our loan diversification measures rely on sectoral aggregation. 7 Areas defined on the basis of these new standards were announced in June 2003. The CBSA is a collective term for both Metropolitan and Micropolitan Statistical Areas (see http://www.census.gov/population/metro/ for more details of delineations and standards). The summary of deposit data to form HHI based on the new definition are observable at the FDIC s website. The SNL data sources are available to map commercial banks into CBSA. We exclude banks not located in either Metropolitan Statistical Areas or newly-created Micropolitan Statistical Areas from the sample. 13

(HHI) calculated by the sum of the squares of the percentages of total deposits across all banks ( i = 1 to n) in each statistical area (CBSA) s and in each quarter t. Concentration( HHI ) st, = Deposits 2 n ist,, (2) n i= 1 Deposits ist,, i= 1 A high HHI indicates more concentrated market, while a low HHI suggests less concentrated market. As an alternative measure of market concentration, we also use the three-firm concentration ratio (C3) measured by the proportion of total deposits held by the largest three banks in each statistical area (CBSA). A high concentration ratio denotes more concentrated market since the market is controlled by a few large firms. The firm concentration ratios have been widely used in the empirical banking literature (e.g., Cetorelli and Gambera, 2001; Uhde and Heimeshoff, 2009). We compute the bank-level concentration HHI and three-firm concentration ratio for every quarter for each statistical area (CBSA). 3.4. Other control variables We include financial statement variables as additional controls, hypothesizing that characteristics of bank balance sheets and income statements are associated with bank insolvency risk. Firm size: The too big to fail hypothesis suggests that larger banks may have more incentives to engage in riskier lending activities due to a government s safety net. However, the charter value acts as a restraint against moral hazard (Keeley, 1990). Larger banks may deter excessive risk-taking behavior to protect their charter or franchise value. Thus, it is difficult to 14

predict a priori the direction of impact of bank size on its insolvency risk. We measure the natural logarithm of total assets as a proxy for firm size. Non-interest share: DeYoung and Roland (2001) show that replacing traditional lending activities with non-interest and fee-based activities is associated with higher volatility of bank earnings. They also find that this shift in product mix is accompanied by an increase in bank profitability, suggesting a risk premium associated with these activities. Stiroh and Rumble (2006) find that increased exposure to non-interest activities is relatively volatile but not more profitable than lending activities. A higher share of non-interest income in total income is expected to be negatively related to a bank s financial stability if increased non-interest income is more exposed to high volatility. In contrast, a positive relationship between the non-interest share and a bank s financial stability is expected if cash flows from banks expanded services are more stable and cross-selling opportunities increase revenues. Liquidity: The liquidity captures the ability of the bank to meet short-term financial obligations without having its investments or fixed assets sold quickly at lower prices. During the recent financial crisis, some financial institutions failed because they were unable to attain liquidity. The larger the liquidity, the less likely is the bank to fail. Thus, the liquidity is expected to be positively related to a bank s financial stability. The liquidity is calculated by dividing liquid assets (cash and marketable securities) by total assets. Brokered deposits: Banks can acquire deposits directly or indirectly through the mediation or assistance of deposit brokers rather than from local customers. The brokers market the pooled deposits to financial institutions for a higher rate and banks often attempt to grow rapidly using riskier funding sources such as brokered deposits. The acceptance of these brokered deposits may lead a bank to take greater risk because the bank must earn more to pay high interest costs 15

(Government Accountability Office, 2013). DeYoung and Torna (2013) and Cole and White (2012) suggest that brokered deposits tend to be positively associated with the likelihood of bank failure. Berger and Bouwman (2013) show that small banks are less likely to survive if they have more brokered deposits. The higher level of brokered deposits is expected to be negatively associated with a bank s financial stability. Core deposits: Core deposits are typically funds of a bank s regular customers and viewed as relatively stable and less costly sources of funding with the lower interest rates. Following the Uniform Bank Performance Report (UBPR) User Guide, we define core deposits as the sum of demand deposits, automatic transfer service (ATS) accounts, money market deposit accounts (MMDAs), savings deposits and time deposits under $100,000, minus brokered deposits under $100,000, normalized by total assets. 8 Berger and Bouwman (2013) show that more core deposits help small and medium-sized banks survive. DeYoung and Torna (2013) find that core deposits are associated with a reduced probability of failure. We expect a positive coefficient on this variable if banks with larger shares of core deposits face a lower chance of bank failure. Commercial real estate: The bank failure and financial distress literature suggests that concentration of commercial real estate loans is one of the most important determinants in identifying risky banks (e.g., Cole and Gunther, 1995; Wheelock and Wilson, 2000; Cole and White, 2012; DeYoung and Torna, 2013). We include an indicator variable equal to one if the ratio of commercial real estate loans to total assets is greater than 90th percentile and zero otherwise to examine whether a focus of commercial real estate loans is directly correlated with a bank s financial stability. 8 As of March 31, 2011, the definition was modified to reflect the FDIC s deposit insurance coverage increase from $100,000 to $250,000 (Federal Deposit Insurance Corporation, 2011). 16

Member of bank holding company: To control for different banking organization, we include an indicator variable equal to one if the bank is a member of bank holding company (BHC) and zero otherwise. BHC membership is predicted to be positively associated with a bank s financial stability if banks affiliated with BHC have ready access to greater financial resources and managerial expertise when needed. De novo banks: Newly chartered banks can be an important source of competition in local markets and tend to specialize in supplying the credit needs of small businesses. However, the literature suggests a relatively high failure rate of these de novo banks compared with established ones (e.g., DeYoung, 2003). To investigate the relationship between de novo banks and their insolvency risk, we include an indicator variable set equal to one for de novo banks and zero otherwise. The coefficient sign of this variable is expected to be negative if de novo banks are less profitable and more financially fragile than their established bank counterparts. Supervisory choice: A bank has the option to choose its supervisor in general. The presence of several supervisors leads to differences in the leniency of supervisory constraints and supervisory costs. Excessive leniency may facilitate the bank s excessive risk taking. To examine how supervisory choice influences the bank s risk taking behavior, we include two indicator variables set equal to one for banks with the Federal Reserve (for state member banks) and the FDIC (for state nonmember banks), respectively and set equal to zero otherwise. The omitted category consists of banks supervised by the Office of the Comptroller of the Currency (for national charter banks). Unemployment rate: To examine the impact of local economic conditions on the bank's insolvency risk, unemployment rate in the state where the bank is headquartered is included as a proxy for the economic conditions in the bank's home market. The state-level unemployment rate 17

is a good indicator of where the economy is headed. We expect this variable to be negatively associated with a bank s financial stability if bank profits fall (rise) in economic downturns (upturns) and if the volatility of the bank profits decreases (increases) during the economic upturns (downturns). 4. Methodology To examine links between loan portfolio diversification, market concentration and downturns while controlling for firm-specific and economic characteristics, we initially conduct multivariate ordinary least squares (OLS) regressions using a series of pooled, cross-sectional, and time-series data. We use unbalanced panel data to avoid survivor bias and to maximize the number of observations. One line of recent research argues that the observed diversification discount is attributable to endogeneity problems (Campa and Kedia, 2002; Villalonga, 2004a). A bank s decision to diversify can be endogenous if the diversification variable is correlated with other omitted variables such as management skill or industry exposure that influences the risk of bank insolvency. The presence of potential endogeneity problems may lead the standard ordinary least squares (OLS) approach to produce biased and inconsistent coefficient estimates. To mitigate some endogeneity concerns, we employ fixed-effect models that include state, time (quarter), and firm dummies. Fixed-effect models enable us to account for unobservable changes at the state-quarter level, such as changes in competition and to control for omitted bankspecific effects which may be correlated with other variables in the model. The basic regression model to test our hypotheses is written as follows: Zit, = α0 + α1 Divit, + α2 MCt + αkx it, + dit, + sit, + fit, + νit, (3) where Z it, denotes the Z-score of bank i at time, t Divit, is a measure of loan portfolio diversification at time t, MC is a measure of market concentration, X it, is a matrix of other control t 18

variables, d it, ( s, and f it, ) is a vector of time (state and firm) fixed-effect, and v, is the error it term. Similar to Laeven and Levine (2009) and Houston et al. (2010), we use the natural logarithm of Z-score considering high skewness of Z-score across our sample banks. The definitions and expected signs of the variables in Equation (3) are presented in Table 1. We supplement our analysis by using two-stage least squares (2SLS) estimation techniques with instrumental variables, which also address the issue concerning the endogeneity of the loan diversification measure. The 2SLS methods involve the selection of appropriate instrumental variables. An instrumental variable must satisfy two requirements that we refer to as the relevance and validity conditions. The relevance condition requires that the partial correlation between the instrument and the endogenous variable not be zero. The validity condition requires the correlation between the instrument and the structural error term to be zero (Wooldridge, 2002). The lagged or historically averaged measures of firm characteristics, industry growth, and general economic growth are suggested as commonly used instrumental variables (Campa and Keida, 2002). Laeven and Levine (2007) include firm size (the log of total assets) and average loan diversification of other banks as instruments for the diversification variable. Following the literature, we take account of three-year average loan diversification of other banks domiciled in the same CBSA as bank i (CBSA loan diversification) and three-year average asset growth rate across all banks domiciled in the same CBSA as bank i (CBSA asset growth) for our instruments. These instrumental variables are estimated at CBSA-level for each quarter. We perform the Durbin-Wu-Hausman (DWH) tests to determine whether the loan diversification variable can be treated as exogenous. We conduct tests to examine whether instrumental variables satisfy two conditions. The relevance condition can be empirically tested it 19

by investigating the fit of the first stage regression. 9 The partial R 2 and the F-statistic are commonly used to assess the relevance of the excluded instruments. The partial R 2 statistic measures the correlation between the excluded instruments and the endogenous variable after partialling out the effect of the included exogenous variables. The F-statistic tests the joint significance of the excluded instruments. If F-statistic is significant, we reject the null hypothesis that the set of instruments is weak. 10 We perform a test of overidentifying restrictions with Hansen s J-statistic to investigate whether the instruments are uncorrelated with the error term. A significant test statistic represents invalid instruments, implying that the instruments do not satisfy the orthogonality (no correlation with error term) conditions. Test results suggest that the aforementioned two variables (CBSA loan diversification and CBSA asset growth) satisfy both the relevance and validity requirements. 5. Data and empirical results 5.1. The sample The financial data representing banks portfolio and operating characteristics are taken from the Call Reports. Our sample consists of an unbalanced panel on a quarterly frequency over the period between 2002: Q1 and 2013: Q3. 11 To avoid survivorship bias, our sample contains both failed and non-failed commercial banks operating at any point over the sample period. Statelevel unemployment data are taken from the Bureau of Labor Statistics Employment and Earnings. 9 The first stage of the 2SLS method involves regressing the endogenous variable (loan diversification) on a set of instruments and other explanatory variables in Equation (3). 10 Stock et al. (2002) indicate that the instruments are only weakly correlated with the endogenous regressor if the first-stage F-statistic is low (less than 10). The weak instruments may cause the 2SLS estimators to be biased. 11 Because of calculating the standard deviation of ROA based on the preceding twelve-quarter rolling periods, other variables for regression analysis span the period from 2005: Q1 through 2013: Q3. 20

Similar to the literature (e.g., Laeven and Levine, 2007), we eliminate banks that report missing values in accounting variables such as assets, equity capital, deposits, total loans, interest income and non-interest income. There are a number of extreme values among the observations of dependent variable (Z-score) and financial ratios constructed from raw data. Similar to the literature (e.g., Stiroh and Rumble, 2006; Houston et al., 2010; Saghi-Zedek, 2016) and to ensure that statistical outcomes are not severely influenced by outliers, we winsorize our main dependent variable (Z-score) along with other control variables at the 1% and 99% levels. Finally, the banks that do not have at least twelve continuous quarterly time series observations are excluded because we need to calculate rolling-window standard deviations of ROA over the preceding twelve quarters. This procedure leads to a final sample of approximately 136,400 quarterly observations. The descriptive statistics on the variables used in the regressions are presented in Table 2. 5.2. Results for primary specification Table 3 reports estimations of the parameters from the Equation (3) using the Z-score as a measure of bank insolvency risk. 12 Columns 1-2 (3-4) present results of the OLS with state and quarter (state, quarter and bank) dummies, and columns 5-6 provide results using the 2SLS method with selected instruments. 13 Standard errors that control for heteroskedasticity and firm-level clustering are reported in parentheses (Petersen, 2009). The Durbin-Wu-Hausman (DWH) tests show that we reject the null hypothesis of exogeneity for loan diversification variable at the 1 percent significant level. The partial 2 R statistic exhibits that the selected instruments are strongly 12 We estimate the Variance Inflation Factor (VIF) to investigate the presence of multicollinearity for the variables used in the regressions. Because values of VIF are less than 3.0 for all variables, there is no multicollinearity concerns in our model. 13 The coefficient estimates of state, quarter and bank dummies are not reported to conserve space. Alternatively, we include CBSA, quarter and bank fixed effects. The main results are not affected. 21

correlated with the endogenous variable. The F-statistic for the joint significance of the excluded instruments exceeds 10 and its p-value is significant at the 1% level, demonstrating that our instruments are not weak and thus, 2SLS estimator is reliable. An insignificant Hansen s J-statistic indicates that the instruments satisfy the validity condition required for their employment. The results in Table 3 show that the estimated coefficients of loan diversification are positive and significant within the 1% significance level, indicating that loan diversification is positively associated with a bank s financial stability. The results suggest that banks diversifying their loan portfolio can reduce the risk of their insolvency more efficiently than banks focusing their loan-making on the specialized area. The coefficients of market concentration measured by CBSA-level deposit HHI and threefirm concentration ratio (C3) are statistically significant and positive, showing that market concentration is positively associated with a bank s financial stability. The result suggests that banks operating in the concentrated market structures with a few large firms that supply most of the market are more financially secure than those in less concentrated market structures with many institutions, each with a small share of the market. This result provides evidence supporting the concentration-stability view and is consistent with previous theoretical and empirical studies (e.g., Boyd et al., 2004; Beck et al., 2006). 14 The coefficient of bank size is statistically significant and positive across all models, suggesting that large banks tend to have lower insolvency risk than small banks. The result might be consistent with the view that charter or franchise value acts against moral hazard incentive (Keeley, 1990). The non-interest income share is negatively related to the Z-score, as expected if 14 Following Berger and Bouwman (2013), we eliminate banks with total assets less than $25 million and perform regressions again with a reduced sample (130,870 quarterly observations). We find that the results are very similar to those presented in this Section. 22

the growing share of non-interest income results in the increased volatility of accounting returns. The coefficient of liquidity is positive and significant in four of the six regressions, suggesting that a greater proportion of liquid assets have a positive effect on the bank s financial strength. The coefficient on the brokered deposits is negative and significant, indicating that greater reliance on brokered deposits may have a negative impact on the bank s financial health. In contrast, core deposits have a positive influence on the bank s financial strength, as expected if core deposits are considered to be a stable and less costly source of funding. The results suggest that the use of brokered deposits rather than core deposits is associated with an increased likelihood of bank insolvency. The negative and significant coefficients on commercial real estate loans suggest that concentration of commercial real estate loans has a negative impact on the bank s financial strength, consistent with prior findings (e.g., Cole and White, 2012). The significant and positive sign of member of BHC variable in OLS results indicates that BHC membership may be advantageous for the bank s financial safety. The negative and significant coefficient of de novo banks shows that de novo banks, all else being equal, have a significantly higher likelihood of being insolvent compared with more mature ones. The coefficients of the FDIC and Federal Reserve are marginally significant only in a few regressions, providing limited information about the impact of the FDIC and Federal Reserve supervision on bank insolvency risk. The coefficient of the state-level unemployment rate is statistically significant and negative across all models, indicating that economic conditions in the markets where a bank operates appear to affect financial health of U.S. commercial banks. The result implies that banks operating in states with robust economies are likely to have a relatively lower probability of insolvency, while banks in depressed states are more likely to suffer financial problems. 23

In columns 5-6 of Table 3, we present the results of first-stage regression of the 2SLS method, where the dependent variable is the loan diversification. The excluded instruments show the statistical relevance with the endogenous variable (loan diversification measure). The threeyear average loan diversification of other banks (CBSA loan diversification) is positive and significant, suggesting that a bank is more likely to diversify loan portfolio if its peers diversify their lending activities. The three-year average asset growth rate (CBSA asset growth) is significant and positive, indicating that banks tend to become more diversified as CBSA-level average asset growth rate increases. Among other explanatory variables in the first-stage regression, three-firm concentration ratio (C3), firm size, liquidity, and core deposits are positively associated with loan diversification, while the focus of commercial real estate, de novo banks and unemployment rate are negatively related to loan diversification. 5.3. Robustness tests and extensions We perform several additional tests to further understand the relationship between the diversification of bank activities, market concentration and bank insolvency risk. In this section, we present mostly the results of our interested variables to save space, although all other control variables and state-quarter indicator variables used in Table 3 are included in the regressions. 5.3.1. Heckman s self-selection model We check the robustness of our results using Heckman s (1979) two-step procedure that controls for potential self-selection bias induced by banks choosing to diversify. In the first step of Heckman method, we estimate a probit model where the dependent variable is an indicator variable equal to one for diversified banks defined as the ones with their loan diversification greater 24

than 75th percentile and zero otherwise. 15 The explanatory variables in the first-stage regression include control variables in Table 3 and instrumental variables used in previous 2SLS. The selfselection parameter (an inverse Mill s ratio) is estimated from the first-stage binary choice model. In the second stage, bank insolvency risk is regressed on the predicted value for loan diversification obtained in the first-stage regression, market concentration, other control variables and the selfselection parameter, while controlling for self-selection bias. The results of Heckman s analysis are presented in Table 4. As an alternative measure of market concentration, we compute CBSA-level HHI and three-firm concentration ratio with the bank s total assets. Columns 1-2 provide results using the market concentration measured by total deposits, while columns 3-4 show results applying the bank s total assets to estimate market concentration. The coefficients of both loan diversification and market concentration are significant and positive in the second-stage regression across all models, confirming the existence of loan diversification benefits and reinforcing the concentration-stability view. The significant coefficients on the inverse Mill s ratio indicate that self-selection has a substantial impact on the estimates of bank insolvency risk in the second-stage regression and thus controlling for selfselection bias is appropriate. Table 4 shows that the results of instruments in the first-stage regression are consistent with those presented in Table 3. 5.3.2. Controlling for mergers and acquisitions Graham et al. (2002) argue that diversification discount arises not because diversification destroys value but because acquiring firms purchase already discounted target firms. Similar to Laeven and Levine (2007), we control for potential influence of M & As in two ways. First, we 15 To explore the sensitivity of the results, we define banks as diversified if loan diversification is greater than 90th percentile and use an indicator variable equal to one for those diversified banks and zero otherwise. We find that the results are unaffected. 25

use an indicator variable equal to one in the year of M & A completion for each bank that involves in acquiring or merging with at least one other bank and zero otherwise. Alternatively, we include indicator variables that take the value of one in the year of M & A completion and all later years for each bank that acquires or merges with at least one other bank and zero otherwise. Second, we exclude observations of acquiring banks for year t when M & A transaction occurs to account for the possibility that M & As can impact the diversification results. Additionally, we exclude bank observations in that year and all later years if the bank engages in merging with at least one other bank in year t. Columns 1-4 in Panel A of Table 5 provide results using the indicator variable in the year of M & A completion and columns 5-8 in Panel A of Table 5 present results with indicator variables in the year of M & A completion and all later years. Similarly, the first (last) four columns in Panel B of Table 5 show the results that exclude acquiring banks for the year of M & A completion (for the M & A year and all later years). As shown in Panels A and B of Table 5, the coefficients of loan diversification and market concentration are statistically significant and positive across all estimations, confirming that controlling for M & A activities does not affect our key findings in Table 3. 16 5.3.3. Sub-sample analysis 16 Following Laeven and Levine (2007), we exclude bank observations in year t and all later years if the bank s total assets increase more than 50% from year t-1 to year t. We exclude acquiring banks that have engaged in M & A activities during the current year, the past three years, or the past five years. We include indicator variables that take the value of one if the bank has acquired or merged with at least one other bank from year t-2 to year t or from year t- 4 to year t, respectively and zero otherwise. We also control for possible impact of M & As on a quarterly basis. The signs of the coefficients on the loan diversification and market concentration are not influenced and remain statistically significant in all modifications. The results are available from the authors upon request. 26