How Do Banks React to Catastrophic Events? Evidence from Hurricane Katrina

How Do Banks React to Catastrophic Events? Evidence from Hurricane Katrina Abstract This paper explores how banks react to an exogenous shock caused by Hurricane Katrina in 2005, and how the structure of the banking system affects economic development following the shock. Independent banks based in the disaster areas increase their risk-based capital ratios after the hurricane, while those that are part of a bank holding company on average do not. The effect on independent banks mainly comes from the subgroup of highly capitalized banks. These independent and highly capitalized banks increase their holdings in government securities and reduce their total loan exposures to non-financial firms, while also increasing new lending to these firms. With regard to local economic development, affected counties with a relatively large share of independent banks and relatively high average bank capital ratios show higher economic growth than other affected counties following the catastrophic event. 1. Introduction How banks navigate through an economic crisis is an important issue to explore, since bank financing is crucial for economic recovery and development. From a policy perspective, it would be desirable if banks could continue to provide financing to borrowers, while also maintaining their stability in an unfavorable market environment that adversely impacts their asset quality. Related studies analyze bank lending in the wake of a crisis (e.g., Gan, 2007; Garmaise and Moskowitz, 2009; Ivashina and Scharfstein, 2010; Puri, Rocholl, and Steffen, 2011; De Haas and Van Horen, 2013; Cortes and Strahan, 2017),1 but little is known about how the business decisions of banks affect their stability and asset allocations during such times, and how the structure of the local banking system is related to economic development in affected areas following a crisis.2 This paper first explores how banks adjust their risk-based capital ratios, which are a key determinant of a bank’s stability and a cornerstone of banking regulation. Second, we analyze the mechanisms of these adjustments, that is, the asset allocations and lending practices of banks. The analysis specifically considers the role of different types of banks with regard to bank structure (independent or part of a bank holding company) and bank capitalization (relatively low or high capital ratios). Finally, we analyze whether characteristics of the local banking system (the share of banks that belong to a bank holding company and the average of banks’ risk-based capital ratios by county) are related to local economic development after a crisis. As different arguments point in different directions, it is not clear what findings to expect from these analyses. More specifically, banks may increase their capital ratios to foster financial stability and protect their franchise values, or they may decrease capital ratios to benefit from risk-shifting in times of crisis. These strategies may be associated with decreased or increased lending, respectively. Furthermore, the presence of a relatively large share of independent banks may be associated with relatively high economic growth, since these banks usually have a local focus and may have advantages in screening and monitoring local borrowers. It could also be associated with relatively low economic growth, since these banks are typically less diversified and face more capital constraints compared with banks belonging to bank holding companies. The challenge for the analysis is twofold: first, to identify the direction of the relationship between a crisis and the banks’ business decisions,3 and second, to control for parallel economic developments, which may also affect banks’ financial figures but may not be the result of active changes in banks’ financing or investment decisions. In order to identify causality between a crisis and banks’ business decisions, we use Hurricane Katrina and two subsequent hurricanes that struck the U.S. Gulf Coast in the third and fourth quarters of 2005 as a natural experiment. Hurricane Katrina ranks among the costliest natural disasters in U.S. history, with estimated property damages ranging from $100 billion to over $200 billion (National Hurricane Center, 2005; Congleton, 2006). The hurricanes exposed banks in the U.S. Gulf Coast region to unexpected losses and weakened their asset quality, as a large part of the damages suffered by borrowers was not insured. It also caused uncertainty for banks with respect to how individual and commercial borrowers would cope with the damages. Asymmetric information between banks and their borrowers increased and it was also uncertain as to how the overall economy in the affected regions would recover from the shock. The FDIC (2006) characterized the situation as follows: Hurricane Katrina had a devastating effect on the U.S. Gulf Coast region that will continue to affect the business activities of the financial institutions serving this area for the foreseeable future. Some of these institutions may face significant loan quality issues caused by business failures, interruptions of borrowers’ income streams, increases in borrowers’ operating costs, the loss of jobs, and uninsured or underinsured collateral damage. In keeping with this, the major rating agencies announced a close monitoring of capital adequacy and the risk-management processes of affected banks in the aftermath of Hurricane Katrina (Moody’s, 2005a, 2005b). Katrina also led to a change in the perceived hurricane risks, as reflected in property insurance premium increases of 30% or more (USA TODAY, 2010). Using this natural experiment, we analyze a large sample of U.S. banks within a difference-in-difference framework. The treatment group comprises affected banks, while the control group comprises unaffected banks in the U.S. Gulf Coast region and neighboring states. When the analysis turns its focus on local economic development after the 2005 hurricane season, we use a corresponding sample of affected and unaffected counties. The key findings of our empirical analysis are as follows: independent banks in the disaster areas increase their risk-based capital ratios after the hurricane compared with the control group (unaffected by the shock), as illustrated in the graph on the left of Figure 1, while those that are part of a bank holding company on average do not. Independent banks thereby strengthen their buffer against future income shocks and thus mitigate bankruptcy risks. Our results therefore suggest that asset quality is an important determinant of a bank’s risk-based capital ratio as long as that bank is not backed by a larger banking organization. When we examine independent banks that are low-capitalized (below the median) or highly capitalized (above the median) separately, we find that this precautionary behavior only holds for the latter of the two. A potential explanation is that banks with high franchise values and/or high bankruptcy costs have incentives to avoid bankruptcy and are thus characterized by high capital ratios (before a hurricane) and precautionary behavior (after a hurricane). Our analysis also shows that highly capitalized independent banks achieve higher risk-based capital ratios by prioritizing lower risk-weighted assets: they increase their holdings in government securities and reduce their existing loan exposures to non-financial firms compared with the control group. The latter is illustrated in the graph on the right of Figure 1, where banks’ loan exposures to non-financial firms are represented by the ratio of the volume of commercial and industrial loans (C&I loans) to assets. We also explore banks’ new lending transactions with non-financial firms based on a sample from the Small Business Administration (SBA) loan program, and we find that highly capitalized independent banks also increase their new lending to non-financial firms in their core markets (where they have a branch presence). This suggests that these banks reduce their loan exposures not through reduced new lending, but rather through other strategies such as loan sales. Figure 1. View largeDownload slide Impact of the 2005 hurricanes on banks’ risk-based capital ratios and loans. Notes: The graphs show the development of banks’ risk-based capital ratios and the ratio of the volume of commercial and industrial loans to assets (C&I loans/Assets) from the third quarter of 2003 through the fourth quarter of 2007. The mean values for independent banks located in areas affected by the 2005 hurricanes are represented by a solid line. The mean values for independent banks located in the U.S. Gulf Coast region or neighboring states, but not affected by the hurricanes, are represented by a dotted line. The solid vertical lines indicate the quarters around the disaster period of the third and fourth quarters of 2005, when Hurricanes Katrina, Rita, and Wilma hit the U.S. Gulf Coast region. Figure 1. View largeDownload slide Impact of the 2005 hurricanes on banks’ risk-based capital ratios and loans. Notes: The graphs show the development of banks’ risk-based capital ratios and the ratio of the volume of commercial and industrial loans to assets (C&I loans/Assets) from the third quarter of 2003 through the fourth quarter of 2007. The mean values for independent banks located in areas affected by the 2005 hurricanes are represented by a solid line. The mean values for independent banks located in the U.S. Gulf Coast region or neighboring states, but not affected by the hurricanes, are represented by a dotted line. The solid vertical lines indicate the quarters around the disaster period of the third and fourth quarters of 2005, when Hurricanes Katrina, Rita, and Wilma hit the U.S. Gulf Coast region. Finally, our analysis provides new evidence on the role of the structure of the banking system to foster growth and employment in the post-Katrina period. We assess whether affected counties that host a relatively small or large number of independent banks (not part of a bank holding company) and banks with relatively low or high pre-Katrina capital ratios develop differently in the post-crisis period. Our evidence shows that counties with a higher share of independent banks and relatively high average bank capital ratios are associated with better economic growth in total personal income and employment compared with other counties (with fewer independent banks or lower average bank capital ratios). This has important policy implications because it suggests that the promotion of independent (locally focused) banks and higher bank capital requirements may mitigate economic costs in areas affected by a crisis. We particularly find that the change (before versus after the natural disaster) in total personal income and employment in an affected county is about 5 and 4 percentage points higher, respectively, if the county has a relatively large share of independent banks and relatively high average bank capital ratios compared with a county with only bank holding company banks (BHC banks) and relatively low bank capital ratios.4 Our research contributes to several strands of literature. First, we contribute to the literature that analyzes the relationship between asset quality (or, related to this, asset risk) and bank capital, using a natural experiment as the identification strategy. Previous studies face the difficulty that asset quality and bank capital are typically determined simultaneously by banks. Using simultaneous equations, two-stage, or standard OLS estimation techniques, these studies typically find a positive relationship between asset risk and capital ratios, that is, a negative relationship between asset quality and capital ratios (e.g., Shrieves and Dahl, 1992; Flannery and Rangan, 2008; Gropp and Heider, 2010). Our findings are in line with the findings from these studies and, using an exogenous shock to banks’ asset quality, provide further evidence on the relationship between a bank’s asset quality and risk-based capital ratio. Importantly, we also consider different bank characteristics, that is, independent banks versus banks that are part of a bank holding company, and low-capitalized versus highly capitalized banks, providing evidence of how these characteristics are associated with banks’ risk-based capital ratio adjustments in the wake of a crisis. Related empirical evidence shows a positive relationship between banks’ risk-based capital ratios and their stability. For example, Berger and Bouwman (2013) find that higher pre-crisis bank capital, measured as equity-to-assets or risk-based capital ratio, is associated with higher survival probability during a banking crisis. Demirguc-Kunt, Detragiache, and Merrouche (2013) show that higher leverage and regulatory capital ratios are associated with better stock market performance during the financial crisis. Hence, our results on banks’ capital ratio adjustments are also relevant for understanding banks’ stability. Furthermore, the results of this paper contribute to studies that assess the consequences of various types of crises on bank lending. For example, using the land-price collapse in Japan in the early 1990s as an exogenous shock, Gan (2007) reports that firms with greater collateral losses receive less credit and reduce investments. Garmaise and Moskowitz (2009) use the 1994 Northridge earthquake in California to show that earthquake risk impacts credit markets with a reduction of over 20% in the provision of commercial real estate loans. Chavaz (2016) finds that banks with more concentrated portfolios in markets affected by the 2005 hurricane season maintain lending in markets hit by the shock and circumvent potential capital constraints through loan sales. Cortes and Strahan (2017) show that, following natural disasters, multi-market banks reallocate funds into markets affected by the disasters (with high credit demand) and away from unaffected markets in which they do not own any branches. They also find that banks do not reduce lending in unaffected core markets in which they own branches. As regards the recent financial crisis, the literature finds that stricken banks reduced lending, which, in turn, also led to lower corporate investment. Ivashina and Scharfstein (2010) find that banks are less likely to cut lending if they have better access to deposit financing and thus rely less on short-term debt. Santos (2011) shows that banks with larger losses during the subprime crisis requested higher loan spreads from their corporate borrowers than did banks with smaller losses. Puri, Rocholl, and Steffen (2011) find that the U.S. financial crisis led to a contraction in bank retail lending in Germany among those banks that experienced losses within their banking organizations. In addition, cross-border lending decreased during the financial crisis (Giannetti and Laeven, 2012), while the deeper financial integration of banks in foreign countries is associated with more stable cross-border lending (De Haas and Van Horen, 2013). A study by Berger, Bouwman, and Kim (2017) finds that—particularly in adverse economic conditions—small banks have comparative advantages over large banks in alleviating the financial constraints of small businesses. Our paper presents complementary evidence that reveals, in particular, that independent banks (not part of a bank holding company) with relatively high pre-Katrina capital ratios reduce their total loan exposures to non-financial firms after the shock caused by Hurricane Katrina, while also increasing new lending to non-financial firms under the SBA loan program. Finally, our paper is related to the literature that analyzes the role of banks in economic recovery and development following a crisis. Cortes (2014) analyzes U.S. county-level data and shows that the presence of local lenders is beneficial for job creation following a natural disaster. Several cross-country studies suggest that higher average bank capital ratios contribute to quicker recoveries from financial crisis recessions (e.g., Cecchetti, King, and Yetman, 2011; Jordà et al., 2017). Our study contributes to these findings by showing that affected counties with a relatively large share of (local) independent banks and relatively high average bank capital ratios show higher economic growth than other affected counties following the catastrophic event. The paper proceeds as follows: Section 2 provides background information on the 2005 hurricane season and how it affected the economy in the U.S. Gulf Coast region. Section 3 presents the identification strategy and the data used in our analysis. Section 4 shows the empirical model and the estimation results. Section 5 concludes. 2. Background on Hurricane Katrina and the 2005 Hurricane Season The heavy winds, rain, and flooding brought by Hurricane Katrina hit the mainland on August 29, 2005, having swept north from the Gulf of Mexico. Only weeks later, on September 24, 2005, Hurricane Rita came ashore, amplifying the effects of Hurricane Katrina. Finally, 1 month later, in October 2005, Hurricane Wilma made landfall in Florida. Overall, the 2005 hurricane season caused massive destruction and had significant negative effects on the economy in the affected U.S. Gulf Coast region. 2.1 Damage Estimates Hurricanes Katrina, Rita, and Wilma rank among the costliest natural disasters in the history of the USA. Estimated property damages from Hurricane Katrina alone range from approximately $100 billion (National Hurricane Center, 2005; Hazards & Vulnerability Research Institute, 2014), $125 billion, to $150 billion (Congressional Research Service, 2013), and up to over $200 billion (Congleton, 2006). Among its destructive effects, Hurricane Katrina made approximately 300,000 homes uninhabitable, which caused more than 400,000 citizens to relocate (Congressional Research Service, 2013). While Hurricane Katrina brought significantly more destruction than Hurricane Rita or Hurricane Wilma, all three hurricanes rank among the most intense and costliest hurricanes over the last 100 years (National Hurricane Center, 2011; Hurricane Research Division, 2015). Estimated yearly losses from natural disasters over the period 1960–2012, based on data from the Hazards & Vulnerability Research Institute (2014), are illustrated in Figure 2. The estimate for 2005 is about $120 billion, which includes losses from Hurricane Katrina (about $100 billion), Hurricane Rita (about $10 billion), and Hurricane Wilma (about $10 billion). The figure shows that the losses from the 2005 hurricane season exceed the losses of previous and subsequent periods by far. Therefore, while areas affected by Hurricanes Katrina, Rita, and Wilma are located in hurricane states where hurricanes are not uncommon, the extraordinary impact of the 2005 hurricane season suggests that a significant part of the damages sustained was unexpected. Figure 2. View largeDownload slide Annual disaster losses since 1960. Notes: The figure shows the total sum of yearly disaster losses for the states in our baseline sample: Alabama, Louisiana, Florida, Mississippi, Georgia, Tennessee, Oklahoma, and Arkansas. The numbers are expressed in US$billions and adjusted to 2011 dollar values. The data source is the Hazards & Vulnerability Research Institute (Sheldus database). Figure 2. View largeDownload slide Annual disaster losses since 1960. Notes: The figure shows the total sum of yearly disaster losses for the states in our baseline sample: Alabama, Louisiana, Florida, Mississippi, Georgia, Tennessee, Oklahoma, and Arkansas. The numbers are expressed in US$billions and adjusted to 2011 dollar values. The data source is the Hazards & Vulnerability Research Institute (Sheldus database). 2.2 Insurance Payments and Federal Disaster Assistance The effects that natural disasters have on households and institutions are mitigated through insurance payments and federal disaster assistance. Such support is significant, but cannot offset the huge losses from a natural disaster. This is especially so when the magnitude of a natural disaster is huge and unexpected, as in the case of the 2005 hurricane season. According to the Insurance Information Institute (2014), about 50% of losses from the 2005 natural disasters were insured. The American Insurance Services Group estimates that Katrina is responsible for a total of $41.1 billion of insured losses in the USA (National Hurricane Center, 2005). As a consequence of these unprecedented losses, insurance prices in catastrophe-prone areas were expected to rise and insurance terms and conditions were expected to be tightened, “as insurers seek to control their aggregate hurricane exposure” (Towers Watson, 2005). Hence, in addition to the immediate losses from the 2005 hurricane season, individuals and firms in affected areas were also facing more expensive and restricted insurance contracts, which made it more difficult to protect against potential disasters in the future. In addition to potential insurance payments, the U.S. federal government offers assistance and funding through a variety of agencies and programs. One such agency, the Federal Emergency Management Agency (FEMA) was specifically created in 1979 to coordinate the response to a disaster. As outlined in the announcement of a Presidential Disaster Declaration, FEMA’s disaster assistance programs provide assistance to individuals (“individual assistance”), jurisdictions (“public assistance”), and funds for “hazard mitigation”.5 Individual assistance is directed to individuals and families whose property has been damaged and whose losses are not covered by insurance. Public assistance supports state or local governments as they rebuild a community’s damaged infrastructure, which includes “debris removal, emergency protective measures and public services, repair of damaged public property, loans needed by communities for essential government functions and grants for public schools” (FEMA, 2015). Funds for hazard mitigation are used to “assist communities in implementing long-term measures to help reduce the potential risk of future damages to facilities” (U.S. Government Accountability Office, 2012). The majority of federal assistance is funded through FEMA’s Disaster Relief Fund, which made obligations of roughly $40 billion with respect to damages caused by Hurricane Katrina (U.S. Government Accountability Office, 2012). 2.3 Implications for the Economy Despite the financial support that came from insurance payments and federal disaster assistance, the hurricanes that swept the region in 2005 were expected to have “substantial and long-term effects on the economies of southern Louisiana and Mississippi” (Congressional Research Service, 2005). The graph on the left of Figure 3 depicts the number of initial jobless claims filed in Louisiana between 2000 and 2009.6 It shows a significant increase in the third and fourth quarters of 2005, reflecting the desolate situation during the 2005 hurricane season. The graph on the right illustrates the development of the CredAbility Consumer Distress Index in Louisiana, which is published by the St. Louis Fed and measures the financial condition of the average consumer. The index incorporates various data including employment, housing, credit scores, household budget, and net worth.7 A higher measure of the index reflects a more favorable situation for the average household. The index shows the dramatic consequences for the financial situation of Louisiana households right after the 2005 hurricane season; an all-time low in the fourth quarter of 2005, reaching even below the levels recorded during the recent financial crisis. Figure 3. View largeDownload slide Impact of Hurricane Katrina on the Louisiana economy. Notes: The left graph shows initial jobless claims (in thousands) and the right graph shows the CredAbility Consumer Distress Index for Louisiana, where a higher score shows a more favorable situation and a score under 70 indicates financial distress. Both graphs reflect quarterly values from the first quarter of 2000 through the fourth quarter of 2009. The solid vertical lines indicate the quarters around the disaster period of the third and fourth quarters of 2005, when Hurricanes Katrina, Rita, and Wilma hit the U.S. Gulf Coast region. The data source for both graphs is the FRED database of the St. Louis Fed. Figure 3. View largeDownload slide Impact of Hurricane Katrina on the Louisiana economy. Notes: The left graph shows initial jobless claims (in thousands) and the right graph shows the CredAbility Consumer Distress Index for Louisiana, where a higher score shows a more favorable situation and a score under 70 indicates financial distress. Both graphs reflect quarterly values from the first quarter of 2000 through the fourth quarter of 2009. The solid vertical lines indicate the quarters around the disaster period of the third and fourth quarters of 2005, when Hurricanes Katrina, Rita, and Wilma hit the U.S. Gulf Coast region. The data source for both graphs is the FRED database of the St. Louis Fed. Previous research also points to the adverse effects of hurricanes on local economic conditions. For example, Strobl (2011) studies hurricanes in the USA over the period 1948–2005 and finds a 0.45 percentage point decline in economic growth rates in affected counties. Deryugina, Kawano, and Levitt (2014) show that Katrina victims face an initial negative wage income shock 1 year after the disaster, but also that the gap in wage income disappears 2 years after the storm. Summing up, the 2005 hurricane season caused significant uninsured losses and—at least temporarily—a significant deterioration of local economic conditions. Moreover, it created uncertainty regarding how households and the economy would recover from the disaster. 3. Identification Strategy and Data This sections starts with a description of our identification strategy. The following subsections provide detailed information on the characteristics of our sample. 3.1 Identification of Affected and Unaffected Counties and Banks Following Hurricane Katrina and the subsequent Hurricanes Rita and Wilma in the second half of 2005, FEMA designated 135 out of 534 counties in the Gulf Coast region (Louisiana, Mississippi, Texas, Florida, and Alabama) as eligible for individual and public disaster assistance.8 We classify these counties as affected by the 2005 hurricane season (dark-gray shaded region in Figure 4). Correspondingly, we classify a bank as affected if its headquarters is located in an affected county. Next, we classify a county as unaffected if it is not eligible for public or individual disaster assistance and located in the U.S. Gulf Coast region or a neighboring state (light-gray shaded area in Figure 4). Banks with their headquarters in these counties are also classified as unaffected. Lastly, we exclude counties that are eligible for public disaster assistance but not eligible for individual disaster assistance—as well as banks with their headquarters in these counties—because this criterion is ambiguous. For example, counties in the northwest region of Texas were very distant from the wind fields, but designated for public assistance. A possible reason is that they were affected indirectly through the accommodation of disaster evacuees or other minor effects. To guarantee that we are dealing with banks and counties that were clearly affected or clearly not affected by the hurricanes, we exclude these counties and banks. Consequently, we are left with a clean identification of affected and unaffected counties and banks. Figure 4. View largeDownload slide 2005 hurricane disaster areas. Notes: This figure shows counties in the U.S. Gulf Coast region and neighboring states that were affected by the 2005 hurricane season (Katrina, Rita, and Wilma). The dark-gray shaded area comprises counties that were eligible for individual and public disaster assistance, which we classify as affected counties. Banks with headquarters in these counties are classified as affected banks. The light-gray shaded area comprises counties that did not receive disaster assistance, which we classify as unaffected counties. Banks with headquarters in these counties are classified as unaffected banks. The white shaded area includes counties that were eligible only for public disaster assistance, which we exclude from the sample. Banks with headquarters in these counties are also excluded from the sample. Figure 4. View largeDownload slide 2005 hurricane disaster areas. Notes: This figure shows counties in the U.S. Gulf Coast region and neighboring states that were affected by the 2005 hurricane season (Katrina, Rita, and Wilma). The dark-gray shaded area comprises counties that were eligible for individual and public disaster assistance, which we classify as affected counties. Banks with headquarters in these counties are classified as affected banks. The light-gray shaded area comprises counties that did not receive disaster assistance, which we classify as unaffected counties. Banks with headquarters in these counties are classified as unaffected banks. The white shaded area includes counties that were eligible only for public disaster assistance, which we exclude from the sample. Banks with headquarters in these counties are also excluded from the sample. 3.2 Data Sources and Sample Description Our data come from several public sources. With regard to the impact of the 2005 hurricanes (Katrina, Rita, and Wilma) along the U.S. Gulf Coast region, we use data from the FEMA, as described above. Our bank data come primarily from the Statistics on Depository Institutions database of the Federal Deposit Insurance Corporation (FDIC).9 This data set includes quarterly balance sheet and income data of all FDIC-insured U.S. banks. We also use bank-level data on mortgage inquiries from bank customers, which are available from the Federal Financial Institutions Examination Council and reported by banks under the Home Mortgage Disclosure Act,10 to control for credit demand before and after the 2005 hurricanes. Data on banks’ lending transactions come from the U.S. SBA loan program.11 Finally, we use income and unemployment data at the county level from the Bureau of Economic Analysis (BEA)12 and the Bureau of Labor Statistics (BLS).13 3.2.a. Bank sample For our main bank-level analysis, we restrict the sample to banks located in the U.S. Gulf Coast region or neighboring states: Alabama, Arkansas, Florida, Georgia, Louisiana, Mississippi, Oklahoma, Tennessee, and Texas.14 This preliminary sample consists of 2,583 banks doing business at the end of the second quarter of 2005, that is, the quarter before the hurricanes hit the U.S. Gulf Coast. Moreover, for our baseline sample, we only consider banks that do not belong to a bank holding company, what we refer to as independent banks, which results in a preliminary sample of 706 banks.15 These independent banks share similar features with the community banks discussed by DeYoung, Hunter, and Udell (2004) and are a viable part of the U.S. banking sector. We only consider independent banks in our baseline regressions because, at first, we want to exclude effects from internal capital markets within bank holding groups that are due to capital allocations or implicit and explicit guarantees (Houston, James, and Marcus, 1997; Froot and Stein, 1998). We include both independent banks and banks that are part of a bank holding company in a set of extended regressions where we evaluate the effects of internal capital markets. Earlier studies that also use FDIC data point out that some of the data are erroneous or include rather atypical institutions. Therefore, similar to Berger and Bouwman (2009), we exclude banks that (1) have no outstanding commercial real estate loans or commercial and industrial loans; (2) have zero or negative equity capital; (3) hold assets below $25 million; or (4) hold consumer loans exceeding 50% of gross total assets. We also leave out atypical institutions with risk-based capital ratios above 40%, which represents five times the regulatory requirement of 8%. This reduces the sample to 532 banks. Since we want to exclude biases from newly founded banks, we require that the banks were in existence 2 years prior to the third quarter of 2005, which leaves us with a sample of 422 banks. Finally, as described in the previous section, we only consider banks that are located in a county that was clearly affected or clearly not affected by the 2005 hurricane season, which results in our final sample of 258 banks, of which 94 were affected and 164 were unaffected by the 2005 hurricane season. When we include banks that are part of a bank holding company in our analysis, the sample is extended to a total of 1,253 banks, of which 307 were affected and 946 were unaffected by the hurricanes. When we use loan transaction data from the SBA loan program, data are only available for a subset of banks, which restricts the sample to 337 independent and BHC banks, of which 73 were affected and 264 were unaffected by the hurricanes. 3.2.b. County sample We use county-level data to assess whether the structure of the banking system matters for local economic developments following Hurricane Katrina. This sample is based on a propensity score matching procedure, which requires that counties in both the treatment and control groups had similar characteristics prior to the 2005 hurricane season. In particular, we match affected and unaffected counties on the basis of their average total personal income in US$, the number of employed persons, the number of unemployed persons, and the unemployment rate in 2004 (1:1 nearest-neighbor matching with a caliper of 0.01). This procedure results in a sample of 176 counties in the U.S. Gulf Coast region and neighboring states (Alabama, Arkansas, Florida, Georgia, Louisiana, Mississippi, Oklahoma, Tennessee, and Texas). 3.3 Variable Descriptions and Summary Statistics Our main explanatory variable is the exogenous adverse shock on the asset quality of banks caused by Hurricanes Katrina, Rita, and Wilma in Q3 and Q4 of 2005. Asset quality reflects the “quantity of existing and potential credit risk associated with the loan and investment portfolios, other real estate owned, and other assets, as well as off-balance sheet transactions” (Federal Reserve Board, 2014). The 2005 hurricanes caused significant unexpected losses and increased credit risks for banks in affected regions. We are thus able to identify a causal relationship between asset quality and our dependent variables. Measures for asset quality, which are frequently used in the literature, are risk-weighted assets (Avery and Berger, 1991), the standard deviation of the return on assets, or the standard deviation of (unlevered) stock price returns (Flannery and Rangan, 2008; Gropp and Heider, 2010). In our study, we circumvent using these traditional measures, which cause concerns about endogeneity, because banks typically determine their asset quality and capital ratio simultaneously. The dependent variable that we use in our baseline regressions is a bank’s quarterly risk-based capital ratio.16 We thereby explore how banks adjust their risk-based capital ratios during the 2-year period following the 2005 hurricane season. The banks in our sample operate within a Basel I regulatory environment. Consequently, they can assign risk weights corresponding to five different categories that range from 0 to 100%. For example, U.S. government securities have a risk weight of zero, residential mortgage loans have a risk weight of 50%, and commercial and industrial loans have a risk weight of 100%. Banks are required to hold capital equal to at least 8% of risk-weighted assets. Other dependent variables used in this study, which allow us to explore the mechanisms banks use to adjust their capital ratios and business decisions, are total capital, risk-weighted assets, U.S. government securities, real estate loans, real estate commercial loans, reconstruction and development loans, consumer loans, and commercial and industrial loans. For some specifications, we use the regional unemployment rate from the county where a bank has its headquarters in order to control for time-varying differences in local economic conditions. In addition, we consider the volume of a bank’s approved small business loans as reported under the SBA loan program. These data reflect new lending transactions and thus complements the analysis of the balance sheet exposures of banks. In particular, we consider the total volume of a bank’s gross SBA lending as well as the volume net of government guarantees from the SBA program. Finally, we are interested in the role of the structure of local banking systems for economic development. We therefore consider per county j the share of banks belonging to a bank holding company (BHCsharej) and the average of banks’ risk-based capital ratios (CAPaveragej). Both measures are based on all banks with a branch in county j before the 2005 hurricane season (June 2005). The data on bank branches come from the Summary of Deposits database of the FDIC, and the data on county-level economic developments come from the BEA and the BLS. For measures of economic activity on the county level, we use total personal income, the number of employed and unemployed persons, and the unemployment rate per county.17 For a description of all variables used in this study, see Table I. Table I. Variable descriptions The source for all variables as well as their descriptions is the FDIC, if not stated otherwise. For more details, refer to https://www.fdic.gov/bank/statistical/. Variable name Description Bank-level variables Affected A dummy variable that separates banks with their headquarters in counties that were affected by the hurricanes (Affected = 1) and banks with their headquarters in counties that were unaffected (Affected = 0). Assets The sum of all assets owned by the institution including cash, loans, securities, bank premises, and other assets (FDIC variable name: asset). Bank size The natural logarithm of the number of full-time employees on the payroll of the bank and its subsidiaries (FDIC variable name: numemp). BHC Bank Holding Company bank: A Dummy variable that we assign a value of 0 if namehcr is blank and a value of 1 otherwise. CAP Risk-based capital ratio: Tier 1 capital and Tier 2 capital divided by the bank’s risk-weighted assets (FDIC variable name: rbcrwaj). For some banks, the nominator also includes Tier 3 capital allocated for market risk, net of all deductions. For details, see “Schedule RC-R—Regulatory Capital” of the FDIC. C&I Commercial and industrial loans: All loans excluding loans secured by real estate, loans to individuals, loans to depository institutions and foreign governments, loans to states and political subdivisions, and lease financing receivables (FDIC variable name: lnci). CON Consumer loans: Loans to individuals for household, family, and other personal expenditures including outstanding credit card balances and other secured and unsecured consumer loans (FDIC variable name: lncon). Credit demand The natural logarithm of the sum of accepted and denied loans. Source: Home Mortgage Disclosure Act (HMDA) database. GOV Government securities: Total U.S. Treasury securities plus U.S. Government agency and corporation obligations (FDIC variable name: scus). NPL Non-performing loans: Total assets past due 90 or more days and still accruing interest (FDIC variable name: p9asset). preCAP Pre-Hurricane Katrina capital ratio: A dummy variable which is zero if a bank’s average risk-based capital ratio during the eight quarters before the event is below the sample median (preCap = 0) and 1 otherwise (preCAP = 1). Source: Own calculations based on FDIC data. RE Real estate loans: Loans secured primarily by real estate, whether originated by the bank or purchased (FDIC variable name: lnre). RECO Commercial real estate loans. Nonresidential loans primarily secured by real estate (FDIC variable name: lnrenres). RECON Construction and development loans: Construction and land development loans secured by real estate held in domestic offices. This item includes loans for all property types under construction, as well as loans for land acquisition and development (FDIC variable name: lnrecons). RoA Return on assets: Net income after taxes and extraordinary items (annualized) to average total assets (FDIC variable name: roaptx). RWA Risk-weighted assets: Assets adjusted for risk-based capital definitions that comprise on-balance-sheet as well as off-balance-sheet items multiplied by risk weights that range from 0 to 100% under Basel I (FDIC variable name: rwaj). Total capital The sum of Tier 1 capital and Tier 2 capital (FDIC variable name: rbct1j+rbct2). SBA loans Small Business Administration loans: The amount of loans to small business. Source: U.S. SBA. Z-score The natural logarithm of the sum of a bank’s return on assets (roaptx) and its core capital ratio (eqv), standardized by the standard deviation (12 quarter rolling) of the bank’s return on assets. Source: Own calculations based on FDIC data. County-level variables BHCshare Share of BHC banks: The share of banks belonging to a bank holding company per county, based on all banks with a branch in a certain county before the 2005 Hurricane season (June 2005). Banks’ sum of deposits per county are used as weights. Source: Own calculations based on the Summary of Deposits database of the FDIC. CAPaverage Average bank capitalization: The average of banks’ risk-based capital ratios per county, based on all banks with a branch in a certain county before the 2005 Hurricane season (June 2005). Banks’ sum of deposits per county are used as weights. Source: Own calculations based on the Summary of Deposits database of the FDIC. Employed The number of employed persons per county and year. Source: Bureau of Labor Economics. Labor force The number of all employed and unemployed persons per county and year. Source: Bureau of Labor Economics. Personal income The income received by, or on behalf of all persons resident in a county per year. It includes wages and salaries, supplements to wages and salaries, proprietors’ income, dividends, interest, and rent, and personal current transfer receipts, less contributions for government social insurance. Source: Bureau of Economic Analysis. Unemployed The number of unemployed persons per county and year. Source: Bureau of Labor Economics. UR Unemployment rate: The percentage of the labor force that is unemployed per county and year. Source: Bureau of Labor Economics. Variable name Description Bank-level variables Affected A dummy variable that separates banks with their headquarters in counties that were affected by the hurricanes (Affected = 1) and banks with their headquarters in counties that were unaffected (Affected = 0). Assets The sum of all assets owned by the institution including cash, loans, securities, bank premises, and other assets (FDIC variable name: asset). Bank size The natural logarithm of the number of full-time employees on the payroll of the bank and its subsidiaries (FDIC variable name: numemp). BHC Bank Holding Company bank: A Dummy variable that we assign a value of 0 if namehcr is blank and a value of 1 otherwise. CAP Risk-based capital ratio: Tier 1 capital and Tier 2 capital divided by the bank’s risk-weighted assets (FDIC variable name: rbcrwaj). For some banks, the nominator also includes Tier 3 capital allocated for market risk, net of all deductions. For details, see “Schedule RC-R—Regulatory Capital” of the FDIC. C&I Commercial and industrial loans: All loans excluding loans secured by real estate, loans to individuals, loans to depository institutions and foreign governments, loans to states and political subdivisions, and lease financing receivables (FDIC variable name: lnci). CON Consumer loans: Loans to individuals for household, family, and other personal expenditures including outstanding credit card balances and other secured and unsecured consumer loans (FDIC variable name: lncon). Credit demand The natural logarithm of the sum of accepted and denied loans. Source: Home Mortgage Disclosure Act (HMDA) database. GOV Government securities: Total U.S. Treasury securities plus U.S. Government agency and corporation obligations (FDIC variable name: scus). NPL Non-performing loans: Total assets past due 90 or more days and still accruing interest (FDIC variable name: p9asset). preCAP Pre-Hurricane Katrina capital ratio: A dummy variable which is zero if a bank’s average risk-based capital ratio during the eight quarters before the event is below the sample median (preCap = 0) and 1 otherwise (preCAP = 1). Source: Own calculations based on FDIC data. RE Real estate loans: Loans secured primarily by real estate, whether originated by the bank or purchased (FDIC variable name: lnre). RECO Commercial real estate loans. Nonresidential loans primarily secured by real estate (FDIC variable name: lnrenres). RECON Construction and development loans: Construction and land development loans secured by real estate held in domestic offices. This item includes loans for all property types under construction, as well as loans for land acquisition and development (FDIC variable name: lnrecons). RoA Return on assets: Net income after taxes and extraordinary items (annualized) to average total assets (FDIC variable name: roaptx). RWA Risk-weighted assets: Assets adjusted for risk-based capital definitions that comprise on-balance-sheet as well as off-balance-sheet items multiplied by risk weights that range from 0 to 100% under Basel I (FDIC variable name: rwaj). Total capital The sum of Tier 1 capital and Tier 2 capital (FDIC variable name: rbct1j+rbct2). SBA loans Small Business Administration loans: The amount of loans to small business. Source: U.S. SBA. Z-score The natural logarithm of the sum of a bank’s return on assets (roaptx) and its core capital ratio (eqv), standardized by the standard deviation (12 quarter rolling) of the bank’s return on assets. Source: Own calculations based on FDIC data. County-level variables BHCshare Share of BHC banks: The share of banks belonging to a bank holding company per county, based on all banks with a branch in a certain county before the 2005 Hurricane season (June 2005). Banks’ sum of deposits per county are used as weights. Source: Own calculations based on the Summary of Deposits database of the FDIC. CAPaverage Average bank capitalization: The average of banks’ risk-based capital ratios per county, based on all banks with a branch in a certain county before the 2005 Hurricane season (June 2005). Banks’ sum of deposits per county are used as weights. Source: Own calculations based on the Summary of Deposits database of the FDIC. Employed The number of employed persons per county and year. Source: Bureau of Labor Economics. Labor force The number of all employed and unemployed persons per county and year. Source: Bureau of Labor Economics. Personal income The income received by, or on behalf of all persons resident in a county per year. It includes wages and salaries, supplements to wages and salaries, proprietors’ income, dividends, interest, and rent, and personal current transfer receipts, less contributions for government social insurance. Source: Bureau of Economic Analysis. Unemployed The number of unemployed persons per county and year. Source: Bureau of Labor Economics. UR Unemployment rate: The percentage of the labor force that is unemployed per county and year. Source: Bureau of Labor Economics. Table I. Variable descriptions The source for all variables as well as their descriptions is the FDIC, if not stated otherwise. For more details, refer to https://www.fdic.gov/bank/statistical/. Variable name Description Bank-level variables Affected A dummy variable that separates banks with their headquarters in counties that were affected by the hurricanes (Affected = 1) and banks with their headquarters in counties that were unaffected (Affected = 0). Assets The sum of all assets owned by the institution including cash, loans, securities, bank premises, and other assets (FDIC variable name: asset). Bank size The natural logarithm of the number of full-time employees on the payroll of the bank and its subsidiaries (FDIC variable name: numemp). BHC Bank Holding Company bank: A Dummy variable that we assign a value of 0 if namehcr is blank and a value of 1 otherwise. CAP Risk-based capital ratio: Tier 1 capital and Tier 2 capital divided by the bank’s risk-weighted assets (FDIC variable name: rbcrwaj). For some banks, the nominator also includes Tier 3 capital allocated for market risk, net of all deductions. For details, see “Schedule RC-R—Regulatory Capital” of the FDIC. C&I Commercial and industrial loans: All loans excluding loans secured by real estate, loans to individuals, loans to depository institutions and foreign governments, loans to states and political subdivisions, and lease financing receivables (FDIC variable name: lnci). CON Consumer loans: Loans to individuals for household, family, and other personal expenditures including outstanding credit card balances and other secured and unsecured consumer loans (FDIC variable name: lncon). Credit demand The natural logarithm of the sum of accepted and denied loans. Source: Home Mortgage Disclosure Act (HMDA) database. GOV Government securities: Total U.S. Treasury securities plus U.S. Government agency and corporation obligations (FDIC variable name: scus). NPL Non-performing loans: Total assets past due 90 or more days and still accruing interest (FDIC variable name: p9asset). preCAP Pre-Hurricane Katrina capital ratio: A dummy variable which is zero if a bank’s average risk-based capital ratio during the eight quarters before the event is below the sample median (preCap = 0) and 1 otherwise (preCAP = 1). Source: Own calculations based on FDIC data. RE Real estate loans: Loans secured primarily by real estate, whether originated by the bank or purchased (FDIC variable name: lnre). RECO Commercial real estate loans. Nonresidential loans primarily secured by real estate (FDIC variable name: lnrenres). RECON Construction and development loans: Construction and land development loans secured by real estate held in domestic offices. This item includes loans for all property types under construction, as well as loans for land acquisition and development (FDIC variable name: lnrecons). RoA Return on assets: Net income after taxes and extraordinary items (annualized) to average total assets (FDIC variable name: roaptx). RWA Risk-weighted assets: Assets adjusted for risk-based capital definitions that comprise on-balance-sheet as well as off-balance-sheet items multiplied by risk weights that range from 0 to 100% under Basel I (FDIC variable name: rwaj). Total capital The sum of Tier 1 capital and Tier 2 capital (FDIC variable name: rbct1j+rbct2). SBA loans Small Business Administration loans: The amount of loans to small business. Source: U.S. SBA. Z-score The natural logarithm of the sum of a bank’s return on assets (roaptx) and its core capital ratio (eqv), standardized by the standard deviation (12 quarter rolling) of the bank’s return on assets. Source: Own calculations based on FDIC data. County-level variables BHCshare Share of BHC banks: The share of banks belonging to a bank holding company per county, based on all banks with a branch in a certain county before the 2005 Hurricane season (June 2005). Banks’ sum of deposits per county are used as weights. Source: Own calculations based on the Summary of Deposits database of the FDIC. CAPaverage Average bank capitalization: The average of banks’ risk-based capital ratios per county, based on all banks with a branch in a certain county before the 2005 Hurricane season (June 2005). Banks’ sum of deposits per county are used as weights. Source: Own calculations based on the Summary of Deposits database of the FDIC. Employed The number of employed persons per county and year. Source: Bureau of Labor Economics. Labor force The number of all employed and unemployed persons per county and year. Source: Bureau of Labor Economics. Personal income The income received by, or on behalf of all persons resident in a county per year. It includes wages and salaries, supplements to wages and salaries, proprietors’ income, dividends, interest, and rent, and personal current transfer receipts, less contributions for government social insurance. Source: Bureau of Economic Analysis. Unemployed The number of unemployed persons per county and year. Source: Bureau of Labor Economics. UR Unemployment rate: The percentage of the labor force that is unemployed per county and year. Source: Bureau of Labor Economics. Variable name Description Bank-level variables Affected A dummy variable that separates banks with their headquarters in counties that were affected by the hurricanes (Affected = 1) and banks with their headquarters in counties that were unaffected (Affected = 0). Assets The sum of all assets owned by the institution including cash, loans, securities, bank premises, and other assets (FDIC variable name: asset). Bank size The natural logarithm of the number of full-time employees on the payroll of the bank and its subsidiaries (FDIC variable name: numemp). BHC Bank Holding Company bank: A Dummy variable that we assign a value of 0 if namehcr is blank and a value of 1 otherwise. CAP Risk-based capital ratio: Tier 1 capital and Tier 2 capital divided by the bank’s risk-weighted assets (FDIC variable name: rbcrwaj). For some banks, the nominator also includes Tier 3 capital allocated for market risk, net of all deductions. For details, see “Schedule RC-R—Regulatory Capital” of the FDIC. C&I Commercial and industrial loans: All loans excluding loans secured by real estate, loans to individuals, loans to depository institutions and foreign governments, loans to states and political subdivisions, and lease financing receivables (FDIC variable name: lnci). CON Consumer loans: Loans to individuals for household, family, and other personal expenditures including outstanding credit card balances and other secured and unsecured consumer loans (FDIC variable name: lncon). Credit demand The natural logarithm of the sum of accepted and denied loans. Source: Home Mortgage Disclosure Act (HMDA) database. GOV Government securities: Total U.S. Treasury securities plus U.S. Government agency and corporation obligations (FDIC variable name: scus). NPL Non-performing loans: Total assets past due 90 or more days and still accruing interest (FDIC variable name: p9asset). preCAP Pre-Hurricane Katrina capital ratio: A dummy variable which is zero if a bank’s average risk-based capital ratio during the eight quarters before the event is below the sample median (preCap = 0) and 1 otherwise (preCAP = 1). Source: Own calculations based on FDIC data. RE Real estate loans: Loans secured primarily by real estate, whether originated by the bank or purchased (FDIC variable name: lnre). RECO Commercial real estate loans. Nonresidential loans primarily secured by real estate (FDIC variable name: lnrenres). RECON Construction and development loans: Construction and land development loans secured by real estate held in domestic offices. This item includes loans for all property types under construction, as well as loans for land acquisition and development (FDIC variable name: lnrecons). RoA Return on assets: Net income after taxes and extraordinary items (annualized) to average total assets (FDIC variable name: roaptx). RWA Risk-weighted assets: Assets adjusted for risk-based capital definitions that comprise on-balance-sheet as well as off-balance-sheet items multiplied by risk weights that range from 0 to 100% under Basel I (FDIC variable name: rwaj). Total capital The sum of Tier 1 capital and Tier 2 capital (FDIC variable name: rbct1j+rbct2). SBA loans Small Business Administration loans: The amount of loans to small business. Source: U.S. SBA. Z-score The natural logarithm of the sum of a bank’s return on assets (roaptx) and its core capital ratio (eqv), standardized by the standard deviation (12 quarter rolling) of the bank’s return on assets. Source: Own calculations based on FDIC data. County-level variables BHCshare Share of BHC banks: The share of banks belonging to a bank holding company per county, based on all banks with a branch in a certain county before the 2005 Hurricane season (June 2005). Banks’ sum of deposits per county are used as weights. Source: Own calculations based on the Summary of Deposits database of the FDIC. CAPaverage Average bank capitalization: The average of banks’ risk-based capital ratios per county, based on all banks with a branch in a certain county before the 2005 Hurricane season (June 2005). Banks’ sum of deposits per county are used as weights. Source: Own calculations based on the Summary of Deposits database of the FDIC. Employed The number of employed persons per county and year. Source: Bureau of Labor Economics. Labor force The number of all employed and unemployed persons per county and year. Source: Bureau of Labor Economics. Personal income The income received by, or on behalf of all persons resident in a county per year. It includes wages and salaries, supplements to wages and salaries, proprietors’ income, dividends, interest, and rent, and personal current transfer receipts, less contributions for government social insurance. Source: Bureau of Economic Analysis. Unemployed The number of unemployed persons per county and year. Source: Bureau of Labor Economics. UR Unemployment rate: The percentage of the labor force that is unemployed per county and year. Source: Bureau of Labor Economics. Summary statistics are provided in Table II. All statistics refer to the average values of the 2-year period prior to the 2005 hurricane season, that is, Q3 2003 to Q2 2005 for the bank sample based on quarterly data, and 2003 and 2004 for the county sample based on yearly data. The table reports mean values and standard deviations separately for the groups of affected banks (or counties) and unaffected banks (or counties), as well as normalized differences, which we discuss in more detail below. Table II. Summary statistics This table reports summary statistics for all variables over the period 2 years before the 2005 hurricane season. The upper panel shows mean values and standard deviations for independent banks (i.e., banks that do not belong to a BHC) with headquarters in counties that were affected or unaffected by the hurricanes. The sample includes banks in Alabama, Arkansas, Florida, Georgia, Louisiana, Mississippi, Oklahoma, Tennessee, and Texas. Descriptive statistics for SBA loans (denoted by *) refer to the sample used in Section 4.5, which includes 73 affected and 264 unaffected independent and BHC banks. The bottom panel provides descriptive statistics for the analysis on county level. The last column shows the normalized differences (Norm. Diff.) according to Imbens and Wooldridge (2009) and compares differences between banks/counties that were affected and banks/counties that were not affected [(1) versus (2)]. As a rule of thumb, groups are regarded as sufficiently equal and adequate for linear regression methods if normalized differences are largely in the range of ± 0.25. A description of all variables is given in Table I. (1) Affected (2) Unaffected Norm. Diff. Bank sample Mean SD Mean SD (1) versus (2) Assets ($millions) 424.4568 1,122.7046 244.8142 836.5487 0.1283 Bank size 3.8657 1.0052 3.5775 0.8791 0.2158 C&I/assets 0.0851 0.0775 0.0866 0.0642 −0.0146 CON/assets 0.0483 0.0485 0.0732 0.0603 −0.3207 GOV/assets 0.1747 0.1408 0.1703 0.1385 0.0222 Loans/assets 0.6466 0.1713 0.6706 0.1543 −0.1042 NPL/assets 0.0013 0.0029 0.0016 0.0030 −0.0657 RE/assets 0.5010 0.2058 0.4861 0.1860 0.0536 RECO/assets 0.1671 0.1284 0.1486 0.1123 0.1082 RECONS loans/assets 0.0663 0.0639 0.0708 0.0788 −0.0445 Risk-based capital ratio (CAP) 0.1701 0.0576 0.1733 0.0630 −0.0368 Risk-weighted assets/assets 0.6426 0.1214 0.6830 0.1208 −0.2356 RoA 0.0094 0.0094 0.0097 0.0096 −0.0207 SBA loans ($000)* 1,608.8426 2,890.6289 1,741.0798 5,049.5701 −0.0227 SBA loans net of guarantees ($000)* 402.8130 721.2119 455.4962 1,338.1543 −0.0347 Total capital/assets 0.1053 0.0276 0.1136 0.0306 −0.2027 Z-score 3.6972 0.8996 3.7628 0.8841 −0.0520 Credit demand 9.7183 1.9764 9.5051 1.6228 0.0834 Unemployment rate (UR) 0.0563 0.0123 0.0542 0.0146 0.1070 Number of banks 94 164 Number of observations 752 1312 (1) Affected (2) Unaffected Norm. Diff. County sample Mean SD Mean SD (1) versus (2) BHCshare (before demeaning) 0.8495 0.2241 0.8693 0.2197 −0.0631 CAPaverage (before demeaning) 0.1533 0.0391 0.1513 0.0376 0.0363 Employed (000) 43.6428 59.2071 54.9142 136.0716 −0.0760 Labor force (000) 46.2746 62.5010 57.8899 143.0961 −0.0744 Unemployed (000) 2.6317 3.4257 2.9757 7.1244 −0.0435 Personal income ($million) 2,418.0714 3,340.6302 2,955.3050 7,277.4510 −0.0671 Unemployment rate (UR) 0.0643 0.0166 0.0660 0.0198 −0.0691 Number of counties 88 88 Number of observations 176 176 (1) Affected (2) Unaffected Norm. Diff. Bank sample Mean SD Mean SD (1) versus (2) Assets ($millions) 424.4568 1,122.7046 244.8142 836.5487 0.1283 Bank size 3.8657 1.0052 3.5775 0.8791 0.2158 C&I/assets 0.0851 0.0775 0.0866 0.0642 −0.0146 CON/assets 0.0483 0.0485 0.0732 0.0603 −0.3207 GOV/assets 0.1747 0.1408 0.1703 0.1385 0.0222 Loans/assets 0.6466 0.1713 0.6706 0.1543 −0.1042 NPL/assets 0.0013 0.0029 0.0016 0.0030 −0.0657 RE/assets 0.5010 0.2058 0.4861 0.1860 0.0536 RECO/assets 0.1671 0.1284 0.1486 0.1123 0.1082 RECONS loans/assets 0.0663 0.0639 0.0708 0.0788 −0.0445 Risk-based capital ratio (CAP) 0.1701 0.0576 0.1733 0.0630 −0.0368 Risk-weighted assets/assets 0.6426 0.1214 0.6830 0.1208 −0.2356 RoA 0.0094 0.0094 0.0097 0.0096 −0.0207 SBA loans ($000)* 1,608.8426 2,890.6289 1,741.0798 5,049.5701 −0.0227 SBA loans net of guarantees ($000)* 402.8130 721.2119 455.4962 1,338.1543 −0.0347 Total capital/assets 0.1053 0.0276 0.1136 0.0306 −0.2027 Z-score 3.6972 0.8996 3.7628 0.8841 −0.0520 Credit demand 9.7183 1.9764 9.5051 1.6228 0.0834 Unemployment rate (UR) 0.0563 0.0123 0.0542 0.0146 0.1070 Number of banks 94 164 Number of observations 752 1312 (1) Affected (2) Unaffected Norm. Diff. County sample Mean SD Mean SD (1) versus (2) BHCshare (before demeaning) 0.8495 0.2241 0.8693 0.2197 −0.0631 CAPaverage (before demeaning) 0.1533 0.0391 0.1513 0.0376 0.0363 Employed (000) 43.6428 59.2071 54.9142 136.0716 −0.0760 Labor force (000) 46.2746 62.5010 57.8899 143.0961 −0.0744 Unemployed (000) 2.6317 3.4257 2.9757 7.1244 −0.0435 Personal income ($million) 2,418.0714 3,340.6302 2,955.3050 7,277.4510 −0.0671 Unemployment rate (UR) 0.0643 0.0166 0.0660 0.0198 −0.0691 Number of counties 88 88 Number of observations 176 176 Table II. Summary statistics This table reports summary statistics for all variables over the period 2 years before the 2005 hurricane season. The upper panel shows mean values and standard deviations for independent banks (i.e., banks that do not belong to a BHC) with headquarters in counties that were affected or unaffected by the hurricanes. The sample includes banks in Alabama, Arkansas, Florida, Georgia, Louisiana, Mississippi, Oklahoma, Tennessee, and Texas. Descriptive statistics for SBA loans (denoted by *) refer to the sample used in Section 4.5, which includes 73 affected and 264 unaffected independent and BHC banks. The bottom panel provides descriptive statistics for the analysis on county level. The last column shows the normalized differences (Norm. Diff.) according to Imbens and Wooldridge (2009) and compares differences between banks/counties that were affected and banks/counties that were not affected [(1) versus (2)]. As a rule of thumb, groups are regarded as sufficiently equal and adequate for linear regression methods if normalized differences are largely in the range of ± 0.25. A description of all variables is given in Table I. (1) Affected (2) Unaffected Norm. Diff. Bank sample Mean SD Mean SD (1) versus (2) Assets ($millions) 424.4568 1,122.7046 244.8142 836.5487 0.1283 Bank size 3.8657 1.0052 3.5775 0.8791 0.2158 C&I/assets 0.0851 0.0775 0.0866 0.0642 −0.0146 CON/assets 0.0483 0.0485 0.0732 0.0603 −0.3207 GOV/assets 0.1747 0.1408 0.1703 0.1385 0.0222 Loans/assets 0.6466 0.1713 0.6706 0.1543 −0.1042 NPL/assets 0.0013 0.0029 0.0016 0.0030 −0.0657 RE/assets 0.5010 0.2058 0.4861 0.1860 0.0536 RECO/assets 0.1671 0.1284 0.1486 0.1123 0.1082 RECONS loans/assets 0.0663 0.0639 0.0708 0.0788 −0.0445 Risk-based capital ratio (CAP) 0.1701 0.0576 0.1733 0.0630 −0.0368 Risk-weighted assets/assets 0.6426 0.1214 0.6830 0.1208 −0.2356 RoA 0.0094 0.0094 0.0097 0.0096 −0.0207 SBA loans ($000)* 1,608.8426 2,890.6289 1,741.0798 5,049.5701 −0.0227 SBA loans net of guarantees ($000)* 402.8130 721.2119 455.4962 1,338.1543 −0.0347 Total capital/assets 0.1053 0.0276 0.1136 0.0306 −0.2027 Z-score 3.6972 0.8996 3.7628 0.8841 −0.0520 Credit demand 9.7183 1.9764 9.5051 1.6228 0.0834 Unemployment rate (UR) 0.0563 0.0123 0.0542 0.0146 0.1070 Number of banks 94 164 Number of observations 752 1312 (1) Affected (2) Unaffected Norm. Diff. County sample Mean SD Mean SD (1) versus (2) BHCshare (before demeaning) 0.8495 0.2241 0.8693 0.2197 −0.0631 CAPaverage (before demeaning) 0.1533 0.0391 0.1513 0.0376 0.0363 Employed (000) 43.6428 59.2071 54.9142 136.0716 −0.0760 Labor force (000) 46.2746 62.5010 57.8899 143.0961 −0.0744 Unemployed (000) 2.6317 3.4257 2.9757 7.1244 −0.0435 Personal income ($million) 2,418.0714 3,340.6302 2,955.3050 7,277.4510 −0.0671 Unemployment rate (UR) 0.0643 0.0166 0.0660 0.0198 −0.0691 Number of counties 88 88 Number of observations 176 176 (1) Affected (2) Unaffected Norm. Diff. Bank sample Mean SD Mean SD (1) versus (2) Assets ($millions) 424.4568 1,122.7046 244.8142 836.5487 0.1283 Bank size 3.8657 1.0052 3.5775 0.8791 0.2158 C&I/assets 0.0851 0.0775 0.0866 0.0642 −0.0146 CON/assets 0.0483 0.0485 0.0732 0.0603 −0.3207 GOV/assets 0.1747 0.1408 0.1703 0.1385 0.0222 Loans/assets 0.6466 0.1713 0.6706 0.1543 −0.1042 NPL/assets 0.0013 0.0029 0.0016 0.0030 −0.0657 RE/assets 0.5010 0.2058 0.4861 0.1860 0.0536 RECO/assets 0.1671 0.1284 0.1486 0.1123 0.1082 RECONS loans/assets 0.0663 0.0639 0.0708 0.0788 −0.0445 Risk-based capital ratio (CAP) 0.1701 0.0576 0.1733 0.0630 −0.0368 Risk-weighted assets/assets 0.6426 0.1214 0.6830 0.1208 −0.2356 RoA 0.0094 0.0094 0.0097 0.0096 −0.0207 SBA loans ($000)* 1,608.8426 2,890.6289 1,741.0798 5,049.5701 −0.0227 SBA loans net of guarantees ($000)* 402.8130 721.2119 455.4962 1,338.1543 −0.0347 Total capital/assets 0.1053 0.0276 0.1136 0.0306 −0.2027 Z-score 3.6972 0.8996 3.7628 0.8841 −0.0520 Credit demand 9.7183 1.9764 9.5051 1.6228 0.0834 Unemployment rate (UR) 0.0563 0.0123 0.0542 0.0146 0.1070 Number of banks 94 164 Number of observations 752 1312 (1) Affected (2) Unaffected Norm. Diff. County sample Mean SD Mean SD (1) versus (2) BHCshare (before demeaning) 0.8495 0.2241 0.8693 0.2197 −0.0631 CAPaverage (before demeaning) 0.1533 0.0391 0.1513 0.0376 0.0363 Employed (000) 43.6428 59.2071 54.9142 136.0716 −0.0760 Labor force (000) 46.2746 62.5010 57.8899 143.0961 −0.0744 Unemployed (000) 2.6317 3.4257 2.9757 7.1244 −0.0435 Personal income ($million) 2,418.0714 3,340.6302 2,955.3050 7,277.4510 −0.0671 Unemployment rate (UR) 0.0643 0.0166 0.0660 0.0198 −0.0691 Number of counties 88 88 Number of observations 176 176 3.4 Similarity between Treatment and Control Group It is important for the validity of the difference-in-difference estimation that banks in our treatment group (affected banks) and banks in our control group (unaffected banks) have similar characteristics before the event. As suggested by Imbens and Wooldridge (2009), we report summary statistics with normalized differences to compare the similarity between both groups with regard to important bank characteristics.18  As a rule of thumb, groups are regarded as sufficiently equal and adequate for linear regression methods if normalized differences are largely in the range of ± 0.25. The summary statistics reported in Table II confirm that the groups of affected and unaffected banks are relatively similar prior to the event. In particular, banks in both groups hold, on average, similar levels of risk-based capital ratios of around 17% during the 2 years prior to the 2005 hurricane season. Normalized differences of 0.037 are clearly in the range of ± 0.25. Note that the level of 17% substantially exceeds the regulatory minimum of 8%. This observation is in line with Flannery and Rangan (2008), who also report relatively high ratios for U.S. banks. We also find for all reported bank-level variables, with the exception of consumer loans, that normalized differences are in the range of ± 0.25. At the county level, we find that all statistics are relatively similar for affected and unaffected counties. Normalized differences are clearly in the range of ± 0.25. Overall, we find that characteristics of affected and unaffected banks and counties are similar prior to the event and, hence, that the treatment and control groups are well suited for our analysis. 4. Empirical Analysis In this section, we explore how a deterioration in a bank’s asset quality due to the 2005 hurricane season affects the bank’s capital ratios, asset allocation, and lending decisions in the aftermath of the hurricanes. In our analysis, we are interested in whether and how different pre-Katrina capital levels affect a bank’s behavior. Our analysis also includes an assessment of the role of the bank’s organizational structure for these developments. Furthermore, we are interested in the economic impact of the disaster, in particular how unemployment and growth developed in the post-Katrina environment in counties with different banking system structures. Anecdotal evidence of the deterioration in a bank’s asset quality following the disaster is provided by the FDIC and rating agencies, as noted in the Introduction of this paper (see, e.g., the quote from the FDIC). Empirical evidence of the adverse short-term effects of the hurricanes on bank profitability and bank stability is provided in the Appendix of this paper, highlighting the adverse effects of the hurricane season on both bank stability and profitability.19 4.1 Do Affected Banks Adjust Their Risk-Based Capital Ratios? In this section, we explore whether the risk-based capital ratios of independent banks change after these banks experience an adverse shock on their asset quality as a result of the 2005 hurricane season. 4.1.a. Baseline model For the empirical analysis, we need to consider potential parallel economic and industry-wide factors that affect all banks independent of the shock. Another concern is that unobservable bank characteristics might influence the analysis. To account for both aspects, we use a difference-in-difference estimation technique with time- and bank-fixed effects. We thus formally estimate the following equation with a fixed effects OLS model for a sample period of 2 years around the 2005 hurricane season (Q3 and Q4 2005): CAPit=νi+τt+β1(Eventt×Affectedi)+ϵit. (1) The dependent variable is the risk-based capital ratio (CAP) of bank i at time t. The terms νi and τt represent bank-fixed effects and yearly time-fixed effects, respectively. The variable Eventt is a time dummy with a value of 0 for the eight quarters before the hurricanes (t  ≤  Q2 2005) and a value of 1 for the eight quarters after the hurricanes (t ≥  Q1 2006). The variable Affectedi is a dummy variable that is 1 if bank i is located in a county classified by FEMA as eligible for “public and private disaster assistance” and thus belongs to the treatment group, and zero otherwise (for the control group). Hence, the interaction term Eventt × Affectedi is one if both the variable Eventt and the variable Affectedi amount to 1, and 0 otherwise. The corresponding coefficient β1 is the main interest. It shows how affected banks adjust their CAPs after the event compared with the control group. The single terms Eventt and Affectedi do not appear in the equation because they are absorbed by the time- and bank-fixed effects, respectively. The term ϵit represents the idiosyncratic error term. Standard errors are clustered at the bank level. For robustness, we reestimate our baseline estimation with three alternative specifications. First, we estimate Equation (1) without bank-fixed effects. The variable Affectedi, which otherwise interferes with bank-fixed effects, then enters the equation. Second, we consider potential concerns that a shortfall in credit demand in affected regions may drive our results. Technically, such a shortfall may lead to lower risk-weighted assets and consequently higher CAPs of the affected banks. However, a shortfall in credit demand is unlikely because of reconstruction activities. As stated by the Federal Reserve Bank of Atlanta (2005), credit demand was rather expected to increase in affected regions in the aftermath of the 2005 hurricanes. Nevertheless, we add a control variable for the banks’ credit demand in a robustness specification. The general difficulty for such a control variable is that it needs to disentangle credit demand from credit supply. Therefore, a bank’s reported loan volume is not suitable. However, we can use data reported by banks under the Home Mortgage Disclosure Act to build a proxy for credit demand. In particular, banks are required under this act to report the volume of all mortgage applications on a yearly basis. We use this data to calculate Credit demand per bank and year as the log of the dollar volume of the mortgage applications (accepted and denied mortgages) of each bank. We then include the variable Credit demand in Equation (1). Note that this specification reduces our sample from 258 to 182 banks, because the data on credit demand are not available for all banks. Third, we estimate Equation (1) with further control variables which are common in the banking literature. In particular, we add the return on assets (RoA), the ratio of non-performing loans to assets (NPL/Assets), and the log of the total number of employees (Bank size).20 Note that these control variables only matter for the estimation to the degree that they are time variant because they are otherwise already included in the bank-fixed effects. Furthermore, in order to capture differences in local economic developments, we use quarterly unemployment rates (UR) at the county level as an additional time-varying control variable. 4.1.b. Baseline results We present our baseline results in Table III. Column (1) shows the difference-in-difference estimation with bank-fixed effects and without further covariates, as reflected in Equation (1). With regard to our main variable of interest, the interaction term Event × Affected, we observe a positive and significant coefficient that shows that affected banks increase their capital ratios after the hurricane compared with the control group. This effect is also economically significant. The risk-based capital ratios of affected banks increase on average by 1.04 percentage points, as shown by the coefficient of the interaction term. Column (2) shows results for our baseline estimation of Equation (1) without bank-fixed effects. The results remain robust and confirm the relatively higher CAPs of banks after the event. The average effect of the hurricanes on banks’ CAPs, which is reflected in the coefficient of the interaction term, is 1.49 percentage points. Next, we again include bank-fixed effects, as well as a proxy for credit demand in the regression, and we find that results remain intact. The effect on banks’ capital ratios, as reflected in a coefficient of 1.24 percentage points, is again in the range of estimation results in Column (1). Finally, we add bank characteristics that are considered to be relevant for banks’ capital ratios, as well as the UR in a bank’s home county, in order to control for economic developments. We find that the banks’ CAPs decrease in bank size, but are not significantly affected by the other new covariates. Most importantly, the coefficient of the interaction term, which is our main interest, remains in the same range as before. Table III. Baseline results These regressions explore how banks adjust their total risk-based capital ratios following the 2005 hurricane season, as specified in Equation (1). The sample includes quarterly data for all independent banks (not part of a bank holding company) in Alabama, Arkansas, Florida, Georgia, Louisiana, Mississippi, Oklahoma, Tennessee, and Texas over the period of ± 2 years around the 2005 hurricane season (Q3 2003–Q4 2007). Event is a dummy variable that is 0 for the pre-hurricane period and 1 after the hurricane season. Affected is a dummy variable that separates banks located in counties that were affected by the hurricanes (Affected = 1) and banks located in counties that were unaffected (Affected = 0). Event×Affected is an interaction term for the variables Event and Affected. Credit demand is the log of a bank’s volume of loan applications as reported under the Home Mortgage Disclosure Act. RoA is a bank’s return on assets. NPL/assets represent a bank’s non-performing loans over assets. Bank size is the natural logarithm of a bank’s number of employees. UR represents the unemployment rate for the county where a bank’s headquarters is based. A description of all variables is given in Table I. The models include bank-fixed effects (bank FE) and year-fixed effects (time FE) as stated in the table. Standard errors are clustered at the bank level and reported in parentheses. The ***, **, and * stand for significant coefficients at the 1%, 5%, and 10% levels, respectively. Risk-based capital ratio (CAP) (1) (2) (3) (4) Event×Affected 0.0104** 0.0149*** 0.0124** 0.0131** (0.0045) (0.0051) (0.0053) (0.0053) Affected −0.0031 (0.0075) Credit demand −0.0062*** −0.0052*** (0.0018) (0.0017) RoA −0.0005 (0.0027) NPL/assets −0.0036 (0.2624) Bank size −0.0237** (0.0093) UR 0.0724 (0.1379) Bank FE Yes No Yes Yes Time FE Yes Yes Yes Yes Number of banks 258 258 182 182 Number of observations 3,795 3,795 2,415 2,415 Adjusted R2 0.8670 0.0037 0.8576 0.8607 Within R2 0.0127 0.0043 0.0399 0.0625 Risk-based capital ratio (CAP) (1) (2) (3) (4) Event×Affected 0.0104** 0.0149*** 0.0124** 0.0131** (0.0045) (0.0051) (0.0053) (0.0053) Affected −0.0031 (0.0075) Credit demand −0.0062*** −0.0052*** (0.0018) (0.0017) RoA −0.0005 (0.0027) NPL/assets −0.0036 (0.2624) Bank size −0.0237** (0.0093) UR 0.0724 (0.1379) Bank FE Yes No Yes Yes Time FE Yes Yes Yes Yes Number of banks 258 258 182 182 Number of observations 3,795 3,795 2,415 2,415 Adjusted R2 0.8670 0.0037 0.8576 0.8607 Within R2 0.0127 0.0043 0.0399 0.0625 Table III. Baseline results These regressions explore how banks adjust their total risk-based capital ratios following the 2005 hurricane season, as specified in Equation (1). The sample includes quarterly data for all independent banks (not part of a bank holding company) in Alabama, Arkansas, Florida, Georgia, Louisiana, Mississippi, Oklahoma, Tennessee, and Texas over the period of ± 2 years around the 2005 hurricane season (Q3 2003–Q4 2007). Event is a dummy variable that is 0 for the pre-hurricane period and 1 after the hurricane season. Affected is a dummy variable that separates banks located in counties that were affected by the hurricanes (Affected = 1) and banks located in counties that were unaffected (Affected = 0). Event×Affected is an interaction term for the variables Event and Affected. Credit demand is the log of a bank’s volume of loan applications as reported under the Home Mortgage Disclosure Act. RoA is a bank’s return on assets. NPL/assets represent a bank’s non-performing loans over assets. Bank size is the natural logarithm of a bank’s number of employees. UR represents the unemployment rate for the county where a bank’s headquarters is based. A description of all variables is given in Table I. The models include bank-fixed effects (bank FE) and year-fixed effects (time FE) as stated in the table. Standard errors are clustered at the bank level and reported in parentheses. The ***, **, and * stand for significant coefficients at the 1%, 5%, and 10% levels, respectively. Risk-based capital ratio (CAP) (1) (2) (3) (4) Event×Affected 0.0104** 0.0149*** 0.0124** 0.0131** (0.0045) (0.0051) (0.0053) (0.0053) Affected −0.0031 (0.0075) Credit demand −0.0062*** −0.0052*** (0.0018) (0.0017) RoA −0.0005 (0.0027) NPL/assets −0.0036 (0.2624) Bank size −0.0237** (0.0093) UR 0.0724 (0.1379) Bank FE Yes No Yes Yes Time FE Yes Yes Yes Yes Number of banks 258 258 182 182 Number of observations 3,795 3,795 2,415 2,415 Adjusted R2 0.8670 0.0037 0.8576 0.8607 Within R2 0.0127 0.0043 0.0399 0.0625 Risk-based capital ratio (CAP) (1) (2) (3) (4) Event×Affected 0.0104** 0.0149*** 0.0124** 0.0131** (0.0045) (0.0051) (0.0053) (0.0053) Affected −0.0031 (0.0075) Credit demand −0.0062*** −0.0052*** (0.0018) (0.0017) RoA −0.0005 (0.0027) NPL/assets −0.0036 (0.2624) Bank size −0.0237** (0.0093) UR 0.0724 (0.1379) Bank FE Yes No Yes Yes Time FE Yes Yes Yes Yes Number of banks 258 258 182 182 Number of observations 3,795 3,795 2,415 2,415 Adjusted R2 0.8670 0.0037 0.8576 0.8607 Within R2 0.0127 0.0043 0.0399 0.0625 Summing up, our results show that independent banks react when confronted with an adverse shock on their asset quality. They do this by increasing their risk-based capital ratios compared with banks that do not experience this shock. The results suggest that banks thereby strengthen their cushion against insolvency. This finding adds to Flannery and Rangan (2008) who suggest that a change in the banking environment rather than supervisory pressure leads to higher capital ratios for U.S. banks during the 1990s. Similarly, Gropp and Heider (2010) argue that banks rely on their “own judgement” to define the appropriate amount of total risk-based capital and that regulatory requirements are of secondary importance. 4.1.c. Alternative regional samples In the following, we discuss results of several robustness regressions. The respective result tables are provided in an Online Appendix. The first robustness check examines whether a smaller or larger sample of the states that we consider for the composition of both control and treatment groups might change our main results. Recall that our main results are based on a sample with 94 affected banks and 164 unaffected banks located in Alabama, Florida, Louisiana, Mississippi, Texas, Georgia, Tennessee, Arkansas, and Oklahoma. To check robustness, we make the following changes: first, we restrict the sample to banks that only operate in Alabama and Florida. The reason for this is that only these states comprise both counties that were affected and counties that were unaffected by the hurricanes. Second, we restrict the sample to counties in the core states affected by the hurricanes (Alabama, Florida, Louisiana, and Mississippi) and thus exclude banks in neighboring states (Arkansas, Georgia, Oklahoma, Tennessee, and Texas) from the control group compared with our baseline sample. Third, we extend the previous sample to banks in Texas. We rerun our main regression and provide results for our baseline sample and the three alternative regional samples in the Online Appendix. Across all groups, we find significant results for the treatment effect from Hurricane Katrina on the risk-based capital ratios of affected banks. We also find that the effect is economically stronger for the Alabama and Florida region. Here, affected banks increase their CAP by 2.28 percentage points compared with their unaffected peers after the event. Overall, these robustness regressions show that our results do not hinge on the choice of a specific regional sample or control group. 4.1.d. Extended control group or treatment group As stated in Section 3.1, the treatment group includes banks with their headquarters in a county that is eligible for individual and public disaster assistance. The control group includes banks with their headquarters in a county that is not eligible for disaster assistance. For robustness, we extend the control group or the treatment group as follows: first, we include in the control group all banks with their headquarters in a county in the U.S. Gulf Coast region or in neighboring states that we previously ignored, because these counties were not eligible for individual but rather only for public disaster assistance. Thus, we add to the control group banks that are located in counties that are in some way affected by the hurricanes, but certainly less so than counties that were also eligible for individual disaster assistance. Second, we include these banks in the treatment group instead of in the control group. For both changes, the sample increases from 258 to 422 independent banks. As shown in regressions with the extended control group and in regressions with the extended treatment group (see the Online Appendix for the full regression results), results are qualitatively unchanged compared with the baseline regressions in Table III. As expected, the coefficients of Event × Affected are relatively smaller in all regressions because the extended control group and the extended treatment group are defined less stringent compared with the baseline sample. 4.1.e. Parallel-trend assumption To alleviate potential biases, we have to guarantee that the parallel-trend assumption prior to the treatment is satisfied. In other words, the risk-based capital ratios should follow a similar trend for the treatment and control groups. In Figure 1, we graphically inspect the trend of the mean of total risk-based capital for both groups and confirm the parallel-trend assumption. In addition, as already discussed in Section 3.3 and shown in Table II, the groups of affected and unaffected banks are largely similar with respect to common bank characteristics. 4.1.f. Collapsed sample In order to show that the results are robust against problems with difference-in-difference techniques in the presence of serial correlation, Bertrand, Duflo, and Mullainathan (2004) suggest ignoring the time structure of the data. Therefore, we take the average of the data before and after the hurricane and rerun the estimation for this collapsed sample. As before, standard errors are clustered at the bank level. We find the treatment effect for all different periods intact and in the range of 1.12–2.03 percentage points (see the Online Appendix for the full regression results). 4.1.g. Time-placebo estimation The possibility that the results are driven by time trends unrelated to Hurricane Katrina must be ruled out. Therefore, we run a “placebo estimation” in which the treatment shifts from the time period when Hurricanes Katrina, Rita, and Wilma actually occurred (Q3 and Q4 of 2005) to a time period 3 years earlier (Q3 and Q4 of 2002). We then rerun the estimation for observations 2 years before and after this “2002 pseudo hurricane” event. As previously done, we run the regressions using total risk-based capital ratio as the dependent variable for four specifications: (1) with bank-fixed effects; (2) without bank-fixed effects; (3) with bank-fixed effects controlling for credit demand; and (4) with bank-fixed effects controlling for demand and some additional covariates that are common in the literature. We do not find an effect for the 2002 pseudo hurricane in any of the specifications (see the Online Appendix for the full regression results). This finding supports our assumption that our results are not driven by factors unrelated to Hurricane Katrina. 4.2 The Role of Bank Capitalization for Capital Ratio Adjustments Bank capitalization is a key determinant of bank risk in the supervisory assessment. Banks with low capital ratios are considered as less stable than banks with high capital ratios.21 In this section, we are interested in whether banks with lower or higher capital ratios before Hurricane Katrina struck the U.S. Gulf Coast, that is, banks that are considered less stable or more stable by the banking supervisor, differ in their capital ratio adjustments after the hurricanes. We construct a bank’s pre-Hurricane Katrina CAP by calculating for each bank the mean value of its risk-based capital ratio over the eight quarters prior to Hurricane Katrina (Q3 2003 to Q2 2005). These values range from 9.2% to 37.8% and have a mean value of 17.2%. Note that more than 95% of banks held an average pre-Katrina CAP above 10%, which is well above the required 8% and considered as “well capitalized” by the FDIC. Furthermore, we separate banks into two groups: banks with a pre-Hurricane Katrina CAP below the median (preCAP=0) or above the median (preCAP=1). Then, we extend Equation (1) by interacting the variables Event and Event × Affected with preCAP. We formally estimate the following equation for the sample of independent banks with a fixed effects OLS model: CAPit= νi+τt+β1(Eventt×Affectedi)+β2(Eventt×preCAPi)+β3(Eventt×Affectedi×preCAPi)+ϵit. (2) The first coefficient of interest is β1 and refers to Event × Affected, which now shows how a bank with a below median pre-Hurricane Katrina capital ratio adjusts its capital ratio after the hurricanes compared with the control group. The second coefficient of interest is β3 and refers to Event × Affected × preCAP, which shows whether (and how) the previous effect differs across the two groups of banks. As before, we run the regressions using CAP as the dependent variable for four specifications: (1) with bank-fixed effects; (2) without bank-fixed effects; (3) with bank-fixed effects controlling for credit demand; and (4) with bank-fixed effects controlling for credit demand and some additional covariates. As shown in Column (1) of Table IV, the coefficient of the interaction term Event × Affected is positive but insignificant with a value of 0.0009. This means that affected banks with a low pre-Hurricane Katrina capital ratio do not significantly increase their risk-based capital ratios after the event compared with the control group (unaffected banks with a low pre-Hurricane Katrina capital ratio). When we consider whether this effect is significantly different for the group of banks with a high pre-Hurricane Katrina capital ratio, we observe a positive and significant coefficient of the triple interaction term Event × Affected × preCAP. Hence, the effect Event × Affected is significantly different across the two groups of banks. At the bottom of Table IV we show that affected banks with a high pre-Hurricane Katrina capital ratio increase their risk-based capital ratios by 1.87 percentage points compared with the control group. This suggests that the key result of the previous regressions that banks in disaster areas increase their risk-based capital ratios after the hurricanes comes from banks with relatively high capital ratios. This finding also holds for the OLS estimation in Column (2), the fixed-effects estimation that controls for credit demand in Column (3), and for the fixed-effects estimation with additional control variables in Column (4). Table IV. The role of bank capitalization These regressions explore the role of bank capitalization for the capital ratio adjustments of banks following the 2005 hurricane season, as specified in Equation (2). In comparison with the baseline model, the terms Event, Affected, and Event×Affected are interacted with a dummy variable that is 0 if a bank’s average risk-based capital ratio during the eight quarters before the event is below the sample median (preCap = 0), and 1 otherwise (preCAP = 1). The sample includes quarterly data for all independent banks (not part of a bank holding company) in Alabama, Arkansas, Florida, Georgia, Louisiana, Mississippi, Oklahoma, Tennessee, and Texas over the period of ± 2 years around the 2005 hurricane season (Q3 2003–Q4 2007). A description of all variables is given in Table I. The models include bank-fixed effects (bank FE) and year-fixed effects (time FE) as stated in the table. Standard errors are clustered at the bank level and reported in parentheses. The ***, **, and * stand for significant coefficients at the 1%, 5%, and 10% levels, respectively. Risk-based capital ratio (CAP) (1) (2) (3) (4) Event×Affected 0.0009 0.0037 0.0059 0.0069 (0.0057) (0.0059) (0.0060) (0.0056) Event×preCAP −0.0269*** −0.0260*** −0.0201*** −0.0222*** (0.0046) (0.0047) (0.0055) (0.0051) Event×Affected×preCAP 0.0178** 0.0132 0.0142 0.0131 (0.0085) (0.0090) (0.0107) (0.0102) preCAP 0.0913*** (0.0062) Affected 0.0040 (0.0027) Affected×preCAP −0.0055 (0.0099) Credit demand −0.0064*** −0.0052*** (0.0017) (0.0016) RoA −0.0004 (0.0026) NPL/assets −0.0059 (0.2055) Bank size −0.0283*** (0.0091) UR 0.0325 (0.1329) Bank FE Yes No Yes Yes Time FE Yes Yes Yes Yes Number of banks 258 258 182 182 Number of observations 3,795 3,795 2,415 2,415 Adjusted R2 0.8757 0.4484 0.8622 0.8665 Within R2 0.0777 0.4493 0.0718 0.1025 Difference-in-difference effect for relatively low (preCAP=0) or highly (preCAP=1) capitalized banks Event×Affected for preCAP=0 0.0009 0.0037 0.0059 0.0069 (0.0057) (0.0059) (0.0060) (0.0056) Event×Affected for preCAP=1 0.0187*** 0.0169** 0.0201** 0.0200** (0.0063) (0.0069) (0.0089) (0.0087) Risk-based capital ratio (CAP) (1) (2) (3) (4) Event×Affected 0.0009 0.0037 0.0059 0.0069 (0.0057) (0.0059) (0.0060) (0.0056) Event×preCAP −0.0269*** −0.0260*** −0.0201*** −0.0222*** (0.0046) (0.0047) (0.0055) (0.0051) Event×Affected×preCAP 0.0178** 0.0132 0.0142 0.0131 (0.0085) (0.0090) (0.0107) (0.0102) preCAP 0.0913*** (0.0062) Affected 0.0040 (0.0027) Affected×preCAP −0.0055 (0.0099) Credit demand −0.0064*** −0.0052*** (0.0017) (0.0016) RoA −0.0004 (0.0026) NPL/assets −0.0059 (0.2055) Bank size −0.0283*** (0.0091) UR 0.0325 (0.1329) Bank FE Yes No Yes Yes Time FE Yes Yes Yes Yes Number of banks 258 258 182 182 Number of observations 3,795 3,795 2,415 2,415 Adjusted R2 0.8757 0.4484 0.8622 0.8665 Within R2 0.0777 0.4493 0.0718 0.1025 Difference-in-difference effect for relatively low (preCAP=0) or highly (preCAP=1) capitalized banks Event×Affected for preCAP=0 0.0009 0.0037 0.0059 0.0069 (0.0057) (0.0059) (0.0060) (0.0056) Event×Affected for preCAP=1 0.0187*** 0.0169** 0.0201** 0.0200** (0.0063) (0.0069) (0.0089) (0.0087) Table IV. The role of bank capitalization These regressions explore the role of bank capitalization for the capital ratio adjustments of banks following the 2005 hurricane season, as specified in Equation (2). In comparison with the baseline model, the terms Event, Affected, and Event×Affected are interacted with a dummy variable that is 0 if a bank’s average risk-based capital ratio during the eight quarters before the event is below the sample median (preCap = 0), and 1 otherwise (preCAP = 1). The sample includes quarterly data for all independent banks (not part of a bank holding company) in Alabama, Arkansas, Florida, Georgia, Louisiana, Mississippi, Oklahoma, Tennessee, and Texas over the period of ± 2 years around the 2005 hurricane season (Q3 2003–Q4 2007). A description of all variables is given in Table I. The models include bank-fixed effects (bank FE) and year-fixed effects (time FE) as stated in the table. Standard errors are clustered at the bank level and reported in parentheses. The ***, **, and * stand for significant coefficients at the 1%, 5%, and 10% levels, respectively. Risk-based capital ratio (CAP) (1) (2) (3) (4) Event×Affected 0.0009 0.0037 0.0059 0.0069 (0.0057) (0.0059) (0.0060) (0.0056) Event×preCAP −0.0269*** −0.0260*** −0.0201*** −0.0222*** (0.0046) (0.0047) (0.0055) (0.0051) Event×Affected×preCAP 0.0178** 0.0132 0.0142 0.0131 (0.0085) (0.0090) (0.0107) (0.0102) preCAP 0.0913*** (0.0062) Affected 0.0040 (0.0027) Affected×preCAP −0.0055 (0.0099) Credit demand −0.0064*** −0.0052*** (0.0017) (0.0016) RoA −0.0004 (0.0026) NPL/assets −0.0059 (0.2055) Bank size −0.0283*** (0.0091) UR 0.0325 (0.1329) Bank FE Yes No Yes Yes Time FE Yes Yes Yes Yes Number of banks 258 258 182 182 Number of observations 3,795 3,795 2,415 2,415 Adjusted R2 0.8757 0.4484 0.8622 0.8665 Within R2 0.0777 0.4493 0.0718 0.1025 Difference-in-difference effect for relatively low (preCAP=0) or highly (preCAP=1) capitalized banks Event×Affected for preCAP=0 0.0009 0.0037 0.0059 0.0069 (0.0057) (0.0059) (0.0060) (0.0056) Event×Affected for preCAP=1 0.0187*** 0.0169** 0.0201** 0.0200** (0.0063) (0.0069) (0.0089) (0.0087) Risk-based capital ratio (CAP) (1) (2) (3) (4) Event×Affected 0.0009 0.0037 0.0059 0.0069 (0.0057) (0.0059) (0.0060) (0.0056) Event×preCAP −0.0269*** −0.0260*** −0.0201*** −0.0222*** (0.0046) (0.0047) (0.0055) (0.0051) Event×Affected×preCAP 0.0178** 0.0132 0.0142 0.0131 (0.0085) (0.0090) (0.0107) (0.0102) preCAP 0.0913*** (0.0062) Affected 0.0040 (0.0027) Affected×preCAP −0.0055 (0.0099) Credit demand −0.0064*** −0.0052*** (0.0017) (0.0016) RoA −0.0004 (0.0026) NPL/assets −0.0059 (0.2055) Bank size −0.0283*** (0.0091) UR 0.0325 (0.1329) Bank FE Yes No Yes Yes Time FE Yes Yes Yes Yes Number of banks 258 258 182 182 Number of observations 3,795 3,795 2,415 2,415 Adjusted R2 0.8757 0.4484 0.8622 0.8665 Within R2 0.0777 0.4493 0.0718 0.1025 Difference-in-difference effect for relatively low (preCAP=0) or highly (preCAP=1) capitalized banks Event×Affected for preCAP=0 0.0009 0.0037 0.0059 0.0069 (0.0057) (0.0059) (0.0060) (0.0056) Event×Affected for preCAP=1 0.0187*** 0.0169** 0.0201** 0.0200** (0.0063) (0.0069) (0.0089) (0.0087) The analysis yields an interesting result about how different banks react to the adverse shock on their asset quality through the hurricanes. Banks that appear ex ante relatively conservative in their business model—or at least hold relatively large capital buffers—also react conservatively to the shock and thus further increase their buffers against future losses. The effect is insignificant for banks that appear more risky in their business model—or at least hold relatively smaller capital buffers. From a supervisory perspective, these banks are presumably those that should increase their buffers against future losses, but as our evidence shows, these banks are not capable or willing to do so. 4.3 The Role of Bank Structure for Capital Ratio Adjustments In this section, we explore whether banks that are part of a bank holding company, which we refer to as BHC banks, adapt their capital ratios in a similar way than independent banks, on which the analysis has focused thus far. In contrast to independent banks, bank holding companies have the opportunity to establish internal capital markets to allocate capital across their various subsidiaries (Houston, James, and Marcus, 1997). As a consequence, BHC banks have greater leeway in case of financial distress and may rely on this flexibility instead of building higher capital ratios by themselves. For the following analysis, we extend the sample of 258 independent banks by 995 BHC banks. This results in a total sample of 1,253 banks, of which 307 were affected by the 2005 hurricanes and 946 were unaffected. We formally extend Equation (1) to differentiate between independent banks and BHC banks and estimate the following equation: CAPit= νi+τt+β1(Eventt×Affectedi)+β2(Eventt×BHCi)+β3(Eventt×Affectedi×BHCi)+ϵit. (3) The variable BHC has a value of 0 for independent banks and a value of 1 for BHC banks. We estimate a difference-in-difference-in-difference model that analyzes whether the effect on affected independent banks is different from the effect on affected BHC banks. In particular, the interaction term Event × Affected captures the effect of the 2005 hurricanes on independent banks, and the coefficient of the triple interaction term Event × Affected × BHC captures whether and how the effect is different for BHC banks.22 As previously done, we run the regressions using CAP as the dependent variable for four specifications: (1) with bank-fixed effects as reflected in the equation above; (2) without bank-fixed effects; (3) with bank-fixed effects controlling for credit demand; and (4) with bank-fixed effects controlling for credit demand and some additional covariates that are common in the literature. As shown in Column (1) of Table V, we find a positive and significant coefficient of the interaction term Event × Affected (0.0104). The coefficient of the triple interaction term, which shows the differential effect for BHC banks, is significant and negative. The magnitude of this coefficient of −0.0089 is close to the magnitude of the coefficient of the double interaction term Event × Affected. Hence, the total effect is 0.0104 for independent banks (BHC = 0) while it is 0.0014 for BHC banks (BHC=1). The bottom rows of Table V show these effects for both groups of banks. Accordingly, the effect is significant for independent banks on the 5% level and insignificant for BHC banks. Results remain robust when we estimate the equation without bank-fixed effects [Column (2)], control for credit demand [Column (3)], or include additional bank covariates [Column (4)]. Table V. The role of bank structure These regressions explore the role of bank structure for the capital ratio adjustments of banks following the 2005 hurricane season, as specified in Equation (3). In comparison with the baseline model, the terms Event, Affected, and Event× Affected are interacted with a dummy variable that indicates independent banks (BHC = 0) or banks that belong to a bank holding company (BHC = 1). The sample includes quarterly data for all banks in Alabama, Arkansas, Florida, Georgia, Louisiana, Mississippi, Oklahoma, Tennessee, and Texas over the period of ± 2 years around the 2005 hurricane season (Q3 2003–Q4 2007). A description of all variables is given in Table I. The models include bank-fixed effects (bank FE) and year-fixed effects (time FE) as stated in the table. Standard errors are clustered at the bank level and reported in parentheses. The ***, **, and * stand for significant coefficients at the 1%, 5%, and 10% levels, respectively. Risk-based capital ratio (CAP) (1) (2) (3) (4) Event×Affected 0.0104** 0.0149*** 0.0126** 0.0130** (0.0044) (0.0051) (0.0054) (0.0054) Event×BHC 0.0043 0.0030 0.0032 0.0034 (0.0027) (0.0029) (0.0030) (0.0029) Event×Affected×BHC −0.0089* −0.0133** −0.0098* −0.0101* (0.0048) (0.0055) (0.0059) (0.0058) Affected −0.0031 (0.0075) BHC −0.0201*** (0.0051) Affected×BHC 0.0076 (0.0084) Credit demand −0.0012 −0.0005 (0.0008) (0.0008) RoA 0.0006 (0.0012) NPL/assets −0.0510 (0.1219) Bank size −0.0129*** (0.0040) UR 0.0548 (0.0409) Bank FE Yes No Yes Yes Time FE Yes Yes Yes Yes Number of banks 1,253 1,253 859 859 Number of observations 19,288 19,288 11,971 11,971 Adjusted R2 0.8883 0.0233 0.8840 0.8855 Within R2 0.0042 0.0222 0.0099 0.0223 Difference-in-difference effect for independent banks (BHC=0) and BHC banks (BHC=1) Event×Affected for BHC=0 0.0104** 0.0149*** 0.0126** 0.0130** (0.0044) (0.0051) (0.0054) (0.0054) Event×Affected for BHC=1 0.0014 0.0016 0.0028 0.0029 (0.0019) (0.0020) (0.0022) (0.0021) Risk-based capital ratio (CAP) (1) (2) (3) (4) Event×Affected 0.0104** 0.0149*** 0.0126** 0.0130** (0.0044) (0.0051) (0.0054) (0.0054) Event×BHC 0.0043 0.0030 0.0032 0.0034 (0.0027) (0.0029) (0.0030) (0.0029) Event×Affected×BHC −0.0089* −0.0133** −0.0098* −0.0101* (0.0048) (0.0055) (0.0059) (0.0058) Affected −0.0031 (0.0075) BHC −0.0201*** (0.0051) Affected×BHC 0.0076 (0.0084) Credit demand −0.0012 −0.0005 (0.0008) (0.0008) RoA 0.0006 (0.0012) NPL/assets −0.0510 (0.1219) Bank size −0.0129*** (0.0040) UR 0.0548 (0.0409) Bank FE Yes No Yes Yes Time FE Yes Yes Yes Yes Number of banks 1,253 1,253 859 859 Number of observations 19,288 19,288 11,971 11,971 Adjusted R2 0.8883 0.0233 0.8840 0.8855 Within R2 0.0042 0.0222 0.0099 0.0223 Difference-in-difference effect for independent banks (BHC=0) and BHC banks (BHC=1) Event×Affected for BHC=0 0.0104** 0.0149*** 0.0126** 0.0130** (0.0044) (0.0051) (0.0054) (0.0054) Event×Affected for BHC=1 0.0014 0.0016 0.0028 0.0029 (0.0019) (0.0020) (0.0022) (0.0021) Table V. The role of bank structure These regressions explore the role of bank structure for the capital ratio adjustments of banks following the 2005 hurricane season, as specified in Equation (3). In comparison with the baseline model, the terms Event, Affected, and Event× Affected are interacted with a dummy variable that indicates independent banks (BHC = 0) or banks that belong to a bank holding company (BHC = 1). The sample includes quarterly data for all banks in Alabama, Arkansas, Florida, Georgia, Louisiana, Mississippi, Oklahoma, Tennessee, and Texas over the period of ± 2 years around the 2005 hurricane season (Q3 2003–Q4 2007). A description of all variables is given in Table I. The models include bank-fixed effects (bank FE) and year-fixed effects (time FE) as stated in the table. Standard errors are clustered at the bank level and reported in parentheses. The ***, **, and * stand for significant coefficients at the 1%, 5%, and 10% levels, respectively. Risk-based capital ratio (CAP) (1) (2) (3) (4) Event×Affected 0.0104** 0.0149*** 0.0126** 0.0130** (0.0044) (0.0051) (0.0054) (0.0054) Event×BHC 0.0043 0.0030 0.0032 0.0034 (0.0027) (0.0029) (0.0030) (0.0029) Event×Affected×BHC −0.0089* −0.0133** −0.0098* −0.0101* (0.0048) (0.0055) (0.0059) (0.0058) Affected −0.0031 (0.0075) BHC −0.0201*** (0.0051) Affected×BHC 0.0076 (0.0084) Credit demand −0.0012 −0.0005 (0.0008) (0.0008) RoA 0.0006 (0.0012) NPL/assets −0.0510 (0.1219) Bank size −0.0129*** (0.0040) UR 0.0548 (0.0409) Bank FE Yes No Yes Yes Time FE Yes Yes Yes Yes Number of banks 1,253 1,253 859 859 Number of observations 19,288 19,288 11,971 11,971 Adjusted R2 0.8883 0.0233 0.8840 0.8855 Within R2 0.0042 0.0222 0.0099 0.0223 Difference-in-difference effect for independent banks (BHC=0) and BHC banks (BHC=1) Event×Affected for BHC=0 0.0104** 0.0149*** 0.0126** 0.0130** (0.0044) (0.0051) (0.0054) (0.0054) Event×Affected for BHC=1 0.0014 0.0016 0.0028 0.0029 (0.0019) (0.0020) (0.0022) (0.0021) Risk-based capital ratio (CAP) (1) (2) (3) (4) Event×Affected 0.0104** 0.0149*** 0.0126** 0.0130** (0.0044) (0.0051) (0.0054) (0.0054) Event×BHC 0.0043 0.0030 0.0032 0.0034 (0.0027) (0.0029) (0.0030) (0.0029) Event×Affected×BHC −0.0089* −0.0133** −0.0098* −0.0101* (0.0048) (0.0055) (0.0059) (0.0058) Affected −0.0031 (0.0075) BHC −0.0201*** (0.0051) Affected×BHC 0.0076 (0.0084) Credit demand −0.0012 −0.0005 (0.0008) (0.0008) RoA 0.0006 (0.0012) NPL/assets −0.0510 (0.1219) Bank size −0.0129*** (0.0040) UR 0.0548 (0.0409) Bank FE Yes No Yes Yes Time FE Yes Yes Yes Yes Number of banks 1,253 1,253 859 859 Number of observations 19,288 19,288 11,971 11,971 Adjusted R2 0.8883 0.0233 0.8840 0.8855 Within R2 0.0042 0.0222 0.0099 0.0223 Difference-in-difference effect for independent banks (BHC=0) and BHC banks (BHC=1) Event×Affected for BHC=0 0.0104** 0.0149*** 0.0126** 0.0130** (0.0044) (0.0051) (0.0054) (0.0054) Event×Affected for BHC=1 0.0014 0.0016 0.0028 0.0029 (0.0019) (0.0020) (0.0022) (0.0021) In summary, we observe that affected independent banks adjust their risk-based capital ratios after the catastrophic event compared with banks in the control group, while affected banks that belong to a bank holding company do not. This indicates that BHC banks rely on potential financial support from their bank holding company, which they can access through internal capital markets and, therefore, do not build up precautionary capital reserves internally. 4.4 The Development of Banks’ Balance Sheet Exposures Next, we are interested in the mechanisms used by banks to adjust their risk-based capital ratios. We therefore explore the development of the balance sheet exposures of banks, including banks’ capital, risk-weighted assets, U.S. government securities, and various loan categories. Furthermore, we consider the previous results that banks’ risk-based capital ratios prior to Hurricane Katrina, as well as bank structures, do indeed matter. Hence, we include interactions with both dummy variables that were introduced in the previous sections, preCAP and BHC, in the regression model: Yit= νi+τt+β1(Eventt×Affectedi)+β2(Eventt×BHCi)+β3(Eventt×preCAPi)+β4(Eventt×Affectedi×BHCi)+β5(Eventt×Affectedi×preCAPi)+β6(Eventt×BHCi×preCAPi)+β7(Eventt×Affectedi×BHCi×preCAPi)+ϵit, (4) where Yit stands for alternative balance sheet variables: the natural logarithms of total capital, risk-weighted assets, government securities, real estate loans, real estate commercial loans, real estate construction and development loans, consumer loans, and consumer and industry loans. All other variables are as previously defined in Equations (1)–(3). Regression results are shown in Table VI. The bottom of the table shows corresponding difference-in-difference effects (Affected × Event) for the four different groups of banks: independent and low-capitalized (BHC=0 and preCAP=0), independent and highly capitalized (BHC=0 and preCAP=1), BHC and low-capitalized (BHC=1 and preCAP=0), and BHC and highly capitalized (BHC=1 and preCAP=1). Table VI. Balance sheet exposures These regressions explore the development of various balance sheet exposures of banks following the 2005 hurricane season: the natural logarithms of total capital; risk-weighted assets (RWA); government securities (GOV); real estate loans (RE); commercial real estate loans (RECO); construction and development loans (RECON); consumer loans (CON); and commercial and industrial loans (C&I). In order to recap our previous results, the first column shows a regression in which banks’ risk-based capital ratio (CAP) is the dependent variable. The model is specified in Equation (4). In comparison with the baseline model, the terms Event, Affected, and Event×Affected are interacted with two variables: a dummy variable that indicates independent banks (BHC = 0) or banks that belong to a bank holding company (BHC = 1), and a dummy variable which is 0 if a bank’s average CAP during the eight quarters before the event is below the sample median (preCap = 0) and 1 otherwise (preCap = 1). The sample includes quarterly data for all banks in Alabama, Louisiana, Mississippi, Florida, Texas, Georgia, Tennessee, Arkansas, and Oklahoma over the period of ± 2 years around the 2005 hurricane season (Q3 2003–Q4 2007). Bank-fixed effects (bank FE) and year-fixed effects (time FE) are included in all regressions. Standard errors are clustered at the bank level and reported in parentheses. The ***, **, and * stand for significant coefficients at the 1%, 5%, and 10% levels, respectively. CAP log(capital) log(RWA) log(GOV) log(RE) log(RECO) log(RECON) log(CON) log(C&I) (1) (2) (3) (4) (5) (6) (7) (8) (9) Event×Affected 0.0009 0.0149 0.0172 −0.3302 0.0289 0.0885 0.3102 −0.1020 0.0560 (0.0057) (0.0512) (0.0592) (0.2265) (0.0658) (0.0862) (0.2940) (0.1168) (0.1282) Event×BHC −0.0041 0.0121 0.0386 −0.3713* 0.0763** −0.0230 0.3638*** −0.0965 −0.0847 (0.0027) (0.0316) (0.0327) (0.2117) (0.0337) (0.0441) (0.0984) (0.0596) (0.0585) Event×preCAP −0.0269*** −0.1862*** −0.0402 −0.5632*** 0.0072 0.1341* 0.3283* −0.0166 −0.0184 (0.0046) (0.0375) (0.0424) (0.2136) (0.0441) (0.0699) (0.1742) (0.0687) (0.0849) Event×Affected×BHC 0.0030 0.0274 −0.0070 0.5433** −0.0201 −0.1048 −0.4030 0.1300 −0.0838 (0.0060) (0.0585) (0.0660) (0.2505) (0.0737) (0.0941) (0.3084) (0.1244) (0.1367) Event×Affected×preCAP 0.0178** −0.0096 −0.1017 0.7552** −0.0706 −0.1576 −0.5840 0.0045 −0.2247 (0.0085) (0.0621) (0.0745) (0.3058) (0.0889) (0.1283) (0.3607) (0.1426) (0.1621) Event×BHC×preCAP 0.0152*** −0.0219 −0.1001** 0.4517** −0.1408*** −0.1802** −0.3821** 0.0249 −0.0210 (0.0049) (0.0408) (0.0460) (0.2200) (0.0484) (0.0757) (0.1911) (0.0740) (0.0894) Event×Affected×BHC×preCAP −0.0206** −0.0162 0.1034 −0.9100*** 0.0775 0.1648 0.5770 −0.1153 0.2673 (0.0092) (0.0707) (0.0832) (0.3268) (0.0987) (0.1389) (0.3862) (0.1575) (0.1734) Bank FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Time FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Number of banks 1,253 1,253 1,253 1,253 1,253 1,253 1,253 1,253 1,253 Number of observations 19,288 19,288 19,288 19,288 19,288 19,288 19,288 19,288 19,288 Adjusted R2 0.8933 0.9812 0.9808 0.8736 0.9801 0.9637 0.8930 0.9406 0.9502 Within R2 0.0493 0.1284 0.0457 0.0126 0.0286 0.0110 0.0035 0.0049 0.0067 Difference-in-difference effect for subgroups of banks Event×Aff. for BHC=0, preCAP=0 0.0009 0.0149 0.0172 −0.3302 0.0289 0.0885 0.3102 −0.1020 0.0560 (0.0057) (0.0512) (0.0592) (0.2265) (0.0658) (0.0862) (0.2940) (0.1168) (0.1282) Event×Aff. for BHC=0, preCAP=1 0.0187*** 0.0053 −0.0845* 0.4251** −0.0416 −0.0691 −0.2738 −0.0976 −0.1687* (0.0063) (0.0351) (0.0453) (0.2057) (0.0597) (0.0950) (0.2088) (0.0818) (0.0991) Event×Aff. for BHC=1, preCAP=0 0.0039** 0.0423 0.0102 0.2132** 0.0089 −0.0163 −0.0928 0.0280 −0.0278 (0.0018) (0.0283) (0.0292) (0.1070) (0.0332) (0.0378) (0.0933) (0.0428) (0.0474) Event×Aff. for BHC=1, preCAP=1 0.0011 0.0165 0.0119 0.0584 0.0158 −0.0091 −0.0999 −0.0828 0.0147 (0.0030) (0.0188) (0.0229) (0.0427) (0.0273) (0.0377) (0.1018) (0.0512) (0.0397) CAP log(capital) log(RWA) log(GOV) log(RE) log(RECO) log(RECON) log(CON) log(C&I) (1) (2) (3) (4) (5) (6) (7) (8) (9) Event×Affected 0.0009 0.0149 0.0172 −0.3302 0.0289 0.0885 0.3102 −0.1020 0.0560 (0.0057) (0.0512) (0.0592) (0.2265) (0.0658) (0.0862) (0.2940) (0.1168) (0.1282) Event×BHC −0.0041 0.0121 0.0386 −0.3713* 0.0763** −0.0230 0.3638*** −0.0965 −0.0847 (0.0027) (0.0316) (0.0327) (0.2117) (0.0337) (0.0441) (0.0984) (0.0596) (0.0585) Event×preCAP −0.0269*** −0.1862*** −0.0402 −0.5632*** 0.0072 0.1341* 0.3283* −0.0166 −0.0184 (0.0046) (0.0375) (0.0424) (0.2136) (0.0441) (0.0699) (0.1742) (0.0687) (0.0849) Event×Affected×BHC 0.0030 0.0274 −0.0070 0.5433** −0.0201 −0.1048 −0.4030 0.1300 −0.0838 (0.0060) (0.0585) (0.0660) (0.2505) (0.0737) (0.0941) (0.3084) (0.1244) (0.1367) Event×Affected×preCAP 0.0178** −0.0096 −0.1017 0.7552** −0.0706 −0.1576 −0.5840 0.0045 −0.2247 (0.0085) (0.0621) (0.0745) (0.3058) (0.0889) (0.1283) (0.3607) (0.1426) (0.1621) Event×BHC×preCAP 0.0152*** −0.0219 −0.1001** 0.4517** −0.1408*** −0.1802** −0.3821** 0.0249 −0.0210 (0.0049) (0.0408) (0.0460) (0.2200) (0.0484) (0.0757) (0.1911) (0.0740) (0.0894) Event×Affected×BHC×preCAP −0.0206** −0.0162 0.1034 −0.9100*** 0.0775 0.1648 0.5770 −0.1153 0.2673 (0.0092) (0.0707) (0.0832) (0.3268) (0.0987) (0.1389) (0.3862) (0.1575) (0.1734) Bank FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Time FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Number of banks 1,253 1,253 1,253 1,253 1,253 1,253 1,253 1,253 1,253 Number of observations 19,288 19,288 19,288 19,288 19,288 19,288 19,288 19,288 19,288 Adjusted R2 0.8933 0.9812 0.9808 0.8736 0.9801 0.9637 0.8930 0.9406 0.9502 Within R2 0.0493 0.1284 0.0457 0.0126 0.0286 0.0110 0.0035 0.0049 0.0067 Difference-in-difference effect for subgroups of banks Event×Aff. for BHC=0, preCAP=0 0.0009 0.0149 0.0172 −0.3302 0.0289 0.0885 0.3102 −0.1020 0.0560 (0.0057) (0.0512) (0.0592) (0.2265) (0.0658) (0.0862) (0.2940) (0.1168) (0.1282) Event×Aff. for BHC=0, preCAP=1 0.0187*** 0.0053 −0.0845* 0.4251** −0.0416 −0.0691 −0.2738 −0.0976 −0.1687* (0.0063) (0.0351) (0.0453) (0.2057) (0.0597) (0.0950) (0.2088) (0.0818) (0.0991) Event×Aff. for BHC=1, preCAP=0 0.0039** 0.0423 0.0102 0.2132** 0.0089 −0.0163 −0.0928 0.0280 −0.0278 (0.0018) (0.0283) (0.0292) (0.1070) (0.0332) (0.0378) (0.0933) (0.0428) (0.0474) Event×Aff. for BHC=1, preCAP=1 0.0011 0.0165 0.0119 0.0584 0.0158 −0.0091 −0.0999 −0.0828 0.0147 (0.0030) (0.0188) (0.0229) (0.0427) (0.0273) (0.0377) (0.1018) (0.0512) (0.0397) Table VI. Balance sheet exposures These regressions explore the development of various balance sheet exposures of banks following the 2005 hurricane season: the natural logarithms of total capital; risk-weighted assets (RWA); government securities (GOV); real estate loans (RE); commercial real estate loans (RECO); construction and development loans (RECON); consumer loans (CON); and commercial and industrial loans (C&I). In order to recap our previous results, the first column shows a regression in which banks’ risk-based capital ratio (CAP) is the dependent variable. The model is specified in Equation (4). In comparison with the baseline model, the terms Event, Affected, and Event×Affected are interacted with two variables: a dummy variable that indicates independent banks (BHC = 0) or banks that belong to a bank holding company (BHC = 1), and a dummy variable which is 0 if a bank’s average CAP during the eight quarters before the event is below the sample median (preCap = 0) and 1 otherwise (preCap = 1). The sample includes quarterly data for all banks in Alabama, Louisiana, Mississippi, Florida, Texas, Georgia, Tennessee, Arkansas, and Oklahoma over the period of ± 2 years around the 2005 hurricane season (Q3 2003–Q4 2007). Bank-fixed effects (bank FE) and year-fixed effects (time FE) are included in all regressions. Standard errors are clustered at the bank level and reported in parentheses. The ***, **, and * stand for significant coefficients at the 1%, 5%, and 10% levels, respectively. CAP log(capital) log(RWA) log(GOV) log(RE) log(RECO) log(RECON) log(CON) log(C&I) (1) (2) (3) (4) (5) (6) (7) (8) (9) Event×Affected 0.0009 0.0149 0.0172 −0.3302 0.0289 0.0885 0.3102 −0.1020 0.0560 (0.0057) (0.0512) (0.0592) (0.2265) (0.0658) (0.0862) (0.2940) (0.1168) (0.1282) Event×BHC −0.0041 0.0121 0.0386 −0.3713* 0.0763** −0.0230 0.3638*** −0.0965 −0.0847 (0.0027) (0.0316) (0.0327) (0.2117) (0.0337) (0.0441) (0.0984) (0.0596) (0.0585) Event×preCAP −0.0269*** −0.1862*** −0.0402 −0.5632*** 0.0072 0.1341* 0.3283* −0.0166 −0.0184 (0.0046) (0.0375) (0.0424) (0.2136) (0.0441) (0.0699) (0.1742) (0.0687) (0.0849) Event×Affected×BHC 0.0030 0.0274 −0.0070 0.5433** −0.0201 −0.1048 −0.4030 0.1300 −0.0838 (0.0060) (0.0585) (0.0660) (0.2505) (0.0737) (0.0941) (0.3084) (0.1244) (0.1367) Event×Affected×preCAP 0.0178** −0.0096 −0.1017 0.7552** −0.0706 −0.1576 −0.5840 0.0045 −0.2247 (0.0085) (0.0621) (0.0745) (0.3058) (0.0889) (0.1283) (0.3607) (0.1426) (0.1621) Event×BHC×preCAP 0.0152*** −0.0219 −0.1001** 0.4517** −0.1408*** −0.1802** −0.3821** 0.0249 −0.0210 (0.0049) (0.0408) (0.0460) (0.2200) (0.0484) (0.0757) (0.1911) (0.0740) (0.0894) Event×Affected×BHC×preCAP −0.0206** −0.0162 0.1034 −0.9100*** 0.0775 0.1648 0.5770 −0.1153 0.2673 (0.0092) (0.0707) (0.0832) (0.3268) (0.0987) (0.1389) (0.3862) (0.1575) (0.1734) Bank FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Time FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Number of banks 1,253 1,253 1,253 1,253 1,253 1,253 1,253 1,253 1,253 Number of observations 19,288 19,288 19,288 19,288 19,288 19,288 19,288 19,288 19,288 Adjusted R2 0.8933 0.9812 0.9808 0.8736 0.9801 0.9637 0.8930 0.9406 0.9502 Within R2 0.0493 0.1284 0.0457 0.0126 0.0286 0.0110 0.0035 0.0049 0.0067 Difference-in-difference effect for subgroups of banks Event×Aff. for BHC=0, preCAP=0 0.0009 0.0149 0.0172 −0.3302 0.0289 0.0885 0.3102 −0.1020 0.0560 (0.0057) (0.0512) (0.0592) (0.2265) (0.0658) (0.0862) (0.2940) (0.1168) (0.1282) Event×Aff. for BHC=0, preCAP=1 0.0187*** 0.0053 −0.0845* 0.4251** −0.0416 −0.0691 −0.2738 −0.0976 −0.1687* (0.0063) (0.0351) (0.0453) (0.2057) (0.0597) (0.0950) (0.2088) (0.0818) (0.0991) Event×Aff. for BHC=1, preCAP=0 0.0039** 0.0423 0.0102 0.2132** 0.0089 −0.0163 −0.0928 0.0280 −0.0278 (0.0018) (0.0283) (0.0292) (0.1070) (0.0332) (0.0378) (0.0933) (0.0428) (0.0474) Event×Aff. for BHC=1, preCAP=1 0.0011 0.0165 0.0119 0.0584 0.0158 −0.0091 −0.0999 −0.0828 0.0147 (0.0030) (0.0188) (0.0229) (0.0427) (0.0273) (0.0377) (0.1018) (0.0512) (0.0397) CAP log(capital) log(RWA) log(GOV) log(RE) log(RECO) log(RECON) log(CON) log(C&I) (1) (2) (3) (4) (5) (6) (7) (8) (9) Event×Affected 0.0009 0.0149 0.0172 −0.3302 0.0289 0.0885 0.3102 −0.1020 0.0560 (0.0057) (0.0512) (0.0592) (0.2265) (0.0658) (0.0862) (0.2940) (0.1168) (0.1282) Event×BHC −0.0041 0.0121 0.0386 −0.3713* 0.0763** −0.0230 0.3638*** −0.0965 −0.0847 (0.0027) (0.0316) (0.0327) (0.2117) (0.0337) (0.0441) (0.0984) (0.0596) (0.0585) Event×preCAP −0.0269*** −0.1862*** −0.0402 −0.5632*** 0.0072 0.1341* 0.3283* −0.0166 −0.0184 (0.0046) (0.0375) (0.0424) (0.2136) (0.0441) (0.0699) (0.1742) (0.0687) (0.0849) Event×Affected×BHC 0.0030 0.0274 −0.0070 0.5433** −0.0201 −0.1048 −0.4030 0.1300 −0.0838 (0.0060) (0.0585) (0.0660) (0.2505) (0.0737) (0.0941) (0.3084) (0.1244) (0.1367) Event×Affected×preCAP 0.0178** −0.0096 −0.1017 0.7552** −0.0706 −0.1576 −0.5840 0.0045 −0.2247 (0.0085) (0.0621) (0.0745) (0.3058) (0.0889) (0.1283) (0.3607) (0.1426) (0.1621) Event×BHC×preCAP 0.0152*** −0.0219 −0.1001** 0.4517** −0.1408*** −0.1802** −0.3821** 0.0249 −0.0210 (0.0049) (0.0408) (0.0460) (0.2200) (0.0484) (0.0757) (0.1911) (0.0740) (0.0894) Event×Affected×BHC×preCAP −0.0206** −0.0162 0.1034 −0.9100*** 0.0775 0.1648 0.5770 −0.1153 0.2673 (0.0092) (0.0707) (0.0832) (0.3268) (0.0987) (0.1389) (0.3862) (0.1575) (0.1734) Bank FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Time FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Number of banks 1,253 1,253 1,253 1,253 1,253 1,253 1,253 1,253 1,253 Number of observations 19,288 19,288 19,288 19,288 19,288 19,288 19,288 19,288 19,288 Adjusted R2 0.8933 0.9812 0.9808 0.8736 0.9801 0.9637 0.8930 0.9406 0.9502 Within R2 0.0493 0.1284 0.0457 0.0126 0.0286 0.0110 0.0035 0.0049 0.0067 Difference-in-difference effect for subgroups of banks Event×Aff. for BHC=0, preCAP=0 0.0009 0.0149 0.0172 −0.3302 0.0289 0.0885 0.3102 −0.1020 0.0560 (0.0057) (0.0512) (0.0592) (0.2265) (0.0658) (0.0862) (0.2940) (0.1168) (0.1282) Event×Aff. for BHC=0, preCAP=1 0.0187*** 0.0053 −0.0845* 0.4251** −0.0416 −0.0691 −0.2738 −0.0976 −0.1687* (0.0063) (0.0351) (0.0453) (0.2057) (0.0597) (0.0950) (0.2088) (0.0818) (0.0991) Event×Aff. for BHC=1, preCAP=0 0.0039** 0.0423 0.0102 0.2132** 0.0089 −0.0163 −0.0928 0.0280 −0.0278 (0.0018) (0.0283) (0.0292) (0.1070) (0.0332) (0.0378) (0.0933) (0.0428) (0.0474) Event×Aff. for BHC=1, preCAP=1 0.0011 0.0165 0.0119 0.0584 0.0158 −0.0091 −0.0999 −0.0828 0.0147 (0.0030) (0.0188) (0.0229) (0.0427) (0.0273) (0.0377) (0.1018) (0.0512) (0.0397) To recap our previous results, the first column of Table VI shows the regression results of Equation (4) using banks’ risk-based capital ratio as the dependent variable. We again find that, in particular, independent banks with relatively high pre-Hurricane Katrina CAPs (BHC=0 and preCAP=1) significantly increase their risk-based capital ratios. The increase for this group of banks is 1.87 percentage points. We also find a significant increase for the group of BHC banks with relatively low pre-Hurricane Katrina CAPs (BHC=1 and preCAP=0), but the increase is much smaller (0.39 percentage points). The second column of Table VI shows regression results using the natural logarithm of total capital as dependent variable. We find no significant effect for any of the four groups of affected banks, which indicates that our baseline results are not driven by a change in banks’—not risk-adjusted—capital ratios. Next, we use the natural logarithm of risk-weighted assets as dependent variable in the regression. Results in Column (3) of Table VI show a significantly negative coefficient of the interaction term for the group of independent banks with a high pre-Hurricane Katrina CAP (BHC=0 and preCAP=1), that is, a decrease of 8.45% compared with the control group. This suggests that these banks achieve higher risk-based capital ratios after Hurricane Katrina, as documented in Column (1), by reducing their exposure to risky assets. In the following, we explore different asset categories that determine the risk-weighted assets of banks. In particular, U.S. government securities have a risk-weight of zero, real estate loans have a risk weight between 50% and 100% (e.g., typically 50% for loans to individuals or families, and 100% for commercial real estate loans), and consumer loans, as well as commercial and industrial loans, have a risk weight of 100%. Regression results with the natural logarithm of U.S. government securities as dependent variable are shown in Column (4) of Table VI. Again, we find a significant and most pronounced effect for independent banks with a relatively high pre-Hurricane Katrina CAP (BHC =0 and preCAP =1). These banks increase their holdings of government securities by 42.5% after the 2005 hurricane season compared with the control group. We also find a significant and positive increase of 21.3% for the group of BHC banks that are relatively low capitalized (BHC =1 and preCAP=0). We now take a closer look at the exposure to different loan categories: real estate loans (RE), commercial real estate loans (RECO), reconstruction and development loans (RECON), consumer loans (CON), and commercial and industrial loans (C&I). We do not find significant difference-in-difference effects (Affected × Event) for the first four loan categories, but we do so for C&I loans. In particular, affected independent banks with a relatively high pre-Hurricane Katrina CAP (BHC=0 and preCAP=1) decrease their exposure to C&I loans by 16.87% compared with unaffected banks, as shown at the bottom of Table VI. The effect is not significant for the other three subgroups of banks. Overall, the evidence in this section shows that the effects of the 2005 hurricane season are most pronounced for independent banks with relatively high pre-Hurricane Katrina risk-based capital ratios. These banks decrease their risk-weighted assets, which is explained by increased exposures to U.S. government securities and decreased exposures to commercial and industrial loans compared with the control group. 4.5 The Lending Activities of Banks Results in the previous section lead to the question of whether declines in commercial and industrial loan exposures of independent and highly capitalized banks are driven by a reduction in lending or other strategies compared with the control group. To shed more light on this issue, this section provides evidence on new lending transactions of banks before and after the 2005 hurricane season. The data for this analysis come from the U.S. SBA 7(a) loan program data set, which provides lending transaction data including bank name, borrower name and location, date, and approved lending volume.23 The SBA typically provides guarantees on parts of the loans in order to strengthen access to finance for small businesses. Information about the gross approved lending volume and the SBA guaranteed part of the loan—in most cases between 50% and 75%—is also included in the data set. Borrowers may use such loans to establish a new business or to expand an existing business. Most importantly, the loans are made through financial institutions, which assess whether borrowers are financially eligible, use the funds for a sound business purpose, and fulfill all other requirements of the program.24 A restriction of this analysis is that not all banks are included in the SBA data set, which reduces the sample for this analysis from 1,253 to 337 independent and BHC banks. The analysis differentiates between a bank’s gross volume of approved SBA loans and the volume of SBA loans net of guarantees. Furthermore, we differentiate between a bank’s lending in its core market and non-core market. We define a bank’s core market as all counties where a bank has established a branch as of June 2005 and all other counties as non-core markets.25 This results in six variables on the bank-year level: (1a) the total volume of a bank’s gross SBA lending, (1b) the volume of a bank’s gross SBA lending in its core market, (1c) the volume of a bank’s gross SBA lending in its non-core market, (2a) the total volume of a bank’s SBA lending net of guarantees, (2b) the volume of a bank’s SBA lending net of guarantees in its core market, and (2c) the volume of a bank’s SBA lending net of guarantees in its non-core market. We then use the natural logarithm of these variables as dependent variables in a regression model equivalent to Equation (4). Regression results are shown in Table VII. The difference-in-difference effects (Event × Affected) that are shown at the bottom of the table provide interesting results. We observe significant effects for the group of affected independent banks with relatively high pre-Katrina CAPs (BHC = 0 and preCAP =1), but not for the other groups. The former significantly increase total SBA lending, as well as SBA lending in their core markets, but not in their non-core markets. These results are similar for gross SBA loans [Columns (1)–(3)] and SBA loans net of guarantees [Columns (4)–(6)]. Table VII. Lending activities (SBA loans) These regressions explore the development of banks’ lending activities following the 2005 hurricane season, based on data from the U.S. SBA loan program. We use the natural logarithm of a bank’s amount of total SBA loans as dependent variable in the first three columns, and the natural logarithm of a bank’s amount of SBA loans net of guarantees as dependent variable in the last three columns. Furthermore, we differentiate between the total amount per bank, the amount lent to borrowers in the banks’ core markets (where they own a branch), and the amount lent to borrowers in the banks’ non-core markets (where they do not own a branch). The regression model is equivalent to the model used in the previous subsection [see Equation (4)], where the terms Event, Affected, and Event×Affected are interacted with two variables: a dummy variable that indicates independent banks (BHC = 0) or banks that belong to a bank holding company (BHC = 1), and a dummy variable which is 0 if a bank’s average risk-based capital ratio during the eight quarters before the event is below the sample median (preCap = 0) and 1 otherwise (preCap = 1). The sample includes yearly data for all banks in Alabama, Louisiana, Mississippi, Florida, Texas, Georgia, Tennessee, Arkansas, and Oklahoma for which SBA data are available over the period of ± 2 years around the 2005 hurricane season (2003–2007). Bank-fixed effects (bank FE) and year-fixed effects (time FE) are included in all regressions. Standard errors are clustered at the bank level and reported in parentheses. The ***, **, and * stand for significant coefficients at the 1%, 5%, and 10% levels, respectively. log(SBA loans) log(SBA loans net of guarantees) Total Core markets Non-core markets Total Core markets Non-core markets (1) (2) (3) (4) (5) (6) Event×Affected 2.0870 −0.1103 2.3074 1.6826 −0.1719 1.9171 (2.5550) (2.4659) (2.0010) (2.2748) (2.1885) (1.7849) Event×BHC 0.9809 −0.4161 0.8428 0.8271 −0.3584 0.6810 (1.1212) (1.0835) (1.2246) (0.9931) (0.9423) (1.0811) Event×preCAP −2.6019 −1.8358 −0.6256 −2.3282 −1.6041 −0.6174 (1.6721) (1.6997) (1.9058) (1.5260) (1.5161) (1.7002) Event×Affected×BHC −0.7373 1.1709 −2.6882 −0.5559 1.0298 −2.2844 (2.7042) (2.6477) (2.1754) (2.4105) (2.3536) (1.9414) Event×Affected×preCAP 2.5299 5.6157 −0.0495 2.5885 5.1589 0.0485 (3.6816) (3.6827) (3.3901) (3.3535) (3.3271) (3.0400) Event×BHC×preCAP 2.3016 1.8918 0.3937 2.0926 1.6240 0.4615 (1.7840) (1.8130) (1.9916) (1.6243) (1.6156) (1.7775) Event×Affected×BHC×preCAP −4.4513 −6.7012* −0.0018 −4.2878 −6.0704* −0.1384 (3.9003) (3.9313) (3.5986) (3.5424) (3.5404) (3.2238) Bank FE Yes Yes Yes Yes Yes Yes Time FE Yes Yes Yes Yes Yes Yes Number of banks 337 337 337 337 337 337 Number of observations 1,341 1,341 1,341 1,341 1,341 1,341 Adjusted R2 0.4252 0.4369 0.5605 0.4445 0.4542 0.5722 Within R2 0.0096 0.0060 0.0033 0.0094 0.0062 0.0032 Difference-in-difference effects for subgroups of banks Event×Affected for BHC=0, preCAP=0 2.0870 −0.1103 2.3074 1.6826 −0.1719 1.9171 (0.4148) (0.9644) (0.2499) (0.4602) (0.9375) (0.2837) Event×Affected for BHC=0, preCAP=1 4.6169* 5.5054** 2.2579 4.2711* 4.9870** 1.9655 (0.0826) (0.0450) (0.4101) (0.0840) (0.0475) (0.4252) Event×Affected for BHC=1, preCAP=0 1.3497 1.0606 −0.3808 1.1267 0.8579 −0.3674 (0.1291) (0.2725) (0.6552) (0.1591) (0.3230) (0.6303) Event×Affected for BHC=1, preCAP=1 −0.5717 −0.0249 −0.4321 −0.5726 −0.0536 −0.4573 (0.5410) (0.9798) (0.6125) (0.4836) (0.9495) (0.5436) log(SBA loans) log(SBA loans net of guarantees) Total Core markets Non-core markets Total Core markets Non-core markets (1) (2) (3) (4) (5) (6) Event×Affected 2.0870 −0.1103 2.3074 1.6826 −0.1719 1.9171 (2.5550) (2.4659) (2.0010) (2.2748) (2.1885) (1.7849) Event×BHC 0.9809 −0.4161 0.8428 0.8271 −0.3584 0.6810 (1.1212) (1.0835) (1.2246) (0.9931) (0.9423) (1.0811) Event×preCAP −2.6019 −1.8358 −0.6256 −2.3282 −1.6041 −0.6174 (1.6721) (1.6997) (1.9058) (1.5260) (1.5161) (1.7002) Event×Affected×BHC −0.7373 1.1709 −2.6882 −0.5559 1.0298 −2.2844 (2.7042) (2.6477) (2.1754) (2.4105) (2.3536) (1.9414) Event×Affected×preCAP 2.5299 5.6157 −0.0495 2.5885 5.1589 0.0485 (3.6816) (3.6827) (3.3901) (3.3535) (3.3271) (3.0400) Event×BHC×preCAP 2.3016 1.8918 0.3937 2.0926 1.6240 0.4615 (1.7840) (1.8130) (1.9916) (1.6243) (1.6156) (1.7775) Event×Affected×BHC×preCAP −4.4513 −6.7012* −0.0018 −4.2878 −6.0704* −0.1384 (3.9003) (3.9313) (3.5986) (3.5424) (3.5404) (3.2238) Bank FE Yes Yes Yes Yes Yes Yes Time FE Yes Yes Yes Yes Yes Yes Number of banks 337 337 337 337 337 337 Number of observations 1,341 1,341 1,341 1,341 1,341 1,341 Adjusted R2 0.4252 0.4369 0.5605 0.4445 0.4542 0.5722 Within R2 0.0096 0.0060 0.0033 0.0094 0.0062 0.0032 Difference-in-difference effects for subgroups of banks Event×Affected for BHC=0, preCAP=0 2.0870 −0.1103 2.3074 1.6826 −0.1719 1.9171 (0.4148) (0.9644) (0.2499) (0.4602) (0.9375) (0.2837) Event×Affected for BHC=0, preCAP=1 4.6169* 5.5054** 2.2579 4.2711* 4.9870** 1.9655 (0.0826) (0.0450) (0.4101) (0.0840) (0.0475) (0.4252) Event×Affected for BHC=1, preCAP=0 1.3497 1.0606 −0.3808 1.1267 0.8579 −0.3674 (0.1291) (0.2725) (0.6552) (0.1591) (0.3230) (0.6303) Event×Affected for BHC=1, preCAP=1 −0.5717 −0.0249 −0.4321 −0.5726 −0.0536 −0.4573 (0.5410) (0.9798) (0.6125) (0.4836) (0.9495) (0.5436) Table VII. Lending activities (SBA loans) These regressions explore the development of banks’ lending activities following the 2005 hurricane season, based on data from the U.S. SBA loan program. We use the natural logarithm of a bank’s amount of total SBA loans as dependent variable in the first three columns, and the natural logarithm of a bank’s amount of SBA loans net of guarantees as dependent variable in the last three columns. Furthermore, we differentiate between the total amount per bank, the amount lent to borrowers in the banks’ core markets (where they own a branch), and the amount lent to borrowers in the banks’ non-core markets (where they do not own a branch). The regression model is equivalent to the model used in the previous subsection [see Equation (4)], where the terms Event, Affected, and Event×Affected are interacted with two variables: a dummy variable that indicates independent banks (BHC = 0) or banks that belong to a bank holding company (BHC = 1), and a dummy variable which is 0 if a bank’s average risk-based capital ratio during the eight quarters before the event is below the sample median (preCap = 0) and 1 otherwise (preCap = 1). The sample includes yearly data for all banks in Alabama, Louisiana, Mississippi, Florida, Texas, Georgia, Tennessee, Arkansas, and Oklahoma for which SBA data are available over the period of ± 2 years around the 2005 hurricane season (2003–2007). Bank-fixed effects (bank FE) and year-fixed effects (time FE) are included in all regressions. Standard errors are clustered at the bank level and reported in parentheses. The ***, **, and * stand for significant coefficients at the 1%, 5%, and 10% levels, respectively. log(SBA loans) log(SBA loans net of guarantees) Total Core markets Non-core markets Total Core markets Non-core markets (1) (2) (3) (4) (5) (6) Event×Affected 2.0870 −0.1103 2.3074 1.6826 −0.1719 1.9171 (2.5550) (2.4659) (2.0010) (2.2748) (2.1885) (1.7849) Event×BHC 0.9809 −0.4161 0.8428 0.8271 −0.3584 0.6810 (1.1212) (1.0835) (1.2246) (0.9931) (0.9423) (1.0811) Event×preCAP −2.6019 −1.8358 −0.6256 −2.3282 −1.6041 −0.6174 (1.6721) (1.6997) (1.9058) (1.5260) (1.5161) (1.7002) Event×Affected×BHC −0.7373 1.1709 −2.6882 −0.5559 1.0298 −2.2844 (2.7042) (2.6477) (2.1754) (2.4105) (2.3536) (1.9414) Event×Affected×preCAP 2.5299 5.6157 −0.0495 2.5885 5.1589 0.0485 (3.6816) (3.6827) (3.3901) (3.3535) (3.3271) (3.0400) Event×BHC×preCAP 2.3016 1.8918 0.3937 2.0926 1.6240 0.4615 (1.7840) (1.8130) (1.9916) (1.6243) (1.6156) (1.7775) Event×Affected×BHC×preCAP −4.4513 −6.7012* −0.0018 −4.2878 −6.0704* −0.1384 (3.9003) (3.9313) (3.5986) (3.5424) (3.5404) (3.2238) Bank FE Yes Yes Yes Yes Yes Yes Time FE Yes Yes Yes Yes Yes Yes Number of banks 337 337 337 337 337 337 Number of observations 1,341 1,341 1,341 1,341 1,341 1,341 Adjusted R2 0.4252 0.4369 0.5605 0.4445 0.4542 0.5722 Within R2 0.0096 0.0060 0.0033 0.0094 0.0062 0.0032 Difference-in-difference effects for subgroups of banks Event×Affected for BHC=0, preCAP=0 2.0870 −0.1103 2.3074 1.6826 −0.1719 1.9171 (0.4148) (0.9644) (0.2499) (0.4602) (0.9375) (0.2837) Event×Affected for BHC=0, preCAP=1 4.6169* 5.5054** 2.2579 4.2711* 4.9870** 1.9655 (0.0826) (0.0450) (0.4101) (0.0840) (0.0475) (0.4252) Event×Affected for BHC=1, preCAP=0 1.3497 1.0606 −0.3808 1.1267 0.8579 −0.3674 (0.1291) (0.2725) (0.6552) (0.1591) (0.3230) (0.6303) Event×Affected for BHC=1, preCAP=1 −0.5717 −0.0249 −0.4321 −0.5726 −0.0536 −0.4573 (0.5410) (0.9798) (0.6125) (0.4836) (0.9495) (0.5436) log(SBA loans) log(SBA loans net of guarantees) Total Core markets Non-core markets Total Core markets Non-core markets (1) (2) (3) (4) (5) (6) Event×Affected 2.0870 −0.1103 2.3074 1.6826 −0.1719 1.9171 (2.5550) (2.4659) (2.0010) (2.2748) (2.1885) (1.7849) Event×BHC 0.9809 −0.4161 0.8428 0.8271 −0.3584 0.6810 (1.1212) (1.0835) (1.2246) (0.9931) (0.9423) (1.0811) Event×preCAP −2.6019 −1.8358 −0.6256 −2.3282 −1.6041 −0.6174 (1.6721) (1.6997) (1.9058) (1.5260) (1.5161) (1.7002) Event×Affected×BHC −0.7373 1.1709 −2.6882 −0.5559 1.0298 −2.2844 (2.7042) (2.6477) (2.1754) (2.4105) (2.3536) (1.9414) Event×Affected×preCAP 2.5299 5.6157 −0.0495 2.5885 5.1589 0.0485 (3.6816) (3.6827) (3.3901) (3.3535) (3.3271) (3.0400) Event×BHC×preCAP 2.3016 1.8918 0.3937 2.0926 1.6240 0.4615 (1.7840) (1.8130) (1.9916) (1.6243) (1.6156) (1.7775) Event×Affected×BHC×preCAP −4.4513 −6.7012* −0.0018 −4.2878 −6.0704* −0.1384 (3.9003) (3.9313) (3.5986) (3.5424) (3.5404) (3.2238) Bank FE Yes Yes Yes Yes Yes Yes Time FE Yes Yes Yes Yes Yes Yes Number of banks 337 337 337 337 337 337 Number of observations 1,341 1,341 1,341 1,341 1,341 1,341 Adjusted R2 0.4252 0.4369 0.5605 0.4445 0.4542 0.5722 Within R2 0.0096 0.0060 0.0033 0.0094 0.0062 0.0032 Difference-in-difference effects for subgroups of banks Event×Affected for BHC=0, preCAP=0 2.0870 −0.1103 2.3074 1.6826 −0.1719 1.9171 (0.4148) (0.9644) (0.2499) (0.4602) (0.9375) (0.2837) Event×Affected for BHC=0, preCAP=1 4.6169* 5.5054** 2.2579 4.2711* 4.9870** 1.9655 (0.0826) (0.0450) (0.4101) (0.0840) (0.0475) (0.4252) Event×Affected for BHC=1, preCAP=0 1.3497 1.0606 −0.3808 1.1267 0.8579 −0.3674 (0.1291) (0.2725) (0.6552) (0.1591) (0.3230) (0.6303) Event×Affected for BHC=1, preCAP=1 −0.5717 −0.0249 −0.4321 −0.5726 −0.0536 −0.4573 (0.5410) (0.9798) (0.6125) (0.4836) (0.9495) (0.5436) A plausible explanation for this finding is that independent banks may be more focused on their core markets compared with BHC banks which are more diversified. They may be better able to collect and process information and have more incentives to do so. Empirical evidence by Berger et al. (2005) suggests that smaller banks have stronger relationships with their borrowers and alleviate credit constraints more effectively than larger banks. Berger, Bouwman, and Kim (2017) show that such comparative advantages of smaller banks over larger banks are even stronger when economic conditions are adverse. In addition, Loutskina and Strahan (2011) find that lenders that are concentrated in few markets are more active in information-sensitive segments of the mortgage market than more diversified lenders. The 2005 hurricane season caused massive destruction of property and uncertainty about the financial situation and business prospects of borrowers. Hence, information about the creditworthiness of borrowers was difficult to assess.26 Furthermore, our results add to existing evidence that shows that bank capital is important for bank lending in times of crisis. In particular, relatively high capital ratios before the 2005 hurricane season are associated with relatively more SBA lending in the subsequent years. Interestingly, our evidence shows that affected banks that are independent and well-capitalized both decrease their exposure to C&I loans (Section 4.4) and increase their lending to small businesses in their core-markets (this section) compared with the control group. Hence, these banks reduce their risk-weighted assets, but not through reduced new lending. These banks presumably engage in loan sales or loan securitization. Unfortunately, such data are very limited for the C&I loans of independent banks, and therefore we cannot test this directly. However, a related study by Chavaz (2016) analyzes mortgage loans from the Home Mortgage Disclosure Act data set. The study finds that more local banks originate a higher share and also sell a higher share of these loans in the aftermath of Hurricane Katrina in 2005 compared with more diversified banks. It is also plausible for our results that independent and well-capitalized banks at the same time increase new C&I lending in their core-markets and reduce balance sheet exposures to C&I loans through loan sales. 4.6 The Structure of the Local Banking System and Economic Development Our previous results reveal that bank behavior in the aftermath of the 2005 hurricane season differs across banks depending on a bank’s pre-hurricane capitalization and structure. This may have effects on the real economy. Hence, we explore in this section whether the structure of the local banking system matters for economic development in counties affected by the 2005 hurricane season. Note that variation across local banking structures is not random because it develops jointly with the local economy. The following analysis thus provides evidence on the relationship between banking structure and economic development, and not necessarily evidence on a causal effect of banking structure on economic development.27 Related research by Cortes (2014) shows that the presence of local lenders is beneficial for job creation following a natural disaster. Furthermore, studies by Cecchetti, King, and Yetman (2011) and Jordà et al. (2017) suggest that a higher average bank capital ratio of a country’s financial sector at the start of a financial-crisis recession is associated with a significantly stronger recovery after the crisis. The analysis in this section contributes to this evidence. The following analysis uses county-level data, in contrast to the bank-level analysis in the previous sections. Counties are classified as affected if they were designated for disaster assistance by FEMA, corresponding to the description in Section 3.1 (see also Figure 4). The economic development in affected and unaffected counties over the period 2003–2007 is illustrated in Figure 5. The graph on the left shows the development of the average total personal income. The graph does not reveal any significant effects of the 2005 hurricane season. Further, the graph on the right of Figure 5 shows the average unemployment rate in affected and unaffected counties over the sample period. Here, we can observe a significant increase in 2005 when the hurricanes hit the U.S. Gulf Coast, as well as a significant decrease in the following years. Note that the unemployment rate alone is an intricate measure because of the migration effects in the aftermath of a natural disaster.28 Hence, we also consider the number of employed persons, the number of unemployed persons, and the total labor force (sum of employed and unemployed persons) in the regression analysis. Figure 5. View largeDownload slide Economic developments in affected and unaffected counties. Notes: This figure illustrates economic developments in counties affected (solid lines) and unaffected (dashed lines) by the 2005 hurricane season. The left graph shows the average of counties’ total personal income (in US$ millions) and the right graph shows the average of counties’ unemployment rates. Figure 5. View largeDownload slide Economic developments in affected and unaffected counties. Notes: This figure illustrates economic developments in counties affected (solid lines) and unaffected (dashed lines) by the 2005 hurricane season. The left graph shows the average of counties’ total personal income (in US$ millions) and the right graph shows the average of counties’ unemployment rates. We consider the share of banks belonging to a bank holding company (BHCsharej) and the average of risk-based capital ratios for all banks with a branch in county j (CAPaveragej) before the 2005 hurricane season as measures of the structure of the local banking system in county j. The measures are calculated using a bank’s sum of deposits by county—a proxy for a bank’s local presence and business activities—as weights. Both variables are demeaned.29 Consequently, the “average” county has values of BHCsharej = 0 and CAPaveragej = 0. When we compare counties with a value of BHCshare (before demeaning) at the 25th and 75th percentiles, this corresponds to 74% and 100% of banks belonging to a bank holding company, respectively. Hence, a value of BHCshare at the 25th percentile can also be interpreted as a more diversified banking system compared with a value of BHCshare at the 75th percentile. The values of CAPaverage (before demeaning) at the 25th and 75th percentiles are 12.35% and 17.09%, respectively. We estimate the following fixed effects OLS model for the 2 years before and after the 2005 hurricane season: Yjy= νj+τy+β1(Eventy×Affectedj)+β2(Eventy×BHCsharej)+β3(Eventy×Affectedj×BHCsharej)+β4(Eventy×CAPaveragej)+β5(Eventy×Affectedj×CAPaveragej)+β6(Eventy×BHCsharej×CAPaveragej)+β7(Eventy×Affectedj×BHCsharej×CAPaveragej)+ϵjy, (5) where Yjy stands for alternative measures of economic development in county j at year y: the natural logarithms of personal income, number of employed persons, number of unemployed persons, number of persons in the labor force and—not in logs—the unemployment rate. The terms νj and τy represent county-fixed effects and yearly time-fixed effects, respectively. Eventy is a dummy variable that is 0 for 2003 and 2004, and 1 for 2006 and 2007. Affectedj is a dummy that categorizes counties as affected or unaffected by the hurricanes. The variables BHCsharej and CAPaveragej reflect the demeaned share of BHC banks and average bank risk-based capital ratio per county before the 2005 hurricane season. The term ϵiy represents the idiosyncratic error term. Standard errors are clustered at the county level. Regression results are provided in Table VIII. First, consider a county with an average share of BHC banks and average risk-based capital ratios of banks (BHCshare=0 and CAPaverage=0). We find that personal income increases significantly by 5.45% in affected counties after the 2005 hurricane season compared with unaffected counties, as reflected in the coefficient of Event × Affected in Column (1). The next column shows regression results for the log of the number of employed persons as the dependent variable. The coefficient of Event×Affected is positive, but not significant. Column (3) shows that the number of unemployed persons decreases significantly by 10.1%. The effect on the labor force, as shown in Column (4), is insignificant. Finally, Column (5) shows that the unemployment rate decreases significantly by 0.45 percentage points in the average affected county after the 2005 hurricane season compared with the control group. Overall, we observe a positive effect of the 2005 hurricane season on economic activity. This presumably reflects reconstruction activities and transfer payments to affected regions. Table VIII. Effects on growth and employment These regressions explore how the structure of the local banking system matters for economic development in counties affected by the 2005 hurricane season. The following measures of economic activity on the county level are used as dependent variables: total personal income; number of employed persons; number of unemployed; number of persons in the labor force (all expressed as natural logarithms); and unemployment rate (UR). The model is specified in Equation (5). Event is a dummy variable that is 0 for the pre-hurricane period and 1 after the hurricane season. Affected is a dummy variable that separates affected counties (Affected = 1) from unaffected counties (Affected = 0). Event×Affected is an interaction term for the variables Event and Affected. The terms Event, Affected, and Event×Affected are interacted with two continuous variables: a variable that indicates the share of banks that belong to a bank holding company per county (BHCshare), and a variable that indicates banks’ average risk-based capital ratio per county (CAPaverage). The sample includes yearly data for all counties in Alabama, Louisiana, Mississippi, Florida, Texas, Georgia, Tennessee, Arkansas, and Oklahoma over the period ± 2 years around the 2005 hurricane season (2003–2007). County-fixed effects (county FE) and year-fixed effects (time FE) are included in all regressions. Standard errors are clustered at the county level and reported in parentheses. The ***, **, and * stand for significant coefficients at the 1%, 5%, and 10% levels, respectively. log (personal income) log (employed) log (unemployed) log (labor force) UR (1) (2) (3) (4) (5) Event×Affected 0.0545*** 0.0134 −0.1014*** 0.0087 −0.0045** (0.0098) (0.0098) (0.0339) (0.0099) (0.0019) Event×BHCshare 0.0344 0.0188 0.0317 0.0179 −0.0007 (0.0279) (0.0210) (0.0846) (0.0200) (0.0053) Event×Affected×BHCshare −0.1002** −0.0589 −0.1631 −0.0664 −0.0070 (0.0454) (0.0448) (0.1982) (0.0475) (0.0107) Event×CAPaverage −0.7880*** −0.6560*** −1.2509** −0.7278*** −0.0647* (0.1366) (0.1401) (0.5619) (0.1323) (0.0376) Event×Affected×CAPaverage 0.4493* 0.5990** 0.2895 0.5844** −0.0149 (0.2355) (0.2544) (0.7933) (0.2524) (0.0447) Event×BHCshare×CAPaverage −1.4290** −0.6531 −3.5332 −0.8180* −0.1520 (0.6466) (0.5221) (2.5921) (0.4908) (0.1679) Event×Affected×BHCshare×CAPaverage 2.0841** −0.0134 2.6334 0.1324 0.1336 (1.0386) (1.1763) (4.0736) (1.2037) (0.2190) County FE Yes Yes Yes Yes Yes Time FE Yes Yes Yes Yes Yes Number of counties 176 176 176 176 176 Number of observations 704 704 704 704 704 Adjusted R2 0.9990 0.9990 0.9800 0.9990 0.7117 Within R2 0.2191 0.0849 0.0588 0.0977 0.0510 Difference-in-difference effects for alternative values of BHCshare and CAPaverage Event×Affected for the 25th and 25th percentiles 0.0601*** 0.0030 −0.0822 0.0002 −0.0028 (0.0139) (0.0149) (0.0519) (0.0153) (0.0027) Event×Affected for the 25th and 75th percentiles 0.0701*** 0.0319*** −0.0829* 0.0276*** −0.0042* (0.0105) (0.0100) (0.0461) (0.0104) (0.0026) Event×Affected for the 75th and 25th percentiles 0.0192 −0.0119 −0.1432*** −0.0178 −0.0056* (0.0175) (0.0194) (0.0547) (0.0195) (0.0029) Event×Affected for the 75th and 75th percentiles 0.0549*** 0.0168* −0.1115*** 0.0112 −0.0054** (0.0115) (0.0101) (0.0409) (0.0100) (0.0025) log (personal income) log (employed) log (unemployed) log (labor force) UR (1) (2) (3) (4) (5) Event×Affected 0.0545*** 0.0134 −0.1014*** 0.0087 −0.0045** (0.0098) (0.0098) (0.0339) (0.0099) (0.0019) Event×BHCshare 0.0344 0.0188 0.0317 0.0179 −0.0007 (0.0279) (0.0210) (0.0846) (0.0200) (0.0053) Event×Affected×BHCshare −0.1002** −0.0589 −0.1631 −0.0664 −0.0070 (0.0454) (0.0448) (0.1982) (0.0475) (0.0107) Event×CAPaverage −0.7880*** −0.6560*** −1.2509** −0.7278*** −0.0647* (0.1366) (0.1401) (0.5619) (0.1323) (0.0376) Event×Affected×CAPaverage 0.4493* 0.5990** 0.2895 0.5844** −0.0149 (0.2355) (0.2544) (0.7933) (0.2524) (0.0447) Event×BHCshare×CAPaverage −1.4290** −0.6531 −3.5332 −0.8180* −0.1520 (0.6466) (0.5221) (2.5921) (0.4908) (0.1679) Event×Affected×BHCshare×CAPaverage 2.0841** −0.0134 2.6334 0.1324 0.1336 (1.0386) (1.1763) (4.0736) (1.2037) (0.2190) County FE Yes Yes Yes Yes Yes Time FE Yes Yes Yes Yes Yes Number of counties 176 176 176 176 176 Number of observations 704 704 704 704 704 Adjusted R2 0.9990 0.9990 0.9800 0.9990 0.7117 Within R2 0.2191 0.0849 0.0588 0.0977 0.0510 Difference-in-difference effects for alternative values of BHCshare and CAPaverage Event×Affected for the 25th and 25th percentiles 0.0601*** 0.0030 −0.0822 0.0002 −0.0028 (0.0139) (0.0149) (0.0519) (0.0153) (0.0027) Event×Affected for the 25th and 75th percentiles 0.0701*** 0.0319*** −0.0829* 0.0276*** −0.0042* (0.0105) (0.0100) (0.0461) (0.0104) (0.0026) Event×Affected for the 75th and 25th percentiles 0.0192 −0.0119 −0.1432*** −0.0178 −0.0056* (0.0175) (0.0194) (0.0547) (0.0195) (0.0029) Event×Affected for the 75th and 75th percentiles 0.0549*** 0.0168* −0.1115*** 0.0112 −0.0054** (0.0115) (0.0101) (0.0409) (0.0100) (0.0025) Table VIII. Effects on growth and employment These regressions explore how the structure of the local banking system matters for economic development in counties affected by the 2005 hurricane season. The following measures of economic activity on the county level are used as dependent variables: total personal income; number of employed persons; number of unemployed; number of persons in the labor force (all expressed as natural logarithms); and unemployment rate (UR). The model is specified in Equation (5). Event is a dummy variable that is 0 for the pre-hurricane period and 1 after the hurricane season. Affected is a dummy variable that separates affected counties (Affected = 1) from unaffected counties (Affected = 0). Event×Affected is an interaction term for the variables Event and Affected. The terms Event, Affected, and Event×Affected are interacted with two continuous variables: a variable that indicates the share of banks that belong to a bank holding company per county (BHCshare), and a variable that indicates banks’ average risk-based capital ratio per county (CAPaverage). The sample includes yearly data for all counties in Alabama, Louisiana, Mississippi, Florida, Texas, Georgia, Tennessee, Arkansas, and Oklahoma over the period ± 2 years around the 2005 hurricane season (2003–2007). County-fixed effects (county FE) and year-fixed effects (time FE) are included in all regressions. Standard errors are clustered at the county level and reported in parentheses. The ***, **, and * stand for significant coefficients at the 1%, 5%, and 10% levels, respectively. log (personal income) log (employed) log (unemployed) log (labor force) UR (1) (2) (3) (4) (5) Event×Affected 0.0545*** 0.0134 −0.1014*** 0.0087 −0.0045** (0.0098) (0.0098) (0.0339) (0.0099) (0.0019) Event×BHCshare 0.0344 0.0188 0.0317 0.0179 −0.0007 (0.0279) (0.0210) (0.0846) (0.0200) (0.0053) Event×Affected×BHCshare −0.1002** −0.0589 −0.1631 −0.0664 −0.0070 (0.0454) (0.0448) (0.1982) (0.0475) (0.0107) Event×CAPaverage −0.7880*** −0.6560*** −1.2509** −0.7278*** −0.0647* (0.1366) (0.1401) (0.5619) (0.1323) (0.0376) Event×Affected×CAPaverage 0.4493* 0.5990** 0.2895 0.5844** −0.0149 (0.2355) (0.2544) (0.7933) (0.2524) (0.0447) Event×BHCshare×CAPaverage −1.4290** −0.6531 −3.5332 −0.8180* −0.1520 (0.6466) (0.5221) (2.5921) (0.4908) (0.1679) Event×Affected×BHCshare×CAPaverage 2.0841** −0.0134 2.6334 0.1324 0.1336 (1.0386) (1.1763) (4.0736) (1.2037) (0.2190) County FE Yes Yes Yes Yes Yes Time FE Yes Yes Yes Yes Yes Number of counties 176 176 176 176 176 Number of observations 704 704 704 704 704 Adjusted R2 0.9990 0.9990 0.9800 0.9990 0.7117 Within R2 0.2191 0.0849 0.0588 0.0977 0.0510 Difference-in-difference effects for alternative values of BHCshare and CAPaverage Event×Affected for the 25th and 25th percentiles 0.0601*** 0.0030 −0.0822 0.0002 −0.0028 (0.0139) (0.0149) (0.0519) (0.0153) (0.0027) Event×Affected for the 25th and 75th percentiles 0.0701*** 0.0319*** −0.0829* 0.0276*** −0.0042* (0.0105) (0.0100) (0.0461) (0.0104) (0.0026) Event×Affected for the 75th and 25th percentiles 0.0192 −0.0119 −0.1432*** −0.0178 −0.0056* (0.0175) (0.0194) (0.0547) (0.0195) (0.0029) Event×Affected for the 75th and 75th percentiles 0.0549*** 0.0168* −0.1115*** 0.0112 −0.0054** (0.0115) (0.0101) (0.0409) (0.0100) (0.0025) log (personal income) log (employed) log (unemployed) log (labor force) UR (1) (2) (3) (4) (5) Event×Affected 0.0545*** 0.0134 −0.1014*** 0.0087 −0.0045** (0.0098) (0.0098) (0.0339) (0.0099) (0.0019) Event×BHCshare 0.0344 0.0188 0.0317 0.0179 −0.0007 (0.0279) (0.0210) (0.0846) (0.0200) (0.0053) Event×Affected×BHCshare −0.1002** −0.0589 −0.1631 −0.0664 −0.0070 (0.0454) (0.0448) (0.1982) (0.0475) (0.0107) Event×CAPaverage −0.7880*** −0.6560*** −1.2509** −0.7278*** −0.0647* (0.1366) (0.1401) (0.5619) (0.1323) (0.0376) Event×Affected×CAPaverage 0.4493* 0.5990** 0.2895 0.5844** −0.0149 (0.2355) (0.2544) (0.7933) (0.2524) (0.0447) Event×BHCshare×CAPaverage −1.4290** −0.6531 −3.5332 −0.8180* −0.1520 (0.6466) (0.5221) (2.5921) (0.4908) (0.1679) Event×Affected×BHCshare×CAPaverage 2.0841** −0.0134 2.6334 0.1324 0.1336 (1.0386) (1.1763) (4.0736) (1.2037) (0.2190) County FE Yes Yes Yes Yes Yes Time FE Yes Yes Yes Yes Yes Number of counties 176 176 176 176 176 Number of observations 704 704 704 704 704 Adjusted R2 0.9990 0.9990 0.9800 0.9990 0.7117 Within R2 0.2191 0.0849 0.0588 0.0977 0.0510 Difference-in-difference effects for alternative values of BHCshare and CAPaverage Event×Affected for the 25th and 25th percentiles 0.0601*** 0.0030 −0.0822 0.0002 −0.0028 (0.0139) (0.0149) (0.0519) (0.0153) (0.0027) Event×Affected for the 25th and 75th percentiles 0.0701*** 0.0319*** −0.0829* 0.0276*** −0.0042* (0.0105) (0.0100) (0.0461) (0.0104) (0.0026) Event×Affected for the 75th and 25th percentiles 0.0192 −0.0119 −0.1432*** −0.0178 −0.0056* (0.0175) (0.0194) (0.0547) (0.0195) (0.0029) Event×Affected for the 75th and 75th percentiles 0.0549*** 0.0168* −0.1115*** 0.0112 −0.0054** (0.0115) (0.0101) (0.0409) (0.0100) (0.0025) Next, consider the role of banking structure, BHCshare, conditional on an average capitalization of banks (CAPaverage = 0), as reflected in the coefficient of the triple interaction term Event × Affected × BHCshare. Results in Column (1) show that a relatively lower share of BHC banks (higher share of independent banks) is beneficial for total personal income in affected counties after the 2005 hurricane season. For example, a share of BHC banks that is 10 percentage points lower is associated with a 1 percentage point higher increase in total personal income in affected counties following the 2005 hurricane season (−0.10 × (−0.1002)). The coefficients of Event × Affected × BHCshare are not significant in the regressions in Columns (2)–(5), where labor market variables are used as dependent variables. Moreover, consider the role of bank capitalization before the 2005 hurricane season, CAPaverage, conditional on an average share of BHC banks (BHCshare =0), as reflected in the coefficient of the triple interaction term Event × Affected × CAPaverage. We find a significant positive effect on personal income [Column (1)], the number of employed persons [Column (2)], and the number of persons in the labor force [Column (4)]. For example, a 1 percentage point higher average bank capital ratio in an affected county is associated with a 0.45 percentage point higher increase in total personal income (0.01 × 0.4493), a 0.60 percentage point higher increase in employed persons (0.01 × 0.5990), and a 0.58 percentage point higher increase in labor force (0.01× 0.5844) after the 2005 hurricane season compared with the “average” affected county with CAPaverage = 0. Finally, we turn to the bottom of Table VIII, where difference-in-difference effects for four groups of counties are calculated based on the 25th and 75th percentiles of BHCshare and CAPaverage. In line with the previous results, we find the strongest and most consistent positive effects on local economic conditions for affected counties that host a relatively large share of independent banks (25th percentile of BHCshare) and banks with relatively high capital ratios (75th percentile of CAPaverage). In particular, total personal income and the number of employed persons increase by 7.0% and 3.2% respectively, in the aftermath of the 2005 hurricane season compared with unaffected counties. In comparison, affected counties that host only BHC banks (75th percentile of BHCshare) and banks with relatively low capital ratios (25th percentile of CAPaverage) show no significant increases in total personal income or the number of employed persons compared with unaffected counties (coefficients are 0.0192 and −0.0119, respectively). To the degree that these differences are caused by differences in the structures of the banking systems, policy measures that strengthen (typically more locally focused) independent banks, as well as the banks’ capital buffers, would have highly significant positive economic effects in the aftermath of a crisis. Note that the decrease in the number of unemployed persons and the unemployment rate is even stronger for counties that host a relatively large share of BHC banks (see the bottom rows of Table VIII). However, in the case of relatively low average capital ratios (75th/25th percentiles), these results are not associated with an increase in total personal income or in the number of employed persons, which points to migration effects. In the case of relatively high average capital ratios (75th/75th percentiles), the effects on total personal income and the number of employed persons are also significantly positive, but the increases are comparatively smaller than in counties that host a relatively large share of independent banks and banks with relatively high capital ratios (25th/75th percentiles). Marginal effects of Event × Affected are illustrated in Figure 6. The graph on the left illustrates that affected counties perform better after the 2005 hurricane season if they hosted a relatively small share of BHC banks (relatively large share of independent banks). The graph on the right shows that affected counties with banks with relatively high risk-based capital ratios before 2005 perform significantly better as well. Figure 6. View largeDownload slide Structure of the local banking system and economic development. Notes: This figure illustrates the marginal effects of Event×Affected on the log of total personal income (on the county-level), conditional on the average share of BHC banks (left graph), and the average risk-based capital ratio (right graph) in affected counties. These effects correspond to the regression results in Column (1) of Table VIII. Note that BHCshare and CAPaverage are demeaned. Original values for BHCshare range from 0 to 1 with a mean of 0.86. Original values for CAPaverage range from 0.10 to 0.31 with a mean of 0.15. Values on the x-axis range from the min to the max of the respective variable. The solid vertical lines represent the means (0), and the dashed lines represent the 25th and 75th percentiles of these variables. Figure 6. View largeDownload slide Structure of the local banking system and economic development. Notes: This figure illustrates the marginal effects of Event×Affected on the log of total personal income (on the county-level), conditional on the average share of BHC banks (left graph), and the average risk-based capital ratio (right graph) in affected counties. These effects correspond to the regression results in Column (1) of Table VIII. Note that BHCshare and CAPaverage are demeaned. Original values for BHCshare range from 0 to 1 with a mean of 0.86. Original values for CAPaverage range from 0.10 to 0.31 with a mean of 0.15. Values on the x-axis range from the min to the max of the respective variable. The solid vertical lines represent the means (0), and the dashed lines represent the 25th and 75th percentiles of these variables. To sum up, our evidence supports the view that local (independent) banks, as well as relatively high bank capital, contribute to economic development in the aftermath of a crisis. 5. Conclusions In this paper, we have explored how banks react to catastrophic events and the role that banking structure plays for local economic development, using Hurricane Katrina and two subsequent hurricanes in 2005 as a natural experiment that exposed the borrowers of the banks to enormous losses and economic stress. The natural experiment allows us to provide evidence on a causal effect of a crisis on the business decisions of banks, which is otherwise difficult to identify because of mutual influences and feedback effects. We find that independent banks based in the disaster areas increase their risk-based capital ratios after the hurricanes, while those that are part of a bank holding company on average do not. Affected independent banks thereby strengthen their buffer against future income shocks and mitigate bankruptcy risks. The effect on independent banks is driven by the subgroup of banks with relatively high capital ratios. Independent and relatively low capitalized banks do not show any significant increases in their risk-based capital ratios following the 2005 hurricane season. This demonstrates that the behavior of banks cannot be generalized for all banks but depends on bank characteristics. Apparently, affected independent banks with a relatively cautious business model (reflected in relatively high risk-based capital ratios before Hurricane Katrina) also behave cautiously after the disaster by increasing their risk-based capital ratios. Independent banks with relatively low risk-based capital ratios, which may be more risky and cause more worries for banking supervisors, are not capable or not willing to build higher capital buffers against potential future losses. Increases in the risk-based capital ratios of independent and highly capitalized banks are associated with decreases in risk-weighted assets. In particular, these banks increase their exposures to U.S. government securities and decrease their loan exposures to non-financial firms. Interestingly, these banks also increase new lending to non-financial firms, as reflected in our evidence on the lending practices of banks under the SBA loan program. This suggests that they do not decrease their exposures to non-financial firms through reduced lending, but rather through other strategies such as loan sales or securitization. Finally, we find that the structure of the local banking system plays a role in economic development following the disaster. Affected counties with a relatively large share of independent banks and a relatively high average bank capitalization show a higher growth in total personal income and employment than affected counties with banks that are predominantly part of a bank holding company or affected counties with relatively low capitalized banks. These findings thus have important policy implications by highlighting the importance of (more locally focused) independent banks, as well as bank capital, in facilitating economic development. Supplementary Material Results of further robustness regressions are available at Review of Finance online. Appendix: Short-Term Effects on Banks' Asset Quality In this section, we provide further empirical evidence that Hurricanes Katrina, Rita, and Wilma had an adverse effect on banks’ asset quality by exploring bank profitability and bank risk in Q3 and Q4 2005. In particular, we expect lower bank profitability and higher bank risk. Formally, we estimate the following equation: Yit=νi+τt+β1(EventSTt×Affectedi)+ϵit. (6) The dependent variable Yit stands for RoA or z-score of bank i at quarter t, which reflects a bank’s profitability and stability, respectively.30 The variables νi and τt represent bank- and quarter-fixed effects, respectively. The variable EventSTt is a short-term time dummy with a value of 0 for the two quarters before the hurricanes (Q1 and Q2 2005) and a value of 1 for the two quarters when the hurricanes occurred (Q3 and Q4 2005). The variable Affectedi is a dummy variable of bank i that is 1 if the bank is located in a county classified by FEMA as eligible for “public and private disaster assistance” and thus belongs to the treatment group, and zero otherwise (for the control group). Hence, the interaction term EventSTt × Affectedi is one if both the variable EventSTt and the variable Affectedi amount to 1, and 0 otherwise. The corresponding coefficient β1 is the main interest. It captures the average effect of the hurricanes on the RoA or z-score of affected banks compared with the control group. Note that the single term Affectedi and the single term EventSTt are not directly included in the equation because they are fully captured by the bank- and quarter-fixed effects, respectively. Finally, ϵit is the idiosyncratic error term. We use clustered standard errors at the bank level. First, as shown in Column (1) of Table AI, we estimate the effect of Hurricanes Katrina, Rita, and Wilma on banks’ RoA. We observe that profits decline significantly for banks that were affected by the hurricanes compared with banks that were not affected. The effect of −0.0009 represents about 10% of banks’ average RoA of 0.01. The true effect on banks may be even larger because banks tend to underreport losses in times of crisis (Gunther and Moore, 2003). Regression results for banks’ z-scores are presented in Column (2). The results show that affected banks became significantly more risky during the two quarters when the hurricanes hit the U.S. Gulf Coast compared with unaffected banks. The coefficient of −0.1099 means that the ratio (RoA + Capital)/SD(RoA), where RoA is the return on assets, Capital is the core capital over assets, and SD(RoA) is the standard deviation of RoA, decreases by about 10.99% for affected banks compared with unaffected banks following the 2005 hurricane season, which is economically highly relevant. Table AI. Evidence on short-term effects This table shows results for regressions of Equation (6) in which banks’ RoAs and banks’ z-scores are the dependent variables. The sample includes quarterly data for all independent banks (not part of a bank holding company) in Alabama, Arkansas, Florida, Georgia, Louisiana, Mississippi, Oklahoma, Tennessee, and Texas over the four quarters of 2005. EventST is a dummy variable that is zero for the first two quarters of 2005 and one for the last two quarters of 2005 (when the hurricanes occured). Affected is a dummy variable that separates banks located in counties that were affected by the hurricanes (Affected = 1) and banks located in counties that were unaffected (Affected = 0). Event×Affected is an interaction term for the variables Event and Affected. Bank-fixed effects (bank FE) and quarterly dummies (time FE) are included in each regression. Standard errors are clustered at the bank level and reported in parentheses. The ***, **, and * stand for significant coefficients at the 1%, 5%, and 10% levels, respectively. RoA Z-score (1) (2) EventST×Affected −0.0009** −0.1099*** (0.0004) (0.0420) Bank FE Yes Yes Time FE Yes Yes Number of banks 258 258 Number of observations 1,023 1,017 Within R2 0.0178 0.0481 RoA Z-score (1) (2) EventST×Affected −0.0009** −0.1099*** (0.0004) (0.0420) Bank FE Yes Yes Time FE Yes Yes Number of banks 258 258 Number of observations 1,023 1,017 Within R2 0.0178 0.0481 Table AI. Evidence on short-term effects This table shows results for regressions of Equation (6) in which banks’ RoAs and banks’ z-scores are the dependent variables. The sample includes quarterly data for all independent banks (not part of a bank holding company) in Alabama, Arkansas, Florida, Georgia, Louisiana, Mississippi, Oklahoma, Tennessee, and Texas over the four quarters of 2005. EventST is a dummy variable that is zero for the first two quarters of 2005 and one for the last two quarters of 2005 (when the hurricanes occured). Affected is a dummy variable that separates banks located in counties that were affected by the hurricanes (Affected = 1) and banks located in counties that were unaffected (Affected = 0). Event×Affected is an interaction term for the variables Event and Affected. Bank-fixed effects (bank FE) and quarterly dummies (time FE) are included in each regression. Standard errors are clustered at the bank level and reported in parentheses. The ***, **, and * stand for significant coefficients at the 1%, 5%, and 10% levels, respectively. RoA Z-score (1) (2) EventST×Affected −0.0009** −0.1099*** (0.0004) (0.0420) Bank FE Yes Yes Time FE Yes Yes Number of banks 258 258 Number of observations 1,023 1,017 Within R2 0.0178 0.0481 RoA Z-score (1) (2) EventST×Affected −0.0009** −0.1099*** (0.0004) (0.0420) Bank FE Yes Yes Time FE Yes Yes Number of banks 258 258 Number of observations 1,023 1,017 Within R2 0.0178 0.0481 Summing up, the regression results provide evidence that the 2005 hurricane season had an adverse effect on bank’s asset quality, as reflected in lower bank profitability and higher bank risk. Footnotes We would like to thank Franklin Allen (the editor), an anonymous referee, Marcel Blum, Franziska Bremus, Martin Brown, Don Chance, Horst Entorf, Igor Goncharov, Florian Heider, Jan P. Krahnen, Gregory Nini, Jörg Rocholl, Joao Santos, Alexander Schäfer, Reinhard H. Schmidt, Isabel Schnabel, Adi Sunderam, Marcel Tyrell, Greg Udell, Laurent Weill, participants at the 2012 AEA Annual Meeting in Chicago, the 2012 CEPR Winter Conference on Financial Intermediation in Lenzerheide, the 2012 Financial Intermediation Research Society Conference in Minneapolis, the 2012 European Finance Association Annual Meeting in Copenhagen, the 2012 DGF Annual Meeting in Hannover, the 2012 GEABA Symposium in Graz, the 2012 Verein für Socialpolitik Annual Conference in Göttingen, the 2012 AFFI Annual Meeting in Strasbourg, as well as seminar participants at the Bank of England, Banque de France, Central Bank of Hungary, Bundesbank, DIW Berlin, Sveriges Riksbank, and the University of Zurich in 2013 for their valuable comments and suggestions. Ulrich Schüwer gratefully acknowledges financial support from the Research Center SAFE, funded by the State of Hesse research initiative LOEWE. The paper represents the authors’ personal opinions and does not necessarily reflect the views of the institutions with which they are affiliated. 1 The different crises explored in these studies include the land market collapse in Japan in the early 1990s (Gan, 2007), natural disasters (Garmaise and Moskowitz, 2009; Cortes and Strahan, 2017), and the recent financial crisis (all other studies). 2 One exception in this respect is Cortes (2014) who shows that the presence of local lenders is beneficial for job creation following a natural disaster. 3 The direction of the relationship is difficult to identify because, on the one hand, a crisis affects banks’ business decisions, and on the other hand, banks’ business decisions may cause a crisis if many banks make similar financing or investment decisions. 4 These numbers may seem very high, but a recent study by Jordà et al. (2017) also finds very high effects for a cross-country sample with data from 1870 to 2013. They find that, “over the 5-year period after the peak of economic activity, the cumulative GDP costs of a financial crisis hitting a below-average capitalized banking sector amount, on average, to more than 13 percentage points lower GDP per capita compared to a financial crisis hitting an above-average capitalized banking sector.” 5 Other minor categories are “mission assignments” and “administration” (U.S. Government Accountability Office, 2012). 6 The source for the number of initial jobless claims is the FRED online database of the St. Louis Fed (http://research.stlouisfed.org/fred2/). 7 For more details, see http://research.stlouisfed.org/fred2/release? rid=260. Note that the index was discontinued in 2013. 8 Source: https://www.fema.gov/disasters. 9 Source: https://www2.fdic.gov/sdi/index.asp. 10 Source: http://www.ffiec.gov/hmda/. 11 Source:https://www.sba.gov/category/lender-navigation/sba-loan-programs. 12 Source: https://www.bea.gov/. 13 Source: http://www.bls.gov/lau/. 14 In a set of robustness regressions, we also use a sample of banks from alternative geographic areas. 15 Technically, we require that the FDIC data field namehcr, which denotes a bank holding company, is left blank. 16 The risk-based capital ratio is equivalent to the sum of the bank’s Tier 1 and Tier 2 capital divided by its risk-weighted assets. For some banks, the nominator also includes Tier 3 capital allocated for market risk, net of all deductions. For details, see “Schedule RC-R—Regulatory Capital” of the FDIC. 17 Another useful measure for our analysis would be the gross domestic product (GDP) by county, that is, the value of production that occurs within the geographic boundaries of a county. However, GDP by county is not available (the smallest geographic area for which it is available is a metropolitan area). 18 Normalized differences are calculated as “the difference in averages by treatment status, scaled by the square root of the sum of the variances” (Imbens and Wooldridge, 2009, p. 24). 19 The adverse effects of natural disasters on bank stability are also shown by Noth and Schüwer (2017) for a sample of more than 6,000 U.S. banks over the period 1994–2012. 20 Results remain qualitatively the same if we use the log of total assets instead of the log of the total number of employees for bank size, or RoE instead of RoA. 21 Furthermore, the various capital ratios of banks are typically related to different business models, which can be more or less risky, and reflect aspects such as investment opportunities, bankruptcy costs, franchise value, value of deposit guarantees, and bank governance. 22 Again, note that the terms BHC and Affected × BHC are captured by the bank-fixed effects and are therefore not included in the equation. 23 See https://www.sba.gov/. 24 Note that an alternative SBA program provides disaster loans, but these loans are approved and administered directly by the SBA and are hence not useful for our analysis. 25 This classification follows Cortes and Strahan (2017) who explore mortgage lending in the banks’ core markets and non-core markets following a natural disaster. Data on bank branching comes from the Summary of Deposits database of the FDIC (see https://www.fdic.gov/bank/statistical/). 26 Cortes and Strahan (2017) also provide evidence on the special role of banks’ core markets. They explore how multi-market banks change their credit supply when local credit demand increases following a natural disaster. They find that bank lending increases in affected markets and decreases in unaffected non-core markets, but not in unaffected core markets where banks have a branch presence. 27 Evidence on a causal effect would require exogenous variation in the structure of the local banking system, which is not the case for our analysis. 28 For example, Boustan, Kahn, and Rhode (2012) document migration away from tornado-struck areas in the USA during the 1920s and 1930s, a period before coordinated public disaster assistance existed. 29 In particular, we subtract the mean value of 85.94% across all counties in order to calculate the demeaned BHCshare. To calculate the demeaned CAPaverage, we first consider the different capital ratios of independent and BHC banks and subtract the mean value of 15.84% for all independent banks and the mean value of 13.65% for all BHC banks. We then use these values and calculate the demeaned average capital ratio by county. 30 The z-score is defined as the natural logarithm of the sum of a bank’s RoA and its core capital over assets, standardized by the standard deviation of the bank’s RoA. A lower z-score indicates a lower bank stability. References Avery R. B. , Berger A. B. ( 1991 ): Risk-based capital and deposit insurance reform , Journal of Banking & Finance 15 , 847 – 874 . Google Scholar CrossRef Search ADS Berger A. N. , Bouwman C. H. ( 2009 ): Bank liquidity creation , Review of Financial Studies 22 , 3779 – 3837 . Google Scholar CrossRef Search ADS Berger A. N. , Bouwman C. H. ( 2013 ): How does capital affect bank performance during financial crises?, Journal of Financial Economics 109 , 146 – 176 . Google Scholar CrossRef Search ADS Berger A. N. , Bouwman C. H. , Kim D. ( 2017 ): Small bank comparative advantages in alleviating financial constraints and providing liquidity insurance over time , Review of Financial Studies 30 , 3416 – 3454 . Google Scholar CrossRef Search ADS Berger A. N. , Miller N. H. , Petersen M. A. , Rajan R. G. , Stein J. C. ( 2005 ): Does function follow organizational form? Evidence from the lending practices of large and small banks , Journal of Financial Economics 76 , 237 – 269 . Google Scholar CrossRef Search ADS Bertrand M. , Duflo E. , Mullainathan S. ( 2004 ): How much should we trust differences-in-differences estimates?, Quarterly Journal of Economics 119 , 249 – 275 . Google Scholar CrossRef Search ADS Boustan L. P. , Kahn M. E. , Rhode P. W. ( 2012 ): Moving to higher ground: migration response to natural disasters in the early twentieth century , American Economic Review 102 , 238 – 244 . Google Scholar CrossRef Search ADS PubMed Cecchetti S. G. , King M. , Yetman J. ( 2011 ): Weathering the financial crisis: good policy or good luck? BIS Working paper 351. Chavaz M. ( 2016 ): Dis-integrating credit markets: diversification, securitization, and lending in a recovery. Bank of England Staff Working paper 617. Congleton R. ( 2006 ): The story of Katrina: new Orleans and the political economy of catastrophe , Public Choice 127 , 5 – 30 . Google Scholar CrossRef Search ADS Congressional Research Service ( 2005 ): The macroeconomic effects of Hurricane Katrina. Report for Congress [prepared by B. Cashell and M. Labonte]. Congressional Research Service ( 2013 ): Federal disaster assistance after Hurricanes Katrina, Rita, Wilma, Gustav, and Ike. Report for Congress [prepared by B. R. Lindsay and J. C. Nagel]. Cortes K. R. ( 2014 ): Rebuilding after disaster strikes: How local lenders aid in the recovery. Working paper. Cortes K. R. , Strahan P. E. ( 2017 ): Tracing out capital flows: how financially integrated banks respond to natural disasters , Journal of Financial Economics 125 , 182 – 199 . Google Scholar CrossRef Search ADS De Haas R. , Van Horen N. ( 2013 ): Running for the exit? International bank lending during a financial crisis , Review of Financial Studies 26 , 244 – 285 . Google Scholar CrossRef Search ADS Demirguc-Kunt A. , Detragiache E. , Merrouche O. ( 2013 ): Bank capital: lessons from the financial crisis , Journal of Money, Credit and Banking 45 , 1147 – 1164 . Google Scholar CrossRef Search ADS Deryugina T. , Kawano L. , Levitt S. ( 2014 ): The economic impact of Hurricane Katrina on its victims: Evidence from individual tax returns. NBER Working paper No. 20713. DeYoung R. , Hunter W. C. , Udell G. F. ( 2004 ): The past, present, and probable future for community banks , Journal of Financial Services Research 25 , 85 – 133 . Google Scholar CrossRef Search ADS FDIC ( 2006 ): Hurricane Katrina Examiner Guidance. Interagency Supervisory Guidance for Institutions Affected by Hurricane Katrina. http://www.fdic.gov/news/news/financial/2006/fil06012.html (download October 25, 2014). Federal Reserve Bank of Atlanta ( 2005 ): Annual Report 2005: Resilience. https://www.frbatlanta.org/about/publications/annual-reports.aspx (download January 20, 2015). Federal Reserve Board ( 2014 ): Supervisory Policy and Guidance Topics: Asset Quality. http://www.federalreserve.gov/bankinforeg/topics/asset\_quality.htm (download December 1, 2014). FEMA ( 2015 ): The Disaster Process and Disaster Aid Programs. https://www.fema.gov/disaster-process-disaster-aid-programs (download January 20, 2015). Flannery M. J. , Rangan K. P. ( 2008 ): What caused the bank capital build-up of the 1990s?, Review of Finance 12 , 391 – 429 . Google Scholar CrossRef Search ADS Froot K. A. , Stein J. C. ( 1998 ): Risk management: capital budgeting and capital structure policy for financial institutions: an integrated approach , Journal of Financial Economics 47 , 55 – 82 . Google Scholar CrossRef Search ADS Gan J. ( 2007 ): Collateral, debt capacity, and corporate investment: evidence from a natural experiment , Journal of Financial Economics 85 , 709 – 734 . Google Scholar CrossRef Search ADS Garmaise M. , Moskowitz T. ( 2009 ): Catastrophic risk and credit markets , Journal of Finance 64 , 657 – 707 . Google Scholar CrossRef Search ADS Giannetti M. , Laeven L. ( 2012 ): The flight home effect: evidence from the syndicated loan market during financial crises , Journal of Financial Economics 104 , 23 – 43 . Google Scholar CrossRef Search ADS Gropp R. , Heider F. ( 2010 ): The determinants of bank capital structure , Review of Finance 14 , 587 – 622 . Google Scholar CrossRef Search ADS Gunther J. W. , Moore R. R. ( 2003 ): Loss underreporting and the auditing role of bank exams , Journal of Financial Intermediation 12 , 153 – 177 . Google Scholar CrossRef Search ADS Hazards & Vulnerability Research Institute ( 2014 ). Spatial Hazard Events and Losses Database for the United States (SHELDUS). http://hvri.geog.sc.edu/SHELDUS/ (download July 1, 2014). Houston J. , James D. , Marcus D. ( 1997 ): Capital market frictions and the role of internal capital markets in banking , Journal of Financial Economics 46 , 135 – 164 . Google Scholar CrossRef Search ADS Hurricane Research Division ( 2015 ). National Oceanic & Atmospheric Administration Revisits Historic Hurricanes. http://www.aoml.noaa.gov/hrd/hurdat/ (download January9, 2015). Imbens G. W. , Wooldridge J. M. ( 2009 ): Recent developments in the econometrics of program evaluation , Journal of Economic Literature 47 , 5 – 86 . Google Scholar CrossRef Search ADS Insurance Information Institute ( 2014 ): Catastrophes: U.S. http://www.iii.org/fact-statistic/catastrophes-us (download December 1, 2014). Ivashina V. , Scharfstein D. ( 2010 ): Bank lending during the financial crisis of 2008 , Journal of Financial Economics 97 , 319 – 338 . Google Scholar CrossRef Search ADS Jordà Ò. , Richter B. , Schularick M. , Taylor A. M. ( 2017 ): Bank capital redux: solvency, liquidity, and crisis. NBER Working paper No. 23287. Loutskina E. , Strahan P. E. ( 2011 ): Informed and uninformed investment in housing: the downside of diversification , Review of Financial Studies 24 , 1447 – 1480 . Google Scholar CrossRef Search ADS Moody’s ( 2005a ): Moody’s comments on Hurricane Katrina’s impact on regional bank ratings. Global Credit Research, 8 September 2005. Moody’s ( 2005b ): Moody’s outlines possible Katrina effect on local banks. Market Watch, 8 September 2005. National Hurricane Center ( 2005 ): Tropical cyclone report. Report [prepared by R. D. Knaab, J. R. Rhome, and D. P. Brown]. http://www.nhc.noaa.gov/pdf/TCR-AL122005\_Katrina.pdf (download January 5, 2015). National Hurricane Center ( 2011 ): The deadliest, costliest, and most intense United States tropical cyclones from 1851 to 2010. Working paper [prepared by E. S. Blake, C. W. Landsea, and E. J. Gibney]. Noth F. , Schüwer U. ( 2017 ): Natural disasters and bank stability: Evidence from the U.S. financial system. Goethe University Frankfurt SAFE Working paper No. 167. Puri M. , Rocholl J. , Steffen S. ( 2011 ): Global retail lending in the aftermath of the US financial crisis: distinguishing between supply and demand effects , Journal of Financial Economics 100 , 556 – 578 . Google Scholar CrossRef Search ADS Santos J. A. C. ( 2011 ): Bank corporate loan pricing following the subprime crisis , Review of Financial Studies 24 , 1916 – 1943 . Google Scholar CrossRef Search ADS Shrieves R. E. , Dahl D. ( 1992 ): The relationship between risk and capital in commercial banks , Journal of Banking & Finance 16 , 439 – 457 . Google Scholar CrossRef Search ADS Strobl E. ( 2011 ): The economic growth impact of hurricanes: evidence from US coastal counties , Review of Economics and Statistics 93 , 575 – 589 . Google Scholar CrossRef Search ADS Towers Watson ( 2005 ): Hurricane Katrina: Analysis of the impact on the insurance industry. http://www.towerswatson.com/en/Insights/IC-Types/Ad-hoc-Point-of-View/Perspectives/2005/impact-of-hurricane-katrina-on-the-insurance-industry (download January 5, 2015). U.S. Government Accountability Office ( 2012 ): Federal Disaster Assistance, improved criteria needed to assess a jurisdiction’s capability to respond and recover on its own. http://www.gao.gov/assets/650/648162.pdf (download January 9, 2015). USA TODAY ( 2010 ): 5-Years after Katrina, homeowners insurance costs more (August 26). http://www.usatoday.com/money/industries/insurance/ (download April 18, 2011). © The Author(s) 2018. Published by Oxford University Press on behalf of the European Finance Association. All rights reserved. For permissions, please email: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Review of Finance Oxford University Press

How Do Banks React to Catastrophic Events? Evidence from Hurricane Katrina

Loading next page...
 
/lp/ou_press/how-do-banks-react-to-catastrophic-events-evidence-from-hurricane-YOM7CdUxTO
Publisher
Oxford University Press
Copyright
© The Author(s) 2018. Published by Oxford University Press on behalf of the European Finance Association. All rights reserved. For permissions, please email: journals.permissions@oup.com
ISSN
1572-3097
eISSN
1573-692X
D.O.I.
10.1093/rof/rfy010
Publisher site
See Article on Publisher Site

Abstract

Abstract This paper explores how banks react to an exogenous shock caused by Hurricane Katrina in 2005, and how the structure of the banking system affects economic development following the shock. Independent banks based in the disaster areas increase their risk-based capital ratios after the hurricane, while those that are part of a bank holding company on average do not. The effect on independent banks mainly comes from the subgroup of highly capitalized banks. These independent and highly capitalized banks increase their holdings in government securities and reduce their total loan exposures to non-financial firms, while also increasing new lending to these firms. With regard to local economic development, affected counties with a relatively large share of independent banks and relatively high average bank capital ratios show higher economic growth than other affected counties following the catastrophic event. 1. Introduction How banks navigate through an economic crisis is an important issue to explore, since bank financing is crucial for economic recovery and development. From a policy perspective, it would be desirable if banks could continue to provide financing to borrowers, while also maintaining their stability in an unfavorable market environment that adversely impacts their asset quality. Related studies analyze bank lending in the wake of a crisis (e.g., Gan, 2007; Garmaise and Moskowitz, 2009; Ivashina and Scharfstein, 2010; Puri, Rocholl, and Steffen, 2011; De Haas and Van Horen, 2013; Cortes and Strahan, 2017),1 but little is known about how the business decisions of banks affect their stability and asset allocations during such times, and how the structure of the local banking system is related to economic development in affected areas following a crisis.2 This paper first explores how banks adjust their risk-based capital ratios, which are a key determinant of a bank’s stability and a cornerstone of banking regulation. Second, we analyze the mechanisms of these adjustments, that is, the asset allocations and lending practices of banks. The analysis specifically considers the role of different types of banks with regard to bank structure (independent or part of a bank holding company) and bank capitalization (relatively low or high capital ratios). Finally, we analyze whether characteristics of the local banking system (the share of banks that belong to a bank holding company and the average of banks’ risk-based capital ratios by county) are related to local economic development after a crisis. As different arguments point in different directions, it is not clear what findings to expect from these analyses. More specifically, banks may increase their capital ratios to foster financial stability and protect their franchise values, or they may decrease capital ratios to benefit from risk-shifting in times of crisis. These strategies may be associated with decreased or increased lending, respectively. Furthermore, the presence of a relatively large share of independent banks may be associated with relatively high economic growth, since these banks usually have a local focus and may have advantages in screening and monitoring local borrowers. It could also be associated with relatively low economic growth, since these banks are typically less diversified and face more capital constraints compared with banks belonging to bank holding companies. The challenge for the analysis is twofold: first, to identify the direction of the relationship between a crisis and the banks’ business decisions,3 and second, to control for parallel economic developments, which may also affect banks’ financial figures but may not be the result of active changes in banks’ financing or investment decisions. In order to identify causality between a crisis and banks’ business decisions, we use Hurricane Katrina and two subsequent hurricanes that struck the U.S. Gulf Coast in the third and fourth quarters of 2005 as a natural experiment. Hurricane Katrina ranks among the costliest natural disasters in U.S. history, with estimated property damages ranging from $100 billion to over $200 billion (National Hurricane Center, 2005; Congleton, 2006). The hurricanes exposed banks in the U.S. Gulf Coast region to unexpected losses and weakened their asset quality, as a large part of the damages suffered by borrowers was not insured. It also caused uncertainty for banks with respect to how individual and commercial borrowers would cope with the damages. Asymmetric information between banks and their borrowers increased and it was also uncertain as to how the overall economy in the affected regions would recover from the shock. The FDIC (2006) characterized the situation as follows: Hurricane Katrina had a devastating effect on the U.S. Gulf Coast region that will continue to affect the business activities of the financial institutions serving this area for the foreseeable future. Some of these institutions may face significant loan quality issues caused by business failures, interruptions of borrowers’ income streams, increases in borrowers’ operating costs, the loss of jobs, and uninsured or underinsured collateral damage. In keeping with this, the major rating agencies announced a close monitoring of capital adequacy and the risk-management processes of affected banks in the aftermath of Hurricane Katrina (Moody’s, 2005a, 2005b). Katrina also led to a change in the perceived hurricane risks, as reflected in property insurance premium increases of 30% or more (USA TODAY, 2010). Using this natural experiment, we analyze a large sample of U.S. banks within a difference-in-difference framework. The treatment group comprises affected banks, while the control group comprises unaffected banks in the U.S. Gulf Coast region and neighboring states. When the analysis turns its focus on local economic development after the 2005 hurricane season, we use a corresponding sample of affected and unaffected counties. The key findings of our empirical analysis are as follows: independent banks in the disaster areas increase their risk-based capital ratios after the hurricane compared with the control group (unaffected by the shock), as illustrated in the graph on the left of Figure 1, while those that are part of a bank holding company on average do not. Independent banks thereby strengthen their buffer against future income shocks and thus mitigate bankruptcy risks. Our results therefore suggest that asset quality is an important determinant of a bank’s risk-based capital ratio as long as that bank is not backed by a larger banking organization. When we examine independent banks that are low-capitalized (below the median) or highly capitalized (above the median) separately, we find that this precautionary behavior only holds for the latter of the two. A potential explanation is that banks with high franchise values and/or high bankruptcy costs have incentives to avoid bankruptcy and are thus characterized by high capital ratios (before a hurricane) and precautionary behavior (after a hurricane). Our analysis also shows that highly capitalized independent banks achieve higher risk-based capital ratios by prioritizing lower risk-weighted assets: they increase their holdings in government securities and reduce their existing loan exposures to non-financial firms compared with the control group. The latter is illustrated in the graph on the right of Figure 1, where banks’ loan exposures to non-financial firms are represented by the ratio of the volume of commercial and industrial loans (C&I loans) to assets. We also explore banks’ new lending transactions with non-financial firms based on a sample from the Small Business Administration (SBA) loan program, and we find that highly capitalized independent banks also increase their new lending to non-financial firms in their core markets (where they have a branch presence). This suggests that these banks reduce their loan exposures not through reduced new lending, but rather through other strategies such as loan sales. Figure 1. View largeDownload slide Impact of the 2005 hurricanes on banks’ risk-based capital ratios and loans. Notes: The graphs show the development of banks’ risk-based capital ratios and the ratio of the volume of commercial and industrial loans to assets (C&I loans/Assets) from the third quarter of 2003 through the fourth quarter of 2007. The mean values for independent banks located in areas affected by the 2005 hurricanes are represented by a solid line. The mean values for independent banks located in the U.S. Gulf Coast region or neighboring states, but not affected by the hurricanes, are represented by a dotted line. The solid vertical lines indicate the quarters around the disaster period of the third and fourth quarters of 2005, when Hurricanes Katrina, Rita, and Wilma hit the U.S. Gulf Coast region. Figure 1. View largeDownload slide Impact of the 2005 hurricanes on banks’ risk-based capital ratios and loans. Notes: The graphs show the development of banks’ risk-based capital ratios and the ratio of the volume of commercial and industrial loans to assets (C&I loans/Assets) from the third quarter of 2003 through the fourth quarter of 2007. The mean values for independent banks located in areas affected by the 2005 hurricanes are represented by a solid line. The mean values for independent banks located in the U.S. Gulf Coast region or neighboring states, but not affected by the hurricanes, are represented by a dotted line. The solid vertical lines indicate the quarters around the disaster period of the third and fourth quarters of 2005, when Hurricanes Katrina, Rita, and Wilma hit the U.S. Gulf Coast region. Finally, our analysis provides new evidence on the role of the structure of the banking system to foster growth and employment in the post-Katrina period. We assess whether affected counties that host a relatively small or large number of independent banks (not part of a bank holding company) and banks with relatively low or high pre-Katrina capital ratios develop differently in the post-crisis period. Our evidence shows that counties with a higher share of independent banks and relatively high average bank capital ratios are associated with better economic growth in total personal income and employment compared with other counties (with fewer independent banks or lower average bank capital ratios). This has important policy implications because it suggests that the promotion of independent (locally focused) banks and higher bank capital requirements may mitigate economic costs in areas affected by a crisis. We particularly find that the change (before versus after the natural disaster) in total personal income and employment in an affected county is about 5 and 4 percentage points higher, respectively, if the county has a relatively large share of independent banks and relatively high average bank capital ratios compared with a county with only bank holding company banks (BHC banks) and relatively low bank capital ratios.4 Our research contributes to several strands of literature. First, we contribute to the literature that analyzes the relationship between asset quality (or, related to this, asset risk) and bank capital, using a natural experiment as the identification strategy. Previous studies face the difficulty that asset quality and bank capital are typically determined simultaneously by banks. Using simultaneous equations, two-stage, or standard OLS estimation techniques, these studies typically find a positive relationship between asset risk and capital ratios, that is, a negative relationship between asset quality and capital ratios (e.g., Shrieves and Dahl, 1992; Flannery and Rangan, 2008; Gropp and Heider, 2010). Our findings are in line with the findings from these studies and, using an exogenous shock to banks’ asset quality, provide further evidence on the relationship between a bank’s asset quality and risk-based capital ratio. Importantly, we also consider different bank characteristics, that is, independent banks versus banks that are part of a bank holding company, and low-capitalized versus highly capitalized banks, providing evidence of how these characteristics are associated with banks’ risk-based capital ratio adjustments in the wake of a crisis. Related empirical evidence shows a positive relationship between banks’ risk-based capital ratios and their stability. For example, Berger and Bouwman (2013) find that higher pre-crisis bank capital, measured as equity-to-assets or risk-based capital ratio, is associated with higher survival probability during a banking crisis. Demirguc-Kunt, Detragiache, and Merrouche (2013) show that higher leverage and regulatory capital ratios are associated with better stock market performance during the financial crisis. Hence, our results on banks’ capital ratio adjustments are also relevant for understanding banks’ stability. Furthermore, the results of this paper contribute to studies that assess the consequences of various types of crises on bank lending. For example, using the land-price collapse in Japan in the early 1990s as an exogenous shock, Gan (2007) reports that firms with greater collateral losses receive less credit and reduce investments. Garmaise and Moskowitz (2009) use the 1994 Northridge earthquake in California to show that earthquake risk impacts credit markets with a reduction of over 20% in the provision of commercial real estate loans. Chavaz (2016) finds that banks with more concentrated portfolios in markets affected by the 2005 hurricane season maintain lending in markets hit by the shock and circumvent potential capital constraints through loan sales. Cortes and Strahan (2017) show that, following natural disasters, multi-market banks reallocate funds into markets affected by the disasters (with high credit demand) and away from unaffected markets in which they do not own any branches. They also find that banks do not reduce lending in unaffected core markets in which they own branches. As regards the recent fi