Local Bankruptcy and Geographic Contagion. in the Bank Loan Market

Transcription

1 Local Bankruptcy and Geographic Contagion in the Bank Loan Market Jawad M. Addoum, University of Miami Alok Kumar, University of Miami Nhan Le, University of Mannheim Alexandra Niessen-Ruenzi, University of Mannheim January 13, 2016 Abstract This paper examines whether corporate bankruptcies influence the bank loan characteristics of geographically proximate firms. We find that, beyond industry contagion (Hertzel and Officer, 2012), firms headquartered near a bankruptcy event experience an 11 basis point increase in loan spreads over the subsequent year, even when the credit default risks of local firms do not increase. There is also a decrease in loan amounts and maturities, as well as an increase in the proportion of secured loans and loan covenants among non-filing local firms. Local bankruptcy has a stronger impact on lenders with geographically concentrated loan portfolios, but even lenders with no exposure to the local economy increase their spreads. The adverse effects of bankruptcy weaken as the distance to bankrupt firms increases. Collectively, these results suggest that lenders are sensitive to local bankruptcies and induce geographic contagion in the bank loan market. Keywords: Corporate bankruptcy; geographic contagion; bank loans; local business cycles. JEL classification: G21, G33, E51. Please address all correspondence to Alexandra Niessen-Ruenzi, University of Mannheim, L9, 1-2, Mannheim; Phone: +49 (0) ; niessen@bwl.uni-mannheim.de. Jawad Addoum can be reached at or jawad.addoum@miami.edu. Alok Kumar can be reached at or akumar@miami.edu. Nhan Le can be reached at +49 (0) or le@corporate-finance-mannheim.de. We would like to thank Sandro Andrade, Tim Burch, Sudheer Chava, Vidhi Chhaochharia, Indraneel Chakraborty, Michael Hertzel, George Korniotis, Justin Murfin, Micah Officer, Meijun Qian (discussant), Francisco Santos (discussant), Sascha Steffen, participants at the 2015 European Finance Association (EFA) meetings, the Research Camp at Australian National University, and seminar participants at the University of Miami and the University of Mannheim for helpful comments. We are responsible for all remaining errors and omissions.

2 Local Bankruptcy and Geographic Contagion in the Bank Loan Market Abstract This paper examines whether corporate bankruptcies influence the bank loan characteristics of geographically proximate firms. We find that, beyond industry contagion (Hertzel and Officer, 2012), firms headquartered near a bankruptcy event experience an 11 basis point increase in loan spreads over the subsequent year, even when the credit default risks of local firms do not increase. There is also a decrease in loan amounts and maturities, as well as an increase in the proportion of secured loans and loan covenants among non-filing local firms. Local bankruptcy has a stronger impact on lenders with geographically concentrated loan portfolios, but even lenders with no exposure to the local economy increase their spreads. The adverse effects of bankruptcy weaken as the distance to bankrupt firms increases. Collectively, these results suggest that lenders are sensitive to local bankruptcies and induce geographic contagion in the bank loan market. Keywords: Corporate bankruptcy; geographic contagion; bank loans; local business cycles. JEL classification: G21, G33, E51.

3 While that c-word contagion provokes alarm, what is also striking is just how little we understand the complex network of modern financial flows. Gillian Tett, Financial Times, July Introduction A growing finance literature investigates the economic impact of corporate bankruptcies. For example, Lang and Stulz (1992) demonstrate that industry competitors stock prices decrease by an average of 1% following a bankruptcy announcement. Hertzel, Li, Officer, and Rodgers (2008) show that this pricing implication extends to bankrupt firms customers and suppliers. More recently, a strand of this literature examines the provision of debt financing and suggests that financial intermediaries such as banks are likely to play an important role in transmitting bankruptcy shocks across their corporate clients. In particular, Benmelech and Bergman (2011) and Hertzel and Officer (2012) document an increase in the cost of debt financing among bankrupt firms industry peers. Benmelech and Bergman (2011) focus on the U.S. airline industry and demonstrate the role of the collateral channel in driving industry contagion effects. Specifically, they show that airline industry bankruptcies lead to reduced collateral valuations among surviving airlines, consequently increasing their loan spreads. Hertzel and Officer (2012) demonstrate the existence of within-industry contagion effects for bankruptcies across all industries. Although a corporate borrower s industry is one important factor in the loan evaluation process (e.g. Treacy and Carey (2000), Pesaran, Schuermann, and Treutler (2005)), other salient characteristics that are common across borrowers have also been shown to affect the pricing of debt capital. In particular, Treacy and Carey (2000) and Pesaran, Schuermann, and Treutler (2005) suggest that in addition to a firm s industry, its geographic region of 1

4 operation is an important characteristic considered by banks when evaluating systemic credit risk factors. While industry contagion effects in the bank loan market have been previously documented, the existence and extent of geographic contagion effects in banks financing decisions remain an open question. In this paper, we attempt to fill this gap in the literature by examining the impact of local bankruptcies on the pricing of debt capital for geographically proximate firms. We also identify potential channels through which geographic contagion effects arise in the bank loan market. Using data on bankruptcy filings of all publicly listed firms in the U.S. between 1990 and 2006, we show that spreads on new and renegotiated loans increase significantly for firms that are located in a 50 kilometer county-radius surrounding a firm entering bankruptcy. This geographic contagion effect lasts for about one year following a local bankruptcy, and is statistically and economically significant. Specifically, in univariate tests, we find that loan spreads increase by over 12 basis points, on average, in the year following a local bankruptcy. Further, in multivariate regressions where we control for a large set of firm- and loan-level characteristics, as well as various sets of time varying fixed effects at the industry-year, state-year, and loan characteristic-levels, we find an average increase in local loan spreads of 11 basis points. These magnitudes represent an increase of 5.3% in the cost of debt financing relative to the average loan spread in our sample. We conduct several robustness checks to ensure that our results are distinct from the industry contagion effects documented in Hertzel and Officer (2012). Specifically, we reestimate our main specification after excluding all geographically proximate firms that operate in the same industry as the firm filing for bankruptcy. In addition, we include industry 2

5 sales growth and industry cash flows as additional control variables. In both cases, the local bankruptcy effect remains statistically significant and economically meaningful. In the next set of tests, we focus further on the role of geography. We relax the definition of geographic proximity and define local firms as those located a 100 or 200 kilometer county radius surrounding a filing firm, respectively. We find that our main result remains statistically significant using these alternative measures. Consistent with bankruptcies having the strongest impact on firms that are located in their immediate neighborhood, the economic impact of local bankruptcies weakens as the radius under consideration grows. When we include both our baseline local bankruptcy indicator and the alternative indicators in the same regression, only the 50 kilometer indicator loads significantly. This also mitigates concerns that local economic conditions are the main driver of our result, as they should result in significant coefficients for a 100 or 200 kilometer radius, too. We also examine the contagion effect of local bankruptcies on a broader set of loan characteristics. In addition to adjusting loan spreads to perceived increases in credit risk, lenders could also react to local bankruptcies by offering loans with higher coupon spreads, lower amounts, and shorter maturities in the year following a local bankruptcy. To examine these possibilities, we re-run our main loan spread specification, replacing the dependent variable with coupon spreads, loan amount, and loan maturity, respectively. In addition, we consider a set of probit regressions where the dependent variables are indicators for whether a loan has covenants and whether a loan is secured, respectively. We find that local firms experience significantly higher coupon spreads, lower loan amounts, and lower loan maturities. In economic terms, the average increase in coupon spreads (9 basis points) and decrease in loan amounts ($39 million USD, implying a more than 15% reduction relative to the sample average) are particularly important. We also find that following local bankruptcies, loans are significantly more likely to include covenants 3

6 and to be secured by collateral. Overall, the results from these additional tests suggest that lenders tighten important non-spread lending terms for geographically proximate borrowers during the year following a corporate bankruptcy. In the last set of tests, we identify the dominant economic mechanism that is likely to generate geographic contagion in the bank loan market. We focus on two potential mechanisms. First, geographic contagion effects could be driven by an overall deterioration in local economic conditions, i.e., the state-level business cycles may affect all firms in the same geographic region (Korniotis and Kumar (2012)). In addition, there could be an increase in credit default risks of local firms following corporate bankruptcies, which could lead to an increase in bank loan spreads. In either instance, local bankruptcies could be accompanied with tighter lending conditions for solvent firms. Alternatively, geographically proximate firms might be affected by local bankruptcies due to lender sentiment. Bankruptcy events are highly salient and covered extensively by the news media. Thus, they could serve as an anchor when lenders consider the credit quality of a borrower. As a result, lenders may tighten credit terms for firms headquartered near bankruptcies, even when the creditworthiness of these firms does not change. If the bankruptcy news is salient, even lenders with no local exposure may react and increase their loan spreads. Distinguishing between these two potential contagion mechanisms is important because it allows us to assess the efficiency of the bank loan market. The first economic mechanism suggests a potential rational response by local banks to a deteriorating local economic environment. In contrast, excess lender sensitivity to local bankruptcies could lead to an inefficient allocation of debt capital to surviving local firms. To determine whether loan spreads increase due to worsening local economic conditions, which in turn affect the creditworthiness of firms headquartered near bankruptcy events, we 4

7 identify geographically proximate firms that have experienced a negative cash flow shock and exclude them from the sample. We also include aggregate sales growth and cash flow measured at the state-level as additional controls in our main specification. Finally, we use an instrumental variable approach similar to that of Parsons, Sulaeman, and Titman (2015) as an alternative approach to addressing this concern. In all cases, we find that our main result remains robust, which suggests that worsening local economic environment is unlikely to drive the geographic contagion in the bank loan market. When we examine changes in the credit default risks of local firms, we find that the Altman s Z-score, default ratings, credit rating, and bankruptcy propensity of local firms do not worsen after bankruptcy events. This evidence suggests that borrowers are charged more by lenders even when there is no significant increase in their credit default risks. Next, we examine whether geographic contagion reflects the impact of local bankruptcies on the cash flows of local borrowers. An increase in loan spread may be associated with a decline in borrowers performance due to a decline in local consumer demand. We find that our key result remains significant for larger and older firms as well as for financially unconstrained firms that may be able to access different types of funding sources. The loan spreads increase even for firms with limited local economic exposure. We also consider the impact of supply-side effects. Following a corporate bankruptcy, an increase in local loan spread may be attributed to the deterioration in the supply of local capital. In a recent study, Murfin (2012) shows that lenders with an experience of a recent default are more sensitive to distress events. To investigate whether there exists a potential contagion effect through a bank s balance sheet, we consider subsamples of lenders that have limited local presence and business exposure. The financial capability of these lenders are less likely to be adversely impaired by a local individual distress event. 5

8 We find that our key results remain very similar among the subset of lenders that have no bankruptcy exposure or limited local presence, as measured by the total deal amount or number of deals in the same geographical region. We also observe the contagion effect among the top 30 largest banks and lenders that operate in competitive local lending markets. Together, these findings suggest that geographic contagion is unlikely to operate through the balance sheet channel. Overall, our results suggest that lenders are sensitive to local bankruptcies even when they are not directly exposed to those events. When banks do have direct exposure to local firms, they increase loan spreads after a local bankruptcy even when the local economic environment or the credit risks of local firms do not change. These findings suggest that lenders are potentially over-reacting to salient bankruptcy events and generate a geographic contagion in the bank loan market. Of course, we cannot completely rule out the possibility that lender reaction is economically motivated since we cannot directly observe their expectations. It is likely that the increase in loan spreads reflect the impact of shifts in lenders economic expectations. Identifying information signals that affect those expectations would be an interesting topic for future research. These empirical findings contribute to the finance literature on the determinants of financial contagion. In particular, Lang and Stulz (1992) and Ferris, Jayaraman, and Makhija (1997) investigate bankruptcy contagion effects on investors of industry peers. Further, Hertzel, Li, Officer, and Rodgers (2008) examine bankruptcy contagion effects along the supply chain of filing firms, while Boone and Ivanov (2012) define proximate non-filing firms as strategic alliance partners. Finally, Jorion and Zhang (2007) investigate bankruptcy contagion effects on industry capital providers. 6

9 We complement this literature by adding the new dimension of geography to bankruptcyinduced contagion. Specifically, we document and quantify geographic contagion effects in loan spreads and other loan characteristics following corporate bankruptcies. In addition, we investigate the drivers of geographic contagion effects, and distinguish the extent to which they are driven by local economic conditions and lenders overreaction to local bankruptcy events. The paper most closely related to ours is Hertzel and Officer (2012), who examine industry loan spread contagion effects surrounding a bankruptcy filing by an industry rival. Their main finding is that loans originated in an industry bankruptcy wave are about basis points higher than spreads on loans originated in other industries. We show that there is a similar effect on the geographic dimension. This result complements the evidence on industry contagion and adds to our understanding of how financial shocks are transmitted across firms. Beyond the literature on financial contagion, our paper contributes to a growing finance literature that examines how geographic proximity affects corporate decisions. For example, Chhaochharia, Kumar, and Niessen-Ruenzi (2012) show that firms with a high fraction of local investors exhibit better corporate governance. Further, Parsons, Sulaeman, and Titman (2015) show that, due to cultural factors, financial misconduct increases with the frequency of misconduct among geographically proximate firms. Our paper contributes to this literature by showing that bankruptcies of geographically proximate firms affect the credit conditions of local non-filing firms as local lenders overreact to news about local bankruptcy. 1 1 Although not directly related to our study, Giannetti and Wang (2014) demonstrate that news about local corporate fraud discourages stock market participation among households. 7

10 2 Data and Summary Statistics 2.1 Bank loan data Our initial sample includes public companies in the United States with financial data available on Standard & Poors Compustat database extracted over the period from 1990 to We exclude firms with missing data on assets, cash, sales and earnings before interest, tax and depreciation (EBITDA). We also remove utility (SIC from 4910 to 4939) and financial firms (SIC from 6000 to 6999) since the financial policies of these firms are subject to statutory capital requirements or regulatory supervision in a number of states. This yields a universal sample of 28,104 firm-year observations. Our bankruptcy sample includes bankruptcy filings of public firms obtained from New Generation Research and the UCLA-LoPucki Bankruptcy Research Database. New Generation Research provides a comprehensive dataset of bankruptcy filings for major public companies. The UCLA Bankruptcy Research Database includes data on all large and public company bankruptcies. The database includes large firms with reported assets worth at least $100 million in 1980 dollars. To be included in our sample, these firms must have geographical information including county and state. Then, they are matched with the Standard and Poor s Compustat database. The final sample comprises 1,903 corporate bankruptcy filings from 1990 to We obtain data on individual loan contracts from LPC-Reuter s Dealscan database. Dealscan provides information on loans made to medium and large-sized U.S. and foreign firms. We obtain information on all dollar-denominated loans made by U.S. lenders to U.S. borrowers from 1990 to The matching of Dealscan and Compustat observations is performed using the linkage database from Chava and Roberts (2008). Following the prior literature (Hertzel and Officer (2012) and Ivashina (2009)), the largest tranche in each deal is one observation. To be included in our sample, we require the borrowers to have non-missing 8

11 location information (i.e., either the ZIP code or county and state of corporate headquarters). There are 28,104 loans in Dealscan that meet these requirements. Table 1 reports the number of loans and the number of bankruptcy filings for each year in our sample. Loans are distributed fairly evenly across the sample. While there are 1,037 loans in 1990, the maximum number of loans granted in 1997 amounts to 2,777. The annual number of bankruptcy filings ranges from 48 in 1994 to 237 in 2001, reflecting the burst of the Tech-bubble and the subsequent economic downturn. 2.2 Location data We obtain historical geographical information of firms headquarters from SEC Analytics Suite and Compustat. We focus on the location of corporate headquarters rather than the state of incorporation, since corporate headquarters form the strategic center of a firm and are presumably more relevant for lenders decisions on granting a loan. Specifically, a firm s location is defined based on county and state of its headquarters. We collect the latitude and longitude at the centroid of each U.S. county from the U.S. Census Bureau Gazetteer. To estimate geographic proximity, we use the Haversine Formula and compute the distance between two counties, d 1,2, as: d 1,2 = R 2 arcsin(min(1, a)); R 6378 kilometers; a = (sin( lat2 lat1 2 )) 2 + cos(lat1) cos(lat2) (sin( lon2 lon1 2 )) 2. We define firms to be geographically close to a bankruptcy event, if they are located in counties within a 50 kilometer radius surrounding the headquarter county of a firm filing for bankruptcy in a given year. 2 The distribution of bankruptcy filings across U.S. states is presented in Figure 1. 2 Consistent with the prior literature, we alternatively define local firms as those located in a 100 or 200 kilometer county radius of a bankruptcy (e.g., Coval and Moskowitz (2001), Malloy (2005), and Kedia and Rajgopal (2009)). See section

12 As expected, bankruptcies are more frequent around economic centers of the U.S. Specifically, most bankruptcies are observed in California, New York, New Jersey, Florida and Texas. In contrast, there are no bankruptcies of publicly listed firms in the state of Wyoming throughout our sample period. 2.3 Summary statistics After merging observations from all databases, our final sample comprises 10,621 loans that are held by 4,244 firms. The sample period is from 1990 to All variables are defined in more detail in Appendix A and their summary statistics are reported in Table 2. Panel A displays average and median loan and firm characteristics as well as their standard deviations. The mean loan spread in our sample is basis points. Loans have an average maturity of about 44 months and the average deal size is $250 million USD. Most of our sample loans are secured (73%) and do not involve relationship lenders. The average firm size in our sample is $5.6 billion USD. Overall, summary statistics for our sample are very similar to the sample statistics reported by Hertzel and Officer (2012), who investigate loan spread contagion within industries. Panel B compares average loan and firm characteristics between local firms, i.e., firms that are geographically close to a bankrupt firm, to non-local firms. Firms entering bankruptcy are excluded from the sample. We find that spreads of local firms are significantly larger than spreads of non-local firms. Deals of local firms do not differ in average loan amounts, but do have significantly lower maturities on average. At the same time, local firms have lower return on assets and higher cash flow volatility. These differences mandate to control for firms return on assets and cashflow volatility in all subsequent regressions. 10

13 3 Effect of Local Bankruptcies on Loan Spreads 3.1 Estimation framework To quantify the impact of local bankruptcies on geographically proximate firms, we define a dummy variable that is equal to one for all firms headquartered in counties within a 50 kilometer radius surrounding a firm that filed for bankruptcy in the previous year, and zero otherwise. We are interested in how these firms cost of debt capital is affected by a local bankruptcy. Following Hertzel and Officer (2012), we define the logarithm of loan spreads as the main dependent variable. Firms entering bankruptcy are dropped from the sample to avoid a mechanical relation between loan spreads and local bankruptcies. Our econometric specification follows Hertzel and Officer (2012) and Chava, Livdan, and Purnanandam (2009). Specifically, our regressions include the following lagged firm characteristics as control variables: firm size measured as the logarithm of total assets, a firm s market-to-book ratio, return on assets, leverage ratio, and asset tangibility. Cash flow volatility and stock return volatility are computed using data from the last 12 months. The results in Panel B of Table 2 suggest that borrowers that are geographically proximate to a filing firm are more risky than borrowers that are not close to a local bankruptcy. There are several channels through which geographically proximate firms could be affected by a local bankruptcy. First, local business cycles might cause firms in specific geographic areas to experience a simultaneous economic downturn (Korniotis and Kumar (2012)). Second, even if just one local firm is hit by a bankruptcy, job loss due to this bankruptcy might slow down regional demand and thereby affect other firms in the same geographic area. To capture the impact of these contagion effects of local bankruptcy filings, we include two control variables that capture the likelihood that geographically proximate firms enter financial distress after a local bankruptcy. The first control variable is a forward looking measure of borrower quality, defined as a dummy variable that is equal to one if a firm enters 11

14 bankruptcy during the three years following loan origination, and zero otherwise. The second control variable is based on the Chicago Fed National Activity recession indicator (CFNAI) generated from the Stock-Watson Experimental Coincident Recession Index (Santos and Winton (2008)). It is defined as a dummy variable, which is equal to one if the CFNAI index is larger or equal to its time series average in the loan origination month and the two prior months, and zero otherwise. We also control for loan-specific characteristics in our regressions. Specifically, we include the log of the loan s deal amount and a dummy indicating the presence of financial covenants. In addition, we control for whether the loan is secured and whether there is a sole or a relationship lender, because the latter might lead to stronger contagion effects (Cai, Saunders, and Steffen (2014). We also control for whether the loan contains a performance pricing feature, and whether the base rate is prime. Finally, we also include a large set of fixed effects. First, industry and state fixed effects allow us to control for cross sectional differences in loan characteristics across industries and states. Second, we include fixed effects for senior and non-senior debt ratings, type of loan, and purpose of loan. Squared (continuous) control variables are also added in one specification to allow for a non-linear impact of these variables. To further address causality concerns, we also use an instrumental variable approach. 3.2 Baseline results The baseline estimation results are reported in Table 3. All t-statistics are based on robust standard errors clustered at the borrower level. Our results suggest that loan spreads of geographically proximate firms are adversely affected by a local bankruptcy. The point estimates of the coefficients suggest that loan spreads of firms that are located within a 50 kilometer county radius of a filing firm increase by 11 basis points the year following a local bankruptcy filing. In column (1), we only include state fixed effects to control for geographic heterogeneity in loan spreads. We then add industry fixed effects (column (2)), fixed effects 12

15 reflecting senior debt ratings, types and purposes of loans (column (3)), and squared control variables (column (4)) as in Hertzel and Officer (2012). However, independent of the regression specification we use, our main result always remains statistically significant at the 1% level. The result is also economically significant. With an unconditional average spread of around 206 basis points over LIBOR, our point estimates suggest that, controlling for loan, market, and borrower characteristics, loan spreads of geographically proximate firms are about 5.3% higher if a local firm has filed for bankruptcy in the previous year. With respect to our control variables, the signs of the coefficients are generally consistent with earlier findings from the literature (e.g., Hertzel and Officer (2012)). We find that loan spreads are significantly higher for borrowers that are smaller, less profitable, and have higher cash flow and stock return volatility over the year prior to origination. At the same time, spreads are higher for borrowers with low relative valuations (market-to-book ratio) and higher pre-loan leverage. Smaller loans, re-financings, and loans that are tied to the U.S. prime rate also have systematically higher spreads (despite the fact that Dealscan converts non-libor spreads into LIBOR-equivalent spreads). Spreads are also significantly higher for secured loans, and lower for loans with a relationship lender and loans with a performance pricing feature. We do not find a significant impact of covenants on loan spreads. As expected, our measures of borrower quality are positively related to loan spreads. Spreads are about 50 basis points higher if the borrower files for bankruptcy within three years following loan origination. This result suggests that lenders are at least partly able to predict greater credit risk and the probability of corporate bankruptcy. Further, loan spreads are about 12 basis points higher during economic downturns. Since we still observe a significant impact of location on loan spreads, changes in borrower quality and market-wide economic conditions can not fully explain the observed geographic contagion effect. 13

16 3.3 Anticipation of bankruptcy? So far, our results suggest that local bankruptcies are associated with a tightening of lending terms for geographically proximate firms in the year following a local bankruptcy. However, it is possible that lenders anticipate local bankruptcies and adjust lending conditions before the actual filing date. Additionally, local bankruptcy shocks could have an impact on lending terms that last for more than one year after the bankruptcy filing date. To investigate anticipation effects of lenders as well as the duration of geographic loan contagion, we relate loan spreads to various leads and lags of corporate bankruptcy events in addition to the standard control variables used in our baseline specification in column (3) of Table 3. These results are reported in Table 4. The results show that lenders do not adjust lending terms for geographically proximate borrowers in anticipation of a local bankruptcy. We observe an economically and statistically insignificant impact of local bankruptcies on contemporaneous loan spreads of geographically proximate firms of about 3 basis points (column 1). Also, loan spreads do not change significantly if a bankruptcy is filed within the next year (column 2), or within two years (column (3)), respectively. These results do not support the view that lenders anticipate local bankruptcies and adjust credit terms of local firms before a bankruptcy actually happens in their geographic vicinity. With respect to the duration of geographic contagion, we only observe a marginally significant increase in loan spreads of geographically proximate firms which amounts to four basis points two years after a local bankruptcy filing (column (3)). This evidence suggests that geographic contagion lasts for a little bit more than one year after a local bankruptcy filing, while the main effect is clearly observed in the first year following a local bankruptcy. Most importantly, our baseline effect of geographic contagion within the year following a local bankruptcy remains economically and statistically significant at the 1% level for all specifications in Table 4. 14

17 3.4 Effect of distance In our next set of tests, we relax our definition of geographic proximity and define local firms as all firms that are located in a 100 or 200 kilometer county radius surrounding a filing firm, respectively. Then, we estimate our baseline regression from column (3) in Table 3 and include these definitions of geographic proximity instead of, or in addition to, our main local bankruptcy dummy variable. The results are reported in Table 5. We continue to observe a positive contagion effect of about 9 basis points (column (1)) and 6 basis points (column (3)) for firms that are located in a 100 or 200 kilometer county radius surrounding a filing firm, respectively. Both coefficients are statistically significant at the 1% level. As expected, the effect becomes economically weaker as the indicator includes firms that are located progressively farther away from the filing firm. In columns (2) and (4), we include our baseline bankruptcy indicator together with the dummy variable reflecting the 100 kilometer radius (column (2)) and the 200 kilometer radius (column (4)). Our baseline bankruptcy indicator variable remains highly statistically and economically significant, while the other two variables are no longer significant. We interpret this result as evidence of lenders overreaction to local bankruptcies, since firms that are most proximate to bankruptcies experience a more adverse deterioration in their credit conditions. If local economic conditions were driving our main result instead, the adverse impact of bankruptcies would have also extended to those firms located 100 or 200 kilometers from a filing firm. Local business cycles almost surely extend beyond a 50 kilometer radius. 3.5 Effect of local bankruptcies on other loan terms In this section, we examine a broader set of loan characteristics that might be affected by a local bankruptcy. In addition to adjusting loan spreads to perceived increases in credit risk, 15

18 lenders could also react to local bankruptcies by offering loans with higher coupon spreads, lower amounts, and shorter maturities. To test these conjectures, we re-estimate our baseline model from column (3) in Table 3, and replace the dependent variable with coupon spreads, loan amount, and loan maturity, respectively. We also estimate two probit regressions where the dependent variables are indicators for whether a loan has covenants and for whether a loan is secured, respectively. The results of this analysis are presented in Table 6. Consistent with our expectations, we find that lenders also tighten other important lending terms for local borrowers the year after a local firm files for bankruptcy. In addition to the substantial increases in loan spreads that we observe in Table 3, the estimates in column (1) of Table 6 suggest that coupon spreads of geographically proximate firms significantly increase by 9 basis points the year following a local bankruptcy. Further, the results in columns (2) and (3) respectively show that local bankruptcies lead to a significant subsequent reduction in the size and maturity of loans for geographically proximate borrowers. In economic terms, the average loan size decreases by $39 million USD during the year following a local bankruptcy. Given that the average loan amount in our sample is $250 million USD, this evidence implies that a local bankruptcy event induces a more than 15% reduction in the average loan amount. We also observe an increase in loan maturities of approximately 1 month following a local bankruptcy, which we deem to be a rather small effect in economic terms. The estimates in columns (4) and (5) of Table 6 show that loans of firms in the vicinity of a bankrupt firm are significantly more likely to have covenants and to be secured in the subsequent year. The coefficients on control variables are again broadly in line with the previous literature. Taken together, the results in Table 6 suggest that local bankruptcies adversely affect local borrowers through various channels, and that there is a broad deterioration of credit conditions for these firms. 16

19 4 What Drives Geographic Contagion? Our results so far suggest that the cost of debt capital increases significantly for firms that are geographically proximate to a firm filing for bankruptcy. During the year following a local bankruptcy, loan spreads of geographically proximate firms increase significantly by about 11 basis points. This effect lasts for about one year following a local bankruptcy. We interpret these results as evidence of geographic contagion induced by corporate bankruptcies. In this section, we identify the dominant mechanism that is likely to generate geographic contagion in the bank loan market. There are three alternative, but not mutually exclusive, explanations for geographic contagion. First, geographic contagion effects might be observed because industries tend to cluster geographically (e.g., IT firms cluster in Silicon Valley). Thus, the results we observe might be an artefact of the well documented industry contagion effects (Hertzel and Officer (2012)). Second, similar to industry contagion, geographic contagion effects might capture a worsening of borrower quality due to deteriorating local economic conditions that are not fully captured by the credit risk measures that we consider. Thus, our results could reflect an appropriate lender reaction to higher credit risks of geographically proximate firms caused by local bankruptcies and worsening economic conditions. This logic would imply that our geographic contagion indicator variable (together with industry contagion variables) better captures the creditworthiness of borrowers in our sample. Thus, they can be usefully employed in determining the credit risk of individual firms beyond firm-specific and market-wide impact factors. In line with this view, Averey, Bostic, Calern, and Canner (2000) criticize standard credit scoring models for not sufficiently adjusting for local economic conditions that might affect loan repayment in a certain geographical area. 17

20 Third, the geographic contagion effect that we document could be the result of a potential lender overreaction to local bankruptcies. Local bankruptcies are highly salient events that might act as an anchor in lenders assessments of credit quality. Thus, even if other firms in the same geographic area are not economically affected by a local bankruptcy, lenders might still refrain from granting credit to these otherwise healthy firms because of merely a perception of higher borrower credit risk. This anchoring effect, that has been shown to exist in other financial contexts (e.g., Choi, Haisley, Kurkoski, and Massey (2014)), could lead to an inefficient supply of capital, and might drive other economically healthy firms into bankruptcy as well. We perform several tests to determine whether and to what extent each of these potential explanations contributes to our main empirical results. 4.1 Geographic or industry contagion? To begin, we ensure that our results do not merely reflect the well-documented industry contagion effects. These tests are motivated by the findings in Chava and Jarrow (2004), who show that industry effects are important for bankruptcy prediction. Further, more recently, Hertzel and Officer (2012) show that bankruptcies adversely affect other firms operating in the same industry. To account for industry effects, we re-estimate our main regression after excluding all geographically proximate firms that operate in the same industry as the firm filing for bankruptcy. Alternatively, we include industry sales growth and industry cash flows as additional control variables. The results from these tests are presented in Table 7. The results in column (1) show that there is a significant geographic contagion effect, even when we exclude all geographically proximate firms within the same industry. The local bankruptcy variable is statistically significant at the 1% level and indicates that there is about 10 basis points increase in loan spreads during the year following a local bankruptcy. We obtain similar results when we include additional control variables to account for industry 18

21 contagion (column 2). The local bankruptcy variable remains economically and statistically significant at the 1% level. Thus, only a very small fraction of our baseline effect of 11 basis points can potentially be explained by industry clustering. 4.2 Geographic contagion or local business cycle effects? Next, we investigate the extent to which our results are driven by local economic conditions that might affect the creditworthiness of all firms in a certain geographic area. Specifically, in columns (3) and (4) of Table 7, we account for the impact of local economic conditions on bank loan spreads. In column (3), we drop geographically proximate firms that have experienced a negative cash flow shock. Motivated by Lemmon, Ma, and Tashjian (2009), we identify cash flow shocks based on the median cash flow ratio (i.e., EBITDA over total assets) of firms in a given state. We exclude all firm-year observations for which the cash flow ratio is in the bottom quartile of the sample distribution. Our results in column (3) still show a significant impact of our local bankruptcy variable on loan spreads that amounts to roughly 9 basis points. In column (4), we include aggregate sales growth and cash flows of all firms located in the same state as the firm that filed for bankruptcy in the previous year as additional control variables. Again, our main results remain very similar and suggest a significant increase in local loan spreads of about 8 basis points. Next, we combine our approaches from columns (1) and (3) and exclude geographically proximate firms that operate in the same industry as the bankrupt firm, as well as firms with a negative cash flow shock. These results are presented in column (5), where we continue to observe a statistically significant impact of our local bankruptcy variable. The impact also continues to be meaningful in economic terms as there is about 8 basis points increase in local loan spreads. A similar effect is observed in column (6), where we combine our approaches from columns (2) and (4) and include all industry and state-level controls at the same time. 19

22 Finally, in columns (7) and (8), we include the number of non-local bankruptcies and the number of total bankruptcies in a given year as additional control variables. The number of non-local bankruptcies reflects the total number of bankruptcies in a given year, excluding those in a 50 kilometer county radius of the filing firm. And the number of total bankruptcies captures the sum of all bankruptcies in our sample in a given year. The estimates in columns (7) and (8) indicate that adding these control variables does not affect our main result. We still observe a significant increase (about 8 basis points) in local loan spreads following local bankruptcies. Taken together, our baseline effect of an 11 basis points increase in loan spreads after a local bankruptcy can partially be explained by local economic conditions. However, we still observe an increase of 8-9 basis points even after controlling for local economic conditions. Thus, more than half of our baseline effect does not seem to be driven by either industry clustering or local economic conditions. Rather, the remaining increase might be due to lenders overreaction to a local bankruptcy. 4.3 Alternative tests: Industry and local economic effects In this section, we adopt an alternative approach to dealing with the concern that the geographic contagion effects we document are driven by unobserved variation in local economic conditions or industries. Specifically, we re-estimate the baseline specification in column(3) of Table 3, but also include several combinations of time-varying fixed effects. The results of this analysis are reported in Table 8. In column (1), we add industry-year fixed effects, obtained by interacting industry indicators with year indicators. This specification controls for any time-varying unobservable industry characteristics that might otherwise drive our main result. Similarly, in column 2 we control for time-varying local economic conditions by adding state-year fixed effects to the 20

23 baseline specification. The state-year fixed effects are obtained interacting state indicators with year indicators. In column 3 we include year fixed effects, controlling for common time variation in spreads across all loans. Across all three specifications, we find that the local bankruptcy indicator continues to be associated with a statistically significant increase in loan spreads. Further, when we simultaneously include industry-year, state-year, and year fixed effects in column (4), we find that the local bankruptcy indicator is statistically significant at the 1% level, with a coefficient indicating that local loan spreads are about 8 bps higher. Finally, we include firm effects and find that our main result obtains. Taken together, the results in Table 8 indicate that our main finding is robust to controlling for unobservable time-varying industry and local economic factors. As an additional test designed to rule out the unobserved effect of local economic conditions on loan spreads, we use an instrumental variable approach similar to that of Parsons, Sulaeman, and Titman (2015). We propose two instrumental variables that capture the likelihood of a firm s bankruptcy, but are plausibly unrelated to local economic conditions. Specifically, we instrument local bankruptcies with the median performance, measured by cash flows or sales growth, of non-local firms belonging to the same industry. While a firm s likelihood of entering financial distress should be correlated with the performance of its nonlocal industry peers, local economic conditions should be uncorrelated with this instrument by construction. We run two-stage least squared regressions where, in the first stage, we estimate the likelihood that a firm files for bankruptcy in a given year. This likelihood is based on a linear probability model including one of our two instruments. In the second stage, we re-estimate our baseline regression in column (3) of Table 3, including the instrumented local bankruptcy dummy as the independent variable of interest. 21

24 The results of the two-stage least squares regressions are reported in Table 9. Columns (1) and (3) present results of the first-stage regressions. As expected, we find that a local firm is less likely to file for bankruptcy if its non-local industry peers had high cash flows or sales growth. Columns (2) and (4) report the results of the second-stage regressions. In both specifications, we find that local bankruptcies continue to have a significantly positive impact on the loan spreads of non-filing local firms. While the two-stage least squares results are qualitatively similar to those of our OLS estimates, one finding of note is that, similar to the results in Kim and Lu (2011), the magnitude of the local bankruptcy coefficient becomes significantly larger. A potential explanation for the large difference is measurement errors in the local bankruptcy variable, biasing the coefficient towards zero in an OLS specification. This problem is usually mitigated by the instrumental variable approach (e.g., Das, Kim, and Patro (2011)). We take comfort in the fact that all results point in the same direction, and suggest that local bankruptcies have a significantly positive effect on the loan spreads of non-filing local firms. Importantly, the instrumental variable results suggest that this impact cannot be explained by the unobserved effect of local economic conditions. In the following sections, we conduct several additional tests to examine a potential overreaction explanation. 4.4 Impact of borrower characteristics In this section, we conduct several cross-sectional tests to investigate whether our main effect is stronger across certain borrower and lender characteristics. For example, we expect that geographic contagion effects of local bankruptcies are stronger for younger, smaller, and less diversified borrowers that might be considered more risky. Also, we expect geographic contagion effects to be stronger for more financially constrained borrowers. To test these conjectures, we estimate the baseline regression from column (3) in Table 3 and interact our local bankruptcy indicator variable with several borrower characteristics 22

25 split at the sample median. The local bankruptcy indicator as well as borrower characteristics are also included in the specifications. The dependent variable in these regressions is the logarithm of loan spreads. The estimation results are reported in Table 10. The results in Panel A show that loan spreads of local borrowers increase for all types of borrowers, independent of their age (column (1)) or size (column (2)). The results are also independent of whether the borrower is financially constrained, as measured by the Whited and Wu (2006) index in column (3) or the Kaplan and Zingales (1997) index in column (4). The loan spreads of local firms also increase independent of whether borrowers business operations are diversified across counties (column (5)) or states (column (6)). In short, none of the regressions in Panel A of Table 10 yield significant interaction terms between borrower type and loan spreads. In contrast, we continue to find a statistically significant impact of our local bankruptcy indicator variable on loan spreads of about 11 basis points. These results suggest that lenders do not differentiate between types of borrowers when they adjust loan conditions for non-filing firms in the geographic proximity of a bankruptcy. This finding potentially hints at a one-size-fits-all approach by lenders, that might adversely affect even those local firms whose creditworthiness has not changed at all after a local bankruptcy. 4.5 Impact of lender characteristics Next, we investigate the role of lender characteristics on the loan market reaction to local bankruptcies. Specifically, it is likely that loans issued by lenders with geographically concentrated loan portfolios are more affected by geographic contagion effects than loans issued by lenders that are more diversified with respect to geographical credit risk. We employ three different measures of lender concentration in Panel B of Table 10. In column (1), we interact our local bankruptcy indicator with a dummy variable equal to one 23

26 if a lender s total lending in a given county is larger than the median county-level lending of all lenders in our sample, and zero otherwise. We adopt a similar approach in column (2), where we interact our local bankruptcy indicator with a dummy variable equal to one if a lender s number of local deals is larger than the median number of local deals by all lenders in the sample, and zero otherwise. Last, in column (3) we use a Herfindahl measure to take into account the effect of local lender concentration. First, we compute the local market share of each lender based on the lender s total deal amount. We then compute the local Herfindahl index (HHI), which is equal to the sum of squared market shares of all local lenders. To interact lender concentration with our local bankruptcy variable, we define an indicator variable that is equal to one if the local HHI index is higher than the median number of all lenders HHI indices in a given year, and zero otherwise. Similar to our findings on borrower heterogeneity, the results in Panel B of Table 10 show that lender characteristics do not affect how lenders react to local bankruptcies. In all three specifications, we find that interaction terms between our local bankruptcy variable and lender characteristics are either marginally significant only (column 1) or insignificant (columns 2 and 3), while our baseline effect remains economically and statistically significant in all columns. These results again suggest that local lenders might overreact to a local bankruptcy event and apply stronger restrictions on newly issued loans because the bankruptcy event is more salient to them. 4.6 Impact of lenders with limited local exposure In a recent study, Murfin (2012) demonstrates that lenders experiencing defaults subsequently write more restrictive loan contracts. This finding suggests that our key result may be driven by lenders whose balance sheets have been adversely affected by local corporate 24

27 bankruptcies. To examine whether our main result reflects lenders direct exposure to local bankruptcies, we re-estimate our baseline specification using two subsamples of lenders that have limited exposure to the local environment. First, we consider a subsample of lenders that are not directly exposed to a local bankruptcy. Since these lenders do not experience any loss due to the default event, their capital supply is unlikely to be affected by the local shock. Second, we look at a subset of lenders that are among the top 30 largest banks in the U.S., as reported by the Federal Reserve Board. The balance sheets of these large banks are also less likely to be strongly affected by a local bankruptcy. The subsample estimates are presented in Table 11. In both subsamples, we continue to observe an economically and statistically significant effect associated with local bankruptcies. The loan spreads are higher even for banks with limited or no exposure to local bankruptcy. These findings suggests that our results are unlikely to reflect a rational reaction by lenders that are either directly affected by a local bankruptcy event or are highly exposed to the local environment. Lender over-reaction to local bankruptcies is a more likely explanation for our findings. 4.7 Reaction to an increase in local default risks? In our last set of tests, we examine whether increased loan spreads are economically justified by higher actual default risks among local firms following a local default event. If this is the mechanism driving our results, we expect to observe a subsequent increase in the likelihood of financial distress among surviving local firms. In Table 12, we regress several proxies of default risk on the local bankruptcy indicator. In column (1), the dependent variable is the modified Altman s Z-score. In column (2), the dependent variable is a default indicator that equals one if a firm s debt is rated either D or SD, and zero otherwise. In column 3, the dependent variable is a downgrade indicator 25

28 that equals one if a firm experiences a downgrade in its debt rating in a given year, and zero otherwise. In column 4, the dependent variable is a forward-looking indicator of a firm s eventual bankruptcy that equals one if the borrower files for bankruptcy within three years of the loan origination date, and zero otherwise. To control for industry and state heterogeneity and macroeconomic conditions, we also include industry, state and year fixed effects in all regressions. Across all four specifications in Table 12, we find that the local bankruptcy indicator is statistically insignificant. Together with our earlier findings,this evidence suggests that although local bankruptcies do not result in higher default risks among geographically proximate firms, lenders demand higher spreads on loans to these borrowers. Thus, lenders appear to overreact to local distress events and, consequently, increase loan spreads. 5 Summary and Conclusion This paper investigates whether local bankruptcies adversely affect credit conditions for geographically proximate firms. Our results suggest that loan spreads of geographically proximate firms increase by about 11 basis points during the year following a local bankruptcy. This effect lasts for about one year and does not seem to be anticipated by lenders. We also find strong evidence that other loan characteristics such as coupon spreads, loan amounts and maturities, as well as covenants and loan securitization are affected by local bankruptcies. Particularly, local and less diversified lenders strongly adjust loan terms for firms in the geographic vicinity of a bankruptcy event. We do not find any evidence that lenders differentiate between different types of borrowers when making their adjustments. Our results suggest a new geographic channel through which financial contagion is transmitted. While previous papers suggest that there are intra-industry contagion effects of bankruptcies (Hertzel and Officer, 2012), we show that geographical proximity also contributes to financial contagion. Evidence in Bernile, Delikouras, Korniotis, and Kumar (2015) 26

29 suggests that the geographical distribution of publicly-traded firms generates an economic network that links the economic environments of all U.S. states. Our results imply that local bankruptcies might spread through this type of network and, if it is large enough, it could also affect firms in other U.S. states. 27

30 References Averey, R. B., R. W. Bostic, P. S. Calern, and G. B. Canner, 2000, Credit Scoring: Statistical Issues and Evidence from Credit-Bureau Files, Real Estate Economics, 28, Benmelech, E., and N. K. Bergman, 2011, Bankruptcy and the Collateral Channel, Journal of Finance, 66, Bernile, G., S. Delikouras, G. Korniotis, and A. Kumar, 2015, Geography of Firms and Propagation of Local Economic Shocks, Working Paper; Available at SSRN: Boone, A. L., and V. I. Ivanov, 2012, Bankruptcy Spillover Effects on Strategic Alliance Partners, Journal of Financial Economics, 103, Cai, J., A. Saunders, and S. Steffen, 2014, Syndication, Interconnectedness, and Systematic Risk, Working Paper; Available at SSRN: Chava, S., and R. Jarrow, 2004, Bankruptcy Prediction with Industry Effects, Review of Finance, 8, Chava, S., D. Livdan, and A. Purnanandam, 2009, Do Shareholder Rights Affect the Cost of Bank Loans?, Review of Financial Studies, 22, Chava, S., and M. R. Roberts, 2008, How Does Financing Impact Investment? The Role of Debt Covenants, The Journal of Finance, 63(5), Chhaochharia, V., A. Kumar, and A. Niessen-Ruenzi, 2012, Local Investors and Corporate Governance, Journal of Accounting and Economics, 54, Choi, J., E. Haisley, J. Kurkoski, and C. Massey, 2014, Small Cues Change Savings Choices, Working Paper. Coval, J. D., and T. J. Moskowitz, 2001, The Geography of Investment: Informed Trading and Asset Prices, Journal of Political Economy, 109,

31 Das, S., K. Kim, and S. Patro, 2011, An Analysis of Managerial Use and Market Consequences of Earnings Management and Expectation Management, The Accounting Review, 86, Ferris, S., N. Jayaraman, and A. Makhija, 1997, The Response of Competitors to Announcements of Bankruptcy: An Empirical Examination of Contagion and Competitive Effects, Journal of Corporate Finance, 3, Giannetti, M., and T. Y. Wang, 2014, Corporate Scandals and Household Stock Market Participation, Journal of Finance, forthcoming. Hertzel, M. G., Z. Li, M. S. Officer, and K. Rodgers, 2008, Inter-Firm Linkages and the Wealth Effects of Financial Distress Along the Supply Chain, Journal of Financial Economics, 87, Hertzel, M. G., and M. S. Officer, 2012, Industry Contagion in Loan Spreads, Journal of Financial Economics, 103, Ivashina, V., 2009, Asymmetric Information Effects on Loan Spreads, Journal of Financial Economics, 92(2), Jorion, P., and G. Zhang, 2007, Good and Bad credit Contagion: Evidence From Credit Default Swaps, Journal of Financial Economics, 84, Kaplan, S. N., and L. Zingales, 1997, Do Investment-Cash Flow Sensitivities Provide Useful Measures of Financial Constraints?, Quarterly Journal of Economics, 112, Kedia, S., and S. Rajgopal, 2009, Neighborhood Matters: The Impact of Location on Broad Based Stock Option Plans, Journal of Financial Economics, 92, Kim, E. H., and Y. Lu, 2011, CEO ownership, external governance, and risk-taking, Journal of Financial Economics, 102, Korniotis, G., and A. Kumar, 2012, State-Level Business Cycles and Local Return Predictability, Journal of Finance, 68,

32 Lang, L., and R. Stulz, 1992, Contagion and Competitive Intra-industry Effects of Bankruptcy Announcements, Journal of Financial Economics, 32, Lemmon, M., Y. Ma, and E. Tashjian, 2009, Survival of the Fittest? Financial and Economic Distress and Restructuring Outcomes in Chapter 11, Working Paper; Available at SSRN: Malloy, C. J., 2005, The Geography of Equity Analysis, Journal of Finance, 60, Murfin, J., 2012, The Supply-Side Determinants of Loan Contract Strictness, The Journal of Finance, 67(5), Parsons, C. A., J. Sulaeman, and S. Titman, 2015, The Geography of Financial Misconduct, Working Paper; Available at SSRN: Pesaran, M. H., T. Schuermann, and B.-J. Treutler, 2005, The Role of Industry, Geography and Firm Heterogeneity in Credit Risk Diversification, Working Paper; Available at SSRN: Santos, J., and A. Winton, 2008, Bank Loans, Bonds, and Information Monopolies Across the Business Cycle, Journal of Finance, 63, Treacy, W. F., and M. Carey, 2000, Credit Risk Rating Systems at Large U.S. Banks, Journal of Banking and Finance, 24, Whited, T. M., and G. Wu, 2006, Financial Constraints Risk, Review of Financial Studies, 19,

33 Table 1: Number of loans and filings This table shows the number of corporate loans from DealScan in each year of our sample (column (1)) and the number of bankruptcy filings from Bankruptcy Research Database and New Generation Research in each year of our sample (column (2)). The sample period is from 1990 to To be included in the loan sample, a borrower needs to be included in the Compustat database and must not have missing county information. The largest tranche is chosen for each deal from Dealscan. In column (1), Year refers to the time when a loan is originated for the loan sample. In column (2), Year refers to the number of bankruptcy filings in a given year. Year Number of loans Number of filings (1) (2) , , , , , , , , , , , , , , , , , Total 28,104 1,737 31

34 Table 2: Summary statistics and univariate differences Panel A of this table presents summary statistics of all loans in our sample as obtained from Dealscan and of all firms in our sample. Panel B of this table presents univariate differences between firms defined as local (i.e. firms located in a 50 kilometers county-radius surrounding a corporate bankruptcy event) and non-local firms (i.e. all firms located outside this radius), respectively. Panel A: Summary statistics Variable Number of obs. Mean Median Std. Dev. Loan characteristics Spread Coupon Deal amount Maturity Covenants Secured loan Sole lender Base is prime Relationship lender Performance pricing Refinancing Firm characteristics Market to book ratio Total assets Returns on assets Leverage Asset tangibility Cashflow volatility Return volatility Bankruptcy propensity Recession

35 Table 2: cont d Panel B: Differences between local and non-local firms Variable Local firms Non-local firms Difference t-stat. Loan characteristics Spread *** Coupon *** Deal amount Maturity *** Covenant Secured loan *** Sole lender ** Base is prime Relationship lender *** Performance pricing *** Refinancing Firm characteristics Total assets 2, , *** Market to book ratio *** Return on assets *** Leverage ratio *** Asset tangibility *** Cashflow volatility *** Return volatility ***

36 Table 3: Local bankruptcies and loan spreads: Baseline estimates The table presents estimates from regressions of firms loan spreads on a dummy variable which is equal to one if a local firm filed for bankruptcy in the previous year, and zero otherwise. All explanatory variables are defined in detail in Appendix A. The sample period is from 1990 to Standard errors are clustered at borrower level. t-statistics are provided in parentheses. ***, **, * indicate significance at the 1, 5, and 10 percent level, respectively. 34

37 (1) (2) (3) (4) Local bankruptcy i,t *** *** 9.189*** 9.189*** (4.99) (4.90) (4.33) (4.33) Total assets i,t *** *** *** *** (-3.10) (-3.93) (-5.03) (-5.03) Market to book ratio i,t *** *** *** *** (-5.68) (-5.71) (-7.35) (-7.35) Return on assets i,t *** *** *** *** (-10.63) (-11.06) (-12.64) (-12.64) Leverage ratio i,t *** *** *** *** (11.64) (10.77) (9.58) (9.58) Asset tangibility i,t ** (-0.88) (-2.54) (0.56) (0.56) Cashflow volatility i,t *** ** *** *** (3.11) (2.51) (6.08) (6.08) Return volatility i,t 1, *** 1, *** 2, *** 2, *** (18.56) (17.51) (11.37) (11.37) Bankruptcy propensity i,t *** *** *** *** (5.68) (6.05) (7.04) (7.04) Recession t *** 9.583*** 9.583*** (4.84) (4.18) (4.18) Deal amount i,t *** *** (-9.63) (-9.61) (1.00) (1.00) Covenant i,t (0.11) (0.20) (-1.02) (-1.02) Secured loan i,t *** *** *** *** (34.09) (32.94) (27.18) (27.18) Sole lender i,t (-0.44) (-0.40) (-0.43) (-0.43) Base is prime i,t *** *** *** *** (5.16) (5.05) (6.79) (6.79) Relationship lender i,t *** *** *** *** (-3.55) (-3.23) (-3.58) (-3.58) Performance pricing i,t *** *** *** *** (-12.96) (-12.69) (-11.35) (-11.35) Refinancing i,t 5.968*** 5.971*** *** *** (2.84) (2.84) (8.34) (8.34) Constant *** *** (15.72) (15.19) (0.96) (0.96) Observations 10,621 10,621 10,621 10,621 Adj. R Industry FE No Yes Yes Yes State FE Yes Yes Yes Yes Senior debt rating FE No No Yes Yes Type of loan FE No No Yes Yes Purpose of loan FE No No Yes Yes Squared control variables No No No Yes 35

38 Table 4: Local bankruptcies and loan spreads: Lead and lag effects The table presents estimates from regressions of firms loan spreads on a dummy variable which is equal to one if a local firm filed for bankruptcy in the previous year, and zero otherwise. In column (1), we include dummy variables for local bankruptcies in the current, and in the previous year. In column (2), we additionally include a dummy variable for local bankruptcies in the subsequent year. In column (3), we also include dummy variables for local bankruptcies two years ago, and two years ahead, respectively. The regression specification is the same as in Table 3, column (3). All explanatory variables are defined in detail in Appendix A. Standard errors are clustered at borrower level. t-statistics are provided in parentheses. ***, **, * indicate significance at the 1, 5, and 10 percent level, respectively. (1) (2) (3) Local bankruptcy i,t * (1.78) Local bankruptcy i,t *** 7.943*** 7.099*** (3.29) (3.23) (2.86) Local bankruptcy i,t (1.12) (1.09) (0.83) Local bankruptcy i,t (0.22) (0.07) Local bankruptcy i,t (-0.01) Constant *** *** *** (11.70) (11.70) (11.76) Observations 10,621 10,621 10,621 Adj. R Control variables Yes Yes Yes Industry FE Yes Yes Yes State FE Yes Yes Yes Senior debt rating FE Yes Yes Yes Type of loan FE Yes Yes Yes Purpose of loan FE Yes Yes Yes 36

39 Table 5: The impact of distance on geographic contagion The table presents estimates from regressions of firms loan spreads on a dummy variable which is equal to one if a local firm filed for bankruptcy in the previous year, and zero otherwise. The regression specification is the same as in Table 3, column (3). In addition to our baseline local bankruptcy variable (50 kilometers county-radius), we also include dummy variables reflecting a 100 and 200 kilometers county-radius surrounding a filing firm, respectively. All explanatory variables are defined in detail in Appendix A. Standard errors are clustered at borrower level. t- statistics are provided in parentheses. ***, **, * indicate significance at the 1, 5, and 10 percent level, respectively. (1) (2) (3) (4) Local bankruptcy i,t 1-100km 8.861*** (3.67) (0.54) Local bankruptcy i,t 1-200km 5.872** (2.53) (1.02) Local bankruptcy i,t ** 8.790*** (2.50) (3.77) Constant *** *** *** *** (11.40) (11.43) (11.42) (11.44) Observations 10,118 10,118 10,287 10,287 Adj. R Control variables Yes Yes Yes Yes Industry FE Yes Yes Yes Yes State FE Yes Yes Yes Yes Senior debt rating FE Yes Yes Yes Yes Type of loan FE Yes Yes Yes Yes Purpose of loan FE Yes Yes Yes Yes 37

40 Table 6: The impact of local bankruptcies on other loan terms Columns (1) to (3) of this table present panel regressions of firms coupon spreads (column (1)), loan amount (column (2)), or loan maturity (column (3)) on a dummy variable indicating a local bankruptcy in the previous year. The regression specification is the same as in Table 3, column (3). In column (4), the dependent variable is equal to one if loan covenants are present, and zero otherwise. In column (5), the dependent variable is equal to one for secured loans, and zero otherwise. Results in columns (4) and (5) are based on probit regressions. All explanatory variables are defined in detail in Appendix A. Standard errors are clustered at the borrower level. t-statistics are provided in parentheses. ***, **, * indicate significance at the 1, 5, and 10 percent level, respectively. 38

41 Coupon Amount Maturity Covenant Secured loan (1) (2) (3) (4) (5) Local bankruptcy i,t *** ** * 0.087** 0.104*** (4.35) (-2.14) (-1.91) (2.31) (2.73) Total assets i,t *** 0.555*** *** *** *** (-6.23) (57.38) (-6.96) (-2.65) (-11.13) Market to book ratio i,t *** 0.038*** * 0.020** *** (-6.03) (6.18) (-1.84) (2.20) (-4.73) Return on assets i,t *** 0.499*** 3.271* 0.319*** *** (-11.23) (7.44) (1.86) (2.78) (-8.59) Leverage ratio i,t *** 0.221*** 2.907** *** 1.344*** (9.28) (4.43) (2.22) (-5.12) (10.74) Asset tangibility i,t * 6.800*** * (-1.64) (-1.83) (4.79) (-0.98) (-1.91) Cashflow volatility i,t ** *** 0.871*** (2.20) (1.64) (-1.07) (3.06) (3.38) Return volatility i,t 1, *** *** *** *** (18.81) (-5.77) (-9.63) (1.12) (10.94) Bankruptcy propensity i,t *** ** (5.78) (-0.05) (-2.58) (-1.34) (1.17) Recession t 9.167*** *** *** ** (4.01) (-7.65) (-3.62) (-2.07) (1.17) Deal amount i,t *** 4.410*** 0.117*** *** (-9.63) (12.64) (5.21) (-3.99) Covenant i,t * *** 0.121*** (-0.41) (1.76) (-2.73) (2.84) Secured loan i,t *** *** ** 0.142*** (30.68) (-3.73) (-2.41) (3.22) Sole lender i,t *** *** (0.19) (-33.23) (-4.91) (0.58) (-1.44) Base is prime i,t *** *** 0.439*** 0.177*** (6.22) (-0.37) (-12.86) (7.86) (3.57) Relationship lender i,t *** 0.073*** *** *** 0.061* (-3.78) (4.67) (-5.47) (-3.65) (1.84) Performance pricing i,t *** 0.212*** 5.200*** 1.130*** *** (-11.54) (10.41) (10.06) (29.54) (-6.42) Refinancing i,t *** 0.056*** * 1.001*** 0.363*** (8.57) (2.67) (-1.66) (23.09) (8.42) Constant *** *** ** ** 6.864*** (11.50) (-26.22) (-2.34) (-2.51) (11.88) Observations 10,455 10,455 11,530 10,448 11,927 (Pseudo) R Industry FE Yes Yes Yes Yes Yes State FE Yes Yes Yes Yes Yes Senior debt rating FE Yes Yes Yes Yes Yes Type of loan FE Yes Yes Yes Yes Yes Purpose of loan FE Yes Yes Yes Yes Yes 39

42 40 Table 7: Robustness checks The table presents estimates from regressions of firms loan spreads on a dummy variable indicating a local bankruptcy in the previous year. The regression specification is the same as in Table 3, column (3). In column (1), we exclude all firms from the same industry as the filing firm. In column (2), we include industry sales growth and industry cash flow as additional control variables. In column (3), we drop local firms with negative cash-flow shocks from the sample. In column (4), we include state sales growth and state cash flows as additional control variables. In column (5), we combine the exclusion restrictions from columns (1) and (3). Column (6) combines the additional control variables from columns (2) and (4). Column (7) includes the number of non-local bankruptcies, and column (8) the total number of bankruptcies in a given year. All explanatory variables are defined in detail in Appendix A. Standard errors are clustered at borrower level. t-statistics are provided in parentheses. ***, **, * indicate significance at the 1, 5, and 10 percent level, respectively.

43 (1) (2) (3) (4) (5) (6) (7) (8) Local bankruptcy i,t *** 9.450*** 8.721*** 7.523*** 8.133*** 7.553*** 7.836*** 7.532*** (4.26) (4.42) (3.69) (3.48) (3.16) (3.50) (3.61) (3.46) Industry sales growth i,t ** (-2.25) (0.33) Industry cash flow i,t (-0.08) (1.07) State sales growth i,t *** *** (-5.27) (-5.03) State cash flow i,t *** *** (-3.49) (-3.77) Non-local bankruptcies i,t 0.149*** Total bankruptcies t 0.145*** Constant *** *** *** *** *** *** *** *** (6.13) (6.09) 41 (11.11) (11.48) (10.46) (11.67) (10.05) (11.43) (11.25) (11.26) Observations 7,753 10,621 8,247 10,621 6,150 10,621 10,455 10,455 Adj. R Control variables Yes Yes Yes Yes Yes Yes Yes Yes Industry FE Yes Yes Yes Yes Yes Yes Yes Yes State FE Yes Yes Yes Yes Yes Yes Yes Yes Senior debt rating FE Yes Yes Yes Yes Yes Yes Yes Yes Type of loan FE Yes Yes Yes Yes Yes Yes Yes Yes Purpose of loan FE Yes Yes Yes Yes Yes Yes Yes Yes

44 Table 8: Controlling for time-varying unobservable industry and location factors The table presents estimates from regressions of firms loan spreads on a dummy variable indicating a local bankruptcy in the previous year. The regression specification is the same as in Table 3, column (3). Additional fixed effects are as indicated below. All explanatory variables are defined in detail in Appendix A. Standard errors are clustered at borrower level. t-statistics are provided in parentheses. ***, **, * indicate significance at the 1, 5, and 10 percent level, respectively. (1) (2) (3) (4) (5) Local bankruptcy 4.816** 7.593*** 4.967** 7.969*** *** (2.39) (3.02) (2.57) (3.02) (4.05) Control variables Yes Yes Yes Yes Yes Observations 10,621 10,621 10,621 10,621 10,621 Adj. R Industry-yr FE Yes No No Yes No State-yr FE No Yes No Yes No Year FE No No Yes Yes No Firm FE No No No No Yes Senior debt rating FE Yes Yes Yes Yes Yes Type of loan FE Yes Yes Yes Yes Yes Purpose of loan FE Yes Yes Yes Yes Yes

45 Table 9: Instrumental variable approach The table presents estimates from IV regressions of firms loan spread on a dummy variable which is equal to one if a local firm filed for bankruptcy in the previous year, and zero otherwise. The list of the control variables is the same as in Table 3, column (3). Columns (1) and (3) present the results of the first-stage regressions. Columns (2) and (4) report the results of the second-stage regressions. Non-local industry cash flows is the median cash flows of non-local industry peers. Nonlocal industry sale growth is the median sale growth of non-local industry peers. All explanatory variables are defined in detail in Appendix A. Standard errors are clustered at borrower level. t- statistics are provided in parentheses. ***, **, * indicate significance at the 1, 5, and 10 percent level, respectively. Local bankruptcy Spread Local bankruptcy Spread 1 st stage 2 nd stage 1 st stage 2 nd stage (1) (2) (3) (4) Non-local industry cash flows *** (-8.13) Non-local industry sale growth *** (-5.14) Local bankruptcy (Instrumented) *** *** (4.63) (2.87) Observations 10,452 10,452 10,452 10,452 Adj. R Control variables Yes Yes Yes Yes Industry FE Yes Yes Yes Yes State FE Yes Yes Yes Yes Senior debt rating FE Yes Yes Yes Yes Type of loan FE Yes Yes Yes Yes Purpose of loan FE Yes Yes Yes Yes 43

46 Table 10: Impact of borrower and lender characteristics Panel A of this table presents estimates from regressions of firms loan spreads on a dummy variable indicating a local bankruptcy interacted with borrower characteristics. The regression specification is the same as in Table 3, column (3). In column (1), we interact the local bankruptcy variable with an indicator for old firms (defined at the median firm age in the sample). In column (2), we interact the local bankruptcy variable with an indicator for large firms (defined at the median firm size in the sample). In columns (3) and (4), we interact the local bankruptcy variable with an indicator for financial distress (defined at the median financial distress level in the sample). In columns (6) and (7), we interact the local bankruptcy variable with an indicator for diversified business at the country, or state level, respectively. Panel B of this table presents estimates from regressions of firms loan spreads on a dummy variable indicating a local bankruptcy interacted with lender characteristics. The regression specification is the same as in Table 3, column (3). In column (1), we interact the local bankruptcy variable with lenders local lending amount. In column (2), we interact the local bankruptcy variable with the number of lenders local deals. In column (3), we interact the local bankruptcy variable with an indicator of local lender concentration (defined at the median Herfindahl concentration in the sample). We also include the logarithm of the number of local lenders as an additional control variable. All explanatory variables are defined in detail in Appendix A. Standard errors are clustered at borrower level. t-statistics are provided in parentheses. ***, **, * indicate significance at the 1, 5, and 10 percent level, respectively. 44

47 Panel A: Borrower characteristics Table 10: cont d (1) (2) (3) (4) (5) (6) Local bankruptcy i,t *** *** 8.651*** 7.151** *** *** Local bankruptcy i,t Old firm i,t 1 (-1.11) Old firm i,t (3.51) (2.95) (3.53) (2.32) (3.91) (3.06) (0.72) Local bankruptcy i,t Large firm i,t 1 (-0.73) Large firm i,t * (-1.83) Local bankruptcy i,t Whited-Wu i,t 1 (0.57) Whited-Wu index i,t (1.39) Local bankruptcy i,t KZ index i,t 1 (0.88) KZ index i,t *** (3.78) Local bankruptcy i,t Country div. i,t 1 (-1.25) Country diversification i,t Local bankruptcy i,t State div. i,t 1 (-0.86) State diversification i,t Constant *** *** *** *** *** *** (1.35) (0.79) (11.75) (11.50) (11.11) (12.03) (10.66) (10.66) Observations 10,621 10,607 10,567 10,457 9,250 9,250 Adj. R Control variables Yes Yes Yes Yes Yes Yes Industry FE Yes Yes Yes Yes Yes Yes State FE Yes Yes Yes Yes Yes Yes Senior debt rating FE Yes Yes Yes Yes Yes Yes Type of loan FE Yes Yes Yes Yes Yes Yes Purpose of loan FE Yes Yes Yes Yes Yes Yes 45

48 Panel B: Lender characteristics Table 10: cont d (1) (2) (3) Local bankruptcy i,t *** 7.808** 7.788*** Local bankruptcy i,t 1 Local lending amount i,t 0.974* (3.36) (2.54) (2.74) (1.66) Local lending amount i,t *** (-2.91) Local bankruptcy i,t 1 Local deals i,t (1.57) Local deals i,t *** (-3.98) Local bankruptcy i,t 1 HHI of local lenders i,t (1.37) HHI of local lenders i,t * (-1.72) Local firms i,t * (0.56) (1.78) (-1.34) Constant *** *** *** (11.19) (11.48) (11.72) Observations 10,621 10,621 10,614 Adj. R Control variables Yes Yes Yes Industry FE Yes Yes Yes State FE Yes Yes Yes Senior debt rating FE Yes Yes Yes Type of loan FE Yes Yes Yes Purpose of loan FE Yes Yes Yes 46

49 Table 11: Estimates for lenders without bankruptcy exposure or limited local presence The table presents estimates from regressions of firms loan spreads on a variable indicating a local bankruptcy in the previous year for subsamples of lenders that do not have exposure to local bankruptcies or have limited exposure to local economic conditions. The regression specification is the same as in Table 3, column (3). In column (1), we only include lenders that do not have bankruptcy exposure. In column (2), we only include lenders from the top 30 largest banks in the US. All explanatory variables are defined in detail in Appendix A. Standard errors are clustered at the borrower level. t-statistics are provided in parentheses. ***, **, * indicate significance at the 1, 5, and 10 percent level, respectively. No bankruptcy exposure Top 30 banks (1) (2) Local bankruptcy i,t *** 9.515*** (4.56) (3.88) Constant *** *** (15.82) (13.40) Observations 9,964 5,469 Adj. R Control variables Yes Yes Industry FE Yes Yes State FE Yes Yes Senior debt rating FE Yes Yes Type of loan FE Yes Yes Purpose of loan FE Yes Yes 47

50 Table 12: Impact of local bankruptcies on local default risks The table presents estimates from regressions of distress probability on a dummy variable indicating a local bankruptcy in the previous year. In column (1), the dependent variable is the modified Altman s Z-score. In column (2), the dependent variable is a default indicator that equals one if a firm s debt rating is rated either D or SD, and zero otherwise. In column 3, the dependent variable is a downgrade indicator that equals one if a firm s debt rating is downgraded in a given year, and zero otherwise. In column 4, the dependent variable is an indicator of borrower s bankruptcy propensity that equals one if a borrower files for bankruptcy within three years of the loan origination date, and zero otherwise. All regressions include industry, year, and state fixed effects. All explanatory variables are defined in detail in Appendix A. Standard errors are clustered at the borrower level. t-statistics are provided in parentheses. ***, **, * indicate significance at the 1, 5, and 10 percent level, respectively. Z-score Default Downgrade Bankruptcy propensity (1) (2) (3) (4) Local bankruptcy i,t (-1.49) (0.79) (1.31) (-0.33) Constant 1.383*** 0.002*** 0.035*** 0.014*** (46.95) (3.60) (19.16) (14.13) Observations 20,619 22,345 22,345 22,345 Adj. R Year FE Yes Yes Yes Yes Industry FE Yes Yes Yes Yes State FE Yes Yes Yes Yes 48

51 Figure 1: Bankruptcies across U.S. States from 1990 to 2006 This figure shows the percentage of bankruptcies in each U.S. state as a fraction of all bankruptcies from 1990 to