Policy Research Working Paper 9202 Owe a Bank Millions, the Bank Has a Problem Credit Concentration in Bad Times Sumit Agarwal Ricardo Correa Bernardo Morais Jessica Roldán Claudia Ruiz-Ortega Development Economics Development Research Group April 2020 Policy Research Working Paper 9202 Abstract This paper studies the dynamics of credit supply when a to borrowers in the sector by 18 percent and reduced loan negative shock impacts a substantial share of bank loans. rates charged by 6 percent, even though borrower’s CDS The analysis exploits the 2014 collapse of energy prices, spreads were widening. Highly exposed banks amplified using the universe of Mexican commercial bank loans. The this sector-specific shock to the rest of the economy by findings show that, after the drop in energy prices, the credit contracting lending to other sectors, with important real default swap spreads (CDS) of firms in the energy sector effects, as the borrowers could not switch credit suppliers. soared, and banks that were more exposed to the energy Finally, the energy price shock had a large negative impact sector increased even more their exposure ex post, by sup- on macro outcomes, especially in the capital-intensive plying loans to their larger debtors in the energy sector at secondary sector. Quantitatively, a one standard deviation lower interest rates. An increase of one standard deviation in increase in the exposure of a state’s banks to the energy ex-ante exposure to the energy sector increased loan volume sector reduces its GDP by 1.8 percent. This paper is a product of the Development Research Group, Development Economics. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at cruizortega@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Owe a Bank Millions, the Bank Has a Problem: Credit Concentration in Bad Times Sumit Agarwal Ricardo Correa Bernardo Morais Jessica Roldán Claudia Ruiz-Ortega 1 JEL codes: E52, E58, G01, G21, G28. Keywords: credit exposure, bank lending, financial stability, commodity prices, emerging markets. 1 This draft is from February 2020. Sumit Agarwal: National University of Singapore, usharki@yahoo.com; Ricardo Correa: Federal Reserve Board, ricardo.correa@frb.org; Bernardo Morais: Federal Reserve Board, bernardo.c.morais@frb.gov (contact author); Jessica Roldán: Casa de Bolsa Finamex, jroldan@finamex.com.mx; and Claudia Ruiz-Ortega: DECFP, World Bank, cruizortega@worldbank.org. We are grateful to Banco de México and Adrián De la Garza, for their support of this project. The data were accessed through the Econlab at Banco de México. The EconLab collected and processed the data as part of its effort to promote evidence-based research and foster ties between Banco de México’s research staff and the academic community. Inquiries about the terms under which the data can be accessed should be directed to: econlab@banxico.org.mx. Jessica Roldán thanks the Directorate General of Economic Research of Banco de México, where she contributed to this project when she was head of the Monetary Research Division at this institution. We thank the seminar participants at Banca d’Italia, Banco de México, Banco de Portugal, CREI, World Bank, Vanderbilt University, and the Federal Reserve Board for helpful comments. We thank Carlos Zarazúa and Ben Smith for outstanding research assistance. The views in this paper are solely the responsibility of the authors and should not be interpreted as reflecting the views of the Board of Governors of the Federal Reserve System or any other person associated with the Federal Reserve System, the World Bank, or Banco de México. Banco de México requested to review the results of the study prior to dissemination to ensure confidentiality of the data. 1. Introduction Risk concentration has been a driver of major banking crises around the world (Acharya and Steffen 2015; Brunnermeier 2009; Westernhagen et al. 2004), forcing regulators to continuously monitor bank exposures to concentrated risks (FSI 2019). However, although regulation considers exposures to single and financially connected counterparties, there is limited knowledge on the strategies that banks adopt when their counterparties face a common negative shock, like a sectoral shock. 2 On the one hand, banks may scale back their lending to the impacted sector to reduce their losses and possibly diversify their loan portfolios. On the other hand, the actions taken by banks may depend on the bargaining power of their borrowers (Rajan 1992; Santos and Winton 2019). More exposed banks may be forced to expand their lending to the struggling sector, especially to their largest borrowers, to preserve their regulatory ratios. In this latter scenario, risk concentration may trigger a credit channel whereby banks inject even more credit to borrowers in a troubled industry, reducing credit to other sectors in the economy. This not only leads to a misallocation of resources away from productive borrowers, but also raises the risk of financial stress, given the increased concentration in a weak segment of the economy. Theoretical studies have stressed the trade-offs faced by banks in their portfolio choices. Although portfolio diversification allows banks to enhance their credit monitoring reputation (Diamond 1984; Boyd and Prescott 1986), bank specialization may provide better bank performance under certain circumstances (Winton 1999). The empirical literature has been mixed on this question. Some studies have stressed the aggregate benefits of bank specialization on systemic risks (Beck, De Jonghe, and Mulier 2017) and its benefits for borrowers with close bank relations (De Jonghe et al. 2019). In contrast, other studies have noted how shocks to specialized banks may affect credit provision (Paravisini, Schnabl, and Rappoport 2017) and how geographic diversification reduces bank risk (Goetz, Laeven, and Levine 2016). Given these mixed results, we analyze the impact of a negative sectoral shock on the supply of credit as the degree of banks’ exposure to the energy sector varies. We study the collapse of global energy prices in late 2014 as a shock to the Mexican energy sector. Although the price drop 2 For example, when prices in the energy sector collapsed in March of 2020 the Financial Times wrote that “investors are confronted with the alarming possibility that a collapse in oil prices could trigger a wave of defaults by borrowers. […] U.S. bank shares had their worst single-session performance since 2009 and the industry was a big contributor to a global stock market rout.” 2 was driven by factors external to Mexico, the credit risk of Mexican firms operating in energy- related sectors ramped up as a result. 3 Importantly, the degree of Mexican banks’ exposure to the energy sector varied substantially before the shock. We exploit this cross-bank variation to identify how banks reallocated their credit depending on their ex-ante exposure to the struggling sector. We show that banks with large exposures to the energy sector had an incentive to maintain borrowers afloat even as their creditworthiness deteriorated, causing a decline in lending to other sectors (Caballero, Hoshi, and Kashyap 2008; Peek and Rosengren 2005). How banks cope with shocks and transmit them to other sectors of the economy remains an important issue in finance. However, there are several challenges to overcome to answer these questions rigorously. One challenge is having a credible counterfactual, since aggregate shocks may affect the entire banking sector simultaneously. To overcome this hurdle, we exploit the late 2014 oil price shock and adopt a difference-in-differences approach with ex-ante similar banks differing in their exposure to energy-related sectors prior to the shock. We define ex-ante bank exposure as the ratio of loans to firms in energy-related sectors over the bank’s tier 1 capital in the month prior to the unanticipated shock. 4 A second challenge in identifying banks’ strategies is to isolate changes in the supply of credit from changes in the demand for credit, as aggregate shocks might impact firms’ credit needs. To control for time-varying credit demand, we rely on loan-level data obtained from the Mexican credit registry on the universe of commercial bank loans from January 2013 to June 2016. Loan- level data allow us to saturate our specifications with bank*firm and firm*month fixed effects, exploiting variation in the credit conditions of a firm-bank pair over time, as well as by the same firm across different banks with varying exposures to the shock (Khwaja and Mian 2008; Morais et al. 2019). 3 The global energy price drop was related to the advent of new oil suppliers, such as shale producers in the United States (World Bank 2018). Figure 1 shows the rapid drop of West Texas Intermediate prices from almost $100 per barrel in mid-2014 to about $50 per barrel over just a few months. This drop was associated with a weakening of the creditworthiness of energy companies in Mexico and other emerging markets, as shown by the increase in their credit default swap premiums. 4 Our measure of exposure follows the definition of credit exposure outlined by the Basel Committee on Banking Supervision (2014). As a robustness check, we use four alternative bank-level measures of ex-ante exposure and confirm that our findings remain unchanged. These measures are (i) August 2014 ratio of energy sector loans to total loans, (ii) August 2014 ratio of energy sector loans to total assets, (iii) August 2014 ratio of energy sector loans to total bank equity, and (iv) December 2012 ratio of energy sector loans to tier 1 capital (Table A4 in the appendix). 3 Using bank-level data, we first investigate the impact that larger exposure to the energy sector had on banks’ balance sheet outcomes and risk dynamics after the shock. We then use the loan-level data to examine separately the lending dynamics of firms in energy-related sectors and all other sectors. Focusing first on borrowers in energy-related sectors, we examine changes in the value and terms of their loans across banks after the shock. Our outcomes of interest include the total amount of credit borrowed by energy firms, as well as the amount borrowed to finance working capital and investment projects. Other outcomes we analyze include interest rates, collateral rates, and loan maturities. To investigate whether banks transmit this sector-specific shock to other sectors, we compare lending to non-energy firms by banks with different degrees of exposure to the energy sector before and after the energy price shock. Since firms may switch to other financing sources to smooth bank credit shocks, we complement our loan-level data with firms’ yearly balance sheet information. The firm-level data allow us to identify whether shocks that affect the credit conditions of firms also affect their real outcomes. We construct a measure of exposure to the energy price shock at the municipality level. In addition, we investigate whether the states with more exposed banks had a sharper slowdown. We use quarterly gross domestic product (GDP) data for Mexican states and construct a state-level measure of exposure to the energy sector of the banks operating in the state. We find that banks with greater ex-ante exposure to the energy sector significantly expanded their ex post lending to firms in the sector. This expansion took the form of larger loans for working capital at lower interest rates, suggesting that banks attempted to keep stressed firms afloat and their own capital ratios intact. For instance, an increase of one standard deviation in ex- ante exposure to the energy sector (relative to the bank’s tier 1 capital) is associated with an increase of 1.2 percentage points in ex post exposure to the sector, an 18 percent increase in the size of loans to firms in the energy sector, and a 0.8 percentage point (roughly 6 percent) decrease in the interest rate. Consistent with a hold-up problem, these economically important magnitudes are concentrated among larger energy firms with which banks have greater exposures. This strategy was associated with a substantial increase in the risk taken by more exposed banks. An increase of one standard deviation in exposure to the energy sector results in 10.1 percent higher 4 credit default swap (CDS) spreads, 3.1 percent lower stock prices, and 0.16 percentage point (roughly 8 percent) higher delinquency rates in the following quarters. 5 The injection of credit to the energy sector did not result in an increase in total bank lending, as credit was redirected from firms in other sectors. The loan-level analysis reveals that the credit contraction among non-energy borrowers was stronger among smaller firms and especially for loans destined to investment projects. An increase of one standard deviation in ex-ante bank exposure leads to a 16.5 percent reduction in credit to smaller firms. 6 We also find significant negative real effects on the activities of non-energy firms as a result of the contraction of bank credit. Non-energy firms headquartered in municipalities where banks had higher ex-ante exposure to the energy sector experienced a decrease in liabilities, investment, and assets after the energy price shock. An increase of one standard deviation in a municipality’s exposure to the energy sector (via its bank branches) leads to a reduction in total liabilities of 2.9 percent and a reduction in assets of 2.6 percent. At a more aggregate level, we find that, compared with energy-producing states, the GDP of non-energy-producing states that were more exposed to the energy sector (via their banks) contracted more during the energy price shock, especially in the capital-intensive secondary sector (Buera, Kaboski, and Shin 2011). An increase of one standard deviation in the exposure of a state’s banks to the energy sector reduces the state’s total GDP by 1.8 percent. We interpret these findings as evidence of a credit channel, whereby banks amplified a sector-specific shock by contracting their lending to non-energy borrowers, which in turn struggled to switch lenders and smooth the shock. Our findings suggest that risk concentration, or specialization, can amplify negative shocks. This is particularly relevant if banks need to provide additional credit to an ailing sector. Not only do bank exposures become riskier, but also banks must curtail credit from other areas of the economy, reducing the performance of nonaffected firms. Although regulations could prevent 5 We also test whether in the post-shock period banks’ nonperforming loans increased or capital levels decreased. This would point to the energy shock affecting the creditworthiness of Mexican banks. However, we do not find any significant relation between banks’ exposure to the energy sector and measures of bank solvency. Alternatively, the banks may have extended lending to creditworthy energy firms that were only facing a transitory shock. However, this is not consistent with the steep increase in CDS premiums shown in Figure 1. 6 These findings are consistent with Bidder, Krainer, and Shapiro (2017), who also find that, in response to the 2014 energy price collapse, U.S. banks did not change the overall size of their credit portfolio, but they reduced the risk of their portfolio. 5 these types of concentrated exposures, supervision may play a better role. Given the mixed evidence on the effects of specialized portfolios, supervisors should not only limit the risks of concentrated sectoral exposures, but also estimate the likelihood that those sectors may suffer from any shocks. This is particularly relevant for commodity sectors where price fluctuations are sharp and frequent and have a strong correlated impact on all firms in the industry. Our paper relates to several literatures. First, it contributes to the literature studying the relationship between bank diversification and performance. Diamond (1984) argues that as banks increase diversification, their vulnerability to economic downturns and risk of default drops. However, recent empirical studies find that diversification is negatively associated with banks’ returns and monitoring effectiveness, while positively related to their risk (Acharya, Hasan, and Saunders 2006; Laeven and Levine 2007; Berger, Hasan, and Zhou 2010; Tabak, Fazio, and Cajueiro 2011). Our paper complements this empirical literature by exploiting an exogenous shock to commodity prices to identify the costs that may stem from banks’ high sectoral exposures. Second, we complement the extensive literature studying the negative effects that volatile commodity price shocks have on financial markets and economic growth (Agarwal, Duttagupta, and Presbitero 2019; Blanchard and Gali 2010; Bruckner and Ciccone 2010; Alesina, Campante, and Tabellini 2008; Dehn 2000; Kinda, Mlachila, and Ouedraogo 2016; Deaton 1999; Deaton and Miller 1996). We contribute to this literature by documenting that banks amplify commodity price shocks via the credit channel. Last, we contribute to the literature studying the effects of liquidity shocks on bank lending. This literature has traditionally examined how banks transmit shocks to the real sector via changes in their credit supply, by exploiting changes in the domestic monetary policy (Kashyap and Stein 2000), local liquidity shocks (Gilje, Loutskina, and Strahan 2016; Khwaja and Mian 2008; Iyer and Peydro 2011), and global liquidity shocks (Schnabl 2012; Morais et al. 2019; De Jonhge et al. 2019; Ippolito et al. 2016). Different from these papers, we examine a liquidity shock that has not been as widely explored. The shock we examine works through troubled borrowers’ demand for short-term funding and its effects on banks’ asset side of their balance sheets instead of their liabilities. 6 The rest of the paper is organized as follows. Section 2 describes the data used in the analysis. Section 3 discusses the empirical strategy we follow. The results are summarized in section 4. Finally, section 5 concludes. 2. Data We use data from three main sources, covering January 2013 to June 2016. The first data set, which we refer to as the loan-level data, consists of the universe of commercial loans in Mexico, which we obtained from regulatory reports sent monthly by every commercial bank to the bank regulator. The reports are mandatory, updated electronically, and include detailed characteristics of all new and continuing commercial loans. All loans, regardless of their size, are reported. Each loan has an identifier of the issuing bank, as well as the borrower’s identifier, location, sector, and number of employees. The data set includes information on the interest rate, outstanding amount, type of financing (i.e., whether the loan is for working capital or investment purposes), and start and end dates (maturity) of each loan. Given that some borrowers have more than one loan issued by the same bank at a given point in time, we adopt a similar approach as La Porta et al. (2003) and aggregate individual loans at the firm-bank-month level. We then report loan characteristics, such as the interest rate, fraction of the loan covered by collateral, and maturity at origination, using a weighted average by loan value. This approach puts greater weight on larger loans, ensuring that our results are economically meaningful. Our second data source is Orbis, a firm-year-level data set compiled by Bureau van Dijk, which contains information on the balance sheets and income statements of a large set of Mexican firms. The data set reports information on assets and revenues of firms as well as their total and bank-specific liabilities by type of financing. As shown by Morais et al. (2019), this sample of firms is representative of the universe of sectors and locations in Mexico, albeit somewhat skewed toward larger firms. We complement this data set with a measure of GDP for Mexico’s 32 states, normalized to 2004, which was obtained from the National Statistics Institute. In addition to the total GDP, we also use information on the GDP contributed by the primary sector of each state, consisting of mining and agriculture; the GDP contributed by the secondary sector, covering manufacturing and construction; and the GDP contributed by the tertiary sector, defined as services. 7 The third data set contains the monthly balance sheet information of the 18 commercial banks in our sample, representing more than 98 percent of commercial bank lending. 7 We merge this data set with information from Bloomberg on the stock prices and CDS spreads of the banks. Overall, our data contain a total of 1,718,740 loans to firms in the energy and non-energy- related sectors. We classify firms as belonging to an energy-related sector according to their 5- digit North American Industry Classification System (NAICS) codes. 8 The summary statistics for our sample are shown in Table 1, grouped in five panels: (i) bank-month-level indicators, (ii) loan- level variables of firms in energy-related sectors, (iii) loan-level variables of firms in non-energy sectors, (iv) real outcomes at the firm-year level, and (v) GDP measures at the state-quarter level. Table A1 in the appendix presents the definitions of all the variables. The first variable in Table 1, panel A, captures the banks’ exposure to borrowers in energy- related sectors as a share of their tier 1 capital. On average, the ratio of exposure to capital is 9.9 percent, with banks in the bottom decile having no exposure to the energy sector, while banks in the top decile have exposure above 25 percent. Our measure of exposure follows the Basel Committee’s assessment of exposure to related entities, which is defined as the credit volume of a bank to related entities as a share of its tier 1 capital. 9 The next variables in panel A correspond to different elements of the banks’ balance sheets, including their tier 1 capital ratios, lending portfolio, and delinquency rates, along with statistics on the banks’ stock prices and CDS spreads. The average tier 1 capital ratio of the banks is 15.4 percent, with the banks in the bottom decile having a capital ratio of 12.5 percent, while the banks in the top decile have a capital ratio of 18.5 percent. The banks vary greatly in size, with the average bank lending more than Mex$3,000 million, and the banks in the top decile lending 100 times as much as the banks in the lowest decile. Delinquency rates are low. On average, 2.4 percent of the banks’ loans are more than 90 days late, and the banks in the top decile have delinquencies of 7 To guarantee the comparability of our results across banks, and given our focus on commercial lending, we exclude from our analysis banks that specialize in consumer lending as well as niche banking. 8 Table A2 in the appendix displays the NAICS energy-related sectors as well as their descriptions. 9 According to the Basel Committee on Banking Supervision (2014), two entities are related: (i) if one of the counterparties, directly or indirectly, has control over the other, or, (ii) if 50 percent or more of one counterparty's receipts comes from transactions with the other counterparty, or (iii) if a significant part of a counterparty’s production is sold to another counterparty, or (iv) if financial problems of one counterparty cause difficulties for the other counterparties, or (v) if counterparties rely on the same source for their funding and an alternative provider cannot be found in a timely manner. 8 around 4.7 percent. Finally, the bottom line of panel A shows the statistics for the banks’ ratio of exposure to the energy sector in the month prior to the energy price shock. On average, the exposure is around 8 percent, with banks in the bottom decile having zero exposure to the energy sector, whereas banks in the top decile have 19.6 percent exposure. Table 1, panel B, reports the loan characteristics of firms operating in energy-related sectors. Although the average bank loan of an energy firm is around Mex$89 million, the median loan size is Mex$1.5 million. Interest rates average 11.4 percent, with loans in the bottom decile having rates as low as 4.4 percent and in the top decile 18 percent. The maturity of loans is on average around two years, with loans in the bottom decile having maturities of around two months, whereas the top decile has maturities of four years. 10 These short maturities are consistent with most loans being destined for working capital. Collateral rates average 14.6 percent of the value of the loan and, again, there is great variation in the amount of collateral required across loans. Although the median loan is uncollateralized, loans in the top decile require collateral of more than half their value. The loan characteristics of firms operating in non-energy sectors display similar patterns as those of firms in the energy sector (Table 1, panel C). The last two panels in Table 1 present summary statistics for real outcomes at the firm-year level and aggregate production at the state-quarter level. As panel D shows, the median bank debt of firms according to the credit registry is around Mex$730,000. From the Orbis data set, we find that the median liabilities of firms are around Mex$350 million, with median assets and revenues of around Mex$1.1 billion and Mex$800 million, respectively. We construct AvgExposureEnergym,Aug14, a measure of banks’ exposure to the energy price shock at the municipality level, as the average exposure to the energy sector in August 2014, weighted by loan value, of the banks serving municipality m. 11 Given that lending tends to be local (Degryse and Ongena 2005), this measure allows us to capture variation in the exposure of firms to banks that were more affected by the shock. Finally, panel E shows two statistics at the state-quarter level. The first one is the state GDP, which is measured as an index normalized for each state to its level in January 2014. The second variable, AvgExposureEnergys,Aug14, corresponds to the average exposure to the energy 10 These maturities do not include revolving loans. 11 In Mexico, there are 2,448 municipalities, with an average population of around 400,000 people. 9 sector of banks in state s in August 2014, the month prior to the oil price shock. Although the average exposure of banks across states is 12.1 percent, for states in the bottom decile it is 9.6 percent, and for states in the top decile it is 14.3 percent. 3. Methodology We use the 2014 collapse of global energy prices as an exogenous shock to the Mexican banking sector to assess the implications for banks of large exposures to a troubled sector. Banks with large exposures to ailing sectors may suffer due to weaker capital ratios as loans become delinquent, or losses on those exposures as loans default. Therefore, the banks might have incentives to expand lending to these borrowers. However, these actions can come at the expense of increased risk and lower returns, by taking lending away from borrowers in unaffected sectors. To investigate the impact of this external shock on banks’ balance sheets and credit allocation, we adopt a difference-in-differences approach in which treatment is continuous and corresponds to the banks’ exposure to borrowers in energy-related sectors in the month prior to the unanticipated shock. This measure of bank exposure consists of the August 2014 ratio of loans to firms in energy-related sectors issued by a bank over its tier 1 capital. 12 In Figure 2, we classify banks into two groups according to their August 2014 exposure to the energy sector. The three panels in the figure provide descriptive evidence that the effect of the global energy price collapse was more pronounced among banks with greater exposure to the energy sector prior to the price drop. The group labeled “high exposure” includes banks with exposures above the median (5 percent) for the selected date, while “low exposure” banks had exposures below the median. For each group, we plot their exposures to the energy sector (panel A), CDS spreads (panel B), and stock prices (panel C) from January 2013 to June 2016. Panel A shows that although there were substantial differences in the level of exposure to the energy sector across banks prior to the shock, the variation across these groups was constant from January 2013 through August 2014 and followed a parallel trend. The shares of lending to energy firms of banks above and below the median exposure were on average around 10 and 3.8 12However, our results remain unchanged using alternative measures of exposure or different periods (Table A6 in the appendix). These measures correspond to the December 2012—the month prior to the start of our sample— exposure to energy firms of banks and the December 2012 number of branches in energy-intensive municipalities over the total number of branches of a bank. 10 percent, respectively. However, after the price shock, banks that were more exposed increased their exposure to the sector, reaching 30 percent by mid-2016, while the share of lending to the energy sector by banks that were less exposed was around 8 percent. The data thus suggests that the increased share of bank lending to the energy sector that followed the drop in oil prices (Figure A1) was driven by banks with greater exposures. Furthermore, panels B and C of Figure 2 show that although both types of banks had similar trends in their CDS spreads and stock prices, these trends diverged after the energy price drop. Normalizing the CDS spreads of both groups of banks to their values in August 2014, we find that through mid-2016, the banks with high exposure saw their CDS spreads reach 100 basis points, whereas the remaining banks reached only 50 basis points. Similarly, the stock prices of banks with high exposure declined by around 12 percent through mid-2016, whereas the stock prices of the banks with low exposure increased 10 percent. All in all, this descriptive evidence suggests that the financial conditions of banks with higher exposure to the energy sector became relatively worse following the collapse of energy prices. We run equation 1 to test more formally the impact that exposure to the energy sector had on the banks’ balance sheets after the collapse of energy prices. yb,m = α + βExposureEnergyb,Aug14*Postm + γm + γb + εb,m (1) Our five outcomes of interest (yb,m) at the bank-month level correspond to the exposure to the energy sector, total lending, CDS spreads, stock prices, and delinquency ratio. We regress these outcomes on the interaction of the August 2014 exposure to the energy sector of bank b— ExposureEnergyb,Aug14—and a dummy variable—Postm—that equals one from September 2014 onward. We also include fixed effects at the bank and month levels, with standard errors double clustered at the bank and month levels. To study the impact of the energy price shock on loans to energy-related firms by banks with varying exposure to the distressed sector, we use our loan-level data and run the regression summarized in equation 2. yf,b,m = α + βExposureEnergyb,Aug14*Postm + γb,f + γm + εf,b,m (2) 11 where yf,b,m corresponds to the amount loaned to firm f by bank b in month m for all types of loans as well as working capital and investment loans. The interest rate, collateral rate, and maturity of loans to firm f by bank b in month m are additional credit outcomes that we analyze. Equation 2 includes firm-bank fixed effects, γb,f, and month fixed effects, γm, with robust standard errors double clustered at the bank and month levels. Furthermore, in some specifications, we include firm-month fixed effects, γf,m, to control for changes in the demand for credit. A key identifying assumption for estimating the causal effects of the change in energy prices is that the trends in the outcomes of interest would have been the same across banks in the absence of the energy price drop. Although this assumption cannot be tested, we test for differences in bank outcomes and their trends before the energy price drop, using the regression outlined in equation 3, constraining the sample to the period before August 2014. yb,m = α + β1ExposureEnergyb,Aug14 + β2ExposureEnergyb,Aug14*Trendm + γm + εb,m (3) In equation 3, yb,m corresponds to the outcomes of interest for bank b at time t, and the term Trendm consists of a linear trend over time. As before, ExposureEnergyb,Aug14 captures bank b’s August 2014 exposure to the energy sector. Coefficient β1 measures whether the average outcomes of banks are statistically different as their exposure to the energy sector varies, whereas coefficient β2 measures differences in the trends of outcome y across banks with varying exposures to the energy sector. Fixed effects at the month level, γm, are included in the regression. The results, summarized in Table A3 in the appendix, give credibility to the identification strategy, as they show that there are no statistically significant differences in the pre-shock averages and trends of the outcomes of interest. We conduct an additional pre-trends test using the loan-level data for energy sector borrowers prior to August 2014. The results, displayed in Table A4, also corroborate that there are no statistically significant differences in the loan terms and trends of banks with varying degrees of exposure to the energy sector in the months prior to the shock. Finally, we test for the existence of nonlinear pre-trends across banks with different exposures to the energy sector prior to the shock. The specification, presented in equation 4, restricts the loan-level data to loans from firms in the energy sector. yf,b,q = α + ∑βmMonthm*ExposureEnergyb,Aug14 + γf,b + γm + εf,b,q (4) 12 The dependent variable consists of the value loaned to firm f by bank b in month m. The covariates of interest are monthly dummies interacted with the bank’s exposure to the energy sector in August 2014. The βm coefficients thus measure the monthly variation in the value of credit to energy firms across banks with varying exposures in August of 2014. We include fixed effects at the firm-bank and quarter levels. The βm coefficients, plotted in Figure 3, give further credibility to our identification strategy. Prior to the energy price drop, banks with varying exposures to the energy sector had the same dynamics on the value of loans to energy firms. However, once the energy prices dropped, the amount of credit to energy firms began increasing significantly as the banks’ exposure to the energy sector rose. 4. Results We start this section by assessing the impact that the collapse of global energy prices had on the balance sheets of banks with varying degrees of exposure to the energy sector around the time of the shock. We then present our loan-level results, which separately analyze the bank lending dynamics of firms in energy-related and all other sectors after the shock. Finally, we summarize the real effects that increased bank exposure to the energy sector had on the economy as a result of the price shock. 4.1. Impact of the Energy Price Shock on Banks’ Balance Sheets and Financials Table 2 summarizes the results of equation 1 for five bank-month variables: exposure to the energy sector (percent), total lending (in logs), CDS spreads (in logs), stock price (in logs), delinquency rates (percent). We find that the sharp drop in energy prices had a substantially greater effect on banks with higher ex-ante exposure to the energy sector. Compared with banks with less exposure, banks with higher ex-ante exposure to the energy sector increased their lending to the affected sector relatively more after the global price of energy plummeted (column 1). An increase of one standard deviation in exposure to the energy sector in August 2014 leads to an increase of around 1.2 percentage points in the following quarters. As column 2 shows, this increase in lending to the energy sector did not come from an increase in the overall lending of more exposed banks, which suggests that banks reallocated their lending away from other sectors and to energy firms. Columns 3 and 4 indicate that the drop in energy prices increased the risk of banks while reducing their 13 stock prices. An increase of one standard deviation in exposure to the energy sector increases CDS spreads by 10.4 percent (column 3) and reduces stock prices by 3.2 percent (column 4) after the shock. One reason why banks with higher exposure to the energy sector were more affected by the shock is that their borrowers were in distress. The results in column 5 confirm this, as the delinquency rate after the shock increased substantially, given the ex-ante bank exposure. An increase of one standard deviation in exposure to the energy sector leads to an increase in delinquencies in the portfolios of banks by about 0.16 percentage point (roughly 8 percent). 4.2. Impact on Credit to Energy Borrowers The results suggest that banks with higher ex-ante exposure to the energy sector expanded their lending to the energy sector after the collapse of energy prices. In this section, we use loan- level data on the universe of loans to energy sector borrowers to document how the credit terms of firms in the energy sector changed in response to the energy price shock. Table 3, panel A, summarizes the results of our benchmark equation 2 on three credit outcomes: (i) total lending, (ii) lending for working capital, and (iii) lending for investment projects. All the regressions include Bank*Firm and Month fixed effects, and the regressions displayed in columns 2, 4, and 6 further include fixed effects at the Firm*Month level. The inclusion of the latter limits our sample to firms that borrowed from more than one bank at a given point in time. However, this helps us isolate time-varying changes in the demand for credit of borrowers in the energy sector. This is important, as the decline in energy prices directly impacted producers’ revenues, forcing them to demand more external funds. Columns 1 and 2 corroborate the earlier finding that banks with higher ex-ante exposure to the energy sector channeled more credit to the sector. Once we control for time-varying changes in the demand for credit, we find that banks that were more exposed ex-ante injected more credit in energy borrowers. An increase of one standard deviation in ex-ante exposure to the energy sector leads to an increase in the value of loans to firms in the energy sector of around 18 percent. Columns 3 to 6 show that the increase in credit was mainly for working capital, reflecting that distressed energy firms financed their working capital needs rather than starting new investment projects. An increase of one standard deviation in ex-ante exposure to the energy sector leads to an increase in lending for working capital of almost 85 percent, but it has no impact on lending for investment. 14 Our evidence suggests that, compared with less exposed banks, banks that were more exposed to the energy sector had a greater increase in lending to energy firms. To understand whether the increased lending was driven by the supply of credit, rather than expansion in the demand for credit, we analyze the credit terms offered. The results on the interest rates, collateral, and maturity of the loans obtained by energy sector borrowers are displayed in Table 3, panel B. Columns 1 and 2 suggest that, compared with less exposed banks, banks with higher ex-ante exposure to the energy sector relaxed the interest rates on loans to the affected firms significantly more. For example, an increase of one standard deviation in ex-ante exposure leads to a 0.7 percentage point decrease in lending rates (roughly 7.5 percent). Furthermore, as columns 3 to 6 indicate, we find no evidence that the collateral requirements or maturity of the loans to energy firms changed differentially as ex-ante bank exposure to the sector varied. These results suggest that the increase in lending to firms in the energy sector by highly exposed banks was driven in large part by an expansion in supply. Finally, we explore the existence of heterogeneity across borrowers in the energy sector with different outstanding loan amounts. We test whether the banks’ response depended on their relative bargaining power over individual borrowers (Rajan 1992; Santos and Winton 2019). Figure A2 in the appendix presents a simple bin scatter plot (to preserve the anonymity of the borrowers) with censored tails. The results suggest that the increase in credit supply in the energy sector was mainly targeted toward borrowers with larger outstanding loan amounts. Table A5 in the appendix presents the results of a series of regressions where we run equation 2 for two samples of borrowers, depending on whether their outstanding loan amounts in August of 2014 were below (Small) or above (Large) the median. The results suggest that banks expanded their credit to borrowers with ex-ante larger credit amounts relatively more, especially credit for working capital. This finding is consistent with borrowers holding up their lenders, given the borrowers’ higher bargaining position. 4.3. Spillovers to Non-Energy Borrowers Our earlier results at the bank-month level show that although banks with higher ex-ante exposure increased their lending to the energy sector, they did so without increasing their total lending. Therefore, the increase in credit toward energy firms should have affected the access to 15 credit of firms in other sectors. In this section, we restrict the sample to borrowers in non-energy- related sectors, to analyze how and which non-energy borrowers were affected by this reallocation. Table 4, panel A, presents the results of equation 2 for the sample of borrowers in non- energy sectors. The three credit outcomes displayed are the log of total bank lending as well as the log of bank lending for working capital and investment projects. We include Bank*Firm and Month fixed effects in all the regressions, and Firm*Month fixed effects in the regressions displayed in columns 2, 4, and 6, to fully control for time-varying changes in the demand for credit. In the table, the first two columns show that, as a result of the collapse of oil prices, banks that were more exposed to the energy sector had a greater reduction in the amount of credit to firms in sectors that were not directly affected by the shock. An increase of one standard deviation in a bank’s ex-ante exposure to the energy sector leads to a reduction in the loan volume to firms in other sectors of around 13 percent. We decompose this result to understand which type of loans— for working capital or investment—contracted the most. The results are displayed in columns 3 to 6. Although loans for working capital contracted on average by 8.5 percent, loans for the investment sector contracted by a full 30 percent. These results suggest that most of the contraction of bank credit was driven by a reduction in loans for investment, which are typically associated with increases in firm productivity. Overall, our evidence suggests that banks with greater ex-ante exposure to the energy sector had greater contractions in their credit to non–energy sector borrowers. This contraction in credit was concentrated in financing for investment projects. Next, we investigate whether there was heterogeneity in the contraction of lending across borrowers. We check whether the impact was higher among smaller firms, which tend to be considered riskier (Morais et al. 2019). We divide non-energy borrowers into two groups, those with more or fewer than 50 employees in 2014 (following Beck and Demirguc-Kunt (2006)), and run equation 2 on each sample. The results, which are summarized in Table 4, panel B, suggest that the contraction of credit almost exclusively affected smaller firms in non-energy sectors. For this subsample of borrowers, an increase of one standard deviation in ex-ante bank exposure to the energy sector leads to a contraction in the volume of lending of around 16.4 percent, whereas for larger firms the impact on total lending volume is statistically indistinguishable from 0. 4.4. Real Effects 16 The results suggest that the energy price collapse impacted the credit allocation of banks that were more exposed to energy-related sectors. Banks that were more exposed increased lending to firms in the affected sectors and contracted credit to firms in other sectors. If borrowers were not able to switch credit suppliers, the contraction in bank lending might have had a material impact on their real outcomes. Using firm-year-level data for the sample of firms in non-energy sectors, we run the following specification: yf,y = α + βAvgExposureEnergym,Aug14*Posty + γf + γy + εf,y (5) where the real outcome yf,y of firm f in year y corresponds to one of the following variables: total lending, loans for working capital, loans for investment projects, total liabilities, assets, and revenue. 13 AvgExposureEnergym,Aug14 is a measure of exposure to the energy sector in August 2014 of a firm headquartered in municipality m. Posty is an indicator that the yearly observation is after 2014. β is the coefficient of interest, as it measures the extent to which the real outcomes of firms in municipalities with more banks with greater ex-ante exposure were affected by the drop in energy prices. Finally, γf and γy are fixed effects at the firm and year levels, respectively, and εf,y is the error term clustered at the municipality level. The results of this exercise are displayed in Table 5. Starting with information from the credit registry of bank loans, we find that an increase of one standard deviation in the exposure of banks with which a firm has relations reduces total lending by 2.1 percent. Again, the impact is much larger for financing for investment projects. Total loans for working capital contract by 1.9 percent, and loans for investment contract by 12.4 percent. These results suggest that non-energy firms are unable to smooth the shock that their banks receive. We also find evidence that other firm outcomes (liabilities, assets, and sales) were negatively impacted by the municipality’s exposure to the energy sector. An increase of one standard deviation in a municipality’s exposure to the energy sector reduces total firm liabilities by around 2.9 percent and total firm assets by around 2.6 percent. However, we do not find any impact on total sales. Overall, we uncover evidence suggesting that non-energy firms experienced a larger contraction in their liabilities Orbis information tends to refer to the month of December. For the credit registry outcomes (total loans, loans for 13 working capital, and loans for investment) for each firm-year pair, we selected the December value. 17 (particularly investment) and assets if they were headquartered in municipalities with high exposure to banks that were more impacted by the decline in energy prices. In addition to these firm-level results, we analyze the impact of the collapse of energy prices at the more aggregated state level. We run a similar specification using state-quarter-level data. In this exercise, we relate quarterly state GDP to the average ex-ante exposure to the energy sector of banks operating in a state. We use the following specification: ys,q = α + βAvgExposureEnergys,Aug14*Postq + γs + γq + εf,y (6) where ys,q is the total GDP of state s in quarter q. Furthermore, we study the decomposition of the GDP in the three sectors: primary, secondary, and tertiary. The regressor— AvgExposureEnergys,Aug14—is the average ex-ante exposure to the energy sector of banks operating in state s, weighted by loan value. Finally, we include state γs and quarter γq fixed effects to control for state-specific, time-unvarying variation as well as aggregate time variation affecting all states, and errors are clustered at the state level. Our coefficient of interest is β, which indicates whether the aggregate production of a given state was differentially affected by the drop in energy prices as the average ex-ante exposure of its banks to the energy sector increased. To isolate the impact of the contraction in bank lending from the drop in energy prices, we present the results for all 32 states in Mexico and the 30 states in Mexico that do not produce energy. 14 Table 6 presents the findings. Focusing on the non-energy-producing states, an increase of one standard deviation in the exposure of a state to the energy sector reduces the state’s GDP by 1.8 percentage points. We interpret this finding as evidence that the reduction in output was caused by the contraction in lending of banks that were highly exposed ex-ante. Furthermore, as the results in the table show, the brunt of the impact was on the GDP of the secondary sector. Relative to the tertiary sector, the secondary sector tends to be more capital intensive and dependent on external financing (Buera, Kaboski, and Shin 2011). An increase of one standard deviation in a state’s ex- ante exposure to the energy sector reduces the state’s GDP from the secondary sector by around 14 Tabasco and Veracruz are the main oil producing states in Mexico. In these states, oil extraction and production represent roughly 40 percent of state-level GDP. For the remaining five producers—Chiapas, Tamaulipas, Puebla, San Luis Potosi, and Hidalgo—energy production is residual and represents less than 2 percent of state GDP. 18 3.9 percent, whereas the GDP from the tertiary sector is not impacted in a statistically significant way. 5. Conclusions We analyzed the credit supply of banks in the event of large exposures to financially stressed borrowers. We studied the impact of the halving of energy prices in late 2014 on the banking sector in Mexico, a large energy producer. As energy prices declined, the CDS spreads of energy producers ramped up, as their working capital and financial needs outpaced their expected revenues. Using the universe of corporate loans to energy and non-energy firms, we found that banks that were more exposed to the energy sector prior to the shock notably increased their exposure to the sector ex post—by offering loans of higher volume and reducing interest rates on those loans. This behavior suggests an attempt on behalf of largely exposed banks to avoid realizing losses on an important part of their loan portfolios, even at the cost of jeopardizing their regulatory liquidity and capitalization ratios. Controlling for demand shocks, we found that banks that were more exposed to the energy sector contracted their credit to firms in non-energy sectors, with important negative real effects. The relation between large and concentrated credit exposures of banks and commodity prices has not been closely studied in the literature. Our findings are particularly relevant for commodity-producing economies that are exposed to global fluctuations in commodity prices. The channel that we identify outlines the need to account for proper risk management of banks with large concentrations in their credit portfolios that are subject to price volatility. 19 References Acharya, V.V., I. Hasan, and A. Saunders. 2006. “Should Banks Be Diversified? Evidence from Individual Bank Loan Portfolios.” Journal of Business 79 (3): 1355–1412. Acharya, V. V., and S. Steffen. 2015. “The Greatest Carry Trade Ever? Understanding Eurozone Bank Risks.” Journal of Financial Economics 115: 215–36. Agarwal, I., R. Duttagupta, and A. F. Presbitero. 2019. “Commodity Prices and Bank Lending.” Economic Inquiry 58 (2): 953–79. Alesina, A., F. R. Campante, and G. Tabellini. 2008. “Why Is Fiscal Policy Often Procyclical?” Journal of the European Economic Association 6 (5): 1006–36. Armstrong, R, S. Morris, R. Smith. 2020. “Bank Investors Confront a New Fear: Oil Company Defaults” Financial Times, March 9, 2020 Basel Committee on Banking Supervision. 2014. “Supervisory Framework for Measuring and Controlling Large Exposures.” Basel Committee on Banking Supervision, Basel, Switzerland. Beck, T., O. De Jonghe, and K. Mulier. 2017. “Bank Sectoral Concentration and (Systemic) Risk: Evidence from a Worldwide Sample of Banks.” CEPR Discussion Paper 12009, Center for Economic and Policy Research, Washington, DC. Beck, T., and A. Demirguc-Kunt. 2006. “Small and Medium-Size Enterprises: Access to Finance as a Growth Constraint.” Journal of Banking & Finance 30: 2931–43. Berger, A. N., I. Hasan, and M. Zhou. 2010. “The Effects of Focus versus Diversification on Bank Performance: Evidence from Chinese Banks.” Journal of Banking & Finance 34 (7): 1417–35. Bidder, R. M., J. R. Krainer, and A. H. Shapiro. 2017. “De-Leveraging or De-Risking? How Banks Cope with Loss.” Working Paper 2017-03, Federal Reserve Board of San Francisco. Blanchard, O., and J. Gali. 2010. “Labor Markets and Monetary Policy: A New Keynesian Model with Unemployment.” American Economic Journal: Macroeconomics 2 (2): 1–30. Boyd, J. H., and E. C. Prescott. 1986. “Financial Intermediary-Coalitions.” Journal of Economic Theory 38 (2): 211–32. 20 Bruckner, M., and A. Ciccone. 2010. “International Commodity Prices, Growth and the Outbreak of Civil War in Sub‐Saharan Africa.” Economic Journal 120 (544): 519–34. Brunnermeier, M. K. 2009. “Deciphering the Liquidity and Credit Crunch 2007-2008.” Journal of Economic Perspectives 23 (1): 77–100. Buera, F., J. Kaboski, and Y. Shin. 2011. “Finance and Development: A Tale of Two Sectors.” American Economic Review 101 (5): 1964–2002. Caballero, R. J., T. Hoshi, and A. K. Kashyap. 2008. “Zombie Lending and Depressed Restructuring in Japan.” American Economic Review 98 (5): 1943–77. De Jonghe, O., H. Dewachter, K. Mulier, S. Ongena, and G. Schepens. 2019. “Some Borrowers Are More Equal Than Others: Bank Funding Shocks and Credit Reallocation.” Review of Finance 24 (1): 1–43. Deaton, A. 1999. “Commodity Prices and Growth in Africa.” Journal of Economic Perspectives 13 (3): 23–40. Deaton, A., and R. Miller. 1996. “International Commodity Prices, Macroeconomic Performance, and Politics in Sub-Saharan Africa.” Journal of African Economies 5 (3): 99–191. Degryse, H., and S. Ongena. 2005. “Distance, Lending Relationships, and Competition.” The Journal of Finance 60 (1): 231-266. Dehn, J. 2000. “Commodity Price Uncertainty and Shocks: Implications for Economic Growth.” CSAE Working Paper Series 2000-10, Center for the Study of African Economies, University of Oxford. Diamond, D. W. 1984. “Financial Intermediation and Delegated Monitoring.” Review of Economic Studies 51 (3): 393–414. FSI (Financial Stability Institute). 2019. “Pillar 3 Framework – Executive Summary.” FSI, Basel, Switzerland, BIS.org. Gilje, E. P., E. Loutskina, and P. E. Strahan. 2016. “Exporting Liquidity: Branch Banking and Financial Integration.” Journal of Finance 71(3): 1159–84. 21 Goetz, M. R., L. Laeven, and R. Levine. 2016. “Does the Geographic Expansion of Banks Reduce Risk?” Journal of Financial Economics 120 (2): 346–62. Ippolito, F., J.-L. Peydro, A. Polo, and E. Sette. 2016. “Double Bank Runs and Liquidity Risk Management.” Journal of Financial Economics 122 (1): 135–54. Iyer, R., and J.-L. Peydro. 2011. “Interbank Contagion at Work: Evidence from a Natural Experiment.” Review of Financial Studies 24 (4): 1337–77. Kashyap, A. K., and J. C. Stein. 2000. “What Do a Million Observations on Banks Say about the Transmission of Monetary Policy?” American Economic Review 90 (3): 407–28. Khwaja, A. I., and A. Mian. 2008. “Tracing the Impact of Bank Liquidity Shocks: Evidence from an Emerging Market.” American Economic Review 98 (4): 1413–42. Kinda, T., M. Mlachila, and R. Ouedraogo. 2016. “Commodity Price Shocks and Financial Sector Fragility.” IMF WP/16/12, International Monetary Fund, Washington, DC. La Porta, R., Lopez-de-Silanes, F. and G. Zamarripa. 2003. “Related Lending.” Quarterly Journal of Economics 118 (1): 231-268. Laeven, L., and R. Levine. 2007. “Is There a Diversification Discount in Financial Conglomerates?” Journal of Financial Economics 85 (2): 331–67. Morais, B., J.-L. Peydro, J. Roldan, and C. Ruiz. 2019. “The International Bank Lending Channel of Monetary Policy Rates and QE: Credit Supply, Reach‐for‐Yield, and Real Effects.” Journal of Finance 74 (1): 55–90. Paravisini, D., P. Schnabl, and V. Rappoport. 2017. “Specialization in Bank Lending: Evidence from Exporting Firms.” CEPR Discussion Papers 12156, Center for Economic and Policy Research, Washington, DC. Peek, J., and E. S. Rosengren. 2005. “Unnatural Selection: Perverse Incentives and the Misallocation of Credit in Japan.” American Economic Review 95 (4): 1144–66. Rajan, R, G. 1992. “Insiders and Outsiders: The Choice between Informed and Arm’s-Length Debt.” Journal of Finance 47 (4): 1367–1400. 22 Santos, J. A. C., and A. Winton. 2019. “Bank Capital, Borrower Power, and Loan Rates.” Review of Financial Studies 32 (11): 4501–41. Schnabl, P. 2012. “The International Transmission of Bank Liquidity Shocks: Evidence from an Emerging Market.” Journal of Finance 67 (3): 897–932. Tabak, B. M., D. M. Fazio, and D. O. Cajueiro. 2011. “The Effects of Loan Portfolio Concentration on Brazilian Banks’ Return and Risk.” Journal of Banking & Finance 35 (11): 3065–76. Westernhagen, N., E. Harada, T. Nagata, B. Vale, J. Ayuso, J. Saurina, S. Daltung, S. Ziegler, E. Kent, J. Reidhill, and S. Peristiani. 2004. “Bank Failures in Mature Economies.” Working Paper 13, Basel Committee on Banking Supervision, Basel, Switzerland. Winton, A. 1999. “Don’t Put All Your Eggs in One Basket? Diversification and Specialization in Lending.” Working Paper 16, Wharton School Center for Financial Institutions, University of Pennsylvania, Philadelphia, PA. World Bank. 2018. Global Economic Prospects. World Bank, Washington, DC, WorldBank.org. 23 Figure 1 – Oil Prices and CDS Spreads of Energy Producers This figure displays the movements in oil prices—West Texas Intermediate—in dollars as well as the movements in the CDS spreads of energy firms in Mexico and in other emerging economies. The sample period spans from January 2013 to June 2016. 400 100 CDS Spreads - Percent 80 300 Oil Prices 60 200 40 100 20 2013m1 2014m1 2015m1 2016m1 Time Oil Price CDS - Mexico CDS - Emerging 24 Figure 2 – Bank Exposure to the Energy Sector and Financial Variables This figure displays the evolution of exposure to the energy sector of Mexican banks as well as their stock prices and CDS spreads. We split the sample into two groups with below and above median exposure to the energy sector, defined as the value of total loans outstanding to the energy sector over total capital in August 2014. The series of CDS spreads of five-year bonds and stock prices are normalized to August 2014. The sample period is from January 2013 to June 2016. Bank Exposure to Energy and CDS Spreads Bank Exposure to Energy and Stock Prices Bank Exposure to Energy Sector 10 100 40 5 30 50 Percentage Percentage 0 Percentage 20 -5 0 10 -10 -15 -50 0 2013m1 2014m1 2015m1 2016m1 Time 2013m1 2014m1 2015m1 2016m1 2013m1 2014m1 2015m1 2016m1 Time Time Low Exposure High Exposure Low Bank Exposure High Bank Exposure Low Exposure High Exposure 25 Figure 3 – Evolution of Bank Exposure to the Energy Sector This figure displays quarterly coefficients of a bank-month regression where the dependent variable is the share of loans to the energy sector by bank b in month m. The coefficients displayed are the interaction of the bank’s ex-ante exposure to the energy sector, defined as the value of total loans outstanding to the energy sector over total capital in August 2014, and month dummies. The coefficients represent the relative changes in banks’ exposure to the energy sector, given their exposure in August 2014. The regression includes bank and month fixed effects. Standard errors are double clustered at the bank and month levels. Vertical bars represent the confidence intervals of the coefficients at 90 percent. The sample period is from January 2013 to June 2016. 1 .5 0 -.5 Sep13 Dec13 Mar14 Jun14 Sep14 Dec14 Mar15 Jun15 Months 26 Table 1. Summary Statistics This table reports the summary statistics of our sample for January 2013 to June 2016. All variable definitions are provided in Table A1. # Obs Average p10 Median p90 Std dev Panel A. Bank-month-level variables ExposureEnergyb,m (%) 897 9.9 0 5.9 25.4 12.1 Tier 1 Capital Ratiob,m (%) 897 15.4 12.5 15.3 18.5 2.2 Total Lendingb,m (logs) 897 21.8 19.6 21.3 24.2 1.8 Delinquencyb,m (%) 897 2.4 0.4 2 4.7 2 CDSb,m (basis points) 367 410 321 425 492 71 Stock Priceb,m (index) 470 3.4 0.7 3.9 5.2 1.6 ExposureEnergyb,Aug14 (%) 897 8.0 0 5.0 19.6 6.5 Panel B. Loan-level variables of firms in energy-related sectors Total Lendingf,b,m (‘000) 34,741 89,790 96 1,560 126,930 307,605 Loans to working capitalf,b,m (‘000) 34,741 78,848 25 1,218 96,521 285,116 Loans to investment,b,m (‘000) 34,741 2,980 0 0 131 13,470 Interest Ratef,b,m (%) 31,257 11.4 4.4 11.8 18.0 10.7 Maturityf,b,m (months) 31,257 25 2.4 20.4 52.0 25.7 Collateralf,b,m (%) 31,257 14.6 0.0 0.0 54.7 29.1 Panel C. Loan-level variables of firms in non-energy sectors Total Lendingf,b,m (‘000) 1,684,329 6,225 48 511 5,179 58,224 - Working capitalf,b,m (‘000) 1,600,896 5,509 46 500 4,511 54,745 - Investmentf,b,m (‘000) 145,324 11,782 53 1,159 20,000 71,286 Interest Ratef,b,m (%) 1,684,329 13.4 7.8 13.0 19.0 4.2 Maturityf,b,m (years) 1,668,951 33 3.0 19.3 43.1 111.8 Collateralf,b,m (%) 1,684,329 13.2 0.0 0.0 50.0 27.2 Panel D. Firm-year-level variables Total Lendingf,y (‘000) 66,592 10,337 69 732 7,962 117,140 - Working Capitalf,y (‘000) 64,561 8,902 67 696 6,894 105,158 - Investmentf,y (‘000) 7,009 15,757 57 1,511 26,872 93,249 AvgExposureEnergyf,Aug14 (%) 66,592 13.4 3.2 14.5 19.6 4.8 Liabilitiesf,y (millions) 2,132 6,466 42 344 20,947 15,655 Assetsf,y (millions) 2,350 19,607 101 1,146 50,739 47,695 Revenuesf,y (millions) 2,350 7,056 179 818 23,431 15,245 AvgExposureEnergym,Aug14 (%) 2,350 9.8 7.7 10.3 11.5 2.1 Panel E. State-quarter-level variables Total GDPs,q (index) 512 4.6 4.6 4.6 4.7 0.1 AvgExposureEnergys,Aug14 (%) 512 12.1 9.6 12.1 14.3 1.6 27 Table 2. Evolution of Bank-Level Indicators after the Shock Regressions at the bank*month level using bank balance sheet data. Dependent variables are listed in the columns. ExposureEnergyb,m represents lending to the energy sector as a share of its tier 1 capital of bank b in month m. Total Lendingb,m are total monthly loans of bank b in logs. CDS Spreadsb,m is the log of CDS spreads of five-year maturity bonds of bank b in month m. Stock Priceb,m is the log of the stock price of bank b in month m. Delinquencyb,m is the share of delinquent loans of bank b in month m. The regressor ExposureEnergyb,Aug14 represents the exposure to the energy sector of bank b in August 2014. Postm is an indicator for month m after the energy price shock in August 2014. All regressions include bank and month fixed effects. The results show that there was relocation of lending toward the energy sector by banks that were more exposed to it. However, other margins were unchanged, suggesting that there was a reallocation across sectors. Robust standard errors are double clustered at the bank and month levels. Detailed variable definitions are provided in Table A1. Observations are at the bank-month level for January 2013 to June 2016. *** p<0.01, ** p<0.05, * p<0.1. ExposureEnergyb,m Total Lendingb,m CDS Spreadsb,m Stock Priceb,m Delinquencyb,m (1) (2) (3) (4) (5) ExposureEnergyb,Aug14*Postm 0.192*** -0.003 0.016* -0.005*** 0.024*** (0.022) (0.002) (0.009) (0.002) (0.007) Observations 612 612 272 350 612 R-squared 0.884 0.992 0.706 0.996 0.896 Bank FE Yes Yes Yes Yes Yes Time FE Yes Yes Yes Yes Yes SD(ExposureEnergyb,Aug14) 6.5 6.5 6.3 6.2 6.5 28 Table 3. Panel A - Lending Volumes to the Energy Sector This panel displays the impact of bank exposure to the energy sector and lending to borrowers in the energy sector. The dependent variables are in logs. Total Lendingf,b,m is the total lending value to firm f by bank b in month m. Working Capitalf,b,m and Investmentf,b,m are total lending value destined to working capital and investment, respectively. ExposureEnergyb,Aug14 represents lending to the energy sector as a share of its tier 1 capital of bank b in August 2014. Postm is an indicator that the month m is after the energy price shock in August 2014. Robust standard errors are double clustered at the bank and month levels. Detailed variable definitions are provided in Table A1. Observations are at the firm-bank-month level for January 2013 to June 2016. *** p<0.01, ** p<0.05, * p<0.1. Total Lendingf,b,m Working Capitalf,b,m Investmentf,b,m (1) (2) (3) (4) (5) (6) ExposureEnergyb,Aug14*Postm 0.03** 0.09*** 0.14*** 0.32*** -0.05 -0.05 (0.01) (0.02) (0.05) (0.09) (0.04) (0.04) Observations 34,998 16,898 34,998 16,898 34,998 16,898 R-squared 0.88 0.94 0.87 0.92 0.87 0.92 Bank-firm FE Yes Yes Yes Yes Yes Yes Month FE Yes - Yes - Yes - Firm-month FE No Yes No Yes No Yes SD(ExposureEnergyb,Aug14) 6.1 6.1 6.1 6.1 6.1 6.1 29 Table 3. Panel B - Terms on Loans to the Energy Sector This panel displays the impact of bank exposure to the energy sector and its lending to energy borrowers. The dependent variables are Interest Ratef,b,m, which is the total interest rate charged to firm f by bank b in month m; Collateralf,b,m , which is the fraction of loans that is guaranteed; and Maturityf,b,m, which is the average length in log months of loan duration. ExposureEnergyb,Aug14 represents lending to the energy sector as a share of its tier 1 capital of bank b in August 2014. Postm is an indicator that the month m is after the energy price shock in August 2014. Standard errors double clustered at the bank and month levels. Detailed variable definitions are provided in Table A1. Observations are at the firm-bank-month level for January 2013 to June 2016. *** p<0.01, ** p<0.05, * p<0.1. Interest Ratef,b,m Collateralf,b,m Maturityf,b,m (1) (2) (3) (4) (5) (6) ExposureEnergyb,Aug14*Postm -0.13** -0.11*** 0.38 0.51 -0.01 0.03 (0.05) (0.04) (0.77) (0.55) (0.02) (0.02) Observations 32,358 16,698 32,358 16,698 32,358 16,698 R-squared 0.29 0.75 0.81 0.89 0.67 0.83 Bank-firm FE Yes Yes Yes Yes Yes Yes Month FE Yes - Yes - Yes - Firm-month FE No Yes No Yes No Yes SD(ExposureEnergyb,Aug14) 6.1 6.1 6.1 6.1 6.1 6.1 30 Table 4. Panel A - Lending Volumes to Borrowers in Non-Energy Sectors This panel presents the coefficients of the regression in equation 2, testing the impact of the price shock on loan value to non-energy borrowers. The dependent variables are in logs. Total Lendingf,b,m is the total lending value to firm f by bank b in month m. Working Capitalf,b,m and Investmentf,b,m are total lending value destined to working capital and investment, respectively. ExposureEnergyb,Aug14 represents lending to the energy sector as a share of its tier 1 capital of bank b in August 2014. Postm is an indicator that the month m is after the energy price shock in August 2014. The results indicate that banks that were more exposed to the energy sector reduced relatively more their lending. Standard errors are double clustered at the bank and month levels. Detailed variable definitions are provided in Table A1. The sample period is from January 2013 to June 2016. *** p<0.01, ** p<0.05, * p<0.1. Total Lendingf,b,m Working Capitalf,b,m Investmentf,b,m (1) (2) (3) (4) (5) (6) ExposureEnergyb,Aug14*Postm -0.022*** -0.017*** -0.014* 0.001 -0.050*** -0.067*** (0.007) (0.005) (0.008) (0.007) (0.007) (0.011) Observations 1,262,712 573,544 1,262,712 573,544 1,262,712 573,544 R-squared 0.794 0.897 0.824 0.899 0.873 0.921 Bank*firm FE Yes Yes Yes Yes Yes Yes Month FE Yes - Yes - Yes - Firm-month FE No Yes No Yes No Yes SD(ExposureEnergyb,Aug14) 6.1 6.1 6.1 6.1 6.1 6.1 31 Table 4. Panel B - Lending to Non-Energy Sectors, by Borrower Size This panel presents the coefficients of the regression in equation 2, testing the impact of the price shock on loan value to non-energy borrowers. The dependent variables, all in logs, are loan value, value to working capital, and value to investment to firm f in month m by bank b. Postm is an indicator that month m is after the energy price shock in September 2014. The results indicate that banks that were more exposed to the energy sector reduced relatively more their lending. Standard errors are double clustered at the bank and month levels. Detailed variable definitions are provided in Table A1. The sample period is from January 2013 to June 2016. *** p<0.01, ** p<0.05, * p<0.1. Total Lendingf,b,m Working Capitalf,b,m Investmentf,b,m (1) (2) (3) (4) (5) (6) ExposureEnergyb,Aug14*Postm -0.027*** -0.005 -0.020** 0.006 -0.047*** -0.062*** (0.007) (0.005) (0.008) (0.010) (0.006) (0.014) Observations 1,026,135 236,519 1,026,135 236,519 1,026,135 236,519 R-squared 0.766 0.847 0.816 0.840 0.872 0.871 Borrower size Small Large Small Large Small Large Bank*firm FE Yes Yes Yes Yes Yes Yes Month FE Yes Yes Yes Yes Yes Yes SD(ExposureEnergyb,Aug14) 6.1 6.1 6.1 6.1 6.1 6.1 32 Table 5 Real Effects – Impact on Firm Outcomes Associated with Banks’ Exposure to the Energy Sector This table reports the real effects associated with the reduction of lending to borrowers in the non-energy sector, given their bank’s exposure to the energy sector. Observations are at the firm-year level. All observations are in logs of thousands of pesos. Bank Liabilities,y is the value of bank loans of firm f in year y. Working Capitalf,y and Investmentf,y are the value of the bank loans of firm f in year y to working capital and investment, respectively. Total Liabilitiesf,y is the value of a firm’s total liabilities of firm f in year y. Assetsf,y is the value of firm f’s total assets in year y. Revenuef,y is the value of firm f’s sales in year y. Posty is an indicator variable that equals 1 after 2014. AvgExposureEnergym,Aug14 is the average exposure to the energy sector in August 2014, weighted by loan value in the municipality, of the banks operating in municipality m in which firm f resides. It proxies the impact at the municipality level of the decline in energy prices through banks that operate in it. Standard errors are clustered at the firm level. Detailed variable definitions are provided in Table A1. The sample period is from 2013 to 2016. *** p<0.01, ** p<0.05, * p<0.1. Total Lendingf,y Working Capitalf,y Investmentf,y Total Liabilitiesf,y Assetsf,y Revenuef,y (1) (2) (3) (4) (5) (6) AvgExposureEnergym,Aug14*Posty -1.00** -0.88* -5.90*** -1.37** -1.24*** -0.55 (0.48) (0.50) (1.56) (0.60) (0.47) (0.96) Observations 122,157 118,581 12,022 1,115 1,239 1,236 R-squared 0.85 0.84 0.86 0.99 1.00 0.98 Firm FE Yes Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Yes Sector-year FE No No No No No No SD( AvgExposureEnergym,Aug14) 2.1 2.1 2.1 2.1 2.1 2.1 . 33 Table 6 - Impact on States’ Output Associated with Banks’ Exposure to the Energy Sector This table reports the results of a regression testing whether activity is impacted by the degree of exposure that states had to banks that were lending to the energy sector. Observations are at the state-quarter level. GDPs,q is the log index—relative to January 2004—of state s’s GDP in quarter q. We further split this indicator by type of sector (primary, secondary, and tertiary). Primary sector includes mining and agriculture, Secondary includes manufacturing and construction, and Tertiary includes services. AvgExposureEnergys,Aug14 is the average exposure to the energy sector in August 2014, weighted by loan value in the state, of the banks operating in state s. It proxies for the impact at the state level of the decline in energy prices through banks that operate in it. Columns indicating Non- Energy refer to states that are non-producers of energy. Standard errors are clustered at the state level. Detailed variable definitions are provided in Table A1. The sample period is from the first quarter of 2013 to the fourth quarter of 2016. *** p<0.01, ** p<0.05, * p<0.1. GDPs,q GDP Primarys,q GDP Secondarys,q GDP Tertiarys,q (1) (2) (3) (4) (5) (6) (7) (8) AvgExposureEnergys,Aug14*Postq -0.67* -1.14*** -3.67* -2.94 -1.92** -2.44*** -0.17 -0.30 (0.42) (0.36) (2.19) (2.32) (0.85) (0.85) (0.28) (0.27) Observations 512 480 512 480 512 480 512 480 R-squared 0.69 0.77 0.31 0.32 0.51 0.53 0.82 0.86 State FE Yes Yes Yes Yes Yes Yes Yes Yes Quarter FE Yes Yes Yes Yes Yes Yes Yes Yes States All Non-Energy All Non-Energy All Non-Energy All Non-Energy SD(AvgExposureEnergys,Aug14) 1.6 1.6 1.6 1.6 1.6 1.6 1.6 1.6 34 Appendix. Additional Figures and Tables Figure A1 – Energy Prices and Firm Leverage This figure displays the bank-level share of lending to the energy sector in Mexico from 2013Q1 to 2016Q2. 120 56 100 54 Leverage - Percent Oil Prices 52 80 50 60 48 40 2013q1 2014q1 2015q1 2016q1 Time Oil Prices Leverage 35 Figure A2 –Loan Size Pre-Shock and Loan Size Growth This figure displays a bin scatter plot for the sample of firms in the energy sector. The figure relates the loan volume of firms in the month prior to the price shock to ex post loan volume growth. Therefore, loans on the x-axis are grouped by size. Log Total Lending – Pre-Shock is the log value of the total loans outstanding in August 2014. We have censored the energy sample of all firm-bank loans above the 90th percentile. Log Loan Growth is the average growth rate of loan volume (in percent) from August 2014 to August 2016. 50 Log Loan Growth - Percent -50 -100 0 12 14 16 18 20 22 Log Total Lending - Pre-Shock 36 Table A1. Variable Definitions Bank variables ExposureEnergyb,m Share of lending of bank b to borrowers operating in the energy sector in month m (percent) Tier 1 Capital Ratiob,m Ratio of a bank's core equity capital to its total risk-weighted assets in month m (percent) Total Lendingb,m Total credit portfolio of commercial loans of bank b in month m (logs of millions of Mexican pesos) Delinquencyb,m Average ratio of nonperforming loans of bank b to total loans in month m (percent) CDSb,m Five-year CDS spread of bank b in month m minus sovereign CDS spread of bank b’s country (percent relative to August 2014) Stock Priceb,m Stock price of bank b in month m (percent relative to August 2014) ExposureEnergyb,Aug14 Share of lending of bank b to borrowers operating in the energy sector in August 2014 (percent), weighted by number observations Loan variables Total Lendingf,b,m Value of outstanding loans of firm f with bank b in month m (logs of thousands of Mexican pesos) Working Capitalf,b,m Value of outstanding working capital loans of firm f with bank b in month m (logs of thousands of Mexican pesos) Investmentf,b,m Value of outstanding investment loans of firm f with bank b in month m (logs of thousands of Mexican pesos) Interest Ratef,b,m Average interest rate of loans of firm f with bank b in month m, weighted by loan value (percent) Maturityf,b,m Average maturity of loans of firm f with bank b in month m, weighted by loan value (years) Collateralf,b,m Fraction of loans with guarantees of firm f with bank b in month m, weighted by loan value (percent) ExposureEnergyb,Aug14 Share of lending of bank b to borrowers operating in the energy sector in August 2014, weighted by number observations (percent) Firm variables Total Lendingf,y Total bank loans of firm f in year y (logs of thousands of Mexican pesos) - Working Capital f,y Total value of working capital bank loans of firm f in year y (logs of thousands of Mexican pesos) - Investmentf,y Total value of investment bank loans of firm f in year y (logs of thousands of Mexican pesos) ExposureEnergyf,Aug14 Average exposure of banks to the energy sector in August 2014 of banks serving firm f, weighted by loans in municipality (percent) Liabilitiesf,y Total liabilities of firm f in year y (logs of thousands of Mexican pesos) Assetsf,y Total assets of firm f in year y (logs of thousands of Mexican pesos) Revenuesf,y Total operational revenue of firm f in year y (logs of thousands of Mexican pesos) Small firmf Indicator that firm f had fewer than 50 employees in 2014 ExposureEnergym,Aug14 Average of exposure to the energy sector in August 2014 of banks serving municipality m, weighted by loans in municipality (percent) State variables ExposureEnergyStates,q Share of lending to the energy sector of banks operating in state s in quarter q, weighted by banks’ loan portfolio (percent) Total GDPs,q GDP of state s in quarter q (index relative to 2014) ExposureEnergys,Aug14 Exposure of banks to the energy sector in August 2014 in the state of firm f, weighted by loans in the state (percent) Note: CDS = credit default sway; GDP = gross domestic product. 37 Table A2 – Bank Comparison, Given Exposure to the Energy Sector This table displays the 5-digit NAICS energy-related sectors as well as their descriptions. NAICS = North American Industry Classification System. 5-digit NAICS sector Description 21111 Oil and gas extraction 21211 Coal mining 21311 Support activities for mining and oil and gas extraction 23712 Oil and gas pipeline related structures construction 32411 Petroleum refineries 32419 All other petroleum and coal products manufacturing 32511 Petrochemical manufacturing 48311 Marine oil and natural gas transportation 48611 Transportation of crude oil through pipelines 48621 Transportation of natural gas through pipelines 48691 Pipeline transportation of refined petroleum products 38 Table A3 – Testing for Pre-Trends, Given Exposure to the Energy Sector in August 2014 This table tests for the existence of pre-trends across banks with varying exposures to the energy sector in June 2014 on a series of bank-level characteristics. Observations are at the bank-month level. Dependent variables are listed in the columns. Loansb,m are total monthly loans of bank b in logs. CDS Spreadsb,m is the log of CDS spreads of five-year maturity bonds of bank b in month m. Stock Priceb,m is the log of the stock price of bank b in month m. Delinquencyb,m is the share of delinquent loans of bank b in month m. ExposureEnergyb,m represents lending to the energy sector of bank b as a share of its tier 1 capital in month m. Standard errors are double clustered at the bank and month levels. Detailed variable definitions are provided in Table A1. The sample period is from January 2013 to August 2014. *** p<0.01, ** p<0.05, * p<0.1. ExposureEnergyb,m Total Lendingb,m CDS Spreadsb,m Stock Priceb,m Delinquencyb,m (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) ExposureEnergyb,Aug14 0.333*** 0.001 -0.032 -0.015 -0.027 (0.103) (0.029) (0.032) (0.047) (0.023) ExposureEnergy b,Aug14*Trendm 0.006 0.004 -0.000 -0.000 0.003 0.003 -0.000 -0.000 0.001 0.001 (0.009) (0.003) (0.003) (0.000) (0.002) (0.002) (0.004) (0.000) (0.003) (0.001) Observations 306 306 306 306 124 124 170 170 306 306 R-squared 0.362 0.946 0.163 0.996 0.205 0.817 0.006 0.998 0.046 0.863 Month FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Bank FE No Yes No Yes No Yes No Yes No Yes SD(ExposureEnergyb,Aug14) 6.5 6.5 6.5 6.5 6.5 6.5 6.5 6.5 6.5 6.5 39 Table A4 – Testing for Pre-Trends, Given Exposure to the Energy Sector in August 2014 This table tests for the existence of pre-trends in loans to firms in the energy sector, given the exposure of banks to the energy sector in August 2014. The sample period is from January 2013 to August 2014. The dependent variables are in logs. Total Lendingf,b,m is the total lending value to firm f by bank b in month m. Working Capitalf,b,m and Investmentf,b,m are total lending value destined to working capital and investment, respectively. The sample is at the firm-bank-month level of observation. Standard errors are double clustered at the bank and month levels. *** p<0.01, ** p<0.05, * p<0.1. Total Lendingf,b,m Working Capitalf,b,m Investmentf,b,m (1) (2) (3) (4) (5) (6) ExposureEnergyb,Aug14 -0.018 -0.058* 0.035 (0.025) (0.030) (0.056) ExposureEnergy b,Aug14*TimeTrendb,m 0.001 0.001 0.001 0.002 0.001 0.001 (0.002) (0.002) (0.001) (0.002) (0.001) (0.002) Observations 7,160 7,160 7,160 7,160 7,160 7,160 R-squared 0.001 0.058 0.004 0.045 0.004 0.090 Bank-firm FE Yes Yes Yes Yes Yes Yes Month FE Yes - Yes - Yes - Firm-month FE No Yes No Yes No Yes SD(ExposureEnergyb,Aug14) 6.2 6.2 6.2 6.2 6.2 6.2 40 Table A5. - Lending Volume to the Energy Sector, Given Bank Exposure to Energy Sector Firms In this panel, we display the impact of bank exposure to the energy sector on its lending to energy sector borrowers, given the exposure of the bank relative to the borrower. We divide the sample of loans into Small or Large based on whether their value in August 2014 is above or below the median weighted by loan size. The dependent variables are in logs. Total Lendingf,b,m is the total lending volume to firm f by bank b in month m. Working Capitalf,b,m and Investmentf,b,m are total lending volume destined to working capital and investment, respectively. ExposureEnergyb,201409 is the ratio of lending to the energy sector of bank b in September 2014 to its tier 1 capital. Postm is an indicator that the observation is after September 2014. Standard errors are clustered at the state*month level. Detailed variable definitions are provided in Table A1. Observations are at the firm-bank-month level and the sample period is from January 2013 to December 2016. *** p<0.01, ** p<0.05, * p<0.1. Total Lendingf,b,m Working Capitalf,b,m Investmentf,b,m (1) (2) (3) (4) (5) (6) ExposureEnergy b,Aug14*Postm 0.00 0.02* 0.00 0.03* -0.00 0.01 (0.00) (0.01) (0.00) (0.02) (0.02) (0.03) Observations 20,610 3,353 19,721 2,932 1,486 890 R-squared 0.78 0.84 0.80 0.88 0.81 0.66 Bank-firm FE Yes Yes Yes Yes Yes Yes Month FE Yes Yes Yes Yes Yes Yes Loan size pre-shock Small Large Small Large Small Large SD(ExposureEnergyb,Aug14) 6.2 6.2 6.2 6.2 6.2 6.2 41 Table A6. Alternative Measures of Bank Exposure to the Energy Sector In this panel, we display the main estimates of a regression of total bank lending on bank exposure to firms in the energy sector, under four alternative bank exposure variables: (i) energy sector loans divided by total loans, (ii) energy sector loans divided by total assets, (iii) energy sector loans divided by total bank equity, and (iv) energy sector loans divided by the tier 1 ratio in December 2012. The top panel displays the results for the energy sector, and the bottom panel displays the results for the non-energy sector. The dependent variable in all regressions is the total lending volume (in logs) to firm f by bank b in month m. Postm is an indicator that the observation is after August 2014. Standard errors are clustered at the bank*month level. Detailed variable definitions are provided in Table A1. Observations are at the firm-bank-month level and the sample period is from January 2013 to June 2016. *** p<0.01, ** p<0.05, * p<0.1. Energy sector Total Lendingf,b,m (1) (2) (3) (4) ExposureEnergyb,Aug14*Postm 0.028* 0.136 0.098*** 0.045* (0.015) (0.104) (0.029) (0.026) Observations 6,994 6,994 6,994 6,994 R-squared 0.947 0.947 0.948 0.947 Bank-firm FE Yes Yes Yes Yes Firm-month FE Yes Yes Yes Yes SD(ExposureEnergyb,Aug14) 6.2 6.2 6.2 6.2 Non-energy sector Total Lendingf,b,m (1) (2) (3) (4) ExposureEnergyb,Aug14*Postm -0.012*** -0.131*** -0.033*** -0.006* (0.002) (0.023) (0.006) (0.003) Observations 1,262,640 1,262,640 1,262,640 1,262,640 R-squared 0.795 0.795 0.795 0.794 Bank-firm FE Yes Yes Yes Yes Firm-month FE Yes Yes Yes Yes SD(ExposureEnergyb,Aug14) 6.1 6.1 6.1 6.1 42