Spillover effects of banks liquidity risk control

Transcription

1 Spillover effects of banks liquidity risk control Yong Kyu Gam July 31, 2018 Abstract This study investigates spillover effects of banks liquidity risk control on the real economy by using the introduction of the Basel III liquidity regulation as shocks to banks. Since the Basel Committee s official endorsement of this new liquidity regulation in December 2010, banks exposed to high liquidity risk have reduced the proportion of loans in their portfolio significantly. In regions with many such banks, these reductions resulted in decreased small business lending and GDP growth. After the introduction of the new liquidity regulation, banks with less liquid balance sheets raised their deposit rates aggressively, generating liquidity problems in nearby banks through deposit competition and ultimately curbing the expansion of these banks credit supply. These results demonstrate that, while more stringent liquidity regulation may make banks safer, it carries real costs as well. JEL Classification: G21, G28. Keywords: Bank, Bank Lending, Bank Regulation, Basel Accords. I am deeply grateful to my dissertation committee, Radhakrishnan Gopalan (chair), Armando Gomes and Jennifer Dlugosz, for their invaluable guidances and supports. For helpful comments, I also thank Emily Gallagher, Philip H. Dybvig, Nonna Sorokina, Matthias Fleckenstein, Peter Morgan and seminar/conference participants at Washington University in St. Louis, Southwestern University of Finance and Economics, 2017 FMA Applied Finance Conference, 2017 Northern Finance Association Conference and 2017 World Finance & Banking Symposium. This is a revised version of the paper previously circulated as Basel III LCR: A Regulatory Shock on a Bank and Beyond The author is from the Institute of Financial Studies at Southwestern University of Finance and Economics and can be reached at gam@swufe.edu.cn. 1

2 Introduction Banks play a fundamental role in the economy, fulfilling functions such as providing loans to small businesses that might not otherwise receive funding (see, for instance, Gorton and Winton (2003)). To ensure well-functioning of liquidity provision to businesses, it is important to reduce banks vulnerability to liquidity shocks by controlling their liquidity risks, but banks strict liquidity control may have harmful effects on real economic activities by curbing credit supply. This paper documents spillover effects of banks liquidity risk control on the real economy in the United States. This study employs the introduction of the Liquidity Coverage Ratio (LCR), the new liquidity regulation announced by the Basel Committee on Banking Supervision (Basel Committee, BCBS) in , as shocks to banks liquidity risk management. First, this study investigates the divergence in real economic trends between two types of regions treated regions, where banks with less liquid balance sheets have substantial market shares, and control regions, where those banks have smaller market shares around the time the LCR standard was introduced. The results show that, after the LCR shock, the treated regions experienced lower growth in small business lending and GDP than the control regions. The changes made to banks deposit interest rates 2 and small business lending after the LCR introduction described in the next paragraphs support the idea that the divergence in real economic trends was driven by the LCR-induced credit supply reduction rather than demand-side shocks. Second, this paper examines differences in asset structure changes between banks with 1 The Liquidity Coverage Ratio (LCR) standard is one part of the Basel III international banking regulation packages. The Basel III consists of strengthening the existing bank capital frameworks (raising the quality of capital base, enhancing risk coverage, introducing a leverage ratio requirement, promoting countercyclical buffer, and addressing systematic risk) and introducing new bank liquidity standards (BCBS, 2011). The Basel III was officially announced in December After the introduction of the LCR standard, banks raised their deposit interest rates more aggressively in the treated regions than the banks did in the control regions. In general, if credit demands decline in local markets, banks in those regions are less likely to increase their deposit interest rates because of the reduced funding needs of those banks. Thus, it is unlikely that, after the LCR shock, credit demands in the treated regions diminished more than the demands in the control regions. 2

3 high liquidity risk (high-risk banks) and banks with low liquidity risk (low-risk banks) around the introduction of the LCR standard to identify the channel responsible for the real effects mentioned above. I find that high-risk banks increased the proportion of liquidities and reduced the proportion of loans in their total assets more substantially than the low-risk banks did after the LCR shock. Finally, I study the changes in deposit rates at high-risk banks after the LCR shock and their spillover effects on the deposit rates and credit supply at neighboring low-risk banks. The empirical findings highlight that, after the introduction of the LCR standard, the highrisk banks increased their deposit interest rates more aggressively than the low-risk banks. Also, low-risk banks located in the treated regions raised their deposit rates more than lowrisk banks in the control regions. These results suggest that deposit competition among banks intensified in the treated regions, resulting in new liquidity shocks to low-risk banks in those regions. Consequently, after the LCR shock, the low-risk banks in the treated regions cut the proportion of loans in their total assets more severely than did their counterparts in the control regions. This may explain why neighboring low-risk banks did not effectively fill the credit supply void in the treated regions created by the high-risk banks. This same conclusion can be reached by MSA-bank level small business lending regressions. After the introduction of the LCR standard, low-risk banks reduced their small business lending more significantly in the treated MSAs than the banks did in the control regions. For high-risk banks, on the other hand, there is no significant difference between their small business lending in the treated MSAs and the lending in the control regions after the LCR shock 3. The LCR measures whether a bank retains an adequate amount of high-quality liquid assets against expected net cash outflows during the next 30 calendar days following the financial market meltdown. The LCR regulation requires banks to meet the minimum LCR 3 This finding also supports the idea that the changes made to the real economic trends are driven by the LCR shock rather than credit demand reduction. If the credit demand shock was the underlying force of the changes to the real economic trends around the time of the LCR introduction, the high-risk-banks should reduce their small business lending more significantly in the treated MSAs than the banks did in the control regions after

4 (100 percent) at all times. The regulation was established by the Basel Committee on Banking Supervision in December 2010 and took effect in January To test the effect of the LCR on banks and the economy as a whole, this paper focuses on trend changes of key variables around the regulatory shock. I compare the period (one year or two years) prior to the approval of the LCR standard in December 2010 and the period (one year or two years) immediately following its approval 5. According to a quantitative impact study (QIS) by the Basel Committee, 40 percent of 212 sample banks had not met even 75 percent of the minimum requirement as of June 2011 (BCBS, 2012). As a result, many banks had yet to increase their LCRs by more than 25 percentage points during the remaining portion of the grace period. Because of the gradual nature of a bank s balance sheet reallocation, it can be assumed that banks with less liquidity spent the majority of the 4-year grace period working to meet the new LCR requirement rather than beginning this work just before official implementation in January In all the analyses, banks with higher liquidity risk (high-risk banks) are selected as the treated observations. Because LCR data for each bank is not available before December 2010, the loan-to-deposit ratio is used as a proxy for liquidity status. The higher the ratio, the lower the bank s liquidity and the more likely the bank is to fall below the minimum LCR requirement. The 20 percent of banks with the highest loan-to-deposit ratios as of the previous quarter-end are defined as high-risk banks. To calculate these banks spillover effects 4 The Basel Committee revised the LCR standard in January One of the major revisions is an extension of the LCR s grace period until January The minimum requirement of the LCR is newly set at 60 percent as of 2015 and rises in equal annual steps to reach 100 percent in 2019 (BCBS, 2013). Despite the change of the LCR standard in 2013, this study focuses on the effects of the LCR around December 2010, which is the point of time of the initial regulatory shock regarding the LCR introduction. 5 For month- or quarter-level panel regressions (bank- or branch-level analyses), I compare months or quarters of the year (2010) prior to the regulatory shock to the corresponding months or quarters of the year (2011) immediately following the shock. For year-level panel regressions (MSA-level analyses), I compare two years ( ) preceding the regulatory shock and two years ( ) after the shock to have sufficient samples for the regressions. 6 Banks may respond to the LCR shock in advance before However, it is less likely that banks respond to the new regulation without identifying its final form because of its uncertainty and the potential costs generated from the banks strengthened liquidity control. The LCR standard including the LCR formula and its grace period was confirmed in December 2010 by the Basel Committee. 4

5 on the real economy and neighboring banks, the treated regions are defined as Metropolitan Statistical Areas (MSAs, for real economy) 7 or counties (for neighboring banks) where aggregate deposit market shares of high-risk banks are at least 25 percent as of June 30th of the prior year. To effectively control for the influence of other regulatory reforms, such as strengthened capital requirements and the introduction of the maximum leverage ratio (BCBS, 2011), the regression model adds banks capital ratios, leverage ratios, and their interactions with post-period dummy variables as controls. The market-level regression model includes the market shares of banks with low capital ratios, the market shares of banks with high leverage ratios, and these variables interactions with post-period dummies as controls. This paper also includes several robustness tests to provide further support for the idea that the introduction of the LCR standard is the main driving force behind the empirical findings. One robustness test compares the key estimators between large and small banks. At the time of its introduction, the LCR was expected to apply solely to large banks, although the scope of its application had not yet been finalized. If the new LCR requirement is the primary cause of the changes observed around December 2010, the large high-risk banks should change their behavior more than small high-risk banks. As another robustness check, I conduct placebo tests by assuming that the treatment event occurred at different years. All robustness test results support the idea that the LCR is the main driving force behind the changes in banks behavior (and the consequent changes in the real economy) around the treatment event. This paper is related to several strands of literature. First, this paper analyzes regulatory impacts on banks. Thakor (1996) predicts that the minimum bank capital requirement will increase equilibrium credit rationing and lower aggregate lending. Aiyar et al. (2014a) highlight that regulated banks reduce lending in response to tighter capital requirements. 7 Because a local GDP is available only from an MSA level, MSAs are sorted into treated and control groups for real economic effects analyses. 5

6 Kisin and Manela (2016) find that, as the capital requirement increases by one percentage point, costs of $220 million per year are imposed on the banks providing liquidity guarantees to asset-backed commercial paper conduits. These are just some of the many studies of the effects of regulation on banks and financial institutions (see also Hellmann el al. (2000), Koehn and Santomero (1980), Peek and Rosenbaum (1995), Aiyar et al. (2014b), Di Maggio et al. (2015), and Rime (2001)). In line with this literature, this paper also addresses the spillover effects of one specific piece of regulatory reform, focusing on the newly established Basel III bank liquidity regulation. Second, this paper contributes to the discussion of real economic effects through the bank lending channel. Peek and Rosenbaum (2000) identify an exogenous loan supply shock connected to the Japanese banking crisis and link this shock to construction activity in the U.S. commercial real estate market. Gilge et al. (2016) demonstrate that banks exposed to liquidity windfalls from the oil and natural gas shale booms increase mortgage lending in nonboom areas through their branch networks. A number of other studies also investigate the bank lending channel (Bernanke and Blinder (1992), Kashyap and Stein (2000), Khwaja and Mian (2008), Ivashina and Scharfstein (2010), Chava and Purnanandam (2011) and Schnabl (2012)). This paper also discusses how bank lending affects the real economy, but adds an analysis of the deposit-competition channel to explain why neighboring low-risk banks do not effectively mitigate the negative real effects from the reduction in bank lending. Finally, this paper is related to the literature that studies the Basel III liquidity regulation. Giordana et al. (2011) simulate the optimal balance sheet adjustment a bank would need to perform in order to adhere to the new Basel III liquidity regulation using Luxembourg banking data. Covas and Driscoll (2014) develop a nonlinear dynamic general equilibrium model to study the macroeconomic impact of introducing a minimum liquidity standard for banks on top of existing capital adequacy requirements. Many other papers also predict and simulate the effects of the LCR regulation (Balasubramanyan and VanHoose (2013) and Keister and Bech (2012)). This study contributes by empirically verifying the predictions 6

7 and simulation results from the existing literature investigating the LCR s effects. The rest of the paper is organized as follows. Section 1 introduces the Basel III liquidity regulation. Section 2 explains the hypothesis of this study. Section 3 presents the data and empirical methodology. Section 4 describes the summary statistics. Section 5 presents empirical results. Section 6 concludes. 1 Review of Basel III Liquidity Coverage Ratio According to the Basel Accord, the LCR is designed to promote short-term resilience of a bank s liquidity risk by making sure that banks have sufficient amounts of high quality liquid assets to survive a significant stress scenario lasting for one month (BCBS, 2010). The LCR is defined as follows: LCR = High quality liquid assets T otal net cash outflow over the next 30 calendar days 100% (1) In other words, the LCR is calculated by dividing the total amount of high-quality liquid assets by total expected net cash outflows over the following 30 calendar days. The minimum requirement for the fully implemented LCR is 100 percent. The definition of a high-quality liquid asset is an asset that can be immediately converted into cash at little or no loss of value (BCBS, 2010). The liquidity of an asset depends on the underlying stress scenario, the volume to be monetized and the time frame (BCBS, 2010). Typical examples of highquality liquid assets are cash, central bank reserves, and government securities like Treasury bonds. Either no haircut or a minimal haircut rate is applied to the principal amount of the liquid assets depending on their degree of liquidity and quality. Net cash outflow over the designated period is calculated by adding expected cash outflows from liability items and subtracting anticipated cash inflows from asset items for the following thirty calendar days. During a financially distressed situation, depositors usually 7

8 run to banks and withdraw funds from their deposit accounts if their balance is not covered by the maximum limit of the national deposit insurance system. In such a credit crunch, banks encounter difficulties in rolling over their existing short-term financing. At the same time, banks try to collect cash by reducing loans origination and making full use of credit lines from other institutions. Expected cash inflows and outflows are computed by multiplying the designated drawdown or inflow rate and the balance amount of each asset and liability item, following the Basel III rules. Finally, the expected net cash outflow can be determined by deducting expected cash inflows from expected cash outflows. I present a hypothetical example to illustrate the LCR calculation. Assume that a bank has $100 million in assets. $5 million are government bonds and the other $95 million are retail loans. Among the retail loans, $15 million will be due within 30 days. The bank s funding structure is assumed to consist of retail deposits ($70 million), wholesale borrowings ($20 million) and equity ($10 million). Among the wholesale borrowings, $10 million will be due within 30 days. According to the Basel rules, the government bond is defined as a highquality liquid asset without any haircut. The rules text also specifies the cash inflow rate as 50 percent for the retail loan receivable in 30 days. The drawdown rates are 5 percent and 100 percent for the retail deposits and the wholesale borrowings payable in 30 days, respectively. Applying the LCR rule to this hypothetical example, the amount of the high quality liquid assets is $5 million and the net cash outflow is $6 million ( ). Thus, the bank s LCR is approximately 83.3 percent (5/6). 2 Hypothesis The LCR standard requires banks to maintain a minimum liquidity coverage ratio; thus, the LCR s effect on a given bank s behavior depends on that bank s liquidity status. The LCR impacts banks exposed to high liquidity risk (high-risk banks) more than banks with low liquidity risk (low-risk banks). 8

9 The LCR regulation was introduced in December 2010 with an effective date of January 2015, providing a grace period of approximately 4 years. According to a QIS conducted by the Basel Committee, many banks failed to meet even 75 percent of the minimum LCR level as of June For these banks, increasing the LCR to meet minimum requirements would take a substantial period of time, due to the long-term nature of bank loans and the negative impact of sudden asset reallocations on bank performance. Thus, I expect highrisk banks to start adapting to the new regulatory environment from the first stages of the transition period, which begins in December 2010, rather than just before the official LCR implementation in January I anticipate the introduction of the LCR standard to have four effects on banks and real economy. First, the LCR regulation changes high-risk banks asset structure by increasing the relative proportion of liquid assets such as cash and securities in their portfolios. While a bank loan generates cash inflows, which can offset expected cash outflows and increase the LCR, expanding the proportion of liquid assets is a more direct and efficient way to increase this ratio. Thus, banks should hold more liquid assets, which by default reduces the proportion of loans they hold. Second, asset reallocation among high-risk banks affects aggregate credit supply and the real economy. These effects vary by region depending on the extent of each bank s behavioral changes in response to the LCR standard and its relative position in the local market. If the majority of banks in a given area are high-risk banks and reduce credit supply in response to the introduction of the LCR standard, the region should be adversely affected. Borrowers have difficulties finding substitute funding sources in a timely manner. For this reason, after the LCR shock, areas where high-risk banks have large market shares should experience lower growth in lending and GDP than other regions. Third, the LCR regulation expands high-risk banks retail deposits at the expense of wholesale funding, because retail deposits are more favorable to the LCR in terms of reducing the expected cash outflows (the denominator of the LCR). Expected cash outflows are derived 9

10 by multiplying designated drawdown rates by the balance of each liability item. Different drawdown rates are applied to individual funding items depending on these items traits; traits include type (retail deposit or wholesale borrowing), coverage by deposit insurance, and remaining maturity. On average, drawdown rates for retail deposits are lower than for short-term wholesale borrowing. For this reason, it is predicted that high-risk banks have more incentives than low-risk banks to aggressively increase interest rates for retail deposits in order to attract such deposits after the LCR shock. Finally, a high-risk bank s deposit rate increase in response to a regulatory shock could have a spillover effect on neighboring low-risk banks deposit funding costs due to deposit competition among banks in local markets. The spillover effects on competing low-risk banks should be more conspicuous in regions where the market shares of high-risk banks are high (treated regions). The competing low-risk banks in these treated regions will likely raise deposit interest rates to protect deposit volumes from the severe deposit competition following the LCR shock. In short, the introduction of the LCR standard may create liquidity shocks for competing low-risk banks through their deposit competition with high-risk banks. This may curb expansion of the credit supply even among low-risk banks. Consequently, low-risk banks in the treated regions are less likely to effectively fill the void in credit supply caused by the introduction of the LCR standard. The testable hypothesis in this study is described as follows. Hypothesis: After the introduction of the LCR standard, banks with low liquidity (high-risk banks) reduce their credit supply and increase their deposit rates aggressively, leading to credit supply reductions in all neighboring banks by triggering liquidity shocks to those banks. As a result, regions where high-risk banks dominate the market experience lower aggregate lending and GDP growth after the LCR standard is introduced. 10

11 3 Data and Empirical Methodology 3.1 Data source Various data sources are used in this study. First, I use three different data sources to identify the effects of the LCR shock on the real economy. I obtain data on local GDP in each MSA from the Bureau of Economic Analysis (BEA). The Federal Financial Institutions Examination Council (FFIEC) provides data on each bank s annual small business lending in each MSA. Finally, I use CoreLogic to obtain a ZIP code-level house price index. This study also relies on the U.S. banks financial statement data from the Call Report. This dataset includes balance sheets, income statements, off-balance sheets, and risk-based capital for all banks that are regulated by the Federal Reserve System, the Federal Deposit Insurance Corporation, and the Office of the Comptroller of the Currency. From this data source, I can identify changes in the composition of a bank s assets and liabilities before and after the adoption of the Basel III liquidity regulation. Other bank-level control variables, such as size, capital ratio, leverage ratio, and non-performing loan ratio, are also constructed using this dataset. Third, I obtain branch-level deposit interest rates from RateWatch to determine the effect of the regulatory shock on banks deposit interest rates. This data source provides interest rates for various deposit products, including Certificates of Deposit with different maturities and balance ranges. The original data is provided weekly, but I convert it to a monthly frequency for this study. To calculate the spread on the deposit rate against a risk-free rate, I use the 3-Month Treasury Bill yield reported in the Federal Reserve Economic Data (FRED). Fourth, I obtain information on the deposit market share of each bank in each ZIP code, county or MSA from the Summary of Deposit provided by the Federal Deposit Insurance Corporation (FDIC). From this dataset, I can observe the branch-level deposit balances as 11

12 of June 30th of each year and calculate each bank s deposit market share for each ZIP code, county or MSA. The market share is a key factor in defining treated areas in this study. I also derive the Herfindahl-Hirschman Index (HHI) for each deposit market on a ZIP code-level, county-level or an MSA-level using this data. The HHI and the market share are included as control variables for branch-, local bank-, or regional economy-level regressions. A local bank is defined as a bank that collects more than 65 percent of its deposits from a given county, following Cortés (2014). 3.2 Empirical Design I first examine the spillover effect of the new liquidity regulation on the real economy. The regression model is defined as follows: Y msa,y = β 0 + β 1 T reated msa,y P ost y + β 2 T reated msa,y + Γ X msa,y + δ msa + δ y + ε msa,y (2) The subscripts msa and y refer to MSA and year, respectively. In this regression, I employ two different outcome variables: GDP and SBL. GDP refers to the annual percent change in the local GDP of each MSA. SBL is the annual percent change in small business lending in each MSA. In this regression, I use panel data consisting of MSA year observations and analyze trend changes of annual growth of GDP and small business lending in each MSA around the regulatory shock. I compare two years before the introduction of the new LCR standard in December 2010 ( pre-shock period ) and two years after its introduction ( post-shock period ). I compare the MSAs where high-risk banks hold higher deposit market shares to MSAs where high-risk banks hold lower market shares. In this study, high market share is defined as 25 percent or higher. Treated regions are defined each year based on high-risk banks market shares as of June 30th of the prior year. Control regions are selected among MSAs located in the same states as the treated regions. I 12

13 remove 12 MSAs with average annual GDPs exceeding 200 billion USD during from the sample in order to control for outliers in terms of MSA economy size. Figure 1 maps the treated and control regions as of X msa,y is a vector of MSA-year level control variables, which include LowCapMktShare, LowLevMktShare, HighNPLMktShare, LocalMkt- Share, SmallMktShare, BHCMktShare and HHI. Interaction terms between the above controls and the post-period dummy, Post, are also added as control variables. The appendix provides detailed variable definitions. δ msa and δ y refer to MSA fixed effects and year fixed effects, respectively. Standard errors are clustered at the MSA level. In my next set of analyses, I focus on the effect of the introduction of the LCR standard on banks asset composition. The regression model is as follows. Y i,t = β 0 + β 1 High i,t P ost t + β 2 High i,t + Γ X i,t + δ i + δ t + ε i,t (3) The subscripts i and t refer to bank and year-quarter, respectively. This regression compares quarterly changes in a bank s asset structure in the year (2010) preceding the shock to the same changes during the corresponding quarters of the year (2011) immediately following the shock. I compare banks with higher liquidity risk (high-risk banks) to banks with lower liquidity risk (low-risk banks). I use the loan-to-deposit ratio as a proxy for a bank s liquidity status. The higher the ratio, the lower the bank s liquidity and the higher the possibility that it will fall below the minimum LCR requirement. The loan-to-deposit ratio is a widely used indicator that measures banks general liquidity and is calculated by dividing total loans by total deposits. If a bank s loans are too large compared to its total deposits, it will be more likely to experience a liquidity deficit during a financial crisis. A lower amount of total deposits relative to total loans implies that part of the bank s loans are financed by wholesale funding, which is vulnerable to rollover risk during a financial crisis, leading to liquidity shortfalls. Thus, the 20 percent of banks in the sample with the highest loan-to-deposit ratios as of the previous quarter-end are defined as high-risk banks. 13

14 When the new LCR regulation was introduced in December 2010, the Basel Committee did not specify whether it would apply to only banks with assets exceeding a certain threshold. In the U.S., the scope of the new standard s application was only announced in For this reason, all high-risk banks, regardless of size, should have responded to the introduction of the LCR regulation in Thus, all large and small banks are included in my sample for this regression analysis. In robustness tests, I run separate regressions for large and small banks in order to observe how size impacts the differences in behaviors between high-risk and low-risk banks. Y i,t represents the outcome variable Liquidity/Assets or Loans/Assets, depending on the regression specification. These variables are expressed in terms of quarterly changes in asset composition from the previous quarter-end to the current quarter-end rather than the absolute amounts of bank holdings. In doing so, I can identify banks behavioral changes in response to the regulatory shock by controlling for a general mean-reverting process, which occurs when banks facing substantial liquidity risk change their balance sheet structures to reduce this risk in the quarter ahead. X i,t is a set of bank-level control variables, which include Size, Capital, Leverage, NPL, Local, Small and BHC. All controls are calculated as of the previous quarter-end. Interaction terms between the above controls and the dummy variable for the post-period, Post, are also added as control variables. The appendix provides detailed definitions for each variable. δ i and δ t represent bank fixed effects and quarter fixed effects, respectively. Standard errors are clustered at the bank level. My third test focuses on the effect of the new regulation on banks deposit interest rate spreads. The regression model is defined as follows: Spread j,m = β 0 + β 1 High j,m P ost m + β 2 High j,m + Γ X j,m + δ j + δ m + ε j,m (4) The subscript j refers to branch and m refers to year-month. In this model, I use a panel 14

15 of branch month data. Spread is the difference between a bank s deposit interest rate on a certificate of deposit (CD) with an account size of $10,000 and 12-month maturity (Drechsler, Savov, and Schnabl (2016)) and the yield on 3-Month Treasury Bills. Control variables consist of market (county)-level variables such as HighLTDMktShare, LowCapMktShare, LowLevMktShare, HighNPLMktShare, LocalMktShare, SmallMktShare, BHCMktShare, HHI and MktShare and the bank-level variables listed above. Also, BranchSize is added as a branch-level control variable. Interaction terms between the above controls and Post are added as control variables. The appendix provides detailed definitions for each variable. Except for the differences mentioned in this paragraph, all other specifications are the same as in the previous bank-level regression model. As a last step, I consider a regression that highlights the regulatory spillover effect on low-risk banks located in markets dominated by high-risk banks. This analysis identifies one channel through which the regulatory shock impacts the real economy: local deposit competition among banks. The regression models are defined as follows: Spread j,m = β 0 + β 1 Compete j,m P ost m + β 2 Compete j,m + Γ X j,m + δ j + δ m + ε j,m (5) Y i,t = β 0 + β 1 Compete i,t P ost t + β 2 Compete i,t + Γ X i,t + δ i + δ t + ε i,t (6) In Equation (5), the subscripts j and m refer to branch and year-month, respectively. In this model, the outcome variable is Spread, which is the difference between a bank s deposit interest rate on a 12-month CD with an account size of $10,000 and the yield on 3-Month Treasury Bills. This regression focuses on branches of low-risk banks. Compete identifies branches of low-risk banks located in treated counties, where high-risk banks hold at least 25 percent of the total deposit market shares as of the previous June 30th 8. Control observations 8 For observations in January to June, the previous June 30th implies June 30th of the prior year. For observations in July to December, the previous June 30th means June 30th of the current year. 15

16 are branches of low-risk banks located in control counties where the market shares of highrisk banks are less than 25 percent as of the previous June 30th 9. This regression employs the set of bank-level, county-level and branch-level control variables described above. Among the variables above, HighLTDMktShare is replaced by Compete, and High is replaced by LTD. Except for the differences mentioned in this paragraph, all other specifications are the same as in the prior branch-level regression models. In Equation (6), the subscripts i and t refer to bank and year-quarter, respectively. The outcome variable represents Loans/Assets or Credit/Assets, depending on the regression specification. Loans/Assets and Credit/Assets are quarterly changes in banks loans-toassets and credit-to-assets ratios from the previous quarter-end to the current quarter-end, respectively. Credit is defined as the sum of total loans and unused commitments. This regression focuses on local low-risk banks. Compete identifies local low-risk banks located in market dominated by high-risk banks (counties where the market shares of high-risk banks exceed 25 percent). Control observations are local low-risk banks located in counties where high-risk banks have smaller market shares. In this regression, each local bank is assigned to one county, where the local bank collects more than 65 percent of its deposits as of June 30th of This regression employs the set of bank-level and county-level control variables described above. As in the previous paragraph, HighLTDMktShare is replaced by Compete, and High is replaced by LTD. BranchSize and Local are removed from the set of control variables because this is a bank-level regression and focuses on local banks. Except for the differences mentioned in this paragraph, all other specifications are the same as in the prior low-risk bank branch level regression models. 9 See footnote 8 16

17 4 Summary Statistics This section presents summary statistics for the key dependent and independent variables in my regression models. Table 1 reports summary statistics for the regression that examines the impact of the LCR regulation on real economic factors (equation 2). In panel A, it can be seen that MSA-level GDP increased by 2.2 percent on average each year during the period The average annual growth in the amount of small business lending in each MSA is around percent during the same period. The proportion of the treated observations (treated MSAs) in the sample is approximately 50 percent. Table 2 shows the results of the univariate tests for the null hypothesis that the differences in the means of key variables between the treated and control groups are equal to zero. Panels A and B report the univariate test results of pre-period ( ) and post-period ( ), respectively. In the pre-period, both the treated and control MSAs experienced reduction of GDP and small business lending growth. The financial crisis may cause those consequences. Also, in this pre-period, the growth rates of GDP and small business lending in the treated MSAs are not significantly different from those in the control regions. In the post-period, the growth rates of GDP and small business lending recover in both types of regions, but the recoveries in the control MSAs are faster than the recoveries in the treated regions. In the next lines of this table, it is reported that, in the post-period, the banks with low leverage ratio or high NPL ratio are more likely to hold larger deposit market shares in the control MSAs than they did in the treated regions. Competitiveness among banks measured by deposit market HHI is higher in the treated MSAs in both periods. For these regions, I add control variables to control for the differences between the two types of MSAs. Panel A of Table A.1 in the Internet Appendix reports summary statistics for the regression that relates changes in banks asset composition to the regulatory shock (equation 3). All variables are based on bank-quarter panel data. During the sample period, the average 17

18 bank s liquidity-to-assets ratio increased by 0.57 percentage points per quarter, while the loans-to-assets ratio decreased by 0.52 percentage points per quarter. Panel B of Table A.1 in the Internet Appendix reports the summary statistics for the regression that measures the effects of the new LCR regulation on banks deposit interest rate spreads (equation 4). The spread on the average bank s deposit interest rate against the risk-free rate is around 44 basis points during the sample period. The data on deposit rate spreads is a panel of branch-month observations. 5 Empirical Results This section discusses five sets of empirical results. First, I report the changes in real economic trends observed around the introduction of the LCR regulation. Second, to determine how banks directly impact local economies through the bank lending channel, I describe the changes in high-risk banks asset composition around the LCR shock. Third, I investigate deposit competition, another channel through which banks affect the real economy; I focus on changes in high-risk banks deposit funding cost after the introduction of the LCR standard. Fourth, I consider high-risk banks spillover effect on neighboring banks operations via an indirect deposit competition channel around the time of the regulatory reform. Finally, results from several robustness tests provide evidence that the above findings can be attributed to the LCR shock. 5.1 Effect of the regulatory shock on the real economy First, I examine the changes in real economic trends around the introduction of the new LCR regulation. Table 3 reports the results of the regression defined in Equation (2). MSAs are sorted into treated and control groups based on the deposit market share of high-risk banks in each region as of June 30th of the prior year. In this regression, the coefficient of interest is the interaction term, Treated Post. This interaction term represents changes in 18

19 real economic factors in the treated regions after December 2010, when the new Basel III liquidity regulation was introduced, compared to changes in the control regions during the same period. In the first column of Table 3, the outcome variable is the annual GDP growth rate in each MSA. Treated Post is negative and statistically significant. In contrast, Treated is positive and significant. These results suggest that the average annual GDP growth rate in treated MSAs was higher than that in control MSAs before the adoption of the Basel III liquidity regulation in December After the regulatory reform, however, the average annual GDP growth rate in the treated MSAs is lower than that in control MSAs. In other words, regions where high-risk banks have greater market shares experience lower GDP growth after the regulatory shock. The first graphs in Figures 2 and 3 show that the parallel trends in GDP growth in treated and control MSAs observed until 2010 start to diverge after the LCR shock. In the second column of Table 3, the outcome variable is the annual percent change in aggregate small business lending in each MSA. Again, the estimated coefficient of Treated is positive, while that of the interaction term, Treated Post, is negative. Interestingly, the absolute value of the Treated Post coefficient is much larger than that of the Treated coefficient. Both coefficients are statistically significant. These results show that average annual growth in small business lending is higher in treated MSAs than control MSAs in the pre-shock period, but lower in the post-shock period. The second graphs in Figures 2 and 3 illustrate that the trends in MSA-level annual growth in small business lending are parallel until the regulatory shock, at which point they begin to diverge. As a robustness test, I run county-year panel regressions by aggregating small business lending up to each county-level rather than MSA-level. The regression results are robust to this alternative panel regression setting, as shown in Table A.3 (Panel A) of Internet Appendix. Taken together, the results in this subsection show that, in regions where banks exposed to higher liquidity risk hold substantial market power, real economic factors such as GDP 19

20 and small business lending are more likely to experience a slowdown after the LCR regulatory shock 10. The next step is to determine why this occurs, identifying reasons for the diverging trends in real economic factors between treated and control areas after the LCR shock. In these analyses, the primary difference between treated and control areas is the market share of high-risk banks, which are exposed to greater liquidity risk. Thus, I next examine what happens to high-risk banks operations after the regulatory reform, as compared to changes in low-risk banks during the same period. 5.2 Effect of the regulatory shock on banks asset structure Table 4 reports the results of the regression defined in Equation (3) to determine whether high-risk banks behave differently than low-risk banks after the LCR shock. In this regression, the main coefficient of interest is the interaction between the treated and time dummy variables, High Post. This interaction term represents changes in high-risk banks quarterly asset structure after the introduction of the new LCR standard, compared to changes in low-risk banks asset structure during the same period. In the first column of Table 4, the dependent variable is the quarterly change in banks liquidity-to-assets ratios. The coefficient of High Post is positive and statistically significant, implying that quarterly increases in high-risk banks liquidity-to-assets ratios grow larger after the regulatory shock relative to low-risk banks. In other words, banks facing higher liquidity risk, as measured by their loan-to-deposit ratios, increase their liquidity more severely after the introduction of the LCR standard than before it, compared to banks with lower liquidity risk. In the second column of Table 4, the outcome variable is the quarterly change in banks loans-to-assets ratios. In this specification, the estimated value for the coefficient of High Post is negative and statistically significant. From the estimated value of High, it can be observed 10 Similar regression results are obtained even when the regressions also cover 12 MSAs with average annual GDPs exceeding 200 billion USD during

21 that the average high-risk bank tends to reduce its loans-to-assets ratio over time more than the average low-risk bank did even before the treatment event. This suggests a meanreverting process before the regulatory shock: banks with weak liquidity tend to expand their liquidity as a percentage of total assets and reduce loans as a percentage of total assets. However, the interaction term, High Post, implies that this mean-reverting process becomes stronger in the post-lcr period. In other words, the quarterly reductions in highrisk banks loans-to-assets ratios become more substantial after the LCR shock compared to the analogous reductions in low-risk banks during this period. Overall, this subsection indicates that, after the introduction of the LCR standard, the average high-risk bank s asset structure changes radically: the relative proportion of liquidity in its portfolio increases, while the relative proportion of loans falls. This latter reduction could decrease aggregate lending in the local market, bringing about a slowdown in local GDP growth. This conclusion leads to another question: why don t nearby low-risk banks fill the localized credit supply voids created by their high-risk counterparts? 5.3 Effect of the regulatory shock on banks deposit interest rates To answer this question, I investigate changes in the operations of low-risk banks located in the treated areas, where high-risk banks dominate the market, following the introduction of the LCR regulation. To investigate this point, I need to consider the changes made to high-risk banks funding costs: the LCR can have a substantial influence on a bank s funding structure and changes in funding costs can affect competition among neighboring banks. This section investigates changes in banks deposit interest rates after the LCR shock (Equation 4). Table 5 reports the regression results. Deposit interest rates are reported at the branch level; I also add county-level control variables such as the HHI for the county-level deposit market and deposit market shares of banks in each county as of the previous June 21

22 30th 11. The total deposit amount of each branch as of the previous June 30th 12 is also used as a control variable. All other specifications are the same as in the previous tables. In this regression, the interaction terms, High Post, are the main coefficients of interest. In the first column, High Post is estimated to be and highly significant. In contrast, the estimated value for High is statistically insignificant. These results imply that the average spread on a high-risk bank s deposit interest rate is not statistically different from that of a control bank before the regulatory shock but becomes significantly higher in the post-lcr period. Turning to column 2, where I compare deposit rates between treated and control branches within the same county, the estimated coefficient for High Post remains both statistically and economically significant. Taken together, these results show that, after the regulatory shock, high-risk banks increase deposit interest rates more aggressively than do low-risk banks. This likely occurs because high-risk banks need to attract stable retail deposits in order to meet the minimum requirement set by the LCR regulation. As a next step, I investigate how these attempts to attract deposits by raising interest rates create spillover effects on neighboring low-risk banks. 5.4 Spillover effect of the regulatory shock on low-risk banks Deposit interest rate This subsection describes the results of the regression defined in Equation (5), which are reported in Table 6. This regression relates the deposit rate changes at branches of low-risk banks around the LCR shock to these branches locations. Low-risk banks branches are divided into treated and control groups based on their location. Branches in counties where banks with high loan-to-deposit ratios dominate the market belong to the treated group. As before, counties where treated banks have market shares above 25 percent are considered to 11 See footnote 8 12 See footnote 8 22

23 be dominated by such banks. In the first column of Table 6, the interaction term, Compete Post, is estimated to be and is statistically significant at the 1 percent level. The result implies that the deposit interest rates at branches located in treated regions increased by 10 basis points more after the regulatory shock compared to those in control regions, even though all branches belong to low-risk banks. In other words, even low-risk banks increased their deposit funding costs significantly after the LCR shock if they faced severe deposit competition with the highrisk banks in their local markets. In conclusion, the aggressive increases in high-risk banks deposit interest rates after the introduction of the LCR standard had a significant spillover effect on nearby low-risk banks. In the second and third columns, I add month-state fixed effects and month-bank fixed effects, respectively. I continue to find statistically significant and positive coefficients for the interaction terms. The second and third columns show that the regression results are robust to a comparison of branches within the same state and branches within the same bank across different counties. In Table 7, I move one step further to examine changes in low-risk local banks deposit funding costs after the regulatory shock. Except for the sample selection, all the regression specifications are the same as in Table 6. The first column of Table 7 encompasses all local banks and the second column focuses on independent local banks. An independent local bank is defined as any local bank that is not affiliated with a bank holding company (BHC). Unlike other banks, independent local banks may only utilize a limited internal capital market and thus depend more heavily on the local deposit market for funding. In the first column, the coefficient of Compete Post is estimated to be and statistically significant at the 1 percent level. Thus, the conclusions from the prior analysis hold when considering only local banks in the treated areas. In the second column, the interaction term is much more significant statistically and economically for independent local banks. Because the local deposit market is the major funding source for these banks, they must respond to deposit competition from high-risk banks by increasing their deposit interest rates more 23

24 aggressively Asset Structure Finally, I compare differences in quarterly changes in local banks asset structures before and after the regulatory shock, described in Equation (6). In the first and second columns of Table 8, the outcome variable is the quarterly change in the loans-to-assets ratio. In the last two columns of the table, the outcome variable is the quarterly change in the credit-toassets ratio. Credit is defined as the sum of total loans and unused commitments. As in the previous regression, the first and third columns encompass all low-risk local banks and the second and fourth columns focus on independent local banks. As in Tables 6 and 7, treated regions are defined as counties where the deposit market share of high-risk banks is at least 25 percent. In all four columns, the interaction terms are negative and statistically significant. This implies that, if a low-risk local bank is located in a treated region, the bank s credit supply is less likely to expand after the regulatory shock than before it, compared to a low-risk local bank located in a control region. The previous regressions attest to severe deposit competition in treated regions vs. control regions after the regulatory shock. This competition acts as a liquidity shock to low-risk banks located in treated regions, forcing them to attract more stable funding or hold more liquidity. This may ultimately reduce credit supply even among low-risk banks. In columns 2 and 4 of Table 6, the interaction terms are more statistically and economically significant than those in columns 1 and 3. Because local deposit markets are critical funding sources for independent local banks, these banks must respond more severely to the increased deposit competition in the local market and the resulting liquidity shock. This analysis shows why low-risk banks located in treated regions failed to effectively fill the credit supply voids created by high-risk banks, leading to a reduction in aggregate credit supply and a slowdown in GDP growth in the treated regions compared to the control regions after the new LCR regulation. 24

25 As the final stage, I move back to the growth rate of MSA-level small business lending, which was initially reported in Table 3. As an extension of this MSA-level tests, I run MSAbank level regressions by employing growth rates of individual banks small business lending to each MSA as outcome variables. The results are reported in Table 9. According to Column 1 of the table, after the introduction of the LCR standard, individual banks reduced small business lending more severely in the treated MSAs than the banks did in the control regions. This finding is valid when the regression limits samples to low-risk-banks (Column 2), but it becomes ambiguous when the regression covers only high-risk-banks (Column 3). These results also support the idea that, after the LCR shock, even low-risk banks deteriorated credit supply to the real economy in the treated regions as a result of spillover effects of the high-risk banks strengthened liquidity risk management on low-risk banks credit supply via the deposit competition. As a robustness check, I aggregate banks small business lending to county-bank-level instead of MSA-bank-level and can reach the same conclusion, as shown in Table A.3 (Panel B) of Internet Appendix. 5.5 Robustness In this paper, a series of analyses highlight that the introduction of the new LCR regulation in 2010 led to substantial changes in MSA-level real economic factors, high-risk banks asset structure, and both low- and high-risk banks operations. The treated and control groups in each regression are selected based on banks liquidity, which can be reasonably assumed to differentiate responses to the new LCR regulation. However, doubts may remain that the observed differences between the pre-shock and post-shock periods may not actually be caused by the announcement of the new Basel III liquidity regulation. Through several robustness tests, this section provides evidence that the driving force behind the empirical findings above is, in fact, the LCR shock. In Table 10, I compare changes in asset structure and deposit interest rates between large 25

26 and small banks. Historically, most Basel standards have been applied only to large banks or large bank holding companies. While the scope of the new LCR standard s application had not been determined at its introduction, historical trends could create an expectation that this standard would apply to large banks alone. For this reason, if the LCR is a leading factor behind the observed changes in banks behavior around December 2010, these changes would likely be more pronounced among large banks than small banks. I run two separate regressions on large and small banks, respectively, and compare the results. A large bank is defined as any bank holding total assets of more than $10 billion or affiliated with a bank holding company with total consolidated assets of more than $50 billion as of year-end 2009, following the criteria for mandatory regular stress tests under the Dodd-Frank Act, which was enacted in July A small bank is defined as any bank holding total assets of less than $2 billion (Cortés and Strahan, 2014) and not affiliated with any bank holding company as of year-end As seen in Table 10, the magnitudes of the interaction terms are much greater for large banks than for the entire sample (reported in Tables 3 and 4). On the other hand, for small banks, the same coefficients are statistically insignificant. These results show that changes in high-risk banks behavior compared to low-risk banks in the post-lcr period are more distinctive among big banks and nonexistent in small banks. These results support the idea that the LCR is a major reason for diverging trends in high-risk and low-risk banks behavior and the ensuring effects on the local economy and nearby banks after December Table 11 presents the results of another robustness test. This is a placebo test that reruns the baseline regressions assuming that the treatment event occurred in December 2009 or December As in the baseline regressions, I compare various trends one year before the shock to the same trends one year afterward. This test focuses on large banks, which are more likely to respond to the LCR shock. The deposit rate regression focuses on bank branches that have the authority to set their own deposit rates. As panels A-C in Table 11 show, the most significant results are observed in panel B, where December 2010 when 26

27 the LCR regulation was actually introduced is used as the date for the treatment event. These results provide further support for the idea that the LCR shock is a major driving force behind the findings reported in Tables 3 to 9. I conduct several other robustness tests. First, I rerun select baseline regressions by difference-in-differences approach with time-invariant treated and control observations throughout the sample period (see Table A.4 of the Internet Appendix). Because a bank s liquidity status changes over time, a bank considered high-risk at the beginning of the sample period may no longer be high-risk when the treatment event occurs. Such shifts in the composition of the treated and control groups can create noise in analyzing transitions of annual or quarterly changes of values around the treatment event, weakening the regression results. Because of this extra noise, values as of year- or quarter-end are employed as outcome variables instead of their annual or quarterly changes. Most coefficients in these difference-in-differences regressions are still significant and similar to those in the baseline regressions. One caveat is that the level changes of the year- or quarter-end values from pre- to post-periods may be driven by a general mean-reverting process, which occurs when banks facing significant liquidity risk transform their assets and liabilities structures to mitigate this risk in the quarter ahead, regardless of the introduction of the LCR standard. Another test explores the LCR s real effects by examining annual percent changes in ZIP code-level house price indices. In this analysis, each ZIP code is classified as a treated or control region based on the market share of high-risk banks in the ZIP code where it is located. Table A.5 in the Internet Appendix reports the results of the regression, which are consistent with my original findings. In ZIP codes where high-risk banks hold substantial market power, house price indices of those ZIP codes are more likely to decline after the LCR regulatory shock. A third robustness test matches high-risk banks and low-risk banks based on bank or branch size 13. The nearest neighboring match, determined via the Mahalanobis distance, is 13 For bank-level regressions, the treated (high-risk banks) and control (low-risk banks) observations are 27

28 employed to select the matched samples (high-risk banks and low-risk banks). Table A.6 in the Internet Appendix presents the regression results; the results are robust to this alternate method of constructing the treated and control samples. Next, Table A.7 in the Internet Appendix reports the results of a robustness test in which high-risk banks are defined using a fixed cutoff level for the loan-to-deposit ratio. In the baseline regressions, the group of high-risk banks consists of the top 20 percent of the sample in terms of the loan-to-deposit ratio. Because a bank s loan-to-deposit ratio changes over time, the absolute cutoff level for the top 20 percent can also change throughout the sample period. In this robustness test, high-risk banks are defined as banks with loan-todeposit ratios higher than 90%; low-risk banks are those banks with loan-to-deposit ratios lower than or equal to 90% 14. Using these definitions, the relative proportions of high-risk and low-risk banks change over time. The results are robust to this alternative definition 15. As the final robustness test, I replace the treated dummy variables with continuous variables in the baseline regressions. Table A.8 in the Internet Appendix reports the results. Treated, High and Compete are replaced with HighLTDMktShare, LTD and HighLTDMkt- Share, respectively. The results are robust to the employment of those continuous variables specification. 6 Conclusion This paper investigates the spillover effects of banks liquidity risk control on the real economy, as well as the channels through which these effects propagate by using the newly matched on bank size. For the branch-level regression, the treated and control observations are matched on branch size. 14 A loan-to-deposit ratio of 90% is around the cutoff level for top 20 percent banks in terms of banks loanto-deposit ratios. According to Forbes, a loan-to-deposit ratio around 80-90% would be a good benchmark. 15 In Panel A of Table A.5 in the Internet Appendix, the regressions cover the entire samples. In Panel B of the same table, the regressions focus on banks with loan-to-deposit ratios between 50% and 100% to control for outliers for banks liquidity status. Loan-to-deposit ratios of 50% and 100% correspond to cutoff levels for bottom 5-10 percent and top 5-10 percent banks in terms of banks loan-to-deposit ratios, respectively. The results are still robust to this alternative sample selection. 28

29 introduced Basel III LCR regulation as shocks to banks. I show that, after the introduction of the new LCR standard, local markets dominated by banks with high liquidity risk (highrisk banks) experience lower growth in GDP and small business lending compared to control regions. The LCR shock directly affects the real economy through the bank lending channel: after the shock, high-risk banks average proportion of loans relative to total assets decreases more severely than that of low-risk banks, reducing high-risk banks credit supply. I also provide evidence of an indirect channel for the LCR s spillover effects on the real economy: deposit competition among banks. After the introduction of the LCR regulation, highrisk banks increase their deposit interest rates more aggressively than low-risk banks in order to attract more retail depositors. The consequent increase in deposit funding costs affects the operations of neighboring low-risk banks. In fact, low-risk banks located in areas where high-risk banks market shares are substantial also increase their deposit interest rates significantly. This means that the deposit competition that followed the introduction of the LCR regulation provides new liquidity shocks to those nearby low-risk banks, causing them to suppress the expansion of credit supply to the real economy. My findings show that the effects of the new banking regulation extend beyond simply solving the problems it was designed to target. Rather, the new LCR standard has widespread spillover effects on the real economy, not only through the regulated banks behaviors but also through those of their nearby competitors. These spillover effects, though they may have been anticipated, were not intended by the Basel Committee in their initial endorsement of the Basel III regulation. In order to maximize the effectiveness of the LCR regulation as a tool for guaranteeing financial stability, its regulatory shocks would ideally be confined to fixing banks liquidity risk management problems, while its spillover effects on the real economy and other financial institutions would be minimized. Exactly how to mitigate these spillover effects is an important question for financial regulators or policymakers. 29

30 References [1] Shekhar Aiyar, Charles W Calomiris, John Hooley, Yevgeniya Korniyenko, and Tomasz Wieladek. The international transmission of bank capital requirements: Evidence from the uk. Journal of Financial Economics, 113(3): , 2014a. [2] Shekhar Aiyar, Charles W Calomiris, and Tomasz Wieladek. Does macro-prudential regulation leak? evidence from a uk policy experiment. Journal of Money, Credit and Banking, 46(s1): , 2014b. [3] Lakshmi Balasubramanyan and David D VanHoose. Bank balance sheet dynamics under a regulatory liquidity-coverage-ratio constraint. Journal of Macroeconomics, 37:53 67, [4] BCBS. Basel III : International framework for liquidity risk measurement, standards and monitoring, Available at [5] BCBS. Basel III : A global regulatory framework for more resilient banks and banking systems, Available at [6] BCBS. Results of the Basel III monitoring exercise as of 30 june 2011, Available at [7] BCBS. Basel III : The liquidity coverage ratio and liquidity risk monitoring tools, Available at [8] Ben S Bernanke and Alan S Blinder. The federal funds rate and the channels of monetary transmission. The American Economic Review, pages , [9] Sudheer Chava and Amiyatosh Purnanandam. The effect of banking crisis on bankdependent borrowers. Journal of Financial Economics, 99(1): , [10] U.S. Congress. Dodd frank wall street reform and consumer protection act, 165(i) [11] Kristle Romero Cortés. Rebuilding after disaster strikes: How local lenders aid in the recovery [12] Kristle Romero Cortés and Philip E Strahan. Tracing out capital flows: How financially integrated banks respond to natural disasters. Journal of Financial Economics, [13] Francisco Covas and John C Driscoll. Bank liquidity and capital regulation in general equilibrium [14] Marco Di Maggio, Amir Kermani, and Sanket Korgaonkar. Deregulation, competition and the race to the bottom [15] Itamar Drechsler, Alexi Savov, and Philipp Schnabl. The deposits channel of monetary policy. The Quarterly Journal of Economics,

31 [16] Fed Board FDIC and OCC. Joint press release: Federal banking regulators finalize liquidity coverage ratio, [17] ERIK P GILJE, Elena Loutskina, and Philip E Strahan. Exporting liquidity: Branch banking and financial integration. The Journal of Finance, [18] Gaston Giordana, Ingmar Schumacher, et al. The impact of the basel iii liquidity regulations on the bank lending channel: A luxembourg case study. Technical report, Central Bank of Luxembourg, [19] Gary Gorton and Andrew Winton. Financial intermediation. Handbook of the Economics of Finance, 1: , [20] Thomas F Hellmann, Kevin C Murdock, and Joseph E Stiglitz. Liberalization, moral hazard in banking, and prudential regulation: Are capital requirements enough? American economic review, pages , [21] Victoria Ivashina and David Scharfstein. Bank lending during the financial crisis of Journal of Financial economics, 97(3): , [22] Anil K Kashyap and Jeremy C Stein. What do a million observations on banks say about the transmission of monetary policy? American Economic Review, pages , [23] Todd Keister and Morten L Bech. On the liquidity coverage ratio and monetary policy implementation. BIS Quarterly Review December, [24] Asim Ijaz Khwaja and Atif Mian. Tracing the impact of bank liquidity shocks: Evidence from an emerging market. The American Economic Review, 98(4): , [25] Roni Kisin and Asaf Manela. The shadow cost of bank capital requirements. Review of Financial Studies, page hhw022, [26] Michael Koehn and Anthony M Santomero. Regulation of bank capital and portfolio risk. The journal of finance, 35(5): , [27] Joe Peek and Eric Rosengren. Bank regulation and the credit crunch. Journal of Banking & Finance, 19(3): , [28] Joe Peek and Eric S Rosengren. Collateral damage: Effects of the japanese bank crisis on real activity in the united states. American Economic Review, pages 30 45, [29] Bertrand Rime. Capital requirements and bank behaviour: Empirical evidence for switzerland. Journal of Banking & Finance, 25(4): , [30] Philipp Schnabl. The international transmission of bank liquidity shocks: Evidence from an emerging market. The Journal of Finance, 67(3): , [31] Anjan V Thakor. Capital requirements, monetary policy, and aggregate bank lending: theory and empirical evidence. The Journal of Finance, 51(1): ,

32 Figure 1: Treated and Control MSAs (as of 2009) 32

33 Figure 2: Real economic trend (MSA-level, ) a. Annual growth rate of GDP b. Annual growth rate of aggregate small business lending 33