Craft Lending: The Role of Small Banks in Small Business Finance

Craft Lending: The Role of Small Banks in Small Business Finance Lamont Black Micha l Kowalik December 2016 Abstract This paper shows the craft nature of small banks lending to small businesses when small banks face competition from large banks. Small banks have a greater ability to monitor their borrowers, but large banks have a lower cost of lending. In our theoretical model, an increase in large bank competition makes small banks especially valuable to intermediate-productivity borrowers. We analyze bank data using loan size as a proxy for borrowers productivity and find results consistent with our craft lending hypothesis. Small banks increase their small business lending in the intermediate loan size category ($250,000 to $1 million) following large bank entry into their market. These findings suggest that this intermediate segment of the small business loan market is a niche for small banks competing with large banks. Keywords: commercial banking, information, competition, small business lending JEL: G21, G28, G32 We thank Jose Berrospide, Kristian Blickle (discussant), Kristle Cortes (discussant), Matthieu Chavaz, Ron Feldman (discussant), Manisha Goel, Elena Loutskina, Ralf Meisenzahl, Andrea Moro (discussant), Ali Ozdagli, Joe Peek, Larry Wall (discussant), and Christina Wang for useful comments, as well as seminar participants at the Bank of England and Federal Reserve Bank of Boston and conference participants at the IAES 2015 in Boston, St. Louis Federal Reserve-CSBS Community Banking in the 21st Century, Eastern Finance Association, Federal Reserve System Meeting for Financial Structure and Stability, Financial Management Association, and summer IBEFA meetings. Andy Barton and Laura Fried provided excellent research assistance. The views expressed herein are those of the authors and do not necessarily represent those of the Federal Reserve Bank of Boston or the Federal Reserve System. DePaul University, 1 E Jackson Blvd, Chicago, IL 60601, tel.: 312-362-5617, email: lblack6@depaul.edu Federal Reserve Bank of Boston, 600 Atlantic Ave, Boston, MA 02120, tel.: 617-973-6367, email: michal.kowalik@bos.frb.org 1

1 Introduction In recent years, small banks in the U.S. have been exposed to increasing competition from large banks for small business borrowers. This increased competition has been facilitated by low cost lending technologies and improvements in information sharing, which have eroded informational advantages of small banks. Because small banks are still significant lenders to U.S. small businesses, an important question arises as to how competition from large banks affects the type of lending provided by small banks. 1 We show that small banks response to increasing competition is craft lending. Their borrowers are neither the smallest nor the largest firms, but rather firms of intermediate size. This is the loan market where soft information matters most, which makes it a niche role for small banks in small business finance. 2 In our theoretical model, small and large banks compete for entrepreneurs of varying productivity. Small and large banks differ along two dimensions. Small banks can better evaluate the risk of projects undertaken by the entrepreneurs due to their ability to monitor borrowers. However, small banks cost of lending (e.g., due to monitoring costs) is higher than the large banks cost. 3 We show that, despite large banks cost advantage, small banks attract entrepreneurs of intermediate productivity. Such entrepreneurs value the small banks ability to monitor projects, because this ability allows these banks to better tailor loan rates, maximize the project s value for the borrower, and outweigh the cost disadvantage. In contrast, high- and low-productivity entrepreneurs choose large banks. High-productivity entrepreneurs do not benefit from the monitoring provided by small banks and low-quality entrepreneurs prefer to undertake riskier projects 1 Recent evidence suggests that greater small bank presence in a market yields significantly more lending to and slightly lower failure rates of small firms during normal times (?). See also? for an overview of small firm finance, relationship lending, and the importance of bank organizational structure. The authors discuss potential implications of technological innovation and shifts in competitive conditions. 2 Importance of soft information in bank lending is highlighted by, among many, (?),?, and (?). 3 The model builds on early theoretical literature on competition between informed and uninformed lenders (???), which examines the role of relationship lending when a lending bank obtains an informational advantage about its borrower over competing lenders. 2

no matter the lender type. Our theoretical model yields the following empirical prediction: an increase in competition from large banks results in small banks increasing their share of lending to intermediate-productivity entrepreneurs and decreasing their share of lending to high- and low-productivity entrepreneurs. We test the above prediction of our model using Call Report data for small banks from 1994 to 2007. To analyze the changes in the nature of small banks small business lending, we study the volumes and shares of commercial and industrial (C&I) loans of different sizes in banks C&I portfolios. Although the data provides information on loan size rather than borrower characteristics, the loan size categories are a plausible proxy for entrepreneurs productivity. In particular, we look at changes in these loan portfolio levels and shares following changes in large bank competition. We proxy changes in large bank competition in markets of small banks using organic entry of large banks, large banks total deposit HHI, and large banks distance from small banks. Consistent with the model, we find that small, single-market banks increase the lending volume and share of their portfolio in loans between $100,000 and $1 million when additional large banks enter the small bank s market. The opposite results hold for loans less than $100,000 and greater than $1 million. Our empirical analysis contributes to the literature on bank size and small business lending. Prior evidence shows that small (less hierarchical) banks are better able to collect and act on soft information than large banks (???). 4 The soft information may be collected by loan officers, branch managers, or at a higher hierarchical level (?). Distance is an important factor in the use of soft information, even for large banks (???). Credit rationing can increase with the pairing of large banks and small firms as odd couples (?). More broadly, our paper contributes to the wider economic literature on competition be- 4? documents the screening efficiencies from having risk managers in a bank s organizational structure. 3

tween large and small firms.? challenge standard theories of firm-size distribution by developing an alternative theory in which industries are made up of large plants producing standardized goods and small plants making custom or specialty goods. The authors obtain empirical results with trade flow data that are consistent with this theory of competition. In related work,? use establishment-level data to quantify the impact of Big-Box store entry and growth on nearby single unit and local chain stores. The authors find a significant negative impact on employment growth at the smaller stores, but only when the Big-Box activity is both in the immediate area and in the same detailed industry. Finally, our paper sheds light on the changing landscape of the U.S. banking industry. There has been significant consolidation in the U.S. banking sector, with the number of banks falling to less than 7000 from over 14,000 in the 1980s. Despite this trend, the findings in our paper suggest that small banks will continue to serve a niche in the small business loan market in the future. This builds on recent literature examining the viability of small, local financial institutions (?). Consistent with?, our model and empirical results identify a potentially exploitable strategic position in the banking industry. We contribute to this argument by showing that small banks serve as craft lenders in the small business loan market. Even as large bank competition increases, small banks will likely have a comparative advantage in making mid-size loans to small businesses that can benefit from the use of soft information. The paper is organized as follows. Section 2 presents the theoretical model of craft lending based on small and large bank competition and develops our theoretical prediction. Section 3 discusses the data and descriptive statistics for the empirical analysis. Section 4 describes the empirical methodology and results. Section 5 concludes. 4

2 Model of craft lending There are three dates, t = 1, 2, 3, and the economy is populated by an entrepreneur and a number of banks. 5 The entrepreneur is risk-neutral and owns a project, but has no means to finance it. To finance the project, the entrepreneur borrows from a bank. There are two types of banks, small and large, with at least two banks of each type (which ensures competition between and within bank type). The payoff structure of the entrepreneur s project is determined by the entrepreneur s productivity and risk choice. Small and large banks observe the entrepreneur s productivity, but they differ from each other in the following way. First, the small banks observe the entrepreneur s risk choice whereas the large banks do not. Second, the large banks have a lower marginal cost of lending than the small banks, ρ L > ρ S. These assumptions capture the idea that the small banks possess better information about their borrowers than the large banks, but acquiring this information is expensive. More specifically, we assume that the small banks loan officers acquire this superior information by monitoring their borrowers actions. This monitoring is costly because the small banks have to pay their loan officers. Large banks, who do not employ such loan officers, have a lower cost of lending, but cannot acquire this additional information and, therefore, are subject to the moral hazard. 6 The timing is as follows. At date 1, each bank offers a loan contract (y; r) consisting of a loan size y and a loan rate r to an entrepreneur. The small banks can condition their contracts on the entrepreneur s risk choice at t = 2. The entrepreneur chooses an offer from one bank and undertakes a project. At date 2, the entrepreneur can shift the risk of her project if she borrowed 5 The model is a simplified version of a working paper by?. It is closest to?, who studies borrowers choice between a monitoring and an arm-length lender in a model of reputation building. 6 We do not model the reasons behind small and large banks technology choices. Such a choice might follow from two reasons. First, the large banks are better able to diversify idiosyncratic risks and therefore do not have to rely on additional monitoring of entrepreneurs. Second, the information acquired by the loan officer might be soft information (?). 5

from the large bank, and chooses the project s risk corresponding to the contract she chose from the small bank at t = 1. At date 3 the project returns are realized. The entrepreneur repays ry if the realized project return is high enough, and defaults otherwise. We model risk-shifting at date 2 as the entrepreneur s choice between a safe and risky project. The safe project pays a return A log y (A > 1) with certainty, where A is a parameter describing the entrepreneur s productivity. The risky project pays δa log y with probability p (0; 1) and 0 otherwise. The risky project represents a change in the entrepreneur s business strategy, which makes the entrepreneur s business riskier but more profitable in case of success, δ > 1. In addition, the moral hazard problem arises, because the choice of the risky project leads in expected terms to a less profitable business, pδ < 1. We solve the model by deriving the small banks offers first. Because the small banks observe the borrower s risk choice at t = 2, they offer two loan contracts, each conditioned on the risk the entrepreneur takes at t = 2. Because the safe and risky projects returns only differ in p and δ, we can simply find the small banks optimal offer for the risky project first, and then find the optimal offer for the safe project by setting p = δ = 1. Because the banks are competitive, they offer loan contracts that maximize the entrepreneur s expected payoff at t = 3 as long as they break even. Formally, for the risky project the small banks offer y and r such that the entrepreneur s expected payoff at t = 3 from the risky project is maximized, max y,r p (δa log y ry), subject to the small banks participation constraint, pry ρ S y, or r ρ S p. Deriving the entrepreneur s expected payoff with respect to y and solving the emerging first order condition for y delivers that the optimal loan amount is y = δa r. Because the entrepreneur s expected payoff 6

decreases in the loan rate r, the optimal loan rate is the one for which the small banks break even, ( ) r = ρ S. Hence, the small banks offer the entrepreneur a contract A p ρ S ; ρ S for the safe project ( ) and a contract for the risky project. pδa ρ S ; ρ S p The competitive large banks offers for a project of given risk differ from the offers of the small banks only due to a different cost of lending (we need to substitute ρ S with ρ L ). However, the large banks do not observe the risk the entrepreneur takes at t = 2. This implies that they can only offer only one contract. Moreover, when offering the contract corresponding to the safe project, the large have to make sure that the entrepreneur does not shift risk under this contract ( ) A at t = 2. Hence, the large banks will offer the entrepreneur a contract ρ L ; ρ L as long as the entrepreneur has no incentive to shift risk under this loan contract. Otherwise, the large banks ( ) pδa will offer the entrepreneur the contract ρ L ; ρ L. Formally, the entrepreneur does not shift risk p ( ) A under the contract ρ L ; ρ L if her payoff from the safe project is not lower than her payoff from the risky project under this contract: A log A ( A ρ L p δa log A ) A ρ L ρ L ρ L ρ L ρ L or A ρ L e 1 p 1 pδ AL. Hence, the large banks offer an entrepreneur with observed productivity A < A L the contract ( ) ( ) pδa ρ L ; ρ L A, whereas they offer an entrepreneur with A A p L the contract ρ L ; ρ L. Now we can study the entrepreneur s choice of the offers. An entrepreneur with a given ( ) A productivity A gets the following offers. The large banks offer her the loan contract ρ L ; ρ L if her ( ) pδa productivity A A L or the contract ρ L ; ρ L if her productivity A < A p L. The small banks offer 7

( ) A the entrepreneur, regardless of her productivity A, two contracts: ρ S ; ρ S if she takes the safer ( ) project and if she takes the riskier project. Hence, the entrepreneur with productivity pδa ρ S ; ρ S p A chooses the contract that delivers her the highest payoff. The entrepreneur with productivity A A L chooses an offer from a large bank and takes a safer project. As shown above such an entrepreneur does not want to shift risk at t = 2 and prefers the safe project when borrowing from a large bank. The loan offer from a large bank delivers higher payoff than the small banks offer for the safe project, because the small banks have a higher cost of lending and therefore their offered loan rates are higher and the offered loan amounts lower than those of the large banks. The entrepreneur with productivity A < A L faces a more complex trade-off. On the one hand, borrowing from a small bank is more costly than from a large bank, because the small banks have a higher cost of lending. On the other hand, when borrowing from a large bank, the entrepreneur has to take the risky project to compensate for the high loan rate that a large bank offers to protect itself against the moral hazard. When borrowing from a small bank, the entrepreneur prefers to take the safe project, because the safe project delivers higher payoff than the risky project. To find out which offer the entrepreneur takes, we compare her payoff from borrowing from a small bank and taking the safer project with the payoff from borrowing from a large bank and undertaking the risky project: A log A ( A ρ S p δa log pδa ρ L ρ S ρ S ρ L p ) pδa ρ L or pδ 1+ A e 1 pδ (log(pδ) log(ρ L))+ 1 1 pδ log(ρs) A S 8

Hence, an entrepreneur with productivity A [A S ; A L ) takes an offer from a small bank and the safe project, and the entrepreneur with productivity A < A S chooses the large bank s offer and the risky project. 7 These results highlight the advantage of small banks superior information for the borrower. Because the small banks observe risk, the entrepreneur can choose the more profitable safe project. For an entrepreneur with relatively high productivity the possibility to undertake the safe project is so valuable that she is willing to borrow from a small bank, whose lending cost is higher than the large banks cost. 8 An entrepreneur with relatively lower productivity prefers to take an offer from a large bank and run a riskier project, because their low productivity makes the safer project less profitable for them in the first place. Finally, our model predicts also that the loan size offered by banks is increasing in the borrowers productivity. 9 Hence, the following proposition summarizes the equilibrium outcome of this simple model. Proposition: High and low productivity entrepreneurs borrow from large banks, whereas entrepreneurs of intermediate quality borrow from small banks. More specifically, large banks makes the largest and smallest loans, and small banks focus on the loans of intermediate size. 3 Data and descriptive statistics To test our theoretical proposition, we use data on the balance sheets of U.S. banks from 1994 to 2007. The Call Report data contains information on banks small business lending in every 7 We assume here that the interval [A S ; A L ) is not empty, which holds when the small banks lending cost is not too high. 8 This is consistent with evidence that bank market power can facilitate access to credit by poorly performing forms and positively affect firm performance (?). Similarly, strong bank-firm relationships have been shown to improve the borrower s corporate governance (?). 9 For some parameters, the loan size around the cutoff A S might be non-monotonic, but in general the loan size is increasing in parameter A. 9

June 30 quarterly filing. Banks report the number and dollar amount of loans within three loan size buckets (described below). We use the size of C&I loans as a proxy for the entrepreneurs productivity, because we do not observe the firm s productivity and our model delivers a plausible prediction that a loan size is positively correlated with an entrepreneur s productivity. The Call Report includes data on the dollar volume of C&I loans with an amount less than $100,000, loans with an amount more than $100,000 but less than $250,000, and loans more than $250,000 but less than $1 million. We construct a category of loans greater than $1 million by subtracting the three above mentioned categories from the total C&I loans. These different loan sizes are likely associated with different lending technologies (?) and experience varying degrees of credit rationing (?).? use these same loan size thresholds to construct a bank survey on small business credit scoring. The Frame et al. survey (using data as of 1998) shows that 100 percent of the banks use credit scoring for loans under $100,000 and 74 percent of the banks score loans between $100,000 and $250,000. In contrast, only 21 percent of the surveyed banks score loans between $250,000 and $1 million. We focus on small banks with less than $1 billion in assets that operate in urban markets. The asset threshold is based on the asset-size limit used by the Federal Deposit Insurance Corporation (FDIC) to define a community bank (?). 10 An urban market is defined as a single Metropolitan Statistical Area (MSA). Market size structure is an important factor in determining the availability of credit to small firms (?). In our analysis of small banks, we distinguish between single-market versus multimarket banks, as in?. The FDIC s Summary of Deposit (SoD) data contains information on bank deposits by market. The SoD data measures every insured bank s deposits in each metropolitan market as 10 The FDIC asset-size limit for a community bank is adjusted upward over time from $250 million in 1985 to $1 billion in 2010. 10

of June 30th each year. We use this data for two key purposes. First, we distinguish between single- and multi-market banks in metropolitan markets. Single market banks are banks that only have deposits in a single metropolitan area. For these banks, we assume that their small business lending is done in the same market where they collect deposits. Second, we use the data to measure changes in large bank competition. Large banks are measured as banks with more than $10 billion in assets. We use three measures of competition from large banks: the large bank entry, the large bank Herfindahl-Hirschman Index (HHI) for deposits and the minimum distance between the small bank s branches and nearby large bank branches. We construct the large bank entry dummy by extracting information about branches that are established by a given large bank for the first time in a given market. We construct an HHI for each bank size category (large, medium, and small) within each market as a way to measure increased competition from large banks in a more continuous manner. Distance is measured in miles. Observations with smaller distances are not included. Figure 1 shows the evolution of these HHI and distance measures over our sample period of 1994 to 2007. Table 1 shows the descriptive statistics for the variables used in the regression analysis. All amounts are reported for the sample of small, single market banks. As can be seen, the loan size shares are meaningful in all four size categories. However, the largest share is in the $250,000 to $1 million category, which is 32% on average. 4 Empirical methodology and results The proposition developed in section 2 allows us to formulate our main hypothesis. Hypothesis: Faced with greater competition from large banks, small banks decrease the concen- 11

tration of their lending in the smallest and largest loans and increase the concentration of their lending in loans of intermediate size. We test our hypothesis in a panel regression on the annual data, where we analyze a sample of small banks from 1994 to 2007. We restrict our sample to the observations with total volume of $1 million dollars. This is especially important in the case of the two largest categories and ensures that the largest category is reported in a meaningful way. 11 The dependent variable is either the log of loan levels or the loan portfolio shares, so that we can analyze more fully the response of the small banks loan portfolio to large bank competition. For each bank in each year, we measure the loan levels and shares in the four size categories: less than $100,000, between $100,000 and $250,000, between $250,000 and $1 million, and greater than $1 million. The key explanatory variable is the Herfindahl-Hirschman index of large banks deposit share, which is our first proxy for large bank competition. The large-bank Herfindahl index is interacted with each of the loan size categories to identify the differential effect of large-bank competition on the volume/share of each loan size category. Our baseline regression specification is the following: LoanVolume/Share itl = β 1 Large bank comp it D 1 + β 2 Large bank comp it D 2 + β 3 Large bank comp it D 3 + β 4 Large bank comp it D 4 + Controls + Fixed Effects + ε itl 11 Rather than imposing the restriction on all observations, we could use an alternative restriction and run our regressions only for banks that have at least $1 million in the largest category. However, the results would be qualitatively the same. 12

where i is a bank, t is time and l is one of the four loan categories: l = 1 is the 0-100K category, l = 2 is the 100-250K category, l = 3 is the 250K-1M category, and i = 4 is the >1M category. D l are the dummies for the loan categories l = 1, 2, 3, 4. Hence, the coefficients of interest are β l, where we identify the effect of an increase in large-bank competition, Largebankcomp it, on the volume/share of each loan size category. As controls we include standard bank- and market-level characteristics: logarithm of total assets, ratios of loans and deposits to assets, a leverage ratio, a share of C&I loans in total loans, number of large banks in a given market, logarithm of population, unemployment rate, change in population and change in market rent. Our baseline specification for single-market banks contains four separate fixed effects at the level of banks, loan categories, markets and years. For multi-market banks we can run the same specification without market fixed effects and we weight the market controls by population of the markets in which these banks operate. For both type of banks we run also a version with two types of fixed effects: bank-loan and year-loan. In this specification we can control of any unobserved heterogeneity, with which banks can deal with different type of loans and any trends in loan categories. We will now describe in more detail our three large bank competition measures. Our first measure of large bank competition, LargeBankEntry it, is a dummy variable, which assumes value of 1 if at least one large bank establishes its first branch ever in a given market in a given year, and 0 otherwise. By definition this is a market-level variable for single-market banks and a bank-level variable for multi-market banks. We prefer to use a dummy specification rather than a number of such large banks directly, because the markets for which we have 2 or 3 large banks entering in a given year are highly skewed towards 2 years and 2 markets. 13

Our second measure of large banks competition, HHI(Deposits of large banks) it, is the sum of the squared deposit share for all large banks operating in all markets of bank b. 12 Although an increase in HHI is associated with lower competition (more concentrated market), our empirical specification allows for a different interpretation. The reason is that we use HHI of deposits for large banks and control in our regression for the HHI of deposits for all banks and for mid-sized banks. Hence, in our regression an increase in HHI of large banks (when holding the HHI of deposits for all banks and for mid-sized banks constant) leads to a mechanical reduction in the HHI of small bank deposits. This can be directly interpreted as large banks taking market share from small banks, which is the competitive effect at the core of our theoretical model. 13 Again, this is a market-level variable for single-market banks and a bank-level variable for multi-market banks. Finally, we also use minimum distance between the small bank and large banks as a measure of large bank competition. This measure is simply the minimum number of miles between the closest branches of a given small bank to any large bank. This is always a bank-level variable. The results of the HHI specification are shown in Tables 2-7. Tables with even numbers present the results for the loan volumes, and those with odd numbers for the loan shares. For simplicity purposes we present only the results for the specification with bank, loan, and year fixed effects. The other above mentioned results can be obtained from the authors upon request. In each table, column 1 shows the results for single-market banks, column 2 for multi-market banks, and column 3 reports the results for all banks. The results for each large bank competition proxy are presented in the following order: entry dummy, large bank HHI, and distance. 12 Specifically, this is measured as ( ) 2 Deposits of a given large bank in all markets of bank b Deposits of all banks in all markets of bank b 13 We prefer to use the HHI rather than the deposit shares directly, because the HHI also takes into account the number of banks in the market. 14

Evidence consistent with our theory would be that the magnitude of β 1 and β 4 is smaller than the magnitude β 2 and β 3. Such results would be mean that, as the competition from large banks increases, small banks increase lending in the two intermediate categories faster than in the smallest and largest loans categories. The regressions using the shares should confirm our results on loan volumes, because it is natural that stronger increases in one category must lead to an increase in the share of this category in the overall portfolio. As can be seen in tables 2-7, coefficients on the parameters of interest are consistent with our theoretical model. 14 Figure 2 depicts the results for the loan volumes visually. The figure plots the magnitude of β 1 (4) for each large bank competition proxy for the cases of single-market and all banks. As can be seen, each test produces a hump shape, in which the intermediate size categories are higher than the smallest and largest size categories. This indicates that small banks shift their small business loan portfolios toward these intermediate loan size categories when facing increased competition from large banks. This is consistent with our theoretical model of craft lending. 5 Conclusion In this paper, we analyze the concept of craft lending. Our theoretical model of craft lending shows that small banks will lend to borrowers of intermediate productivity when competing with large banks. The empirical results in the paper are consistent with this prediction. Using bank balance sheet data, we find that small, local banks increase their lending in the range of $250,000 to $1 million when faced by greater large bank competition. The results are robust to several different measures of competition. 14 The only exception is the coefficient for the largest category for single-market banks for distance. 15

Although the future of small banks remains somewhat uncertain, our results suggest that small banks will continue to serve some small business borrowers despite increased competition from large banks. It will likely not be the smallest borrowers or the largest borrowers. Small banks will likely serve borrowers of intermediate size and quality. These borrowers benefit most from the proximity of small banks and the ability of small banks to tailor loans to specific credit needs. Small businesses are important to economic growth, which makes credit availability for small businesses an important issue for economic policy (?). Our future research will continue to analyze this changing role of small banks in the U.S. financial system. 16

Figure 1: Evolution of HHI and distance as measures of large bank competition These figures show the changes in HHI and distance for small banks from 1994 to 2007. Only market data for banks with assets less that $1 billion are included. The HHI shows the evolution of competition from each of the bank size categories for small, single-market banks. Distance is measured as the minimum distance from one of the small bank s branches to a large bank branch. This distance measure is reported for both single-market and multi-market small banks. HHI for Large, Medium, and Small Banks Minimum Distance to Large Bank 17

Figure 2: Estimated response of small banks lending to large bank competition These figures show the estimated coefficients based on tables 2 through 5. The coefficients are reported for each of the four loan size categories. The four panels show the results for our two measures of competition: large bank HHI and the minimum distance to a large bank branch. Under each competition heading, we plot the coefficient results for log portfolio levels and loan portfolio shares. Black squares in the line plots indicate coefficients that are significant at the 10% level. Single-market banks All banks 18

Table 1: Descriptive statistics for single market banks This table reports the descriptive statistics for the variables used in the empirical analysis. All values are reported for the sample of small, single-market banks. (1) (2) (3) (4) (5) (6) (7) (8) VARIABLES N mean sd p50 p25 p75 min max 19 Assets in Mill. 29,719 186.3 171.4 127.5 73.63 234.3 2.641 999.8 Deposits-to-assets ratio 29,719 83.69 8.250 85.59 80.83 88.97 0 102.9 Leverage ratio 25,722 10.42 5.927 9.100 7.940 11.05 0.300 282.9 Loans-to-assets ratio 29,719 64.37 14.21 65.86 55.95 74.45 2.069 98.84 C&I-to-total loans ratio 29,719 0.127 0.0858 0.108 0.0680 0.166 0 0.968 0-100K-to-total C&I 29,718 28.38 16.07 26.26 16.27 38.56 0.00255 95.26 100-250K-to-total C&I 29,718 19.53 8.586 19.01 13.64 24.66 0.00359 74.64 250K-1M-to-total C&I 29,718 32.28 14.43 31.57 22.23 41.28 0.0326 94.07 +1M-to-total C&I 29,718 19.81 18.50 15.05 4.721 30.00 0.00126 99.50 Large Bank Deposit HHI 22,386 0.0758 0.0657 0.0669 0.0301 0.106 0 0.814 Total Deposits HHI 22,386 0.109 0.0599 0.0983 0.0723 0.136 0.0224 0.815 Mid Bank Deposit HHI 22,386 0.0215 0.0371 0.00510 0.00175 0.0253 0 0.534 Distance to large bank 17,996 1.309 4.099 0.145 0.0456 0.559 0 63.75 Population in Mill. 21,181 3,292 4,041 1,838 457.0 4,489 55.94 19,254 Population growth rate 20,755 9.769 53.69 0.0107 0.00450 0.0204-0.952 482.0 Market rent growth rate (pop-wgh) 20,755 9.896 54.40 0.0266 0.0123 0.0442-0.453 475.9 Unemployment growth rate (pop-wgh) 20,740 8.078 43.24-0.0202-0.0930 0.105-1 373.2

Table 2: Small bank loan portfolio levels and large bank entry This table shows results for loan portfolio levels and the large bank entry. The dependent variable is the log of loan levels for each of the four loan size categories. Only banks with assets less that $1 billion are included. Column 1 is for single-market banks; column 2 is for multimarket banks; and column 3 is for all banks. Robust standard errors clustered by market are in parentheses. *** p<0.01, ** p<0.05, * p<0.1. (1) (2) (3) VARIABLES Single Multi All Large Bank Entry * 0-100K loans -0.0734*** -0.0867*** -0.0862*** (0.0150) (0.0260) (0.0132) Large Bank Entry * 100-250K loans 0.0185 0.0292* 0.0250*** (0.0116) (0.0172) (0.00968) Large Bank Entry * 250K-1M loans 0.0718*** 0.0836*** 0.0772*** (0.0142) (0.0223) (0.0120) Large Bank Entry * +1M loans 0.0343 0.0175 0.0253 (0.0288) (0.0424) (0.0241) Log of assets 0.757*** 0.864*** 0.797*** (0.0130) (0.0198) (0.0105) Loans-to-assets ratio 0.00127*** 0.00190*** 0.00115*** (0.000387) (0.000570) (0.000323) Deposits-to-assets ratio -0.000174 0.000696 0.000858* (0.000633) (0.000787) (0.000509) Leverage ratio -0.00157-0.000668 0.000538 (0.00142) (0.00267) (0.00119) Ratio of CI to Total Loans 4.748*** 5.569*** 5.002*** (0.107) (0.169) (0.0899) Number of Large Banks -0.00239* -0.000209-0.00231*** (0.00133) (0.000850) (0.000720) Log of Population (unwgh) -0.186** -0.00525 0.0102 (0.0776) (0.0102) (0.00748) Population growth rate (unwgh) -0.0608* 0.000445-0.000969 (0.0346) (0.000972) (0.000946) Market rent growth rate (pop-wgh) 0.0626* -0.00900 0.0152 (0.0324) (0.0287) (0.00974) Unemployment growth rate (pop-wgh) -0.00913-0.0166-0.0167* (0.0137) (0.0163) (0.0101) Observations 79,005 48,398 127,433 R-squared 0.686 0.721 0.707 20

Table 3: Small bank loan portfolio shares and large bank entry This table shows results for loan portfolio shares and the large bank entry. The dependent variable is the loan portfolio shares for each of the four loan size categories. Only banks with assets less that $1 billion are included. Column 1 is for single-market banks; column 2 for multimarket banks; and column 3 for all banks. Robust standard errors clustered by market are in parentheses. *** p<0.01, ** p<0.05, * p<0.1. (1) (2) (3) VARIABLES Single Multi All Large Bank Entry * 0-100K loans -2.392*** -1.948*** -2.432*** (0.396) (0.687) (0.344) Large Bank Entry * 100-250K loans 0.336 0.654** 0.463** (0.259) (0.329) (0.209) Large Bank Entry * 250K-1M loans 1.910*** 1.863*** 2.011*** (0.439) (0.647) (0.366) Large Bank Entry * +1M loans 0.427 0.665 0.409 (0.701) (1.032) (0.586) Log of assets -4.184*** -2.213*** -3.659*** (0.287) (0.387) (0.205) Loans-to-assets ratio -0.0150* -0.0208* -0.0215*** (0.00846) (0.0109) (0.00656) Deposits-to-assets ratio -0.0135 0.00765 0.00295 (0.0136) (0.0167) (0.0105) Leverage ratio 0.129*** 0.169*** 0.161*** (0.0388) (0.0608) (0.0324) Ratio of CI to Total Loans -34.64*** -24.47*** -31.07*** (1.696) (2.024) (1.269) Number of Large Banks 0.111*** 0.0120 0.0102 (0.0325) (0.0209) (0.0170) Log of Population (unwgh) -3.373 0.345** 0.393*** (2.081) (0.163) (0.134) Population growth rate (unwgh) -1.557* 0.0128-0.0194 (0.919) (0.0248) (0.0194) Market rent growth rate (pop-wgh) 1.282-1.306** -0.175 (0.819) (0.624) (0.216) Unemployment growth rate (pop-wgh) -0.00342 0.00843 0.0499 (0.361) (0.280) (0.211) Observations 79,005 48,398 127,433 R-squared 0.325 0.266 0.303 21

Table 4: Small bank loan portfolio levels and large bank HHI This table shows results for loan portfolio levels and the large bank HHI of the small bank s market(s). The dependent variable is the log of loan levels for each of the four loan size categories. Only banks with assets less that $1 billion are included. Column 1 is for single-market banks; column 2 is for multimarket banks; and column 3 is for all banks. Robust standard errors clustered by market are in parentheses. *** p<0.01, ** p<0.05, * p<0.1. VARIABLES Single Multi All Large Bank Deposit HHI * 0-100K loans 0.00765-0.270-0.0348 (0.248) (0.236) (0.167) Large Bank Deposit HHI * 100-250K loans 0.564** 0.363* 0.565*** (0.229) (0.201) (0.146) Large Bank Deposit HHI * 250K-1M loans 0.780*** 0.425** 0.712*** (0.249) (0.212) (0.158) Large Bank Deposit HHI * +1M loans -0.273 0.226 0.0546 (0.285) (0.263) (0.197) Total Deposits HHI -0.320-0.113-0.298** (0.240) (0.200) (0.147) Mid Bank Deposit HHI -0.0244 0.00590 0.000258 (0.198) (0.190) (0.129) Log of assets 0.758*** 0.863*** 0.796*** (0.0130) (0.0197) (0.0105) Loans-to-assets ratio 0.00131*** 0.00191*** 0.00118*** (0.000386) (0.000568) (0.000323) Deposits-to-assets ratio -0.000223 0.000704 0.000825 (0.000631) (0.000789) (0.000509) Leverage ratio -0.00157-0.000815 0.000495 (0.00142) (0.00266) (0.00119) Ratio of CI to Total Loans 4.744*** 5.569*** 4.998*** (0.107) (0.170) (0.0898) Number of Large Banks -0.00251* -0.000117-0.00229*** (0.00133) (0.000842) (0.000716) Log of Population (unwgh) -0.177** -0.00804 0.00419 (0.0773) (0.0110) (0.00775) Population growth rate (unwgh) -0.0666* 0.000390-0.00101 (0.0347) (0.000973) (0.000945) Market rent growth rate (pop-wgh) 0.0678** -0.00668 0.0162* (0.0323) (0.0288) (0.00977) Unemployment growth rate (pop-wgh) -0.00858-0.0175-0.0165* (0.0137) (0.0163) (0.00999) Observations 79,005 48,342 127,377 R-squared 0.686 0.722 0.708 22

Table 5: Small bank loan portfolio shares and large bank competition This table shows results for loan portfolio shares and the large bank HHI of the small bank s market(s). The dependent variable is the loan portfolio shares for each of the four loan size categories. Only banks with assets less that $1 billion are included. Column 1 is for single-market banks; column 2 for multimarket banks; and column 3 for all banks. Robust standard errors clustered by market are in parentheses. *** p<0.01, ** p<0.05, * p<0.1. (1) (2) (3) VARIABLES Single Multi All Large Bank Deposit HHI * 0-100K loans -13.02* -5.174-11.29** (7.155) (6.143) (4.760) Large Bank Deposit HHI * 100-250K loans 0.827 9.475** 3.213 (6.462) (4.737) (4.036) Large Bank Deposit HHI * 250K-1M loans 11.94* 13.34** 11.25** (7.152) (5.315) (4.483) Large Bank Deposit HHI * +1M loans -18.99** 6.097-8.930* (7.778) (6.509) (5.261) Total Deposits HHI 6.271-4.734 1.977 (6.650) (4.837) (4.124) Mid Bank Deposit HHI -6.147 2.768-3.836 (5.657) (4.446) (3.649) Log of assets -4.173*** -2.231*** -3.664*** (0.286) (0.387) (0.206) Loans-to-assets ratio -0.0148* -0.0204* -0.0214*** (0.00846) (0.0109) (0.00656) Deposits-to-assets ratio -0.0143 0.00773 0.00233 (0.0135) (0.0168) (0.0105) Leverage ratio 0.129*** 0.166*** 0.160*** (0.0388) (0.0608) (0.0324) Ratio of CI to Total Loans -34.67*** -24.46*** -31.06*** (1.696) (2.016) (1.267) Number of Large Banks 0.102*** 0.0144 0.00924 (0.0323) (0.0208) (0.0169) Log of Population (unwgh) -3.334 0.264 0.389*** (2.071) (0.172) (0.143) Population growth rate (unwgh) -1.626* 0.0114-0.0196 (0.919) (0.0251) (0.0196) Market rent growth rate (pop-wgh) 1.354* -1.254** -0.162 (0.816) (0.630) (0.216) Unemployment growth rate (pop-wgh) -0.00637-0.00321 0.0410 (0.361) (0.281) (0.211) Observations 79,005 48,342 127,377 R-squared 0.327 0.267 0.305 23

Table 6: Loan portfolio levels and distance to large bank competitors This table shows results for loan portfolio levels and the minimum distance to a large bank. The dependent variable is the log of loan levels for each of the four loan size categories. Only observations with distance greater than one mile are included and only banks with assets less that $1 billion are included. Column 1 is for single-market banks; column 2 is for multimarket banks; and column 3 is for all banks. Log of population, population growth rate, market rent growth rate, and unemployment growth rate are deposit-weighted averages across markets for multimarket banks. Robust standard errors clustered by market are in parentheses. *** p<0.01, ** p<0.05, * p<0.1. (1) (2) (3) VARIABLES Single Multi All (-)Min. Distance to Large Bank * 0-100K loans -0.0151*** -0.00904*** -0.0120*** (0.00146) (0.00156) (0.00110) (-)Min. Distance to Large Bank * 100-250K loans -0.00209 0.000299 0.000286 (0.00139) (0.00113) (0.000885) (-)Min. Distance to Large Bank * 250K-1M loans 0.0119*** 0.00719*** 0.00984*** (0.00173) (0.00140) (0.00112) (-)Min. Distance to Large Bank * +1M loans 0.0138*** 0.00524** 0.00775*** (0.00377) (0.00262) (0.00219) Log of assets 0.769*** 0.856*** 0.805*** (0.0152) (0.0206) (0.0116) Loans-to-assets ratio 0.00144*** 0.00210*** 0.00151*** (0.000424) (0.000590) (0.000342) Deposits-to-assets ratio -0.000385 0.000593 0.000571 (0.000690) (0.000804) (0.000529) Leverage ratio -0.00241-0.00151-0.000982 (0.00172) (0.00274) (0.00138) Ratio of CI to Total Loans 5.073*** 5.570*** 5.269*** (0.145) (0.178) (0.110) Number of Large Banks -0.00228-0.000221-0.00195*** (0.00147) (0.000852) (0.000740) Log of Population (unwgh) -0.279** -0.00407 0.0107 (0.116) (0.0113) (0.00854) Population growth rate (unwgh) -0.0575 0.000355-0.000950 (0.0369) (0.00108) (0.00101) Market rent growth rate (pop-wgh) 0.0559-0.0132 0.0113 (0.0343) (0.0294) (0.0102) Unemployment growth rate (pop-wgh) -0.00477-0.0145-0.0133 (0.0152) (0.0166) (0.0106) Observations 65,993 45,137 111,153 R-squared 0.686 0.722 0.707 24

Table 7: Loan portfolio shares and distance to large bank competitors This table shows results for loan portfolio shares and the minimum distance to a large bank. The dependent variable is the loan portfolio shares for each of the four loan size categories. Only observations with distance greater than one mile are included and only banks with assets less that $1 billion are included. Column 1 is for single-market banks; column 2 for multimarket banks; and column 3 for all banks. Log of population, population growth rate, market rent growth rate, and unemployment growth rate are deposit-weighted averages across markets for multi-market banks. Robust standard errors clustered by market are in parentheses. *** p<0.01, ** p<0.05, * p<0.1. (1) (2) (3) VARIABLES Single Multi All (-)Min. Distance to Large Bank * 0-100K loans -0.548*** -0.280*** -0.400*** (0.0523) (0.0490) (0.0361) (-)Min. Distance to Large Bank * 100-250K loans -0.0805** 0.0164-0.00320 (0.0399) (0.0273) (0.0225) (-)Min. Distance to Large Bank * 250K-1M loans 0.229*** 0.143*** 0.206*** (0.0515) (0.0381) (0.0314) (-)Min. Distance to Large Bank * +1M loans 0.288*** 0.138** 0.175*** (0.0960) (0.0653) (0.0553) Log of assets -4.155*** -2.329*** -3.459*** (0.308) (0.403) (0.220) Loans-to-assets ratio -0.00740-0.0216* -0.0158** (0.00886) (0.0112) (0.00693) Deposits-to-assets ratio -0.00765 0.00393 0.00742 (0.0146) (0.0173) (0.0111) Leverage ratio 0.0818* 0.129** 0.113*** (0.0434) (0.0633) (0.0354) Ratio of CI to Total Loans -32.87*** -23.17*** -28.92*** (1.757) (2.047) (1.307) Number of Large Banks 0.111*** 0.00982 0.0123 (0.0340) (0.0210) (0.0173) Log of Population (unwgh) -2.660 0.215 0.314** (2.750) (0.176) (0.139) Population growth rate (unwgh) -0.818 0.0159-0.0168 (0.932) (0.0271) (0.0214) Market rent growth rate (pop-wgh) 0.756-1.233* -0.133 (0.828) (0.651) (0.221) Unemployment growth rate (pop-wgh) -0.175-0.0248-0.0288 (0.383) (0.281) (0.212) Observations 65,993 45,137 111,153 R-squared 0.316 0.263 0.294 25