Bank size, lending technologies, and small business finance Allen N. Berger a,*, Lamont K. Black b a Moore School of Business, University of South Carolina, Columbia, SC 29208, U.S.A. a Wharton Financial Institutions Center, Philadelphia, PA 19104, U.S.A. a CentER, Tilburg University, The Netherlands b Board of Governors of the Federal Reserve System, Washington, DC 20551 U.S.A. This version: August 12, 2010 Abstract Under the current paradigm in small business lending research, large banks tend to specialize in lending to relatively large, informationally transparent firms using hard information, while small banks have advantages in lending to smaller, less transparent firms using soft information. We go beyond this paradigm to analyze the comparative advantages of large and small banks in specific lending technologies. Our analysis begins with the identification of fixed-asset lending technologies used to make small business loans. Our results suggest that large banks do not have equal advantages in all of these hard lending technologies and these advantages are not all increasing monotonically in firm size, contrary to the predictions of the current paradigm. We also analyze lines of credit without fixed-asset collateral to focus on relationship lending. We confirm that small banks have a comparative advantage in relationship lending, but this appears to be strongest for lending to the largest firms. JEL classification: G21; G28; G34; L14 Keywords: Banks; Lending technologies; Relationship lending; Small business. The views expressed do not necessarily reflect those of the Federal Reserve Board or its staff. The authors thank the anonymous referee, Bob Avery, Brian Bucks, Bob DeYoung, Traci Mach, Greg Udell, John Wolken, and participants at a seminar at the Federal Reserve Board for helpful comments and suggestions, and Dan Grodzicki and Phil Ostromogolsky for valuable research assistance. * Corresponding author. Tel: +1 803 576 8440; fax: +1 803 777 6876. E-mail addresses: aberger@moore.sc.edu (A.N. Berger), lamont.black@frb.gov (L.K. Black).
1. Introduction The current research paradigm in small business lending emphasizes the advantages of large banks in lending to large, informationally transparent firms and the advantages of small banks in lending to small, opaque firms. In this paradigm, loan officers at large banks are hypothesized to focus on lending to large, transparent firms using their comparative advantages in lending technologies based primarily on hard quantitative information that the loan officers may credibly communicate to others in the bank such as financial ratios from certified audited financial statements, collateral values, and credit scores. Loan officers at small banks have more flexibility to evaluate credit using techniques based primarily on soft qualitative information that is difficult to quantify and communicate by the loan officers such as personal knowledge about the subjective circumstances of the firm, its owner, and its management. In this paper, we go beyond the current paradigm to analyze bank size and the use of different lending technologies in small business lending. Our tests allow for the possibility that large banks have a comparative advantage in lending to small businesses, including the smallest, least transparent firms, using hard-information lending technologies. We allow for the possibility that large banks use techniques such as the leasing of assets and lending based primarily on collateral values to lend to the smallest firms. We also analyze more closely the comparative advantages of small banks in using soft information to lend to small firms. Our results provide new evidence that does not always fit the predictions of the current paradigm. One of the key motivations for our paper is to understand the role of large banks in small business lending. Large banks appear to have been aggressively pursuing very small business credits using hard information-based technologies, at least before the recent financial crisis. Banking giants, such as Bank of America, were loosening their standards on small credits to 1
small businesses by relying on hard information such as owners personal credit scores (Enrich 2007). As well, recent research shows that large banks provide large amounts of funding and other services to small firms in other nations (e.g., de la Torre, Martínez Pería, and Schmuckler 2010). Our data and data from bank regulatory reports are consistent with the fact that most small business loans are made by large banks. We find in our data that banks with over $1 billion in assets make about 60% of all small business loans, similar to their share of bank branch offices. Likewise, the June 2006 Call Report shows that over 65% of the dollar value of business loans of $1 million or less and over 68% of the value of such loans of $100,000 or less were made by banks with over $1 billion in assets. Our empirical analysis matches data on U.S. small businesses, the banks that lend to them, the contract characteristics of these loans, and information from several other data sources to test the empirical implications of the current paradigm. The data include information about the loan contract, the borrower, the bank, and the bank-borrower relationship for 1811 small business loans. Using these data, we analyze the comparative advantages of large and small banks when using different lending technologies to lend to firms of different sizes. First, we empirically identify five fixed-asset lending technologies used by the banks to make small business loans. Second, we analyze the role of relationship lending in lines of credit without fixed-asset collateral. This approach is more comprehensive than prior empirical studies, which usually either identify one or two lending technologies or rely on a single measure of relationship strength using the complete set of loans. Our analysis also allows for differences in the comparative advantages for different bank sizes in lending to firms of varying size. This more general approach to studying small business lending by bank size, lending technologies, and firm 2
size yields some new and interesting findings. Under the current paradigm, large banks generally have a comparative advantage in using hard-information lending technologies also known as transactions-based lending. The reasons for this comparative advantage are discussed below. Loan officers at large banks are hypothesized to make lending decisions using lending technologies based primarily on hard information. In most cases, the research tends to lump these hard technologies together, which often originates from an assumption that hard technologies may be represented by a single technology financial statement lending which relies primarily on statistics in firms financial statements. In contrast, we allow for the possibility that large banks may not have equal advantages in all of the individual hard technologies. This implies that financial statement lending may not be representative of hard technologies as a whole. The assumption about the representativeness of financial statement lending implies that large banks comparative advantage in using hard-information lending technologies should be monotonically increasing in the size of the firm. As firms increase in size, they tend to have higher-quality financial statements, yielding an implied increasing advantage in hard technologies (see Berger and Udell 2006 for a summary of the current paradigm). However, we permit the comparative advantage of large banks to be increasing, decreasing, or nonmonotonic in firm size. If financial statement lending is not representative of hard-information lending technologies, then large banks may have differing comparative advantages across these technologies when lending to firms of different sizes. The current paradigm also predicts that small banks tend to have comparative advantages in using soft-information technologies to lend to the smallest firms. Loan officers at small banks are hypothesized to have more flexibility to evaluate credit using techniques based primarily on 3
soft qualitative information that is difficult to quantify and communicate by the loan officers such as personal knowledge about the subjective circumstances of the firm, its owner, and its management. In particular, relationship lending which is based primarily on information gathered over the course of a bank-borrower relationship, such as the owner s character or reliability is often analyzed as a soft-information technology. We take a step beyond this analysis by allowing the comparative advantage of small banks in relationship lending to be increasing, decreasing, or nonmonotonic in firm size. Our main empirical findings are: 1) Large banks appear to have different comparative advantages in each of the fixedasset lending technologies, which implies that no single hard technology is representative of all of the hard lending technologies; 2) The measured comparative advantages of large banks in hard technologies do not all appear to be monotonically increasing in firm size; and 3) Small banks appear to have a comparative advantage in relationship lending, but this advantage seems to be strongest for lending to the largest firms. All of these major results are new to the literature and conflict with the predictions of the current paradigm. The remainder of the paper is organized as follows. Section 2 reviews the literature and explains our contribution. Section 3 describes the data and our approach to analyzing the lending technologies used by banks to lend to small businesses. Section 4 shows our methodology for testing the implications of the current paradigm, and Section 5 gives the empirical results from 4
those tests. Section 6 concludes. 2. The literature and our contribution The current paradigm for small business lending concentrates mainly on two categories of lending technologies, hard- and soft-information technologies. It is often explicitly or implicitly assumed under the current paradigm that hard technologies as a whole may be represented by the financial statement lending technology alone. Based on this assumption, the conclusion is often drawn that hard technologies are best suited for serving the largest, most transparent small businesses that tend to have the highest quality financial statements. Thus, for most of the research in the current paradigm, as firms increase in size and transparency, banks tend to substitute from the use of a soft technology to one of the hard technologies. The assumptions employed about the technologies in the current paradigm may result in biased or misleading empirical results. The empirical research in most cases does not separately identify the individual hard-information technologies employed by the lending banks. Instead, researchers often focus solely on the soft technology of relationship lending (e.g., Petersen and Rajan 1994, Berger and Udell 1995, Degryse and van Cayseele 2000). This research generally uses a measure of bank-borrower relationship strength, such as relationship length or breadth, as a continuous indicator of the degree to which the relationship lending technology versus a hard technology is effectively applied. This practice effectively groups the hard-information technologies together, so any measured effect of these technologies at best reflects an overall average effect across the individual lending methods, and may not accurately measure the effects of financial statement lending or any other single hard technology. Moreover, the measured effect of hard technologies may be biased from the inadvertent 5
inclusion of the effects of soft technologies other than relationship lending. That is, the measured effect of hard technologies may also mix in the effects of soft technologies that are associated with weak banking relationships. We postulate a soft-information technology that we call judgment lending, which is lending based primarily on the judgment of a loan officer relying on experience and training, as well as any other available hard and soft information. While judgment of the loan officer is important for virtually any lending technology, it may be the principal information source for lending to some firms, such as small businesses that do not have significant hard information available and have not established a strong banking relationship. The exclusion of soft technologies such as judgment lending suggests that measured effects of relationship lending may not accurately reflect the effects of soft technologies as a whole and/or may give biased effects of relationship lending. Some recent research recognizes the possibility that financial statement lending may not represent hard technologies as a whole, and that some of the other hard technologies may be particularly useful in lending to the smallest, least transparent firms (e.g., Berger and Udell 2006). To illustrate, small business credit scoring may be applied even when there is very limited information about the overall quality of the firm, as long as the firm has a good credit score based mostly on the credit history of the owner. Similarly, fixed-asset lending can be used to extend credit when the firm has high-quality fixed assets (real estate, motor vehicles, or equipment) that may be leased or pledged as collateral, even if the small business is not sufficiently transparent based on other hard and soft information. In the Berger and Udell (2006) framework, lending technologies employed by commercial banks include the hard technologies of financial statement lending, fixed-asset lending, asset-based lending, and small business credit scoring, and the soft technology of 6
relationship lending. All lending technologies employ some combination of hard and soft information, but hard and soft technologies are defined by the principal or most critical source of information employed in the screening, underwriting, and monitoring of the credit. 1 As described above, the principal data source for financial statement lending is a firm s financial statements. For fixed-asset lending, the main data are appraised values of the real estate, motor vehicles, or equipment leased or pledged as collateral, while the key data for asset-based lending are valuations of accounts receivable and/or inventory pledged. Small business credit scoring decisions are based principally on credit scores generated from the owner s personal credit history and limited financial data on the firm. Relationship lending is based on proprietary information gathered over the course of the relationship. As noted above, judgment lending is based mainly on the judgment of a loan officer relying on experience and training. A few recent empirical studies have made progress by identifying one or two specific lending technologies, rather than simply using a measure of relationship strength to separate relationship lending from hard technologies as a whole. For example, some studies empirically identify small business credit scoring based on survey data regarding whether, when, and how U.S. banks employ this lending technology (e.g., Frame, Srinivasan, and Woosley 2001, Berger, Frame, and Miller 2005, Berger, Espinosa-Vega, Frame, and Miller 2005, forthcoming, DeYoung, Glennon, and Nigro 2008, DeYoung, Frame, Glennon, and Nigro forthcoming). These studies confirm the possibility of banks using a hard technology to expand their small business lending or improve their information sets about very small customers, depending on how the technology is implemented. Some recent studies of Japan have information on whether small businesses have certified audited financial statements, which may be an indicator that their banks 1 Underwriting any loan requires at least some numbers about the firm, the owner, and/or the collateral (hard information), and some judgment of the loan officer based on experience and training (soft information). 7
use financial statement lending on the loans to these customers (e.g., Kano, Uchida, Udell, and Watanabe 2006). This research finds, for example, that the beneficial effect of relationship length is smaller for firms with audited statements, consistent with a predicted shifting of technologies from relationship lending to financial statement lending. 2 The current paradigm also relates to the role of banks organizational structure in determining their comparative advantages in different lending technologies. Large banks are considered to have comparative advantages in hard technologies because they have economies of scale in the processing and transmission of hard information, and may be better able to quantify and diversify the portfolio risks associated with hard-information loans. Conversely, large banks may be disadvantaged in processing and transmitting soft information through the communication channels of large organizations (e.g., Stein 2002). Lending based on soft information may also be associated with agency problems within the financial institution because the loan officer is the main repository of the information, giving a comparative advantage to small institutions with fewer layers of management (e.g., Berger and Udell 2002) or less hierarchical distance between the loan officer and the manager that approves the loans (e.g., Liberti and Mian 2009). 3 In this paper, we go beyond the current paradigm in three ways. First, we allow for the possibility that large and small banks may have different comparative advantages for individual 2 One study of Japan identifies the use of six lending technologies, but takes a very different approach from the one employed here. While we focus principally on contract terms and the bank-borrower relationship, Uchida, Udell, and Yamori (2008) focus principally on the borrower s perception of how much the bank relied on different information. In their application, a loan may be made using multiple technologies. They find no significant differences in comparative advantages for large and small banks in the different technologies in Japan, which is very different from our findings below. The reasons for these different findings may be the different methodology, the application to a different country, or some combination of these. 3 One recent paper finds results suggesting that banks may use different lending technologies when lending at different distances (Jiménez, Salas, and Saurina 2009). 8
hard technologies. As explained above, all hard technologies employ some combination of both hard and soft information. Thus, for some hard-information technologies, the comparative advantage of large banks in using the hard-information component may be offset by a comparative advantage of small banks in using the soft-information component. For example, commercial real estate lending is a hard-information technology which is based mainly on the appraised value of the property. However, there may also be a significant soft-information component. Large banks may have only a slight comparative advantage in obtaining and processing the appraised values, whereas small banks may have a significant advantage in the soft-information component, based on relationships with the borrowing firms or loan officers knowledge of the market and local business conditions. This implies that large banks may not have comparative advantages in hard technologies with significant soft-information components. In our analysis, we examine whether large banks have equal comparative advantages in different fixed-asset lending technologies. Second, we allow banks comparative advantages in each lending technology to differ by the size of borrower to which it is applied. The advantages in a given technology may differ by firm size because the relative importance of the hard- and soft-information components may differ by the size of the borrowing firm. In the example of commercial real estate loans, the benefit of assessing the soft-information component related to information other than the appraised value of the property may differ with the size of the firm. Collecting soft information is costly (Diamond 1984) and may only be cost effective for larger loans when hard information is already present. For relationship lending, the access to soft information may also have more or less value depending on the size of the firm. Recent evidence on small banks suggests that the value of relationships with small firms may have changed with the increased usage of credit 9
scoring for small business loans by small institutions (Berger, Cowan, and Frame forthcoming). Third, we analyze the effects of relationships on the comparative advantages of large and small banks apart from hard information based on fixed-asset valuations. To accomplish this step in our analysis, we analyze lines of credit without fixed-asset collateral. As explained in the next section, we argue that a bank will use a hard-information lending technology over a softinformation lending technology if sufficient hard information is available. Therefore, we exclude lines of credit with fixed-asset collateral because these loans were likely underwritten on the value of the fixed assets. This is an important step forward for analyzing the comparative advantage of banks of different sizes based on the strength of a relationship between a bank and a firm. When lines of credit secured by fixed assets are included in a sample of lines of credit used to analyze relationship lending, results may be significantly biased or misleading. Our analysis provides a cleaner test of the effects of relationships on the comparative advantages of large and small banks among loans that are the most likely to involve relationship lending. 3. Data and lending technologies In this section, we describe our data on small businesses and our approach toward identifying some of the technologies used by banks to lend to these firms. Although we are not able to identify all of the technologies which are likely used by banks, we believe that our approach provides a framework for moving the current paradigm forward. By incorporating lending technologies into our analysis, we are able to evaluate the comparative advantages of large and small banks in the context of specific types of hard and soft information. Our primary data are from the 1998 Survey of Small Business Finance (SSBF). The 10
SSBF collects information on small businesses (fewer than 500 employees) in the United States. 4 Because we are interested in the size of the lending institutions as a proxy for their organizational form, we first reduce the dataset to loans from banks for which we are able to match the bank that provided the credit. 5 This involves merging the 1998 SSBF data on small businesses with the December 1997 Call Report data for banks. Although some firms have loans from other types of financial institutions, we confine attention to bank loans because banks are the only institutions that use almost all the major technologies, giving the best representation of the use of comparative advantages in the most technologies. Banks are also the most popular source of loans, and at least some banks are conveniently available to virtually every small business. We also merge the data with market characteristics taken from the 1997 Summary of Deposits data. 6 As explained below, we further reduce the dataset to a sample of fixed-asset loans (Step 1) and a sample of lines of credit without fixed-asset collateral (Step 2). We focus on these two samples because the fixed-asset loans provide a clean test of some of the hard technologies and the lines of credit without fixed-asset collateral provide a relatively clean test of one of the soft technologies. Table 1 shows descriptive statistics for these samples, including brief descriptions and the means and standard deviations of the variables used in the analysis. Column (1) gives the statistics for the sample of 1000 fixed-asset loans, which is over 40 percent of the full 4 It oversamples certain types of firms, including larger small businesses, which have been shown to use leasing less often than smaller firms (see Eisfeldt and Rampini 2009). 5 About 13% of the bank loans are excluded because the bank could not be determined. We also exclude five bank loans to firms with total assets 0 and another four for which the type of collateral could not be determined. 6 Note that we do not analyze the lending technologies that are used or would be used on borrowers whose loan applications are rejected, or are discouraged from applying because they expect to be rejected. Recent research suggests that the small businesses that are discouraged tend to be riskier than firms that apply for credit (Han, Fraser, and Storey 2009). 11
sample. Column (2) gives the statistics for the sample of 811 non-fixed-asset lines of credit. 7 Excluded from these samples are 649 loans which do not fall into either of the two categories. 8 Before considering lending technologies, we first observe the average bank size, firm characteristics, and market characteristics within each sample shown in Table 1. The mean of the large bank dummy is smaller for the sample of fixed-asset loans than for the sample of nonfixed-asset lines of credit, 55.4% and 69.7%, respectively. As for firm characteristics, the two samples have similar proportions of small, medium, and large firms around 20% of loans to small firms (total assets (TA) $100,000), and about 40% each to medium firms ($100,000 < TA $1 million), and large firms (TA > $1 million). The data on firms return on equity (ROE) are winsorized at the 10% level for both samples to reduce the influence of extreme observations. While the winsorized means for ROE of 37% and 43% are still relatively high, the corresponding medians of 14% and 17% are much lower. One reason for the relatively high ROE for small firms is that our sample only contains firms that survive and leaves out the relatively large percentage of small firms that fail and tend to have significant negative returns. 9 The market characteristics are similar across both the samples. Large banks have branch market shares between 55% and 65%, the Herfindahl concentration index averages about 0.20, and about 70% of the firms are in metropolitan competitive environments. We follow two basic principles in identifying the lending technologies. First, we argue that to evaluate each potential borrower, the bank will choose the lending technology that is most 7 Return on equity is missing for one observation in the fixed-asset loan sample and two observations in the non-fixed-asset lines of credit sample, which reduces the number of observations in the regressions using this variable. 8 These excluded loans include term loans used to purchase a fixed asset but not backed by the fixed asset itself and lines of credit backed by fixed assets. 9 Recent statistics show that only 51 percent of new employer establishments born to new firms in 2000 survived five or more years. This statistic is taken from the U.S. Small Business Administration s Office of Advocacy website http://web.sba.gov/faqs/faqindex.cfm?areaid=24. 12
efficient for that firm based on the available collateral and other information that the firm brings to the table, as well as any information that the bank already has about the firm. Based on this principle, we argue that the bank will generally choose a hard-information technology over a soft-information technology if sufficient hard information is available. Soft-information techniques tend to be labor-intensive on the part of the loan officer (high processing costs) and the information generated is difficult to communicate, so a hard-information technology will be chosen if possible. Second, we argue that lending based on the values of fixed assets that are leased or pledged as collateral is generally more efficient than other hard-information lending technologies if this collateral is available. Fixed assets are long-lived assets that are not sold in the normal course of business (i.e., are immovable ), and are uniquely identified by a serial number or a deed. These include real estate, motor vehicles, and equipment. The value of fixed assets meets the definition of hard information as quantitative data that may be credibly transmitted by the loan officer. We argue that fixed-asset lending is more efficient than other hard technologies because a bank with a loan secured by fixed assets can usually collect most of its owed repayment with higher priority before other creditors in the event of default or bankruptcy. Fixed-asset lending may also be particularly effective by providing a strong incentive for firms to make their payments in many cases, the businesses may be crippled without access to their real estate, motor vehicles, or equipment. The threat of removal of fixed assets may therefore provide a powerful incentive for borrows to repay their loans. Following these two principles, we believe that a bank first evaluates whether there are fixed assets to lease or pledge as collateral. Therefore, loans which are secured by a fixed asset are identified as having been underwritten using a fixed-asset lending technology. We exclude 13
lines of credit from this sample in order to reduce the possibility of mixing cases of relationship lending into our analysis of fixed-asset lending. Based on this approach, we are reasonably certain of the identification of the technology used for over 40 percent of the loans term loans using valuations of fixed assets leased or pledged as collateral. Researchers often focus on the comparative advantages of large and small banks without specifying the lending technology, so this may offer a first opportunity to examine banks comparative advantages using a more detailed breakout of the fixed-asset technologies. In Step 1, we analyze fixed-asset technologies. To identify these technologies, we use the contract type (lease versus loan) and the type of collateral pledged. The fixed-asset lending technologies include leasing (LEASE) as well as loans with fixed assets pledged as collateral commercial real estate lending (CRE), residential real estate lending (RRE), motor vehicle lending (MV), and equipment lending (EQ). For convenience, we refer to leasing as a lending technology, although the asset is directly owned by the financing institution, and no loan is issued. However, a lease is similar to a loan with an almost perfect collateral lien the bank owns the asset and can sell it or lease it to another customer if the loan is not repaid. 10,11 We consider leasing to be a fixed-asset lending technology, because the leased assets are generally fixed. We also note that residential real estate collateral is somewhat different from the other fixed assets in that residential real estate is generally outside collateral (owned outside the firm), whereas the other types are generally inside collateral (owned by the firm). The collateral literature has found different results when using outside collateral versus inside collateral (e.g., 10 We acknowledge that there are other differences between leasing and lending against fixed assets such as real estate, motor vehicles, and equipment. For example, under leasing, the leasee/borrower can usually deduct the payment from taxable income, and the lessor can claim depreciation, whereas when a secured loan is made, the borrower deducts the interest and claims the depreciation. 11 The argument that leasing provides a more perfect lien than pledging collateral on fixed assets is not new. Eisfeldt and Rampini (2009) argue that the benefit of leasing is that repossession of a leased asset is easier than foreclosure on the collateral of a secured loan. 14
John, Lynch, and Puri 2003, Brick and Palia 2007). Among the fixed-asset technologies, we first identify the use of leasing (LEASE), which is simply based on whether the contract type is a lease. For leases, the principal source of information used to evaluate the credit is the valuation of the real estate, motor vehicles, or equipment that is leased. We hypothesize that large banks have the strongest comparative advantage in leasing, because the almost perfect collateral lien against the assets means that the bank has to rely very little on any secondary sources of hard or soft information. The remaining fixed-asset technologies commercial real estate lending (CRE), residential real estate lending (RRE), motor vehicle lending (MV), and equipment lending (EQ) are identified by loan purpose and collateral. For instance, if the purpose of the loan is to purchase specific motor vehicles and those motor vehicles are pledged as collateral, then we identify the lending technology as MV, whether or not another type of collateral is pledged. The same is true for real estate loans secured by the real estate being purchased (CRE and RRE) and equipment loans secured by the equipment being purchased (EQ). Because these fixed-asset technologies have a less perfect lien than in the case of leasing, they may depend more on a softinformation component in addition to the hard information about the value of the fixed asset. For instance, knowledge of local neighborhood business conditions may have value in underwriting a commercial real estate loan. As shown in column (1) of Table 1, we identify 11.6%, 21.4%, 7.9%, 36.3%, and 22.8% of the fixed-asset loans as LEASE, CRE, RRE, MV, and EQ loans, respectively. The five fixedasset technology dummy variables cover the full sample of fixed-asset loans, so the means of these technologies sum to 1.00. In total, we identify 41% of all bank small business loans as using fixed-asset lending technologies. 15
In Step 2, we analyze lines of credit (LCs) which do not have fixed assets as collateral. Berger and Udell (1995) argue that lines of credit are ideally suited for relationship lending and support this by showing that small firms are more likely to have all of their LCs consolidated at a single lender than other types of loans. The principal source of information in relationship lending is the loan officer s processing of information through contact over time with the firm, its owner, and others in the local community that may be suppliers or customers, and so forth. The information is primarily soft because such information cannot be easily reduced to hard numbers that can be easily communicated by the loan officer. We only analyze LCs without fixed assets as collateral because this eliminates one of the sources of hard information which banks might use in lending. Specifically, we exclude LCs collateralized by real estate, equipment, or a motor vehicle. As we explained in our principles for identifying lending technologies, a bank will generally choose a hard-information technology over a soft-information technology if sufficient hard information is available. In this case, even for lines of credit, the banks would likely use the valuations of the fixed assets as the primary source of information. Based on this principle, the exclusion of LCs secured by fixed-asset collateral can be an important first step for avoiding biased and misleading results in an analysis of the effect of relationship strength. With this exclusion, we can compare the advantages of banks which have strong bank-borrower relationships relative to banks that do not have strong relationships. The absence of a strong relationship suggests that the bank may be relying on technologies other than relationship lending, such as financial statement lending, small business credit scoring, asset-based lending, or judgment lending. Using the sample of lines of credit without fixed-asset collateral, we test for the comparative advantages of large and small banks based on an overall measure of relationship 16
strength as well as individual relationship characteristics. Although we cannot identify relationship lending directly because the bank may be relying on other sources of hard and soft information, we rely on the strength of the relationship as an indicator of the importance of the relationship as a source of information. For instance, lines of credit underwritten using small business credit scoring are not as likely to have long relationships. Therefore, our approach to using relationship strength will help identify the comparative advantage of these banks in relationship lending. We first quantify relationship strength based on an overall indicator of a strong relationship. To measure whether a relationship is strong, we combine several measures of the length, breadth, and exclusivity of the relationship between the firm and the bank extending the loan. Most measures of relationship strength in the literature focus on length, as longer relationships allow more time for the lending bank to garner proprietary soft information about the firm (e.g., Petersen and Rajan 1994, Berger and Udell 1995). Some strength measures include breadth in the form of a checking account through which the bank may gain information from monitoring the firm s cash flows (e.g., Mester, Nakamura, and Renault 2007). Others focus on lender exclusivity, the accumulation of all of a firm s credits in a single bank, which may maximize the information advantage of that bank (e.g. Berger, Klapper, and Udell 2001, Berger, Miller, Petersen, Rajan, and Stein 2005). For a strong relationship, we require that the firm s relationship with the bank have a sufficient combination of length, breadth, and exclusivity. If the relationship is long over 10 years we require either breadth or exclusivity or both, which means that the bank must either have a checking account of the firm and/or serve as the firm s exclusive lender. If the relationship length is medium over 5 years and up to 10 years we require that the bank have 17
both breadth and exclusivity, such that the bank must have a checking account of the firm and be the exclusive lender. If the relationship is short less than 5 years or there is neither breadth nor exclusivity, the firm does not have a strong relationship with the bank. As shown in column (2) of Table 1, 40.7% of the non-fixed-asset lines of credit are to firms with strong relationships with their lending banks by our definition. The category of loans without a strong relationship likely includes several types of lending technologies. As already explained, we focus in this part of our analysis solely on lines of credit and we have eliminated lines of credit which use fixed assets as collateral. However, these LCs may still have accounts receivable/inventory pledged as collateral. Additionally, the loan could be based on financial statements, model-based credit scores, or non-relationshipbased soft information. Therefore, the excluded group of loans likely contains several types of lending technologies, including asset-based lending, financial statement lending, small business credit scoring, and judgment lending. The identification of these other technologies is not pursued here because their characteristics are not as easily observed. Column (2) of Table 1 shows the descriptive statistics for the individual relationship characteristics used to measure a strong relationship. The mean of each relationship characteristic indicates the average relationship strength based on that measure. To check our results for the strong relationship indicator, we use the same empirical model to evaluate the direct role of the individual relationship characteristics. With respect to relationship length, the mean of the log length of relationships in months is 4.202 (the average relationship length is 105 months). With respect to relationship breadth, 87.7% of the firms also have a checking account at the bank, and for 42.3%, the bank is the firm s exclusive lender, providing all of the firm s loans (from either bank or nonbank financial institutions). 18
4. Methodology Our empirical methodology is designed to test the empirical predictions of the current paradigm regarding whether large versus small banks have comparative advantages in the different lending technologies, and how these advantages differ by firm size. In this section, we briefly describe the general model used in our empirical tests and how we change the specification to test different hypotheses when different subsets of the small business loans are included. Our first specification is an analysis of the lending technologies associated with fixedasset loans. The five fixed-asset technologies include leasing (LEASE), commercial real estate lending (CRE), residential real estate lending (RRE), motor vehicle lending (MV), and equipment lending (EQ). In this specification, we model the probability that a given bank loan is made by a large bank as a function of firm size, lending technology, interactions of firm size and technology, and control variables for firm profitability, banking market competitive conditions, and firm industry. We interpret a significantly higher probability of a loan being made by a large bank, conditional on competitive conditions, as evidence of a comparative advantage for large banks. The general specification is a logit equation of the following form: (1) ln [P(loan is from a large bank) / (1 - P(loan is from a large bank))] = f(firm size, lending technology, firm size lending technology, firm ROE, large-bank branch market share, bank market concentration, MSA dummy, industry dummies) 19
where P( ) indicates probability, loan is from a large bank is a dummy variable that is one if the loan is made by a large bank and zero if it is made by a small bank. 12 The key exogenous variables are dummies for firm size class, lending technology employed, and their interactions, denoted by firm size lending technology. These dummies allow for tests of whether large or small banks have net comparative advantages in lending to different firm sizes, and in using the different technologies. The interaction terms are particularly important here because some of the empirical predictions of the current paradigm concern how the comparative advantages in the lending technologies vary with firm size class. The remaining variables in equation (1) control for firm profitability, local market competitive conditions, and firm industry. We include return on equity (ROE) as a measure of firm profitability to control for the possibility that profitability would affect the bank from which the firm borrows. ROE is generally hard information, and so it is expected that large banks would more often lend to firms with high ROE relative to small banks. Consistent with prior research and anti-trust guidelines, we define the firm s local banking market as the Metropolitan Statistical Area (MSA) or non-msa rural county in which the small business is located. 13 The market conditions specified are the large bank market share of branches, the Herfindahl concentration index of market bank deposits, and an MSA indicator dummy. It is important to control for the large-bank branch share because this accounts for the relative presence of large 12 In some cases, multiple bank loans to the same firm are included and may be made using different lending technologies. 13 In some cases, we use New England County Metropolitan Areas (NECMAs), but for convenience, we simply use the term MSA to cover both MSAs and NECMAs. 20
and small banks. 14 Prior research shows that local market share of large banks is a powerful predictor of lending bank size (e.g., Berger, Rosen, and Udell 2007), which suggests that firms may generally choose an institution based on convenience. The Herfindahl index is the most standard measure of market power used in bank research and anti-trust analysis, and the MSA dummy proxies for the generally greater degree of competition in metropolitan markets. The effect of market concentration may be either favorable or unfavorable for small business borrowers (e.g., see Scott and Dunkelberg forthcoming). We also add 8 industry dummies for the industry of the borrower to control for differences in transparency, tangibility of assets, and loan types across industries. In the interest of brevity, the industry dummies are not shown in the tables and their results are not discussed. The regression analysis of the five fixed-asset technologies allows us to test the prediction of the current paradigm that large banks have equal comparative advantages in all of these hard technologies versus our hypothesis above that large banks are likely to have the strongest comparative advantage in leasing. As explained above, we hypothesize that large banks have the strongest comparative advantage in leasing because of the almost perfect collateral lien against the assets. If we find different comparative advantages for individual fixed-asset technologies, this suggests that hard technologies as a whole may not be well represented by financial statement lending. In our analysis of the fixed-asset loans, we also test whether the comparative advantages of large banks in these hard technologies are all monotonically increasing in firm size. Under the current paradigm, the comparative advantages of large banks in hard technologies are predicted to increase monotonically for all hard technologies. Our analysis extends beyond the current 14 The large-bank branch market share variable plays a similar role here to the log of median bank size in the market in the model of the relationship between bank size and firm size in Berger, Miller, Petersen, Rajan, and Stein (2005) controlling for the local availability of large versus small banks. 21
paradigm by allowing for the possibility that large banks comparative advantages in the fixedasset technologies may depend differentially on the size of the borrower. As noted above, some of the recent literature recognizes the possibility that some of the hard technologies other than financial statement lending may be particularly useful in lending to the smallest, least transparent firms, but this has rarely been applied in practice. Our second specification is an analysis of lines of credit without fixed-asset collateral. The sample for this specification includes loans underwritten using relationship lending as well as financial statement lending, small business credit scoring, asset-based lending, and judgment lending. Our exclusion of fixed-asset-secured lines of credit allows us to rule out fixed-asset technologies, which is a step forward from previous empirical work. However, we cannot identify relationship lending directly from the data because the remaining lines of credit may have been underwritten based on one of the other technologies. Therefore, we focus our analysis specifically on measured relationship strength. In this specification, we model the probability that a given bank loan is made by a large bank as a function of firm size, relationship strength, interactions of firm size and relationship strength, and the same control variables as in equation (1): (2) ln [P(loan is from a large bank) / (1 - P(loan is from a large bank))] = g(firm size, relationship strength, firm size relationship strength, firm ROE, large-bank branch market share, bank market concentration, MSA dummy, industry dummies) In the regression shown in equation (2), we first analyze relationship strength using our 22
index measure of Strong Relationship. This allows us to test the predictions of the current paradigm that small banks have a comparative advantage in relationship lending and that their comparative advantage should be greatest for the smallest firms. We also take a step further to analyze the underlying relationship characteristics used to define a relationship loan for the sample of non-fixed-asset lines of credit. In this specification, we use the same model of the probability that a given bank loan is made by a large bank, but we replace the strong relationship indicator variable with the three separate characteristics: relationship length, checking account, and exclusive lender. This specification allows us to identify the characteristics that generate a comparative advantage in relationship lending based on the strength of the relationship. 5. Empirical results Tables 2 through 4 show the regression results. In Table 2 Panel A, we show four columns based on the specification in equation (1) one with firm size dummies only, one with technology dummies only, one with both sets of dummies, and one complete specification with all the dummies and the interaction terms. Tables 3 and 4 Panels A also show four columns, but replace lending technologies with relationship strength, as specified in equation (2). All estimations include the control variables for banking market characteristics and firm industry. The control variable for firm profitability is included as an additional firm characteristic in all estimations that include the firm size dummies. For the complete specifications in column (4), we also show the predicted probabilities for each firm size when combined with either a lending technology or relationship-strength indicator, as explained below. Instead of presenting the logit coefficients, we report odds ratios which are obtained by exponentiating the original coefficients. When a variable has an odds ratio that is greater than (is 23
less than) 1, a higher level of the variable is associated with higher (lower) odds of the loan being from a large bank. We also report the absolute value of robust z-statistics based on the Huber- White sandwich method to correct for heteroskedasticity. We first turn to Table 2 Panel A, which shows the results of testing the five fixed-asset technologies. We exclude the small-firm dummy and the LEASE dummy as the base case and conduct most of our tests on the comparative advantage differences of large and small banks for the four collateral-based fixed-asset technologies versus leasing. As discussed above, we hypothesize that large banks have the strongest comparative advantage for LEASE because of the almost perfect collateral lien against the leased assets. This characteristic of leasing makes it the purest hard information technology with the least requirement for secondary sources of soft information for which small banks may have the advantage. Column (1) of Table 2 Panel A shows the logit regression with firm size and control variables only. This version of the model is the most similar to prior empirical research, which focuses on the effects of firm size without identifying or separating out the effects of the different technologies. One key difference here is that we include three firm size classes and allow for a nonmonotonic relationship between firm size and bank size, rather than using a continuous measure of firm size that forces a monotonic relationship. Our goal is to allow for the possibility that large banks may tailor specific fixed-asset technologies to reach different firm sizes. Under the current paradigm employed in much of the small business finance literature, the odds ratios on the firm size class dummies are expected to be greater than 1 and increasing. Larger firms tend to be more informationally transparent, and therefore better served by large 24