How important are small banks to small business lending?

How important are small banks to small business lending?

Journal of Banking & Finance 23 (1999) 427±458 How important are small banks to small business lending? New evidence from a survey of small ®rms Jith...

183KB Sizes 5 Downloads 112 Views

Journal of Banking & Finance 23 (1999) 427±458

How important are small banks to small business lending? New evidence from a survey of small ®rms Jith Jayaratne a, John Wolken

b,*

a

b

The Tilden Group, 5335 College Avenue, Oakland, CA 94618, USA Board of Governors of the Federal Reserve System, Financial Structure Section, Division of Research and Statistics, Washington, DC 20551, USA

Abstract Typically, small banks lend a larger proportion of their assets to small businesses than do large banks. The recent wave of bank mergers has thinned the ranks of small banks, raising the concern that small ®rms may ®nd it dicult to access bank credit. However, bank consolidation will reduce small business credit only if small banks enjoy an advantage in lending to small businesses. We test the existence of a small bank cost advantage in small business lending by conducting the following simple test: If such advantages exist, then we should observe small businesses in areas with few small banks to have less bank credit. Using data on small business borrowers from the 1993 National Survey of Small Business Finance, we ®nd that the probability of a small ®rm having a line of credit from a bank does not decrease in the long run when there are fewer small banks in the area, although short-run disruptions may occur. Nor do we ®nd that ®rms in areas with few small banks are any more likely to repay trade credit late, suggesting that such ®rms are no more credit constrained than ®rms in areas with many small banks. Ó 1999 Elsevier Science B.V. All rights reserved. JEL classi®cation: G21; G32 Keywords: Small business lending; Bank mergers

*

Corresponding author. Tel.: +1 202-452-2503; fax: +1 202-728-5838; e-mail: [email protected]

0378-4266/99/$ ± see front matter Ó 1999 Elsevier Science B.V. All rights reserved. PII: S 0 3 7 8 - 4 2 6 6 ( 9 8 ) 0 0 0 8 5 - 5

428

J. Jayaratne, J. Wolken / Journal of Banking & Finance 23 (1999) 427±458

1. Introduction The recent wave of bank mergers ± and the resulting decrease in the presence of small banks ± has raised the concern that bank consolidation may reduce the availability of credit to small businesses. 1 This concern is justi®ed, however, only if small banks enjoy a cost advantage over large banks in lending to small businesses. Cost advantages for small business loans may arise in connection with loan origination and monitoring costs, and consequently smaller banks may be more likely to lend to small ®rms than large banks. In this paper, we test for the ``Small Bank Advantage'' hypothesis employing data on the use of credit obtained from small business owners. Until now, studies of the e€ects of bank consolidation on small business lending have relied exclusively on loan data from banks, producing mixed results. In this study, we use data from the 1993 National Survey of Small Business Finances. These data are obtained from small businesses ± the borrowers. Focusing on borrower data has several advantages over data obtained from banks. In particular, the data set contains information on the ®rm and owner's credit quality and other characteristics as well as a complete inventory of ®nancial services used by that ®rm. If a ``Small Bank Advantage'' exists, then we should observe small businesses in areas with few small banks to have less bank credit. We test this implication in two ways. First, we examine the e€ect of small bank presence (measured, for example, as the fraction of deposits in a Metropolitan Statistical Area (MSA) controlled by banks with assets under $300 million) on the probability of a small business having a line of credit (a form of credit with few non-bank providers). Second, we examine the e€ect of small bank presence on whether a small business repaid trade credit after the due date (an indicator of credit constraints faced by ®rms since trade credit is potentially an expensive source of credit). We also test the Small Bank Advantage hypothesis by testing a second implication of that conjecture: If small banks enjoy a cost advantage in originating loans that require signi®cant monitoring (such as small business loans), and if marginally creditworthy small ®rms require especially intensive scrutiny, then we should observe marginal ®rms borrowing from small banks more often than non-marginal small ®rms. The results generally do not support the ``Small Bank Advantage'' hypothesis. We ®nd that the probability of a small ®rm having a line of credit from a bank does not decrease in the long run when there are fewer small banks in the area, although short-run disruptions may occur. Nor do we ®nd that

1

Texas legislators cited this concern in 1995 when they voted to opt out of the provisions of the 1994 Riegle±Neal Act ± which allowed full interstate banking ± thereby limiting the bank consolidation that could have resulted from more out-of-state banks entering Texas.

J. Jayaratne, J. Wolken / Journal of Banking & Finance 23 (1999) 427±458

429

®rms in areas with few small banks are any more likely to repay trade credit late, suggesting that such ®rms are no more credit constrained than ®rms in areas with many small banks. Even when we examine marginally creditworthy ®rms separately, we ®nd that small bank presence a€ects neither their use of bank credit lines nor the probability of late repayment of trade credit. Finally, we ®nd that marginal ®rms ± small ®rms with poor or short credit histories ± are as likely to borrow from small banks as are non-marginal ®rms. The unimportance of small banks appears to be the result of large banksÕ willingness to lend to small borrowers, suggesting that bank consolidation will have little long-run e€ect on credit availability to these ®rms. 2. Background Large banks make proportionately fewer small business loans than small banks. This is the most common statistical evidence cited to support the claim that bank consolidation will reduce credit availability to small businesses. Banks with less than $100 million in assets have nearly 9 percent of their assets in small business loans; banks with more than $5 billion in assets have only 3 percent of their assets in loans to small ®rms (Table 1). Similar results are reported by Keeton (1995) who ®nds that banks aliated with out-of-state bank holding companies (BHCs) originate fewer small business loans than independent banks. 2 Based on these di€erences in lending patterns between banks of di€erent size and complexity, several observers have inferred that consolidation that reduces the number of independent small banks will also reduce small business lending. For example, Berger et al. (1995) predict that bank consolidation resulting from full nationwide banking would reduce small business lending by 32 percent in just ®ve years. This inference is plausible ± and bank consolidation will reduce small business lending ± if small banks are able to lend to small businesses at a lower cost than large banks (and this explains in part why small banks lend proportionately more to small businesses, as seen in Table 1). Why might this be true? Perhaps small business loans are mostly ``relationship loans'' that require substantial individual monitoring and nurturing by a bankÕs loan ocers. Consequently, such lending may require individual loan ocers to have considerable autonomy to set underwriting standards and discretion to monitor and evaluate. But autonomy to loan ocers might be abused, and banks may have to pocket the cost of poor loan decisions by its loan ocers. Small banks

2

However, Whalen (1995) ± using a di€erent sample of banks from Keeton (1995) ± ®nds that banks owned by out-of-state holding companies did not lend any less to small ®rms than other banks.

430

J. Jayaratne, J. Wolken / Journal of Banking & Finance 23 (1999) 427±458

Table 1 Portfolio shares of small C&I loans, by bank size Banks by asset size

Small C&I loans/total C&I loans (percent)

Small C&I loans/total assets (percent)

Less than $100 million $100 million±$300 million $300 million±$1 billion $1 billion±$5 billion Greater than $5 billion

96.7 85.2 63.2 37.8 16.9

8.9 8.8 6.9 4.9 2.9

Source: P. Strahan and J. Weston, 1996, Small Business Lending and Bank Consolidation: Is There Cause for Concern?, Federal Reserve Bank of New York, Current Issues in Economics and Finance 2. Based on the June 1995 Reports of Condition and Income.

may contain such costs because their small size allows upper management to monitor their loan departments fairly easily. However, large banks, given their size and complexity, may ®nd it more costly to monitor their agents. If bigger and more complex banking organizations try to contain these costs by setting rigid underwriting standards, then large banks will ®nd it more dicult to originate and maintain the relationship loans required by small businesses. We shall call this conjectured disadvantage of large banks the ``Small Bank Advantage'' hypothesis. 3 Existing tests of this conjecture have used loan data from banksÕ balance sheets and have produced mixed results. One test of the Small Bank Advantage hypothesis is to see what happens to small business lending by a bank after it is acquired by another bank or by a bank holding company. If larger banks su€er from higher costs of making relationship loans (as the Small Bank Advantage hypothesis claims), then the new bank formed by the merger should originate fewer small business loans after the merger. Consistent with this prediction, Berger et al. (1998) ®nd that after a merger, the new bank originates fewer small business loans than the independent banks prior to the merger. Similarly, Peek and Rosengren (1997) conclude that mergers reduced small business lending except when small banks bought other small banks ± then small business lending actually increased. Keeton (1997) ®nds that small business lending by small urban banks decreased when they were acquired by out-of3 This conjectured cost disadvantage of large banks when making small business loans (and it is only a conjecture) is still not sucient to conclude that bank consolidation will reduce the supply of credit to small businesses. Even if large banks are at a disadvantage in this loan market, small banks (existing or new entrants) may pick up the excess demand and lend to small businesses denied credit because of bank consolidation. If the supply of loans by small banks is elastic, then consolidation will have little e€ect on the supply of credit to small ®rms. However, small banks may ®nd it hard to raise the extra funds needed to ®nance additional lending or entry barriers (regulatory or otherwise) may limit the number of small banks that may enter a market to lend to small businesses cut out by large banks.

J. Jayaratne, J. Wolken / Journal of Banking & Finance 23 (1999) 427±458

431

state holding companies or when they were bought by in-state organizations that held these small banks as junior aliates; other types of acquisitions did not reduce lending to small ®rms. In contrast, Strahan and Weston (1998) conclude that bank mergers actually increased such lending ± at least for those mergers where both participants were small (other types of mergers appear to have no e€ect). 3. Data and empirical model Data on small business ®nance are scarce. One of the few available sources is the National Survey of Small Business Finance (NSSBF), a nationally representative sample of non-®nancial, non-farm small businesses sponsored by the Board of Governors of the Federal Reserve System and the US Small Business Administration. We use data from the December 1993 Survey, which was conducted in 1994/95 and which records ®nancial information for 1993. The sample was strati®ed by census region, urban/rural location and employment size (500-employee ®rms being the largest ®rms sampled), and minority partition. 4 The sample for this paper includes 4630 ®rms, of which 3697 ®rms were located in MSAs and 933 were located in non-MSA rural counties. The NSSBF is a rich source of information on the current ®nancial condition of small businesses as well as on the ®nancial history of these ®rms. Firms report their obligations by type of ®nancial institution as well as non-institutional funding sources. In this paper, we rely primarily on those parts of the survey that provide information on the lines of credit and on trade credit from suppliers. Using these data, we test two implications of the Small Bank Advantage conjecture: First, small ®rms in areas with few small banks should be more credit constrained, and second, marginally creditworthy small ®rms should be more likely to obtain credit from small banks than non-marginal ®rms. 5 To test these two implications, we conduct a total of three tests. The ®rst examines the use of bank lines of credit. 6 The second examines the repayment

4 For additional information on the survey methodology, see National Survey of Small Business Finances, ``Methodology Report'', mimeo., March 1996. (Available through the Federal Reserve BoardÕs public web site.) 5 Note that we do not argue that the Small Bank Advantage hypothesis implies that marginal ®rms rely on small banks more often than on large banks (i.e., more than 50 percent of marginal ®rms borrow from small banks). This may not follow from the Small Bank Advantage conjecture simply because marginal ®rms may not have enough small banks that they can turn to. If most bank funds in a market are controlled by large banks, most marginal ®rms may resort to large banks even if small banks have a cost advantage in lending to such ®rms. 6 Commercial banks and thrifts are included in our working de®nition of ``bank'' in this paper.

432

J. Jayaratne, J. Wolken / Journal of Banking & Finance 23 (1999) 427±458

of trade credit after the due date, a proxy for excess demand for institutional credit. Next, we re®ne these two tests and examine whether small bank presence has a stronger e€ect on marginal (smaller, younger, poorer credit history) ®rms. And the third test examines whether marginal ®rms are more likely than non-marginal ®rms to obtain credit from small banks. Before describing these three tests in more detail, we discuss the measure of small bank presence. De®ning small bank presence. In estimating the empirical model above, the measurement of small bank presence (``SMALL'') is debatable. The de®nition of what constitutes a ``small bank,'' and the relative size of small banks vis-a-vis the market in which they operate are both subject to interpretation. We test the robustness of our results by using four de®nitions of SMALL. In all four measures of SMALL, we assume that the MSA is the relevant geographic area for urban banking markets and the non-MSA county for rural markets (an assumption routinely made by antitrust regulators). In our ®rst de®nition, SMALL was measured as the fraction of banking market deposits held by banks that had fewer than $300 million in assets as of June 1993. To construct this measure, we ®rst identi®ed all banks (and thrifts) in a banking market that had assets less than $300 million ± these are our ``small banks''. Next, we calculated the amount of deposits held at branches of such small banks in that banking market and divided that number by the total deposits held at all banks in that market to generate ``SMALL''. 7 We used the $300 million cuto€ because banks with more than $300 million in assets appear to lend to small businesses sharply less than smaller banks (Table 1). Our second de®nition of SMALL allows for the possibility that even ``small'' banks aliated with bank holding companies may behave like large banks (as noted by Keeton, 1995), and we de®ne ``small banks'' as only those banks that have less than $300 million in assets and that are unaliated with a large holding company (de®ned as holding companies with assets greater than $300 million as of June 1993). 8 The remainder of the construction of SMALL is as before. Our third de®nition of SMALL relaxes an implicit assumption in our previous de®nitions of this variable, i.e., that a bankÕs deposits in an MSA proxies for the resources available to that bank to originate small business loans in that MSA (although this assumption is routinely made by bank regulators and the Department of Justice in their antitrust review of bank mergers). The third

7

Bank asset data are from the Quarterly Reports of Condition ®led by all banks, and branch deposit data are from the Summary of Deposits database created by the Federal Deposit Insurance Corporation. 8 Bank ownership information are from the Federal Reserve BoardÕs National Information Center database.

J. Jayaratne, J. Wolken / Journal of Banking & Finance 23 (1999) 427±458

433

measure of SMALL is simply the fraction of banks in a banking market that have assets less than $300 million. Our fourth de®nition of SMALL is the fraction of banks in a banking market that have assets less than $300 million and that are unaliated with a large BHC. 9 Test 1. Lines of credit and small bank presence. Our ®rst test measures access to bank credit in terms of bank lines of credit. We focus here on lines of credit (and not on other types of bank lending) partly because few non-banks provide a similar service. Only 1.5 percent of small businesses in the NSSBF obtain a line of credit from a source other than a bank, while 23.5 percent have credit lines from banks. In contrast, other types of bank lending ± such as mortgage loans, equipment loans and vehicle loans ± are also provided by non-banks, so that even if consolidation reduces mortgage lending by banks, borrowers may go to mortgage ®nance companies and others for such credit. Another reason to focus on credit lines is that ®rms in the NSSBF used credit lines more often than any other type of credit. Twenty-®ve percent of ®rms in the sample had credit lines, while only 6 percent had mortgage loans and 14 percent had equipment loans. 10 We estimate the following Logit model using ®rm-level data from the NSSBF to see if small borrowers ®nd it more dicult to access bank lines of credit in areas with few small banks: Probability …Firm j has a bank line of credit† ˆ f …SMALL; change in SMALL; firm j0 s creditworthiness; demand for credit†:

…1†

Variable de®nitions are summarized in Appendix A. SMALL is a measure of small bank presence in the MSA or non-MSA county where ®rm j is located. We use several measures of small bank presence discussed earlier including ± for example ± the fraction of total deposits in a MSA held by banks with assets under $300 million. CHANGE IN SMALL is measured as the change in SMALL over the ®ve years prior to 1993 (divided by the average value of SMALL over the period). As indicators of a ®rmÕs creditworthiness, we use

9 As a further test of the robustness of our results, we also de®ned SMALL using a $100 million asset cuto€, and we found similar results as those reported here using the $300 million asset cuto€. 10 These fractions do not agree with those in Table 2 because these are based on weighted data whereas Table 2 relies on unweighted data. The weights are used to adjust for the fact that the NSSBF over-sampled certain categories of ®rms (especially large ®rms and minority-owned ®rms). The weighting process attaches a lower weight to observations of ®rms that belong to over-sampled categories, and thereby recreates a sample that is representative of the population of small ®rms in the US (The regressions in this paper approximate the same procedure by including controls for sampling strata such as minority ownership, ®rm size and census region [although several of these are not included in the tables with regression results]).

434

J. Jayaratne, J. Wolken / Journal of Banking & Finance 23 (1999) 427±458

variables for personal delinquency, business delinquency, bankruptcy, and a ®rmÕs age. 11 A ®rmÕs demand for credit is proxied by its pro®tability, ®rm size, by the growth rate of the stateÕs economy in 1993 and the growth rate of total bank lending in that state in 1993. 12 We also included the Her®ndahl±Hirschmann Index (HHI) of deposit concentration in a banking market in the above model. This controls for competitive conditions in the ®rmÕs banking market. 13 We interpret SMALL to capture the long-run e€ects of small bank presence on credit supply. We include the CHANGE IN SMALL to identify short-run e€ects of a diminished small bank presence in an area. Changes in small bank presence may produce only short-run disruptions in credit supply when ± for example ± loan ocers are laid o€ or underwriting criteria are revised after bank mergers and current borrowers switch to alternative lenders; this adjustment process may take some time to be completed. 14 A positive coecient on SMALL (or CHANGE IN SMALL) would con®rm the conjecture that small businesses in areas with few small banks ®nd it relatively dicult to access bank lines of credit. Bank consolidation is likely to reduce small business credit. A zero coecient on SMALL suggests that a

11

The NSSBF records whether a ®rmÕs owner has been delinquent in his/her personal or business obligations within the last three years or declared bankruptcy within the seven years prior to the end of 1993. We include ®rm age as a control variable because moral hazard problems such as risk shifting are likely to be greater for younger ®rms ± as such, younger ®rms will ®nd it more costly to borrow (Diamond, 1989). 12 Firms in the NSSBF report the city and state where they are located. Although this information is not available in the public access version of the NSSBF, location information was made available to us by the Board of Governors. 13 The standard argument for including the HHI is that the supply of credit is likely to be constrained in less competitive markets. Alternatively, Petersen and Rajan (1995b) argue that small businesses in more concentrated banking markets ®nd it easier to get bank credit because market power makes loan contracts easier to enforce. For a more complete explanation of this argument, see Petersen and Rajan (1995b) who ®nd evidence consistent with this result using the 1987 NSSBF. For our purposes here, omitting the HHI could produce biased estimates of the e€ects of SMALL because the HHI and SMALL are inversely correlated. 14 Another reason to include CHANGE IN SMALL is more complicated. If the Small Bank Advantage hypothesis is correct, then small banks may follow demand and locate in areas with many small businesses (and small ®rms may follow supply and locate in areas with many small banks). In frictionless markets for small business lending, this process will equalize loan rates across geographic markets. Firms in areas with many small banks will not have lower funding costs than areas with few small banks. The coecient of SMALL will be zero ± even though the Small Bank Advantage hypothesis is true. But this argument assumes that banks and ®rms can relocate quickly and at low cost. If such relocation costs are signi®cant (or if regulatory and other entry barriers produce an inelastic supply of banks), then the adjustment process may slow and markets need not clear instantaneously, producing positive coecients on CHANGE IN SMALL and (perhaps) SMALL.

J. Jayaratne, J. Wolken / Journal of Banking & Finance 23 (1999) 427±458

435

decline in the presence of small banks does not a€ect the supply of bank lines of credit to small ®rms. 15 Test 2. Repaying trade credit after the due date and small bank presence. Lines of credit are only one type of bank credit. As a somewhat more general indicator of access to institutional ®nance, we look at the repayment of trade credit after the due date ± the focus of our second test. This test starts from the premise that trade credit is an expensive source of liquidity, used as a last resort by most ®rms. Trade credit is typically provided by suppliers as an interest-free loan up to a due date (often 30 days). Payment by that date will incur no interest. But payment after the due date incurs an extremely high interest rate and may constitute a very expensive loan. 16 Firms use trade credit to facilitate obtaining supplies and paying for these supplies at di€erent points in time. However, ®rms may as a last resort ``borrow'' from their trade suppliers (i.e., pay after the due date) when they have trouble accessing institutional credit. 17 As such, repaying trade credit after the due date is a summary indicator of access to institutional credit. If small businesses in areas with few small banks are more likely to repay after the due date, we conclude that such borrowers ®nd it more dicult to access bank credit of all types. We implement this second test by estimating the following model: 18 Probability …Firm j repays its suppliers after the due date† ˆ f …SMALL; CHANGE IN SMALL; firm j0 s creditworthiness; demand for credit; late payment penalty†:

…2†

The dependent variable takes on multiple values since the NSSBF asks ®rms whether they paid (1) none, (2) less than half, (3) about half, (4) more than half, or (5) almost all or all of their trade credit after the due date. Consequently, we estimate Eq. (2) as an Ordered Logit.

15 Other possible reasons for why SMALL may have a non-signi®cant coecient are discussed in Section 5. 16 Although trade credit is generally granted for a period (e.g., 30 days) without an explicit interest charge, most trade credit accounts do o€er cash discounts of one or two percent if the bill is paid within a short period, generally within 10 days. Hence, there is also an opportunity cost associated with the ``discount'' period. 17 This assumption is supported by Petersen and Rajan (1995a) who ®nd that ®rms use more trade credit when they do not have a bank line of credit. Petersen and Rajan (1994) ®nd that ®rms repay trade credit late more often when they have a relatively short relationship with their primary bank lender. Also supporting this assumption is the ®nding by Calomiris et al. (1995) that high credit-quality ®rms issue commercial paper to ®nance accounts receivables during downturns in the business cycle (when bank credit is tight) ± that is, they act as ®nancial intermediaries who lend via trade credit to ®rms of lesser credit quality when credit is tight. 18 Not all ®rms repaying trade credit after the due date face an explicit penalty. We include a variable to control for the cost of late payments.

436

J. Jayaratne, J. Wolken / Journal of Banking & Finance 23 (1999) 427±458

A negative coecient on SMALL implies that small ®rms in areas with many small banks are less likely to pay their suppliers after the due date (i.e., they are less likely to use the high-interest loan component of trade credit). We interpret a negative coecient on SMALL to mean that small ®rms in areas with many small banks demand less trade credit, supporting the conjecture that small businesses in such areas ®nd it easier to access bank credit. As with Eq. (1), a zero coecient on SMALL suggests that a decline in the presence of small banks does not a€ect the supply of bank credit to small ®rms. Similarly, we interpret a negative coecient on CHANGE IN SMALL to re¯ect increased short-run demand for trade credit in areas with recent decreases in small banks, re¯ecting disruptions in credit supply. In the ®rst test we interpret the coecients of SMALL and CHANGE IN SMALL to be ``supply'' parameters and in the second test, as ``demand'' parameters. So, for example, a negative coecient of SMALL is taken to mean that the demand for trade credit decreases when there are more small banks in the area (because such banks are more willing to lend to small ®rms). 19 This is despite the fact that both empirical equations above are clearly reduced forms, and not structural demand or supply equations. We believe that such a structural interpretation of the coecients of SMALL and CHANGE IN SMALL is reasonable here for the following reason: while the null hypothesis explains why SMALL (and CHANGE IN SMALL) a€ects the supply of bank lines of credit (in Eq. (1)) and the demand for trade credit (in Eq. (2)), there is no obvious reason for why SMALL (and CHANGE IN SMALL) would a€ect demand for bank credit lines in Eq. (1) or the supply of trade credit in Eq. (2). For all other variables in the two empirical models above, no such structural interpretation is possible since they probably in¯uence both demand for and supply of credit. Re®nements of Tests 1 and 2: Marginal ®rms and small bank presence. Even if the average small ®rm is not credit constrained by a decrease in the number of small banks, marginally creditworthy small ®rms may ®nd it harder to access credit when there are fewer small banks. Perhaps large banks make cookiecutter loans because of in¯exible underwriting standards, making it dicult for ®rms without a long credit history or with a poor credit history to borrow because such borrowers require ¯exible pricing and underwriting standards. 20

19 Consistent with this, we do not use accounts payable data as an indicator of trade credit use because most accounts payable on ®rmsÕ balance sheets are driven by the supply of trade credit, not demand for trade credit (Petersen and Rajan, 1995a). The fraction of trade credit paid back after the due date (the high interest loan part of trade credit) is as close as we can get to estimating ``demand'' for trade credit (although even this captures supply to some extent in that a ®rm can use such credit conditional on it being o€ered trade credit). 20 Cole et al. (1997) show that large banks rely more on ®nancial ratios based on balance sheets and income statements when making small business loans ± more so than small banks.

J. Jayaratne, J. Wolken / Journal of Banking & Finance 23 (1999) 427±458

437

Consequently, marginal ®rms may be more adversely a€ected by changes in small bank presence than non-marginal ®rms. We test for this possibility by estimating separate coecients on SMALL and CHANGE IN SMALL for the marginal ®rms (separate from non-marginal ®rms). To implement this test, we include dummy variables for young ®rms (5 or fewer years), small ®rms (5 or fewer employees), and ®rms with poor credit histories, and interact these terms with SMALL and CHANGE IN SMALL. In this test involving marginal ®rms, we assume that young ®rms, ®rms with poor credit histories and small ®rms are ``marginal borrowers'', requiring especially close monitoring and screening. Admittedly all three qualities of age, credit history and size are observable equally to large and small banks. However, we interpret these ®rm qualities to be proxies for information problems. For example, Diamond (1989) shows that older ®rms are less likely to choose relatively risky investments than young ®rms. This reduces the lenderÕs cost of monitoring. Moreover, an older ®rm has a longer performance history, providing a potential lender with more initial information about the ®rmÕs prospects, thereby reducing the need for ex post monitoring. Test 3. The propensity of marginal ®rms to use small banks. Our third test examines whether marginal ®rms (®rms that are small, young and have poor credit histories) have a greater propensity than non-marginal ®rms to use small banks for lines of credit. We use the following model: Probability …Firm j obtains its line of credit from a large bank† ˆ f …firm j0 s age; firm j0 s credit history; firm j0 s size; HHI; demand for credit†:

…3†

4. The e€ects of small bank presence on credit availability: Results In this section, we report results from estimating the models discussed in Section 3. 21 The NSSBF sample is described in Table 2. Because the sample was strati®ed by size and minority partition, all models estimated include

21 Although we report here results from estimating the regressions using all ®rms, both rural and urban, we also estimated the models separately for urban and rural ®rms to ®nd similar results. We separated urban and rural ®rms because urban and rural credit markets are likely to have di€erent characteristics. For example, urban areas may have more non-bank sources of credit, and consequently, the e€ects of SMALL on credit access may be quite di€erent in urban and rural areas. Another reason for splitting urban ®rms from rural ®rms is that our de®nition of SMALL is di€erent for urban areas: we use MSAs to de®ne the geographic area for banking markets and individual counties to de®ne rural banking markets.

438

J. Jayaratne, J. Wolken / Journal of Banking & Finance 23 (1999) 427±458

Table 2 The 1993 NSSBF survey: Selected descriptive statistics (all ®rms) Mean

Market characteristics SMALL De®nition 1: Proportion of deposits held by banks with less than $300 million in assets De®nition 2: Proportion of deposits held by banks with assets less than $300 million and if part of holding company, holding company assets are less than $300 million De®nition 3: Proportion of banks with less than $300 million in assets De®nition 4: Proportion of banks with assets less than $300 million and if part of holding company, holding company assets are less than $300 million Her®ndahl Index of deposit concentration Firm characteristics Number of ®rms No. of employees in ®rm Age of ®rm Return on assets

Median

Standard deviation

Mean change 1988±93

0.25

0.15

0.26

0.06

0.19

0.12

0.21

0.02

0.65

0.69

0.21

0.05

0.55

0.54

0.20

0.01

0.20

0.17

0.10

4637 32 15 0.76

6 12 0.14

Percentage of ®rms responding ``yes'' Does the ®rm have a line of credit from a bank? Does the ®rm use trade credit? If using trade credit, did the ®rm repay after the due date? Has the owner declared bankruptcy over the past seven years? Has the owner been delinquent in repaying personal debt over past three years? Has the owner been delinquent in repaying business debt over past three years?

33 68 58 3 13 20

62 14 4.7

J. Jayaratne, J. Wolken / Journal of Banking & Finance 23 (1999) 427±458

439

controls for ®rm size, census region of the ®rm, and the sampling partition. 22 Test 1. Lines of credit and small bank presence. Estimates for Eq. (1) ± which models the odds of having a bank line of credit ± are reported in Table 3. Thirty three percent of ®rms report having a bank line of credit. The key result in Table 3 is that SMALL has no signi®cant e€ect on the odds of a ®rm having a bank line of credit, while CHANGE IN SMALL has a signi®cant positive e€ect in three of four regressions (columns (1)±(3). The insigni®cance of SMALL suggests that, in the long run, small businesses in areas with few small banks (or in areas where small banks have few resources) are no less likely to have a bank credit line than small ®rms in areas with many small banks. This result is robust to di€erent measures of SMALL. In fact, two of the four de®nitions of SMALL have the ``wrong sign'' (negative, but insigni®cantly di€erent from zero) on their coecients, suggesting that a decrease in SMALL is associated with an increase in the odds of having a bank line of credit (columns (1) and (2)). 23 However, the positive e€ect of CHANGE IN SMALL suggests that recent changes in small bank presence a€ects the supply of bank credit in the short run. Nevertheless, that short run e€ect is small ± a 12 percentage point increase in small bank presence (as measured by the fraction of deposits held by banks with less than $300 million in assets ± twice the mean increase observed over the 1988±1993 period) increases the probability of having a bank line of credit by only 0.6 percentage points. Most other variables have predictable e€ects on the dependent variable. If a ®rmÕs owner has a ¯awed credit history, then that ®rm is less likely to have a credit line (but, puzzlingly, business delinquency actually increases the likelihood of having a credit line). More pro®table businesses are less likely to have lines of credit, probably because internal funds (such as retained pro®ts) are a cheaper source of funds than external funds (such as bank debt) because of capital market frictions. As such, ®rms resort to internal funds ®rst when looking for investment funds (Myers and Majluf, 1984). This ``demand'' e€ect

22 Throughout, we present only unweighted regression results. The NSSBF sample does not accurately reproduce the population of small ®rms in the US because minority-owned ®rms and larger ®rms were over sampled in the survey. To account for the e€ects of sample design, we included right-hand side variables for region, industry, minority partition and size (sales and employment). We also estimated the regression models weighted, which produced similar results. 23 Perhaps small ®rms in areas with an especially thin presence of small banks may ®nd it dicult to access bank credit even if ®rms in areas with a moderate presence face the same supply of credit as ®rms in cities with many small banks. A linear SMALL term need not detect this possibility. Allowing for this, we also estimated the model in Table 3 with SMALL and SMALL squared. Neither the linear nor the quadratic SMALL term is signi®cant. Similarly, including a second order term for SMALL did not change the results reported in Table 4.

Firm age squared

Firm age

Business delinquency

Personal delinquency

Creditworthiness Bankruptcy

Market structure Small bank presence (SMALL) Change in SMALL, 1988±93 Her®ndahl Index of deposit concentration Metropolitan statistical area (MSA)

Independent variable

ÿ1.05 (ÿ3.95) ÿ0.46 (ÿ3.64) 0.402 (4.14) 0.001 (0.28) 8  10ÿ6 (0.17)

ÿ0.091 (ÿ0.48) 0.239 (2.47) 0.043 (0.11) ÿ0.139 (ÿ1.16)

(1)

SMALL ˆ Fraction deposits held by banks with assets less than $300 million

ÿ1.05 (ÿ3.97) ÿ0.461 (ÿ3.65) 0.402 (4.15) 0.001 (0.28) 8  10ÿ6 (0.17)

ÿ0.2 (ÿ0.92) 0.157 (1.96) 0.047 (0.12) ÿ0.185 (ÿ1.6)

SMALL ˆ Fraction deposits held by banks with assets less than $300 million and if part of holding company, holding company assets less than $300 million (2)

Measure of small bank presence

ÿ1.05 (ÿ3.95) ÿ0.459 (ÿ3.63) 0.406 (4.19) 0.001 (0.27) 9  10ÿ6 (0.20)

0.154 (0.72) 0.322 (2.15) 0.071 (0.18) ÿ0.156 (ÿ1.53)

(3)

SMALL ˆ Fraction of banks that have assets less than $300 million

Table 3 E€ects of small bank presence on the probablility of having a bank line of credit (logit estimates; all ®rms)

ÿ1.06 (ÿ3.99) ÿ0.461 (ÿ3.65) 0.405 (4.18) 0.001 (0.27) 9  10ÿ6 (0.18)

0.041 (0.19) 0.141 (1.3) 0.025 (0.06) ÿ0.154 (ÿ1.51)

(4)

SMALL ˆ Fraction of banks with assets less than $300 million and if part of holding company, holding company assets less than $300 million

440 J. Jayaratne, J. Wolken / Journal of Banking & Finance 23 (1999) 427±458

4629 0.14

Number of employees squared Pro®tability (return on assets) Growth rate of the state's economy Growth rate of bank lending statewide

Sample size Pseudo R2

4629 0.14

0.02 (13.3) ÿ5  10ÿ5 (ÿ11.5) ÿ0.031 (ÿ2.85) ÿ2.73 (ÿ0.85) 1.29 (1.79) 4629 0.14

0.02 (13.3) ÿ5  10ÿ5 (ÿ11.5) ÿ0.031 (ÿ2.81) ÿ1.48 (ÿ0.46) 1.22 (1.70) 4629 0.14

0.02 (13.3) ÿ5  10ÿ5 (ÿ11.5) ÿ0.031 (ÿ2.84) ÿ2.27 (ÿ0.71) 1.31 (1.83)

Notes: 1. Bankruptcy, personal delinquency and business delinquency are indicator variables that equal 1 if the owner of the ®rm has, respectively, declared bankruptcy in the last seven years, been late in paying personal debts or been late in paying business debt over the three years prior to 1993. 2. The ``Change in Small, 1988±93'' was calculated by dividing the change in SMALL over the 1988±93 period by the average of SMALL over the 1988±1993 period. 3. Z-statistics are given within parentheses. 4. Although not shown, controls for sales and sampling partitions (census region of ®rm and minority partitions) were also included in the regression. * Indicates signi®cance at a 5% level. ** Indicates signi®cance at a 1% level.

0.02 (13.3) ÿ5  10ÿ5 (ÿ11.5) ÿ0.031 (ÿ2.81) ÿ1.44 (ÿ0.45) 1.28 (1.77)

Credit demand Number of employees

J. Jayaratne, J. Wolken / Journal of Banking & Finance 23 (1999) 427±458 441

Firm age squared

Firm age

Business delinquency

Personal delinquency

Creditworthiness Bankruptcy

Market structure Small bank presence (SMALL) Change in SMALL, 1988±93 Her®ndahl Index of deposit concentration Metropolitan statistical area (MSA)

Independent variable

0.354 (1.62) 0.757 (6.64) 1.98 (20.0) ÿ0.003 (ÿ0.65) ÿ6:8  10ÿ6 (ÿ0.13)

(1.01)

(0.76)

0.353 (1.61) 0.756 (6.64) 1.98 (20.0) ÿ0.003 (ÿ0.67) ÿ6:2  10ÿ6 (ÿ0.12)

ÿ0.205 (ÿ0.89) ÿ0.032 (ÿ0.39) ÿ0.039 (ÿ0.10) 0.123

SMALL ˆ Fraction deposits held by banks with assets less than $300 million and if part of holding company, holding company assets less than $300 million (2)

ÿ0.204 (ÿ1.02) ÿ0.112 (ÿ1.17) ÿ0.058 (ÿ0.14) 0.095

(1)

SMALL ˆ Fraction deposits held by banks with assets less than $300 million

Measure of small bank presence

0.351 (1.61) 0.753 (6.60) 1.98 (20.0) ÿ0.003 (ÿ0.68) ÿ7  10ÿ6 (ÿ0.14)

(1.75)

0.057 (0.25) ÿ0.118 (ÿ0.81) ÿ0.023 (ÿ0.06) 0.185

(3)

SMALL ˆ Fraction of banks with assets less than $300 million

0.349 (1.6) 0.755 (6.63) 1.98 (19.9) ÿ0.003 (ÿ0.68) ÿ6  10ÿ6 (ÿ0.13)

(1.71)

0.088 (0.4) 0.001 (0.01) 0.007 (0.02) 0.181

(4)

SMALL ˆ Fraction of banks with assets less than $300 million and if part of holding company, holding company assets less than $300 million

Table 4 E€ects of small bank presence on the amount of trade credit repaid after the due date (ordered logit estimates; all ®rms that used trade credit) 442 J. Jayaratne, J. Wolken / Journal of Banking & Finance 23 (1999) 427±458

3126 0.11

0.329 (4.68) 0.003 (2.16) ÿ3  10ÿ6 (ÿ0.68) ÿ0.023 (ÿ2.08) ÿ2.38 (ÿ0.74) 0.042 (0.06) 3126 0.11

0.33 (4.7) 0.003 (2.18) ÿ3  10ÿ6 (ÿ0.72) ÿ0.022 (ÿ2.06) ÿ1.78 (ÿ0.55) ÿ0.055 (ÿ0.07) 3126 0.11

0.33 (4.69) 0.003 (2.18) ÿ3  10ÿ6 (ÿ0.71) ÿ0.023 (ÿ2.08) ÿ1.82 (ÿ0.57) ÿ0.029 (ÿ0.04) 3126 0.11

0.33 (4.69) 0.003 (2.16) ÿ3  10ÿ6 (ÿ0.69) ÿ0.022 (ÿ2.05) ÿ1.82 (ÿ0.56) ÿ0.056 (ÿ0.08)

Notes: 1. Only those ®rms that reported using trade credit are included. 2. Bankruptcy, personal delinquency and business delinquency are indicator variables that equal 1 if the owner of the ®rm has, respectively, declared bankruptcy in the last seven years, been late in paying personal debts or been late in paying business debt over the three years prior to 1993. 3. The ``Change in Small, 1988±93'' was calculated by dividing the change in SMALL over the 1988±93 period by the average of SMALL over the 1988±1993 period. 4. Z-statistics are given within parentheses. 5. Although not shown, controls for sales and sampling partitions (census region of ®rm and minority partitions) were also included in the regression. ** Indicates signi®cance at a 1% level. * Indicates signi®cance at a 5% level.

Sample size Pseudo R2

Number of employees squared Pro®tability (return on assets) Growth rate of the stateÕs economy Growth rate of bank lending statewide

Credit demand No penalty for paying late on trade credit Number. of employees

J. Jayaratne, J. Wolken / Journal of Banking & Finance 23 (1999) 427±458 443

444

J. Jayaratne, J. Wolken / Journal of Banking & Finance 23 (1999) 427±458

appears to dominate the opposite ``supply'' e€ect produced by the fact that banks are likely to lend to more pro®table ®rms, which are better credit risks. Finally, bigger ®rms are more likely to have lines of credit, probably because bigger ®rms need the liquidity provided by lines of credit. Interestingly, ®rm age has no e€ect. Test 2. Repaying trade credit after the due date and small bank presence. Our second test of the e€ects of small bank presence on credit access focuses on trade credit. Of the 4637 ®rms in the sample, 68 percent reported using trade credit in 1993. Of the ®rms using trade credit, 58 percent repaid their suppliers after the due date. Table 4 reports ordered Logit estimates for the model of determinants of late repayment of trade credit by small businesses. This model uses only those 3126 ®rms that reported using trade credit during 1993. 24 Again, the key result is that neither SMALL nor CHANGE IN SMALL has a statistically signi®cant impact, and ± in all but two cases ± the coecient is positive, the opposite of that predicted by the null hypothesis. That is, small ®rms in areas with few small banks (or in areas where small banks have fewer resources) do not pay back their suppliers late any more than small ®rms in areas with many small banks. SMALL does not appear to a€ect the demand for trade credit, implying that SMALL does not a€ect access to bank credit, even in the short run. This result is robust against changes in the de®nition of SMALL. 25 Other variables that are statistically signi®cant have predictable e€ects on late repayment. Poor credit histories by the owners is associated with more late repayments, while increased cash ¯ows reduce late repayments. Interestingly, larger ®rms pay late more frequently; at the very least this suggests that small ®rms are not necessarily more liquidity constrained than larger ®rms once we control for pro®tability and ownersÕ credit history. Firm age has no e€ect. Finally, we added an indicator variable that equals one if a ®rm reported that it did not have to pay a ®ne if it repaid trade credit late. This variable has a predictable e€ect: those ®rms who do not face such a penalty are more likely to repay after the due date. Re®nements of Tests 1 and 2: Marginal ®rms and small bank presence. The results thus far suggest that small ®rms in areas with a thin presence of small banks do not have greater diculty accessing bank credit. But this result is obtained only after we control for the creditworthiness of small ®rms. Perhaps

24

Because we are estimating the odds of repaying trade credit late conditional on using trade credit, our inferences here are limited to only those ®rms that actually use trade credit. 25 SMALL and HHI are inversely correlated, raising the possibility that the non-signi®cance of SMALL in both Tables 3 and 4 is due to the collinearity between these two variables (despite the large sample size). However, dropping the HHI did not change the results.

J. Jayaratne, J. Wolken / Journal of Banking & Finance 23 (1999) 427±458

445

marginally creditworthy ®rms, those ®rms that require substantial monitoring by their creditors, may face greater diculty accessing bank credit when there are fewer small banks in the area because larger banks have more diculty in customizing loans to the needs of individual borrowers (although this may well be the ecient outcome if such ®rms are inecient). Our results so far do not rule out this possibility because the average ®rm in our sample is probably not a marginal ®rm (only 3 percent of ®rms in our sample have owners who have declared bankruptcy, for example). To test whether marginal borrowers have greater diculty accessing bank credit in areas with few small banks, we begin by de®ning marginal ®rms as those which satisfy one or more of the following conditions: (1) their owners have ¯awed credit histories; (2) are ®ve years or younger; (3) ®ve or fewer employees. 26 We create three indicator variables to ¯ag marginal ®rms as per these three criteria: POOR CREDIT equals one only if a ®rmÕs owner has a ¯awed credit history, YOUNG FIRM equals one only if the ®rm is ®ve years or younger and SMALL FIRM equals one only if the ®rm has ®ve or fewer employees. In Tables 5 and 6, we essentially replicate Tables 3 and 4 ± but now we interact the three indicator variables for marginal ®rms with SMALL and CHANGE IN SMALL. 27 The results in Table 5 are from estimating a model that is identical to that in Table 3 (with the probability of having a bank line of credit as the dependent variable), except that ®rm age and employees has been converted into discrete variables as described above (and the second order terms of ®rm age and employees dropped). Moreover, the three indicator variables for poor credit history used in Table 3 has been collapsed into one indicator variable in Table 5. Only the small bank presence variables and their interactions with marginal ®rm indicators are shown to save space. If small bank presence a€ects credit access especially for marginal ®rms, then we would expect to see the following pattern in Table 5: the coecients of SMALL and CHANGE IN SMALL (which capture the e€ects of these variables on non-marginal ®rms) would be insigni®cant (or positive and signi®cant), but the interaction termsÕ coecients should be positive and

26 Twenty ®ve percent of ®rms report their owners as declaring bankruptcy or being delinquent in their business or personal loans over the 1990±1993 period. Fifteen percent of ®rms are ®ve years or younger and 68 percent have ®ve or fewer employees. 27 Using indicator variables to ¯ag ``marginal ®rms'' and interacting them with SMALL and CHANGE IN SMALL allows easier interpretation of results ± easier than interacting continuous measures of ®rm age and size with small bank presence measures. Nevertheless, we also estimated the models that interacted continuous measures of ®rm age and size with small bank presence measures to ®nd similar results to those reported below.

SMALL FIRMSMALL

YOUNG FIRMChange in SMALL SMALL FIRM

YOUNG FIRMSMALL

POOR CREDITChange in SMALL YOUNG FIRM

POOR CREDITSMALL

POOR CREDIT

Change in SMALL, 1988 ± 93

Small bank presence (SMALL)

Independent variable

ÿ0.38 (ÿ1.65) 0.32 (2.45) ÿ0.031 (ÿ0.25) ÿ0.026 (ÿ0.082) ÿ0.087 (ÿ0.41) ÿ0.55 (ÿ3.7) 0.86 (2.5) ÿ0.27 (ÿ1.02) ÿ1.34 (ÿ11.53) 0.42 (1.58)

(1)

SMALL ˆ Fraction deposits held by banks with assets less than $300 million

ÿ0.6 (ÿ2.1) 0.14 (1.24) 0.019 (0.17) ÿ0.13 (ÿ0.35) 0.19 (1.04) ÿ0.49 (ÿ3.9) 1.11 (2.7) ÿ0.1 (ÿ0.48) ÿ1.34 (ÿ12.81) 0.56 (1.76)

SMALL ˆ Fraction deposits held by banks with assets less than $300 million and if part of holding company, holding company assets less than $300 million (2)

Measure of small bank presence

0.098 (0.35) 0.32 (1.51) 0.16 (0.57) ÿ0.25 (ÿ0.6) 0.24 (0.67) ÿ0.21 (ÿ0.64) ÿ0.031 (ÿ0.07) 0.59 (1.2) ÿ1.5 (ÿ5.8) 0.37 (1.05)

(3)

SMALL ˆ Fraction of banks that have assets less than $300 million

ÿ0.07 (ÿ0.25) 0.14 (0.92) 0.29 (1.2) ÿ0.55 (ÿ1.3) 0.42 (1.6) ÿ0.33 (ÿ1.1) 0.12 (0.25) 0.086 (0.29) ÿ1.6 (ÿ7.2) 0.69 (1.89)

(4)

SMALL ˆ Fraction of banks with assets less than $300 million and if part of holding company, holding company assets less than $300 million

Table 5 E€ects of small bank presence on the probability of having a bank line of credit: Marginal vs. non-marginal ®rms (logit estimates; all ®rms)

446 J. Jayaratne, J. Wolken / Journal of Banking & Finance 23 (1999) 427±458

4629 0.15

Sample size Pseudo R2

4629 0.15

ÿ0.03 (ÿ0.19) 4629 0.15

ÿ0.16 (ÿ0.54) 4629 0.15

ÿ0.27 (ÿ1.24)

Notes: 1. POOR CREDIT is an indicator variable that equals 1 if the owner of the ®rm has either declared bankruptcy in the last seven years or been late in paying personal debts or been late in paying business debt over the three years prior to 1993. 2. YOUNG FIRM is an indicator variable that equals one if the ®rm has been under the current ownership ®ve or fewer years. Otherwise it equals 0. 3. SMALL FIRM is an indicator variable that equals one if the ®rm has ®ve or fewer employees. Otherwise it equals 0. 4. The ``Change in Small, 1988±93'' was calculated by dividing the change in SMALL over the 1988±93 period by the average of SMALL over the 1988±1993 period. 5. Z-statistics are given within parentheses. 6. Although not shown, controls for HHI, MSA dummy, ®rm pro®tability, growth rate of the state economy, growth rate of bank lending statewide, sales and sampling partitions (census region of ®rm and minority partitions) were also included in the regression. * Indicates signi®cance at a 5% level. ** Indicates signi®cance at a 1% level.

ÿ0.05 (ÿ0.26)

SMALL FIRMChange in SMALL

J. Jayaratne, J. Wolken / Journal of Banking & Finance 23 (1999) 427±458 447

YOUNG FIRMChange in SMALL SMALL FIRM

YOUNG FIRMSMALL

POOR CREDITChange in SMALL YOUNG FIRM

POOR CREDITSMALL

POOR CREDIT

Change in SMALL, 1988±93

Small bank presence (SMALL)

Independent variable

0.008 (0.031) ÿ0.21 (ÿ1.55) 2.2 (17.4) 0.02 (0.06) 0.08 (0.4) 0.088 (0.64) ÿ0.29 (ÿ0.82) ÿ0.0005 (ÿ0.002) ÿ0.41 (ÿ3.6)

(1)

SMALL ˆ Fraction deposits held by banks with assets less than $300 million

0.065 (0.21) ÿ0.1 (ÿ0.84) 2.2 (19.4) ÿ0.18 (ÿ0.5) ÿ0.035 (ÿ0.2) 0.06 (0.53) ÿ0.29 (ÿ0.68) ÿ0.052 (ÿ0.26) ÿ0.42 (ÿ4.1)

SMALL ˆ Fraction deposits held by banks with assets less than $300 million and if part of holding company, holding company assets less than $300 million (2)

Measure of small bank presence

0.52 (1.7) ÿ0.28 (ÿ1.6) 2.7 (11.1) ÿ0.9 (ÿ2.2) 0.38 (1.5) 0.14 (0.5) ÿ0.21 (ÿ0.45) 0.19 (0.8) ÿ0.33 (ÿ1.5)

0.67 (2.2) ÿ0.63 (ÿ2.9) 2.7 (9.55) ÿ0.71 (ÿ1.8) 0.47 (1.4) 0.31 (1.0) ÿ0.39 (ÿ0.88) 0.52 (1.3) 0.006 (0.022)

(3)

SMALL ˆ Fraction of banks with assets less than $300 million and if part of holding company, holding company assets less than $300 million (4)

SMALL ˆ Fraction of banks that have assets less than $300 million

Table 6 E€ects of small bank presence on the amount of trade credit repaid after the due date: Marginal vs. non-marginal ®rms (ordered logit estimates; all ®rms that used trade credit)

448 J. Jayaratne, J. Wolken / Journal of Banking & Finance 23 (1999) 427±458

3126 0.11

ÿ0.24 (ÿ0.85) 0.12 (0.66)

3126 0.11

ÿ0.32 (ÿ0.9) 0.12 (0.75)

3126 0.11

ÿ0.69 (ÿ1.9) 0.644 (2.2)

3126 0.11

ÿ0.28 (ÿ0.8) 0.26 (1.2)

Notes: 1. POOR CREDIT is an indicator variable that equals 1 if the owner of the ®rm has either declared bankruptcy in the last seven years or been late in paying personal debts or been late in paying business debt over the three years prior to 1993. 2. YOUNG FIRM is an indicator variable that equals one if the ®rm has been under the current ownership ®ve or fewer years. Otherwise it equals 0. 3. SMALL FIRM is an indicator variable that equals one if the ®rm has ®ve or fewer employees. Otherwise it equals 0. 4. The ``Change in Small, 1988±93'' was calculated by dividing the change in SMALL over the 1988±93 period by the average of SMALL over the 1988±1993 period. 5. Z-statistics are given within parentheses. 6. Although not shown, controls for HHI, MSA dummy, ®rm pro®tability, growth rate of the state economy, growth rate of bank lending statewide, sales, an indicator variable for late payment penalties and sampling partitions (census region of ®rm and minority partitions) were also included in the regression. * Indicates signi®cance at a 5% level. ** Indicates signi®cance at a 1% level.

Sample size Pseudo R2

SMALL FIRMChange in SMALL

SMALL FIRMSMALL

J. Jayaratne, J. Wolken / Journal of Banking & Finance 23 (1999) 427±458 449

450

J. Jayaratne, J. Wolken / Journal of Banking & Finance 23 (1999) 427±458

signi®cant. 28 We do not see this in Table 5. SMALL is never positive and signi®cant (although it is negative and signi®cant twice ± the wrong sign), and CHANGE IN SMALL is positive and signi®cant in only one of four regressions (column (1)). The interactions are positive and signi®cant in only two of twenty-four instances (columns (1) and (2)) ± the interaction between SMALL and the YOUNG FIRM indicator. Even in those two instances, we cannot conclude that SMALL a€ects young ®rmsÕ credit access. The total e€ect of SMALL on young ®rmsÕ likelihood of having a bank credit line is captured by the sum of the coecient of SMALL and the coecient of the interaction between SMALL and YOUNG FIRM. A Wald test showed that this sum is not signi®cantly di€erent from zero (not shown here). Table 5 suggests that small bank presence has little impact on marginal ®rmsÕ access to bank credit lines. However, does small bank presence a€ect trade credit use by marginal ®rms? The results in Table 6 are from estimating a model that is similar to that in Table 4 (with the probability of repaying trade credit late as the dependent variable), except for the presence of indicator variables for marginal ®rms and interactions between such variables and the small bank presence measures (as in Table 5). If small bank presence a€ects credit access especially for marginal ®rms, then we would expect to see the following pattern in Table 6: the coecients of SMALL and CHANGE IN SMALL (which capture the e€ects of these variables on non-marginal ®rms) would be insigni®cant (or negative and signi®cant), but the interaction termsÕ coecients should be negative and signi®cant. We do not see this in Table 6. SMALL is never negative and signi®cant (although it is positive and signi®cant once ± the wrong sign), and CHANGE IN SMALL is negative and signi®cant in only one of four regressions (column (4)). The interactions are negative and signi®cant in only one of twenty-four instances (column (4)) ± the interaction between SMALL and the POOR CREDIT indicator. Even in this instance, we cannot conclude that SMALL a€ects the likelihood of late repayment of trade credit by ®rms with poor credit histories. The total e€ect of SMALL on the likelihood of late repayment by a ®rm with a poor credit record is captured by the sum of the coecient of SMALL and the coecient of the interaction between SMALL and POOR CREDIT. A Wald test showed that this sum is not signi®cantly di€erent from zero (not shown here).

28 Or more accurately, the sum of the coecients of SMALL and CHANGE IN SMALL and the coecients of the interaction terms should be positive and signi®cant. Such sums capture the total e€ects of small bank presence on credit access for marginal ®rms. For example, if the sum of the coecient of SMALL and the coecient of interaction between SMALL and POOR CREDIT is positive and signi®cant, then we conclude that SMALL has an e€ect for ®rms with poor credit records.

J. Jayaratne, J. Wolken / Journal of Banking & Finance 23 (1999) 427±458

451

We conclude that there is no evidence that credit access by marginal ®rms (®rms that are young or very small or have poor credit histories) is a€ected by the presence of small banks in the banking market. 29 Test 3. The propensity of marginal ®rms to use small banks. Marginal ®rms ± ®rms with poor credit histories, small ®rms and young ®rms ± are especially likely to require close scrutiny by lenders. If the Small Bank Advantage hypothesis is correct, i.e., small banks can monitor problem borrowers at a lower cost than large banks, then marginal ®rms are likely to borrow from small banks more often than non-marginal ®rms. In Table 7, we test this implication. There, we ask whether marginal small businesses are less likely than non-marginal small ®rms to go to large banks for credit lines. The Logit model in Table 7 estimates the probability of a ®rm getting a bank line of credit from a large bank, conditional on that ®rm having a bank line of credit at all. 30 That is, the dependent variable equals one when the ®rm has a credit line from a large bank, and zero if it has a credit line from a small bank. We use the same independent variables as those in Tables 3 and 4, except that we drop the small bank presence variables. Table 7 shows that ®rms with poor credit histories and young ®rms are as likely to get a loan from a large bank as ®rms with good credit histories and older ®rms. To the extent that ®rms with poor or short credit histories require closer scrutiny, this result suggests that large banks do not su€er from higher monitoring costs in small business lending. 31 However, Table 7 also shows that smaller ®rms are more likely to get credit lines from small banks. But this ®nding does not necessarily support the Small

29

Tables 5 and 6 test for the e€ects of small bank presence on bank credit availability to marginal ®rms when a ``marginal ®rm'' is de®ned as a ®rm that (1) has a ¯awed credit history, or (2) is ®ve years or younger, or (3) has ®ve or fewer employees. Of course, potentially the most marginal borrowers may be those ®rms that satisfy all three conditions. The total e€ect of SMALL on such a ®rm is the sum of the coecient on SMALL and the coecients on the three interaction terms between SMALL and the three marginal ®rm indicators (likewise for CHANGE IN SMALL). A Wald test revealed that this sum is not signi®cantly di€erent from zero. That is, SMALL has no e€ect on the probability that a small, young ®rm with poor credit has a bank line of credit or on the likelihood of late repayment of trade credit. We found similar results for CHANGE IN SMALL. 30 In Table 6 we use only those 1324 ®rms with bank credit lines where we were able to identify the bank lender as ``large'' or ``small''. Of the 1530 ®rms in our sample with bank lines of credit, we were unable to identify the bank lender for 206 ®rms. (Firms in the NSSBF report the names and locations of the banks that they borrow from. Although this information is not available in the public version of the NSSBF, banks' information was made available to us by the Board of Governors.) 31 Firms in areas with no small banks do not have a choice between large and small banks. We mitigated this (potential) problem by re-estimating the model in Table 7 using only those ®rms in areas where both small and large banks were present (i.e., we dropped those observations where the small bank presence measures equaled 1 or 0). Results were similar to those in Table 7.

452

J. Jayaratne, J. Wolken / Journal of Banking & Finance 23 (1999) 427±458

Table 7 Likelihood of having a line of credit from a large bank: Marginal vs. non-marginal ®rms (logit estimates; all ®rms with bank lines of credit) Independent variable

Market structure Her®ndahl Index of deposit concentration Metropolitan statistical area (MSA) Creditworthiness Bankruptcy Personal delinquency Business delinquency Firm age Firm age squared

Credit demand Number of employees Number of employees squared Pro®tability (return on assets) Growth rate of the state's economy Growth rate of bank lending statewide Sample size Pseudo R2

De®nition of large bank Banks with assets greater than $300 million (1)

Banks with assets greater than $300 million or if part of a holding company, holding company assets greater than $300 million (2)

ÿ0.224 (ÿ0.29) 1.46 (7.79)

ÿ0.632 (ÿ0.81) 1.01 (5.22)

ÿ0.694 (ÿ1.24) ÿ0.027 (ÿ0.10) ÿ0.195 (ÿ1.03) 0.001 (0.06) 3  10ÿ5 (0.17)

ÿ1.04 (ÿ1.85) ÿ0.154 (ÿ0.53) ÿ0.099 (ÿ0.49) 0.002 (0.19) 2  10ÿ5 (0.10)

0.011 (3.95) ÿ2  10ÿ5 (ÿ2.86) 0.0003 (0.01) 5.08 (0.75) 1.49 (1.02)

0.011 (3.64) ÿ3  10ÿ5 (ÿ3.45) 0.011 (0.59) 7.45 (1.04) 0.821 (0.53)

1324 0.17

1324 0.15

Notes: 1. Only those ®rms that reported having a bank line of credit are included. 2. Bankruptcy, personal delinquency and business delinquency are indicator variables that equal 1 if the owner of the ®rm has, respectively, declared bankruptcy in the last seven years, or been late in paying personal debts or been late in paying business debt over the three years prior to 1993. 3. Z-statistics are given within parentheses. 4. Although not shown, controls for sales and sampling partitions (census region of ®rm and minority partitions) were also included in the regression.  Indicates signi®cance at a 5% level. ** Indicates signi®cance at a 1% level.

J. Jayaratne, J. Wolken / Journal of Banking & Finance 23 (1999) 427±458

453

Bank Advantage hypothesis. Small banks may specialize in lending to small businesses even when they do not have any advantage over large banks when lending to small businesses simply because small banks may ®nd it dicult to lend to large ®rms. Large ®rms require large loans, and small banks will be unable to construct diversi®ed loan portfolios if they make a few large loans to large ®rms. This process of specialization is an alternative explanation of why smaller borrowers turn to small banks more often than larger borrowers. 5. Conclusion In this paper, we bring a new and rich source of information on small business ®nance ± the 1993 National Survey of Small Business Finance ± to bear on the question of whether small business lending depends on small banks (and hence whether such lending decreases when small banks disappear due to bank consolidation). We conduct a simple test of the conjecture that large banks are at a disadvantage in lending to small ®rms: if such disadvantages are signi®cant, then small ®rms in areas with few small banks should be more credit constrained than ®rms in areas with many small banks. We ®nd no evidence that small businesses in areas with few small banks are any more credit constrained than ®rms in areas with many small banks, although we ®nd weak evidence of short run e€ects of small bank presence. In the long run, the number of small banks in an area does not a€ect either the odds of a ®rm having a bank line of credit or the probability of repaying trade credit after the due date, a proxy for excess institutional credit demand. This result holds for even the more marginally creditworthy ®rms (such as those with poor credit histories or short credit histories). Moreover, we ®nd that marginal small ®rms are as likely as non-marginal ®rms to get a bank credit line from a small bank. There are several explanations of our main ®nding here that a reduced presence of small banks may not reduce access to bank credit by small businesses. One possibility is that the MSA is not the relevant geographic market area for small business loans in urban areas (and the county for rural areas). If so, then our measure of small bank presence, SMALL (which assumes the MSA or the county is the relevant market area) is mismeasured, and that measurement error biases its coecient in Table 3±6 toward zero. Worse, there may exist some other geographic area (smaller or larger than an MSA, for example) ± the true small business market area ± where a decrease in small bank presence does lead to a decline in credit access for small ®rms. We discount this possibility because there is ample evidence from small business loan pricing studies to suggest that the MSA approximates the relevant market area for small business lending in urban areas and the county for rural areas. A long tradition of previous work has found that small business loans in MSAs that are relatively concentrated (i.e., MSAs with high bank deposit

454

J. Jayaratne, J. Wolken / Journal of Banking & Finance 23 (1999) 427±458

HHIs, for example) are priced higher (i.e., have higher interest rates) than loans in less concentrated MSAs (see Berger and Hannan, 1993, for instance). In other words, the structure of banks within MSAs a€ects the prices of small business loans. This implies that MSAs approximate the relevant market area for small business loans in urban areas. Similar results are obtained for rural areas. Indeed, regulators typically use MSAs as urban small business market areas and counties for rural areas when conducting antitrust reviews of bank mergers. As such, we believe that using MSAs to de®ne SMALL is reasonable. 32 A second possible reason for the non-signi®cance of SMALL, at least in Tables 4 and 6, is that non-bank ®nancial intermediaries (mortgage ®nance companies, ®nance companies, factors, etc.) step in to lend to small businesses when there are fewer small banks. We discount this possibility partly because Table 3 shows that bank lending (lines of credit) does not decline with small banks, and partly because the NSSBF highlights the importance of banks and the relative unimportance of non-banks as providers of short term liquidity to small businesses (although non-banks are important sources of long-term loans for equipment and vehicles). For example, while 23.4 percent of all ®rms in our sample had a line of credit from a depository institution, only 1.5 percent had a line of credit from a non-depository institution (mostly ®nance companies). 33 We believe that if banks reduce credit to small ®rms signi®cantly, then nonbanks are unlikely to completely o€set this shrinking supply of bank credit (and hence SMALL should appear to have some e€ect in Eqs. (1) and (2)). We then turn to a third reason for why SMALL may have no e€ect on small ®rmsÕ access to credit: large banks may be as willing as small banks to lend to small ®rms because large banks do not ®nd it signi®cantly costlier to lend to small ®rms than small banks. That is, the evidence here suggests that the Small Bank Advantage hypothesis is not true, and bank consolidation will have little impact on small business credit. However, individual ®rms and areas may experience diculty in accessing credit when small banks vanish, and short run disruptions to credit supply from mergers may occur. In fact, we ®nd evidence of short-run credit supply e€ect when small bank presence decreases. Another caveat to our conclusion is that our results apply primarily to short-run credit and not to longer-run credit

32 We could potentially use states (an area larger than an MSA) to de®ne small business loan market areas. We do not use states as the market area because the NSSBF shows that small ®rms overwhelmingly go to local banks for credit. Ninety-®ve percent of small ®rms borrow from banks that are no farther than 25 miles from the ®rmÕs headquarters ± an area substantially smaller than a state. Note also that our regressions using rural ®rms used counties as the relevant market area, a smaller geographic region than multi-county MSAs, to also ®nd SMALL has a zero coecient. 33 These ®gures are based on the pooled sample of urban and rural ®rms, and after ®rms have been weighted to re¯ect sampling strata used in the NSSBF (and hence the di€erence with the unweighted sample described in Table 2).

J. Jayaratne, J. Wolken / Journal of Banking & Finance 23 (1999) 427±458

455

such as mortgage, equipment and vehicle loans (although the presence of many non-bank providers of such credit suggests that bank consolidation will have little e€ect). Finally, our results may be limited by the fact that the data are from 1993, which is from the ``credit crunch'' period. If the credit crunch had forced small banks ± who may specialize in lending to marginal borrowers ± to ration out marginal borrowers and instead lend to better borrowers (in a ¯ight to quality in loans), then this would explain why we do not observe small bank presence to have any e€ect on marginal ®rmsÕ access to credit in 1993. Both large and small banks may have rationed out marginal ®rms to the same extent during this atypical period. (However, the credit crunch hit larger banks harder since they were on average less capitalized than small banks.) Acknowledgements This research was completed while Jith Jayaratne was employed by the Federal Reserve Bank of New York. The views in this paper do not necessarily represent those of the Federal Reserve Bank of New York or the Federal Reserve System. The authors thank Robert Avery, Allen Berger, Rebecca Demsetz, Mitchell Petersen, Philip Strahan, seminar and conference participants at the Federal Reserve Bank of New York, and two anonymous referees for their suggestions. Thanks also to Nicole Meleney and Ronnie McWilliams for research assistance. Appendix A: De®nitions of analysis variables Name of variable Dependent variables Bank line of credit

Frequency of late payments on trade credit

Description of variable Equals one if the ®rm has a line of credit at a commercial bank, savings bank or savings and loan association. Equals 0 otherwise. The dependent variable in Tables 3 and 5. Indicates how often the ®rm made payments to trade credit after the due date. Takes on the following values: 0: never 2: less than half of the time 3: about half of the time 4: more than half of the time 5: almost all or all of the time The dependent variable in Tables 4 and 6.

456

J. Jayaratne, J. Wolken / Journal of Banking & Finance 23 (1999) 427±458

Large bank line of credit 1

Large bank line of credit 2

Equals 1 if the ®rm has a line of credit at a large commercial bank, savings bank or savings and loan association. Equals 0 otherwise. A large bank is one with bank assets of $300 million or more (as of June 1993). Dependent variable in Table 7 (column (1)). Equals 1 if the ®rm has a line of credit at a large commercial bank, savings bank or savings and loan association. Equals 0 otherwise. A large bank is one with bank assets of $300 million or more or if part of a holding company, the holding company assets are $300 million or more (as of June 1993). Dependent variable in Table 7 (column (2)).

Market structure variables CHANGE IN SMALL (1993) ÿ SMALL (1988)/[(SMALL SMALL, 1988±1993 (1993) + SMALL (1988))/2] Metropolitan Equals one if the ®rm is located in a metropolitan Statistical Area (MSA) statistical area. Zero otherwise. Her®ndahl Index of Her®ndahl±Hirschman Index of deposit concendeposit concentration tration for the MSA or non-MSA county in which (HHI) the ®rmÕs headquarters are located. Creditworthiness Bankruptcy Personal delinquency Business delinquency Poor credit (POOR CREDIT) Firm age Young ®rm (YOUNG FIRM)

Equals 1 if the ®rmÕs principal owner has declared bankruptcy within the past seven years. Zero otherwise. Equals 1 if the ®rmÕs principal owner has had 1 or more personal obligations that were 60 or more days delinquent within the past three years. Zero otherwise. Equals 1 if the ®rmÕs principal owner has had 1 or more business obligations that were 60 or more days delinquent within the past three years. Zero otherwise. Equals 1 if the bankruptcy, personal delinquency or business delinquency variables equal 1. Zero otherwise. Number of years since the ®rm was founded/ purchased/acquired. Equals 1 if the ``Firm age'' is ®ve years old or less. Zero otherwise.

J. Jayaratne, J. Wolken / Journal of Banking & Finance 23 (1999) 427±458

Other variables Number of employees Small ®rm (SMALL FIRM) No penalty for paying late on trade credit Pro®tability Growth rate of stateÕs economy Growth rate of bank lending statewide

457

Number of full-time equivalent employees in the ®rm, where a part-time employee is 12 of a full-time employee. Equals 1 if the ®rm has ®ve full-time equivalent employees or less. Zero otherwise. Equals 1 if the ®rm used trade credit and incurred no penalty for paying after the due date. Equals zero if the ®rm used trade credit and incurred a penalty for paying after the due date. Ratio of ®rmÕs pro®ts to assets. State personal income (1993)/State personal income (1992). Total bank loans at all banks in state (31 December 1993)/Total bank loans at all banks in state (31 December 1992).

Sample control variables Sales FirmÕs total sales for 1992 in millions of dollars. Sample partition Indicates which sample partition that the ®rm was included in. Census region Indicates which of nine census regions that the ®rm was located in. References Berger, A., Hannan, T., 1993. Using eciency measures to distinguish among alternative explanations of the structure±performance relationship in banking. Finance and Economics Discussion Series 93-18. Board of Governors of the Federal Reserve System, Washington, DC. Berger, A., Kashyap, A., Scalise, J., 1995. The transformation of the US banking industry: What a long, strange trip itÕs been. Brookings Papers on Economic Activity 2, 55±218. Berger, A., Saunders, A., Scalise, J., Udell, G., 1998. The e€ects of bank mergers and acquisitions on small business lending. Journal of Financial Economics 50, 187±229. Calomiris, C., Himmelberg, C., Wachtel, P., 1995. Commercial paper, corporate ®nance, and the business cycle: A microeconomic perspective. Carnegie-Rochester Conference Series on Public Policy 42, 203±250. Cole, R., Goldberg, L., White, L., 1997. Cookie-cutter versus character: The micro structure of small business lending by large and small banks. Mimeo. Diamond, D., 1989. Reputation acquisition in debt markets. Journal of Political Economy 97, 828± 862. Keeton, W., 1995. Multi-oce bank lending to small businesses: Some new evidence. Federal Reserve Bank of Kansas City Economic Review 80, 45±57. Keeton, W., 1997. The e€ects of mergers on farm and business lending at small banks: New evidence from tenth district states. Working Paper. Federal Reserve Bank of Kansas City, Kansas City, MO.

458

J. Jayaratne, J. Wolken / Journal of Banking & Finance 23 (1999) 427±458

Myers, S., Majluf, N., 1984. Corporate ®nancing and investment decisions when ®rms have information that investors do not have. Journal of Financial Economics 13, 187±221. Peek, J., Rosengren, E., 1997. Bank consolidation and small business lending: itÕs not just bank size that matters. Working Paper. Federal Reserve Bank of Boston, Boston, MA. Petersen, M., Rajan, R., 1995a. Trade credit: Theories and evidence. Manuscript. Petersen, M., Rajan, R., 1995b. The e€ects of credit market competition on lending relationships. Quarterly Journal of Economics 60, 407±444. Petersen, M., Rajan, R., 1994. The bene®ts of lending relationships: Evidence from small business data. The Journal of Finance XLIX, 3±37. Strahan, P., Weston, J., 1998. Business lending and the changing structure of the banking industry. The Journal of Banking and Finance 22, 821±845. Whalen, G., 1995. Out-of-state holding company aliation and small business lending. Economic and Policy Analysis Working Paper 95-4. Comptroller of the Currency, Washington, DC.