CHANGES IN COMPETITION AND BANKING OUTCOMES FOR SMALL FIRMS

Transcription

1 CHANGES IN COMPETITION AND BANKING OUTCOMES FOR SMALL FIRMS Abstract This paper examines how a set of small firm banking outcomes are related to changes in the state of competition among financial institutions. The state of competition is measured by an owner assessment of bank competition for their financial business as well as a traditional measure of deposit concentration. A significant positive association is found between changes in bank competition reported by owners and their reports of changes in service delivery and credit availability, but no association with loan terms. Deposit concentration was not significant in explaining any outcome. The possibility that owners assume that a good outcome is a result of increased competition for their banking business is rejected. Evidence is provided that is consistent with owners recognizing changes in bank competition in their markets regardless of the outcomes experienced. Keywords: small business, competition, market concentration JEL Classification Codes: G21, G28

2 CHANGES IN COMPETITION AND BANKING OUTCOMES FOR SMALL FIRMS 1. Introduction A key regulatory and public policy goal is the maintenance of competitiveness in markets for financial services. Regulators frequently use concentration ratios as a first filter for assessing market competitiveness in merger reviews, with further analysis of product pricing changes for cases where a concentration threshold is violated. Implicit in this approach is an assumption that more concentration leads to worse outcomes for market participants. This framework can be problematic for small business loan markets where the strength of banking relationships is central to the firm s interaction with its primary financial institution. In these cases, the results of lender production of hard and soft information and its effect on credit availability and the quality of service may be just as important as the price of credit. The empirical evidence on bank market competition and small firm credit outcomes has generally focused on concentration measures (e.g., Petersen and Rajan, 1994, Jayanthe and Wolken, 1999, Cavalluzzo, Cavalluzzo, and Wolken, 2002). The results have been mixed in the sense that higher deposit concentration has been found in some cases to improve credit availability rather than reduce it. Whether increased competition benefits small, opaque firms depends on the incentives to invest in private information, the lender s portfolio strategy, and market structure. Peterson and Rajan (1995) suggest that at least with respect to credit availability, increased competition (i.e. less concentration) may be detrimental to credit availability if banks are strictly relationship lenders. However, Boot and Thakor (2000) and Dell Arricia and Hauswald 1

3 (2004) develop models that show conditions under which increased competition will result in more credit allocated to certain information-captured firms. This paper provides a different perspective on how local market competition affects small firm banking outcomes using 1995 and 2001 data collected from the membership of the National Federation of Independent Business (NFIB), a U.S. trade organization of non-public firms. One question, Have you noticed any change in competition for your firm s business among financial institutions now compared to 3 years ago? provides the owner s assessment of the behavior of lenders in their market. Responses to this question are related to a set of banking outcomes that are important to small, information opaque firms: credit availability, service quality, and loan terms. The underlying premise of the paper is that the behavior of the banks in local markets provides clearer a signal about the state of competition for small firm banking business than the number of banks in a market or their share of deposits. Measures of concentration as an indicator of competition can be ambiguous. Fewer competitors may result in monopoly power that worsens banking outcomes for all firms or an improvement in banking outcomes for small firms because it is more profitable for banks to invest in private information. With the responses to the survey question, there is a unique opportunity to test the hypothesis that more competition based on the action of lenders - leads to better banking outcomes and still control for bank concentration. The most important empirical concern is whether the survey question is capturing exogenous changes in the competitive behavior of banks, or whether it reflects the owner erroneously crediting changes in bank competition for the firm s banking outcomes. This causality issue presents an identification problem that cannot be easily addressed. 2

4 Using a restricted sample of firms that have not recently applied for a loan or do not use credit, we test for whether it is more likely that an increase (decrease) in competition leads to a better (worse) assessment of the change in service quality over the same period. As reported in detail below, the evidence suggests that owners understand the concept of competition embodied in the question and are more likely to report changes in competitive behavior independent of the banking outcomes for the firm. Even if the causality runs from (reports of) changes in competition to banking outcomes, the error term in the model may still be correlated with the competition variable. Banks may devise a contact strategy for new customers based on a favorable risk/return profile, such as recent sales growth or industry improvements. Thus, a positive association between reported changes in competition and banking outcomes in the entire sample may simply reflect a correlation with an omitted variable that captures the bank s solicitation strategy based on firm attributes. Firms whose performance is improving will experience favorable banking outcomes and are more likely to be solicited by banks for their business. This endogeneity issue is addressed through the use of a set of instruments for the change in competition responses. Using these instruments, the reports of increased competition are found to be positively associated with improved credit availability and service quality, but no significant association is found with the reported loan rate or non-rate terms such as the incidence of collateral or compensating balances. The lack of any association between the owner assessment of competition and loan pricing could be because each individual term is only part of the effective loan price and thus cannot adequately capture the total rate of return (or cost) of the loan. 3

5 Banking concentration was found to have no independent effect on credit availability. However, the favorable effect of owner reports of competition was mitigated in markets with an increase in deposit concentration. Banks in highly concentrated markets can increase their contact of potential customers, but the the overall potential effect on small firm banking outcomes is reduced when there are fewer banks in the market. The remainder of the paper is organized as follows. A brief review of the literature and the competing theories about the effect of market structure and competitive forces on credit market experiences of small firms is presented in Section 2. The data and variable construction are presented in Section 3, followed by the hypotheses and empirical specification in Section 4. Analysis of the results is given in Section 5 and conclusions are presented in Section Competition and banking outcomes A number of theoretical models have been developed to assess the effect of competition on banking outcomes. The Structure-Conduct-Performance (SCP) model predicts that more market power or concentration will negatively affect credit supply and cost (e.g. Hannan, 1991). However, as noted by Shaffer (2004), other definitions of equilibrium may give a different association between conduct and concentration (e.g. contestable markets). Another model of the relationship between competition and market outcomes is the asymmetric information hypothesis (Petersen and Rajan, 1995). In this model, market power could be positively related to credit supply and costs (at least 4

6 initially) because of the increased incentives of lenders to invest in private information (relationship banking) in more concentrated markets. Boot and Thakor (2000) derive a model that predicts a non-linear relation between competition and credit availability for relationship loans. In their model, a bank can make both transaction (or arms-length ) loans and relationship loans that require the production of soft information resulting in what they call sector specialization. An initial increase in inter-bank competition gives the same result as the asymmetric information models, a reduction in the benefits from investing in sector specialization. However, in their model, profits from both transactions and relationship lending are affected by more competition with a greater effect on transactions lending profitability. Thus, inter-bank competition will cause banks to allocate more credit to its captured borrowers. Dell Ariccia and Marquez (2004) provide a similar prediction in their model, where informed banks increase lending to information captured borrowers when faced with greater competition. In his review of the SCP literature, Shaffer (2004) notes the existence of different theoretical predictions about the association between concentration and conduct, so that ultimately the question of competition and market structure is an empirical one. Not surprisingly, the empirical tests of the relationship between concentration and pricing using bank level data are inconclusive. Some evidence is found of market power in the pricing of deposits (e.g., Hannan and Liang, 1993), but not for loan pricing, even in highly concentrated markets such as Canada (e.g., Nathan and Nieve, 1989) or markets with a banking duopoly (Shaffer and DiSalvo, 1994). The size of the market used may also complicate the detection of a relationship between concentration and pricing. For 5

7 example, Radecki (1998) examined the relationship between retail deposit rates and market concentration between the mid-1980s and mid-1990s and concluded that changes in the supply side of retail banking have created markets at least as large as a state. To further complicate the empirical association between outcomes and concentration, Brevoort and Hannan (2004) note that competition may be from out-ofmarket lenders that would not be included in a typical concentration ratio computation. Improvements in technology clearly facilitate the entry of these out-of-market lenders, which over time could further weaken the link between the number or size of competitors within a local market and the competition viewed from their customer s perspective. The rapid growth in credit card lines for small businesses is a prime example. Another segment of the literature indirectly related to the SCP model examines the effect of mergers on the small loan market share in commercial bank portfolios. Bank level data is used in these studies and focuses on the quantity of credit made available to small firms, not the price (e.g., Peek and Rosengren, 1998, and Strahan and Weston, 1996). 1 A common finding in these papers is that mergers between small banking institutions appear to increase the share of assets held as small loans. However, no clear, predictable outcome is found when two large banks or a large and a small bank merge. Berger et al (1998) conclude that adjustments of other banks in the local market offset most, if not all, of the impact of a merged bank s portfolio reduction in small-sized loans. In another study, Berger, Bonime, Goldberg and White (2004) find that mergers and 1 A small firm is defined by a maximum loan size ($1 million or less) which is very large compared to average loan size for small firms included in the Federal Reserve and NFIB surveys of small firms. 6

8 acquisitions are associated with significant increases in the probability of new bank entry in both metropolitan and rural markets. Using firm-level data from the Board of Governors 1987 Survey of Small Business Finance (SSBF), Petersen and Rajan (1995) found evidence in favor of the asymmetric information hypothesis. Firms in more concentrated markets were less credit constrained and had an initial cost of funds that was lower compared to less concentrated markets. Over time, however, the cost of funds was higher, reflecting the recovery of the higher cost of the initial investment in private information for these information-opaque firms. These results were argued by the authors to be consistent with what we label the asymmetric information hypothesis. Jayarantne and Wolken (1999), however, found no significant relationship between a measure of deposit concentration and the amount of trade credit paid late (a proxy for credit availability) using the Board of Governors 1993 SSBF. Yet, DeYoung, Goldberg and White (1999) found a positive association between concentration and small business lending in urban markets, but a negative association in rural markets. Beck, Kunt and Maksimovic (2004), using firm-level data from 74 countries, found a negative association between concentration and financing obstacles for firms of all size, although the relation turned insignificant in countries with high levels of GDP and well-developed financial institutions. Overall, the cumulative firm-level empirical evidence is mixed regarding the effect of deposit concentration on small firm banking outcomes. 2 While there is some 2 The effect of competition on the lender s outcomes is mixed as well. Ergungor (2005) presents evidence that relationship lending has not added value to community bank lenders and he cites changes in the 7

9 evidence that increased market concentration improves availability (supporting the asymmetric information prediction), this effect has not been consistently observed over time or for all banking products. This result is not surprising given the mixed theoretical predictions about the effect of the number of competitors on banking outcomes. Concentration ratios may be poor proxies for the behavior of banking firms in local markets if they do not capture the nature of the competitive behavior of the financial service firms in local markets (e.g. Boot and Thakor, 2000) or because the technology of credit scoring and on-line access to bank credit products have widened the opportunity set for banks to contact small firms. Assessments of the nature of competition provided by business owners may give a more accurate characterization of bank competition in a market than a concentration ratio. This paper makes three contributions to the literature on banking outcomes and market structure. First, a measure of the change in competition is employed that directly reflects the behavior of banks in a local market. As such, it permits a direct test of whether increased competition improves banking outcomes for small firms that is independent of the number of competitors in the market. By including a market concentration variable, we can also test for whether the number of banks in a market is related to their incentives to invest in private information, thereby improving banking outcomes for information opaque firms. Reported changes in competition may, however, capture the reaction of banks adjusting their loan portfolios to changes in market concentration as described by Boot and Thakor (2000). Unfortunately, because the competitive structure of the market as one of the reasons. But Carter et al (2004) find that risk-adjusted yields for small business loans decrease as bank size increases. 8

10 shares of relationship versus transaction lending of the owner s primary banker is not known, a test of the Boot and Thakor (2000) model is not possible. Second, the effects of reported changes in competition on an expanded set of banking outcomes is examined, while still controlling for the strength of banking relationships and deposit concentration in local markets. Similar to previous research, tests are conducted for the association of the survey measure of competition with a measure of credit availability and a set of loan terms (rate and non-rate). The association between changes in the competition variable and a proxy for service quality is also examined. This expanded set of outcomes is especially important for small information opaque businesses that rely on relationship banking to access capital markets. And third, empirical results are provided at two points in time: late 1994 and early 1995, the beginning of a period of active bank consolidation; and late 2001, the end of the expansion and the start of a recession. The results from these two periods, separated by seven years of economic growth and bank consolidation, provide an opportunity to assess the effect of changing market structure on credit costs and availability for U.S. small businesses and to revisit the results of earlier research that used firm-level data before this period of rapid bank consolidation. 3. Data 3.1 Survey description The data used in this study come from the 1995 and 2001 Credit, Banks and Small Business Survey conducted by the National Federation of Independent Business (NFIB). The purpose of the surveys was to collect information about the credit market 9

11 experiences of a random sample of the NFIB s 600,000 members. 3 In the 1995 survey, eighteen thousand questionnaires were sent in the initial mailing and 3,642 completed questionnaires were available after the second mailing for a response rate of 20 percent. In the 2001 survey, questionnaires were mailed to 12,500 firms and responses were received from 2,223 after two mailings for a response rate of 18 percent. A comparison of the distribution of the 1995 and 2001 NFIB sample to the distribution of the 1993 and 1998 SSBF conducted by the Board of Governors of the Federal Reserve System is presented in Appendix 1. In both surveys the NFIB firms are slightly larger than those in the SSBF survey in terms of employees, sales, and assets. The 2001 NFIB survey respondents are somewhat older, larger and less urban than the 1995 survey. Consequently, the 2001 data are weighted by employment, industry and region using proportions calculated from the 1997 U. S. Bureau of the Census Enterprise survey. The definition and summary statistics of the key variables used in each survey in this study are shown in Table 1. Not all of the statistics in Table 1 are based on 3,642 (1995 Survey) or 2,223 (2001 Survey) observations because the sample is restricted to those firms responding to the change in competition question, excluding the no answer responses in the summary statistics. The no answer responses to the change in competition question comprise fewer than five percent of the total observations in each sample. 3 Details about the surveys can be found in (Dunkelberg, 1998 and Scott, Dunkelberg, and Dennis, 2003). 10

12 3.2 Competition proxies: Survey reports of change in competition and bank concentration In response to the NFIB question about observed changes in competition for the firm s financial business, forty-two (42) percent reported much more or slightly more competition for their business in the three-year period prior to the survey compared to thirty-eight (38) percent in 1995 (and thirty-two (32) percent in the 1987 study, not used in the analyses presented here). 4 In 2001, nine percent reported slightly less or much less competition, up from six percent in The responses to this question are recoded into a diffusion-type index, where much less = -2, slightly less = -1, no change = 0, slightly more = 1, and much more = 2. The mean value of the change in competition using this index is.43 for both surveys. The use of this variable in empirical work poses several conceptual challenges. No specific measure of competition is provided to the respondent (e.g. the number of unsolicited contacts by banks or a statistical measure of concentration). Thus improvements (or deterioration) in banking market outcomes may be construed by the respondents as evidence of improved (or degraded) competition, which creates a potential causality problem. Section 4.1 provides evidence that the respondents are likely to know competition when they see it, supporting the premise that reports of changes in competition are attributable to changes in the competitive structure of the market observed by owners. Even so, firms that are successful are more likely to be targeted by bank business development officers and thus may be more likely to experience favorable 4 The specific question, noted in the Introduction, is: Have you noticed any change in competition for your firm s business among financial institutions now compared to 3 years ago? 11

13 outcomes. This problem is addressed in section 4.3 by using a set of instruments for the competition variable. An alternate measure of competition frequently used in the literature is a deposit concentration variable (the Herfindahl-Hirshmann Index) computed for each Metropolitan Statistical Area (or county level for non-msas). The correlation coefficients between reported changes in competition and the HHI is shown in Table 2-A. The association between HHI and reported changes in competition could, a priori, be either positive or negative. 5 In highly concentrated markets, new entrants (de novo charters or out-of-market branches) may see an opportunity to gain new customers, resulting in an increase in competition. Higher concentration could also be the result of a reduction in the number of institutions within the market, resulting in fewer contacts between small firms and financial institutions. The actual correlation coefficient between reported changes in competition and HHI is negative for both years, but small and not significant. This outcome may reflect the dynamics of market entry by other institutions and the inability of a concentration ratio to reflect the behavior of institutions in a local market (Berger et al, 1998 and 2004) or the widening of markets beyond MSAs (Radecki, 1998). Reported changes in competition are also not significantly correlated with the percentage change in HHI (over the previous three years), which is perhaps a better comparison. Reported changes in competition are negatively related to reports of a recent merger of the firms primary financial institution, which may reflect lower levels of bank contact in the short-term caused by a reduction in the number of banking organizations 5 We want to thank Sherill Shafer for making this point. 12

14 remaining in the local market (Table 2-A). This association is only significant for the 1995 survey. In the longer run, as noted by Berger et al (2004), new banks will enter the market. Consistent with this prediction is the lack of significance in 2001 between reported mergers and reports of changes in competition. The stronger economy after the 1995 survey was associated with an increase in new charters, as well as the emphasis of many larger banks on growing their small business loan portfolio. Overall, the correlation coefficients in Table 2-A suggest that the reported changes in competition from the NFIB surveys show little relationship to the traditional measures of market concentration. Table 2-A also presents the correlation coefficients between reported changes in competition and a number of firm characteristics. Years in business is not correlated with reported changes in competition in either survey. The change in competition variable is positively correlated with measures of size (FTEs and sales) in the 1995 survey, but not in the 2001 survey. Sales growth is positively associated with reported changes in competition, but the association is only significant in Reported increases in competition for the firm s business shows some tendency to be associated with better credit quality in 1995, but this relation is much weaker in the 2001 survey. 3.3 Banking outcomes Credit availability A measure of credit availability frequently used in the literature is the frequency with which firms take discounts associated with supplier financing (e.g. Petersen and Rajan, 1994, 1995, Jayaratne and Wolken, 1999). Trade credit is usually a very expensive source of funds and is often used when short-term bank financing is 13

15 unavailable (Petersen and Rajan, 1994). In markets where competition improves credit availability, owners should be less credit constrained and more frequently take discounts for early payment. Both surveys ask how frequently discounts are taken, with four possible responses: never, rarely, sometimes, and always. Reports of increased competition are positively correlated with increases in the frequency of discounts taken for both surveys (Table 2-B), but the correlation is only significant in Bank concentration is positively related to trade credit discounts taken an outcome that would support the asymmetric information hypothesis but the association is not significant Service quality change Both surveys asked owners about a change in the characteristics of financial institutions they deal with most often. The common set of questions between both surveys include accessibility of account manager, services offered, capability of staff, continuity of service manager, and lending terms. For each of the attributes, the owners rate the change as better, no change or worse without regard to its importance to the owner s banking relationship. A simple service quality change index is computed by adding each of the individual ratings, which are coded by worse = -1, no change = 0, and better =1. 6 As seen in Table 2-B, all of the characteristics are significantly positively correlated with reported changes in competition in the 1995 survey, and all except for staff turnover in the 2001 survey. Bank concentration as measured by the HHI is not significantly correlated with any of the service quality change components or the index. 6 An argument can be made that the lending terms component should be excluded from the index because that activity is not strictly related to service. All of the empirical analysis was conducted without this characteristic and there was no change in the results. 14

16 3.3.3 Loan terms If owners were successful in their most recent loan request, they were then asked to report some of the characteristics of their most recent loan, such as the type, size, maturity, interest rate, fees, and the incidence of both collateral and compensating balance requirements. Three loan term outcomes are used in this analysis: the interest rate reported on the most recent loan, the incidence of collateral (1 if business and/or personal collateral was required as a condition of the loan, 0 otherwise), and the incidence of compensating balances (1 if checking account balances were required as a condition of the loan, 0 otherwise). The association of these variables with reported changes in competition is not as strong as the other banking outcome measures (Table 2- B). While the interest rate and collateral requirements are negatively correlated with reports of changes in competition in the 2001 survey and the interest rate is negatively correlated in the 1995 survey, none of the loan term correlations are significant. As was the case with trade credit discounts taken and service quality change, no significant correlation exists between bank concentration and loan terms. 3.3 Other control variables Several survey questions capture important elements of the firm s risk profile: size (FTEs), years in business, form of business (e.g., proprietorship, partnership, corporation, S-corporation), and industry classification. Years in business and FTEs are entered in log form, while a set of 1/0 variables is used for form of business and industry classification. Both of these variables have been used in other firm level research (e g, Cole, 1998, Berger and Udell, 1995, Petersen and Rajan, 1995 among others). Also 15

17 included are the log of sales reported for the most recent fiscal year and annualized sales growth over the previous three years. Two proxies for the strength of banking relations that have been found to be important factors affecting credit outcomes in prior research are used. The first is the length of time since the owner last changed its primary bank. Evidence of the effect of the banking relation strength on credit market outcomes is widespread, primarily using length of time at the owner s current bank (e.g., Berger and Udell, 1995, Petersen and Rajan, 1994, Cole, 1998). The other banking relation proxy is the number of banks used by the owner. Cole (1998) and Petersen and Rajan (1995) have used this variable with the expectation that the more financial institutions used, the weaker the banking relation. Both these variables are entered in log form for estimation purposes. Prior research has also documented that bank size affects small firm credit market outcomes. Cole, Goldberg, and White (2004) have provided evidence that small firms are more successful in obtaining credit at small banks because the banks rely more on relationship factors than financial ratios in their credit underwriting. The estimation includes a proxy for bank size, CFI (community financial institution), which takes a value of 1 if the owner currently does business at a bank with less than $1 billion in assets and 0 otherwise. Loan maturity and purpose (working capital, refinancing, or fixed asset acquisition) are included as control variables in the equations that relate loan terms to change in competition. Three additional control variables are included in the loan rate equation that were also used by Petersen and Rajan (1994): 1) a risk-free U.S. Treasury rate for a maturity matching that of the loan in the quarter it was made: 2) a macro default 16

18 premium based on the spread between the BAA and U.S. Treasury rate by maturity; and 3) a term premium reflecting the difference between the risk free rate associated with the maturity of the loan and the three-month Treasury rate at the time of disbursement. 4. Empirical Specification 4.1 Causality A critical specification issue is whether outcomes are determined by the change in competition as reported by the owner, or is the change in competition is determined by the outcomes. In other words, do owners report more competition simply because they experience a favorable outcome or do they know competition when they see it and provide reports on changes in competition in their market independent of the experience of the firm itself? We do not have a variable that could empirically identify the direction of causality. However, we can restrict our analysis to a subset of the owners who would not be influenced by recent credit application outcomes: those who do not report applying for a loan or those who applied for a loan over three years ago a period outside the time frame for the change in competition assessment. With this restricted sample we can examine the service quality ratings and be assured that the influence of a recent credit decision is minimized. Table 3 Panel A presents a hypothetical cross-tabulation of a hypothetical banking outcome and reported changes in competition. The lower case letters (a, b, c, and d) identify the probability, p, of the off-diagonal cell entries. A chi-square test is used to test the following null hypotheses. If H 0 : p a =p b is rejected, then competition is a function of outcomes. In other words, if a and b are significantly different, it must be due 17

19 to the outcome (even though unchanged) affecting the change in competition assessment. Conversely, if H 0 : p c =p d is rejected, outcomes are a function of competition. If the chisquare test rejects both H 0 : p a =p b and H 0 : p c =p d, then there is a functional relationship in both directions. Using the hypothetical frequencies given in Panel A, the null hypothesis that p a =p b (competition is a function of outcomes) cannot be rejected, but the null hypothesis that p c =p d, i.e. outcomes are a function of competition, is rejected. Panel B of the table reports the chi-square value for the two tests (H 0 : p a =p b and H 0 : p c =p d ) for both surveys, using the change in service quality questions as well as the service quality change index. Three of the five characteristics in the 2001 survey and four of five in the 1995 survey reject the null hypothesis that competition is not a function of outcomes at a.001 level. For the other characteristics, the data are consistent with a functional relationship on both sides. The magnitude of the chi-square statistics is much higher for the null hypothesis H 0 : p c =p d, which is consistent with the direction of causality supported by the tests that were significant. The test of these hypotheses using the service quality change index the sum of the individual characteristic responses provides much stronger results in favor of the causality running from reported changes in competition to outcomes. The null hypothesis of no causality from outcomes to competition cannot be rejected at the.05 level in 2001 and.01 level in 1995; but the null hypothesis of no causality from competition to outcomes is rejected. At a minimum, the tests reject the hypothesis that the causality runs exclusively from outcomes to reported changes in competition. The least favorable interpretation would be that causality runs in both directions for most of the characteristics (e.g. a.05 level of significance where the critical value is 3.841). However, the results of the chi- 18

20 square test on the service quality change index support the proposition that owners know competition when they see it, independent of the banking outcomes for their firm. 4.2 Reduced form model The reduced form of the model used in the empirical tests is shown in equation (1), where banking outcomes are a function of the reported change in competition, firm risk characteristics, bank relationship strength, and bank/market structure proxies. (1) Banking Outcome i = a 0 + b 1 change in competition + b 2 HHI + + b 3 ln(length of time at primary bank) + b 4 ln(no. of banks used) + b 5 ln(years in business) + b 6 ln(fte) + b 7 form of business +b 8 1-digit SIC industry classification +b 9 sales growth + b 10 ln(sales) + b 11 CFI-primary bank + u Banking Outcome i includes the frequency of trade credit discounts taken, the service quality change index, the rate on the most recent loan, the incidence of collateral, and the incidence of compensating balances. The coefficient on change in competition, b 1, should be positive if the actions of banks are a good proxy for the state of competition in a local market. The coefficient on b 2 should be positive if deposit concentration is the appropriate proxy for an asymmetric information view of competition, but could be negative if increased concentration reflects less competition due to monopoly power. 4.3 Endogeneity concerns and instrumental variables The reduced form model in equation (1) still may suffer from an endogeneity problem if banks develop a contact strategy based on identifying firms with the highest credit quality. Thus, the significance of the change in competition variable may be due to correlation with an omitted variable that captures the bank s solicitation strategy based on firm characteristics. A two-stage model is used similar to a Hausman test for endogeneity described in Wooldridge (2002). In the first stage, shown in equation (2), the change in competition 19

21 variable is estimated as a function of the control variables (x) in equation (1) and a set of instruments (y) described below. The residuals from equation (2), u, are used in the second stage where they are entered into equation (1a) shown below. If the coefficient on the residual term u is significant, the null hypothesis of no endogeneity is rejected, and a two-stage least squares approach should be used with the instruments in y as the identifying variables for change in competition. (2) Change in competition i = α 1 + βx +γy + u, where x is the set of predictors in (1) and y includes the instruments for change in competition described below. (1a) Banking outcome i = a 0 + b 1 change in competition + bx + cu + e The instruments that comprise y in equation (2) include variables for location, change in market structure, and real growth. Two proxies are used for location. The first is a 1/0 variable to indicate whether the firm is located in a metropolitan area and the second is a set of 1/0 variables for the region of the country based on U.S. Bureau of the Census definitions. Two proxies are used for the change in market structure: the change in the HHI (by MSA or rural county); and the incidence of merger activity using a 1/0 variable to reflect whether the owner reports a recent merger or acquisition of their primary bank. The level of deposit concentration within a market may influence how the change in concentration affects the change in reported competition. For example, a market with a high level of concentration may be less affected by a given percentage increase in concentration than a market with a lower level of concentration because the effect on the number of competitors would be less. An interactive variable is created (HHI x HHI) and added as an additional instrument. This variable should vary negatively with 20

22 reported changes in competition if the number of participants in the local market determines the strength of competition. The proxy for real growth is the percentage change in employment for the MSA location of the firm over the three years prior to the survey. These figures are computed from 1990 to 1993 for the 1995 survey and from 1997 to 2000 for the 2001 survey using wage and salary employment figures provided by the Bureau of Economic Analysis. The estimation of the instruments for the change in competition variable is shown in Appendix Estimation strategy Two different econometric approaches are employed to obtain consistent parameter estimates of the effect of reported changes in competition on the banking outcome measures. Ordered probit is used for the trade credit variable and the service quality change index, as well as for the estimation of the set of instruments for the change in competition variable. Linear regression would treat all of the differences within each variable as equal, e.g. a change between -2 (much less) and -1 (less) for the change in competition variable is equivalent to a change between 1 (more) and 2 (much more). These differences, however, are only a ranking and thus ordered probit would be more appropriate (Greene, 2000). 7 The downside is that the marginal effects of the independent variables on the dependent variable are more difficult to interpret. 7 Only 60% of the respondents reported using trade credit in the 2001 survey (and 70% in the 1995 survey), which leads to a potential sample selection bias when trying to estimate the association between the change in competition variables and the frequency of trade credit discounts taken. A Heckit estimate was used to model the use of trade credit but we were unable to reject the null hypothesis of independent equations. Thus ordered probit was used on the reduced samples for both surveys. 21

23 The loan term equation needs to consider that the loan rate, collateral, and compensating balances may be determined jointly by the lender and the borrower, possibly leading to a correlation between the error terms. To account for this possibility, seemingly unrelated regression is used to jointly estimate the three loan term equations. This system of equations relies on a linear probability model structure to explain the incidence of collateral and compensating balances, both 1/0 variables. While the problems of a linear probability model are well known (the estimated probabilities are not bounded between 0 and 1), our only interest is in significance, not an estimate of the marginal probabilities. The downside, however, is that the simultaneous equation approach reduces the number of observations for both surveys because not all respondents report a complete set of terms. This reduction in observations is higher for the 2001 survey because more owners reported revolving credit lines without specific interest rates. 5. Empirical Results 5.1 Credit availability The multivariate estimate of the effect of reported changes in competition on the credit availability proxy, trade credit discounts taken, is presented in Table 4. Panel A shows the estimates for the 1995 survey and panel B shows the estimates for the 2001 survey. Only the key independent variables are presented with the other control variables omitted for ease of presentation. The estimates of equation (1a) for the endogeneity test are shown in column 1 with coefficients for the change in competition variable and the residual term u constructed from the coefficients estimated in equation (2). The null hypothesis of no endogeneity is rejected because the residual term is significant and thus 22

24 the instruments for competition are used to evaluate the effect of the reports of change in competition on trade credit discounts taken. The results in Table 4, column 2, reject the null hypothesis that reported changes in competition have no effect on credit availability for both surveys. An increase in reported competition is significantly associated with an increased frequency of taking trade credit discounts. The market concentration variable, HHI, is also positively related to credit availability - consistent with the asymmetric information hypothesis - but is not significant. Alternative specifications that included the change in HHI or the percentage change in HHI still resulted in no significant association. Thus for credit availability, the data provide more support for the SCP than for the asymmetric information hypothesis. Table 6, Panel A presents the observed and predicted values for the frequency of trade credit discounts taken and with the marginal effects of change in competition on each category (never, rarely, sometimes, always). The predicted probabilities follow the observed probabilities more closely in the 1995 survey than in the 2001 survey. The marginal effects are as expected: an increase in the change in competition reduces the probability of reporting never and rarely and increases the probability of reporting sometimes or always (Figure 1). These effects appear to be economically meaningful in both years: the partial derivative with respect to the change in competition variable shows a increase in the probability of always taking trade credit discounts in 1995 and a reduction in the probability of never taking trade credit discounts in The change in competition question was asked in the context of the last 3 years. Some firms could have changed banks in response to bank solicitations in order to be less credit constrained. Thus, the results may be biased in favor of finding a positive 23

25 relationship between reported changes in competition and trade credit discounts taken. To control for this possibility, equation (1) is estimated using only those firms that had not changed banks within the past three years, a restriction that reduced the number of observations by about half (Table 4, column 3). The conclusions from the baseline estimates in column 2 are unchanged, except that now the HHI is significant in the 1995 survey. The persistent significance of the proxies for the strength of banking relations is noteworthy, even after controlling for reported changes in competition, and market concentration. Both coefficients on the relation proxies have the expected sign (positive for the length of the relationship and negative for the number of banking relations) and all are significant, suggesting that it is still worthwhile for banks to invest in private information even with reports of increased competition. Perhaps banks are able to profit from the investment in private information and small businesses are willing to pay but the switching costs may be low enough that more local market competition keeps the incumbent bank on their toes. Unfortunately, the data are not rich enough to rigorously test these ideas. We further examine the relation between market structure and credit availability in Table 4, columns 4 and 5. First we investigate whether the positive association of change in competition varies with the change in deposit concentration. An interactive variable, the change in competition x the % HHI (over the three year period from the survey), is created to test this conjecture and is included with reported changes in competition in the base model. Increased concentration would mitigate the effect of more competition because there would be fewer lenders in the market to make potential 24

26 contacts. Thus, the coefficient on the interactive variable would be negative. However, increased concentration could reduce the potential free-rider costs of investing in private information, so the coefficient would be positive or zero, depending on where the breakeven level of concentration is, i.e. the point where the benefits of relationship banking are just offset by the potential free-rider costs. The coefficient on the interactive variable shown in column 4 is negative and significant for both surveys supporting the conjecture that effective competition depends on the behavior of banks in the market, and the absolute level of potential contacts for a given firm is lower the higher the level of concentration. Banks in highly concentrated markets can increase their competitive activities, but the overall potential impact on outcomes is reduced the fewer the number of banks in the market. Next we examine whether the significance of the competition variable is just reflecting the actions of lenders in urban markets. Firms located in urban markets may be receiving more contact regardless of the number of banks in the market because of the likelihood of a higher level of economic activity and geographic proximity of financial institutions. To test for this effect, the coefficient on the change in competition variable is allowed to vary by market size (MSA versus non-msa) and the null hypothesis is there is no effect by market size. Rejection of the null hypothesis would support the idea that the change in competition result is driven more by the consequences of spatial distribution of banks, rather than their actions. The estimated coefficients, shown in column 5, are both significant but a linear restriction test cannot reject the null hypothesis of equal effects of reported competition on credit availability by market size. 5.2 Service quality change index 25

27 The estimates of the effect of changes in reported competition on the reports of change in service quality incorporated in the service quality change index are presented in Table 5. The results generally correspond with those for credit availability shown in Table 4. The null hypothesis of no endogeneity is rejected using the instruments for reported changes in competition in column 2. An increase in reported competition is positively associated with the service quality change index in both surveys, but unlike the results for credit availability, the HHI is also positively associated with the service quality change index. Cole et al (2004) has shown that small firms experience better credit availability as small banks. Thus, the significance of the HHI may be attributable to a correlation with the number of owners banking at CFIs (see Table 2-A). One way to control for this effect is to create an interactive term between HHI and CFI and omit the CFI variable in equation 1. 8 The better outcomes attributable to banking at CFIs may increase with the level of concentration for one of two reasons. First, if the high concentration is due to the presence of a few large banks, the CFIs may be more likely to receive high service ratings. Second, if located in a rural market, a CFI may have more latitude for capturing rents associated with relationship banking and be better able to deliver better quality service. The results of estimating equation 1 with the CFI x HHI interactive term is shown in column 2. The coefficient on the interactive term is positive as expected and the HHI 8 Alternatively, the HHI variable could be dropped and the CFI variable included. When this specification was run the CFI variable, like the HHI variable in the test presented, was not significant. 26

28 term is no longer significant. The remainder of the sensitivity tests includes the interactive HHI x CFI term. Table 6, Panel B, presents the observed and predicted values for each outcome of the dependent variable, the service quality change index, which takes values from -5 (a rating of worse on each of the five attributes of the index) to 5 (a rating of better on each of the five attributes of the index). Similar to the results with trade credit discounts taken, the predicted probabilities follow the observed probabilities closer for the 1995 survey than the 2001 survey. The marginal effects (derivative of the service quality index with respect to change in competition) are as expected (Figure 2): an increase in the change in competition reduces the probability of reporting lower values of the index (e.g. -3, -4, -5) and increases the probability of reporting higher values of the index (e.g. 3, 4, or 5). The marginal effect is economically meaningful in both years, especially for low ratings of change in service quality: a 0.08 reduction in the probability of the lowest rating in 1995 and a 0.11 reduction in The marginal effect for higher values is much lower, especially in The significance of the association between change in competition and the service quality change index is not attributable to recent bank changes the owner may have made to obtain better service. The change in competition variable for this restricted sample, as seen in column 3, is still significant for both surveys. The effect of the strength of banking relations is only significant in 1995, although the signs are still in the expected direction for However, it is not clear how strength of banking relations should affect changes in service quality because not all of the attributes incorporated in the index 27