Lending technologies, lending specialization, and ... - Springer Link

Small Bus Econ (2011) 37:277–304 DOI 10.1007/s11187-009-9243-1

Lending technologies, lending specialization, and minority access to small-business loans Karlyn Mitchell • Douglas K. Pearce

Accepted: 27 October 2009 / Published online: 10 December 2009 Springer Science+Business Media, LLC. 2009

Abstract We investigate minority access to smallbusiness loans using a probit model of loan application denial that recognizes two loan types (line-of-credit loans and non-line-of-credit loans) made by two lender types (commercial banks and nonbank financial institutions). We estimate our model on data from the 1998 Survey of Small Business Finances. We find evidence consistent with minority equal access to bank credit lines and nonbank non-line-of-credit loans in highly competitive loan markets; in less competitive markets we find evidence consistent with unequal access to these loans. We also find evidence consistent with unequal minority access to bank nonline-of-credit loans, regardless of loan market competitiveness. Our findings differ from previous research which treats small-business loans as a homogenous product and finds evidence consistent with unequal minority access to small-business loans generally. We argue that the existence of multiple small-business lending technologies and loan specialization by lenders account for our findings and demonstrate the need to treat small-business loans as a heterogeneous product when investigating equal access to small-business credit. Keywords Small-business loans Lending technologies Lending specialization K. Mitchell (&) D. K. Pearce North Carolina State University, Raleigh, NC, USA e-mail: [email protected]

Relationship lending Fair lending Discrimination Asymmetric information SME finance JEL Classifications

D21 G21 J15 L26

1 Introduction Improving minority small-business owners’ access to US credit markets remains a serious public concern. Recent academic studies find evidence consistent with unequal access decades after the passage of the Equal Credit Opportunity Act of 1974 (Blanchflower et al. 2003; Blanchard et al. 2008). A recent study by the Government Accountability Office (GAO) concludes that the problem still remains poorly understood (GAO 2008). As the GAO notes, small-business loans comprise a heterogeneous collection of credits ranging from lines of credit to term loans and made by lenders ranging from commercial banks to lease companies. Small-business lending is the subject of a growing academic literature which explores the design and monitoring of loan contracts for firms of varying degrees of informational opacity as well as lenders’ incentives to lend under different conditions. Yet, fairaccess studies rarely draw on this literature and instead treat small-business loans as a fairly homogenous commodity. We seek to advance understanding of minorities’ access to small-business loans by drawing on the

123

278

small-business lending literature. We develop a model of the loan application denial decision that recognizes two types of loans and two types of lenders. The model also permits the criteria for loan denial to differ between White and minority applicants on the two loan types from the two lender types. We estimate our model on data from the 1998 Survey of Small Business Finances (SSBF) and use it to test three predictions about credit-market access for minority small-business owners relative to White owners. Our evidence suggests a mixture of equal and unequal credit-market access for minority smallbusiness owners.1 In markets with high competition among lenders we find evidence consistent with equal access to credit lines at commercial banks and to nonline-of-credit loans at nonbanks. We also find evidence consistent with unequal access to non-lineof-credit loans at banks. In low-competition markets we find evidence consistent with unequal access to both loan types at both lender types. We attribute our findings to differences in the technologies lenders use to extend loans to informationally opaque small businesses under different competitive conditions and to lender specialization. We believe our paper makes two important contributions. First, we bring insights from the literature on small-business lending to the problem of fair access to small-business loans. Second, our evidence that minority small-business owners have relatively equal access to some small-business credits may be helpful to policymakers crafting measures to make credit-market access truly fair.

1

We use the terms ‘‘equal access’’ and ‘‘fair access’’ to credit as concise terms to connote lending decisions that are impartial with respect to demographic identity. We recognize that a lender’s impartiality cannot be directly observed or measured. We also recognize that we cannot distinguish between situations in which lenders deny loans on the basis of racial profiling versus information observable by the lender but not in our data set that happens to be correlated with demographic identity. While we acknowledge these problems, we also note that the 1998 SSBF is the most comprehensive data set with respect to borrower characteristics and coverage of minorityowned small businesses available to academic researchers.

123

K. Mitchell, D. K. Pearce

2 Prior evidence, model, and hypotheses 2.1 Prior evidence Small businesses pose formidable information problems to lenders regardless of the business owner’s demographic group affiliation. Lenders have poorer information about the quality of firms’ projects than do firms’ owners, increasing risk to lenders. Raising loan rates to reflect risk invites adverse selection which could lower lenders’ profits, however. Stiglitz and Weiss (1981) show how these conditions lead to equilibrium credit rationing, whereby loan rates are replaced by another credit-rationing mechanism, say a loan application process. An implication of credit rationing for fair-access researchers is that evidence of unequal loan-market access may take the form of unexplained differences in loan application denial rates for different demographic groups as well as unexplained differences in loan rates. To date, researchers have found little evidence of ethnic disparities in loan rates (Cavalluzzo and Cavalluzzo 1998; Cavalluzzo et al. 2002; and Blanchard et al. 2008; Blanchflower et al. 2003, is an exception). This outcome invites a closer inquiry into possibly disparate rates of loan application denials. In modeling the loan application acceptance decision, researchers view lenders as gathering data on observable applicant characteristics relevant to loan profitability, weighting them to form an index, and then extending a loan if the index exceeds a threshold level or denying the loan if it does not. Since researchers observe whether applications are approved or denied but not the index, they use data on outcomes and applicant characteristics to estimate probit models of loan application denial. Most existing equal-access studies estimate denial models having one of the following two forms: ProbðYi ¼ 1Þ ¼ Zi ¼ U a þ Rj cj Dji þ Rk bk Xki ; ð1aÞ ProbðYi ¼ 1Þ ¼ Zi ¼ U a þ Rj cj Dji þ Rk bk Xki þ HHIi j þ Rj kj Dji ;

ð1bÞ

where Prob(Yi = 1) = Zi is the probability a lender denies an application; Dji is a zero–one indicator of demographic group affiliation; the Xki are observable loan-applicant characteristics; the bk are the weights that lenders apply to the characteristics; and HHIi is

Lending technologies, lending specialization, and minority access

279

the Herfindahl–Hirschman index (HHI) of loanmarket concentration.2 Models (1a) and (1b) provide different perspectives on equal access to credit. In Model (1a) a positive, statistically significant estimate of cj suggests a higher denial probability for a loan applicant from demographic group j compared with a White applicant, but without suggesting a reason. Model (1b) seeks to disentangle statistical discrimination (the belief that group membership is correlated with unobserved individual characteristics) from tastebased discrimination (prejudice). Becker (1957) theorizes that greater loan-market competition (lower HHI) should discourage taste-based discrimination. In Model (1b) a positive, statistically significant estimate of kj suggests taste-based discrimination while a positive, statistically significant estimate of cj suggests statistical discrimination. Researchers who estimate Models (1a) and (1b) usually report evidence consistent with unequal access to small-business credits. Blanchflower et al. (2003) and Blanchard et al. (2008) estimate Model (1a) on data from the 1998 SSBF, and Blanchflower et al. estimate (1a) on data from the 1993 National Survey of Small Business Finances (NSSBF). Both find positive and significant cj estimates, especially for African-Americans and Hispanic-Americans.3 Cavalluzzo and Cavalluzzo (1998), Cavalluzzo et al. (2002), and Cavalluzzo and Wolken (2005) (hereafter CC, CCW, and CW, respectively) estimate Model (1b) using different data sets: the 1988–1989 NSSBF, the 1993 NSSBF, and the 1998 SSBF, respectively. All three studies find some evidence consistent with unequal access in the form of positive and significant cj and kj estimates, particularly for African-Americans. Models (1a) and (1b) produce differences in estimated loan denial rates for White and minority

owners that appear economically significant as well as statistically significant.4 Fair-access researchers generally estimate Models (1a) or (1b) on data for loan applications of all kinds with minimal controls for different lender and loan types;5 but this approach may lead to biased inference about minority credit-market access for two reasons. First, the approach assumes that lenders use the same decision criteria to approve small-business loans of all kinds. This assumption seems at variance with the small-business lending literature, which recognizes that lenders use different ‘‘lending technologies’’— combinations of information sources, loan contract structure, and monitoring mechanisms—to make loans of different costs and risks (Berger and Udell 2006; Berger and Black 2007). Second, the standard approach assumes that all lenders use the same decision criteria to approve small-business loans of a given kind. This assumption seems inconsistent with several studies on lender specialization by credit risk and expertise (Remolona and Wulfekuhler 1992; Carey et al. 1998; Daniels and Ramirez 2008). These observations argue for modifying the standard loan denial models to allow for differences in lending technologies and lender specialization, and for testing minority access to different types of credits by permitting minority denial probabilities to differ by loan type and lender type.

2

4 CC (1998) estimate a difference in denial rates for White and minority owners of about 33 percentage points; CCW (2002) estimate a difference of about 28 percentage points in highconcentration markets. Blanchflower et al. (2003) report differences of 20 to 30 percentage points, while Blanchard et al. (2008) report differences of about 15 percentage points. 5 CCW (2002) and Blanchard et al. (2008) include in their models indicator variables which permit the probability of loan denial to differ by loan type and lender type but neither permits denial probabilities for Whites and minorities to differ by loan type or lender type as does Model (2), presented below. Neither paper makes a connection to the literatures on lending technologies and lending specialization.

The HHI index measures market power by shares of deposits. HHI is computed by squaring percentage shares and summing. Thus a market with a single lender has an HHI of (1002 =) 10,000 and a market with 100 equal-sized lenders has an HHI of (100 9 12 =) 100. 3 Blanchflower et al. (2003) and Blanchard et al. (2008) are each subject to a criticism that is addressed in this paper. Blanchflower et al. estimate a loan denial model without a loan application model, opening their results to possible sample selection bias. Blanchard et al. estimate their loan denial model jointly with a loan application model but use unweighted survey data, which prevents the drawing of population inferences.

2.2 Model and hypotheses In this section we make three predictions about minority access to credit which we test using a modified loan denial model recognizing two categories of loans (credit lines and non-line-of-credit loans) and two categories of lenders (commercial banks and nonbank lenders).

123

280

Our first prediction concerns minority access to new bank credit lines relative to other small-business loans.6 We hypothesize that the lending technology that banks use to extend new credit lines produces soft information which reduces the observance of statistical discrimination towards minorities and makes minorities’ access to credit fairer. Credit lines are bank commitments to lend up to preset amounts for relatively short periods, usually less than 3 years, usually for working-capital purposes, and often without project- or asset-specific collateral. Applying for credit lines produces soft information: by seeking short-maturity loans, applicants send credible signals of firm quality to asymmetrically informed lenders (Myers 1977; Flannery 1986).7 Ortiz-Molina and Penas (2008) find evidence that banks shorten the maturities of credit lines offered to acutely opaque firms, presumably to improve the information content of the signal in firms’ loan applications. We hypothesize that banks use the signal in applicants’ choice of loan type together with hard information on applicants’ observable risk characteristics when deciding whether to extend credit lines. When asymmetrically informed banks limit their information sets to applicants’ observable risk characteristics, minority group affiliation as a proxy for unobservable risk characteristics has its greatest apparent value. However, when banks include the soft information on loan-type choice in their information sets, minority status loses value as an information proxy because the soft information provides a more reliable signal of lenders’ unobservable risk characteristics. This leads us to expect that banks’ line-of-credit lending to minority owners is less likely to exhibit statistical discrimination.8 Hence we predict: 6

We hypothesize that banks are not observed to discriminate statistically against minority firm owners when considering applications for credit-line renewals because the soft information that banks obtain from monitoring original credit lines supersedes the noisy information in minority status about unobservable applicant risk characteristics. As discussed in Sect. 3, the credit-line data in the 1998 SSBF is exclusively on new credit lines, unlike the 1993 NSSBF or the 2003 SSBF. 7 Footnote 16 presents statistics on the average terms to maturity of credit lines and non-line-of-credit loans in our sample. 8 The use of credit scoring does not preclude the use of soft information. Berger et al. (2005) classify banks that adopt credit scoring as either ‘‘rules banks’’ or ‘‘discretion banks,’’ with the latter leaving discretion to loan officers to accept loans

123


Prediction 1 With respect to loan applications, minority access to credit appears fairer for bank credit lines than for bank non-line-of-credit loans. Our second prediction concerns minority access to non-line-of-credit loans (motor vehicle loans, mortgages, equipment loans, and leases) from banks relative to nonbank lenders. These loans typically have longer terms than credit lines and are usually related to specific assets, which often collateralize the loan. We hypothesize that lender specialization leads to the observance of less statistical discrimination by nonbanks than banks when lenders are asymmetrically informed.9 We hypothesize that banks are more likely than nonbanks to use the noisy signal in minority status for several reasons: (i) the longer terms on non-line-ofcredit loans do not produce a low-cost (to the bank) credible signal of firm quality, unlike credit lines; (ii) non-line-of-credit loans are potentially one-shot loans, unlike credit lines, so the probability of recouping information costs over several loans is lower; (iii) the more standardized contracts for non-line-of-credit loans facilitate more competitive loan markets, which further encourages banks to minimize informationgathering costs. We hypothesize that nonbank lenders have fewer incentives than banks to use the noisy signal in minority status, again for several reasons: (i) they may put less weight on applicants’ creditworthiness because loans are tied to a product sale, as in the case of loans from captive finance companies; (ii) they may have superior ability to repossess and redeploy loan collateral should a borrower default due to complementary lines of business; (iii) higher capital ratios may allow nonbank lenders to take greater risks in lending. Thus we predict: Prediction 2 With respect to loan applications, minority access to credit appears fairer for non-lineof-credit loans from nonbanks than from banks.

Footnote 8 continued and set loan terms. They note that all rules banks permit some judgmental overrides. 9 Again, we hypothesize that neither banks nor nonbanks are observed to discriminate statistically when they have a strong prior relationship with a minority applicant because they regard the applicant’s soft information as a more reliable indicator of unobserved risk characteristics than minority status. Chakraborty and Hu (2006) find evidence that banks adjust collateral requirements for non-line-of-credit loans the more bankprovided services a loan applicant uses.


Our third prediction concerns minority access to credit in markets having different degrees of competitiveness. Following prior research we observe that lenders in less competitive markets potentially have less incentive to incur the costs of acquiring specialized information about loan applicants and potentially have greater opportunity to exercise market power.10 Thus we predict: Prediction 3 With respect to loan applications, minority access to credit appears fairer in credit markets that are more competitive than less competitive (have lower market concentration than higher market concentration). We test our predictions using Model (2) shown in Table 1. Model (2) has Models (1a) and (1b) as special cases.11 Model (2) includes two sets of terms not found in (1a) and (1b). The first set uses LOC, a zero–one indicator for a credit-line application: LOCi [d ? Rj gj Dji]. Inclusion of LOC permits the probability of denial to differ between credit lines and other loans, with the coefficient d measuring the difference for White loan applicants. The LOC 9 Dj terms permit each demographic group to have a difference in denial probabilities on credit lines and other loans that is different from d, the difference being gj. The second set of terms uses BANK, a zero– one indicator for a loan application made to a commercial bank: BANKi [f ? Rj hj Dji]. The f coefficient measures the difference in the probability of denial between a bank and a nonbank lender on a loan application presented by a White firm owner. The BANK 9 Dj terms permit each demographic group to have a difference in denial probabilities at banks and nonbanks that differs from f, the difference being hj. Table 1 also shows special cases of our model for different combinations of applicant, loan, lender, and market concentration. Consider an applicant in a low-concentration loan market applying for a non-line-of-credit loan to a nonbank (i.e., HHI_HIi = LOCi = BANKi = 0). For White applicants (i.e.,

10

It is also possible that less competitive markets allow lenders to discriminate in the sense of Becker (1957), whereas competitive markets should eventually drive out prejudicially discriminating firms because they would be less profitable. 11 Model (2) uses HHI_HIi, an indicator variable for a highconcentration loan market, rather than the value of the HHI index, which is not included in the 1998 SSBF.

281

Dji = 0 V j), Model (2) simplifies to Zi = U(Ci) = U(a ? Rk bk Xki); hence, a is the adjustment that a lender makes to the index of weighted applicant characteristics for a White firm owner, reflecting judgment and/or soft information. Changing the loan market concentration from low to high changes the White-owner adjustment from a to a ? j. Changing the lender from a nonbank to a bank changes the adjustment from a ? j to a ? f ? j. Finally, changing the loan applied for from a bank non-line-of-credit loan to a bank credit line changes the White-owner adjustment from a ? f ? j to a ? d ? f ? j. We present analogous special cases for non-White applicants. Finally, we show the restrictions needed to test fair access to credit for minority applicants, that is, minority adjustments jointly zero and, hence, equal to the adjustments made for White-owned firms. We test whether minority applicants have fair access to nonbank non-line-of-credit loans (the NB_NLOC hypothesis), bank non-line-of-credit loans (the B_NLOC hypothesis), and bank credit lines (the B_LOC hypothesis). We interpret failure to reject a hypothesis as evidence of fair access to credit.12 Tests of the market-access hypotheses also test our three predictions. Our prediction of fairer minority access to bank credit lines than to bank non-line-ofcredit loans (Prediction 1) would be supported by failure to reject the B_LOC hypothesis plus rejection of the B_NLOC hypothesis. Our prediction of fairer access to non-line-of-credit loans from nonbanks than from banks (Prediction 2) would be supported by failure to reject the NB_NLOC hypothesis plus rejection of the B_NLOC hypothesis. Our prediction of fairer access to markets that are more competitive than less competitive (Prediction 3) would be supported by failure to reject the B_LOC, B_NLOC, and NB_NLOC hypotheses in high-competition markets but not in low-competition markets. Estimates of Model (2) will be biased if the observed attributes (values of the Xki) of firm owners that did and did not apply for loans differ systematically. To prevent this sample selection problem we follow Van de Ven and Van Pragg (1981) and estimate our loan denial model jointly with a loan application model using maximum-likelihood 12

We do not test for equal access to nonbank line-of-credit loans due to the small number of minority applicants for such loans in our sample.

123

282


Table 1 Loan application denial model and hypothesis tests ð2Þ ProbðYi ¼ 1Þ ¼ Zi ¼ U a þ Rj cj Dji þ Rk bk Xki þ HHI HIi j þ Rj kj Dji þ LOCi d þ Rj gj Dji þ BANKi f þ Rj hj Dji where Prob(Yi = 1) = probability the ith loan application is denied Dji = a zero–one indicator of membership in the jth demographic group, j = 1,…,J Xki = kth attribute pertinent to the ith firm-owner’s creditworthiness observable by lenders, k = 1,…,K HHI_HIi = a zero–one indicator of high-concentration (HHI_HIi = 1) and low-concentration (HHI_HIi = 0) loan markets LOCi = a zero–one indicator of line-of-credit (LOCi = 1) and non-line-of-credit (LOCi = 0) loans BANKi = a zero–one indicator of commercial bank (BANKi = 1) and nonbank (BANKi = 0) lenders The table below shows Model (2) for different combinations of applicant, loan, lender, and market concentration type. Restrictions to test the fair-access hypotheses are also shown. To simplify notation we let Ci = a ? Rk bk Xki (Market concentration)

BANKi = 0 (nonbank lender)

BANKi = 1 (bank lender)

LOCi = 0 (non-lineof-credit loan)

LOCi = 0 (non-line-of-credit loan)

LOCi = 1 (line-of-credit loan)

Dji = 0 (White owner) HHI_HI = 0 (low)

Zi = U(Ci)

Zi = U(Ci ? f)

Zi = U(Ci ? d ? f)

HHI_HI = 1 (high)

Zi = U(Ci ? j)

Zi = U(Ci ? f ? j)

Zi = U(Ci ? d ? f ? j)

Dji = 1 (non-White owner) HHI_HI = 0 (low)

Zi = U(Ci ? cj)

Zi = U(Ci ? f ? cj ? hj)

Zi = U(Ci ? d ? f ? cj ? gj ? hj)

Restrictions

NB_NLOC hypothesis: cj = 0

B_NLOC hypothesis: cj ? hj = 0

B_LOC hypothesis: cj ? gj ? hj = 0

HHI_HI = 1 (high)

Zi = U(Ci ? cj ? j ? kj) Zi = U(Ci ? f ? cj ? hj ? j ? kj) Zi = U(Ci ? d ? f ? cj ? gj ? hj ? j ? kj)

Restrictions

NB_NLOC hypothesis: c j ? kj = 0

B_NLOC hypothesis: cj ? hj ? kj = 0

estimation. The dependent variable of our loan application model is 1 if the ith firm applied for a loan, and 0 otherwise; the independent variables are similar to those in Model (2). 3 Data and descriptive statistics We estimate our model on data from the 1998 Survey of Small Business Finances, conducted in 1999 and 2000 on a nationally representative sample of 3,561 small businesses (fewer than 500 full-time employees) in existence in December 1998. The SSBF includes extensive financial and nonfinancial information on participating firms, including detailed data about the most recent loan applications of 962 firms.13 13

Data considerations led us to winnow the sample down slightly. We excluded 76 of the 3,561 firms because they had

123

B_LOC hypothesis: cj ? gj ? hj ? kj = 0

The 1998 SSBF is well suited for examining fair access to credit for two reasons. First, the survey extensively sampled firms owned by African-Americans, Hispanics, and Asians, potentially producing

Footnote 13 continued zero or negative assets; this left 3,485 firms including 701 minority-owned firms, of which 259, 199, and 258 were owned by African-Americans, Asians, and Hispanics, respectively. (We use these observations to estimate our loan application models.) Of these 3,485 firms, 952 firms applied for a loan; however 64 firms lacked data on loan type applied for or lender type applied to, and another 18 reported having their loan applications both denied and approved. We excluded these observations, leaving 870 firms that applied for credit. Firms owned by African-Americans, Asians, Hispanics, ‘‘Other’’ minorities, and White Americans number 68, 43, 70, 7, and 688, respectively. Because the number of ‘‘Other’’ minority loan applicants is so small we dropped them when estimating Model (2), leaving 863 observations on which to estimate our loan application denial models.


283

Table 2 Loan application denial rates Demographic groupa:

WHITE

MINORITY

AFROAM

HISPANIC

Nonbank non-line of credit (NB_NLOC)

11%

25%

47%

24%

Bank non-line of credit (B_NLOC)

(148) 9%

(38)** 53%

(9)** 56%

(19) 63%

(316)

(58)***

(20)***

(28)***

32%

50%

48%

39%

(174)

(73)***

(33)

(23)

17%

46%

53%

47%

(686)

(177)***

(68)***

(70)***

Loan type applied for

Bank line of credit (B_LOC) All loans applied for

This table shows denial rates for firms that reported on their most recent loan applications. Percentages of respondents that were denied loans are based on population-weighted data. Numbers of respondents who applied for loans appear in parentheses (863 firms) a

Classifications are based on the majority of a firm’s owners. MINORITY firms are at least 50% owned by African-Americans (AFROAM), Asian-Americans or Hispanic-Americans. WHITE firms are firms not classified as MINORITY ***, ** Statistically different from the percentage for White-owned firms at the 1 and 5% levels, respectively

more reliable data on these firms, which are often underrepresented in other databases (Wolken et al. 2001).14 Second, the survey was conducted shortly before 2002, the possible start of the easy-credit period that preceded the financial crisis of 2007– 2008. Although the 1993 NSSBF also has these two advantages, we do not use this survey because it combines applications for new credit lines with applications for credit-line renewals, and distinguishing between them is important for testing fair access: if lenders use ethnicity as a proxy for unobservable borrower characteristics when asymmetrically informed, minority owners are more likely to encounter statistical discrimination on applications for new credit lines than for renewals, when the lender’s information should be more symmetric due to the past monitoring of the existing line. Although the 2003 SSBF is the most recent survey, it did not oversample minority-owned firms, resulting in far fewer observations on minority-owned firms. The drop in number of

14

The 1998 SSBF actually oversampled minority-owned firms. Oversampling biases summary statistics (e.g., means and medians) with respect to population parameters unless the observations are weighted. All statistical and econometric analyses reported below were weighted using weights provided with the 1998 SBBF. For technical details of the 1998 SSBF, see Haggerty et al. (2001).

African-American-owned firms is particularly sharp, falling from 259 in the 1998 SSBF to 109 in the 2003 SSBF. Stratifying the minority observations by loan type, lender type, and market competitiveness, as required by Model (2), produced categories with too few observations to permit reliable estimation. Table 2 reports descriptive statistics for denial rates on loan applications. White small-firm owners report substantially higher denial rates on bank credit lines than other loan types: 32% versus about 10%. Non-White firm owners report denial rates consistently higher than White owners, and the differences are usually statistically significant. Differences in minority/White denial rates are greater for bank nonline-of-credit loans than for bank credit lines, and greater for non-line-of-credit loans from banks than from nonbanks. These results are consistent with Predictions 1 and 2. Table 3 defines the independent variables used in Model (2). The list is extensive because all attributes relevant to credit quality and observable by lenders must be included for the model to yield evidence on fair credit-market access. Many of the variables have appeared in other fair-access studies (CC 1998; CCW 2002; CW 2005; Blanchflower et al. 2003; Blanchard et al. 2008). In addition to variables representing owners’ demographic, educational, ownership, and wealth characteristics, we include variables representing firm characteristics and loan

123

284


Table 3 Independent variable definitions Demographic group MINORITY = 1 if firm is at least 50% owned by African-, Hispanic- or Asian-Americans or other ethnic minority AFROAM = 1 if firm is at least 50% owned by African-Americans ASIAN = 1 if firm is at least 50% owned by Asian-Americans HISPANIC = 1 if firm is at least 50% owned by Hispanic-Americans FEMALE = 1 if firm is at least 50% owned by females Market and lending technology characteristics HHI_HI = 1 if the Herfindahl index for the firm’s location is 1,800 or more using bank and thrift deposits LOC = 1 if firm’s most recent loan application was for a line-of-credit loan BANK = 1 if firm’s most recent loan application was to a commercial bank Owner characteristics Education/experience POST_HS = 1 if principal owner received some education beyond high school but not a 4-year college degree COLLEGE = 1 if principal owner holds at least a 4-year college degree LN_EXPER = Log of 1 ? principal owner’s years of business experience Control/wealth OWNR_% = Percentage of firm owned by principal owner, in decimal terms MANAGER = 1 if principal owner manages the firm LN_NET_WRTH = Log of principal owner’s net worth FAMILY = 1 if firm is more than 50% owned by a single family Firm characteristics Financial LN_ASSETS = Log of firm’s 1998 total assets LN_SALES = Log of firm’s 1998 sales revenue ROA = Firm’s 1998 pretax profits/firm’s 1998 total assets, winsorized at the 1st and 99th percentiles LN_EQUITY = Log of firm’s 1998 equity NEG_EQUITY = 1 if firm’s 1998 equity is negative Credit record BANKRUPT = 1 if firm or principal owner declared bankruptcy in the last 7 years JUDGMENT = 1 if legal judgment was made against principal owner in the last 3 years OWNR_PAY_LATE = 1 if principal owner of a proprietorship or partnership was delinquent on a financial obligation BUS_PAY_LATE = 1 if firm was delinquent on a financial obligation, including trade credit CREDIT_SCORE_LO = 1 if firm’s Dun & Bradstreet credit rating indicates a significant or high credit risk DENIED_TRADE_CR = 1 if firm was ever denied trade credit Relationships #_OF_RELATIONS = Number of sources of financial services used by the firm USE_TRADE_CR = 1 if firm used trade credit during fiscal year 1998 USE_OWNR_CCARD = 1 if firm used owners’ personal credit card for businesses expenses in 1998 USE_BUS_CCARD = 1 if firm used business or corporate credit cards for businesses expenses in 1998 Nonfinancial LN_AGE = Log of 1 ? firm’s age, in years LN_EMPLOYEES = Log of number of employees, including working owners LN_LOCATIONS = Log of number of firm’s locations C_CORP = 1 if firm is a C-corporation S_CORP = 1 if firm is an S-corporation NATIONAL = 1 if firm’s market is national or international

123


285

Table 3 continued MSA = 1 if firm is located in a metropolitan area Loan source characteristics LN_LENGTH = Log of 1 ? number of months firm has had a relationship with the loan source applied to NO_RELATION = 1 if the firm had no relationship with the loan source prior to the loan application PRIMARY = 1 if loan source applied to is the firm’s primary financial institution Control variables APPLY_N = 1 if loan was applied for in year N; N = 1997, 1998, 1999 or 2000 REGION_N = 1 if firm is headquartered in geographical region N; N = 1–8 INDUSTRY_N = 1 if firm is in industry N, based on its two-digit SIC code; N = 1–8

source characteristics.15 Model (2) also includes HHI_HI, LOC, and BANK. HHI_HI is an indicator for a banking market with a Hirschman–Herfindahl index above 1,800. Credit lines (LOC = 1) represent 35% of the loan applications; applications for capital leases, mortgages, motor vehicle loans, equipment loans, and other loans account for the other 65%. These non-line-of-credit loans usually have fixed interest rates, are collateralized, and have longer terms to maturity than do credit lines.16 Applications to commercial banks (BANK = 1) represent 72% of applications, with the remaining 28% of applications made to finance companies (10.6%), savings banks (3.4%), lease companies (3.0%), credit unions (2.6%), and other lenders (8.4%). Nonbank lenders differ from commercial banks in their regulation and scope of product and service offerings.17 15

The three loan source variables—LN_LENGTH, NO_RELATION, and PRIMARY—are highly collinear, leading us to drop NO_RELATION when estimating the models reported in Tables 6 and 7. Estimated models that include NO_RELATION in place of LN_LENGTH and PRIMARY are nearly identical in all other respects. 16 In the 1998 SSBF 82% of approved non-line-of-credit loans have fixed rates, compared with 45% of line-of-credit loans. Sixty-seven percent of approved non-line-of-credit loans are collateralized, versus 33% of approved credit lines. In addition, the average term on an approved non-line-of-credit loan is 68 months, versus 27 months on approved credit lines. These differences are significant at the 1% level for a two-tailed test. The 1998 SSBFs also includes a miscellaneous (‘‘other’’) loan application category, which we include with non-line-of-credit loans. In the 1998 SSBF the contract terms of approved ‘‘other’’ loans resemble the terms on approved leases, mortgages, auto, and equipment loans. 17 The results reported in Tables 6 and 7 are qualitatively similar if we drop savings banks and credit unions from the sample.

Table 4 presents summary statistics for the independent variables, by demographic group, for all surveyed firms and for all firms that applied for loans, metrics relevant to our loan application and loan denial models, respectively. Among all surveyed firms White-owned firms are better established, on average, than minority-owned firms. White owners have more years of experience and higher net worth; their firms have more assets, sales, equity, and employees, are older, more often C-corporations, and more often use trade credit. Only about one-quarter of White-owned firms have Dun & Bradstreet ratings in the significantor high-credit-risk categories compared with more than one-third of minority-owned firms. Although few surveyed firms reported prior bankruptcies, legal judgments against them, prior denials of trade credit or owners who made late payments, those that did were more often minority-owned. Among firms that applied for loans, firms owned by Whites are again larger and better established, on average, but other patterns also appear. White- and minority-owned applicant firms show almost no statistical difference in the six creditworthiness variables, unlike all White- and minority-owned firms. White-owned applicant firms more frequently reported having an established relationship with the lender they applied to and reported relationships of greater average length than did minority-owned applicant firms. A significantly higher proportion of African-American owners than White owners reported applying to banks for credit lines, behavior consistent with reputationbuilding by selecting lenders who monitor as suggested by Diamond (1991). Table 5 reports summary statistics for the independent variables by demographic group and loan applied for. For White-owned firms few variables

123

123 W

0.92 6.62 0.88

MANAGER

NET_WRTH (US $100k) FAMILY

2.41 0.22

EQUITY (US $100k)

NEG_EQUITY

0.07 0.31 0.27 0.05

OWNR_PAY_LATE

BUS_PAY_LATE

CREDIT_SCORE_LO

DENIED_TRADE_CR 2.08

0.03

JUDGMENT

Relationships #_OF_RELATIONS

0.02

BANKRUPT

Credit record

1.81

SALES (US $100k)

ROA

4.63 10.95

ASSETS (US $100k)

Financial

Firm characteristics

0.84

OWNR_%

Control/wealth

18.76

0.49

COLLEGE

EXPER (years)

0.28

POST_HS

Education/experience

Owner characteristics

2.00

0.08

0.37

0.32

0.13

0.06

0.04

0.21

0.86

2.04

4.85

2.01

3.33 0.91

0.94

0.88

15.02

0.48

0.28

1.73

5.47**

18.06***

0.15

19.45***

7.29***

4.20**

0.69

71.07***

0.62

48.62***

74.77***

52.20*** 4.70**

1.49

9.53

60.65***

0.25

0.00

0.42

2.94

0.08

0.32

0.43

0.09

0.06

0.02

0.30

2.77

1.36

17.90

6.82

8.46 0.83

0.91

0.81

17.25

0.47

0.27

0.68

12.49***

BANKb

0.29

W

0.35

0.36

F-statistic

0.47

0.25

0.74

0.42

0.29

2.84

0.09

0.35

0.36

0.16

0.05

0.05

0.23

0.86

1.40

7.30

2.98

2.99 0.89

0.94

0.86

14.21

M

0.50

0.11

0.48

2.45

3.76*

0.33

1.83

2.79*

29.30***

0.01

16.84***

24.85***

14.00*** 3.23*

2.58

4.81**

12.32***

0.00

0.38

1.89

1.78

7.47***

F-statistic

Firms that applied for credit

LOCb

HHI_HI

M

All firms in sample

Market and lending technology characteristics

Demographic group : Independent variables

a

Table 4 Univariate statistics: sample population and loan applicants by demographic group

3.00

0.15

0.39

0.41

0.19

0.09

0.12

0.31

0.52

1.69

5.59

1.83

2.38 0.84

0.95

0.88

14.86

0.43

0.34

0.81

0.56

0.31

AA

0.05

1.98

0.89

0.10

3.38*

0.71

4.74**

0.01

40.67***

0.16

17.57***

40.50***

17.41*** 0.04

1.53

5.53**

13.86***

0.42

3.29*

5.07**

7.87***

2.23

F-statistic

2.74

0.09

0.41

0.30

0.11

0.02

0.02

0.23

0.71

1.22

6.28

2.63

2.97 0.93

0.93

0.85

13.58

0.37

0.22

0.75

0.27

0.29

H

1.10

0.07

1.66

4.46**

0.09

3.26*

0.00

1.33

36.17***

0.05

15.72***

21.35***

11.90*** 7.85***

0.51

2.05

8.39***

2.20

0.76

1.20

1.65

3.86**

F-statistic

286 K. Mitchell, D. K. Pearce

a

0.35

USE_BUS_CCARD

0.14 0.78

S_CORP

NATIONAL

MSA

0.30

0.46

0.54

0.73 0.45

0.52

686

0.60

0.45

3.51 0.37

0.90

0.17

0.33

0.15

1.31

8.86

9.12

0.42

0.44

177

M

2.27

14.27*** 5.78**

24.03***

0.62

0.73

4.96**

0.32

8.47***

15.03***

0.21

2.92*

7.13***

F-statistic

68

0.51

5.06 0.33

0.85

0.20

0.34

0.08

1.37

7.38

9.14

0.41

0.44

0.64

AA

0.05

0.25 1.32

5.06**

1.06

0.59

15.13***

0.01

12.02***

6.85***

0.35

1.31

1.45

F-statistic

70

0.43

2.50 0.38

0.89

0.11

0.30

0.20

1.23

8.34

11.05

0.35

0.43

0.52

H

1.78

27.45*** 3.63*

11.48***

0.36

0.04

0.18

2.68

7.91***

8.37***

2.13

1.70

8.47***

F-statistic

Survey data available only for firms that applied for credit

***, **, * Statistically different from zero at the 1, 5, and 10% levels, respectively

b

Demographic group classifications are based on the affiliation of the majority of a firm’s owners. Minority-owned firms are at least 50% owned by African-Americans, AsianAmericans, Hispanic-Americans, Hawaiians, Alaskans or Native Americans. Analogous definitions apply to African-American- and Hispanic-owned firms. White-owned firms are firms not classified as minority-owned

a

W White-owned, M minority-owned, AA African-American-owned, H Hispanic-owned

This table presents population-weighted mean values of the independent variables defined in Table 3 by firm-owner demographic group affiliation and F-statistics for tests of differences in means between White-owned firms and other demographic groups

# of observations

0.52

0.74

0.14

0.29

0.23

1.36

12.38

11.99

W

PRIMARYb

49.40***

2.03

0.38

13.81***

1.19

37.24***

50.28***

6.92***

0.04

22.39***

F-statistic

5.54 0.25 742

0.89

0.16

0.23

0.15

1.21

6.20

10.77

M

Firms that applied for credit

LENGTHb (years) NO_RELATIONb 2,751

0.24

C_CORP

Loan source characteristics

1.41 0.21

LOCATIONS (#)

9.15

EMPLOYEES (#)

AGE (years)

13.85

0.46

Non-financial

0.65

USE_OWNR_CCARD

W

All firms in sample

USE_TRADE_CR

Demographic group : Independent variables

Table 4 continued

Lending technologies, lending specialization, and minority access 287

123

123

6.53

0.88

NET_WRTH (US $100k)

FAMILY

2.93

0.26

EQUITY (US $100k)

NEG_EQUITY

0.01

0.09 0.36

0.33

0.08

JUDGMENT

OWNR_PAY_LATE BUS_PAY_LATE

CREDIT_SCORE_LO

DENIED_TRADE_CR

3.26

0.67

#_OF_RELATIONS

USE_TRADE_CR

Relationships

0.02

BANKRUPT

Credit record

1.21

SALES (US $100k)

ROA

6.95

14.46

ASSETS (US $100k)

Financial characteristics

Firm characteristics

0.92

0.81

OWNR_MANAGER

Control/wealth OWNR_%

17.55

0.40

EXPER (years)

0.30

COLLEGE

0.35

POST_HS

Education/experience

Owner characteristics

HHI_HI

0.78

2.79

0.07

0.31

0.10 0.47

0.05

0.02

0.31

3.17

1.44

18.57

8.10

0.78

10.03

0.88

0.78

18.00

0.48

0.20

0.46

0.74

2.83

0.10

0.35

0.06 0.48

0.10

0.02

0.31

2.47

1.44

24.07

5.90

0.86

8.68

0.92

0.81

16.39

0.53

0.30

0.42

NB_NLOC B_NLOC B_LOC

1.75

3.33**

0.53

0.25

0.68 2.25

3.97**

0.01

0.53

0.31

0.13

1.01

1.07

3.09**

0.85

1.09

0.91

1.20

1.65

2.60*

1.58

0.74

3.31

0.07

0.39

0.16 0.45

0.08

0.03

0.18

0.52

2.33

6.91

2.51

0.90

2.12

0.92

0.88

12.72

0.47

0.20

0.21

0.49

2.66

0.04

0.32

0.11 0.32

0.01

0.05

0.15

1.17

0.85

7.69

3.62

0.88

3.27

0.93

0.82

15.21

0.43

0.26

0.32

0.64

2.73

0.11

0.34

0.18 0.33

0.06

0.03

0.26

0.84

1.70

7.65

2.87

0.90

3.32

0.97

0.87

14.15

0.55

0.23

0.29


Means, by loan type

Means, by loan type

F-stat

Minority-owned firms

White-owned firms

Market and lending technology characteristics

Independent variables

Table 5 Univariate statistics: loan applicants by demographic group and loan type

2.60*

1.36

0.85

0.19

0.53 0.78

1.69

0.20

1.02

1.95

1.38

0.03

0.45

0.05

1.37

0.70

0.57

0.80

0.79

0.17

0.62

F-stat

0.67

0.02

0.06

0.35

1.14 0.91

1.35

0.19

0.98

14.39***

1.05

3.59*

7.34***

0.09

7.75***

0.00

1.83

7.60***

0.37

1.22

2.32

NB_NLOC

11.83**

0.31

0.42

0.03

0.07 3.63*

3.26*

1.08

6.45**

12.99***

1.55

11.67***

11.76***

2.11

7.37***

1.35

1.23

3.24*

0.50

0.64

2.85*

B_NLOC

1.74

0.19

0.04

0.00

3.72* 3.85*

1.17

0.21

0.49

4.09**

0.06

4.47**

4.85**

0.57

2.19

2.40

2.09

3.10*

0.06

1.18

2.72

B_LOC

F-statistics testing differences in means between Whiteand minority-owned firms, by loan type

288 K. Mitchell, D. K. Pearce

0.46

0.66

0.75

0.26

PRIMARY 316

0.67

0.17

7.37

0.58

174

0.61

0.26

5.09

0.82

0.39 0.19

0.19

1.41

13.02

10.81

0.50

1.13

25.36***

3.98**

7.39***

5.92***

3.55** 4.30**

1.12

0.25

2.99*

3.46**

0.52

38

0.05

0.74

1.23

0.85

0.42 0.08

0.15

1.14

9.71

7.68

0.50

0.56

58

0.56

0.23

4.42

0.90

0.27 0.12

0.18

1.48

8.59

9.61

0.37

0.22

73

0.61

0.24

4.07

0.91

0.36 0.26

0.11

1.24

8.76

9.25

0.45

0.55


1.44

2.75* 0.06

2.81*

5.77**

0.11

9.01***

0.15

0.18

NB_NLOC

34.15*** 11.98***

16.17*** 21.12***

9.41*** 11.92***

0.37

1.00 2.86*

0.41

2.63*

0.10

1.14

0.58

8.75***

F-stat

1.58

0.54

8.84***

20.28***

0.04 0.02

1.59

0.88

11.80***

7.58***

0.70

16.14***

B_NLOC

0.00

0.02

1.25

3.08*

0.14 1.17

1.92

1.77

2.08

2.01

0.36

0.15

B_LOC

F-statistics testing differences in means between Whiteand minority-owned firms, by loan type

***, ** and * denote means that are statistically different at the 1, 5, and 10% levels, respectively, as determined by F-statistics

This table presents population-weighted mean values of the independent variables defined in Table 3 for White- and minority-owned firms by loan type applied for together with F-statistics testing equality of the means. Loan types are non-line-of-credit loans from nonbanks (NB_NLOC), non-line-of-credit loans from banks (B_NLOC), and credit lines from banks (B_NLOC)

148

0.32

NO_RELATION

# of observations

4.02

LENGTH (years)

Loan source characteristics

MSA

0.25 0.12

0.25 0.06

S_CORP NATIONAL

1.34 0.26

1.40

0.27

14.37

13.17

C_CORP

10.54

EMPLOYEES (#)

0.49 0.44

LOCATIONS (#)

11.47

AGE (years)

Nonfinancial

0.51

USE_BUS_CCARD


Means, by loan type

Means, by loan type

F-stat

Minority-owned firms

White-owned firms

USE_OWNR_CCARD

Independent variables

Table 5 continued

Lending technologies, lending specialization, and minority access 289

123

290

show significant differences in means across the three loan types, and those that do mainly involve nonfinancial and loan source characteristics. The average White-owned firm that applied for a bank credit line is younger, more often an S-corporation, and more often selling nationally than firms applying for other loans. This pattern is consistent with informationally opaque firms having good growth opportunities seeking to build reputation by choosing lenders that monitor, as suggested by Diamond (1991). Among White-owned firms, firms that applied to banks for non-line-of-credit loans were better known to lenders than other firm applicants: they more often applied to lenders with whom they had prior contact, more often applied to their primary financial service providers, and reported longer relationships with those providers. Minority-owned firms show even more similarity across the three loan types than do White-owned firms, with the main differences occurring in loan source characteristics. Minority-owned firms that applied to nonbanks were virtually unknown to lenders: about three-quarters had no prior relationship with the lender, and when prior relationships existed they were short, about 1.25 years on average. The final three columns of Table 5 test equality of means of the independent variables for White- and minority-owned firms by loan type applied for. For all three loan types White-owned firms have greater assets, sales, and equity. If larger firms more often use standard accounting practices and obtain audited financial statements, the finding of larger size may suggest that White-owned firms are informationally more transparent. White- and minority-owned firms that applied for bank credit lines (B_LOC) show few other significant differences. More White–minority differences appear between firms that applied for nonbank non-line-of-credit loans (NB_NLOC): in comparison with White owners, minority owners have less experience, have less net worth, and own younger firms; minority owners are also less well known to the lenders applied to, judging from the means for LN_LENGTH, NO_RELATION, and PRIMARY. Minority owners that applied for bank non-line-of-credit loans (B_NLOC) also have less experience, have less net worth, and own younger firms than do White owners; in addition they employ fewer people, on average. Minority B_NLOC applicants appear nearly as well known to lenders as

123


White-owned firms: F-statistics fail to show significant White–minority differences in proportions of firms applying to primary banks or having no prior relationship with banks applied to, although the average length of prior relationships is less, 4.4 years compared with 7.4 years. In addition, minority applicants for B_NLOC loans less frequently report using trade credit and owners’ credit cards for business purposes, suggesting less credit information available on these firms.18

4 Empirical results 4.1 Main results Table 6 Panel A reports selected parts of our estimated loan-application denial models.19 Equations 6.1–6.4 include FEMALE and MINORITY as demographicgroup indicators. Equation 6.1 has the same form as Model (1a). The estimated coefficient of MINORITY is positive and significant, as reported by other researchers. Equation 6.2 has the same form as Model (1b), adding HHI_HI and HHI_HI 9 MINORITY.20 The new terms reduce the estimated coefficient of MINORITY but it remains positive and significant. The estimated coefficient of HHI_HI 9 MINORITY is also positive and significant, consistent with less equal market access in high-concentration loan markets as theorized by Becker (1957). The estimated coefficient of HHI fails to achieve significance.

18

White- and minority-owned firms that applied for B_NLOC loans differ significantly in the proportion of firms with negative equity, having prior legal judgments against them, and paying business obligations late, with minority-owned firms having the more favorable characteristics. 19 Because the loan approval equation is estimated with the loan application model, there is a potential identification issue. As discussed in Cavalluzzo and Wolken (2005), the data collected in the SSBF surveys are likely relevant to both application and approval decisions. We follow them in achieving identification through the nonlinearity of the model. As robustness tests, we also estimated models that excluded some variables from the denial equation that were retained in the application equation and obtain qualitatively similar results. We also estimated a denial equation without the application equation and again get similar results for our hypothesis tests. 20 Equation 6.2 also includes FEMALE 9 HHI_HI, as do (6.3) and (6.4). Its estimated coefficient is always statistically insignificant at the 10% level for a two-tailed test.


291

Table 6 Estimated loan application denial models and hypothesis tests Panel A. Model estimatesa Independent variables

Equation: (6.1) Coefficient

(6.2)

(6.3)

(6.4)

HHI_HI

j

-0.014

0.012

(0.177)

(0.198) 0.647*** (0.186)

LOC BANK

d f

MINORITY

cj

MINORITY 9 HHI_HI

kj

MINORITY 9 LOC

gj

MINORITY 9 BANK

hj

AFROAM

cj

(6.6)

(6.7)

(6.8)

0.037

-0.017

0.008

0.016

(0.190)

(0.175)

(0.194)

(0.182)

0.727***

0.639***

0.674***

(0.202)

(0.190)

(0.215)

0.263

-0.001

0.222

-0.007

(0.182)

(0.197)

(0.176)

(0.185)

0.611***

0.441**

0.550***

0.352

(0.231)

(0.191)

(0.209)

(0.282)

0.726**

0.787**

0.642**

(0.307)

(0.331)

(0.304)

(6.5)

-0.890*** (0.280) 0.802** (0.352) 0.716**

0.514*

0.528*

0.948*

(0.341)

(0.276)

(0.310)

(0.515)

0.987**

1.031**

0.773*

(0.460)

(0.484)

(0.462)

AFROAM 9 HHI_HI

kj

AFROAM 9 LOC

gj

AFROAM 9 BANK

hj

HISPANIC

cj

HISPANIC 9 HHI_HI

kj

HISPANIC 9 LOC

gj

HISPANIC 9 BANK

hj

#_OF_ RELATIONS

bk

LN_LENGTH

bk

PRIMARY

bk

(0.170)

(0.165)

(0.206)

(0.196)

(0.176)

(0.160)

(0.196)

(0.206)

Constant

a

2.181***

2.130**

1.819*

2.203**

2.271***

2.278***

1.984**

2.376***

-0.576 (0.411) -0.035 (0.584) 0.479*

0.415*

0.544*

0.217

(0.292)

(0.244)

(0.278)

(0.343)

0.439

0.622

0.214

(0.371)

(0.426)

(0.397) -1.454*** (0.446) 1.013** (0.457)

-0.165*** -0.162*** -0.178*** -0.196*** -0.186*** -0.177*** -0.191*** -0.227*** (0.060)

(0.056)

(0.065)

(0.058)

(0.060)

(0.055)

(0.066)

(0.054)

0.025

0.025

0.048

0.044

0.026

0.026

0.047

0.035

(0.033)

(0.034)

(0.039)

(0.036)

(0.029)

(0.032)

(0.037)

(0.034)

-0.249

-0.303**

-0.563*** -0.523*** -0.245

-0.318**

-0.550*** -0.497**

(0.781)

(0.851)

(1.014)

(0.929)

(0.723)

(0.825)

(0.984)

(0.860)

q

-0.760

-0.710

-0.584

-0.663

-0.834

-0.737

-0.620

-0.739

Pseudo-R2

0.277

0.302

0.355

0.340

0.253

0.300

0.352

0.329

# of observations, denial model

863

863

863

863

863

863

863

863

123

292


Table 6 continued Panel B. Hypothesis testsb Market concentration

Equation: Demographic group: Restriction tested

(6.4) MINORITY

(6.8) AFROAM

(6.8) HISPANIC

NB_NLOC hypothesis (BANK = 0, LOC = 0) HHI_HI = 0 (low)

cj = 0

1.56

3.38*

0.40

HHI_HI = 1 (high)

c j ? kj = 0

7.20***

6.94***

0.78

14.53***

4.89**

8.85***

18.86***

9.89***

8.61***

B_NLOC hypothesis (BANK = 1, LOC = 0) HHI_HI = 0 (low)

c j ? hj = 0

HHI_HI = 1 (high) c j ? hj ? kj = 0 B_LOC hypothesis (BANK = 1, LOC = 1) HHI_HI = 0 (low)

c j ? gj ? hj = 0

1.13

0.91

0.58

HHI_HI = 1 (high)

cj ? gj ? hj ? kj = 0

6.86***

5.29**

0.00

a

This table reports selected coefficient estimates and standard errors of the loan application denial model presented in Table 1. The dependent variable is 1 if a loan application is denied and 0 if approved. Each model was estimated jointly with a loan application model to correct for possible sample selection bias. The estimated models include all the variables shown in Table 3. ***, **, * Statistically different from zero at the 1, 5, and 10% levels, respectively, for a two-tailed test b This table presents F-statistics testing restrictions on estimated coefficients in Eqs. 6.4 and 6.8 reported in Panel A corresponding with the B_LOC, B_NLOC, and NB_NLOC hypotheses presented in Table 1 Panel B. ***, **, * The F-statistic rejects the restriction at the 1, 5, or 10% level, respectively

Equation 6.3 adds LOC and BANK. The estimated coefficient of LOC is positive and significant, indicating a higher probability of denial on credit-line applications than other applications. The estimated coefficient of BANK is also positive but not significant, implying similar denial probabilities on applications to banks and nonbanks. The estimated coefficients of MINORITY and HHI 9 MINORITY are little changed from (6.2). Equation 6.4 adds MINORITY 9 LOC and MINORITY 9 BANK, giving it the form of Model (2).21 The estimated coefficients of MINORITY 9 LOC and MINORITY 9 BANK are both statistically significant, and negative and positive, respectively. Estimates of the remaining coefficients are little changed from (6.3) except that the estimated coefficient of MINORITY is now insignificant.22 21

Equation 6.4 also includes FEMALE 9 LOC and FEMALE 9 BANK. The estimated coefficients of both variables are statistically insignificant at the 10% level for a two-tailed test. 22 To explore whether different urban–rural distributions of small-business owners across demographic groups might exert a separate influence on loan denial probabilities, we reestimated (6.2–6.4) after adding an interaction term between minority and MSA. The coefficient estimates of this interaction term were always statistically insignificant at conventional levels, as were the estimated coefficients of MSA. Thus

123

Table 6 Panel B reports F-statistics for tests of the B_LOC, B_NLOC, and NB_NLOC hypotheses on Eq. 6.4. In low-concentration markets we are unable to reject the NB_NLOC hypothesis (equal access to nonbank, non-line-of-credit loans) or the B_LOC hypothesis (equal access to bank credit lines); we do, however, reject the B_NLOC hypothesis (equal access to bank non-line-of-credit loans). In high-concentration markets we reject all three equal-access hypotheses. This result stems from the large estimated coefficient of MINORITY 9 HHI_HI, kj, which appears in all the restrictions pertaining to high-concentration markets. The estimates support Predictions 1, 2, and 3. Prediction 1—fairer minority access to bank credit lines than bank non-line-of-credit loans—is supported in low-concentration markets by failure to reject the B_LOC hypothesis plus rejection of the B_NLOC hypothesis. Prediction 2—fairer minority access to non-line-of-credit loans from nonbanks than from banks—is supported in low-concentration markets by failure to reject the NB_NLOC hypothesis plus

Footnote 22 continued differences in the rural–urban distributions of White- and minority-owned firms do not appear to exert a separate influence. We thank an anonymous referee for pointing out this possible problem.


rejection of the B_NLOC hypothesis. Prediction 3— fairer minority access to credit in low-concentration markets than in high-concentration markets—is supported by failure to reject the B_LOC and NB_NLOC hypotheses in low-concentration markets plus rejection of these hypotheses in high-concentration markets. Returning to Panel A we re-estimate Eqs. 6.1–6.4 after replacing MINORITY with AFROAM, ASIAN, and HISPANIC; we report the estimates as Eqs. 6.5– 6.8.23 Estimates of the second set of equations are broadly similar to the first. In (6.5) the estimated coefficients of AFROAM and HISPANIC are positive, significant, and slightly higher and lower, respectively, than the estimated coefficient of MINORITY in (6.1). In (6.6) adding HHI_HI and the HHI_HI interactions reduces the estimated coefficients of AFROAM and HISPANIC but leaves them positive and significant. Applying for loans in high-concentration markets appears to increase denial rates for African-American owners but not for Hispanic owners: the estimated coefficients of HHI_HI 9 AFROAM and HHI_HI 9 HISPANIC are both positive, but only the former is significant. Equation 6.7 adds LOC and BANK; both have positively signed estimated coefficients but only the former is significant. Equation 6.8, which like (6.4) has the form of Model (2), shows slightly different patterns of loan denial for African-American and Hispanic owners, as evidenced by the estimated cj, kj, gj, and hj parameters, the coefficients of Dj, Dj 9 HHI_HI, Dj 9 LOC, and Dj 9 BANK, respectively. For Hispanic owners the estimates of cj, kj, gj, and hj have the same algebraic signs as in (6.4) but the estimate of kj is insignificant, implying similar access in low- and high-concentration markets. For African-American owners the estimates of cj, kj, and gj have the same signs as in (6.4) but the estimate of cj is larger in absolute value and the estimate of hj is effectively zero. The larger cj estimate implies higher denial rates for African-American owners, irrespective of lending technology or market concentration.24

293

Table 6 Panel B reports F-statistics for tests of the B_LOC, B_NLOC, and NB_NLOC hypotheses on Eq. 6.8. The results for Hispanic owners are the same in low- and high-concentration markets: we cannot reject the NB_NLOC or B_LOC hypotheses but can reject the B_NLOC hypothesis. The results for African-American applicants differ between low- and high-concentration markets. In low-concentration markets we cannot reject the B_LOC hypothesis, nor can we reject the NB_NLOC hypothesis at the 5% significance level, though we can at the 10% level; we can easily reject the B_NLOC hypothesis, however. In high-concentration markets we can easily reject all three market-access hypotheses. The estimates largely support Predictions 1, 2 and 3. For Hispanic firm owners Prediction 1—fairer access to bank credit-lines than bank non-line-of-credit loans—and Prediction 2—fairer access to non-lineof-credit loans from nonbanks than from banks—are supported by failure to reject the B_LOC and NB_NLOC hypotheses plus rejection of the B_NLOC hypothesis. Prediction 3—fairer access in low- than high-concentration markets—is not supported; instead market access to bank credit lines and nonbank noncredit-line loans appears to be fair, regardless of market competition. For African-American owners Prediction 1 is supported in low-concentration markets as is Prediction 2 (weakly). Prediction 3 is also supported: African-American owners’ access to bank line-of-credit and nonbank non-line-of-credit loans appears fairer in markets where concentration is low rather than high.25 4.2 Extension: differences in underwriting standards Like prior loan denial models, Model (2) assumes that all lenders use the same underwriting standards, weighting loan applicants’ observable attributes

25

23

We do not report results for Asian owners due to the small number of observations. Equations 6.5–6.8 also include FEMALE and interaction terms with FEMALE. 24 The estimated models for both the loan denial equation and the loan application equation for Eqs. 6.4 and 6.8 are presented in the Appendix. Similar estimates for the other models are available from the authors.

As noted by an anonymous referee, the results in Table 6 could reflect differences in the size distribution of firms owned by White and non-White owners if, for example, firms owned by non-Whites are smaller and, therefore, more prone to failure. To explore this possibly we re-estimated the models in Table 6 on two subsamples of firms: firms having 100 or fewer employees and firms having 50 or fewer employees. (These estimates are available upon request.) The re-estimated models and associated hypothesis tests show qualitatively similar results to those reported in Table 6.

123

294


similarly when computing credit quality scores. Blanchard et al. (2008) observe that, if lenders use different weights, imposing identical weights may bias the coefficient estimates of the demographic variables used to measure equality of credit-market access. In reference to Model (2), applicants’ observable attributes are the Xki, the weights are the bk, and the coefficients of the demographic variables—the Dj, Dj 9 HHI_HI, Dj 9 LOC, and Dj 9 BANK—are the cj, kj, gj, and hj used to test the NB_LOC, B_NLOC, and B_LOC hypotheses. In response to Blanchard et al.’s observation we explore two possibilities: first, that banks and nonbanks use different weights; and second, that banks use different weights on line-ofcredit applications than banks and nonbanks use on other loans. We modify Model (2) by replacing one term in the model, the weighted average of applicant attributes shown as (3a) below, first with (3b) and then with (3c): Ci ¼ a þ Rk bk Xki ;

ð3aÞ

Ci ¼ a þ Rk bk Xki þ Rk lk Xki BANKi ;

ð3bÞ

Ci ¼ a þ Rk bk Xki þ Rk mk Xki BANKi LOCi þ p BANK LOCi :

4.3 Extension: disparate treatment discrimination ð3cÞ

BANKi and LOCi are zero–one indicators of a bank loan application and a line-of-credit application, respectively. In (3b) Xki has the coefficient bk for nonbank loan applications and bk ? lk for bank loan applications. Hence if banks and nonbanks assess applicants differently, estimates of the lk should be statistically significant. Analogously, in (3c) Xki has the coefficient bk ? mk for bank credit-line applications and bk for all other loan applications; statistically significant mk estimates imply different weightings. Table 7 Panel A reports estimates of Model (2) modified by (3b) and (3c) as Eqs. 7.1 and 7.2, respectively; Panel B reports F-statistics for tests of the market-access hypotheses based on (7.1) and (7.2).26 Equation 6.4 and the associated F-statistics

26

We do not include interaction terms between BANK or BANK 9 LOC and the following variables due to lack of variation in the data: BANKRUPT, JUDGMENT, OWNR_PAY_LATE, and DENIED_TRADE_CR. In addition, we do not include interaction terms between BANK or BANK 9 LOC and the indicators for application year, region or industry. Equations 7.1–7.3 also include FEMALE, FEMALE 9 BANK, FEMALE 9 LOC, and FEMALE 9 HHI_HI.

123

from Table 6 appear alongside them to facilitate comparisons. We find little evidence that different lenders use different underwriting standards. Regarding differences between banks and nonbanks, an F-test performed on (7.1) fails to reject the null hypothesis that the lk estimates are jointly zero (F27,3417 = 0.47; p-value = 0.99); F-tests on subsets of the lk also fail to reject the null.27 Analogous statements apply to differences in underwriting standards between bank line-of-credit loans and other loans: the data fail to reject restricting all the mk in (7.2) to be zero (F27,3417 = 0.69; p-value = 0.88) or restricting some of the mk.28 A comparison of the cj, kj, gj, and hj estimates in (7.1), (7.2), and (6.4) shows only small differences. Consequently F-tests of the NB_NLOC, B_NLOC, and B_LOC hypotheses on (7.1) and (7.2) yield results qualitatively similar to those from (6.4). Thus it appears unlikely that imposing uniform underwriting standards biases test statistics for our market-access hypotheses.

The Federal Financial Institutions Examination Council (FFIEC), which sets uniform standards for examinations of financial institutions, notes that under the Equal Credit Opportunity Act of 1974 (ECOA) lenders may not use ‘‘different standards in determining whether to extend credit’’ to applicants in different demographic groups (FFIEC, p. ii). In terms of Model (2) compliance with the ECOA seems to enjoin the use of different attribute weights for applicants in different demographic groups; but if compliance with the ECOA is imperfect, imposing 27

The only lk estimates close to statistical significance at conventional levels appear on interactions of BANK with LN_SALES, USE_BUS_CCARD, and MSA. The probability of loan denial is unrelated to sales revenue at a nonbank but weakly negatively related to sales at a bank, a result consistent with the conventional wisdom that banks lend on the basis of cash flow. For firms that use business credit cards and firms located outside of metropolitan areas, loan denial probabilities are weakly lower at banks than at nonbanks. 28 The only mk estimate close to statistical significance at conventional levels appears on the interaction term between BANK 9 LOC and LN_NET_WRTH, the log of owner net worth. The coefficient estimate implies that greater net worth reduces an applicant’s denial probability more sharply on a bank credit-line application than on other applications.


295

Table 7 Extensions: estimated loan application denial models and hypothesis tests Panel A. Model estimatesa Independent variables

Equation: Coefficient

(6.4)

(7.1)

(7.2)

(7.3)

HHI_HI

j

0.037

-0.014

0.047

0.053

(0.190)

(0.177)

(0.185)

(0.214)

0.727***

0.769**

0.542*

0.866***

(0.202)

(0.304)

(0.256)

(0.246)

-0.001

2.655

-0.169

-0.080

(0.197)

(1.678)

(0.208)

(0.224)

0.352

0.398

0.409

2.780

(0.282)

(0.270)

(0.262)

(2.009)

0.642**

0.601*

0.658**

0.946**

(0.304)

(0.358)

(0.323)

(0.423)

d

LOC BANK

f

MINORITY

cj kj

MINORITY 9 HHI_HI MINORITY 9 LOC

gj

-0.890*** (0.280)

-0.905** (0.362)

-1.125*** (0.360)

-1.053*** (0.374)

MINORITY 9 BANK

hj

0.802**

0.629

0.788***

1.357***

(0.352)

(0.428)

(0.366)

(0.501)

2.203**

1.019

1.805*

1.900

(0.929)

(1.518)

(1.044)

(1.283) No

a

Constant ? Rk lk Xki 9 BANKi

No

Yes

No

? Rk mk Xki 9 BANKi 9 LOCi

No

No

Yes

No

? Rk nk Xki 9 MINORITY

No

No

No

Yes

q

-0.663

-0.756

-0.747

-0.510

Pseudo-R2

0.340

0.333

0.337

0.419

863

863

863

863

# of observations, denial model Panel B. Hypothesis tests Market concentration

b

Restriction tested

Equation (6.4)

(7.1)

(7.2)

(7.3)

NB_NLOC hypothesis (BANK = 0, LOC = 0) HHI_HI = 0 (low)

cj = 0

1.56

2.16

2.44

1.92

HHI_HI = 1 (high)

cj ? kj = 0

7.20***

5.50**

7.27***

3.39*

B_NLOC hypothesis (BANK = 1, LOC = 0) HHI_HI = 0 (low)

cj ? hj = 0

14.53***

5.96**

10.39***

3.85**

HHI_HI = 1 (high)

cj ? hj ? kj = 0

18.86***

6.19**

12.59***

5.56**

B_LOC hypothesis (BANK = 1, LOC = 1) HHI_HI = 0 (low)

cj ? gj ? hj = 0

1.13

0.26

0.09

2.26

HHI_HI = 1 (high)

cj ? gj ? hj ? kj = 0

6.86**

2.88*

4.06*

3.77*

a

This table reports selected coefficient estimates and standard errors of the model presented in Table 1 augmented with the interaction terms shown. Estimates of Eq. 6.4 from Table 6 are shown for comparison. The dependent variable is 1 if a loan application is denied and 0 if approved. Each model was estimated jointly with a loan application model to correct for possible sample selection bias. The estimated models include all the variables shown in Table 3 b

This table presents F-statistics testing restrictions on estimated coefficients in Eqs. 7.1, 7.2, and 7.3 reported in Panel A corresponding with the B_LOC, B_NLOC, and NB_NLOC hypothesis presented in Table 1 Panel B. F-statistics reported for Eq. 6.4 in Table 6 are shown for comparison. ***, **, * The F-statistic rejects the restriction at the 1, 5, or 10% level, respectively

123

296


uniform weights when estimating Model (2) potentially biases the estimated coefficients of the Dji, Dji 9 HHI_HIi, Dj 9 LOCi, and Dji 9 BANKi terms used to test for fair credit-market access (Blanchard et al. 2008). To explore this possibility we modify Model (2) by replacing (3a) above with: Ci ¼ a þ Rk bk Xki þ Rk nk Xki MINORITYi ;

ð3dÞ

where MINORITYi is the zero–one indicator of a minority loan applicant. The coefficient of Xki is bk for White owners and bk ? nk for minority owners; statistically significant estimates of the nk may suggest ‘‘disparate-treatment discrimination’’ (FFIEC, p. iii). We report our estimate of modified Model (2) in Table 7 Panel A as Eq. 7.3; we report F-statistics based on (7.3) in Panel B.29 We find little evidence that lenders use different underwriting standards for White and minority loan applicants. When we test the null hypothesis that the nk estimates in (7.3) are jointly zero, the data fail to reject the null by a wide margin (F27,3417 = 0.69; p-value = 0.88). Three terms have nk estimates statistically significant at the 5% level: the terms interacting MINORITY with CREDIT_SCORE_LO, USE_TRADE_CR, and LN_AGE. Minority owners with credit ratings in the significant- or high-risk categories (i.e., CREDIT_SCORE_LO = 1) have higher estimated denial rates than comparable White-owned firms. Use of trade credit (i.e., USE_TRADE_CR = 1) lowers estimated denial rates for minority owners but not for White owners, suggesting a certification effect from using trade credit. Greater firm age increases denial rates for minority owners relative to White owners, possibly because minority-owned firms have shorter average lifespans or greater management succession problems. Three other nk estimates narrowly achieve statistical significance at the 10% level: the terms interacting MINORITY with LN_ASSETS, LN_SALES, and PRIMARY. All three nk estimates are negatively signed, indicating that minority owners lower their denial rates relative to White owners by increasing assets, increasing sales or submitting loan applications to their primary

financial services providers instead of other lenders, actions which should tend to increase informational transparency. Despite including in the model the Xki 9 MINORITYi terms, estimates of the cj, kj, gj, and hj parameters in (7.3) show the same basic pattern as in the previous models. Consequently when we perform F-tests of the NB_NLOC, B_NLOC, and B_LOC hypotheses, the results parallel those reported for the previous models. 4.4 Economic significance of apparent differences in denial rates The estimated loan application denial models reported in Tables 6 and 7 indicate some statistical differences in denial rates for applicants in different demographic groups but not whether these differences are economically significant. To assess economic significance we use our estimated models to predict denial probabilities for applicants who differ only by demographic group. Specifically, we predict denial probabilities for White (W) and minority (M) applicants having attributes (Xki) equal to the sample medians for White applicants (White medians) and compare the predicted probabilities. We repeat the exercise for applicants having attributes equal to the sample medians for minority applicants (Minority medians). We report predicted denial probabilities based on (6.4), (6.8), and (7.3) in Table 8. Asterisks denote predictions for minority applicants statistically different from predictions for White applicants based on the F-statistics reported in Panel B of Tables 6 and 7. The leftmost pair of columns report predicted denial probabilities based on Eq. 6.4 for White and minority applicants both having White-median attributes.30 Denial predictions for White owners (W column) differ sharply between line-of-credit and non-line-ofcredit loans: 13–14% on a non-line-of-credit loan regardless of lender applied to or market competitiveness, versus 35–37% on a bank credit line. These denial predictions often appear economically different 30

29

We do not include interaction terms between MINORITY and BANKRUPT, JUDGMENT, OWNR_PAY_LATE, and DENIED_TRADE_CR due to lack of variation in the data, or between MINORITY and the indicators for application year, region or industry.

123

To predict denial probabilities for White (minority) owners we set MINORITY equal to 0 (1); we report the predictions in the W (M) column. To predict probabilities for line-of-credit (non-line-of-credit) loans at banks (nonbanks) we set LOC and BANK equal to 1 (0); we set HHI_HI equal to 0 or 1 to predict denial probabilities in low- and high-concentration markets, respectively.


297

Table 8 Predicted probability of loan application denial by demographic group Equation:

(6.4)

(6.8)

White Medians

Minority Medians

Minority Medians but

(7.3)

White Medians

Minority Medians but

White Medians

(PRIMARY = 1) (PRIMARY = 0) PRIMARY = 1 (PRIMARY = 1) PRIMARY = 1 (PRIMARY = 1) Demographic groupa:

W

M

W

M

W

M

W

AA

H

W

AA

H

W

M

Loan type and market concentrationb Nonbank non-line-of-credit loan Low 13 concentration

23

30

43

15

24

21

56*

28

23

58*

30

7

3

High 14 concentration

47*

31

69*

16

49*

22

83*

37

24

84*

39

8

18*

Bank non-line-of-credit loan Low 13 concentration

52*

30

74*

15

54*

21

54*

67*

23

57*

67*

6

26*


76*

31

90*

16

78*

21

82*

75*

23

83*

76*

7

64*

Low 35 concentration

45

58

68

37

48

45

58

36

47

60

38

25

20


71*

59

87*

39

73*

46

84*

45

48

85*

47

26

56*

Bank line-of-credit loan

This table shows predicted denial probabilities, as percentages, for White (W), minority (M), African-American (AA), and Hispanic (H) loan applicants. The predictions were produced by combining the parameter estimates from Eqs. 6.4, 6.8, and 7.3 with values of the explanatory variables equal to the sample medians for White-owner applicants (White medians) and minority-owner applicants (Minority medians). Predicted probabilities are also shown based on minority medians except that the loan application is made to the applicant’s primary financial institution (Minority medians but PRIMARY = 1). Asterisks denote minority-owner percentages statistically different from White-owner percentages based on F-statistics reported in Tables 6 and 7 a Predictions for White-owned firms (W) were produced by setting MINORITY = 0 in (6.4) and (7.3) and by setting AFROAM = 0 and HISPANIC = 0 in (6.8). Predictions for minority-owned firms (M) were produced by setting MINORITY = 1 in (6.4) and (7.3). Predictions for African-American- and Hispanic-owned firm (AA and H) were produced by setting AFROAM = 1 and HISPANIC = 1 in (7.3) b

Predictions for different loan types and market types were produced by setting the indicator variables BANK, LOC, and HHI_HI to 0 or 1 as follows: Nonbank, non-line-of-credit loan: BANK = 0, LOC = 0; low market concentration: HHI_HI = 0

Bank, non-line-of-credit loan: BANK = 1, LOC = 0; high market concentration: HHI_HI = 1 Bank line-of-credit loan: BANK = 1, LOC = 1

to those for minority owners having the same attributes (M column). The differences are greatest for bank nonline-of-credit loans: the denial predictions for a minority applicant are 52 and 76% in low- and highconcentration markets, respectively, 39 and 62 percentage points more than for a White applicant. In lowconcentration markets White and minority denial predictions differ for nonbank non-line-of-credit loans and bank credit lines by economically smaller (and statistically insignificant) amounts, about 10 percentage points; but in high-concentration markets the

differences are a statistically significant 33 or 34 percentage points, amounts that seem economically significant. The second pair of columns report predicted denial probabilities based on Eq. 6.4 for White and minority applicants, both having minority-median attributes. These applicants are presumably more opaque to lenders for two reasons: smaller firm size and weaker ties with lenders, as the median minority applicant applied to a lender other than the applicant’s primary financial institution (i.e., PRIMARY = 0), unlike the

123

298

median White applicant (PRIMARY = 1). This opacity raises the predicted denial probabilities in the second pair of columns by 14–23 percentage points compared with the first pair. However even when White and minority applicants are similarly opaque, the statistically different predicted denial probabilities remain seemingly economically different, with differences in denial probabilities ranging from 26 to 59 percentage points. The third pair of columns also uses minoritymedian attributes except that we let PRIMARY = 1, i.e., we assume that all applicants apply to their primary financial institutions. The predicted denial probabilities for White and minority applicants fall by 12–20 percentage points to nearly the same levels as in the first pair of columns, which use Whitemedian attributes. These declines in predicted denial probabilities suggest an important role for relationships in small-business lending. Statistical differences in minority and White denial probabilities appear to be economically significant based on Eq. 6.8, as shown in the next six columns. We report predicted denial probabilities for White-, African-American-, and Hispanic-owned firms having White-median attributes (the first three columns) and having minority-median attributes except for applying to the applicant’s primary financial institution (the second three columns).31 The two sets of predictions are similar, so we focus on the former. For an African-American owner in a low-concentration market the predicted denial probability on a nonline-of-credit loan is 33–35 percentage points greater than for a White owner; in a high-concentration market the predicted denial probability is about 61 percentage points greater. Both differences seem economically significant. An African-American owner’s predicted denial probability on a bank credit line in a high-concentration market is about 38 percentage points more than for a White owner, also seemingly economically significant. Denial probabilities are statistically greater for Hispanics than Whites on bank non-line-of-credit loans, with differences of 46 and 54 percentage points predicted in low- and

31

To predict denial probabilities for African-American (Hispanic) owners we set AFROAM = 1 (HISPANIC = 1); we report the predictions in the AA (H) columns. To predict denial probabilities for White owners we set AFROAM and HISPANIC equal to 0; we report the results in the W column.

123


high-concentration markets, respectively, differences which seem economically significant. The last pair of columns show that statistical differences in predicted denial probabilities appear to be economically significant based on the model that permits lenders to weight differently the observable attributes of White and minority applicants, Eq. 7.3. The columns report predicted denial probabilities for White and minority applicants having White-median attributes. For minority applicants the predicted denial probabilities on a bank non-line-of-credit loan are 26 and 64% in low- and high-concentration markets, respectively, 20 and 57 percentage points more than predicted for comparable White applicants. In high-concentration markets the denial probabilities for minority applicants exceed those for White applicants by 10 percentage points for nonbank non-line-of-credit loans and by 30 percentage points for bank credit lines, amounts that seem economically significant. 4.5 Discussion Our evidence suggests that, relative to comparable White small-business owners, minority owners in low-competition markets have less access to credit— consistent with Prediction 3—and in high-competition markets have less access to non-line-of-credit loans from banks but equal access to bank credit lines and non-line-of-credit loans from nonbanks—consistent with Predictions 1 and 2.32 Unequal access in low-competition markets might occur for the reasons originally offered by Becker (1957): market forces are too weak to prevent lenders from exercising tastebased or statistical discrimination. However, this explanation does not account for our other finding. Minority firm owners appear to have better access to new line-of-credit loans because the technology that banks use to grant them avert statistical discrimination by producing a reliable, low-cost signal of

32

As an anonymous referee notes, an interesting question is whether minority borrowers learn which loan types are more likely to be approved and subsequently apply for those loans types more often. Current regulations do not allow lenders to collect information about race and gender on loan applications, so the availability of this information is limited; however there is evidence consistent with learning, as we note in the final paragraph of the paper.


loan-applicant quality. When contracting is costly, borrowing with a loan that matures before information about firm quality becomes symmetric is optimal only if the firm owner knows firm quality to be high (Myers 1977; Flannery 1986). Thus a minority firm owner’s loan-type choice produces a positive signal about firm quality that is potentially less noisy than the signal in the owner’s minority status. The information content in credit-line applications may extend beyond the signal from the short loan maturity: the suitability of credit lines for financing recurring needs (i.e., working capital) implies that credit-line applications are more likely to lead to a series of loans, which signals an intention to build reputation (Diamond 1991). Minority access to non-line-of-credit loans appears to be uneven. The longer term of most non-line-ofcredit loans prevents a loan application from sending a low-cost signal of firm quality. Lenders can protect themselves against observably riskier loan applicants by requiring owners to pledge more collateral in the form of business assets or owners’ personal assets. Assessing collateral quality is not costless, however, and the cost of liquidating collateral in the event of loan default may be substantial. Nonbank lenders appear better able to bear these costs than banks because they are better capitalized and possibly better able to dispose of collateral in the event of loan default (Carey et al. 1998). While our finding of unequal minority access to bank non-line-of-credit loans is consistent with statistical discrimination, we cannot rule out alternative explanations. One explanation involves demographic differences in the quality of loan collateral. We have no variable to control for collateral quality. In our sample, minority applicants for non-line-of-credit loans own firms with substantially less assets than White applicants (US $362,000 versus US $810,000), and minority owners themselves have substantially less wealth than their White counterparts (US $327,000 versus US $1,003,000). Although we control for both asset size and owner wealth in our empirical estimations, we have no way of knowing whether the quality of minority firms’ collateral is lower or less readily allocated to other uses than White owners’ collateral. Collateral quality may be a smaller issue for non-line-of-credit loans from nonbanks if these lenders are better able to redeploy collateral in the event of loan default. Collateral quality may also

299

be a smaller issue for credit lines if the collateral is generic (e.g., accounts receivable).33

5 Summary and conclusion The Federal Financial Institutions Examination Council observes that the heterogeneous nature of small-business loans complicates the Council’s job of enforcing the Equal Credit Opportunity Act of 1974 (FFIEC, pp. 24–25); yet prior studies of minority access to credit generally treat small-business loans as a single, homogenous category. This paper draws on the small-business lending literature to develop empirical loan denial models that recognize two types of loans (credit lines and non-line-of-credit loans) made by two types of lenders (commercial banks and nonbank lenders). We estimate our models on data from the 1998 Survey of Small Business Finances (SSBF). We find some evidence consistent with unequal minority access: access appears to be unequal to bank non-line-of-credit loans in highly competitive markets and to all loans in less competitive markets. However, we also find some evidence consistent with equal access: access appears to be equal to bank credit lines and to nonbank non-line-ofcredit loans in highly competitive loan markets. These conclusions continue to hold after we perform several robustness checks. We also find that, where market access differs statistically, these differences are economically significant. We attribute disparities in market access to differences in lending technologies and specialization, which cause lenders to differ in their capacities and incentives to develop and use soft information about a new loan applicant in place of the noisy signal that comes through the applicant’s minority status. Equal minority access to short-term bank credit lines is consistent with theoretical arguments and empirical evidence about informationally opaque firms’ use of short-term monitored bank loans to signal superior project quality (Myers 1977; Flannery 1986; Ortiz-Molina and Penas 2008). Equal minority

33

Shleifer and Vishny (1992) discuss the problem of asset specificity and its implications for capital structure decisions. Ghosal (2007) presents empirical evidence on how asset specificity (sunk costs) affects the entry and exit decisions of small manufacturing firms.

123

300


access to nonbank non-line-of-credit loans is consistent with theory and evidence about lenders’ incentives to specialize their lending (Carey et al. 1998; Remolona and Wulfekuhler 1992; Daniels and Ramirez 2008). Less equal minority access to bank non-line-of-credit loans in competitive loan markets is consistent with the weak credit-quality signal in this choice of loan type when lenders are asymmetrically informed and with lenders’ incentives to minimize information-gathering costs (Petersen and Rajan 1995). Less equal minority access to credit in less competitive loans markets is consistent with reasoning presented by Becker (1957). Our findings have implications for both policymakers and minority small-business owners. Our findings suggest that policymakers encouraging fair access may gain the greatest benefit per taxpayer dollar spent by focusing on those credit-market

segments where minority access appears to be less equal: less competitive loan markets and markets for bank non-line-of-credit loans. Our findings also suggest a strategy for minority small-business owners: cultivate relationships with loan sources and use bank credit lines when possible, followed by nonbank non-line-of-credit loans. Data from the 1998 SSBF suggest that African-American owners appear to be pursuing this strategy: among owners surveyed, a significantly higher proportion of African-American owners than White owners applied to commercial banks for credit lines.

Appendix See Table 9.

Table 9 Complete estimated loan application and loan application denial models Equation:

(6.4)

Dependent variable: Independent variables

Denied

Applied

Denied

Applied

0.037

0.016*

0.016

0.150*

(0.190)

(0.083)

(0.182)

(0.083)

HHI_HI LOC BANK

(6.8)

0.727***

0.674***

(0.202)

(0.215)

-0.001

-0.007

(0.197) FEMALE FEMALE 9 HHI_HI FEMALE 9 LOC FEMALE 9 BANK MINORITY MINORITY 9 HHI_HI

(0.185)

-0.563

0.004

-0.620

0.000

(0.422)

(0.092)

(0.426)

(0.091)

0.114

-0.113

0.147

-0.103

(0.291)

(0.146)

(0.281)

(0.146)

0.246 (0.297)

0.204 (0.288)

0.595

0.662*

(0.364)

(0.367)

0.352

0.077

(0.282)

(0.089)

0.642**

-0.082

(0.304)

(0.152)

MINORITY 9 LOC

-0.890***

MINORITY 9 BANK

0.802**

(0.280) (0.352) AFROAM

123

0.948*

0.106

(0.515)

(0.133)


301

Table 9 continued Equation:

(6.4)


Denied

(6.8) Applied

AFROAM 9 HHI_HI

Denied

Applied

0.773*

-0.172

(0.462)

(0.227)

AFROAM 9 LOC

-0.576

AFROAM 9 BANK

-0.035

(0.411) (0.584) ASIAN

0.217

-0.090

(0.566)

(0.159)

ASIAN 9 HHI_HI

0.230

-0.016 (0.279)

ASIAN 9 LOC

(0.440) 0.171

ASIAN 9 BANK

0.009

(0.490) (0.489) HISPANIC HISPANIC 9 HHI_HI HISPANIC 9 LOC

0.217

0.124

(0.343)

(0.129)

0.214

0.064

(0.397)

(0.230)

-1.454*** (0.446)

HISPANIC 9 BANK

1.013** (0.457)

POST_HS

0.003

-0.177**

-0.034

-0.174**

(0.182)

(0.085)

(0.177)

(0.085)

COLLEGE

0.035

-0.224***

0.026

-0.217***

LN_EXPER

(0.185) 0.107

(0.081) -0.050

(0.180) 0.156

(0.081) -0.044

(0.129)

(0.055)

(0.127)

(0.055)

OWNR_%

-0.005

0.001

-0.005

0.001

(0.004)

(0.002)

(0.004)

(0.002)

MANAGER

-0.521**

0.074

-0.494**

0.079

(0.228)

(0.112)

(0.241)

(0.112)

-0.088**

-0.005

-0.084**

-0.005

(0.036)

(0.013)

(0.037)

(0.013)

0.594**

-0.262**

0.660**

-0.267**

(0.241)

(0.114)

(0.261)

(0.114)

-0.120*

0.093***

-0.132**

0.093***

(0.064)

(0.035)

(0.064)

(0.035)

-0.010

0.025

-0.010

0.027

(0.039)

(0.021)

(0.038)

(0.021)

-0.055*** (0.020)

-0.004 (0.007)

-0.051** (0.021)

-0.004 (0.007)

LN_NET_WRTH FAMILY LN_ASSETS LN_SALES ROA

123

302



(6.4)


Denied

Applied

Denied

Applied

LN_EQUITY

(6.8)

0.031

-0.077***

0.042

-0.075**

(0.061)

(0.030)

(0.058)

(0.030)

0.189

-0.641**

0.260

-0.618*

(0.638)

(0.321)

(0.609)

(0.324)

2.332***

-0.105

2.151***

-0.110

(0.489)

(0.197)

(0.528)

(0.198)

0.551*

0.271*

0.488

0.265

(0.304)

(0.164)

(0.315)

(0.165)

OWNR_PAY_LATE

0.957***

0.262**

0.875**

0.267**

BUS_PAY_LATE

(0.321) 0.526**

(0.119) 0.118

(0.358) 0.484*

(0.120) 0.113

(0.226)

(0.073)

(0.255)

(0.073)

-0.057

-0.007

-0.048

-0.009

(0.142)

(0.067)

(0.137)

(0.067)

0.335

0.124

0.295

0.120

(0.214)

(0.135)

(0.216)

(0.135)

-0.196***

0.271***

-0.227***

0.275***

(0.058)

(0.025)

(0.054)

(0.028)

-0.116

0.022

-0.078

0.030

(0.170)

(0.075)

(0.166)

(0.076)

0.044

0.078

0.051

0.076

(0.138)

(0.064)

(0.132)

(0.064)

-0.356**

0.145**

-0.339**

0.147**

(0.140)

(0.066)

(0.137)

(0.067)

LN_AGE

-0.108

-0.135***

-0.124

-0.115***

LN_EMPLOYEES

(0.110) 0.076

(0.043) 0.014

(0.120) 0.074

(0.045) 0.011

(0.085)

(0.038)

(0.080)

(0.038)

0.153

-0.008

0.165

-0.005

(0.165)

(0.081)

(0.160)

(0.081)

-0.075

-0.136

-0.057

-0.135

(0.203)

(0.095)

(0.194)

(0.095)

-0.015

-0.006

0.003

-0.001

(0.175)

(0.087)

(0.169)

(0.087)

0.345*

-0.025

0.318*

-0.027

(0.177)

(0.090)

(0.174)

(0.091)

0.280

-0.144*

0.272

-0.152*

(0.174)

(0.082)

(0.167)

(0.082)

NEG_EQUITY BANKRUPT JUDGMENT

CREDIT_SCORE_LO DENIED_TRADE_CR #_OF_RELATIONS USE_TRADE_CR USE_OWNR_CCARD USE_BUS_CCARD

LN_LOCATIONS C_CORP S_CORP NATIONAL MSA LN_LENGTH

0.044 (0.036)

(0.034)

PRIMARY

-0.523***

-0.497** (0.206)

(0.196)

123

0.035


303


(6.4)


Denied

APPLY_97

-0.197

(6.8) Applied

Denied

Applied

-0.252

(0.312)

(0.301)

APPLY_98

0.053

0.011

(0.277)

(0.265)

APPLY_99

0.108

0.082

(0.274)

(0.264)

APPLY_00

0.364

0.303

REGION_1

-0.095

0.016

-0.083

0.008

REGION_2

(0.348) 0.198

(0.150) -0.061

(0.331) 0.170

(0.150) -0.069

(0.213)

(0.112)

(0.207)

(0.112)

-0.193

-0.172

-0.156

-0.180

(0.319)

(0.110)

(0.319)

(0.110)

-0.676**

0.161

-0.641**

0.159

(0.302)

(0.132)

(0.301)

(0.133)

-0.224

0.017

-0.230

0.000

(0.204)

(0.101)

(0.204)

(0.102)

-1.098**

-0.049

-1.068**

-0.053

(0.488)

(0.137)

(0.526)

(0.138)

REGION_7

-0.192

0.256**

-0.208

0.235**

(0.220)

(0.106)

(0.211)

(0.106)

REGION_8

-0.181

0.027

-0.127

0.004

(0.264)

(0.129)

(0.249)

(0.130)

INDUSTRY_1

0.167

0.116

0.105

0.120

INDUSTRY_2

(0.259) 0.284

(0.123) -0.010

(0.249) 0.250

(0.123) -0.018

(0.321)

(0.171)

(0.307)

(0.171)

-0.289

0.135

-0.368

0.141

(0.297)

(0.162)

(0.298)

(0.163)

-0.421

0.120

-0.426

0.127

(0.362)

(0.158)

(0.352)

(0.158)

-0.252

-0.005

-0.248

-0.004

(0.335)

(0.144)

(0.314)

(0.144)

0.094

-0.027

0.071

-0.021

(0.225)

(0.102)

(0.211)

(0.102)

0.159

0.137

0.114

0.134

(0.313)

(0.142)

(0.289)

(0.141)

0.109

0.016

0.062

0.016

(0.212)

(0.099)

(0.200)

(0.099)

2.203** (0.929)

-1.314*** (0.360)

2.376*** (0.860)

-1.223*** (0.364)

(0.368)

REGION_3 REGION_4 REGION_5 REGION_6

INDUSTRY_3 INDUSTRY_4 INDUSTRY_5 INDUSTRY_6 INDUSTRY_7 INDUSTRY_8 CONSTANT

(0.354)

123

304



(6.4)

(6.8)


Denied

Applied

Denied

Applied

# of observations

863

3,444

863

3,444

This table reports complete coefficient estimates and standard errors of the loan application denial models reported as Eqs. 6.4 and 6.8 in Table 6, and the loan application models estimated jointly with them. In the loan application denial models the dependent variable is 1 if a loan application is denied and 0 if approved. In the loan application models the dependent variable is 1 if a firm applied for a loan and 0 if it did not ***, **, * Statistically different from zero at the 1, 5, and 10% levels, respectively, for a two-tailed test

References Becker, G. S. (1957). The economics of discrimination. Chicago: University of Chicago Press. Berger, A. N., & Black, L. K. (2007). Bank size and small business finance: Test of the current paradigm. Working paper. Berger, A. N., Frame, W. S., & Miller, N. H. (2005). Credit scoring and the availability, price, and risk of small business credit. Journal of Money, Credit and Banking, 37, 191–222. Berger, A. N., & Udell, G. F. (2006). A more complete conceptual framework for SME finance. Journal of Banking & Finance, 30, 2945–2966. Blanchard, L., Zhao, B., & Yinger, J. (2008). Do lenders discriminate against minority and woman entrepreneurs? Journal of Urban Economics, 63, 467–497. Blanchflower, D. G., Levine, P. B., & Zimmerman, D. J. (2003). Discrimination in the small business credit market. The Review of Economics and Statistics, 84, 930–943. Carey, M., Post, M., & Sharpe, S. A. (1998). Does corporate lending by banks and finance companies differ? Evidence on specialization in private debt contracting. The Journal of Finance, 53, 845–878. Cavalluzzo, K. S., & Cavalluzzo, L. C. (1998). Market structure and discrimination: The case of small businesses. Journal of Money, Credit and Banking, 30, 771–792. Cavalluzzo, K., Cavalluzzo, L., & Wolken, J. D. (2002). Competition, small business financing, and discrimination: Evidence from a new survey. Journal of Business, 75, 641–679. Cavalluzzo, K., & Wolken, J. (2005). Small business loan turndowns, personal wealth and discrimination. Journal of Business, 78, 2153–2177. Chakraborty, A., & Hu, C. X. (2006). Lending relationships in line-of-credit and nonline-of-credit loans: Evidence from collateral use in small business. Journal of Financial Intermediation, 15, 86–107. Daniels, K., & Ramirez, G. G. (2008). Information, credit risk, lender specialization and loan pricing: Evidence from the DIP financing market. Journal of Financial Services Research, 34, 35–59. Diamond, D. W. (1991). Monitoring and reputation: The choice between bank loans and directly placed debt. Journal of Political Economy, 99, 689–721.

123

Federal Financial Institutions Examination Council. (undated). Interagency fair lending examination procedures. Accessed February 21, 2009, from http://www.ffiec.gov/PDF/fair lend.pdf. Flannery, M. J. (1986). Asymmetric information and risky debt maturity choice. The Journal of Finance, 41, 19–37. Ghosal, V. (2007). Small is beautiful but size matters: The asymmetric impact of uncertainty and sunk costs on small and large businesses. Working paper. Government Accountability Office. (2008). Fair lending: Race and gender data are limited for nonmortgage lending. GAO-08-698. Haggerty, C. C., Grigorian, K. H., Harter, R., & Stewart, A. K. (2001). The 1998 survey of small business finances, methodology report. The Board of Governors of the Federal Reserve System. Myers, S. C. (1977). Determinants of corporate borrowing. Journal of Financial Economics, 5, 147–175. Ortiz-Molina, H., & Penas, M. F. (2008). Lending to small businesses: The role of loan maturity in addressing information problems. Small Business Economics, 30, 361–383. Petersen, M. A., & Rajan, R. G. (1995). The effect of credit market competition on lending relationship. The Quarterly Journal of Economics, 110, 407–443. Remolona, E. M., & Wulfekuhler, K. C. (1992). Finance companies, bank competition, and niche markets. Federal Reserve Bank of New York Quarterly Review, 17, 25–38. Shleifer, A., & Vishny, R. (1992). Liquidation values and debt capacity: A market equilibrium approach. The Journal of Finance, 47, 1343–1366. Stiglitz, J. E., & Weiss, A. (1981). Credit rationing in markets with imperfect information. The American Economic Review, 71, 393–410. Van de Ven, W., & Van Pragg, B. (1981). The demand for deductibles in private health insurance: A probit model with sample selection. Journal of Econometrics, 17, 229– 252. Wolken, J., Bitler, M., & Robb, A. (2001). Codebook for 1998 survey of small business finances. Board of Governors of the Federal Reserve. Accessed February 21, 2009, from http://www.federalreserve.gov/pubs/oss/oss3/ssbf98/ssbf98 home.html#ssbf98docs.