| Popis: |
This article seeks to evaluate discrimination in home mortgage originations by examining the performance of mortgage loan portfolios. This approach follows from the theoretical foundations of the economics of discrimination (Becker, 1971). The basic premise is that biased lenders will require higher expected profits for loans to minority borrowers and hold minority applicants to underwriting standards in excess of those required for other applicants. Thus discrimination results in lower expected default costs for loans originated for marginally qualified nonminority borrowers. This study employs a rich FHA data set, comprising a large number of individual loan records, to evaluate the performance of mortgage borrowers. Results of the analysis fail to find evidence of better performance on loans granted to minority borrowers. Indeed, black borrowers are found, all else being equal, to exhibit a higher likelihood of mortgage default than other borrowers. These findings argue against allegations of substantial levels of bias in mortgage lending. Many recent studies of mortgage lending activity have documented large and persistent racial disparities, including the provision of information to prospective home loan applicants, mortgage loan instrument selection, and the loan application decision process. (See, for example, Board of Governors of the Federal Reserve System, 1991; Canner, Gabriel, and Woolley, 1991; and Munnell, Browne, McEneaney, and Tootell, 1992, hereafter to be referred to as MBMT.) For the most part, findings of those analyses indicate significant race effects that are not well explained by objective factors. Hence the findings have led to allegations of widespread racial discrimination in mortgage lending. This article seeks to evaluate discrimination in home mortgage originations by examining the performance of mortgage loan portfolios. This approach follows from the theoretical foundations of the economics of discrimination (Becker, 1971), which are based on the Cityscape: A Journal of Policy Development and Research • Volume 2, Number 1 • February 1996 U.S. Department of Housing and Urban Development • Office of Policy Development and Research Berkovec, Canner, Gabriel, and Hannan 10 Cityscape premise that biased lenders will require higher expected profits from loans to minority applicants. As applied to lending by Richard Peterson (1981), this premise implies that biased lenders may hold minority applicants to more stringent underwriting standards than those required for other applicants. Thus discrimination results in lower expected default costs and higher expected profits for loans originated for marginally qualified minority mortgage borrowers in comparison with those observed for marginally qualified nonminority borrowers. It is important to note that this theory assumes that discrimination against minorities occurs at the margin, affecting those who are near the borderline for creditworthiness, and excludes the possibility that the discrimination is unrelated to credit risk. The theory predicts that this discrimination changes loan performance at the margin. Thus inferences about discrimination that are made from loan performance data must distinguish between average and marginal loan performance. As noted by Peterson (1981) and by Ferguson and Peters (1995), simple comparisons of average loan performance between two groups of borrowers can be misleading if the groups do not exhibit similar distributions of expected returns in the absence of discrimination. If, for example, the proportion of highly qualified nonminority borrowers is substantially higher than that of highly qualified minority borrowers, default rates of nonminority borrowers—observed without controlling for other determinants of credit quality—would be lower than those associated with minority borrowers. This finding, however, would simply reflect the differences in average creditworthiness for the two groups of borrowers and would not necessarily indicate differential underwriting standards. Our study employs a rich Federal Housing Administration (FHA) data set to evaluate the determinants of loan performance as measured by both the likelihood of default and the losses that occur in the event of default. The data consist of a large number of individual loan records recently made available by the U.S. Department of Housing and Urban Development (HUD). That information is augmented with 1980 and 1990 census tract characteristics to account for neighborhood location attributes associated with default risk. These data are particularly well suited to the investigation, given the vast array of detail concerning characteristics of loans, borrowers, and neighborhoods in which the homes are located. The following section of this article presents the theoretical foundations for the tests of discrimination in mortgage loan performance as they apply to the likelihood of default. The section entitled “Discrimination and Loan Performance” provides a description of the data used in the analysis and empirical specifications of the models. The section entitled “Data and Model Specification” presents the results of model estimations, and the final section provides a summary of the findings. Discrimination and Loan Performance The starting point for our analysis is a simple rationing model of loan origination. One must assume that lenders observe a creditworthiness index (C) for each loan applicant. For our purposes we assume that there is a direct relationship between the level of C for an applicant and the default risk of that applicant. The applicant’s default risk is represented by an expected default probability, D(C), where 0 A + B, Then: CONVENTIONAL LOAN If: A + B > C > F + B', Then: FHA LOAN If: C < F + B', Then: REJECTED APPLICATION where A represents the minimum level of creditworthiness required for approval of a conventional loan, and F is the minimum level necessary for an FHA-insured loan. The values of B and B', assumed to be positive, indicate the degree of discrimination faced by the applicant. Discrimination can occur at either one or both of the two margins. Greater values of B represent increased discrimination in the conventional loan market, while higher levels of B' indicate increased bias in the underwriting of FHA loans. If C were observed directly, discrimination could easily be detected by comparing the minimum levels of C for accepted loans for the borrower groups within each loan type. One could also compare maximum levels of C for conventional loan rejections to identify B, or FHA loans to identify B'. However, outsiders cannot observe the creditworthiness index directly. Our assumption is that outside analysts observe instead a set of characteristics of the loan and applicant that are related to C. Formally, this is expressed as: C = Xβ + e, where X is a vector of observed characteristics, β is a vector of known constants, and e is an error term observable only to the lender. In this framework borrowers with the same observed characteristics X will have different default risks, and get different receptions from lenders, because of differences in the unobservable e. As lenders observe e, the highest default risk applicants at every level of X will be rejected. In the presence of discrimination, the rejection probability of an applicant with characteristics X is given by: d(X) = ∫ f(e) de e < F – Xβ + B' and the probability of approval for an FHA-insured loan is given by |
