Academic journal article Real Estate Economics

Evaluating Statistical Models of Mortgage Lending Discrimination: A Bank-Specific Analysis

Academic journal article Real Estate Economics

Evaluating Statistical Models of Mortgage Lending Discrimination: A Bank-Specific Analysis

Article excerpt

The Boston Fed's study on mortgage lending discrimination at the market or MSA level (Munnell et al. 1996, 1992) has prompted a considerable rise in the use of statistical methods to evaluate specific banks' lending behavior. The federal banking agencies and a growing number of banks are using, developing or at least considering such models to test for unfair lending practices by their mortgage underwriting departments.

In this study, we outline our efforts to develop bank-specific models for three nationally chartered banks. Our results demonstrate that, by incorporating the specific underwriting guidelines of each individual bank, the bank-specific regression models uniformly outperform a more broadly defined, generic specification typified by the Boston Fed model. The bank-specific approach significantly improves the ability of the model to explain the outcomes of the mortgage lending decision-making process in terms of the significance of most individual variables, the overall goodness of fit of the model, and tests for the accuracy of the model specification.

In the context of developing the alternative bank-specific models, we also discuss methodological issues of model validity, specification and accuracy that have been raised in the literature that has appeared since the Boston Fed study (LaCour-Little 1996; Phillips and Yezer 1996; Glennon and Stengel 1994; Horne 1994; Bauer and Cromwell 1994; Yezer, Phillips and Trost 1994; Carr and Megbolugbe 1993; Day and Liebowitz 1993). In particular, we address the issues of selection and simultaneous-equation bias. We acknowledge the potential for problems along these lines, but nonetheless argue that a single-equation rejection model is appropriate for our specifically delineated application.

We start with a specification that closely resembles that developed by the Boston Fed (Munnell et al. 1996, 1992) in their evaluation of applicants from the Boston MSA. This specification reflects the aggregation of underwriting guidelines inherent in the application of a single specification to the combined sample drawn from over 130 lending institutions. Because individual banks' underwriting standards deviate from the secondary market guidelines, a Boston Fed-type specification may well be inconsistent with the process that generates the data - a result that will surely lead to specification error if applied to an individual bank's lending practices.

Although the results of our comparison of model specifications using data from the three separate banks are mixed, at this stage of our research we conclude that a statistical model can be useful as a screening device to identify institutions that may use race as a decision variable. There remain, however, significant data issues - issues that evolve out of anomalies in the sample data developed from the Home Mortgage Disclosure Act (HMDA) reports - that make it difficult to rely solely on the results of a statistical model; therefore, the results of the statistical model should be verified by a judgmental review of selected loan files. Although such a review is imperfect and may suffer from other methodological deficiencies, comparing the results of the two approaches will allow us to develop better statistical techniques for fair-lending enforcement purposes.

The remainder of the paper is structured as follows: The next section discusses briefly the circumstances under which a broad-based (or market-level) modeling methodology may be appropriate. We then describe the sampling methodology and data collection procedures used in the present study. Model specification and regression results are presented next, followed by a discussion of the data and modeling issues encountered in applying this model to the three national banks. In the concluding section, we examine questions for further research and issues about the usefulness of statistical models as tools for bank regulators in the fair-lending area. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed


An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.