Evaluating Statistical Models of Mortgage Lending Discrimination: A Bank-Specific Analysis (WP 95-3)

Abstract

The Boston Fed's study on mortgage lending discrimination has prompted a considerable rise in the use of statistical methods to evaluate banks' lending behavior. However, several issues concerning the validity, accuracy, and reliability of statistical models remain unresolved. In this paper, we attempt to address several of those issues within the context of a bank-specific analysis.

Using data gathered from three nationally chartered banks, we extend the "standard" model outlined in the Boston Fed study. We conclude that a model specification that incorporates the specific underwriting guidelines of the individual bank is more appropriate than a broadly-defined, generic specification applied across all banks. We also show that the process of incorporating the bank-specific guidelines is itself difficult and can be complicated further by the lack of accurate data upon which to build a valid representative sample.

We discuss several additional methodological issues concerning model validity, specification, sample design, and data accuracy that must be addressed if a statistical approach is to become a useful and accurate supervisory tool.

Authors

Mitchell Stengel and Dennis Glennon