Rejoinder to Berkovec, Canner, Gabriel, and Hannan

This rejoinder addresses the response to articles by Berkovec, Canner, Gabriel, and Hannan. While our earlier comments emphasized the biases in the default approach, we focus here on the basic disagreement: whether, given those biases, the default approach provides a viable test for mortgage lending discrimination. We argue that the default approach does not provide such a test. The Berkovec, Canner, Gabriel, and Hannan research (BCGH, 1994 and in this issue) provides a careful empirical look at racial differences in loan default rates using data on Federal Housing Administration (FHA) mortgages. In this rejoinder, however, we focus on the key issues that separate us. First, we express our deep concerns about the basic conclusion of the BCGH article; then, with those concerns in mind, we suggest a change in future default specifications. In addition, we discuss the problems inherent in their suggestion that default rates may be useful for partitioning the observed incidence of racial discrimination into two different types: statistical and prejudiced based. Our fundamental disagreement with BCGH centers around the following statement, which is found in their article (in other words) and appears in their response to our critiques: “… FHA loan performance data do not support a finding of widespread systematic discrimination in mortgage lending due to lender prejudice.” Although they acknowledge that their approach may suffer from a variety of biases and cannot prove that discrimination does not exist, they argue that proof is too high a standard for any empirical test. In response, we propose an alternative standard: Performance analyses should not be used to identify discrimination in mortgage lending unless it is demonstrated that these analyses are at least as effective as relevant alternative analyses, such as traditional analyses based on loan approval data. We conclude that the default approach fails to meet this standard under virtually all circumstances. Consider the two key assumptions referred to by BCGH as no omitted variable bias and no statistical discrimination. The first assumption implies that all unobserved variables known to the lender are uncorrelated with race. If that is not so, the BCGH results are biased away from finding discrimination. In this case, BCGH point out—and we agree— loan application studies are also biased, and the bias is in favor of finding discrimination.