The Equal Credit Opportunity Act (ECOA) forbids discrimination by a creditor on the basis of race, color, religion, national origin, sex, marital status or age. As the Consumer Financial Protection Bureau (CFPB) has acknowledged, particularly in the non-mortgage context, lenders are generally not permitted to collect demographic information about borrowers. But, regulators, including the CFPB, continue to look for potential discrimination—and, more precisely, potential disparate impact liability—in a variety of non-mortgage credit contexts, including indirect auto lending. They have also made clear that entities are expected to proactively evaluate their lending patterns using proxy data as well. Regulators and regulated entities alike, then, need to find some way to know, or at least estimate, the race, ethnicity, and gender of borrowers. The CFPB has described the lack of demographic information about borrowers as a “gap” in data, and has noted that it uses a proxy for the race and ethnicity of borrowers.
Regulated entities have, naturally, therefore been asking for a while how the CFPB calculates the proxy for race and ethnicity. In response, the CFPB published a white paper describing in general terms its methodology for determining which borrowers belong to which racial and ethnic groups. As Director Richard Cordray explained in prepared remarks that coincided with the release of the white paper, “in order to estimate consumers’ race and national origin, our researchers use borrowers’ last names and their place of residence” and further noted that “Census Bureau data allows us to calculate the probability that an individual belongs to a particular race and ethnicity based on their last name and the demographics of the area where they live.” The methodology of combining race and place of residence to create a proxy for race and ethnicity is referred to as the Baysien Improved Surname Geocoding.
The white paper falls far short of answering all of the questions posed by regulated entities. For instance, a fair lending analysis involves a comparison of outcomes for applicants in a protected class against “similarly situated” applicants not in the protected class. But how does one choose the “similarly situated” borrower? The question remains unanswered. In this and many other ways, the white paper serves to provide only a general overview of the proxy methodology that the Bureau and other agencies have been using. As a result, regulated entities will have to keep working hard, using their best statisticians and advisors, to proactively put together the most accurate and sensible proxy methodology for their business.