The Problems In CFPB Process For Identifying Race
Jul 15, 2015
The Consumer Financial Protection Bureau recently blogged that “settlement packages” from its 2013 consent order with Ally would be mailed soon — but who will get the checks? Almost two years from the settlement, the CFPB seems no closer to identifying the minority borrowers purportedly harmed by Ally, and so far has refused to identify even how many settlement packages it will send.
In December 2013, the CFPB and the U.S. Department of Justice ordered Ally Financial Inc. and Ally Bank to pay $80 million in damages to minority customers who were allegedly harmed by Ally’s automobile loan policies. On June 15, the CFPB announced that “over the next weeks” the settlement administrator will begin mailing packages “to identified borrowers with instructions for how to participate in the settlement” with an ultimate goal of locating and sending checks to “affected borrowers who were overcharged.” However, this announcement raises questions that the CFPB has not answered: Who is receiving notice that they may be eligible to participate in the settlement and how did the CFPB identify them?
In their investigation, the CFPB and DOJ did not allege discriminatory differences in Ally’s buy rates, which vary with the income and creditworthiness of the loan recipient, and the size of the loan and the down payment. Rather, they alleged that the “dealer markup” — the difference between the ultimate interest rate negotiated between the customer and the dealer and the initial buy rate agreed to by Ally — resulted in higher interest rates for African-American, Hispanic and Asian/Pacific Islander borrowers.
In their consent order, the CFPB and the DOJ argued that between April 1, 2011, and Dec. 31, 2013, approximately 100,000 African-American borrowers paid, on average, over $300 more in interest than non-Hispanic white borrowers. Similarly, they argued that over that same period, approximately 125,000 Hispanic borrowers and 10,000 Asian/Pacific Islander borrowers paid approximately $200 more in interest, on average, than non-Hispanic white borrowers.
As outside observers, we have serious questions about the CFPB’s approach. Its determination of racial discrimination, and the resulting $80 million settlement, was based on analyses of Ally automobile loan data that included no information about a borrower’s race or ethnicity. As a result, not only does the CFPB not know which loan recipients are minority borrowers, it also cannot know which minority loan recipients were allegedly harmed by Ally’s actions. Although the CFPB says it will mail packages to “identified borrowers,” the methodology it has used to impute race will likely cause the CFPB to contact many borrowers who are not African-American, Hispanic or Asian/Pacific Islanders.
The CFPB Process for Identifying Race
The CFPB monitors lenders for compliance with the Equal Credit Opportunity Act, and the ECOA’s Regulation B generally prohibits a lender from inquiring “about the race, color, religion, national origin, or sex of an applicant or any other person in connection with a credit transaction.” As a result, a borrower’s race and ethnicity are unknown to lending companies — and to their regulators. The CFPB attempts to address this limitation by using “Bayesian Improved Surname Geocoding,” or BISG. BISG uses information from the 2000 census about the self-identified race and ethnicity of individuals with relatively common surnames, and the racial composition of neighborhoods, to construct a set of probabilities describing the race and ethnicity of a given borrower.
For example, suppose an Ally borrower named Ms. Smith (the most common surname in the U.S.) lives in the neighborhood near the CFPB headquarters in Washington, D.C. BISG first uses information about the self-reported race of people named Smith in the 2000 census: 73.6 percent were non-Hispanic white, 22.2 percent were African-American, and the remaining 4.2 percent were distributed across other race and ethnicities. Next, BISG uses more recent information about the demographic composition of the relevant neighborhood (67.2 percent of residents are non-Hispanic white, 11.7 percent are African-American, etc.) to update the original estimate and impute that there is a 74.7 percent chance that Ms. Smith is non-Hispanic white, a 20.4 percent chance she is African-American, a 0.8 percent chance she is Asian/Pacific Islander and a 0.5 percent chance she is Hispanic.
The CFPB applied the BISG methodology to all Ally borrowers who have one of the roughly 150,000 surnames that appeared 100 or more times in the 2000 census and summed the race and ethnicity probabilities across borrowers to estimate the number of minority borrowers. Ms. Smith’s loan in Washington, D.C., would have contributed 74.7 percent to the non-Hispanic white borrower total, 20.4 percent to the African-American borrower total and fractions of 1 percent to Asian and Hispanic borrower totals. But neither Ally nor the CFPB knows the race or ethnicity of Ms. Smith.
The CFPB used this methodology to study approximately 800,000 Ally loans between April 1, 2011, and March 31, 2012. It further estimated that Ally made an additional 1.3 million loans between April 1, 2012, and Dec. 31, 2013. Based on the BISG methodology, the CFPB estimated that more than 20 percent of those loans had borrowers or co-borrowers who were African-American, Hispanic or Asian/Pacific Islander.
Problems Identifying Allegedly Injured Borrowers
Because the BISG methodology has not actually determined the race or ethnicity of a given borrower, but rather the probability that a given borrower belongs to each demographic group, it alone cannot be used to identify who is eligible to participate in the settlement. Returning to the example of Ms. Smith, based on the BISG methodology, there is a possibility that her race/ethnicity is non-Hispanic white and she is ineligible for any settlement payment. However, it is also possible that she is African-American, Hispanic or Asian/Pacific Islander and eligible for a range of different payments. In the administration of the settlement, does Ms. Smith receive notification that she is potentially eligible to participate? This question applies to essentially every borrower. Is the CFPB sending notices to approximately 2.1 million borrowers because nearly all borrowers have some probability of being in one of the affected groups?
Further complicating matters are the CFPB’s statements in the consent order regarding the number of individuals allegedly affected. The CFPB states that approximately 100,000 African-American borrowers, 125,000 Hispanic borrowers and 10,000 Asian/Pacific Islander borrowers “paid higher markups” during the “relevant time period” (between April 1, 2011, and Dec. 31, 2013). Yet the CFPB estimated that more than 20 percent of Ally’s 2.1 million loans over this period (approximately 420,000) were made to individuals from these same races and ethnicities. Given that the CFPB doesn’t know the actual race and ethnicity of any individual borrower, it would be unable to identify who specifically has “paid higher markups” due to alleged discrimination. If the CFPB is only planning to send approximately 235,000 notices, in what way did it depart from the BISG methodology in order to determine who should receive these notices?
Other Methodological Problems with BISG
This relatively simple example of attempting to identify allegedly injured borrowers is indicative of a number of other potential problems whenever the BISG methodology is used to estimate race and ethnicity in an effort to ensure fair lending. For many surnames, there can be a large percentage of people in different races sharing the same name (for example, African-American and non-Hispanic white borrowers often have the same surnames). Similarly, the BISG methodology is further limited because the census surname file used is based on data from 2000 and the racial and ethnic composition of the U.S. population has changed considerably since then.
As the CFPB continues its efforts to ensure fair lending in accordance with the ECOA, it is doing so despite the fact that it must manufacture the central variable in its analyses — race and ethnicity. The pending administration of the Ally settlement only further illustrates the limitations of using an estimated race and ethnicity variable for every borrower in the analysis. If every borrower is contacted, a large number of ineligible and uninjured borrowers will be incorrectly notified of potential eligibility for a settlement payment. If a subset of borrowers is notified, the process of identifying these individuals will have occurred using a different set of criteria than was used to estimate the alleged injury to the borrowers. These flaws raise serious questions about how the $80 million settlement will ultimately be distributed to borrowers.
This piece is co-authored with Christopher Craig of O'Melveny & Myers and was originally published in Law360 on July 2, 2015.