Is the CFPB Overestimating Racial Bias in Auto Loans?

WASHINGTON — The financial services industry believes it has finally discovered proof that the Consumer Financial Protection Bureau uses flawed methods to detect possible discrimination in auto loans.

A recently released industry-commissioned review of 8.2 million auto loans found that the CFPB's method was overestimating African-Americans and other minority populations, potentially leading to inflated enforcement actions. If the agency doesn't know a borrower's race accurately, it casts doubt on whether it can detect racial bias, the study argued.

Though the CFPB and consumer groups dispute some of the study's findings, the industry is using it as they press their case on Capitol Hill and elsewhere that the agency is going too far in its crackdown on auto lending.

The CFPB's methodology is "flawed in its application and subject to significant measurements that lead to overestimation of racial disparities and calculations of alleged harm," said Chris Stinebert, president and chief executive of the American Financial Services Association, which commissioned the report conducted by Charles River Associates. The study "was even more aggressive in its findings than we had expected."

The industry has been battling the CFPB over its methods since March 2013, when the agency first warned lenders that they could be cited for so-called disparate impact when their partnering dealerships charge a different rate to minorities based on a portfolio-wide analysis.

At issue is how the CFPB identifies minorities and other legally protected classes. For home loans, a borrower's race is included in the application. But it's not the same in auto lending, where that information isn't available.

As a result, regulators like the CFPB use a proxy method to estimate the race of a borrower by instead looking at trends in surnames and geography. They often use a system called the BISG (Bayesian Improved Surname Geocoding), which has become the most common and arguably best option regulators use to estimate a borrower's race.

But even regulators acknowledge the system is imperfect, identifying some borrowers as minorities when they are not.

The CFPB recently released a white paper meant to answer concerns raised by the industry and some policymakers on its proxy methodology. In the report, the agency did a sampling study and noted that the BISG system was overestimating African American and, to a lesser degree, Hispanic populations based on certain probability thresholds.

"As we stated in our September white paper, the bureau is committed to continuing our dialogue with other federal agencies, lenders, advocates, and researchers regarding the BISG proxy methodology, and we view the American Financial Services Association-funded policy paper as an effort to inform that dialogue," said Sam Gilford, a CFPB spokesman.

But the Charles River study released Nov. 19 said the problems are worse than the CFPB acknowledged. The study found that the BISG system was overestimating African Americans by more than 40% based on a sample size of the actual population. The CFPB did a similar case study in its white paper and found the BISG system overestimated by about 20% of the actual African American population.

That has raised concerns even among some former CFPB officials familiar with the agency's program.

"One, the paper reports that there's a significant error rate in the BISG results, such as a significant rate of false positives. And two, that there are some controls that the CFPB could implement that might reduce the error rate," the former official said on condition of anonymity. "The bureau should analyze this paper carefully."

The CFPB's Gilford said the agency is reviewing the study and the suggestions it made, but he cautioned there were problems with the industry's report, including justifying why certain minorities paid more for car loans.

"We continue to review these recommendations, but note that several of them are premised on unsubstantiated justifications for African-American, Hispanic, and Asian and Pacific Islander borrowers paying higher markups than similarly-situated white borrowers," he said.

But the industry argues that in addition to overestimating the population of a race, the agency is also overestimating the pricing disparities for minorities who get auto loans. Lenders fear that when a firm is cited for disparate impact and required to send checks to harmed consumers, they could be sending money to the wrong people if the BISG system is overestimating the affected population.

"The study reveals that a real challenge to the CFPB's testing methodology is how it determines who gets a check or a rate adjustment. Because of the high error rate, it is likely that checks or rate adjustments will be provided to people not assumed to be harmed," said Paul Metrey, chief regulatory counsel at the National Automobile Dealers Association. "This is among the inherent flaws in the bureau's methodology."

The Charles River study further contends that the loan pricing disparities identified in the BISG system are inflated by 87% for African Americans and 57% for Hispanics relative to the raw pricing disparities identified using the actual race.

The CFPB has "got to come to grips with flaws in its method, some of which are revealed in its own white paper," Metrey said. The "CFPB needs to explain how it corrects for those flaws and it needs to build in controls to ensure the consumers it is comparing are similarly situated."

But consumer groups argue that the industry-commissioned Charles River study set out to diminish any possible auto loan pricing among minorities.

"The way the analysis was done included a lot of factors that artificially diminish the impact of race in their model" but "you also have to factor in if additional variables that have a high correlation to race," said Delvin Davis, senior researcher at the Center for Responsible Lending.

For example, loan data reports have historically shown that minorities tend to fall into the subprime loan category based on credit scores and history. Subprime borrowers tend to get less favorable rates, or receive fewer offers than prime borrowers so they have less to bargain with in reducing their rate at the dealership.

"Even in comparisons of consumers with the same credit score, race is still a significant factor and you have to be able to mitigate that," Davis said. "We have a building case of evidence where there is a problem in pricing disparities with minorities and I think the CFPB knows that."

Since the CFPB took its first significant stance on the topic by citing Ally Financial for disparate impact last December in conjunction with the Justice Department, the industry seems to have largely accepted that the agency would not back down from using a portfolio-wide analysis to find statistical disparities.

"We're very hopeful that the study and data can be used in our continuing discussions with the CFPB," Stinebert said. "We all have the same goal: to reduce any overt discrimination or disparate impact that might occur in the auto finance industry."

However, the industry is pressuring the CFPB to consider other controls in its methodology, largely business and competitive factors that can influence the price of a loan and which differ greatly among dealers. For example, one dealer might have more inventory than another dealer so they cut the price of the loan to move more cars off the lot. That would likely not show in a portfolio-wide analysis at the indirect auto lender without some clearly marked explanation from the dealer. The NADA launched a fair lending compliance program earlier this year where dealers would voluntarily document their reasons for deviating from the wholesale rate offered by the lender.

"As demonstrated in the AFSA study, there are clear limitations in using the BISG proxy methodology to assess disparate impact in auto financing," said Gina Proia, chief communications officer at Ally Financial. "Ally believes that the most prudent course is for finance providers, dealers and other participants to come together to determine an industry-wide solution that addresses the issue of disparate impact without triggering significant unintended consequences."

As for establishing more controls in its methodology, the CFPB argues that is already taking into account certain "economic" controls that the industry suggested.

"We also note that the bureau seeks to maintain a robust dialogue with every lender subject to its supervisory and enforcement authority, and in that context has considered every economic control suggested by lenders and in some instances agreed to incorporate such controls into its analysis," Gilford said. "In assessing proffered controls, however, we require more than mere correlation with markup. Instead, consistent with the legal standard applicable to disparate impact, we look for evidence that allowing differential markups on the basis of a particular factor is supported by a legitimate business need of the lender under evaluation. Here as elsewhere, our goal is to make our financial markets work better for everyone involved, and to create a fair marketplace for all consumers."