First in a series about the CFPB's use of the disparate impact theory.
WASHINGTON — The Consumer Financial Protection Bureau's method for detecting discrimination by indirect auto lenders can overestimate potential bias, resulting in higher payments for lenders cited by the agency, according to internal CFPB documents.
In a series of private documents that were reviewed by American Banker, CFPB officials repeatedly acknowledge its methodology could overcount the potential discrimination by firms, but say they prefer that to the alternative where bias is underestimated.
The agency's methodology "is valid and reasonable under the circumstances here, and although there may be some risk of overestimating disparities, the alternative presents an equal (and perhaps greater) risk of underestimating disparities and thus consumer harm," Patrice Ficklin, assistant director of the CFPB's Office of Fair Lending, stated in a memo to agency officials written in April 2013.
The memo was in response to objections raised by a large bank under investigation. The institution argued that the disparities were less severe than what the CFPB found when the method included other variables.
The documents illustrate how and why the agency's "disparate impact" methodology remains such a controversy within the financial services industry.
Several agencies, including the Justice Department, cite companies under disparate impact, a legal theory that says lenders can be penalized if they have a neutral policy that creates an adverse impact against a protected class of borrowers, regardless of intent. But the methods involved, which vary agency to agency, are essentially sophisticated guesses. It is particularly challenging in certain cases such as auto lending, which unlike mortgages lacks concrete data about the race and gender of a borrower.
As a result, agencies including the CFPB use so-called "proxy" data that attempts to divine whether a borrower is a minority by looking at surnames, geographic location or a combination of both.
But observers and former officials suggest the CFPB's method might be problematic because it often overestimates the disparities — and could ultimately lead to ineligible consumers getting refunds, such as white borrowers who were never discriminated against.
"It's an inherent issue with the use of a proxy and there really isn't a good answer for it, which I suspect is why you don't see much detail on the remuneration of borrowers in the two settlements that have been made public to date" by the CFPB, said Joe Rodriguez, who formerly worked in the CFPB's fair-lending group and is now of counsel at Morrison & Foerster. "In other words, the way the bureau applies the proxy for determining the overall damage calculation vis-á-vis a lender's portfolio doesn't really work when applied to individual consumers for remuneration because it would result in remuneration being distributed to large numbers of people who in all likelihood aren't [minority or other protected class] borrowers, or even disadvantaged by the dealer markup policy."
Yet some consumer groups agree that it's better for the CFPB to err on the side of overestimating potential discrimination, or it risks failing to spot it where it might exist.
"No proxy method is going to have 100% accuracy. However, the methodology the CFPB has released publicly has a high enough accuracy rate and, given the size of the data set, makes it a statistically valid and appropriate method," said Chris Kukla, senior vice president at the Center for Responsible Lending. "Any argument that the CFPB is overcounting is to distract you from the significant number of people who can be affected by dealer markups. Even in a case where 235,000 consumers were assumed to be affected, if the CFPB overcounted by 10,000, that's still 225,000 people who were overcharged."
But the industry argues that the CFPB is being far too aggressive in its assessments, and fails to apply important data — such as credit scores of borrowers in one case — that would show fewer consumers are really being harmed.
The CFPB released a white paper last year offering some details on its disparate impact methodology, but car dealers, the financial industry and several members of Congress say it does not tell them nearly enough, particularly how the method is applied in enforcement.
The Consequences of Overestimating Discrimination
The impact of overestimating versus underestimating potential harm to consumers is not yet clear because so few cases have reached the level where borrowers are being made whole.
Only two indirect auto lenders have so far been publicly cited by the CFPB for discrimination, and only one of those, Ally Financial, is far enough along in a settlement to refund borrowers.
In December 2013, Ally and its bank reached a $98 million settlement with the CFPB and Justice Department. The regulators allege that while Ally had a neutral policy in which dealerships can increase the interest rate on a loan up to a certain percentage, that policy created a disparate impact on an estimated 235,000 minority borrowers who were charged higher rates than white borrowers.
Ally was ordered to pay $80 million to the agencies for consumer refunds, which an administrator would distribute to consumers who were contacted by the agencies. The lender paid the $80 million in January 2014 to an escrow account for purposes of the settlement fund and provided the customer information to the CFPB and DOJ in March 2014, an Ally spokesperson said. (The remaining $18 million in the settlement were civil money penalties.)
Yet more than a year and a half later, borrowers have yet to be refunded. A letter written in March by CFPB and DOJ officials to Ally customers asked consumers to identify themselves to the administrator if they were a minority and got a car loan financed by Ally under specified time frames. "We anticipate that checks will be mailed to all eligible customers later this year or early next year," the March letter said.
Consumers have until October 24 to respond.
Industry representatives see the delay in refunds as proof that the CFPB overestimated the number of people Ally allegedly discriminated against. Doing so would mean that there are fewer consumers who were actually harmed by Ally's policies — and that the company paid more than it had to in order to fix the problem.
"In order to have an accurate measurement of potential consumer harm, you have to isolate out legitimate pricing factors that can cause a deviation in the results. There are business factors that the Justice Department has recognized as legitimate that the CFPB appears to be blatantly ignoring during investigations," said Paul Metrey, chief regulatory counsel at the National Automobile Dealers Association. "There's also an added reputational harm to the actors when the CFPB is basing damages off overestimated figures. If you look at the press releases of these enforcement actions, they are not holding back."
But a CFPB spokesman, Sam Gilford, said the agency has a process in place to make sure checks will not go to the wrong people.
"The remuneration process was designed to avoid sending checks to ineligible consumers by seeking consumer responses regarding whether a borrower or co-borrower is African-American, black, Latino, Hispanic, of Spanish origin, Asian, Native Hawaiian, or other Pacific Islander," Gilford said.
This process would still depend on the consumer responding to the letters, which is a complication for any regulator. There is also the issue of what happens to the funds that aren't claimed by consumers.
"Alternatively, the bureau could decide to cut off remuneration at a certain confidence level" -- e.g., the proxy says the person is greater than 50% likely to be a prohibited basis borrower, Rodriguez said -- "but even then you will still have significant instances where people who aren't prohibited-basis [protected] borrowers get checks and some actual prohibited-basis borrowers won't get checks. And because the damages are based on larger assumptions, there also remains the issue of what to do with the inevitable leftover money after checks have been sent out."
Asked what happens to extra funds, the CFPB's Gilford said that "in Ally and other supervisory and enforcement matters, we anticipate distributing checks for the entire settlement fund[s] to harmed borrowers." Any leftover money "would be deposited in the U.S. Treasury as disgorgement."
American Banker reached out to the other federal regulators to find out how long it typically takes to remit money to affected consumers. But observers agreed it was difficult to determine a time frame since it varied so much case by case.
How the CFPB Overestimates Discrimination
The CFPB said it has previously acknowledged in its white paper that its disparate impact method could overestimate potential discrimination. That overestimation was based on using comparisons to mortgage loans, which, unlike auto loans, show the race and gender of the borrower.
"As we said in the white paper, we believe the likely cause of the overestimation is that the racial and ethnic makeup of mortgage applicants is not particularly representative of the general population," Gilford said. "When the proxy is applied to data where the applicants are more representative of the general population, such as data on auto loan borrowers, this perceived overestimation may disappear or decrease significantly."
In at least several cases with indirect auto lenders, however, internal documents suggests the agency has not considered using other data provided by the lender that shows fewer disparities. For example, in the April 2013 memo responding to the large bank, Ficklin notes that the bank used a method similar to the CFPB's and showed lower disparities when accounting for factors like whether the car is new or used. That result appeared to surprise CFPB staff. After analyzing both the lender's method and its own, the CFPB concluded both were valid.
"Our initial expectation was that OR's analysis of the two estimation methods would reveal that one or the other was plainly superior," Ficklin wrote, referring to the Office of Research. "However," the Office of Research "has concluded that the two methods are both reasonable, but under different assumptions about the underlying cause of the disparities."
The CFPB ultimately stuck with its own method, according to available documents. The case is ongoing.
The overarching question is what types of assumptions the CFPB is making in its analysis — and whether it is including all relevant data.
Each financial regulator has its own method for determining disparities and harm in fair-lending cases, and each of those cases can differ depending on the business model of the bank and what variables the regulators will consider. The Federal Reserve, for instance, generally adds controls, such as geography, to the statistical model if the bank's business model indicates that certain pricing criteria can influence the price or markup, according to a 2013 Fed presentation.
Some of the lenders who were in the early stages of the CFPB's investigative sweep argued that the CFPB did not appear to take into account certain controls that were legitimate business factors that would mitigate the amount of inadvertent discrimination the CFPB found.
In the case of Ally, for example, the bank argued that the CFPB's method did not account for items such as a borrower's credit score in order to estimate harm and total amount of damages. Ally claimed that placing such controls as a borrower's creditworthiness would reduce the disparities to one-third of what the CFPB claimed with certain minority groups, according to CFPB documents.
"The failure of the [CFPB's] preliminary analysis to control for creditworthiness in the preliminary analysis cited … significantly undermines the credibility of the analysis," Ally said in one response to the CFPB prior to reaching a settlement.
Though the CFPB did run tests using such controls, it ultimately disagreed with Ally, arguing that creditworthiness controls "neither model the markup policy nor reflect a legitimate business need."
During settlement discussions with Ally, CFPB also significantly raised its estimate of the harm to consumers based on its own method. The agency had initially calculated that Ally's policy caused more than 213,000 minorities to receive higher rates than white borrowers, totaling more than $41 million in possible direct damages. The initial analysis the CFPB used to identify disparity figures was "with no controls," such as "characteristics of the vehicle, and the timing, location, and structure of the deal," according to documents.
If the agency accounted for controls such as those Ally argued for, the amount of consumer harm would be much lower, just $5.3 million in damages to consumers — 13% of the CFPB's calculation, agency staff noted in one memo.
Ultimately, however, the CFPB nearly doubled its own initial estimate of damages to $80 million to include other factors that are not direct damages, such as "emotional distress." Damages also include projections for what consumers would pay extra during the life of loan, not just in past payments. The CFPB's assumption of the number of affected borrowers also jumped to 235,000 by the time it announced the settlement.
Ally's Precarious Position
In the case with Ally, some sources said the CFPB chose to act aggressively because it wanted to send a message to the industry and do so quickly. (An upcoming story will explore the CFPB's intentions in more detail.)
The agency also had extra leverage against the company. During the time the settlement was being discussed, Ally was waiting for the Federal Reserve to approve its application to convert from a bank holding company to a financial holding company so it could retain certain businesses. This put Ally in a precarious position, CFPB officials said, according to documents reviewed by American Banker.
If Ally did not get approval, it "could have a material adverse effect on Ally's business, resulting in operations, and financial position," CFPB officials said in an October 2013 memo directed to and signed by the agency's director, Richard Cordray.
The Fed had given Ally an extension that was set to expire in late December 2013 — the same month that Ally eventually settled with the CFPB. That deadline gave Ally an incentive to cut a deal, CFPB officials said.
"In order to convert to a financial holding company, [Ally Financial] and its bank subsidiary must, among other requirements, be considered well-managed under the BHCA [Bank Holding Company Act]. As such Ally may be strongly inclined to reach a timely and robust resolution of this matter if it can potentially result in [Ally Financial] successfully converting to a financial holding company," CFPB staff said in the memo to Cordray. "Recent discussions with Ally on October 3 in which Ally expressed a strong willingness to settle this matter quickly are consistent with this analysis."
It's unclear whether the results from Ally's settlement would have come out differently had there not been external pressures at the time.
Some in the industry argue that Ally might have fought harder against the CFPB and its methodology.
The CFPB would not discuss the circumstances of the negotiations but said the harm exists regardless.
"We will decline to comment on the settlement with Ally, beyond noting again that the CFPB and DOJ's investigation found that minority borrowers were being charged hundreds of dollars more in interest than white borrowers of similar credit profiles. As a result of the action, thousands of borrowers will receive remediation for the harm they suffered," Gilford said.
Despite external pressures on Ally, some agree that the CFPB is doing the best it can to actively weed out potential discrimination in light of the fact that dealerships were exempted from its jurisdiction in the Dodd-Frank Act.
"You've got to be able to tie something back to a decision or practice of the lenders themselves. And with indirect auto lending, the only thing the lender has done is given the dealership the ability and discretion to mark up the loans," said Raj Date, managing partner at the venture investment firm Fenway Summer and the former No. 2 at the CFPB. "It would be a lot more straightforward to just look at direct evidence of dealer conduct, of course, but that's not a policy lever that Congress gave to the CFPB."