Our CEO on FHFA's (Federal Housing Finance Agency) Session on Credit Scores

At the Federal Housing Finance Agency’s recent listening, SolasAI CEO Nicholas Schmidt provided much-needed context on the analyses required to understand the fair lending impact of proposed changes to how GSEs use credit scores. Nicholas emphasized that a simple comparison of scores’ relative coverage rates is not sufficient when considering whether the FHFA should increase the number of scores submitted to GSEs by lenders.

It is, in fact, possible that additional coverage may harm minorities and traditionally underserved communities. For this reason, he cast a critical eye on the association between having multiple credit scores and better credit outcomes without further testing, and he reiterated the need for the GSEs to perform disparity testing and discrimination mitigation in the models they use to score borrowers.

-

My name is Nicholas Schmidt, and I am a partner at BLDS.?We are a consultancy based out of Philadelphia that uses statistics and economics to assess evidence of discrimination in employment, housing, and lending.?I am also the CEO of SolasAI, which provides software that tests for and mitigates potential discrimination caused through algorithmic decisioning.?I have worked at the intersection of law, regulation, and economics for over twenty years.?My clients have included many of the largest U.S. lenders and Fortune 50 companies, as well as many Federal, state, and local government agencies charged with enforcing anti-discrimination laws.

As we all understand, it is essential to consider the fair lending implications of any changes in how credit scores are reported to or utilized by the Enterprises.?Understanding the impact of these changes requires looking beyond the superficial appeal of requiring multiple scores, such as the assumption that increased coverage rates result in better outcomes for consumers.?The Enterprises utilize credit scores in nuanced ways throughout the lifecycle of a funding decision.?As a result, even seemingly beneficial elements of the proposal may not result in more favorable fair lending outcomes.

In fact, allowing or requiring lenders to provide multiple scores to the Enterprises may, counterintuitively, harm minorities and traditionally underserved communities.?This may be true across any of the three proposed methods for considering multiple scores.?Determining the impact of these changes is an empirical question, answered through a review of the data available.?It is essential that this analysis not only incorporate high-level effects such as coverage rates, but also consider the many downstream ways that scores affect outcomes for borrowers.

In this, I would like to underscore three key elements that must be considered:

• First, because no loan is rejected solely because borrowers lack a credit score, unscored consumers are not necessarily harmed by lacking coverage;
• Second, borrowers who do have credit scores can have their loans automatically rejected if those scores are too low.?As a result, extending coverage by giving borrowers low scores may actually lead to more rejections of loans for these borrowers than if they did not have any score at all; and,
• Third, inequitable or differential outcomes in loan acceptances may be better improved by having the Enterprises minimize disparities in the scores that they create for determining eligibility.?Here, it is worth stating that the Enterprises make little or no use of the publicly available credit scores when they determine eligibility using their DU and LP algorithms.?

Regarding the first two points, we know that the Enterprises have attempted to increase the pool of eligible borrowers by using different data and different models for unscorable customers.?This means that the lack of a credit score does not, per se, result in the rejection of a loan.?On the other hand, the Enterprises do use credit score cutoffs when the credit scores of borrowers are available.?As a result, a borrower who is rejected because of a low score, might have been accepted under the Enterprises’ nonscorable models.?This is the counterintuitive result: if one credit score has larger coverage than another, but the increased coverage primarily comes at the lower end of the score distribution, then borrowers may be made worse off by making multiple scores available to the Enterprises.

To state it another way, since the Enterprises have methods for evaluating unscored borrowers, increased credit score coverage through lower scores may result in fewer acceptances.?Whether this is the case, and whether it affects minorities disproportionately is an empirical question.?But given its possibility, the assumption that increased coverage leads to increased access to loans is worth evaluating.

Further, beyond cutoffs, the Enterprises make relatively little use of credit scores to determine loan acceptance.?Instead, they primarily use their own models for this purpose.?As a result, under the current processes and proposals, it is likely that the simplest and most effective way for the Enterprises to expand access to credit would be through performing discrimination testing and disparity mitigation of their own models.?

To summarize, the mechanisms by which the Enterprises utilize credit scores is complicated and the impact of these proposed changes will be as well.?The most effective way to determine the impact of the different proposals is through testing.?That is, we should determine if more minorities and underserved borrowers would get access to conforming loans when using one credit score or when using multiple credit scores.?If the results of these tests show that requiring additional scores leads to significant changes the number of minorities accepted, then this should be given an appropriate weight relative to any other potential benefits that may come by changing the scoring process. However, simply giving weight to differences in coverage is inappropriate and can be misleading.