On November 30, 2020, the Consumer Financial Protection Bureau (“CFPB” or “Bureau”) granted a no-action letter (“NAL” or “Letter”) to Upstart Network, Inc. (“Upstart”), a company that that has developed a model incorporating alternative data and machine learning for use in making credit underwriting and pricing decisions. The NAL specifically addresses Upstart’s “automated model for making underwriting and pricing decisions with respect to applications by consumers for unsecured, closed-end loans.” Under the terms of the NAL, the Bureau will not make supervisory findings or bring an enforcement action against Upstart under certain sections of the Equal Credit Opportunity Act and Regulation B or under the Bureau’s authority to prevent, unfair, deceptive, or abusive acts or practices (“UDAAPs”) concerning alleged discrimination on a prohibited basis arising from Upstart’s use of its model for making underwriting and pricing decisions on applications by consumers for unsecured, closed-end loans. The CFPB updated its NAL Policy last year, and this Letter was issued consistent with those guidelines. The NAL expires three years after the date of the Letter. The NAL represents another step forward toward regulatory acceptance of the use of alternative data and machine learning models for credit underwriting.
The CFPB issued its first ever NAL to Upstart in September 2017, allowing the company to use alternative credit data to evaluate borrowers. As a condition for receiving this new NAL, Upstart agreed to a “model risk management and compliance plan” that required it “to analyze and appropriately address risks to consumers, as well as assess the real-world impact of alternative data and machine learning.”
Pursuant to the original NAL, Upstart provided the Bureau with information comparing outcomes from its underwriting and pricing model against outcomes from a hypothetical model that used “traditional application and credit file variables” and did not employ machine learning. Upstart also independently validated the traditional model with fair lending testing to ensure that it did not violate antidiscrimination laws. A review of 2017-2019 lending activity by Upstart found that Upstart’s machine learning model approved 27% more applicants than a traditional underwriting model with 16% lower average APRs for approved loans. Moreover, applicants under 25 years of age were 32% more likely to be approved.
Both the 2017 NAL and the new NAL apply narrowly to the specific details of the lending products and modeling techniques that Upstart provided in its application. While the new NAL may encourage innovation in lending using alternative data and machine learning technology, the narrow scope of the NAL leaves open questions about how industry should conduct fair lending testing of a machine learning credit underwriting model and how creditors can incorporate alternative data and machine learning models into credit underwriting without raising fair lending or UDAAP concerns.