On August 4, 2019, the Consumer Financial Protection Bureau (“CFPB”) published a blog summarizing some of the results of simulations and analyses on an alternative data and credit model developed by Upstart Network, Inc. (“Upstart”), which received a no-action letter (“NAL”) from the CFPB in 2017. The blog is significant because it shows that a Fintech was able to successfully navigate the NAL process, and its research findings may support the further use of alternative data and artificial intelligence in the development and successful deployment of such credit underwriting models.

Upstart received its NAL from the CFPB in 2017 which was predicated upon adhering to an approved model risk management and compliance plan. Upstart further agreed to share key highlights from the simulations and analyses that it conducted pursuant to the model risk management and compliance plan approved by the CFPB. The blog revealed encouraging results from the NAL modeling exercise. It showed that the use of nondiscriminatory alternative data and the machine learning methodology used in Upstart’s model as applied to the applicant pool yielded positive outcomes in terms of expanded access to credit. Under the tested model:

  • 27% more applicants were approved over the traditional model and yielded 16% lower average APRs for approved loans
  • Improved access occurred across all races, ethnicities, and sexes
  • Acceptance rates increased by 23-29% and APRs decreased an average of 15-17%
  • So-called “near prime” consumers with FICO scores ranging from 620 to 660 were approved approximately twice as frequently
  • Applicants less than 25 years old were 32% more likely to be approved
  • Consumers with incomes under $50,000 were 13% more likely to be approved

From a fair lending perspective, approval rates and APR analysis results for minority, female, and age 62 and older applicants showed no disparities. The potential for adverse fair lending outcomes from unvalidated alternative credit models has been a major concern of regulators and policy makers. This concern has held back a wider use of alternative data in consumer credit transactions to date.

The blog’s summary of the results from the first NAL should encourage other Fintechs to utilize the CFPB’s NAL process. From a public policy perspective, it also may undercut the CFPB’s pending NAL proposal, which does not require the sharing of simulations and analyses data in the case of all approved NALs.

The most significant impact of the Upstart NAL findings is that it demonstrates that properly validated alternative credit underwriting models can be successful in expanding access to credit to otherwise creditworthy borrowers who might not meet traditional underwriting models’ credit criteria.

Initial Fintech in Brief Advisory: The CFPB’s Proposed Policy on No-Action Letters and Product Sandbox