CFPB Releases Outcomes from Use of Alternative Data in Credit Underwriting

The Consumer Financial Protection Bureau (CFPB) has released information from its review of alternative data underwriting by Upstart Network, Inc. (Upstart).  Upstart received a No-Action Letter from the CFPB in 2017 that allowed the company to engage in credit underwriting and pricing operations using alternative data and machine learning techniques, subject to oversight by the Bureau and adoption of a compliance plan.  On August 6, 2019, the Bureau announced some of its findings from close observations of Upstart’s models in comparison to a hypothetical model that relied on traditional data and tools.

According to the Bureau, the model employed by Upstart led to 27% more approvals than the traditional models, leading to a reduction in the average APR on approved loans by 16%.  The Bureau found that these results were distributed across “all tested race, ethnicity, and sex segments.”  The Bureau also found “no disparities that require further fair lending analysis under the compliance plan.”

While these findings demonstrate a positive application of new data and techniques to credit underwriting, no analysis should be considered complete without observation of losses.  Only by determining the performance of loans over time can the impact of new underwriting techniques be completely assessed.  Likewise, while a reduction of APR is a positive initial observation, determining whether a loan is priced properly requires assessment of underlying operational costs as well as losses.  AFSA urges the Bureau to extend its research of alternative data and machine learning to cover the entire lifecycle of consumer finance.