Two of those tools — the no‐action letter and the compliance assistance sandbox — equip the CFPB with broad authorities to address various regulatory questions, including fair‐lending risk associated with the use of machine learning and alternative data in credit underwriting.
One example of this is in an August blog post that updated its first issued no‐action letter to Upstart Network, an online marketplace lender that uses alternative data for underwriting. In the post, the CFPB encouraged fintech lenders to take advantage of such policy tools to reduce their own fair lending compliance risk.
More of these no‐action letters that offer a “safe harbor” from the CFPB might benefit a handful firms, but the market as a whole will not reap the rewards until the agency issues generally applicable guidance.
When Upstart applied for the no‐action letter in 2017, there was a tremendous amount of regulatory uncertainty around disparate impact testing — when disparities are found between groups, though unintentional — as related to the use of machine learning and nontraditional data.
Regulatory agencies had little experience with those new and innovative credit models. And there was little regulatory guidance to help new fintech lenders monitor and manage the enhanced fair‐lending risk inherent in those models.
It was against that backdrop that CFPB staff issued a no‐action letter to Upstart in 2017. In addition to market signaling, one primary goal of the letter was to afford the CFPB a ringside seat to gain experience and expertise that would enable the agency to formulate a sound, general policy in the future.
The Upstart letter has a number of novel ideas.
For example, a (very welcome) regulatory innovation is the use of a hypothetical model that contains traditional application and credit variables, but does not use machine learning as the baseline for credit‐access analysis and disparate impact testing.
Too often, regulators compare the outcomes of innovation to a distant ideal rather than an imperfect status quo. Regulatory realism that recognizes the value of incremental improvements and gradual harm reduction is a step in the right direction.
The Upstart no‐action letter for the first time provides a detailed roadmap of fair lending compliance. Unfortunately, all of the regulatory and compliance innovation in the letter is confidential and so far, benefits just one company.
The regulatory uncertainty that existed in 2017 remains unchanged. What has changed, however, is that the CFPB (through the Upstart collaboration) has now developed a wealth of knowledge about how to manage and mitigate fair lending risk for machine learning models.
Now is the time to leverage those insights to develop policies that benefit not just Upstart but the entire industry.