A briefing on the enhancement of credit availability and the introduction of the Consumer Financial Protection Bureau’s (CFPB) inaugural No-Action Letter.

For certain consumers, leveraging non-traditional information sources, or “alternative data,” for credit assessment could pave the way to broader credit access or reduce the cost of borrowing. This alternative data, which is not typically included in the major consumer reporting agencies’ core credit files, may offer insights into a consumer’s likelihood of timely fulfilling financial obligations that traditional credit histories might miss.

Moreover, advancements in computing capabilities and the increased application of machine learning technologies have the potential to uncover correlations not identifiable through traditional credit scoring methods. These innovations could lead to improved credit availability or reduced costs for borrowers who currently face unfavorable credit terms.

Bureau Initiatives: In 2017, the Bureau sought public input through a Request for Information (RFI) on the use of alternative data and modeling techniques in credit processes. The RFI highlighted the potential benefits and risks of using alternative data and models, covering a range of topics from current applications to potential impacts and regulatory considerations. The Bureau expressed its intention to facilitate responsible use of alternative data and minimize barriers to its application.

Following this, the Bureau granted a No-Action Letter to Upstart Network, Inc., a firm that incorporates alternative data and machine learning in credit decisions. This No-Action Letter, while specific to Upstart’s context, aimed at clarifying the application of the Equal Credit Opportunity Act (ECOA) and its Regulation B in the context of using alternative data and machine learning for credit underwriting and pricing. It is not an endorsement of specific variables or models.

As a condition of the No-Action Letter, Upstart agreed to implement a risk management and compliance plan addressing consumer risks and assessing the impact of alternative data and machine learning. This involved comparing their model’s outcomes against a traditional model’s to evaluate access to credit, fair lending implications, and other metrics.

Findings:

  • The model employing alternative data and machine learning approved 27% more applicants than the traditional model and offered 16% lower average APRs for approved loans.
  • This expansion in credit access and reduction in APRs was consistent across various demographic groups, including race, ethnicity, and gender, with acceptance rates increasing by 23-29% and APRs decreasing by 15-17%.
  • Notably, “near prime” consumers, younger applicants, and those with incomes below $50,000 saw significant improvements in credit access.
  • Fair lending analysis comparing both models showed no disparities that necessitated further examination under the compliance plan.

Looking Ahead: Despite these advances, challenges remain in ensuring access to affordable credit, especially for the 26 million Americans without a credit history (“credit invisibles”) and the 19 million with outdated or insufficient credit histories. The Bureau encourages lenders to explore innovative solutions to provide fair, equitable, and non-discriminatory credit access to these groups while maintaining effective risk and legal compliance management programs. The Bureau’s ongoing efforts include reviewing feedback on proposed policies such as No-Action Letters, Trial Disclosure, and Product Sandbox initiatives to address credit access challenges.

Check the original article on this link.

Leave a Comment