A Profit-Based Measure of Lending Discrimination

• In Event: Machine Learning and Fairness: Applications in Lending, Benefits, and Clinical Decision-Making

Abstract

Algorithmic lending has transformed the consumer credit landscape, with complex machine learning models now commonly used to make or assist underwriting decisions. To comply with fair lending laws, these algorithms exclude legally protected characteristics, such as race and gender. Yet algorithmic underwriting can still inadvertently favor certain groups, prompting new questions about how best to audit lending algorithms for potentially discriminatory behavior. Building on prior theoretical work, we operationalize a profit-based measure of lending discrimination to audit fintech lending decisions. We define discriminatory loan pricing as pricing disparities that correlate with protected characteristics and that cannot be explained by differences in creditworthiness. Our approach for detecting discriminatory lending  relies on straightforward intuition: If a lender prices loans accurately, then---after setting interest rates to reflect risk---we would expect loans to be similarly profitable across groups. Consequently, if loans made to a particular group are systematically more profitable, this suggests that the lender priced those borrowers too aggressively relative to their true risk, consistent with discrimination. To compare profitability, we rely on the annualized internal rate of return (IRR), a common measure of profit in lending that directly connects loan pricing with repayment outcomes. Applying our approach to approximately 80,000 personal loans from a major U.S. fintech platform we find that loans made to White, Hispanic, and Asian borrowers earn similar returns (between 8.3% and 8.8%), whereas loans made to Black borrowers earn noticeably less (7.7%). Likewise, the profits of loans made to men (8.3%) fall below those of women (9.1%). These results thus suggest the platform's algorithmic pipeline ultimately benefits Black borrowers and men, offering them relatively favorable loan terms. We trace these disparities to miscalibration in the platform’s underwriting model, which underestimates credit risk for Black borrowers and overestimates risk for women. We show that one could correct this miscalibration— and the corresponding lending disparities—by explicitly including race and gender in underwriting models, illustrating a tension between competing notions of fairness.

Author

• Madison Elisse Coots, Harvard University
  Presenting Author