A Profit-Based Measure of Lending Discrimination (arxiv.org)

A new study has unveiled a profit-based measure of lending discrimination within algorithmic underwriting, spotlighting the unintended biases that can arise from machine learning models in consumer credit. Analyzing around 80,000 loans from a major U.S. fintech platform, researchers discovered that loans issued to men and Black borrowers generated lower profits than those granted to other demographic groups. This anomaly revealed a miscalibration in the platform's underwriting model, which undervalued credit risk for Black applicants while overestimating it for women. This finding is significant for the AI/ML community as it underscores the complexities of fairness in algorithmic decision-making. While existing models often exclude sensitive characteristics like race and gender to comply with fair lending laws, the study suggests incorporating these variables could enhance the accuracy of risk assessments and promote equitable lending practices. This research highlights an important tension between algorithmic fairness and profitability, urging a re-examination of how algorithms are built and audited to prevent discrimination while optimizing yield in the lending sector.