Measuring and Mitigating Racial Disparities in LLMs: Evidence from a Mortgage Underwriting Experiment
Preprint
- 3 June 2024
- preprint
- Published by Elsevier in SSRN Electronic Journal
Abstract
We evaluate LLM responses to a mortgage underwriting task using real loan application data. Experimentally manipulated race is signaled explicitly or through borrower name/location proxies. Multiple generations of LLMs recommend more denials and higher interest rates for Black applicants than otherwise-identical white applicants, with larger disparities for riskier loans. Simple prompt engineering can cost-effectively mitigate these patterns. Race-blind recommendations correlate strongly with real lender decisions and predict delinquency, but LLMs incorporate racial signals when available despite similar delinquency rates across groups. Our findings show potential costs of adopting this new technology in financial settings and raise important questions for regulators.Keywords
This publication has 31 references indexed in Scilit:
- Lender Automation and Racial Disparities in Credit AccessThe Journal of Finance, 2024
- Biases in Large Language Models: Origins, Inventory, and DiscussionJournal of Data and Information Quality, 2023
- Racial Disparities in the Auto Loan MarketThe Review of Financial Studies, 2022
- Consumer-lending discrimination in the FinTech EraJournal of Financial Economics, 2022
- Predictably Unequal? The Effects of Machine Learning on Credit MarketsThe Journal of Finance, 2021
- Color and credit: Race, regulation, and the quality of financial servicesJournal of Financial Economics, 2021
- Achieving fairness in medical devicesScience, 2021
- Does Borrower and Broker Race Affect the Cost of Mortgage Credit?The Review of Financial Studies, 2020
- What Drives Racial and Ethnic Differences in High-Cost Mortgages? The Role of High-Risk LendersThe Review of Financial Studies, 2017
- Taste‐Based or Statistical Discrimination: The Economics of Discrimination Returns to its RootsThe Economic Journal, 2013