Measuring and Mitigating Racial Bias in Large Language Model Mortgage Underwriting

52 Pages Posted: 3 Jun 2024

See all articles by Donald E. Bowen III

Donald E. Bowen III

Lehigh University

S. McKay Price

Lehigh University - Perella Department of Finance

Luke C.D. Stein

Babson College

Ke Yang

Lehigh University

Date Written: April 30, 2024

Abstract

We conduct an audit study of loan approval and interest rate decisions suggested by large language models (LLMs). Using a dataset of real loan applications and experimentally manipulated race and credit scores, we find that LLMs recommend denying more loans and charging higher interest rates to Black applicants than otherwise-identical white applicants. This racial bias is largest for lower-credit-score applicants and riskier loans, but present across the credit spectrum. Surprisingly, simply instructing the LLM to make unbiased decisions eliminates the racial disparity in approvals and moderates the interest rate disparity. LLM recommendations correlate strongly with real lenders' decisions, despite having no fine tuning or specialized training, no macroeconomic context, and access to only limited data from each loan application. A number of different leading LLMs produce racially biased recommendations, although the magnitudes and patterns vary. Our results highlight the critical importance of auditing LLMs and demonstrate that even basic prompt engineering can help reduce LLM bias.

Keywords: Mortgage underwriting, Real estate, Consumer protection, Lending decisions, Fintech, AI, Large Language Models, ChatGPT, Discrimination, Bias, Financial inclusion, Access to finance, Fair Lending, ECOA

JEL Classification: G21, G28, J15, R30, L85, R20

Suggested Citation

Bowen III, Donald E. and Price, S. McKay and Stein, Luke C.D. and Yang, Ke, Measuring and Mitigating Racial Bias in Large Language Model Mortgage Underwriting (April 30, 2024). Available at SSRN: https://ssrn.com/abstract=4812158 or http://dx.doi.org/10.2139/ssrn.4812158

Donald E. Bowen III (Contact Author)

Lehigh University ( email )

Bethlehem, PA 18015
United States

S. McKay Price

Lehigh University - Perella Department of Finance ( email )

621 Taylor Street
Bethlehem, PA 18015
United States
610-758-4787 (Phone)

HOME PAGE: http://www.mckayprice.com

Luke C.D. Stein

Babson College ( email )

Tomasso Hall
231 Forest Street
Babson Park, MA 02457-0310
United States
781-239-5060 (Phone)

HOME PAGE: http://lukestein.com

Ke Yang

Lehigh University ( email )

621 Taylor Street
Bethlehem, PA 18015
United States
6107583684 (Phone)
6107586429 (Fax)

Do you have a job opening that you would like to promote on SSRN?

Paper statistics

Downloads
120
Abstract Views
431
Rank
432,003
PlumX Metrics