New Report Outlines Method for Improving Mortgage Fairness
FOR IMMEDIATE RELEASE
April 24, 2024
Contact: Janelle Brevard | jbrevard@nationalfairhousing.org
Washington, D.C. — Algorithmic fairness techniques can eliminate bias in mortgage underwriting without compromising accuracy, according to a new report by the National Fair Housing Alliance™ (NFHA) and Los Angeles-based fairness-as-a-service fintech company FairPlay AI.
“Disparity in mortgage approval rate persists despite decades of legislation and policy actions aimed at eliminating it. As artificial intelligence becomes more ubiquitous in housing and lending, experts have begun looking for ways to leverage fairness techniques to root out rather than entrench biases in AI-based underwriting systems as a necessity to ensure that AI does not become a technology that perpetuates or exacerbates the legacy of discrimination in housing and lending,” said Michael Akinwumi, NFHA’s Chief Responsible AI Officer. “While accuracy-fairness tradeoff is often wrongly used as a defense against search for less discriminatory models in mortgage underwriting as expected by regulators, our study indicates that it is possible to reduce disparity in mortgage approval rate without undercutting accuracy and profitability. A real-world application of our findings could help eliminate lending bias, expand fair housing opportunities for all, especially for Black and Hispanic applicants, and reduce the racial homeownership gap which is currently wider than it was before the passing of the Fair Housing Act in 1968.”
Through a method known as Distribution Matching (DM), AI-based underwriting models can be constrained so that outcome distribution for any one protected group, like Black or Hispanic applicants, closely resembles the outcome distribution for the corresponding control group, such as White applicants, and improve the fairness of the underwriting model for protected groups without diminishing accuracy for all groups.
“The lending industry has had a long tradition of thinking that greater fairness and greater profits are mutually exclusive—which is not the case,” said Kareem Saleh, Founder & CEO of FairPlay AI. “Today’s research gives compelling evidence that, by training algorithms to regard disparities in lending outcomes as another form of error, mortgage approval rates for Black and Hispanic homebuyers can be increased by 5-13% with no corresponding increase in risk.”
NFHA conducted this study using about six million records representing about four million distinct loans, a dataset courtesy of CoreLogic®. The dataset contains records of loans originated from December 1987 through January 2021.
As a follow on to this release, NFHA and FairPlay AI will host a webinar tomorrow, April 25, 2024 from 3:00 pm – 4:30 pm ET to explore the study’s preliminary findings and present recommendations for future research to expand upon these positive findings.
Click here to read the study. Click here to register for the webinar.
The National Fair Housing Alliance (NFHA) is the country’s only national civil rights organization dedicated solely to eliminating all forms of housing and lending discrimination and ensuring equal opportunities for all people. As the trade association for over 170 fair housing and justice-centered organizations and individuals throughout the U.S. and its territories, NFHA works to dismantle longstanding barriers to equity and build diverse, inclusive, well-resourced communities.