Report finds racial bias in mortgage AI decisioning, MBA fires back

Even decades after the establishment of the Fair Housing and Equal Credit Opportunity Acts, loan denial rates skew heavily toward non-white borrowers, according to an investigation by The Markup.

Compared to white mortgage applicants with similar financial characteristics, Black consumers faced an average 80% higher likelihood of rejection, according to the non-profit’s analysis of 2019 Home Mortgage Disclosure Act data. Denial probability reached 70% higher for Native Americans, 50% for Asian or Pacific Islanders and 40% for Latinos.

The report analyzed 2 million home purchase applications of conventional mortgages across the country. The analysis did not include loans backed by the Federal Housing Administration or Department of Veterans Affairs — which have lower qualification standards and lower borrower profiles compared to conventional loans.

The Mortgage Bankers Association disputed the suggestion that the industry perpetuates racial discrimination. It said the report misrepresented the inequality issue and its findings didn’t provide a complete picture.

FICO scores weren’t used in the report since they are not made public. However, traditional credit scores can be inherently disadvantageous toward people of color since they historically haven’t had equitable access to banks or assets that build credit. Credit assessments haven’t accounted for rental payments, though Fannie Mae will begin using a positive rental payment history as an underwriting factor starting Sept. 18 (those who have an inconsistent record of on-time rental payments will not have that record counted against them, as such payments can be made in cash, which would not have an electronic record).

“[The Markup’s] analysis of HMDA data, and its predetermined conclusions regarding mortgage lending, fail to take into consideration several key components that form the backbone of lending decisions, including a borrower's credit score and credit history,” the MBA said in a press release. “As we told the authors, the Federal Reserve, the CFPB, and other regulators have been clear that denial disparities in the HMDA data alone cannot be used to assess fair lending.”

Today, much of the loan decisioning and underwriting process is automated. The algorithms driving the technology are intended to be “color blind” and non-inclusive of race or possible human bias. However, the issue could have less to do with the fairness of the algorithms themselves, and more to do with the foundational data they’re based on.

“This is how structural racism works,” Chi Chi Wu, a staff attorney at the National Consumer Law Center, said in the report. “This is how racism gets embedded into institutions and policies and practices with absolutely no animus at all.”

Credit history was the reason cited in mortgage denial for 36.9% of Native American applicants, 32.5% of Black, 21.3% of Latino, 20.9% white and 12.1% Asian according to the HMDA data.

Some vendors have sought to offer a more inclusive assessment for those who do not have an extensive credit history. VantageScore, a third-party credit score provider, estimated that under its model, 37 million borrowers who have no FICO score would become credit-worthy, and a third of that number would be either Black or Latino.

“They’ve been testing alternate scores for years, and I don’t know why the process is taking so long,” said Lisa Rice, president and CEO of the National Fair Housing Alliance, a consortium of hundreds of fair housing organizations. “Well-deserving consumers are being left behind.”