The Secret Bias Hidden in Mortgage-Approval Algorithms

{snip}

An investigation by The Markup has found that lenders in 2019 were more likely to deny home loans to people of color than to white people with similar financial characteristics — even when we controlled for newly available financial factors the mortgage industry for years has said would explain racial disparities in lending.

Holding 17 different factors steady in a complex statistical analysis of more than 2 million conventional mortgage applications for home purchases, we found that lenders were 40% more likely to turn down Latino applicants for loans, 50% more likely to deny Asian/Pacific Islander applicants, and 70% more likely to deny Native American applicants than similar white applicants. Lenders were 80% more likely to reject Black applicants than similar white applicants. These are national rates.

{snip}

In every case, the prospective borrowers of color looked almost exactly the same on paper as the white applicants, except for their race.

The industry had criticized previous similar analyses for not including financial factors they said would explain disparities in lending rates but were not public at the time: debts as a percentage of income, how much of the property’s assessed worth the person is asking to borrow, and the applicant’s credit score.

The first two are now public in the Home Mortgage Disclosure Act data. Including these financial data points in our analysis not only failed to eliminate racial disparities in loan denials, it highlighted new, devastating ones.

We found that lenders gave fewer loans to Black applicants than white applicants even when their incomes were high — $100,000 a year or more — and had the same debt ratios. In fact, high-earning Black applicants with less debt were rejected more often than high-earning white applicants who have more debt.

{snip}

We sent our complete analysis to industry representatives: The American Bankers Association, The Mortgage Bankers Association, The Community Home Lenders Association, and The Credit Union National Association. They all criticized it generally, saying the public data is not complete enough to draw conclusions, but did not point to any flaws in our computations.

{snip}

In written statements, the ABA and MBA criticized The Markup’s analysis for not including credit scores and for focusing on conventional loans only and not including government loans, such as those guaranteed by the Federal Housing Administration and Department of Veterans Affairs.

Isolating conventional loans from government loans is common in mortgage research because they are different products, with different thresholds for approval and loan terms. Government loans bring people who wouldn’t otherwise qualify into the market but tend to be more expensive for the borrower.

Even the Federal Reserve and Consumer Financial Protection Bureau, the agency that releases mortgage data, separate conventional and FHA loans in their research on lending disparities. Authors of one academic study out of Northeastern and George Washington universities said they focus on conventional loans only because FHA loans have “long been implemented in a manner that promotes segregation.”

As for credit scores, it was impossible for us to include them in our analysis because the CFPB strips them from public view from HMDA data — in part due to the mortgage industry’s lobbying to remove them, citing borrower privacy.

When the CFPB first proposed expanding mortgage data collection to include the very data that industry trade groups have told us is vital for doing this type of analysis — credit scores, debt-to-income ratio, and loan-to-value ratio — those same groups objected. They didn’t want the government to even collect the data, let alone make it public. They cited the risk of a cyberattack, which could reveal borrowers’ private information.

{snip}

In addition to finding disparities in loan denials nationally, we examined cities and towns across the country individually and found disparities in 89 metropolitan areas spanning every region of the country.{snip}

Black applicants in Chicago were 150% more likely to be denied by financial institutions than similar white applicants there. Lenders were more than 200% more likely to reject Latino applicants than white applicants in Waco, Texas, and to reject Asian and Pacific Islander applicants than white ones in Port St. Lucie, Florida. And Native American applicants in Minneapolis were 100% more likely to be denied by financial institutions than similar white applicants there.

{snip}

“Redlining,” the now-outlawed practice of branding certain Black and immigrant neighborhoods too risky for financial investments that began in the 1930s, can be traced back to Chicago. Chicago activists exposed that banks were still redlining in the 1970s, leading to the establishment of the Home Mortgage Disclosure Act, the law mandating the collection of data used for this story.

{snip}

Who makes these loan decisions? Officially, lending officers at each institution. In reality, software, most of it mandated by a pair of quasi-governmental agencies.

{snip}

Freddie Mac and Fannie Mae were founded by the federal government to spur homeownership and now buy about half of all mortgages in America. If they don’t approve a loan, the lenders are on their own if the borrower skips out.

And that power means Fannie and Freddie essentially set the rules for the industry, starting from the very beginning of the mortgage-approval process.

Fannie and Freddie require lenders to use a particular credit scoring algorithm, “Classic FICO,” to determine whether an applicant meets the minimum threshold necessary to even be considered for a conventional mortgage, currently a score of 620.

This algorithm was developed from data from the 1990s and is more than 15 years old. It’s widely considered detrimental to people of color because it rewards traditional credit, to which white Americans have more access. It does not consider, among other things, on-time payments for rent, utilities, and cellphone bills — but will lower people’s scores if they get behind on them and are sent to debt collectors. Unlike more recent models, it penalizes people for past medical debt even if it’s since been paid.

{snip}

Potentially fairer credit models have existed for years. A recent study by Vantage Score — a credit model developed by the “Big Three” credit bureaus to compete with FICO — estimated that its model would provide credit to 37 million Americans who have no scores under FICO models. Almost a third of them would be Black or Latino.

Yet Fannie and Freddie have resisted a steady stream of plaintive requests since 2014 from advocates, the mortgage and housing industries, and Congress to update to a newer model. Even the company that created Classic FICO has lobbied for the agencies to adopt a newer version, which it said expands credit to more people.

{snip}

Fannie’s and Freddie’s regulator and conservator, the Federal Housing Finance Agency, continues to allow the companies to stick with Classic FICO, more than five years after ordering them to study the effects of switching to something newer. The FHFA has also expressed concern about the “cost and operational implications” if they would have to continually test new credit scoring models.

{snip}

Fannie’s and Freddie’s approval process also involves other mysterious algorithms: automated underwriting software programs that they first launched in 1995 to much fanfare about their speed, ease and, most important, fairness.

“Using a data base as opposed to human judgment can avoid influences by other forces, such as discrimination against minority individuals and red-lining,” Peter Maselli, then a vice president of Freddie Mac, told The New York Times when it launched its software, now called Loan Product Advisor. A bank executive told Congress that year the new systems were “explicitly and implicitly ‘color blind,’” since they did not consider a person’s race at all in their evaluations.

But, like similar promises that algorithms would make colorblind decisions in criminal risk assessment and health care, research shows that some of the factors Fannie and Freddie say their software programs consider affect people differently depending on their race or ethnicity. These include, in addition to credit histories, the prospective borrowers’ assets, employment status, debts, and the size of the loan relative to the value of the property they’re hoping to buy.

{snip}

Research has shown that payday loan sellers usually place branches in neighborhoods populated mainly by people of color, where bank branches are less common. As a result, residents are more likely to use these predatory services to borrow money. This creates lopsided, incomplete credit histories because banks report both good and bad financial behavior to credit bureaus, while payday loan services only report missed payments.

Gig workers who are people of color are more likely to report that those jobs are their primary source of income — rather than a side hustle they’re using for extra cash — than white gig workers. Having multiple sources of income or unconventional employment can complicate the verification process for a mortgage, as Crystal Marie and Eskias McDaniels learned.

Considering an applicant’s assets beyond the down payment, which lenders call “reserves,” can cause particular problems for people of color. People with fatter bank accounts present a lower risk because they can more easily weather a setback that would leave others unable to pay the mortgage. But, largely due to intergenerational wealth and past racist policies, the typical white family in America today has eight times the wealth of a typical Black family and five times the wealth of a Latino family. People of color are more likely to have smaller savings accounts and smaller (or nonexistent) stock portfolios than white people.

{snip}

In written statements, Fannie said its software analyzes applications “without regard to race,” and both Fannie and Freddie said their algorithms are routinely evaluated for compliance with fair lending laws, internally and by the FHFA and the Department of Housing and Urban Development. HUD said in an email to The Markup that it has asked the pair to make changes in underwriting criteria as a result of those reviews but would not disclose the details.

{snip}

The Markup’s analysis does not include decisions made by Fannie’s and Freddie’s underwriting algorithms because, while lenders are required to report those decisions to the government, the CFPB scrubs them from public mortgage data, arguing that including them “would likely disclose information about the applicant or borrower that is not otherwise public and may be harmful or sensitive.” Lenders’ ultimate mortgage decisions are public, however. Borrowers’ names are not reported to the government and addresses are not in the public data.

{snip}

In addition to using Fannie’s or Freddie’s software, many large lenders also run applicants through their institutions’ own underwriting software, which may be more stringent. How those programs work is even more of a mystery; they are also proprietary.

When we examined the reasons lenders listed for denying mortgages in 2019, the most common reason across races and ethnicities, with the exception of Native Americans, was that applicants had too much debt relative to their incomes. When lenders did list “credit history” as the reason for denial, it was cited more often for Black applicants than white ones in 2019: 33% versus 21%.

{snip}