Algorithms Fail to Eliminate Consumer-lending Discrimination
Consumer discrimination continues in banking despite the advancement of algorithmic models used to automate the underwriting process, at least according to an academic study from Berkley.
The study analyzed nearly 9 million loan applications for mortgages between 2008 and 2015. All mortgages were for 30-year, fixed-rate, single-family homes, and all loan applicants had credit scores between 630 and 770.
Overall, the study confirmed previous studies’ findings that African Americans and Latinos pay a higher interest rate (5.6 basis points on average) for mortgages than the general population when obtaining a loan, even when the minority borrower has the same FICO score and loan-to-value ratio. In addition, the study discovered that among FinTech lenders who employ algorithmic credit scoring, a similar level of discrimination exists against African Americans and Latinos (5.3 basis points on average).
The higher interest rates charged to minorities groups can be extremely costly for the borrower. The extra fees add up to about $500 million a year in extra mortgage interest for African-American and Latino borrowers.
The finding appears counterintuitive since algorithmic models are generally thought of as a means to remove racial bias in consumer lending, yet it seems to have continued nonetheless. The study’s authors argue that algorithms are designed by people, and thus could have design flaws that produce biased results.
“The mode of lending discrimination has shifted from human bias to algorithmic bias,” said Adair Morse, one of the study’s authors. “Even if the people writing the algorithms intend to create a fair system, their programming is having a disparate impact on minority borrowers — in other words, discriminating under the law.”
The study is timely since more consumer lenders today are offering online options for potential borrowers. In fact, more than half of the largest mortgage providers offer complete online mortgage applications. It is important that lenders who offer online services design their algorithms carefully to both be in accordance with the law and offer fair loans to all credit-worthy people, regardless of race or gender.