Upstart says it’s improving AI models after report finds race approval disparities
The online consumer lender Upstart Holdings says it’s working to improve its artificial intelligence models after a law firm that monitors its fair-lending compliance found lower loan approval rates for Black applicants.
The law firm, Relman Colfax, said in a recent report that it found significant disparities early this year in how often Upstart approved loans to Black and non-Hispanic white borrowers. The disparities do not by themselves “demonstrate a fair lending violation” and are not unusual, the law firm wrote, but it recommended a “less discriminatory” model for Upstart to adopt.
The report points to shortcomings in the San Mateo, California-based fintech company’s loan decisioning model, which Upstart has said is more inclusive than the underwriting methods that banks have long used. Upstart uses a range of so-called alternative data to gauge borrower creditworthiness, including a person’s education and employment history, and often works with banks to make the loans.
Adobe Stock
In a written statement, the company said its model approves 43% more Black borrowers at lower interest rates than traditional credit scoring models. It also noted that the audit found there is no “pricing bias on our platform,” so borrowers pay the same interest rates no matter their background.
But Upstart also said that loan approval disparities for Black applicants is “an industry-wide challenge driven by decades of racial inequities, including credit scores.”
“We believe our model is one of the most fair and inclusive in the market, and we are committed to continuing to work with Relman to improve it to ensure we remain a leader in fair lending,” said Nat Hoopes, Upstart’s head of government and regulatory affairs. “We look forward to similarly transparent and rigorous testing being applied to all lenders — both old and new — so that all Americans can access inclusive and affordable credit.”
The report stems from an agreement Upstart reached with the NAACP Legal Defense and Educational Fund and the Student Borrower Protection Center in December 2020, when Upstart agreed to establish Relman Colfax as a monitor of its fair lending practices. The agreement followed a 2020 SBPC study that found higher borrowing costs for a sample graduate of Howard University, a historically Black institution, compared with a graduate of New York University.
In a statement Friday, SBPC Executive Director Mike Pierce said that “new evidence that Upstart’s model appears to discriminate against its Black customers is completely unacceptable and demands immediate action.”
On Monday, Pierce added: “Assuming they stick with it, Upstart’s commitment to addressing evidence of discrimination in its model is commendable. Others in the industry with similar disparities should take note of this process.”
Hoopes, the Upstart executive, said: “There was no finding of unfair discrimination in our model. I’m not aware of any company more committed to rigorous fair lending testing and building inclusive credit products than Upstart.”
Borrower advocates had hailed the agreement as a novel way of systemically tracking whether AI models may have racial biases that make it more expensive for Black customers to get credit. Relman Colfax’s report was its third on Upstart; an earlier report flagged similar disparities but had not yet recommended an alternative approach.
Matthew Bruckner, a Howard University law professor, credited Upstart for opening itself to scrutiny and said it’s likely “far from the only” company with such disparities.
Bloomberg News, for example, found that major mortgage lenders approved 87% of refinancing applications from white applicants in 2020 and just 71% from Black applicants. The disparities were particularly evident at Wells Fargo, which approved 72% of refinance applications from White applicants and just 47% from Black applicants. Wells Fargo, which has called the analysis “oversimplified,” has since been sued for those findings.
Howard University’s Bruckner called for “robust public oversight” of fair lending at banks and nonbanks alike. Such a commitment would include making large investments in data scientists who can conduct analyses for regulators such as the Consumer Financial Protection Bureau and state attorneys general, Bruckner said.
Bruckner also argued that the law firm’s findings show a potential shortcoming of so-called no-action letters from regulators such as the CFPB. In 2017, the consumer bureau reached a no-action arrangement with Upstart that let the company pursue innovation with less worry about a regulatory crackdown.
That arrangement ended in June, when Upstart requested an end to the no-action letter to “keep our risk models accurate and up-to-date during a period of significant economic change.”
The disparities that the law firm found occurred during the first quarter of 2022 and “on a set of data” from the third quarter of 2021, while Upstart remained subject to the no-action letter.
In a 2019 blog post about the no-action letter, CFPB officials wrote that they were testing whether the company’s models resulted in “greater disparities than the traditional model with respect to race, ethnicity, sex, or age.”
The agency wrote that its tests showed “no disparities that require further fair lending analysis under the compliance plan.”
Upstart is part of a new organization called MoreThanFair focused on using technology to make lending more inclusive, while ensuring that AI models are “properly supervised and rigorously tested,” according to its website. The group, which launched this month, includes other AI-based lenders such as Oportun Financial and LendingClub, as well as community groups such as the National Community Reinvestment Coalition, the National Consumer Law Center and the Latino civil rights group UnidosUS.
“Overcoming decades of inertia in access to credit is a huge challenge, so we’re fortunate to be joined in this effort by a diverse group of leading organizations,” Upstart CEO and co-founder Dave Girouard said in a press release announcing the group’s launch.
Comments are closed.