By Paul Gu, Co-Founder and Head of Product, Upstart
We were surprised to read this recent report from the Student Borrower Protection Center claiming that Upstart’s model is biased. We would like to clarify several inaccuracies and misunderstandings in this study.
First, we agree that it’s important to test for bias, and Upstart has worked closely with the Consumer Financial Protection Bureau to do this since our early days. As such, we conduct rigorous and quarterly fair-lending testing across millions of actual applicants. These large-scale tests consistently show that the addition of alternative data such as education, occupation, or employment reduces the bias in lending rather than increasing it.
As reported by the CFPB based on analysis by Upstart, Upstart’s AI model increases access to credit across all tested race, ethnicity, and gender segments by 23-29% while also decreasing average rates by 15-17%. Too many people are unfairly disadvantaged by traditional lending models primarily based on credit score and income, and solving that requires considering more data.
SBPC’s study was based on a single individual misrepresenting his education 26 times over a two and a half month period to request a loan rate from Upstart. SBPC’s claim that the loan applications were identical except for the applicant’s college attended is inaccurate and misleading. The applications had many differences among them, including changes to the applicant’s credit report. These differences affected the interest rate offered. Additionally, despite fabricating applications from 26 hypothetical consumers, the analyst selected the 3 results that best suited SBPC’s hypothesis. As a matter of fact, our model doesn’t consider individual schools but groups of schools that have similar economic outcomes and educational characteristics.
Anna, Dave and I founded Upstart to improve access to affordable credit. While well- intentioned, the SBPC analysis was misguided. While the traditional credit score and income-focused lending model particularly disadvantages African-Americans, Upstart’s model increases approval rates for graduates from schools such as Howard University by 46% with 18% lower APRs.
We were not contacted by this organization, but would welcome the opportunity to discuss their research. Ultimately we share the same goals – improving access to credit for the underserved.