How Much Does Racial Bias Affect Mortgage Lending? Evidence from Human and Algorithmic Credit Decisions

Abstract

We compare mortgage lenders’ credit decisions to algorithmic recommendations – on the same set of loan applications – from widely used Automated Underwriting Systems (AUS) to assess discrimination. In 2018-19, lenders were more likely to deny minority applicants than non-Hispanic white applicants. This paper is the first to document that “color-blind” AUS also recommend higher denial rates for minorities. Controlling for AUS recommendation, credit score, debt-to-income ratio, and loan-to-value ratio explains most of the racial and ethnic gaps in denials, although not the entirety. We show that lenders with the largest unexplained racial and ethnic denial gaps tend to also have the largest unexplained denials for non-Hispanic white applicants, suggesting that tight standards on unobservables might explain part of the remaining gaps. Additionally, our analysis of lenders’ reported denial reasons suggests that the remaining gaps could partially reflect differences by race and ethnicity in the successful completion of the final stages of loan approval (e.g. documentation of income). Overall, this evidence suggests a much more limited role for disparate treatment by lenders in the approval process than has been suggested in recent research.