Do credit scores have a disparate impact on racial minorities?
I am troubled when I read allegations in the press that FICO® Scores discriminate against people of color. That’s because a credit score is nothing more than the output of a mathematical formula built to rank-order the likelihood that a person will repay the debts they have incurred. To say that FICO Scores are unfair suggests that the formula was built with racial bias. To say that the use of a FICO Score is unfair suggests that it would be fair for a creditor to use the score to assess the credit risk presented by most borrowers, but not if the borrower is a member of a racial minority.
The first suggestion, that the formula was built with racial bias, is untrue. Building a score that’s both “fair” and predictive is certainly a priority for us at FICO. As part of this, FICO® Scores have never used race, gender, marital status or other legally prohibited bases as inputs.
As to whether the use of credit scores has an unfair impact on racial minorities, researchers at the Federal Reserve Board (FRB) have studied the relationship between credit scores and race to see if empirically derived credit scores produce a disparate impact on specific populations. Their finding: credit scoring does not produce a disparate impact.
As part of that study, the FRB team compared credit scores from two types of models. One type produced a more traditional credit score and was developed on a national consumer population. The other type was a series of “race-neutral” models. Specifically, these models were developed for individual demographic populations (e.g., Hispanic, Black, Asian). Disparate impact would be apparent if a minority group scored higher on the race-neutral scorecard than on the baseline credit score. In other words, for each scoring model developed for a specific minority group, evidence of disparate impact would be found if this model yielded higher scores for the group than did the benchmark scoring model.
On page 26 of their 2010 study, the FRB’s Avery, Brevoort and Canner wrote:
“Our results provide little or no evidence that the credit characteristics used in credit history scoring models operate as proxies for race and ethnicity. The distributions of credit scores for different racial or ethnic groups or across genders are essentially unaffected by the reestimation or redevelopment of the baseline credit scoring model in any of the race- or gender-neutral environments. This suggests that credit scores do not have a disparate impact across race, ethnicity, or gender.” (Bold is mine.)
FICO will continue to build credit scores that are objective measures of risk, while at the same time working to demystify credit scores for consumers. We want to empower everyone, regardless of race, to manage their credit scores well.