Loan Performance and Race

Robert E. Martin and R. Carter Hill address the issue of discrimination in lending to consumers in an article with the above title published in the respected economics journal Economic Inquiry. They point out that the statutes prohibiting discrimination do not distinguish between preference or taste-based discrimination vs. statistical discrimination based on differences in risk and administrative costs. That is, even if a well-defined but protected group (under ECOA) can be statistically identified as more likely to default, the law prohibits disparate treatment relative to other groups. By way of analogy, it is similar to a law prohibiting bettors at a racetrack from considering the past performance of the horses. Much of the literature claiming racial discrimination is based on differences in loan approval rates of blacks vs. whites. To continue the racetrack analogy, your betting at the track would be nondiscriminatory under the current law if you bet on 10 percent of the brown horses, 10 percent of the black horses and 10 percent of the gray horses.

The authors address a basic question: "Does a statistical basis for discrimination exist?" Even if it does exist, current laws prohibit such discrimination. A number of writers supporting the current statutes have based their tests for discrimination on approval rates of white borrowers vs. those in the protected class. Instead, these authors examine differences in risk and administrative costs among minorities and white consumers financing used cars. At present, lenders hoping to avoid litigation apply the same scoring system to all applicants regardless of their rate or national origin. Non-mortgage lenders typically do not gather data on the race or national origin from their applicants, also to avoid litigation. As a result, it is not possible to calculate loan approval rates by these classes, and statistical scoring systems do not include these prohibited variables. The basic question is as follows: Under these statistical scoring systems that do not include race or national origin as predicting variables, are the default rate and administrative costs of loans made to the protected classes higher or lower than among non-Hispanic whites?

To examine this question, the authors obtained a database of 120,000 loans on used cars from a large, multi-state lender specializing in C and D loans. This database was truncated to include only loans that could have "run their natural course" between January 1990 to September 1994. Used car loans were particularly appropriate, since credit decisions are based largely on the credit standing of the borrower and not on the value of movable collateral.

Since the lender did not collect data on the race or national origin of borrowers for the reasons given earlier, the authors designed a statistical model that would infer race based on other information in the file. The lender's data contained many other variables related to the borrower, such as age, monthly income, occupation code, and home ownership. The authors counted borrowers as "white" only if they lived in a zip code where at least 98 percent of the residents were white and as "nonwhite" only if at least 85 percent of those living in the zip code were nonwhite.

Detailed information on each loan in the final sample of about 4,100 was somewhat surprising. In comparison to white borrowers, "minority borrowers receive lower interest rates, borrow larger amounts . . . for longer periods of time, make lower downpayment . . . and have the same combined monthly income as whites." In addition, the researchers had data on each loan's performance, such as the borrower's failure to make regular monthly payments, defaults, records of bad checks, requests for extensions or outright default.

Given all of this information, the basic empirical question was whether race was associated with missed payments, defaults and other outcomes reflecting risk or leading to higher administrative costs. If the lender had been discriminating against qualified minorities (by setting a higher acceptance standard), their default rates and costs of administration should be lower than that for whites. However, the statistical analysis reveals that nonwhites were significantly (at the one percent level) more likely to default that whites. (There was no significant difference between whites and white Hispanics.) Note that the lender's turndown rate is irrelevant. It could have had these same results even while rejecting a higher proportion of nonwhite minority applicants than white applicants.

Their conclusion: "In general, minorities have higher default rates and contribute to higher administration costs. In other words, the foundation for statistical discrimination appears to exist."