Credit risk, credit scoring, and the performance of home mortgages - includes related information

Federal Reserve Bulletin, July, 1996 by Robert B. Avery, Raphael W. Bostic, Paul S. Calem, Glenn B. Canner

Limitations of Scoring

Although credit scoring can reduce costs and bring more consistency to the underwriting process, its reliability depends upon the accuracy, completeness, and timeliness of the information used to generate the scores. For example, credit scores based on erroneous or seriously incomplete credit report information are not likely to accurately measure the risk posed by an individual applicant and may lead to unwarranted actions on an application (see box "How To Obtain Your Credit Report and What To Do To Correct Errors in the Report").

Also, concerns have been expressed that credit scores may not accurately gauge the creditworthiness of individuals whose experiences differ substantially from those on whom the index is based. If the baseline population used to generate the scoring index is not sufficiently diverse, then scores may lack predictive power for the underrepresented segments of the overall population. For example, rent, utility, and other nonstandard payment histories, which are often considered important for low-income populations, are frequently left out of scoring models. Thus, scores for these populations may not reliably assess individual risk.

Another set of concerns surrounds the use of credit scores more generally in the underwriting process. Lenders relying too heavily on scores might not give adequate consideration to special circumstances, such as a recent illness, that might mitigate a low score. Further, scores may lack predictive power if the underlying model used to generate the scores does not reflect current relationships between risk characteristics and measures of loan performance. Builders of credit scoring models report that model performance deteriorates over time. Thus, periodic validation may be necessary to ensure that scoring models retain their accuracy.

Credit scoring and its application to mortgage markets are evolving. Credit history scores, for example, traditionally have been based on the payment performance of a cross-section of consumers who have used credit, not all of whom have incurred mortgage debt. But consumer behavior with respect to mortgage debt may differ from behavior with respect to consumer debt. Consumers facing financial difficulties may, for instance, choose to pay their mortgage obligations first and postpone payments on other debts. For this reason, one might expect that a credit scoring model developed specifically for the mortgage market would provide more accurate predictions of future mortgage payment performance than a generic credit history score, even before the borrower has obtained a mortgage.

The development of models for credit history scores and application scores based on the payment performance of mortgage holders has historically been hampered by incomplete information about which consumers have mortgages and about other characteristics of these consumers. Also, many individual lenders have made too few mortgages to develop a sound mortgage credit scoring model. Recently, however, developers of scoring models have integrated information from several sources to develop both mortgage credit history scores and mortgage application scores.