Utility Analysis of Five Weighting Methods for Making Consumer Loan Decisions

The economic utility of five weighting methods (unit, weighted application blank, chi-square, Bayes, and regression) for evaluating consumer loan applications was determined. A sample of 443 consumer loans which had been classified as either good or bad accounts was analyzed via 11 predictor variables. The predictor variables were weighted by each of the five weighting methods. Cross-validated correlations revealed that consumer credit risk is highly predictable. The five methods differed in the extent to which they produced false positive and false negative selection decisions. Utility assessments revealed that the best methods were the unit and weighted application blank procedures. The findings were discussed in terms of the assessment of credit risk and issues associated with the implementation of decision models by real world decision makers.