A BEHAVIORAL MODEL OF RISK IN CONSUMER CREDIT*

Abstract

THE PROBLEM of determining the risk involved in extending credit to a potential borrower has been with the businessman since the first credit transaction. Within the last fifteen years a concerted, if somewhat inconsistent, effort has been made to quantify risk through the use of financial and demographic variables-the credit scoring model of risk analysis. Since the 1920's writers have suggested the importance of behavioral data in determining credit risk. However, no general functional model has been proposed or tested. This dissertation bridges the gap existing in credit theory. The objectives of this dissertation are divided into four parts: (1) proposal of a generally applicable environmentally induced behavior model, (2) testing the model for validity, (3) comparison of the behavioral model with a financialdemographic model, and (4) testing the assumptions of discriminant analysis to see if they were satisfied by the data. A theoretical behavior model of credit risk was proposed as the basis for the empirical model. An individual's risk is given as being a direct function of the individual's concept of debt responsibility which is a function of the individual's credit character which is a function of environmental forces and his ability to pay which is a function of non-behavioral environmental forces. Two groups of subjects were used in the study. Group I consisted of good credit risks as determined by a credit check and their past credit payment record. Group II consisted of individuals who had a history of overextending their credit purchases and were considered poor risks by most rating criteria. Group II subjects were delinquent or in default on one to thirty credit purchases each. Most Group II subjects were near personal bankruptcy. Behavioral data were collected on thirty-six parameters using a selfadministered response form questionnaire with a summated ratings scale. Financial and demographic data were also obtained from each of the two hundred respondents. Using a stepwise discriminant analysis program on a split sample, several equations were produced for each type of data. The behavioral equation with the greatest classificatory accuracy (96 per cent) required only eight variables. The most accurate financial equation required thirteen variables and classified 94 per cent of the validation sample correctly. A test of the normality and the equal variance-covariance assumptions of discriminant analysis were tested with both sets of data generally violating the variance equality assumption. Also, although both sets of data were skewed to various degrees, the evidence of non-normality was not conclusive. All other discriminant analysis assumptions were corrected through allowable adjustments in the computer program's methodology. Past researchers and authors have demonstrated the need for a theoretical foundation for risk models. This dissertation proposes such a model and presents