A discriminant analysis of Farmers Home Administration housing loans

The objective of this study was the estimation of discriminant functions for the Section 502 Housing Loan Program of the Farmers Home Administration in Fentress, Grundy, and Haywood counties, Tennessee. The hypothesis of the study was that good and problem borrowers were distinguishable from one another based upon information available on the application form. Two samples were collected from the loan portfolios of each county. One sample represented problem borrowers and the other good borrowers. Data from each application sampled included the socio economic characteristics of the borrower. The samples were analyzed in three steps: first, in a tabular analysis; second, in stepwise regressions; and third, in discriminant functions.

The tabular analysis revealed that within each county several of the borrower characteristics were related to the loan classification but that the relationships were not consistent among the counties. The stepwise regressions were used to further study the variables. The most predictive variables for each county were then used in the three discriminant models. The models differed reflecting the local employment, creditor, and household types, as well as the financial measures which could be best identified with loan classification in each county.

The discriminant models for Fentress, Grundy, and Haywood counties were tested, found significant, and correctly classified 81, 79, and 81% of the respective loan application samples.