Trait Recognition: An Alternative Approach to Early Warning Systems in Commercial Banking

The present study reports the empirical results of trait recognition (TR) as an alternative early warning system for identifying failing US commercial banks. TR has previously been employed in the sciences, and unlike previous statistical and nonparametric models, incorporates a large number of interaction variables based on the independent variables taken two and three at a time. Discriminatory original and interaction variables (or traits) are selectively retained for use in classifying observations based on a voting procedure. Comparative results for failed and nonfailed US commercial banks using Call Report data indicate that the TR model generally outperformed logit regression models, in some cases by a considerable margin. A major implication of these results is that TR could be useful in other binary choice problems in business finance and accounting, including predictions of nonbank failures, bond rating changes, and other firm events.