Predicting corporate bankruptcy and financial distress: Information value added by multinomial logit models

Abstract

Efforts to explain and/or predict corporate bankruptcy continues to be of interest from finance, economics, and accounting perspectives. Corporate failure prediction models have generally progressed from univariate financial ratio analysis to multivariate models, and from discriminant models to logit model that offer an opportunity to estimate directly the probability of failure under less restrictive statistical assumptions. This study examines the added value of two types of information provided by multinomial logit models used to explain and predict corporate bankruptcy: (i) the information obtained by expanding the outcome space by including a third state of financial distress and (ii) secondary classification information. Samples of nonbankrupt, financially weak, and bankrupt firms are identified. Multinomial logit models, reflecting the reformulation of two traditional bankruptcy prediction models, then are used to classify firms during the 1970–1983 period. The status of each financially weak and bankrupt firm is monitored for five years after its initial classification. Significant reductions in misclassification error rates for the multinomial model are documented. Results also suggest that secondary classification information can be used to augment primary classifications to improve the ability to correctly predict bankrupt firms, as well as predict financially weak firms that will suffer severe financial distress in the future.