A comparison of statistical techniques and artificial neural network models in corporate bankruptcy prediction

A number of accounting research studies have focused on the use of statistical procedures in the development of bankruptcy classification and prediction models. Recent accounting and finance research has begun to apply algorithms from the discipline of artificial intelligence to a variety of prediction models. This study compares and evaluates the performance of logistic regression and nonparametric discriminant analysis and two artificial neural network models, backpropagation and counterpropagation, in predicting Chapter XI bankruptcy. Bankruptcy data drawn from a ten-year time horizon is input into each of the four models, with predictive ability tested at one, three, and five years prior to bankruptcy filing. Two probability levels,.5 and.8, were used to test accuracy, false positive and negative rates, sensitivity, and specificity. Results suggest that the logistic regression model and the backpropagation network model are generally evenly matched as prediction techniques. These findings were consistent with other applications of artificial neural networks to financial modeling. The results obtained through the use of the nonparametric technique and the counterpropagation network, both based on nearest-neighbor approaches, were generally disappointing. The predictive abilities of both of these methods were clearly surpassed by logit and backpropagation.