Forecasting bankruptcy to promote organizational survival: A comparison of neural network and discriminant analysis methodologies

This research examined corporate continuation through the forecasting of business survival and failure. The purpose of this exploratory research was to determine the suitability and reliability of a bankruptcy forecasting model for alerting management to deteriorating corporate financial health, which in turn may reduce the social burden of bankruptcy.; A backpropagation artificial neural network (BANN) was evaluated as a tool to predict corporate failure and nonfailure. Fifty-four BANN configurations were tested with respect to efficiency, consistency, and accuracy for forecasting bankruptcy; the configurations were compared using a multiattribute decision model. The predictive accuracy of the BANN selected as superior was examined and the results compared to Altman's Z-score (i.e., discriminant analysis methodology), which is considered the benchmark for bankruptcy prediction.; One year prior to bankruptcy (failure) or nonbankruptcy (survival) the BANN's accuracy was 93.69% compared to 72.97% for Altman's Z-score, a 28.39% increase. For the 5 years prior to bankruptcy or nonbankruptcy the BANN classified 80.91% to Z-score's 62.05%. Overall, the neural network was 30.39% more accurate than the discriminant analysis formula.