Financial Distress Prediction Using Neural Networks

This exploratory research examines and models the financial distress prediction using neural network approach. The study is based on financial ratios. Nine different neural network models are constructed to test the predictive capability of the models by considering: (1) the impact of time varying information structure prior the distressed situation using first, independent annual financial ratios (four models)and second, different panel data sets (three models) and, (2) the influence of time varying probability estimates of financial distress in panel data sets (two models). Results support that it is not necessary to have complex architecture in neural models to predict firm's financial distress. Besides more the predictability horizon is shorter and the input information structure is most recent, more the predictive capability of the neural model is better