Forecasting with neural networks. An application using bankruptcy data

In the business environment, Least-Squares estimation has long been the principle statistical method for forecasting a variable from available data with the logit regression model emerging as the principle methodology where the dependent variable is binary. Due to rapid hardware and software innovations, neural networks can now improve over the usual logit prediction model and provide a robust and less computationally demanding alternative to nonlinear regression methods. In this research, a back-propagation neural network methodology has been applied to a sample of bankrupt and non-bankrupt firms. Results indicate that this technique more accurately predicts bankruptcy than the logit model. The methodology represents a new paradigm in the investigation of causal relationships in data and offers promising results.