Alici, Yurt, "Neural Networks in Corporate Failure Prediction: The UK Experience", Working Paper, University of Exeter, (1995).

Abstract: This study makes a comparison between artificial neural networks (ANN) and the traditional statistical techniques of discriminant analysis (DA) and logistic regression (LR) in corporate failure modeling. The comparison was made at every step of the corporate failure prediction modeling process, using principal components analysis (PCA) and self-organizing feature maps (SOFM) in a variable reduction process, since a large number of financial ratios have been employed in financial risk measurement, when using DA and LR. As for using stepwise skeletonization backpropagation was employed in order to establish an optimum neural network structure. The main problem in ANN applications of corporate failure prediction has been the lack of understanding of the financial data and stochastic properties of financial ratios due to creative accounting practices, and of the different accounting policies and the diverse financial patterns of healthy and failed companies. In this research, every step employed in conventional financial failure studies was compared with the equivalent processes in the ANN field. The purpose is to establish a path for a fair comparison, and present ANNs as another tool in corporate failure prediction modeling.