Tyree, Eric K. and J. A. Long, "Assessing Financial Distress with Probabilistic Neural Networks", Working Paper, City University of London, (1994).
Abstract: One of the more common techniques employed in forecasting financial distress involves the use of linear discriminant analysis (LDA). Using various combinations of financial ratios LDA techniques have produced results on the order of 90% - 95% accuracy. A problem with LDA, however, is that it requires that the groups to be classified possess multivariate normality and equal dispersion matrices. The lack of these properties can have severe consequences on the performance of the LDA model. The following report presents and analyses the results of a Probabilistic Neural Network (PNN), also known as a Parzen-Bayes classifier (PBC), for the prediction of financial distress. The prime advantage of the PNN algorithm is that its ability as a statistical classifier does not suffer from departures from multivariate normality or equal dispersion matrices in the groups to be classified. The PNN algorithm utilized in this study not only produced significantly more accurate results than linear discriminant analysis but was also found to provide an interpretable solvency rating for individual companies.