Prediction of corporate failure with an expanded variable set, using multiple discriminant analysis



Journal Title

Journal ISSN

Volume Title



Analyzing the insolvency potential of a business is of interest to firm management, investors, lenders, and auditors. Models for prediction of corporate failure were initiated by Beaver in 1966. Other accounting researchers have expanded on his initial work since that time. This dissertation analyzes the inclusion of additional information in models of corporate failure. Failed and nonfailed firms are matched on the basis of size, industry, and financial statement dates. The dependent variable in the models is failure or nonfailure of the sample firms. The independent variables are separated into four groups. The first group includes changes in financial ratios, decomposition measures, and disaggregate decomposition measures. The second group is based on the first group as adjusted for industry averages. The third group of independent variables is a combination of all the variables of the first group and the fourth group is a combination of all the variables of the second group. All groups include a macroeconomic variable. Using multiple discriminant analysis, models for prediction of corporate failure are developed for each group of independent variables. The models are developed with and without the macroeconomic variable. The difference in overall predictive accuracy of the models is tested using the chi-square test for differences in probabilities. The accuracy of the models is also compared on the basis of Type I and Type II errors observed when the models are applied to the analysis sample and to the holdout sample of firms. The financial ratios of the nonfailed firms are tested for normal distribution. In general, the ratios are not normally distributed. Correlations between independent variables are found to be generally low. This result supports the choices for independent variables. Inclusion of the macroeconomic varaible does not improve the predictive accuracy of the models. In most models, the industry adjusted independent variables classify more accurately although the differences in classification accuracy are not statistically different. Inclusion of the disaggregate decomposition measure does not conclusively improve the predictive accuracy of the models.



Financial statements, Business failures