Discriminant analysis, factor analysis and linear regression analysis to classify financially distressed firms and predict bankruptcy using financial ratios and macroeconomic predictors: Model application to selected M and A

This thesis reports results provided by Factor Analysis (FA), Discriminant Analysis (DA) and Linear Regression Analysis (LRA) to differentiate between financially distressed bankrupt firms and stable nonbankrupt firms in the manufacturing industry for the period of 1995 to 2003. Additional models were developed to classify high financial distress firms that filed for bankruptcy versus high financial distress firms that had a successful turnaround. Combination of these models provides a successful way to classify companies as solvent, financially distressed or bankrupt. Models developed were tested for statistical significance and classification accuracy. LRA is used to establish correlations between financial ratios of selected models and economic indexes. Correlations are used to reevaluate two of the weaker models. This study examines degree of financial distress, data stationarity, sample homogeneity, and effect of economic index correlation upon classification accuracy. Stationarity of the population is established in order for the model to be used as a predictive tool. A comparison was made between FA coupled with simultaneous DA and stepwise DA in classifying bankrupt and nonbankrupt firms. The latter technique achieves higher classification results, including models with 100% accurate classification. Improvement in classification accuracy was achieved by manual selection of stepwise DA-determined variables and by the introduction of macroeconomic indicators. Criteria for building discriminant models include the use of financial ratios from the year before filing for bankruptcy protection as well as from two years before, an average of the two, and a differential trend for up to two years before bankruptcy. Selected models are applied to top manufacturing companies M&A of the years 1997 and 1998 using data prior to the M&A event as well as data from five years later.