A COMPARISON OF SIX MODELS FOR PREDICTING CORPORATE BANKRUPTCY: MULTIPLE LINEAR REGRESSION ANALYSIS, MULTIPLE LINEAR DISCRIMINANT ANALYSIS, STEPWISE REGRESSION ANALYSIS, STEPWISE DISCRIMINANT ANALYSIS, MULTIPLE LINEAR REGRESSION ANALYSIS WITH RIDGE REGRE

The purpose of this research was to examine the estimation of coefficients using six corporate bankruptcy prediction models based upon the following statistical analyses: (1) multiple linear regression analysis; (2) multiple linear discriminant analysis; (3) stepwise regression analysis; (4) stepwise discriminant analysis; (5) multiple linear regression analysis with ridge regression; (6) multiple linear discriminant analysis with biased minimum (chi)('2) rule. As part of the examination, accuracy in predicting corporate bankruptcy by the models and the relative importance of financial ratios were investigated. In short, the following three research questions were examined: (1) coefficient comparisons; (2) ranking of financial ratios; (3) corporate bankruptcy prediction. Specifically, the study focused on a shortcoming of previous research. The shortcoming referred to as multicollinearity is a problem with data behavior.;Often the financial ratios which are used as independent variables in bankruptcy, prediction models are highly correlated (multicollinearity) with each other. Thus, the coefficient estimates, in many of these models, tend to be very unstable and not useful for future predictive purposes on a new sample. The study presents a useful procedure (ridge regression) for "taming" the coefficient estimates.;Based on the research reported it is suggested that models incorporating ridge regression have the following advantages: (1) smaller coefficients; (2) smaller standard errors of these coefficients; (3) yield more reliable information about the effects of individual ratios on corporate bankruptcy; (4) appears to have the edge in predicting on future samples.