A description and analysis of the corporate failure process using an events-based logit model

Abstract

Prior bankruptcy studies have primarily used financial ratio based models to discriminate between matched pairs of bankrupt and non-bankrupt firms. Typically, the focus is on the year of bankruptcy, and the years(s) preceding bankruptcy is (are) used as the estimation period. It can be argued that the results of these prior studies are biased because the matched paired design resulted in the oversampling of bankrupt firms. In addition, the exclusive use of financial ratios may neglect significant events that contain useful information regarding an entity's continued existence. The events-based model proposed herein seeks to overcome these limitations through the use of qualitative independent variables, the method of sample selection, and the particular period of examination. The qualitative variables used included reduction in dividends and bond ratings, change in accounting policies, change in auditors, qualified opinions, troubled debt restructurings, and violations of debt covenants. In addition, a matched pair design was not used. Instead, all firms experiencing financial distress during the period 1975-1981 were sampled and examined over a five year period. There were 669 firms in the total sample--53 bankrupt (8%) and 616 non-bankrupt (92%). Logistic regression analysis was used to test models containing event data as well as traditional financial ratios as independent variables. Overall, the results indicate that certain events, particularly reduction in dividend payments and going concern qualifications, can be used to discriminate between bankrupt and non-bankrupt financially distressed firms. The results are significant despite: (1) the prior probabilities of the bankrupt and non-bankrupt samples not being equal; and (2) the use of financially distressed firms in both the bankrupt and non-bankrupt samples.