Comparing predictor methods; the case of bankruptcy predictions
Sood, Ashvin Raj
The aim of this study was to compare three methods of bankruptcy prediction, which were Multivariate Discriminant Analysis, Logit Analysis and Artificial Neural Networks. The methods were compared based on their ability to correctly classify and predict bankruptcy on American public listed companies one year prior to the year of failure. The window of interest in this study was the ten year period from 1984 to 1994. Eleven financial ratios, based on a study by Karels & Prakash (1987), were used in the three prediction methods. It was expected, based on previous literature, that the Artificial Neural Network would be the superior method. This study used a more robust testing technique, incorporating different proportions of bankrupt and financially stable companies as well as different sample mixes between the classification and holdout sample. Overall, it was Multivariate Discriminant Analysis which proved to be the best method of prediction in this study.
Degree Name: Bachelor of Commerce with Honours
Degree Discipline: Accountancy and Business Law
Keywords: Artificial Neural Networks; bankruptcy prediction; Multivariate Discriminant Analysis
Research Type: Dissertation