Bankruptcy Prediction in South Africa

Bankruptcy prediction modeling and studies are known to have existed since the 1960s. In this report a brief overview is given of the theory with reference to Altman's use of Multivariate Discrimination in the formulation of his Z-score model. This study looks at the impact of time in the classification or prediction accuracy. Two sets of 71 failed and non-failed companies listed in the Johannesburg Stock Exchange were used. The companies come from various sectors and range from the period of 1998 to 2007 excluding 2006. The data were subdivided into three parts according to periods and models were formulated and tested with five multivariate discriminate analysis functions and one classification tree algorithm. The results show that there is no impact in the classification or prediction accuracy due to the specific time period the model was used. It was further verified that the classification accuracy is independent of whether the model was chronologically closer or further than the test sample. The third test verified that the classification accuracy is independent of whether the time period is relatively narrow or wider. The overall result implies that the time period the model was created has no effect in the classification or prediction accuracy. This view is best when limited to a specific business cycle