An evaluation of decision tree and survival analysis techniques for business failure prediction

Adrian Gepp, Bond University

Abstract

Accurate business failure prediction models would be extremely valuable to many industry sectors, particularly in financial investment and lending. The potential value of such models has been recently emphasised by the extremely costly failure of high profile businesses in both Australia and overseas, such as HIH (Australia) and Enron (USA). Consequently, there has been a significant increase in interest in business failure prediction, from both industry and academia.

Statistical models attempt to predict the failure or success of a business based on publicly available information about that business (or its industry and the overall economy), such as accounting ratios from financial statements. Discriminant and logit analyses have been the most popular approaches, but there are also a large number of alternative techniques available. This thesis reviews the various techniques used in previous studies, and presents the results obtained from using two alternative techniques that have shown promise, namely survival analysis and decision trees. These techniques have been compared with the more popular techniques and to each other. Overall, the results suggest that decision trees provide the most accurate predictions of business failure. In addition, while survival analysis techniques are slightly less accurate than other techniques they provide more information that can be used to further the understanding of the business failure process.