A comparative study of different survival analysis models for bankruptcy prediction

• Main Author: Li, Ting
• Other Authors: School of Humanities and Social Sciences
• Format: Theses and Dissertations
• Published: 2008
• Subjects: DRNTU::Business::Law::Bankruptcy; DRNTU::Business::Finance
• Institution: Nanyang Technological University

Survival analysis is one of the most advanced techniques in bankruptcy prediction. However, to date, only few nonlinear techniques in survival analysis have been implemented in financial applications. This study introduces four nonlinear survival analysis, namely, partial logistic artificial neural networks (“PLANNs”) (Biganzoli et al., 1998), the Cox’s survival artificial neural networks (“Cox’s ANNs”) (Faraggi, 1995), the Weibull parametric survival artificial neural networks (“Weibull ANNs”) (Ripley, 1998) and the log-logistic parametric survival artificial neural networks (“log-logistic ANNs”) (Ripley, 1998) into bankruptcy prediction. Based on the data of about 1,000 US corporations in consumer goods/services industries, estimation and prediction results of linear regression and neural networks are presented. A comprehensive comparison among the outputs from different models is conducted. Relevant topics such as misclassification costs and the optimal structure of neural networks are also discussed. The results of this study show that survival artificial neural networks (“ANNs”) are superior to linear survival approaches in terms of prediction performance.