A Comparative Study of Financial Distress Warning Model－Logit Analysis, Discriminant Analysis, and Cox Survivals Analysis

• Main Authors: Sheng-Shing Lin, 林順興
• Other Authors: Kuei-Yen Wu
• Format: Others
• Language: zh-TW
• Published: 2003
• Online Access: http://ndltd.ncl.edu.tw/handle/92739069142173485525

碩士 === 輔仁大學 === 管理學研究所 === 91 === If business enterprises occur financial distress, not only the investors but also society will encounter the loss. Hence, the study is tries to establish a financial distress warning model to detect financial crisis in advance. This research basically adopts the Cox survivals analysis model suggested by Cox and Oakes in 1984, logit analysis model, and discriminant analysis model to develop three forecasting models respectively. Besides, we also compare efficiencies of these three models. We collect all full-delivery, temporary-suspend and stopping-suspension stocks in Taiwan Stock Exchange 1980 to September in 2002. By including financial variables, non-financial variablies respectively and both above variablies to test the time period prior to the crisis. To test the models’ efficiency, we use three statistical sampling methods: 1.time 2.Split-sample 3.Cross-validation. The empirical result of this study can be summarized as follows: 1.Comparing financial and non-financial variablies, non-financial variablies is more efficient to detect the crisis when the model is established by using the data when two years before the crisis occurred.. 2.Logit analysis has the best predictive abilitity in the in-smaple analysis. Cox survivals analysis model has better predictive ability at the time period when 2 and 3 year before the crisis occurred. 3.Logit analysis and discriminant analysis models will increasingly declined the predictive ability in out-sample analysis. The predictive ability of Cox survivals analysis models is better than those of logit analysis and discriminant analysis models in out-sample. 4.In summary, the predictive ability of Cox survivals analysis models is better than those of logit analysis and discriminant analysis models in in-sample and out-sample analysis.