The Usefulness of Neural-Network and Neural-Fuzzy Techniques in the Prediction of Firm Failures

• Main Authors: Yu-Chun Chen, 陳昱均
• Other Authors: Ming-Hsiang Huang
• Format: Others
• Language: zh-TW
• Published: 2003
• Online Access: http://ndltd.ncl.edu.tw/handle/nt5wj4

碩士 === 逢甲大學 === 財務金融學所 === 91 === Abstract Numerous studies have documented that survival analysis provides extra important information─time to failure in predicting firm failures, but encounter a relative weakness in the accuracy of prediction. To accommodate such shortcoming of Survival analysis, this study incorporates survival model into two artificial intelligent frameworks including both Neural-Fuzzy and Neural-Network. A sample of all Taiwan Stock Exchange listed firms having operation difficulties over the period between 1991 and 2001 is used as hazard firms. To investigate the usefulness of underlying model, an extensive comparison of weighted efficiency (W. E.) between the following models have been performed: the survival analysis, Logit model, Neural-Fuzzy survival analysis, Neural-Fuzzy Logit, Neural-Fuzzy, Neural-Network, Neural-Network Survival and Neural-Network Logit. Our empirical result suggests that an incorporation of artificial intelligent framework do improve the accuracy and W.E. of both survival analysis and Logit approach. Moreover, an incorporation of Neural-Network contributes the most significant improvement to the Survival analysis. Unlike the norm that Logit outperform survival analysis, our evidence suggests that accuracy of the Neural-Network Survival is as good as which of Neural-Network and Neural-Network Logit. Thus, the result implies that after inclusion of Neural-Network, Survival analysis will dominate Logit technique in the sense that Survival provide same accuracy but present an extra important information on time to failure.