| Summary: | 碩士 === 樹德科技大學 === 金融保險研究所 === 92 === Business distress has always been the focal point of financial analysis. Since 1998, most listed companies and business groups had a series event of financial distress. When a business, especially a stock listed company, is in its financial distress, the negative impact to the investors as well as the industries is serious. Therefore, it is necessary to build a financial distress prediction model to predict the potential financial distress in advance.
This dissertation first develops three neural networks：Back propagation network、Cascade-Correlation back propagation network、Elman network for the purpose of predicting financial distress. A holdout sample test enables us to show that the model have it validity. The data were collected from Taiwan stock listed companies that encountered financial distress between 1998 and 2003. Besides, We identity five different categories for the models.
Empirical results show that the models for all companies can achieve excellent accuracy for the trained samples. However, for the test samples, we find that the Back propagation network and Elman network have better accuracy. For the models of classified companies, we find the models of three networks have its merits. If the sample size is large enough, the correct prediction is more accurate. Further, for the prediction accuracy , cascade-correlation back propagation network is lower than the other two models. Therefore, we can ratiocination that for the study of financial distress prediction, using Back-Propagation Network methodology is applied to distressed firms and healthy firms. Results indicate that this technique can accurately predict distressed firms. Back-Propagation Neural Network really represents a new paradigms in the investigation of financial prediction and offers promising results.
|
|---|
