Using Data Mining Techniques to build the Financial Distress Prediction Model

• Main Authors: Weng Shao-Yuan, 翁韶遠
• Other Authors: 薛榮棠
• Format: Others
• Language: zh-TW
• Published: 2009
• Online Access: http://ndltd.ncl.edu.tw/handle/sxyn49

碩士 === 元培科技大學 === 企業管理研究所 === 97 === Financial distress of companies makes the enterprises which invest these companies, financial companies and investors suffer great losses as well as a huge amount of payments for social cost. In order to prevent to invest the companies which have financial distress by the sponsors of investment enterprises, financial companies and investors, a set of financial distress alarm model is essential. Before the companies confronted the financial distress, the financial ratio and non-financial information usual divulged some connected omens. Therefore, this study integrates enterprise financial ratio index and non-financial information index to build an enterprise financial crises alarm model. Research samples are taken from 178 companies which have financial distress, and 178 normal companies are selected, total 306 samples are used to build a model. Regarding to our data gathering and former researchers’ experiences, 27 financial indicators and 5 non-financial indicators for the input data are utilized. This study uses the Group method of Data Handling and General Regression Neural Network to construct the financial distress prediction model and compares with the Logistic Regression Analysis and Discriminant Analysis. The result illustrates that General Regression Neural Network are better prediction accuracy in the former one and two and three year of financial crisis. The important variables of judging financial crisis are director holding, directorate holding, pledged shares ratio of directors, valuation of accounting, PS-Cashflow, Return On Assets, Return on Equity, Gross Margin, and BPS. Final, this study confirm that the samples classify according the common stocks can increase the forecast strength.