- 學位論文
利用存活分析模型建立企業破產預測模式
A Bankruptcy Prediction Model of Using Survival Analysis
摘要
本研究以電子產業產為研究對象，以2000年至2010年中挑選38家破產之公司與76家正常公司，利用存活分析-Cox迴歸、Logistic迴歸和類神經網路建立台灣電子產業之破產預測模型，研究先以統計檢定篩選中適當的變數，並以全部輸入法和向前逐步迴歸法篩選出最具顯著性之變數。研究結果顯示，Cox迴歸模型都是最佳的破產預測模型，其預測準確率將近90%，且分類錯誤成本為三個模型中最小。另外，針對Cox迴歸模型進行分析，發現破產公司有以下之特色：流動比率、應收帳款周轉率和每股盈餘偏低，董監事持股質押比率偏高等情況。因此，若有上述之情況，管理者需有警覺在破產發生之前採取有效之改善措施，防止破產之發生。最後，建議後續的研究可利用不同的統計方法、資料探勘技術或資訊揭露指標等變數來建構破產預測模型。
並列摘要
Taiwan economic development has relied on the electronics industry for a long time. In 2008, the global financial tsunami caused by Taiwan's electronics industry showed negative growth. The study selects the 38 bankrupt companies and 76 normal companies from 2000 to 2010 year in electronics industry in Taiwan, and uses survival analysis-Cox model, logistic regression and neural network to differently construct bankruptcy prediction models. Firstly, select appropriate variables through the statistical test. Next, uses all enter method and forward stepwise regression method to filter out the most significant variables. The results show that financial ratios and corporate governance indicators are indeed helpful for bankruptcy prediction models. Cox model is the best bankruptcy prediction model, its accuracy of prediction is close to 90%, and it has the least classification error in the three models. In addition, the study also found that bankrupt companies have the following characteristics: Current ratio is low, accounts receivable turnover ratio is low, and earnings per share is low, but the directors' shareholding pledged rates is high. Finally, the follow-up studies can use different statistical methods and data mining technologies, and new variables to construct the bankruptcy prediction model.