探討電子業及傳統產業於破產預測差異之研究

A Research of Bankruptcy Prediction Comparison between Electronic and Traditional Industries

朱玉婷(Yu-Ting Chu)

指導教授 ： 李維平

中原大學/商學院/資訊管理研究所/碩士(2006年)

https://doi.org/10.6840/cycu200600405

摘要

企業財務狀況一直以來是社會大眾所關注的課題，而企業財務危機更是攸關企業生存與否最重要的關鍵點，因此若能及早預測出企業財務危機將能減少對企業甚至社會大眾的損失，故企業財務危機預警模式逐漸發展起來。從文獻研究中可以知道有相當多的財務危機預警系統，經由研究發現資料探勘所建構的模型優於傳統統計模型，其中過去廣受歡迎且用的較好的即是決策樹與類神經網路模型，而近年來興起的支援向量機也用於財務危機模型的建構，成效上具有不錯的表現，另外文獻研究也皆著重於混合所有產業別進行模式的建構，少有針對個別特定產業建構模式。基於以上問題，本研究主要以電子產業與傳統產業為研究樣本，運用決策樹、類神經網路及支援向量機方法個別建構電子產業與傳統產業預警模型，然而，類神經網路在學習上的收斂表現，常會受到是否有適當的輸入變數所影響，故本研究也採用了過去學者所曾使用的輸入變數篩選方法，即因素分析法、逐步迴歸分析法及敏感度分析法，再結合類神經網路建構模型。本研究欲透過所建構的不同模型，藉以瞭解電子產業與傳統產業在破產預測的差異。實驗結果顯示：
1.電子產業及傳統產業皆適合運用敏感度分析結合類神經網路模式，在電子產業於危機前一年可達82.5%的準確率，傳統產業於危機前一年更可以達到96.08%。
2.與過去學者針對傳統產業建構的預警模式相比，本研究以敏感度分析結合類神經網路建構的模式能優於過去學者的研究，並觀察發現加入非財務比率變數對預測效果具有重要影響，另針對電子產業建構的預警模式相比，準確率也有在平均水準以上。
3.本研究發現支援向量機並非是最好的，而且支援向量機的預測準確率掌控於參數的決定，也不容易找到最佳參數，然而對於敏感度分析結合類神經網路卻沒有這方面的問題。
4.透過敏感度分析篩選變數後，可發現電子產業最具影響的指標是「營業外收入」、「營業費用率」與「存貨週轉率」，傳統產業最具影響的指標則是「負債比率」、「總資產成長率」與「董監事質押比率」，由此可知，電子產業受到了獲利能力、成本費用與經營能力構面之影響，傳統產業則受到了償債能力、成長率與公司治理構面之影響。

關鍵字

敏感度分析法 ； 支援向量機 ； 決策樹 ； 逐步迴歸分析法 ； 類神經網路 ； 因素分析法 ； 財務危機

並列摘要

Enterprise's financial situation has not only been considered as a critical issue to the enterprise and the public, but it also is an key point that affect the survival of each enterprise. Early and correct bankruptcy prediction could effectively reduce the lost of the enterprises and the correspondent investors, therefore the early-warning mode of the enterprise finance crisis was developed gradually. According to the literature review, it was known that there were plenty of the early-warning systems of enterprise finance crisis, but it prospect the model built and constructed and surpasses the traditional statistics model via the data mining, decision tree and neural network models were also well applied in the past among them, support vector machine which has good behavior on the effect rise in recent years and was used to build and construct the financial crisis model, literature review only emphasized the combination of different industries and tailor make the crisis model for them instead of individual and specific development to single industry. On behalf of the above complex and practical commercial challenges, this research aims to build and construct the electronic and traditional industries’ early-warning model individually with application of decision tree, neural network and support vector machine method to develop a prediction model suitable for electronic and traditional industries. This research also adopts the methods of importation variable of the adequacy to avoid possible effects of neural network which could reframe by improper variables, these methods includes the factor analysis, stepwise regression analysis and sensitive analysis together with a type neural network to build the model of constructing. This research wants to use to understand that bankruptcy prediction comparison between the electronic and traditional industries through different models constructed built. The results of the experiments are:
1. Sensitive analysis plus neural network are suitable for electronic and traditional industries. The accuracy rate is up to 82.5% for electronic industries to predict their financial crisis in the year before it happen, and it is up to 96.08% for the traditional industries.
2. Compare to the traditional industry’s early-warning model, this research which applied the methods of sensitive analysis plus neural network is better. It also shows that non- financial rate has important influence according to the final results. Moreover, compared to the electronic industry’s early-warning model, the accuracy rate is above the average level as well.
3. The research discover that the support vector machine is not the best choice. In addition, the accuracy rate of the support vector machine is controlled by the decision of the parameter, but it’s hard to find the best parameter. However, there are no problems in this respect to the sensitive analysis combine neural network.
4. Through the variable sieving of sensitive analysis, this research discovered that the electronic industry's most influential indexes are ‘non-operating revenue’ , ‘operating expenses rate’ and ‘inventory turnover’, and the traditional industry's most influential indexes are ‘liabilities ratio’ , ‘total assets growth rate’ and ‘supervisory board hypothecation rate’. Furthermore, it showed that the electronic industry has influenced by profitability, cost expenses, and management capacity, and the traditional industry has influenced by the debt paying capacity, growth rate, and corporate goverance.

並列關鍵字

Sensitive Analysis ； Support Vector Machine ； Stepwise Regression ； Decision Tree ； Financial Crisis ； Neural Networks ； Factor Analysis

參考文獻

1.王美慧，「運用類神經網路於證券業網路下單服務品質之研究」，顧客滿意學刊，第一卷第一期，頁1~頁30，2005。

20.黃振豐、呂紹強，「企業財務危機預警模式之研究－以財務及非財務因素建構」，當代會計期刊，第一卷第一期，頁19~頁40，2000。

25.蔡永泰，「運用資料探勘技術建構企業財務危機預警模型」，私立中原大學資訊管理研究所碩士論文，2005。

11.陳生祥，「以總體指標與企業財務資料預測企業財務危機」，私立中原大學資訊管理研究所碩士論文，2005。

21.葉怡芳，「構建財務危機預警模式─支向機與羅吉斯之應用」，元智大學會計學系碩士論文，2004。