建構企業危機預警系統-以台灣電子業為例

Alerting Model on Business Failure-A Study of Electronic Company in Taiwan.

胡婉鈞(Wan-Chun Hu)

指導教授 ： 胡為善

中原大學/商學院/企業管理研究所/碩士(2005年)

https://doi.org/10.6840/cycu200500409

摘要

近兩三年台灣也發生多起地雷股事件，讓社會措手不及遭受重挫，其中財務危機的公司多數為一般投資大眾所注意的電子產業且許多導致財務危機的預兆，並非可完全從公司的財務報表中看出端倪，甚至包括企業經營者的個人道德操守。本研究目的為以下兩點：1、探討電子業危機發生前一年度至前三年度，影響危機企業最主要之變數。 2、比較以羅吉斯分析及決策樹分析建構財務預警系統之不同。
本研究採取羅吉斯分析與決策樹分析建構危機預警系統。進行羅吉斯分析又可分為三階段，第一階段將變數分為兩部分再進行羅吉斯分析，一部分的變數包含了非財務性指標與財務性指標，另一部分的變數僅包含了財務性指標。進行決策樹分析時，將訓練資料分為：危機發生前一年、危機發生前兩年、危機發生前三年及危機發生前兩年加前三年之資料。測試資料皆為危機發生前一年。
本研究之結論如下:1.若單純以羅吉斯建構財務預警之模型，其模型預測能力相當高，但無法偵測出影響危機發生之變數為何。但若僅以財務性指標卻不適用現實的情況，因此加入非財務性指標是必要的。2.本研究發現，先進行主成分析後再進行羅吉斯分析，較先進行逐步回歸的預測能力來得低。但相同的是，在危機發生前一年之影響變數皆包括了非財務變數中的「有無變更會計師」。3.若以決策樹建構危機預警模式，越靠近危機發生年度其預測準確度越高。樣本數越大，決策樹的深度越大，能偵測出較多影響企業發生危機之變數，但預測的能力不一定會更準確。4.影響企業發生危機最主要之非財務變數為---公司股權結構。其中包含政府機構持股比例、金融機構持股比例及董監事持股比例。5.雖然決策樹能偵測出較多影響危機發生之變數，但危機預測能力較傳統羅吉斯分析為低。原因在於決策樹屬於監督式之模型，因此需要將資料分為訓練樣本與測試樣本，若在樣本數較少的情況下，決策樹無法建立十分精確之企業危機預警模型。

關鍵字

決策樹 ； 財務預警 ； 羅吉斯分析

並列摘要

For the past three years, there have been many companies in Taiwan suddenly announced bankruptcy and caused investors huge loses. Practically, it is very hard for investors to find out the failed companies in advance by simply reviewing the financial statements of these companies, especially if their managers had business ethical problems. This study attempts to achieve the following objectives: (1) To investigate the major factors affecting financial crisis for the electronic companies for the years ahead of the occurrence of bankruptcy；(2) To compare the Logistic model with the decision tree analysis in order to construct an accurately financial alerting model.
The conclusions of this work are summarized as follows:
(1) The financial alerting model based on the simple Logistic model could not completely describe the key factors causing the financial crisis. It suggests that the non-financial factors should be included.
(2) The findings indicate that the explanatory power of the stepwise regression is better than that of principal component analysis. However, both methods confirmed that the same question that a company changes its Accountants or not be an important determinant factor to predict the financial problem of a company.
(3) When a decision-tree method is used to construct a financial alerting model, the result shows the following two points: (i) the closer the year of financial crisis is, the higher the accuracy of prediction is; (ii) the larger the sample size is, the deeper-stage the decision-tree is, and the more factors on predicting financial crisis are found.
(4) This study finds that the key non-financial variable affecting a company’s financial crisis is the equity structure of the company, which consists of the percentage of stock holdings by government; that of stock holdings by financial institutions and that of stock holdings by the directors and supervisors.
(5) This investigation also finds that, although decision-tree method could detect some factors influencing financial crisis than Logistic model does, yet the explanatory power of the decision-tree is inferior to that of Logistic model. The possible reason is that the decision-tree method is a kind of supervisory model, which requires the entire data to be classified into training sample and testing sample. If the sample size is too small, then the decision-tree method cannot construct a financial alerting model which has an accurately predictive power

並列關鍵字

Decision tree model ； Logistic Model ； Financial alerting model ； Predictive power ； Non-financial factors

參考文獻

施正宏 (2003)「結合總體經濟指標及個股財報資料以預測個股漲跌--以台灣電子類股為例」，中原大學碩士論文。

27.龔志明 (2000)「財務危機預測模型之跨期性研究」，中山大學財務管理研

鄭忠樑 (2002)「運用分類樹於股價報酬率預測之研究」，元智大學碩士論文。

Altman, E. I., “Financial Ratios, Discriminant Analysis and the Prediction of

Altman, E. I., Rovert G. Haldeman and P.Narayanan,(1997) “Zeta Analysis-AnewModel to Identify Bankruptcy Risk of Corporation.” The Journal of Banking and Finance, pp29-54.