整合多元邏吉斯回歸與多元序列探勘技術建構企業財務危機預警模型--以台灣電子產業為例

Applying Multiple Logistic Regression and Multiple Sequential Pattern Mining to Construct Corporation Financial Crisis Predictive Model－A Case of Taiwan's Electronic Industry

王宏綸(Hong-Lun Wang)

指導教授 ： 羅淑娟

國立臺北科技大學/管理學院/工業工程與管理研究所/碩士(2009年)

摘要

近年來我國經歷了金融海嘯，導致有些公司可能被迫列入全額交割股、暫停股票交易甚至為下市公司，造成的財務危機不僅波擊企業本身，連一般投資人及外在大環境都可能受到影響；從這次金融風暴中可以發現投資的不確定性以及準確投資的重要，因此本研究希望建構一套有效之財務危機預警模型提供使用者作為參考的依據，則可讓使用者儘早採取因應措施。
投資若能洞燭先機知道哪家的公司將出現危機，那麼投資的損失就可以得到控制，進而創造最大利潤，財務預警的研究通常利用分類模型來判斷公司是否為危機公司。本研究也針對這問題利用多元的方法給予這模型更公平的分類準則，本研究討論多元邏吉斯迴歸(Multiple Logistic Regression)與二元的邏吉斯迴歸(Binary Logistic Regression)在不同時間模型分類正確率比較，並且在有分類值的情況下，分別帶入序列探勘來預測未來的資料，並對預測的結果進行論述。
從研究證實在二元的分類正確率是大於多元的分類，但若多元分類降階為二元，也就犧牲掉一個目標值則模型將促使正確率提高至超越二元邏吉斯迴歸之分類，在結束分類之討論再加入序列探勘(Sequence Mining)其結果發現二元分類的正確率一樣大於多元分類，但二元有較高的未判率，因此在方法的選擇要看資料的型態以及投資的決策方針來決定使用的分類方法。

關鍵字

財務危機預警模型 ； 二元邏吉斯迴歸 ； 多元邏吉斯迴歸 ； 序列探勘

並列摘要

In recent years, our country experienced a financial tsunami, which may lead some firms force themselves suspend stock trading, shares or even became delivery companies. The financial crisis not only affected enterprises themselves, but also the general investors and external environment. From the financial turmoil, we can be found the uncertainty of investing, as well as the importance of accurate investing. Therefore, if we build an effective financial alert model for investors as a reference, investors can take early response accordingly.
If investors are able to notice crisis in advance, the loss of investing will be decreased and gain more profit. The most financial alert studies used classification models to determine whether it was a crisis company. In this study, the multiple methods is adoped which is more fair to give the guidelines for classifying. Then we discuss the accurate rates of binary logistic regression and multiple logistic regression model during different time period. And the values of classification of this study were substitured into the sequence mining respectively to forecast the financial crisis and discuss them.
The results showed that the classifying accurate rate of binary were better than multiple. But the time we ransformed the multiple into binary, that was we sacrificed a targer value, the accurate rate of multiple would surpass binary. Then the result substituting sequential mining showed that the accurate rates of binary were better than multiple. But the binary-classification had higher rates of non-definition. Therefore, the method we select depends on the types of the information as well as the approach of investion making.

並列關鍵字

Financial Crisis ； Predictive Model ； Binary Logistic Regression ； multiple logistic regression ； Sequential Pattern Mining

參考文獻

[9] 劉泓辰「用一般迴歸類神經網路與序列探勘技術建構企業財務危機預警模型－以台灣電子產業為例」，碩士論文，工業工程與管理研究所，台北2006。

[7] 陳仕昇、陳彥良、許秉瑜，「以可重複序列探勘網路瀏覽規則之研究」，資管評論，第九期，1999。

[43] R. Barniv, A. Agarwal and R. Leach, “Predicting Bankruptcy Resolution,” The Journal of Business Finance and Accounting, 29(3), 2002, pp.479-520

[12] E. I. Altman, “Discriminant Analysis and the Prediction of Corporate Bankruptcy,” The Journal of Finance, Vol.23, No.4, 1968, pp.589-609.

[13] A. Savasere, E. Omiecinski, and S. Navathe, “An Efficient Algorithm for Mining Association Rules,” Proceedings of the 21st Conference of Very Large Databases(VLDB) ,1995 ,pp. 432-444。