運用資料探勘技術建構電子業財務危機預警模型

Using Data Mining Approach To Build Financial Crisis Prediction Model For Taiwan Electronic Industry

許佳豪(Chia-Hao Hsu)

指導教授 ： 陳大正

國立虎尾科技大學/管理學院/資訊管理研究所在職專班/碩士(2010年)

摘要

隨著企業經營型態的轉變及瞬息萬變的外在環境，讓企業的經營受到內、外在因素的影響，包含內部管理與決策及外在經濟環境景氣等因素。財務危機的發生原因眾多，爆發後往往事後都難以收拾，不但造成很多員工的失業，也影響到大眾投資人的利益。一旦企業遭遇危機而面臨倒閉，勢必造成社會不安定與金融秩序嚴重的衝擊，影響到社會大眾許多層級的損失。因此，建立一套穩定且有效的財務危機預警模型，便可儘速採取措施以預防危機發生，形成極有參考價值且可防範於未然。本研究以上市上櫃電子產業公司為樣本，本研究樣本資料從台灣經濟新報資料庫(TEJ)蒐集2004年至2009年間符合本研究定義之財務危機公司，以採取1:1的比例選樣，104家發生財務危機的上市公司配對104家正常上市公司，也就是說用一家危機公司配對一家正常公司做為研究資料。將資料區分為訓練樣本及測試樣本兩組。訓練樣本，其中有52家發生財務危機的上市公司配對52家正常上市公司，建構財務危機預警模型；而測試樣本，其中有52家發生財務危機的上市公司配對52家正常上市公司，用來測試預警模型的預測能力。本研究利用以決策樹為基礎的資料探勘出“若則”的分類預測法則，並且藉由所探勘的法則建立分類預測模型。顯示結合財務比率和公司治理的變數所建構的財務危機預警模型，比單純利用財務比率建構的財務比率預警模型和公司治理建構的公司治理預警模型有更佳的準確率。更進一步利用C5.0決策樹為基礎之boosting集成法，顯示由多重分類器所建立的分類模型比單一分類器模型，具有較高的分類準確率，且型一錯誤和型二錯誤都可以降至最低。

關鍵字

財務危機預警 ； 資料探勘 ； 分類法則

並列摘要

With the changing business patterns and respond to the external environment for enterprises operating under the internal and external factors including internal management, decision-making, external economic environment, and other factors. The financial crisis is always caused by many unknown reasons. When it has become more acute, it is usually difficult to clean up after all. It also caused the serious unemployment for many people, and the loss for the investing people’s interests. Company will inevitably cause social unrest, financial order and a serious detriment to the community in many levels of loss, once it is facing the financial crisis or collapse. Therefore, to building effective financial distress predicting models, which is stable and effective, can quickly be taken measures to prevent crisis.
This study is to investigate some Electronic Industry Company of Listed and OTC. The data in this study are collected from 2004 to 2009 to fully meet definition of research company’s financial crisis based on Taiwan Economic Journal Co., Ltd. To take the ratio of 1:1 sample, 104 listed companies with financial crisis 104 matching normal firms, that is paired with a crisis in a normal as company research information.
The information is divided into two groups of training sample and test samples. As follows:Training sample is construction of financial distress prediction model with 52 listed companies in financial distress the normal 52 pairs of listed companies.Test sample is used to test the predictive power of early warming model with 52 listed companies in financial distress the normal 52 pairs of listed companies.
As the result, it shows that the combination of financial ratios and corporate governance variables in the construction of early warning models of financial crisis is with better accuracy than use of the financial early-warning model and corporate governance of financial early warning model. Moreover boosting method based on decision-tree is also applied. The experimental results demonstrates the multi-classifier classification model is with higher classification model than a single classification model.

並列關鍵字

financial crisis forecast ； Data mining ； Classification rule

參考文獻

[1] Altman, E. I. , “Financial ratios: Discriminant analysis and the prediction of corporate bankruptcy.? Journal of Finance , 23 (4) :589-609, 1968.

[2] Beaver, W. H. , “Financial ratio as predictors of failure.?Journal of Accounting Research, 4: 71-111, 1966.

[3] Berry, M. J. and Linoff, G., Data Mining Techniques for Marketing. Wiley Press, 1997.

[4] Blum, M. , “Failing company discriminant analysis.? Journal of Accounting Research ,12(1):1-25, 1974.

[5] Breiman, L. , “Bagging predictors ”, Machine Learning, 24 (2):123-140, 1996.