臺灣資電業公司財務危機預警模型之探討 -Logit與類神經網路之應用

A Study of Financial Distress Diagnostic Models for Taiwanese High-tech firms－ Application of Logit and Neural Network Models

周郁菁(Yu-Ching Chou)

指導教授 ： 陳瓊蓉

中原大學/商學院/企業管理學系/碩士(2006年)

https://doi.org/10.6840/cycu200600266

摘要

臺灣在2004年陸陸續續傳出博達、訊碟及皇統等公開上市公司會計假帳、掏空疑雲、資金流向過大或不明等問題，引爆了本土性的財務危機風暴，且這些財務危機事件更是警惕我們單看公司的財報資訊來判斷ㄧ家公司的營運狀況已不再可行。有關企業財務危機模型的諸多研究中，針對企業的財務比率、股權結構等均已做過探討，但鮮少有人針對企業爆發財務危機前的警訊做過探討與分析。因此本研究希望藉由抽絲剝繭這些財務危機公司之事件始末，從中找出共同警訊，並藉此建立財務危機預警模型，以幫助投資人或主管機關即早採取適當的監督與對策。
本研究首先整理過去國內外文獻較常使用的財務變數與公司治理變數，另外觀察2004年博達、訊碟等公司爆發財務危機事件始末，找出共同財務危機警訊納入本研究之公司治理變數，再參考Tirapat and Nittayagasetwat(1999)多因子模型中的五項總體經濟變數來分別建立Logit迴歸模型與倒傳遞類神經網路模型在財務危機發生前一年、前二年、前三年之財務危機預警模型。
實證結果發現：
1.距離發生財務危機的時間點越近，財務危機公司與財務正常公司越多的財務變數 與公司治理變數有顯著差異。
2.以Logit迴歸所建立的財務危機預警模型，發生財務危機前一年的迴歸式變數為速動比率、大股東持有股份比例及存貨週轉率；發生財務危機前二年的迴歸式變數為速動比率、稅前淨利率、大股東持有股份比例及變更財務主管次數，發生財務危機前三年的迴歸式變數為流動比率及總資產報酬率，其中前一年及前二年具有「速動比率」與「大股東持有股份比例」共同變數。
3.Logit迴歸模型的正確區別率在發生財務危機前一年、前二年及前三年分別為85.19%、83.33%、76.85%；倒傳遞類神經網路模型中訓練樣本整體正確區別率在發生財務危機前一年、前二年及前三年分別為98.85%、95.40%、89.66%；測試樣本則分別為90.48%、85.71%、71.43%。比較後發現在各年度中，倒傳遞類神經網路模型中的訓練樣本整體區別率均高於以Logit迴歸所建立的財務危機預警模型之區別率；而測試樣本的整體區別率僅在發生財務危機的前一年、前二年高於以Logit迴歸所建立的財務危機預警模型之區別率。

關鍵字

Logit ； 多層感知器 ； 類神經 ； 資電業 ； 財務危機

並列摘要

Procomp Informatic and Infodisc in 2004 had been found to be in financial failures due to accounting fault, defalcation and cash flow illegally. These events alert us shouldn’t to judge a firm’s operation only by financial statements. Prior researches often focused on financial ratios and ownership structure but few noticed the signs of financial distress. In this research, we analyze the financial distress firms in order to find some common signs to build up a financial distress model. We attempt to let the investors or managers find timely signs by the model, so as to adopt some solutions in advance to avoid the happening of financial distress.
We sum up some financial and corporate governance variables which were to be used mostly in the past. In addition, we observe the financial distress firms to find out the signs. This study integrates financial, corporate governance and macro economics variables which are referred from the research of Tirapat and Nittayagasetwat (1999) to build up the financial crisis models.
The results are as follow:
1. It exist some differences between the financial and the corporate governance variables of the normal and failure firms. The closer to the time of the failure, the more different they would be.
2. The variables of Logit model are as follow:
(1) Previous year of the happening financial distress are quick ratio, the ownership ratio of the major shareholders and inventory turnover.
(2) Previous two years of the happening financial distress are quick ratio, earning before tax, the ownership ratio of the major shareholders and the times of the chief financial officer (CFO) turnover.
(3) Previous three years of the happening financial distress are current ratio and return on assets.
3. The predication ratio of previous year, previous two years, previous three years are as follow:
(1) The results of the Logit model are 85.19%、83.33%、76.85%.
(2) The results using the backpropagation neural networks model of the test samples are 90.48%、85.71%、71.43%.
The prediction results using backpropagation neural networks model of the test samples are higher than the Logit model in previous year and previous two years.

並列關鍵字

Financial distress ； Logit ； High-tech firms ； Backpropagation neural networks ； Multi-Layer Perceptron

參考文獻

11.林金賢、劉沂佩、鄭裕書、陳育成，具學習性之模糊專家系統在財務危機上之應用，管理學報，第21 卷，第三期， pp.291-309。

25.陸怡伸，2003，公司財務危機預測－考慮公司治理及關係人交易影響之實證研究，中原大學會計研究所未出版碩士論文。

1.Alkhafaji, A.F., 1990. Effective boards of directors: an overview. Industrial Management and Data Systems: 18-26.

2.Altman, E.I., 1968. Financial Ratios Discriminant and the Prediction of Corporate Bankruptcy. Journal of Finance (September): 589-609.

4.Altman, E. I., G. G. Haldeman, and P. Narayanan, 1977. ZETA Analysis: A new model to identify the bankruptcy risk of corporations. Journal of Banking and Finance: 29-54.