台灣財務危機公司預警模型之探討︰ 景氣衰退期與成長期之比較

A study of the Financial Distress Diagnostic Model for Taiwanese Firms：Comparison between Recession and Prosperity

林俊男(Chun-Nan Lin)

指導教授 ： 陳瓊蓉

中原大學/商學院/企業管理研究所/碩士(2006年)

https://doi.org/10.6840/cycu200600259

摘要

摘要
隨經濟進步及加入WTO等影響，台灣企業在日趨開放的金融體系和資本市場下蓬勃發展。然而近年來不論是何種產業，財務危機的負面消息卻經常出現。因此如何建構有效預警模式供產官學界使用，以即時察覺問題並作適當決策來避免風險，便成為當務之急。
過去研究多使用財務變數來建立財務危機預警模式，然而Jensen and Meckling (1976)認為代理成本會因利益衝突而產生，且或多或少與財務危機有關，因此也須考量非財務因子並納入模型中來加強預測能力。 此外，Mensah (1984)認為使用跨年度資料上未考慮外在經濟會影響模型效果。Platt and Platt (1990)指出資料會隨研究時期不同而受到某些外在經濟因子影響。因此，本研究根據Grice and Ingram（2001）指出Altman（1968）原始模型中並未考量的七項缺失中的時點區分問題及外在經濟環境因素進行深入探討。並透過經建會所發布的景氣對策信號與經濟成長率的综合比較圖及Kane (1996)對景氣擴張期與衰退期的起迄時點定義，選取樣本時點分別為景氣衰退期（共三期，分別為1998/01-1998/12；2000/10-2001/9；2004/7-2005/3）及景氣成長期（共兩期，分別為1999/01-2000/9；2001/10-2004/6），分別建立兩套財務危機預警模式，並分別比較說明這兩者判別結果。
研究發現Logit模型預測結果在衰退期時，財務危機發生前一、二、三年分別為86.55%、83.04%、80.70%，在成長期時各年度分別是94.34%、88.05%、83.65%。而類神經網路模型測試組結果顯示幾乎可完全預測破產情況發生。景氣衰退及擴張兩期間的模型在失敗前一年的區別率皆明顯高於距破產時點較遠年度的模型。此外，景氣成長期的模型預測力的確有稍高於景氣衰退期的模型結果，足以支持考慮經濟景氣來建立不同經濟狀況的企業失敗預警模式是重要的。而類神經網路模型透過學習效果顯示不同景氣下各年的預測力皆高於Logit模型，顯示的確有較高的區別能力。
最後，企業應該從獲利能力及成長能力的財務變數等下手來降低發生財務危機風險。然而，大股東持股比率和董事持股比率兩個非財務變數也是研究中常出現變數，是一般財務報表外，投資大眾、主管機關、國內外法人投資機構、金融機構甚至是往來上下游客戶可著重指標之一。如此，無論是對改善公司內部體質或提供給外部各界偵查檢視都可獲得最大效益。

關鍵字

Logit模型 ； 景氣循環 ； 因素分析 ； 類神經網路

並列摘要

ABSTRACT
With the influence of the economic improvement and the entrance into WTO, the Taiwan enterprises flourish in the gradually open financial system and capital market. However, recently no matter in which industry, the negative news of the financial distress shows up frequently. How to build up an efficient financial distress diagnostic model for public to use to be aware of the problems and adopt the suitable decisions to avoid the risk becomes the most urgent task due to the reasons mentioned above.
The past researches often used the financial variables to build up the financial distress diagnostic model. Nevertheless, Jensen and Meckling (1976) believe that the agency cost would come into existence owing to the interest conflict and it may relate to the financial distress. We should consider the non-financial factors into the model to enhance the predict power. Besides, Mensah (1984) believes that using the yearly data without considering the outside economic situation would effect the model results. Platt and Platt (1990) point out that the data would be influenced by some outside economic factors in different research periods.
Our research bases on the problems of the period separateness and taking the outside economic environment factors into account which Grice and Ingram（2001）point out the two of seven shortcomings in the original Altman’s model. We use the synthetic compared picture of the business indicators and the economic growth rates released by the Council for Economic Planning and Development and Kane’s (1996) definition for the beginning and the end of the recessionary and prosperous period to select the research samples. We separate the samples into the recessionary period（Three periods︰1998/01-1998/12；2000/10-2001/9；2004/7-2005/3）and the prosperous period（Two periods︰1999/01-2000/9；2001/10-2004/6）to build up two different financial distress diagnostic models. Finally, we would compare and explain the discriminate results of the two different models.
We find out that the prediction results of the Logit model in the recessionary period before the financial distress are 86.55%、83.04% and 80.70% and prosperous period are 94.34%、88.05%、83.65% respectively. The prediction results of the artificial neural network model using the teat samples show that it could almost predict the happening of the financial distress completely. We also observe that the discriminate rates of the both periods in the previous year before the happening of the financial distress are higher than the earlier ones in both models. Besides, the prediction results of the prosperous period are truly higher than the recessionary one and can support our consideration for using the economic situation to build up the different financial distress models. Finally, the artificial neural network model through learning has greater discriminate ability than the Logit model.
To sum it up, the enterprises should focus on some financial variables categorized in the profit and the growth classifications to lower the financial distress risk. Besides, the ownership ratios of the major shareholder and the directors are two non-financial variables showing all the time in the research and important indexes for the public、the government offices、the investment and financial institutions. By considering these variables, we could get the most effect not only in improving the financial situation of the enterprises but also in providing financial information for outsiders to inspect.

並列關鍵字

Logit Model ； Business Cycle ； Artificia ； Factor Analysis

參考文獻

9.陸怡伸（民92）“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”，私立中原大學會計研究所碩士論文。

11.黃振豊、呂紹強（民89）“企業財務危機預警模式之研究-以財務及非財務因素構建”，當代會計，頁19-40。

1.AICPA（1987）Statement on Auditing Standards No. 59. The Auditor’s Consideration of an Entity’s Ability to Continue as a Going Concern. New York.

2.Alkhafaji Abbass F. (1990),“ Effective Boards of Directors：An Overview,” Industrial Management ＆ Data Systems, Vol.4, pp.18-26.

3.Altman et al. (1977),“Zeta analysis: a new model to identify bankruptcy risk of corporations,” Journal of Banking Finance, Vol.1, pp.29– 54.