台灣近幾年來企業陸續傳出財務危機、掏空公司資產、及對外發佈虛偽的財務報表等事件。上述這些事件的發生,皆導因於未能對企業做有效持續性的內部控制。然而現今內控聲明書或相關內控制度聲明,已出現形同虛設的情形,導致弊案地雷股頻傳,因此本研究希望能以內部控制觀點建構危機預警模型,提供投資大眾、企業高階管理人員及相關查核人員作為決策之參考。 國內過去對於內部控制之相關研究幾乎都採用問卷調查法或深度訪談法來進行研究。本研究試著從公司治理觀點及COSO內部控制五大構面角度,找出能代表衡量內控之影響因子,尤其是可量化變數,以達到建構準確偵測出內部控制優劣之預警模型,以協助公司了解現有內控制度之缺失並提出改進方法。 本研究樣本,以1999年至2006年間發生財務危機之108家公司,再以1:2之配對方式選出建全216家公司,共計324家公司為研究對象。為了檢驗內控模型之預測正確率與穩定性,本研究更將其研究樣本區分為訓練組 (1999至2004年)與測試組 (2005至2006年)進行相關測試。最後,為比較各種不同內控預警模型之優劣,本研究採用MDA、Logit、類神經、支向機及基因演算等模型,比較各種模型在相同內控變數資料下,檢驗是否在建構企業內控危機預警模式下,預測能力上有哪些不同結果。 實證結果顯示: 1. 公司發生財務危機與否,與公司內部控制之嚴謹與否確實存在密切關係,在COSO內部控制五大構面中,除了資訊溝通構面中無內控變數呈現統計顯著性外,其餘COSO四大構面皆有內控變數在檢定後呈現統計顯著性發生。 2. 以非財務變數建構內控財務危機預警模式中,在「訓練組」模式部分,基因演算法、支向機及倒傳遞網路三種新理論方法所建構之內控危機預警模式之預測能力皆高達100%且明顯優於傳統的Logit及區別分析方法。 3. 以非財務變數所建構內控財務危機預警模式中,在「測試組」模式部分,倒傳遞類神經網路之預測能力為最高,其次為基因演算法、Logit、支向機及區別分析法。 4. 內部控制變數能建構有效之內控危機預警模型,在「訓練組」部分,大部分的內控預警模型皆達到100%準確之區別能力,區別分析模型最低亦有92%之正確率;在檢驗「測試組」部份,除了傳統區別分析模型僅有79.6%之預測準確率外,其餘內控預警模型皆能達到接近85%之正確率,代表以內部控制變數為主而建構之內控早期危機預警模型,能相當有效地偵測出公司發生財務危機之原因,並能幫助公司早期防範與改善。 5. 以內部控制非財務變數所建構之內控財務危機預警模型而言,區別分析模型所建構之內控危機預警模型與其他4種預警模型比較,區別分析預測正確率為最低。
In the recent year, many firms have financial distress, transfer company funds into the accounts of family members, and declare financial statement to outside users. Everything we described above is due to that these companies could not have effectively continuous internal control system. Many illegal cases were happened, because statement of internal control or related announce of internal system are not valid today. Thus, this research will offer some empirical results for investors, high class manager in companies and related auditors to make decision by viewpoints of the internal control system of establishment of an Early Warning Model for Financial distress. In Taiwan, related internal control researches had to be done by survey or in-deepth talking in the past. This research tries to use corporate governance and five COSO internal control perspectives to build up Taiwan financial distress prediction models to find effective factors which can be represented by balanced internal system. This research also helps company to understand the shortage and address improved ways. This study takes 324 companies as a total sample, and 108 of these companies are distress firms from 1999 to 2006. In order to check the accuracy and stability of internal control model, this research divides these samples into training set (1999 to 2004) and validation set (2005 to 2006) to do related testing. Finally, in order to compare differently internal control early warning model’s pro and con, this research use models of MDA、Logit、BPN, SVM, and GA-SVM. Comparing 5 different models using the same internal control information variables. The empirical results of this paper indicated we follows: 1. Internal control variables is highly related to distress companies. In COSO’s five perspectives of internal control variables, all IC variables in 5 different perspectives are attain significant level without information communication perspective. 2. The empirical results of training set data indicated that the predictive accuracy of GA.SVM and BPN models are all better than traditional models. 3. In validation dataset, the highest predictive accuracy is BPN model, and the second and others ranked are GA.Logit and SVM, MDA performed is the poorest. 4. Selecting the useful internal control variables to build up an early warning financial distress model. In the training set, almost each models can reach 100 % accuracy. But the lowest accuracy rate of MDA is 92 %. However, in the part of testing set, except for 79% of accuracy rate in MDA, The other models all can reach 85% accuracy. The results indicated using internal control variables can help company to build up a good early warning distress prediction model.