投資人對於企業的評估已無法從最簡單的財務變數進行投資決策，有鑑於過去的財務危機預警機制之文獻鮮少加入風險調整的概念，因此除了財務變數及非財務變數，本研究特別加入風險值變數，以增加模型之預測能力。此外，由於傳統支援向量機(SVM)在解決二次規劃的問題會是一個相當棘手的問題，在資料量過多的情況下會導致模型的訓練時間相當冗長，為了提升模型的實際應用範圍，本論文將採用雙支援向量機(TSVM)，它不僅保有將傳統支援向量機的優越預測效力，更透過將二次規劃的問題做細部拆解，以降低模型的運算成本，對於重要變數的選取上，將採用逐次前饋式搜尋法(SFS)及羅吉斯迴歸，透過逐次前饋式搜尋法(SFS)可以降低維度魔咒的問題，本研究將比較逐次前饋式搜尋法(SFS)與羅吉斯迴歸所篩選之變數做比較，再將相關的結果匯入到雙支援向量機以構建企業危機預警的模型，本研究結果顯示，加入風險值之模型確實有助於提升模型準確率，在所選用的預測模型中，又以雙支援向量機(TSVM)具有優越的預測能力。
In the past, many scholars used financial ratios to construct financial crisis prediction model. But now, we can’t just using financial ratios to predict financial distress. It is because the economic environment is becoming more and more complex. In view of the past, seldom studies considered about Value at Risk (VaR) factor, which could affect performance of financial crisis prediction model. In addition, as traditional support vector machine (SVM) in solving the problem of quadratic programming problem (QPP) will be a very difficult problem. It will take model a longer time for training. In order to improve efficient of our study, we use twin support vector machine (TSVM) for our study. In variables screening, we use sequential forward selection (SFS) and logistic regression. Through sequential forward selection can reduce the curse of dimensionality. The results of our study show that the model with VaR does help to improve the accuracy of the model, also we find that twin support vector machine (TSVM) has better forecasting ability than support vector machine (SVM).
