結合灰關聯分析與倒傳遞類神經網路兩種方法,本研究期望能以更精簡的數據資料獲得更好的財務危機預警準確度,建立一套更便捷的企業危機預警模式,以協助上市櫃公司易於使用,於財務危機發生前,有效防範公司發生危機之可能性,避免惡化損失。本研究經由台灣經濟新報TEJ資料庫,實證研究顯示,經灰關聯分析共得出6項關鍵財務指標與1項關鍵公司治理指標,且經倒傳遞類神經網路的學習訓練發現,模型I的預警能力最佳,高達92.6%之正確率,模型II、III、IV的正確率相似且理想,皆為88.9%;6個關鍵財務指標與1個關鍵公司治理指標的預測力皆佳,惟關鍵財務指標的預測力略優於關鍵公司治理指標。因此,本研究建議財務危機預警模式經由精簡的數據資料亦可獲得優異的預警準確度,並非所使用的指標數量越多或時期數跨度越長,預警準確度就越高,若能善用灰關聯選取最具關聯性的指標與倒傳遞類神經網路的學習與訓練,少量的數據資料亦能得到非常良好的預警準確度。
Based on grey relational analysis and back-propagation neutral network, this study was proposed to examine if it is possible to use less information to obtain a better early warming model for financial crisis. It will assist business corporations to effectively forecast their financial issues before causing tremendous loss. A total of 37 corporations having crises happened during 2009 to 2012 were randomly sampled from the data base of Taiwan Economics Journal. Research results showed that 6 financial indicators and 1 corporate governance indicator are identified as the key factors and Model I, with a correctness of 92.6%, has a higher prediction level than Models II, III, and IV. This study reached a conclusion that it is not really necessary to use numerous data to build an early warning model, gray relational analysis and neural network can be applied to improve the efficiency and effectiveness of model building process.