風險管理長久以來一直是財金、經濟與會計領域的熱門議題，再加上近年來國內外會計醜聞事件頻傳，國內外學者紛紛致力於財務危機預警模型之構建，然而，企業遭受到財務危機的主因為管理階層之經營效力不彰，卻鮮少研究對企業管理階層的經營效力做探討，因此，本計畫第一年將提出一個企業繼續經營效力之評估模型，並進一步將多角化的經營策略與營運之風險因子予以考量，在評估模型的選用上將採用可以同時處理多投入與多產出變數的模型，而過多的變數會降低模型之評估效果，因此，本模型亦將降維之技術予以整合。此外，為了降低投資者對企業繼續經營的疑慮，並且加速其決策制定的時效性，本計畫第二年將更進一步提出企業繼續經營之預測模型，此模型結合了兩階段分群技術、特徵擷取、整體學習架構下之預測模型與啟發式知識萃取技術。兩階段的分群技術整合了聚合式階層分群法與遞迴式k平均法，其用來選取具代表性的樣本；特徵擷取技術是用來降低維度魔咒之困擾；整體學習概念融入預測模型之構建可以降低預測偏誤與提高效果；啟發式知識萃取技術用來呈現預測模型內部的決策邏輯，並以易於使決策者瞭解的方式呈現，此模型可有效降低投資大眾的風險暴露與穩定金融市場之發展。 Risk management has been an essential research domain in finance, economic and accounting for the last three decades. Recently, numerous accounting scandals started to impair the economic development of many countries. The financial crisis pre-warning model turns to be much more important and doubtlessly catches public attention when the global capital market goes to depression. However, most proportion of the related studies laid much emphasis on financial crisis pre-warning model construction, research on corporate operating performance is rather limited which is the main cause of financial crisis. Thus, this study proposes an emerging corporate operating performance assessment model which integrated risk factor under diversification operating strategy. The result form this model can give a suitable direction for managers to modify corporate’s operating strategies and realize corporate its real operating condition. Furthermore, the study advanced proposes a hybrid forecasting model for decision maker to assess the corporate operating performance. The hybrid model consists of two-stage clustering algorithms, feature selection, multi-stage SVM ensemble, and rule induction by ant colony optimization. Two-stage clustering algorithm is used to determine the representative dataset which can decrease large amount of redundant and useless data. Feature selection is used to pick up the essential features from the original dataset without eliminate the model’s performance. The integration of ensemble learning and support vector machine is used to enhance forecasting model’s performance and eliminate the forecasting bias. The fundamental idea of ensemble learning is used to complement the error made by singular mechanism. Outstanding forecasting performance of multi-agent SVM ensemble comes with a critical defect is lacking of realization ability which will impedes its practical application. To overcome the opaque nature of multi-agent SVM ensemble, the study implements rule induction algorithm by ant colony optimization to extract the inherent decision logic for decision maker. The decision makers can use this model to modify his investing portfolio and eliminate the risk explosion. The public sector can use this model to adjust his policies to sound the stability of global financial market.
