運用類神經網路建構台灣地區農會信用部金融預警系統

Application of the Neural Networks to Establish A Financial Early Warning System for Credit Departments of Farmers' Associations in Taiwan

本文旨在結合企業經營危機理論與投資組合理論，並搭配倒傳遞類神經網路預警模型，量身裁製符合台灣地區農會信用部經營特性之金融預警系統。研究結果顯示，農會信用部經營良窳係屬不同投資組合下之槓桿操作結果，而複雜的投資組合間存在程度上不同之抵換關係，須藉由金融預警系統綜合判定其營運等級。至於擠兌應以單純引發事件視之，並無法衍生為經營不善關係，否則將嚴重產生統計上型與型I誤差。另農會信用部於經營失敗過程中具有明顯的危機警訊可供金融預警系統事前偵測，蓋農會信用部經營失敗過程具有連續軌跡可供搜尋，而此連續過程亦代表不同營運評等等級之差異展現。

This study combines the business crisis and portfolio theories, together with the back-propagation neural network, to establish a financial early warning system catering to the operation needs of credit departments of farmers' associations in Taiwan.The empirical results show that the management performance of a credit department is highly related to its financial leverage operations among different portfolios, among which there exist trade-off relations, and that the ratings of the portfolio operations may be derived by the financial early warning system. Cases of bank runs should be viewed as exceptional due to their weak linkages to operational crises. Otherwise, Type-I and Type-II errors in statistics might occur. Furthermore, this early warning system is able to detect in advance the signs of crises caused by the operation failures of a credit department because there are continuous traces of the process of operation failures and these continuous traces signify the differences of various operation ratings.