應用支援向量機建立農會信用部財務預警系統

農會信用部在地域性金融機構裡扮演重要的角色，金融海嘯之後，金融機構的風險控制日益重要，然而建立財務預警系統則可監控農會信用部的財務狀況，預防並降低其營運風險。本研究之目的在利用美國聯邦金融機構檢查評議委員會 (FFIEC)所制定的金融檢查制度CAMELS 等級方法建立農會信用部評等系統並使用支援向量機建立農會信用部財務預警系統。本研究首先按CAMELS 評等方式選出21 項財務變數，再透過因素分析篩選出15 項主要變數作為建立評等系統之投入變數。然後使用支援向量機建立財務預警系統，將2008 年至2011 年共976 筆資料做為訓練資料，透過格子點演算法計算出最佳參數組合，並預測2012 年農會信用部財務狀況表現，實證結果顯示，使用支援向量機建立財務預警系統預測評比等級落點準確率可達96.3115%，然而使用傳統統計方法邏輯斯迴歸準確率則為89.7541%。實證結果顯示支援向量機預測能力較邏輯斯迴歸準確。

關鍵字

農會信用部；財務預警系統；支援向量機；邏輯斯迴歸

並列摘要

Credit Departments of Farmers’ Associations play an important role in local financial institution. Controlling risk of financial institution becomes more critical after Financial Tsunami. Nevertheless, establishing early warning system can monitor the financial situation of Credit Departments of Farmers’ Associations. This research selects 21 financial indicators through CAMELS score, then sifts 15 main indicators from 21 indicators as variables of scoring system. This research establishes early warning system by support vector machines. Let 976 data as training data during 2008 to 2011. It calculates the optimal parameter combination through grid search and predicts the performance of Credit Departments of Farmers’ Associations’ financial situation. The experiment results show that the accuracy of establishing early warning system by support vector machines is 96.3115%. However establishing early warning system by logistic regression is 89.7541%. According to this result, support vector machines have better performance of prediction than logistic regression.