應用資料採礦技術建置符合新巴賽爾協定之企業信用風險模型

巴塞爾銀行監理委員會於2001年1月公佈新版巴塞爾資本協定，並於2006年底正式實施。新協定鼓勵銀行能建立自己的內部評等系統評估違約風險，並重視銀行放款風險考量資訊的量化和降低計提所需資本，進而提高金融機構風險敏感性，以彌補傳統標準法的不足。為因應此趨勢，本研究以台灣公開資料庫的資料為實例，資料的觀察期間為1996至2005年，透過資料採礦流程，以製造傳統產業公司之授信樣本為主要的研究對象，建構企業違約風險模型及其信用評等系統。本研究分別利用類神經網路、羅吉斯迴歸和C5.0決策樹三種方法建立模型並加以評估比較其預測能力。結果發現羅吉斯迴歸模型對違約戶的預測能力及有效性皆較其他兩者為佳，因此，以羅吉斯迴歸方法所建立的模型為本研究最終模型。接下來便針對該模型進行各項驗證，驗證後發現此模型即使應用到不同期間或其他實際資料，仍具有一定的穩定性與預測效力，確實能夠在銀行授信流程實務中加以應用。

關鍵字

新巴塞爾資本協定；資料採礦；信用評等；羅吉斯迴歸模型

並列摘要

In January 2001 the Supervisory Review Process committee of Basel has announced the new version of Basel capital requirements and has officially set forth the implementation of the requirements in 2006. The revised accord aims to improve risk violation assessment, make regulatory capital more risk sensitive, lower asset minimum requirements and enhanced risk sensitivity management in the banking industry. The data adopted in this study is based on the publicly published data from Taiwan banking industry database. The data used is from the period of 1996 to 2005. The study attempts to construct the model in assessing risk violation and credit evaluation against the traditional manufacturing industry. The methods used in this study is through the process of data mining using Neural Network, Logistics Regression and C5.0 Decision Tree Analysis to construct the model and evaluate the prediction accuracy among the three models. The result shows that the prediction accuracy of Logistics Regression analysis is more favorable than the other two models and therefore is the chosen choice of the model in this study. In addition, the study has adopted multiple testing criteria to validate the validity and accuracy of the model which after testing has confirmed its reliability to be accepted in its application in the banking industry.