應用資料採礦技術建置中小企業傳統產業之信用評等系統

中小企業是台灣經濟貿易發展的命脈，過去以中小企業為主的出口貿易經濟體系，是創造台灣經濟奇蹟的主要動力。隨著2006年底新巴賽爾協定的正式實施，金融機構為符合新協定規範，亦需將中小企業信用評分程序，納入其徵、授信管理系統，以求信用風險評估皆可量化處理。故本研究將資料採礦技術應用於建置中小企業違約風險模型，針對內部評等法中的企業型暴險，根據新協定與金管會的準則，不僅以財務變數為主，也廣泛增加如企業基本特性及總體經濟因子等非財務變數，納入模型作為考慮變數，計算違約機率進而建置一信用評等系統，作為金融機構對於未來新授信戶之風險管理的參考依據。而本研究將以中小企業中製造傳統產業公司為主要的研究對象，建構企業違約風險模型及其信用評等系統，資料的觀察期間為2003至2005年。本研究分別利用羅吉斯迴歸、類神經網路、和C&R Tree三種方法建立模型並加以評估比較其預測能力。研究結果發現，經評估確立以1:1精細抽樣比例下，使用羅吉斯迴歸技術建模的效果最佳，共選出六個變數作為企業違約機率模型之建模變數。經驗證後，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，且符合新巴塞資本協定與金管會的各項規範，表示本研究之信用評等模型，確實能夠在銀行授信流程實務中加以應用。

關鍵字

中小企業；信用評等模型；資料採礦；新巴賽爾協定

並列摘要

In the developing history of Taiwan economy, the SMEs are always the key branches of it. Along with the formal implementation of Basel II in the end of 2006, the banking institutions also need to establish a credit scoring process for the SMEs into its credit checking system in order to conform with the new accords and to quantify the credit risk assessment process. Therefore, in this reserch we use the data mining techniques in establishing the default risk model for the SMEs. According to the new accords and the guidelines published by the FSC (the Financial Supervisory Commission), we not only take the financial variables as the core but also increase the non- financial variables such as the enterprise's basic characteristics and overall economic factors extensively into the default risk model for the enterprise-based exposure at default in the IRB in the second pillars of the Basel II. By the model we can get the probability of default then to establish the credit scoring system which could be one of the reference for the banking institutions to hold the risk management of any newly furture incoming customers. In this reserch, we take the traditional industry of the SMEs as the mainly object of study to establish the default risk model and the credit scoring system and the data was collected from 2003 to 2005.We use each of the following three methods, the Logistic Regression, the Nerual Network and the C&R Tree, to bulid the model and by evaluating several statistics test results to compare the capability of prediction of each model. As the result of this reserch, under the 1:1 oversampling proportion, using the Logistic Regression techniques to bulid the model would get the best outcome and there are six variables being chosen from the dataset as the final significant variables in the default risk model. After several conscientious and formal evaluations, we believe that this model would still hold the constancy and produce a good capability of prediction even with a different period of data or other practical data. And the model conform with the Basel II and the norm published by the FSC. It means that the credit scoring system can be put in practice in the banking credit process indeed.