資料採礦應用於中小企業信用保證授信風險預測～以T銀行為例

根據亞洲開發銀行（ADB）估計，亞洲地區將逐漸成為全球經濟發展的重心，因此金管會期許本國銀行，持續提升資產品質，以掌握亞洲經濟迅速成長商機、提高獲利率。為幫助T銀行能由整體逾放比已相當低的授信案件中，再提升資產品質，爰切割出逾放比仍高之中小企業信用保證授信案件作為改善標的。經蒐集T銀行2011年至2013年所貸放中小企業信用保證基金保證之案件2,474件，包含貸放後一年內即逾期之案件80件及繳息還本均正常之案件2,394件，以其貸放前一年之財務資訊及非財務資訊，篩選出影響中小企業信用保證基金保證授信案件逾期之主要變數，並運用資料探礦方法建立合適之逾期模型。實證結果，以「逾期戶」及「正常戶」1：2之超取樣比例搭配羅吉斯廻歸所建構模型，兼具預測準確率及穩定度，是為本次研究評估後之最佳模型，其比不用任何模型預測相較，可增加42.56%機率篩選出短期內逾放之案件。

關鍵字

資料採礦；超取樣；中小企業信用保證；逾期放款

並列摘要

The Asian Development Bank estimates that Asia will increasingly become the focus of global economic development, and therefore Financial Supervisory Commission R.O.C (FSC) expect Domestic banks to continue improving their asset quality in order to grasp the business opportunities in fast-growing economies of Asia. Although T Bank has good credit quality assets, in order to achieve the expectations of the FSC, screening out the small business credit guarantee (SMEG) loans as the object to be improved, because of its overdue loan ratio remains high. By collecting SMEG loans 2,474 cases of T Bank from 2011 to 2013, including 80 overdue cases within a year after receiving loans, as well as to pay interest repayment of principal of the cases were normal 2,394 cases. With the above cases in the financial information and non-financial information before grant loans one year, selected key variables affect the SMEG loans overdue, and use Data Mining methods to establish an appropriate forecasting model of overdue. The test results, ＂the overdue case＂ and ＂normal case＂ of 1:2 ratio of oversampling constructed with Logistic Regression model, both forecast accuracy and stability, assessed as the best model of this study, and this model can be increased 42.56% probability screened overdue short-term case than not compared to any model prediction.