本研究在於探討銀行在信用卡產品，如果利用行內既有資料是否可建置出預測能力足夠的信用評分模型，以協助銀行進行信用風險管理。文中以國內某銀行為實際案例，針對該行信用卡循環客戶為對象，在2008年9月至2013年7月的歷史資料中抽樣建置行為評分模型，以評估其預測效果，研究結果如下： 一、無論在訓練組、樣本外測試組和時間外測試組，模型Gini係數穩定維持在64%左右，K-S值穩定維持在50%左右，代表模型跨樣本、跨時間的預測能力皆能保持良好，足以作為該銀行在信用卡產品信用風險管理的參考依據。 二、該行為評分模型以「繳款行為變數」為判別違約與否之關鍵指標，且明顯優於「個人屬性變數」。 本研究相關數據與結論可作為未來建置信用卡行為評分模型時的參考資料。;The purpose of this study is to explore whether banks can build credit scoring models with sufficient predictive ability by using internal data to assist banks in credit risk management of credit card products. In this paper, a domestic bank as a practical case, and build a behavioral scoring model for the credit card revolving customers from September 2008 to July 2013 to evaluate the forecasting results. The results are as follows: 1.The Gini coefficient remained stable at about 64%, and the K-S test stably remained at about 50%, which means that the prediction ability of the model across the sample and across time could be kept good, both in the training group, the out-of-sample testing group and the out-of-time testing group. Which can be used as the reference for credit risk management of credit card products. 2.The behavioral score model uses "payment behavior variables" as a key indicator of the probability of default, and is clearly superior to the "personal attribute variables". The data and conclusions of this study can be used as a reference for building of credit card behavioral scoring models in the future.
