銀行信用卡客戶信用風險評估模型之建立

Abstract

本研究是以國內某金融機構信用卡客戶為研究對象，利用Logistic廻歸分析來建立客戶的信用風險模型，研究目的為：利用客戶個人基本資料及繳款行為資料，找出顯著影響客戶信用風險之因素，並建立信用風險評估模型，進一步提供金融機構於信用風險之控管。 研究結果顯示：依繳款行為變數建立之模型（MODEL II）區別能力優於個人屬性變數所建立之模型（MODEL I），而以繳款行為變數為主，個人屬性變數為輔，所建模型（MODEL III）之區別能力又優於前二者。因此，以該模型為本研究之最終模型，其對逾期戶之區別正確率為91%，非逾期戶之區別正確率為88.8%，整體之正確率為89.6%。

In this study, we attempt to build a credit risk evaluation model for cardholders by using the logistic regression. We hope to find out significant behavior and individual characteristic variables that may cause overdue payment or default behaviors. Furthermore, the model could be used by bank managers to control and manage credit risk. Results show that behavior variables (MODEL II) are better than individual characteristic variables (MODEL I) when distinguishing defaulter or not. Moreover, the MODEL III which includes both individual characteristic variables and behavior variables is much better than the former two models. Hence, MODEL III is the final model in this study. The accurate rate of MODEL III is 89.6% of all samples, 91% of defaulters and 88.8% of non-defaulters.