近幾年來本國銀行不斷的追求完善的風險管理模式,然而對消費金融而言,信用評分表對於客戶群之區隔能力,是一門重要的課題;過去眾多針對此問題所提出的相關研究中,對於變數的選取大部分未考慮變數間的交互效果對模型的影響,故本研究在對本國某商業銀行所承做的信用貸款客戶做分析時,除找出對違約具有顯著之變數外,更嘗試找出具有交互效果之變數,並將這些變數整合出新的結合變數,一併納入Logistic Regression模型測試,其測試結果發現,納入結合變數之Logistic Regression模型其對樣本分類準確率較未考慮結合變數之模型為佳,此一結論可作為樣本公司未來建立評分模型及風險控管之重要參考依據。
In recent years, local banks have been pursuing ideal models for risk management. In the field of consumer loans, the issue of whether the credit scorecard is able to segment customers is especially important.In the past, most of the researches on this issue did not consider the influence of variable’s interaction effect on models when choosing variables. To take a further step, this research tries to find the variables which have interaction effect besides variables which is significant in default when analyzing the data of unsecured-loan customers acquired by one local commercial bank. Furthermore, these variables with interaction effect are integrated to become new combined variables and tested in the Logistic Regression model along with other variables. The result of the test finds that the Logistic Regression model with combined variables has a better performance on correctly classifying samples than the Logistic Regression model without combined variables. This finding can serve as an important reference for the case study bank to establish score models for risk management in the future.