本研究將信用卡違約模型依據使用時機分成信用卡申貸評分卡及行為卡，前者於客戶申請信用卡時使用，後者則於信用卡覆審、調整額度或續卡時使用。信用卡違約模型建模過程則分成資料蒐集、資料清理、因子分析、模型調校等步驟，探討各項信用評等因素與持卡人違約機率之間的關係，並得到信用卡申貸評分卡及行為卡之最終模型及模型調校函數，最終得到信用卡戶評等等級。
本研究發現信用卡申貸評分卡中客戶基本資料、聯徵中心與票據查詢資料、客戶往來行為等變數對違約／正常客戶有顯著區分效果；信用卡行為卡中客戶與本行往來行為、交易與付款資料、額度使用率、循環與預借現金餘額、消費者行為分群等變數對違約／正常客戶有顯著區分效果。
本研究將信用卡客戶分為26等級，等級越高信用卡戶未來1年的違約機率越高。未來可進一步發展違約損失率及違約暴險額模型，估算單一客戶的預期損失及信用風險最低資本需求，作為信用卡戶授信定價的基礎。另外可透過所預測之信用卡戶信用風險等級給予不同的促銷方案，加強顧客忠誠度。也可藉由了解因子特性與信用卡戶風險的關係，掌握風險較低的目標族群，並對該族群開發推廣信用卡業務以提升銀行獲利。

In this study, the author classifies Credit Card Default Model into two models, Applicant Model and Behavior Model, according to the time in which the model is implemented. Applicant Model, is adopted when customers apply for credit cards, while Behavior Model is utilized when credit of customers is reviewed, credit lines are adjusted, or the duration of the credit is extended. The process in which Credit Card Default Model is built comprise several steps, such as data collection, data cleaning, factor analysis, and adjustments to the model, etc. This study also probes into the relationship between various factors in credit ranking and default probability, and thus derives model calibration function and rating grade.

This study discovered factors in Applicant Model that can differentiate customers who tend to repay from those whose credits would turn delinquent. Among these factors are customers’ basic profiles, records from Joint Credit Information Center and transaction behavior. This study also discovered factors in Behavior Model that can distinguish between those who tend to pay off and those who would fail to follow through their obligations. Among those factors are customers’ transaction behavior, payment records, use of credit lines, balance of revolving credits, and classification of consumption behavior, etc.

This study measures customer by 26 grades. The higher the grade, the greater the likelihood of default in the coming one year. Future studies may further develop Loss Given Default Model and Exposure at Default Model. Based on expected loss and minimum capital requirement, loans can be priced more accurately. Various marketing strategies can be targeted at different groups of customers, thus fomenting loyalty among customers. Through the understanding of the relationship between factors and risk in credit cards, banks can identify less risky customer groups, therefore making more profits.

1. 王濟川、郭志剛(2003)：Logistic迴歸模型：方法與應用，台北：五南出版。
2. 江世傑(2001)：利用模糊類神經網路系統評估消費性貸款，成功大學企業管理研究所碩士論文。
3. 江淑娟(2003)：信用評等因素與信用卡違約風險之關係－以台灣Ａ金融機構所發行之信用卡為例，逢甲大學保險所碩士論文。
4. 呂美慧(2000)：銀行授信評等模式—Logistic.Regression之應用，政治大學金融研究所碩士論文。
5. 金融監督管理委員會(2007)：銀行自有資本與風險性資產之計算方法說明及表格。
6. 施孟隆(1999)：Logit.Model應用於信用卡信用風險審核之研究，財務金融月刊十月號。
7. 馬芳資(1994)：信用卡信用風險預警範例學習系統之研究，政治大學資訊管理所碩士論文。
8. 張仁哲(1982)：我國信用卡現代化問題之研究，政治大學企業管理研究所碩士論文。
9. 梁德馨、黃高鴻(2007)：小額信用貸款違約風險評分評等模型之建構－依循新巴塞爾資本協定零售型暴險內部評等法之規範，風險管理學報第九卷第二期。
10. 郭銘輝(1985)：個人信用評等制度，財團法人金融人員研究訓練中心經營管理研究班講義。
11. 曾俊堯(1991)：信用卡信用管理之研究，政治大學企業管理研究所碩士論文。
12. 曾國烈(2006)：檢視消費金融的需求及債務處理措施－協助雙卡債務人，台灣智庫我國消費金融的發展、問題與對策研討會會議資料。
13. 葉秋南(1996)：美國消費者信用管理，台北：財團法人金融聯合徵信中心編輯委員會。
14. 銀行公會(2007)：銀行風險管理實務範本－信用風險管理分論及案例彙編。
15. 簡安泰(1971)：消費者信用評分制度之研究，政治大學企業管理研究所碩士論文。
1. Rock.A.(1984)：”Sure.Ways.to.Score.With.Lender”，Money.
2. Steenacker.A..&.Goovaerts.M.J.(1989)：”A.credit.scoring.model.for.personal.loans”，Insurance.Mathematics.Economics,6,p31-p34.
3. Updegrave(1987)，”How.Lender.Size.You.Up”，Money.