銀行業者為審核信用卡顧客之信用品質，常藉信用評等與行為評等建構其風險管理機制，其中對於顧客之後續持卡行為管理卻相形困難，往往待顧客已發生逾期違約之示警始行催收作業，徒增利益與成本之耗費。故本研究試圖建構顧客行為評等模式，期能掌握顧客未來還款行為，並精確預防不良壞帳產生以達預警之效，且可因人制宜設計策略性的顧客管理，為企業掌握顧客資訊與創造利潤最大化。 有鑑於此，本研究藉由國內某家發卡銀行之顧客資料進行實證分析，試圖整合二階段研究方法，先以卡方自動互動檢視與類神經網路建構第一階段之分類模式，將顧客分為正常繳款、循環納息與逾期違約等三類，進而尋得最佳預測分類結果；第二階段則援引資料包絡分析法，不僅可強化上階段所預測顧客類別之分類正確性，更有別於過去資料包絡分析法多用於評估整體經營績效，本研究則應用於探討個人行為評等，建構個人績效評估，試圖降低銀行對顧客信用評等判斷錯誤之機會成本，且針對未達效率之顧客提供目標改進方案，進而提升信用持卡人之高效率與貢獻度。 實證結果顯示，藉由資料包絡分析法可輔佐分類模式之鑑別結果，將上階段分類後之同質性顧客再予以異質化，藉以發掘具有能力成為潛在利潤或低風險顧客之額外訊息，此外又可利用資料包絡分析法之效率分析以助銀行業者強化個別顧客之信用風險管理，並可制定其相關決策方案，期能提供予銀行業者較為具體的改良依據，有效提升整體之信用品質與管理意涵。 Constructing a risk management mechanism with credit scoring and behavioral scoring has turned out to be the prerequisite of banks to evaluate customer’s credit quality. However, behavioral scoring has long been recognized as a difficult task because its complexity of delinquency prediction. Accordingly, this study attempts to propose a two-stage method that integrates ANNs and DEA to assess individual cardholder’s behavioral scoring. Generally, DEA is applied to evaluate enterprises performance appraisal as a common method; nevertheless, it evaluates the behavioral scoring in this study. This empirical study shows that the DEA efficiency analysis not only provides improvement suggestions for inefficient DMUs, but also identifies the ANNs’ classification results as protective measures to ensure weather the prediction is accurate or not. The former information can enhance customers’ contribution and the later is helpful to observe the potential profit customer for banks. Consequently, this proposed two-stage method is applicable to banks for the real world in terms of its classification accuracy guidance and managerial implication.
