- 學位論文
信用卡使用特性對未來消費靜止行為之預測
Predicting Dormant Accounts through Credit Card Holders’ Consumption Behavior.
摘要
當市場愈來愈競爭，商品愈來愈龐雜，資訊愈來愈氾濫的時代下，企業想取得一位新顧客已變得比過去更為困難，且新顧客所能帶來的貢獻也往往不如企業的預期，故行銷人員們開始關注在顧客的保留，找出可能流失的顧客，並給予有效的行銷刺激。 為了找出正確目標，本研究以某知名銀行之信用卡資料庫建立持卡戶的靜止預測模型，採羅吉斯及普羅比迴歸分析方式，顧客未來短中長期是否靜止為依變數，帶入各項變數，來預測顧客的靜止行為，並研究各變數與靜止行為的相關性，以推論出長短期靜止顧客的消費特性。 本研究設立了幾種指標變數以加強對顧客行為的描述，「活躍性指標」描述顧客刷卡區間是否愈趨縮短；「穩定性指標」描述顧客消費次數相較於群體的差異；「危險率」考量顧客的最近購買期間及過去購買行為後在當下購買的機率；及「強弱比」顧客在此銀行的消費金額相較於他行的比率。 而根據研究結果顯示，本研究所設立的靜止預測模型，其預測能力確實較銀行過去的判斷方式更為有效。且所建立的各項指標也具有相當的顯著力，並顯示長短期靜止顧客的特性。行銷人員未來可藉這些指標或預測模型，作為靜止的警訊，以增加行銷活動，強化顧客與銀行的往來關係。
並列摘要
Under the increasingly competitive marketplace, especially consumers being affected by numerous products and information floods, it’s extremely difficult for enterprises to acquire new customers. Moreover, the earnings generated by those of whom are not as many as they expected. As a result, the marketing specialists are then turning their attention to customer retention, i.e., they identify lost customers and try to let them come back by providing effective marketing incentives. The object of this study is to use the credit card database to construct a prediction model for dormant accounts. We used Logistic and Probit regression models as analyzing tools, dormant status in short term and long term as independent variable, and consumption behavior as dependent variables to predict the probability to dormant. And we also analyzed correlation and significant level of every dependent variable to find out the difference between short term dormant and long term dormant. In this study, we established four indexes to describe customers’ behavior. Customer Active Index could tell whether the interval of consumptions is getting more frequent periods; Customer Reliability Index could tell whether the numbers of consumption is strongly different from the group; Hazard Ratio is the probability that the tested party shop at the moment considering his nearest purchasing period and previous patterns; and Strength ratio is the same as share of wallet. In conclusion, the dormant prediction model brings better result than the traditional way the bank used. Furthermore, those four indexes we created were strongly significant, which could explain the character of short-term and long-term dormant users. Based on this study, the marketing specialists could use these indexes and model as alarm to predict dormant accounts and then take action in advance to consolidate their relationship further.