本研究運用二階段變數精煉方法,將關聯性分析、決策樹與羅吉斯迴歸篩選出對房屋貸款提前清償之重要變數,代入存活分析中常用之比例危險模型,建構出提前清償預測模型。結果顯示:本研究模型之提前清償整體預測正確率達78.69%。依據本模式之預測,能有效預測提前清償發生之時點,在未發生提前清償之顧客中,有20.14%未來兩年可能會提前清償。為避免客戶提前清償而產生的利息損失及客戶流失,針對此目標主群,一方面要加強服務品質,另一方面可針對其需求提供相關服務或轉投資以維繫舊客戶。透過本研究的模式,亦可發現對提前清償具有影響之變數包含:調息方式、房屋座落位置、鑑價標準、利率基準、遲延紀錄、貸款溢價、罰則費率、月付金、限制提前清償期限。
This study constructs the mortgage prepayment predicted model by Proportional Hazard Model (PHM) which is popularly used in survival research. Before constructing the predicted model, the Contingency Table, Decision Tree and Logistic Regression are used to select the significantly influenced variables of the prepayment. The result shows: the accuracy rate of the stepped prepayment model is 78.69%, and it segments 20.14% clients of prepayment in the future two years from the clients of non-prepayment. There are some kinds of preventive measures to these targets. The measures include the communication with the customers and the introduction of other products. The result of this study shows that the significantly influenced variables of the prepayment include as follows: adjustable rate style, building address, the standard of appraisal, benchmark of interest rate, delay record, mortgage premium, (prepayment) penalty rate, installment amt., the deadline of abridge prepayment.