Analysis of Residuals for Accelerated Failure Time Models：On Mortgage Loan Data

[[abstract]]金融制度自由化使得競爭更加公平與多元化。壽險公司之業務不再只著重於傳統保險業務，更將其資金在規範許可之範圍內推動住宅抵押貸款。從事住宅抵押貸款除了應預防關係著企業體系安全及穩健營運之違約風險外，借款戶之提前清償風險也是應重視之課題，原因乃在於當提前清償發生，將影響公司利息收入之獲利，降低壽險公司預期現金流量之流入。 本論文採用存活分析方法中之加速失敗時間模型〈AFT--accelerated failure time model〉，建立出三種風險預警模式：偉柏模式〈Weibull model〉，對數常態模式〈Log-normal model〉及對數邏輯斯模式〈Log-logistic model〉，再利用殘差分析法：Cox-Snell residuals，Martingale residuals及Deviance residuals來判斷此三種風險預警模式之配適程度，實證結果顯示，對數常態模型〈log-normal model〉之配適最為良好，故建議使用此模型做為預測借款戶提前清償之工具。[[abstract]]The liberalization of financial system makes the competition fair and more plural The business of the life insurance company no longer only focuses on the traditional insurance business they use their fund to promote resident mortgage loan in the allowed range Engaging in resident mortgage loan should prevent the default risk that relates to security and steadiness of the business operation And it should pay attention to the prepayment risk too The prepayment risk is an important subject because when it happens it will influence the profit of the interest income and then will reduce the inflow of the expected cash flow in the life insurance company This thesis adopts the Accelerated Failure Time model of Survival Analysis to set up three risk predicted model: Weibull model Log-normal model and Log-logistic model Then it uses analysis of residuals to judge the good-of-fit of they three The result shows that the log-normal model is the best model so it is proposed to be the predicted tool for prepayment from the borrower