查詢結果分析
來源資料
頁籤選單縮合
題 名 | 房屋抵押貸款客戶違約預測模式之比較研究=A Comparative Study of Predictive Models for Mortgage Loan Default |
---|---|
作 者 | 黃嘉興; 謝永明; 劉宗哲; | 書刊名 | 東吳經濟商學學報 |
卷 期 | 48 2005.03[民94.03] |
頁 次 | 頁103-125 |
分類號 | 562.33 |
關鍵詞 | 房屋抵押貸款; 貸款違約; 預測模式; 信用評等; Logistic迴歸; Mortgage loan; Prediction model; Credit rating; Debtors' loan default; Logistic regression; |
語 文 | 中文(Chinese) |
中文摘要 | 本研究以國內某銀行一南部分行的房屋貸款客戶為研究對象,運用logistic迴歸與區別分析兩種統計方法,尋求最佳預測變數組合以建立房屋貸款客戶是否違約之預測模式,並與五種預測模式(四種模式來自過去研究及一種來自樣本銀行的個人信用評估表)進行比較。此外,亦特別蒐集已經認列損失或拍賣抵押房屋之違約房貸戶資料,同時檢驗模式的預測準確度與可降低錯誤分類損失之能力。 實證結果發現,本研究所建構模式在第一階段分析中,其準確率與模式配適度均較其他預測模式為佳。但由於諸多預測變數未符合常態分配,區別分析的結果將受到違反基本假設的限制,故而在後續降低錯誤分類損失能力之檢驗上,僅採用logistic 迴歸進行分析。在第二階段的分析中,本文所建構模式對違約客戶的分類準確率為60.7%。進一步分析後發現,模式違約預測準確率較高者,其所能降低錯誤分類成本(或損失)通常也較大。整體而言,本研究所建立的預測模式未來可作為銀行房屋抵押放款之參考,以協助銀行有效降低其潛在違約損失。 |
英文摘要 | The debtors’ default on mortgage loan is an important issue for Taiwan local banks. It does not only deteriorate the capital structure of banks but also damage their profit margin especially during financial crisis or economic depression. Therefore, a default prediction model for banks to improve the mortgage loan decisions and/or to reduce their potential losses is highly needed. This study tries to establish a better prediction model of the debtors’ default for bank’s mortgage loan decision by searching a better combination of predicting variables. Using two statistical techniques (binary logistic regression and discriminant analysis), this study analyzes the accuracy rate of prediction and the goodness of fit for two models built in this study and five other models (four from past research and one from the“individual credit evaluation form” of sample bank) and then compares their performance with these indicators. In addition, the abilities for all models to minimize misclassification costs are also compared. The result shows that the combination of predicting variables formulated by this study is a great model either in the performance indicators or in the ability to minimize the misclassification costs. |
本系統中英文摘要資訊取自各篇刊載內容。