本研究係將「負債倍數」、「投資客與否」、「整批房貸戶與否」、「房屋坪數」等四個風險變數,納入傳統房貸信用評分系統中,建立客戶違約之預警模式,以供金融機構做為授信審核之參考,以期能有效降低逾放比率、提升銀行經營績效。研究結果顯示:傳統信用評分模型具有77.5%預測準確率,若進一步將「負債倍數」、「投資客與否」、「整批房貸戶與否」、「房屋坪數」等四個風險變數納入模式,預測準確率提高到89.3%,顯示此四個風險變數在預測違約行為有相當顯著的預測力;其中又以投資客與否的變項最具有預測力,該變數之勝算比為24.493 最高,與授信品質的相關聯程度亦明顯較其他變項來的高,未來銀行可考慮將此四個風險變數納入信用評分系統中。
This study extends the traditional credit scoring system with further four factors, i.e. 「debt ratio」,「real-estate speculator」,「batch processing mortgage loan」and「size of the house」in the logistic regression model to build a dichotomous prediction model which can be adopted by financial institutes to prevent default risk of residential mortgage loan and improve the quality of risky asset. The empirical results indicate that the overall accurate predict rate of this new model is 89.3%, which is significantly higher than the traditional model (77.5%). Among the four factors, the factor of「real-estate speculator」has the highest odds ratio (24.49) indicating that factor of speculator has the most ability to predict default rate.