| 摘要: | 本研究係以嘉義市某商業銀行從民國87年至96年止已貸放之房貸授信案件中，經隨機抽樣311件為研究樣本，其中285件為正常戶，26件為逾期戶。透過文獻探討暨實務經驗選取的變數合計16個變數(性別、年齡、婚姻狀況、教育程度、職業、職稱、償還方式、保證人、擔保物之所有權人、擔保品地點是否為嘉義市地區、透天住宅或公寓、借款用途、是否有二棟以上之房屋貸款、是否有保證債務、信用卡或現金卡近三個月是否有循環動用餘額、三個月內是否有被同業查詢紀錄)納入羅吉斯迴歸模式中，並設為LR1，另將該16個變數進行逐步迴歸(Stepwise)來篩選顯著變數，並設為LR2(顯著變數有: 教育程度、職稱、保證人、擔保品地點是否為嘉義市地區、透天住宅或公寓、信用卡或現金卡近三個月是否有循環動用餘額)。 經羅吉斯迴歸分析，得知LR1模式及LR2模式之總分類正確率分別為97.4%及96.5%，且LR1模式正常戶之預測正確率為99.3%，違約戶之預測正確率為76.9%，足見 LR1模式有較佳之預測能力，是以選擇LR1為最適授信風險評估模式。
The mortgage customers in the period time of the Republic of China 87 years to 96 years of a selected commercial bank in Chiayi City were the subjects for this study, by random sampling for the study 311 samples, of which 285 were normal mortgage customers and 26 for the abnormal ones. The 16 variables through literature review, practical experience selected (gender, age, marital status, education level, occupation, title, repayment, the guarantor, the owner of the collateral, whether collateral place of Chiayi City , town house or apartment, loan purpose, whether two or more housing loans, whether guaranteed debt, credit card or cash card have cycle balance nearly three months,within three months checked by other Financial institutions) are entered into the logistic regression model, and set LR1, then install 16 variables in the stepwise regression (Stepwise) to filter a significant variable, and set LR2 (significant variables are: education, job title, the guarantor, whether collateral place Chiayi City, town house or apartment , redit card or cash card have cycle balance nearly three months). LR1 and LR2 model are processed by The logistic regression,which total classification accuracy rates were 97.4% and 96.5%, and the LR1 model to predict thecorrect rate of normal mortgage customers about 99.3% ,correct prediction rate of abnormal mortgage customers about 76.9%, LR1 model shows better predictioncapability, select the LR1 is the best credit risk assessment model. |
