| 摘要: | 本研究係以某商業銀行1994~2003年之房屋貸款戶為對象，並將借款人分為正常戶與逾期違約戶，並以性別、年齡、婚姻狀況、教育程度、借戶與貸款行有無地緣關係、職位、服務機關、償還方式、(連帶)保證人之有無、擔保物之所有權人、擔保物與借戶之地緣關係、是否越區承作、貸款成數、是否為購置住宅、貸款型態、貸款期間、核准權限等風險變數來進行探討，並將篩選之變數進行羅吉斯迴歸模式(Logistic Regression Model)分析。 本研究將所有原始變數經過卡方獨立性檢定及交叉分析後選取的變數設定為LR模式Ⅰ，並將選取之變數以逐步迴歸選取法篩選出顯著的變數設定為LR模式Ⅱ，經比較二個模式之整體預測正確率，其中模式Ⅰ之整體預測正確率90.3%略高於模式Ⅱ之90.0%，故以模式Ⅰ為最終授信風險評估模式。
During 1994~2003 the mortgage customers of a selected commercial bank are the subjects for this study. Those mortgage customers are divided into normal accounts and overdue accounts. The variables used for discussion in this study include sex, age, marriage, education, whether a geographic tie exists between the lender and the borrower, position, institution the borrower serves, terms of repayment, whether there is a (severally and jointly liable) surety, the ownership of the collateral, the geographic tie between the collateral and the borrower, whether the case is made beyond the defined region, loan to value ratio, whether a purchase is of a residential dwelling or not, the type of loan, loan period, and authority level of approval. The sieved variables are then analyzed by the Logistic Regression Model. This study assignes the selected variables as LR Model I after analyzing all original variables through the Chi-square test and Crosstabs. In addition, those variables which are found to be significant are assigned as LR Model II after analyzing the selected variables through the statistical Stepwise regression analysis. Model I has the overall predictive accuracy rate of 90.3%, which was slightly higher than the accuracy rate predicted by Model II (90.0%). After comparing the overall predictive accuracy rates between these two models, LR Model I is then chosen as the final evaluation model of credit risk for this study. |
