| Summary: | 碩士 === 國立高雄應用科技大學 === 金融資訊研究所 === 95 === ABSTRACT
This research will investigate the forecast of credit default on mortgage loans of purchasing house by Logistic Regression model. Seven hundred mortgage customers are the samples of this research from X bank. （Performing Loans 500 customers, Non-performing Loans 200 customers）. Each customer would have sixty variables. The empirical research adopts Logistic Regression: Enter method, Forward LR method and Backward LR method, puts 560 training samples in Logistic Regression. Let's choose the significant variable by P＜0.05,under Wald Test， Among them, variables selected by the Enter method, listed as the model Ⅰ；variables selected by the Forward LR method , listed as the modelⅡ; variables selected by the Backward LR method, listed as the model Ⅲ. Then, we used the Regression analysis on these three models orderly, and we got three different Regression equations. Using these three groups of regression equations, we tested 140 test samples individually, and then obtained three groups of different forecast rate. Comparing the forecast accuracy rate of these three groups of models, we found that the model Ⅲ(Backward LR method) had the highest accurate rate 77.86%. Further more, we used Cross-Validation,（the forecast accurate rates of 5 groups on validation samples are
82.14%~-76.43%, average rate is 79.14%）,Goodness of fit-Test and Kolmogorov-Smirnov Test to test the model Ⅲ, and confirmed that it equipped with abilities of forecast and distinction. As the result of these procedures, we confirmed that the model Ⅲ(Backward LR method) is the most suitable model for the forecast of credit default. In this research, we choose 19 variables from Macroecnomic data for forecasting the credit default of mortgage loans of purchasing house. As the result from the research, some variables would have connection with forecasting model of credit default, (For example: the banking basic loans rate, the economical growth rate, the price index, the exchange rate US dollar, the index of Japan stock market, Currency supplies M1A, Currency supplies M1B, Currency supplies M2).Connected with the variables of traditional correlation literature, Macroecnomic data would raise accurate rate of forecasting on Credit Default of purchase house mortgage loans.
Key Words: Mortgage Loans of Purchasing house , Forecast of the Credit Default, Logistic Regression model, Goodness of fit Test, The Kolmogorov
-Smirnov Test.
|
|---|
