以樣本配對觀點探討國內壽險業抵押貸款違約風險之實證分析An Empirical Analysis of Default Risk on Mortgage Loansof Taiwan’s Life I

本研究以國內某知名壽險公司的客戶借貸資料為研究對象，將該原始資料分組成為數個不同正常戶與違約戶之比例的樣本配對建立區別模型，分別以一比一、二比一、三比一、四比一與全體樣本五組，使用邏輯斯迴歸分析、倒傳遞類神經網路與多層函數連結網路，以模型正確率及型一誤差之結果分析模型之區別能力。由實證結果發現，不同樣本配對導致不同的實證結果，其中以一比一樣本配對下，邏輯斯迴歸分析具最佳的區別正確率，而倒傳遞類神經網路具最低的型一誤差。
關鍵字： 抵押貸款、樣本配對、邏輯斯迴歸分析、倒傳遞類神經網路、多層函數連結網路
Abstract
The purpose of this research is to propose a methodology of analyzing the default risk on mortgage loans of Taiwan’s life insurance industry from the perspective of sample pairwise. The original data is divided into 5 groups with a proportion non-default than default individually – 1:1, 2:1, 3:1, 4:1 and 449:53. The models built in this study include Logistic Regression Analysis (LR), Back Propagation Neural Network (BPN) and Multilayer Functional Link Network (MFLN). The empirical results show that different groups of sample pairwise of the data cause different effects, and the models with the proportion 1:1 built by BPN has the lowest Type I Error and LR has the highest discriminate rate.
Keywords: mortgage loans, sample pairwise, Logistic Regression Analysis, Back Propagation Neural Network, Multilayer Functional Link Network