影響機車分期付款違約行為關鍵因子之研究

Abstract

本研究以個案資融公司機車分期付款案件為樣本，採用羅吉斯迴歸方法對造成該產品的違約的因子進行分析。本研究主要的目的除了分析個案資融公司評分表，瞭解對違約預測力顯著、不顯著的因子外，同時期望透過不同抽樣組合之設計，找出最適抽樣比例並以之建構違約判別模型。實證結果簡述如下： 1. 對違約預測力達中度以上的評分因子有「信用卡使用情形」、「收入及支付能力照會」、「住所穩定度及掌握度照會」三者。 2. 對違約毫無預測力的評分因子有「住宅狀況」、「居住年資」、「收入」、「有無勞保」、「外部信用_巨閎」、「外部信用_中華徵信」、「電話照會結果是否正常」共七者。 3. 最適的建模抽樣比例為「違約:非違約=1:1」。而模型顯著變數有「年齡_20歲以下」、「婚姻狀態_無配偶」、「婚姻狀態_無法判斷」、「教育程度_無法判斷 」、「居住所在地_南部」、「行業別_家管」、「行業別_其他(無資料)」、「信用卡使用狀況_無信用卡資料」、「信用卡使用狀況_有資料且異常」、「同業往來狀況_有資料且異常」、「支付能力照會_收入與支付能力不佳」與「住所穩定度照會_無法判斷」共12個。

This thesis takes 5,000 motorcycle installment contracts of a finance company as study sample. The logistic regression model is constructed to analyze this data to find out which factors have more influence on default. The main purpose of this thesis is to determine whether factors of this company’s credit scoring sheet can forecast probability of default or not. Besides, by using multiple combination of sampling result, we try to build up the best risk model. The empirical results are summarized below: 1.There are three factors having medium default forecast power, which are “Condition of Credit Card Usage”, “Inquiry of Income and Payment Ability”, and “Inquiry of State of Residence”. 2.Some factors in the credit scoring sheet do not have any default forecast power. They are “Residence Condition”, “Years of Residence”, “Income”, “Labor Insurance”, “External Credit Check_ Ju-Hong”, “External Credit Check_ CCIS”, and “Condition of Telephone Inquiry”. 3.The optimal sampling proportion to build risk model is one default sample combined with one non-default sample. The statistically significant factors are “Age_ under 20 years old”, “Marriage Status_ without spouse”, “Marriage Status_ invalid information”, “Level of Education_ invalid information”, “Residence Area_ southern Taiwan”, “Occupation_ housewife”, “Occupation_ invalid information”, “Condition of Credit Card Usage_ invalid information”, “Condition of Credit Card Usage_ bad payment record”, “History with Peers_ bad payment record”, and “Inquiry of State of Residence_ invalid information”.