基于Logistic和决策树模型的P2P网络借贷信用风险研究——以HLCT为例  

Research on Credit Risk in Online Peer-to-Peer Lending Based on Logistic and Decision Tree Model——The Case of HLCT

作　　者：王文怡 程平[1] WANG Wenyi;CHENG Ping

机构地区：[1]重庆理工大学,重庆400054

出　　处：《上海立信会计金融学院学报》2018年第3期42-55,共14页Journal of Shanghai Lixin University of Accounting and Finance

基　　金：国家社会科学基金项目"财务共享服务中心的功能定位与实现路径研究"(17BGL194)

摘　　要：本文以具有代表性的HLCT平台为例,通过Python网络爬虫所采集的真实交易数据,从借款人的信用等级、借款信息以及历史表现三方面出发,综合运用Logistic和ID3决策树模型对借款人的信息与其违约信用风险的关系进行实证探究。研究结果表明,决策树模型整体上要优于Logistic回归的判别,对违约样本的识别准确率约达87%。在影响因素中,借款额度、贷款额度、评级得分以及按时还款笔数四个变量是影响借款人是否逾期还款的关键指标,且均存在显著的负相关关系。HLCT可根据分类结果设立平台阈值,当借款人的状态进入上述几个风险警报区域之内时,可采取风险防范措施,减少信用风险发生的概率。Taking an example of HLCT, which is a typical online P2 P platform, the author uses the internal d ata acquired by Python web crawler from the perspective of credit rating, loan information and history performance to empirically analyze the relevance between these factors and borrower default behavior by means of logistic and ID3 decision tree model. The research shows that discriminant result of decision tree model is superior to that of the logistic regression overall, which can correctly distinguish approximately 87 percent of the default behavior. The borrower’s borrowing amount,lending amount, credit score and the number of punctual repayment are the vital four factors which can influence his/her overdue repayment rate and all have a significant negative relationship. HLCT can set some threshold value according to the situation of its own platform. When the borrower’s state enters the risk warning area of those indexes mentioned above, the platform can take some risk prevention measures to alleviate the occurrence of credit risk.

关 键 词：P2P网络借贷 信用风险 LOGISTIC模型 决策树模型 

分 类 号：F832.4[经济管理—金融学]