I 摘要 近几年消费金融的不断兴起和蓬勃发展，为不同阶层的客户提供了小额资金， 有效刺激了消费市场发展。但是由于信息不对称现象的存在，以及受到国内缺乏 健全个人征信体系的影响，使得借款人必然需要面对信用风险问题，导致投资人 的权益受损，加重了消费金融公司的负担。越来越多的消费金融公司选择与第三 方科技公司合作来进行风控筛查，例如芝麻信用分、同盾分、蜜源等，第三方科 技公司通过多种信息来源交叉验证客户信用情况并进行评分，较好的补充了人民 银行征信体系的不完整。充分利用第三方数据后，哪一种评分方法能够更好的识 别客户的违约情况，进行降低违约风险，是本文需要探讨的话题。 根据上述问题，本文拟以借款人的不同信息指标为对象，运用定性分析的方 式进行对比与分类，找出信用风险受哪些信息指标所影响，同时利用消费金融常 用的五种评估方法对违约情况进行评估，并对误判情况进行总结，为后续的分析 提供具有创新性的、且较为全面的评价指标体系。 按照以上思路为依据，拟将论文分为三部分：首先，介绍消费金融及其风险 评分发展现状；然后，以借款人发生违约与否作为被解释变量，选取客户行业、 收入、学历、所在城市、期限、利率、性别、婚姻状况、有无自有房产、年龄、 工作年限、贷款金额、受理渠道和芝麻信用分共 14 个指标作为解释变量，先用这 14 项指标进行描述性统计分析，结果显示，年龄在 26-35 岁，初中及以下学历、 男性、离异、职业是商业贸易类、工作期限在 5-10 年，无房且芝麻信用分较低的 客户违约风险最高；最后，通过机器学习方法筛选出 2513 个样本数据，建立逻辑 回归、随机森林、支持向量机、人工神经网络、贝叶斯分类器 5 种信用分析常用 评估模型对违约风险进行评估与比较，通过综合准确率、违约贷款准确率以及模 型误判来判定模型好坏，并用 AUC 进行模型检验，判定可信度。经分析发现，在 对借款人的信用风险进行评价方面，运用随机森林模型能够实现准确率的提升， 而支持向量机模型能够较好的识别借款人的违约风险及误判情况，对消费金融的 风险评价和管理具有一定的现实意义。 关键词：消费金融，违约风险，风控，模型ABSTRACT II ABSTRACT Withthe continuous rise and vigorous development of consumer finance in recent years, micro-credit has been provided to borrowers from different social classes, which has effectively stimulated the growth of the consumptive market. However, due to the asymmetry of information and the lack of a sound credit system, the credit risk of borrowers is inevitable, which may infringe stakeholder’s rights and increase the burden of consumer finance companies.Thus,anincreasing number of consumer finance companies choose to cooperate with third-party technology companies to conduct risk assessment,such as Zhima Credit, Tongdun Credit,Miyuan Credit,etc.Third-party technology companies would cross-check borrowers’credit through multiple information sources, and assess their credit status.The method is a good complement to the credit system of the People's Bank of China.With the support of third-party data, the main question of this dissertation is which assessment method can better identify the defaults of borrowers and diminish the risk. Based on the question, this dissertation focuses on different indicators of borrowers, and adopts qualitative analysis to compare and classify these indicators, and to identify how credit risk is influenced by these indicators. This dissertation potentially provides an innovative and comprehensive assessment methodology for future qualitative analysis. This dissertation is divided into three sections. First, consumer finance and current consumer credit assessment method would be reviewed in section one. Second, in section two, default is selected as an independent variable, and 14 dependent variables are selected, including the customer’s job, income, education, city, loan term, interest rate, gender, marital status, owned property, age, lengthofemployment, loan amount, receiving channels, and Zhima Credit. A descriptive statistical analysis is adopted with the 14 indicators, and the results show that divorced male borrowers aged between 26-35,with junior high school education or below, working in business or trade industry for 5-10 years, and with no real estate and low Zhima Credit have the highest risk of default. Third, in section three, a machine learning method is used to screen 2513 sample data, and five often used assessment models - Logistic Regression,Random Forest Model,Support Vector Machine,Artificial Aeural Network,and Naive BayesianABSTRACT III algorithm,are used to evaluate and compare the risk of default.General accuracy rate, loan default accuracy rate and model misjudgment are used to verify the model, and AUC is used to test the model’s accuracy. The dissertation has found that the random forest model could increase the accuracy rate in the credit assessment of borrowers, and the support vector machine model can better identify the risk of default and misjudgment, providing empirical evidence for credit assessment and management. Keywords: Consumer Finance, Default Risk, Risk Control, Model目录 IV 目录 第一章 绪论.................................................................................................................... 1 1.1 研究背景与意义................................................................................................ 1 1.2 国内外消费金融研究综述................................................................................ 2 1.2.1 消费金融相关概念综述........................................................................ 2 1.2.2 消费金融违约风险成因综述................................................................ 3 1.2.3 消费金融信用风险控制综述................................................................ 4 1.2.4 国内外研究总结.................................................................................... 5 1.3 本文主要研究思路与内容................................................................................ 6 第二章 消费金融及其风险评估发展现状.................................................................... 8 2.1 消费金融发展现状............................................................................................ 8 2.1.1 消费金融业务概况................................................................................ 8 2.1.2 消费金融市场参与主体分析................................................................ 9 2.1.3 消费金融业务规模及行业情况.......................................................... 10 2.1.4 消费金融的科技运用及现状...............................................................11 2.2 消费金融信用评分体系发展现状.................................................................. 13 2.2.1 信用评分系统发展历程...................................................................... 13 2.2.2 信用评分体系发展现状...................................................................... 14 2.2.3 新兴互联网评分机构简介.................................................................. 15 2.2.4 传统征信与互联网评分机构的比较.................................................. 17 2.2.5 征信行业机构介绍——以芝麻信用为例.......................................... 18 2.3 消费金融信用风险评分方法现状.................................................................. 19 第三章 数据来源与描述性统计分析.......................................................................... 22 3.1ZYS 消费金融公司概况及业务发展情况...................................................... 22 3.2 数据来源.......................................................................................................... 25 3.3 基于 ZYS 公司样本数据的借款人信息指标性分析.................................... 25 3.3.1 借款人基本信息分析.......................................................................... 25 3.3.2 借款人贷款信息分析.......................................................................... 27 3.3.3 借款人工作情况分析.......................................................................... 29 3.3.4 借款人资产负债水平分析.................................................................. 30 3.3.5 借款人受理渠道分析.......................................................................... 31目录 V 3.3.6 芝麻信用分分析.................................................................................. 31 3.3.7 本章小结....................