摘要
识别手机分期消费贷款违约因子是防范手机消费贷款业务信用风险的关键所在。为此,基于融合随机森林(RF)和逻辑回归(Logistics)两阶段模型,通过数据挖掘揭示风险特征重要性含义,并结合经济计量方法诠释异质性客户信用违约的基准逻辑。结果表明:入网时长、终端个数、客户月流量、终端时长是影响手机分期消费贷款客户信用风险的重要性特征变量,且边际影响分别为-0.039%、3.18%、-0.01%、-1.06%,模型泛化能力强,准确率达到74%。所以,要完善手机分期消费贷款信用风险管理应从交叉数据获取、社交网络、兴趣热点和消费习惯等方面着手。
Identifying default factors of mobile phone installment consumer loan is the key to prevent credit risk of consumer loan business. This paper proposes a two-stage model which combines stochastic forest(RF) and logistic regression. The model reveals the importance of risk characteristics from the perspective of data mining, and interprets the benchmark logic of heterogeneous customer credit default by combining measurement methods. The results show that the length of network access, the number of contacts, the monthly flow of customers, and the length of terminal are the important characteristic variables affecting the credit risk of mobile phone installment consumer loans, and the marginal impact is -0.039%, 3.18%, 0.01% and -1.06%. The model has strong generalization ability and accuracy rate of 74%. It provides decision support for the construction of individual credit risk early warning system step by step.
作者
龙海明
邹汉铮
朱建
LONG Haiming;ZOU Hanzheng;ZHU Jian(School of Finance and Statistics, Hunan University, Changsha, Hunan 410079, China)
出处
《财经理论与实践》
CSSCI
北大核心
2019年第5期27-33,共7页
The Theory and Practice of Finance and Economics
基金
湖南省社会科学评委项目(XSP18XBZ065)
国家社会科学基金项目(17FJY013)
关键词
信用风险
随机森林
变量重要性
逻辑回归
Credit risk
Random forest
Importance of variables
Logistic regression
