基于Logit与SVM的银行业信用风险预警模型研究  

Study on credit risk early warning based on Logit and SVM

作　　者：张奇[1,2] 胡蓝艺[1] 王珏[1,2] 

机构地区：[1]中国科学院数学与系统科学研究院预测科学研究中心,北京100190 [2]中国科学院国家数学与交叉科学中心数学与经济金融交叉研究部,北京100190

出　　处：《系统工程理论与实践》2015年第7期1784-1790,共7页Systems Engineering-Theory & Practice

基　　金：国家自然科学基金(71271202);国家数学与交叉科学中心全球经济监测预警与政策模拟仿真项目

摘　　要：本文以银行业金融机构大额授信风险及零售贷款违约风险数据为基础,从宏观经济环境、客户信贷行为、企业经营水平三个维度出发,对客户风险预警的相关指标进行系统分析,构建了企业客户风险预警指标体系,并利用统计学和数据挖掘方法,从企业财务、企业信贷行为等客户数据信息中挖掘出隐含在背后的客户风险特征.在上述分析的基础上,引入一种基于Logit与SVM的混合预警模型.该模型除了具有单个模型的良好基本性质,还能够充分捕捉和有效刻画影响因素对于客户违约的线性和非线性的复杂特征.实证结果表明,新的模型具有更好的泛化能力,对客户信贷风险具有较高的预警准确率.In this study, based on the data of large credit risk and retail loan default risk of banking and financial institutions, we conducted a systematic analysis of indicators related to client risk early warning. Considering macroeconomic environment, customer credit behavior and enterprise management level, the indicator system of enterprise customer risk warning is established. Meanwhile, with the use of statistical methods as well as data mining skills, we find out the characteristics of customer risk implied in customer data concerning enterprise finance, credit behavior, and so on. According to the above analysis, a hybrid early warning model based on Logit and SVM is proposed, which has good basic properties of a single model and can effectively describe the linear and non-linear features of customer default influenced by different factors. Finally, the empirical results indicate that the new model has more generalization ability and higher accuracy of the credit risk early warning.

关 键 词：信用风险 LOGIT SVM 混合预警模型 

分 类 号：F832.33[经济管理—金融学]