非参数下贝叶斯判别与聚类分析的信用指标筛选模型  

Credit Index Screening Model of Nonparametric Bayesian Discriminant and Cluster Analysis

作　　者：刘雨萌 李战江[1] 尹伟 Liu Yumeng;Li Zhanjiang;Yin Wei(College of Economics and Management,Inner Mongolia Agricultural University;Inner Mongolia Banking Regulatory Bureau,Huhhot 010010,China)

机构地区：[1]内蒙古农业大学经济管理学院 [2]内蒙古银监局,呼和浩特010010

出　　处：《统计与决策》2018年第22期5-10,共6页Statistics & Decision

基　　金：国家自然科学基金重点项目(71731003);中国博士后科学基金资助项目(2015M582746XB);内蒙古自治区自然科学基金资助项目(2016MS0714)

摘　　要：文章通过非参数贝叶斯判别对所有信用指标进行第一轮筛选,通过非参数聚类方法对保留的信用指标进行第二轮筛选,提供了一套指标分布未知下筛选信用评价指标的非参数方法,并以某商业银行的860个企业信贷数据为样本进行了应用分析。其要点一是通过非参数核密度分布函数构建违约客户与非违约客户的二分类贝叶斯判别模型,删除判别精度影响度大于等于0的信用指标,保留判别精度影响度小于0的信用指标。二是通过非参数聚类将保留的指标聚为19类,在聚为一类的指标中保留判别精度影响度比重最大的信用指标,最终构建具有显著信用判别能力且信息不重复的指标体系。实证结果表明,最终构建的20个指标的企业信用评价指标体系符合5C要素模型。最终构建指标的判别精度高于全部指标判别精度3个百分点。This paper screens the first round of all credit indexes by nonparametric Bayesian discriminant,and screens the second round of retained credit indexes by nonparametric clustering method.The paper also provides a set of nonparametric methods to screen credit evaluation indexes under unknown index distribution,and analyses a sample of 860 corporate credit data of a commercial bank.The main points are as follows：Firstly,the Bayesian discriminant model of the default and non-default customers is constructed by using the nonparametric kernel density distribution function to delete the credit index with the influence degree greater than or equal to 0 and retain the credit index with the influence degree less than 0.Secondly,the retained indexes are classified into 19 categories by non-parametric clustering.In each category of index clustering,the credit index with the largest proportion of discrimination accuracy influence is maintained so as to finally construct the index system with significant credit ability and information non-duplications.The empirical results show that the credit evaluation index system with 20 indexes conforms to the 5 C factor model,and the discrimination accuracy of the final constructed indexes is 3 percentage points higher than that of all indexes.

关 键 词：信用评价 指标筛选 非参数贝叶斯判别 非参数聚类 

分 类 号：F832.42[经济管理—金融学] O212.7[理学—概率论与数理统计]