多维数据下小企业违约风险过程性评价研究  

Research on Process Evaluation of Default Risk of Small Enterprises under Multidimensional Data

作　　者：赵志冲 严丽霞 ZHAO Zhi-chong;YAN Li-xia(School of Mangement Science and Engineering,Dongbei University of Finance and Economics,Dalian 116025,China;School of Finance,Jiangxi University of Finance and Economics,Nanchang 330013,China)

机构地区：[1]东北财经大学管理科学与工程学院,辽宁大连116025 [2]江西财经大学金融学院,江西南昌330013

出　　处：《中国管理科学》2023年第5期84-92,共9页Chinese Journal of Management Science

基　　金：国家自然科学基金资助青年项目(71901055);国家自然科学基金资助重点项目(71731003);国家自然科学基金资助面上项目(72071026,71873103,71971051);大连银行小企业信用风险评级系统与贷款定价项目(2012-01)。

摘　　要：违约风险过程性评价是在借贷过程中考虑评价特征维度增加的情况下,对小企业的违约风险进行评价。现有违约风险评价大都是考虑相同的评价特征在某一时刻(某一时间序列)的静态评价(动态评价)。大数据环境下用于评价的特征呈爆炸式持续增长,如果每增加一些特征都需要重新构建评价模型,对银行实际操作而言,是不现实也是不可行的。本研究分别从增加新的准则和增加新的指标两个特征增加的角度,提出信贷过程中,违约风险评价模型构建的四种方式,并以中国某商业银行1994年以来的小企业实际贷款数据为对象,通过建立神经网络模型进行实证研究。研究表明,上述四种方式构建的违约风险评价模型的判别精度无显著性差异,同时方式2(以每个准则独立构建评价模型的评价结果作为输入变量,构建违约风险评价模型)有较高的判别精度,为大数据背景下及时更新违约风险评价模型提供了新的思路,也给银行的实际操作节省运算时间和复杂度。The process evaluation of default risk is to evaluate the default risk of small enterprises in the process of borrowing and considering the increase of evaluation characteristics.The existing default risk evaluation mostly considers the static evaluation(dynamic evaluation)of the same evaluation characteristics at a certain time(a certain time series).In the big data environment,the features used for evaluation are growing explosively.It is unrealistic and infeasible for banks to rebuild the evaluation model for each additional feature.The process evaluation of pedagogy is introduced into the default risk evaluation for the first time.Considering the multi-dimensional data background,from the perspective of adding new criteria and new indicators,four ways are proposed to construct the default risk evaluation structure in the credit process:one is to reconstruct the default risk evaluation model every time an index or a criterion layer is added;the second is to construct the evaluation model independently for each criterion layer,and take the evaluation result of each criterion evaluation model as the input variable to construct the default risk evaluation model;the third is to construct the default risk evaluation model by taking the latest evaluation result and all indicators in the newly added as the input variable;the fourth method is to build a default risk evaluation model by taking"the latestkevaluation results and all indicators in the newly added as input variables.Taking the actual loan data of small enterprises of a commercial bank in China since 1994 as the object,an empirical study is made by establishing a neural network model.The research shows that there is no significant difference in the discrimination accuracy of the default risk evaluation model constructed by the above four methods.Meanwhile,mode 2 has higher discrimination accuracy,which can dynamically update the evaluation model in time under the background of big data,and save operation time and complexity.

关 键 词：多维数据 违约风险 过程性评价 小企业 

分 类 号：F830.56[经济管理—金融学]