Abstract:

E-commerce micro and small enterprises have exposed the defect of insufficient risk resistance in the process of Internet financing. Its high dynamism and great influence from subjective expectation of enterprises also pose higher requirements for the constructed credit risk early warning model. In order to further improve the dynamic early warning model of credit risk of e-commerce micro and small enterprises, a study is conducted based on 10 periods of real transaction data of 337 micro and small enterprises in Taobao fresh food industry. First, a subjective and objective two-dimensional index system is established. Then, starting with the mapping principle of early warning, different risk measurement methods are adopted for the state index and time series index in the index system to extract dynamic information. Meanwhile, considering the expectation of decision-makers, the parameters of risk measurement model are set, a dynamic early warning model of credit risk is built, which is compared with the static model. The empirical results show that the overall discrimination accuracy and the early warning effect of the dynamic model are better, and the discrimination accuracy of medium risk enterprises is higher.

Key words: risk measurement, e-commerce micro and small enterprises, status indicators, time series index, random forest

摘要：

电商小微企业在互联网融资的过程中暴露出风险抵抗力不足的缺陷，其动态性高、受企业主观预期影响大的特点，也对所构建的信用风险预警模型提出更高要求。为进一步完善电商小微企业信用风险动态预警模型，基于淘宝生鲜行业337家小微企业的10期真实交易数据展开研究。首先，建立主客观两维度指标体系；其次，从预警的映射原理入手，分别对指标体系中的状态指标和时序指标采用不同的风险度量方法以提取动态信息；同时，考虑决策者的预期期望设置风险度量模型参数，构建信用风险动态预警模型，并与静态模型进行对比。实证结果表明，动态模型总体判别准确率较好、预警效果更好，其对中风险类别的企业判别准确性更高。

关键词: 风险度量, 电商小微企业, 状态指标, 时序指标, 随机森林