《保险研究》20211007-《基于舆情的信用风险预警模型》（苏罡、余尚兵、李凡）

［摘   要］近几年,债券市场信用风险事件频发,传统基于财务的信用风险评估模型数据更新频率低,难以及时反应发债主体信用变化。随着人工智能技术的发展,金融中可利用的另类数据越来越多,如用自然语言处理技术对新闻进行处理形成的标签数据。本文利用新闻的标签数据对新闻负面程度进行打分,通过对某主体过去几年负面新闻得分进行分析得出新的统计特征,再借助人工智能技术对违约主体和非违约主体过去几年负面新闻统计特征的训练,得到基于新闻舆情的信用风险预警模型,最后利用训练的模型对样本外的发债主体进行违约概率预测,以达到信用风险预警的目的。本研究发现,该模型能够及时对信用违约风险进行预警,能有效提升信用风险管理水平。

［关键词］信用风险；新闻舆情；机器学习；不平衡数据分类

［作者简介］苏罡,中国太平洋保险(集团)股份有限公司拟任首席投资官,金融学博士,高级经济师；余尚兵,中国太平洋保险(集团)股份有限公司博士后科研工作站；李凡,长江养老保险股份有限公司资产管理与监督部。

Credit Risk Forecasting Model Based-on Corporate News

SU Gang,YU Shang-bing,LI Fan

Abstract：In recent years,there have been frequent credit risk events in the bond market,and the traditional financial-based credit risk assessment model has a low data update frequency,making it difficult to respond to changes in the credit risk of bond issuers in a timely manner.With the development of artificial intelligence technology,more and more alternative data can be used in finance,such as label data formed by processing news with natural language processing technology.We used the label data to score the negative news,and then analyzed the scores of issuers in the past few years to obtain new statistical characteristics,and then used artificial intelligence technology to analyze the negative news of both defaulted and undefaulted issuers.The statistical characteristics were trained to obtain a credit risk forecasting model based on corporate news.Finally,the trained model was used to predict the default probability of out-of-sample issuers.We found that this model could provide forecast of credit risks in time and improve credit risk management.

Key words：credit risks; corporate news; machine learning; classification on imbalanced data