摘要
以2008—2019年制造业上市企业为研究样本,爬取年度财务报告与互联网股民评论文本数据,建立包含风险透明度和股民信心的上市公司信用评价指标体系,并利用随机森林、AdaBoost和Bagging三种集成学习模型构建评价模型。研究发现,年度财务报告与股民评论文本中所蕴涵的文本数据对信用评价具有有效性,且随机森林具有更优的预测效果与模型泛化能力。
This paper selected the listed manufacturing enterprises from 2008 to 2019 as research samples,and crawled annual financial report and internet shareholders'comments text data.Then we established the credit evaluation index system,including risk transparency and shareholders'confidence,and constructed the evaluation model using three integrated learning models,that is,random forest,AdaBoost and Bagging.The results showed that,the text data contained in annual financial report and shareholder review text were effective for credit evaluation,and random forest had better prediction effect and model generalization ability.
作者
韩嵩
李晓俊
HAN Song;LI Xiao-jun(Information Institute,Beijing Wuzi University,Beijing 101149,China)
出处
《统计学报》
2021年第5期72-81,共10页
Journal of Statistics
基金
北京市社会科学基金项目(19YJC033)。
关键词
互联网文本大数据
信用评价
风险透明度
股民信心
集成学习模型
internet text big data
credit evaluation
risk transparency
shareholder confidence
integrated learning model
