摘要
良好的财务状况是企业可持续发展的基础,但由于制度不完善、市场结构缺陷、管理水平较低等原因,我国上市公司财务欺诈屡禁不止,本文旨在使用新闻媒体对上市公司财务欺诈的报道等文本数据识别欺诈行为,并创新性地引入涉诉相关、监管问询、内部控制与风险管理等观测指标,探讨中国上市公司财务欺诈识别问题.研究表明,企业涉诉次数和监管问询频次与财务欺诈行为呈正相关;当企业内部控制有效并采取较好的风险应对时,其财务欺诈行为减少.本研究创新性地分析了年度净利润指标在欺诈分析中的临界值,当年度净利润小于1500万元时,欺诈的可能性大幅提高.最后,本文使用抽样技术开发不平衡的机器学习模型继续检验上述指标,基于Cluster Centroid抽样技术的随机森林模型能够准确识别出98%的财务欺诈样本.本文结合计量分析和机器学习模型,为上市公司财务欺诈识别提供了大量新的实证证据,扩充了指标体系,创新了研究视角.
A solid financial condition serves as the foundation for the sustainable development of enterprises.However,due to imperfect systems,structural deficiencies in the market,and relatively low managerial proficiency,financial fraud remains rampant among listed companies in China.This paper aims to identify fraudulent behavior in listed companies’financial statements using textual data from news media disclosure.Additionally,it innovatively introduces fraud indicators incluidng litigation involvement,regulatory inquiries,internal controls,and risk management to explore the issue of identifying financial fraud in China.Research indicates a positive correlation between the frequency of litigation involvement and regulatory inquiries with financial fraud.Furthermore,when companies have effective internal controls and adopt robust risk management measures,instances of financial fraud decrease.This study innovatively analyzes the threshold value of annual net profit indicators in fraud analysis,suggesting that when annual net profit is less than 15 million RMB,the likelihood of fraud significantly increases.Finally,this paper employs resampling techniques to develop imbalanced machine learning models to further test these new-introduced fraud indicators.The empirical results find that random forest model integrated with Cluster Centroid sampling technique can accurately identify the highest 98%of financial fraud samples.By integrating econometric analysis and machine learning models,this paper provides abundant new empirical evidence for identifying financial fraud in listed companies,expands the indicator system,and innovates research perspectives.
作者
马溪远
刘尚超
吴德胜
MA Xiyuan;LIU Shangchao;WU Desheng(School of Economics and Management,University of Chinese Academy of Sciences,Beijing 100190,China)
出处
《计量经济学报》
CSSCI
CSCD
2024年第3期699-726,共28页
China Journal of Econometrics
基金
科技部重大项目(2020AAA0108400)
国家自然科学基金(71825007,72210107001)。
关键词
财务欺诈识别
欺诈指标创新
计量分析
机器学习
文本数据
financial fraud detection
innovative fraud indicators
econometric analysis
machine learning
textual data
作者简介
马溪远,博士研究生,研究方向:财务风险管理,E-mail:maxiyuan21@mails.ucas.ac.cn;刘尚超,博士,研究方向:城市与房地产,E-mail:liycsl@163.com;通信作者:吴德胜,博士,中国科学院大学经济与管理学院教授,研究方向:金融风险管理、智能决策、环境经济,E-mail:dwu@ucas.ac.cn.
