上市公司财务违规特征分析及预测研究——基于企业画像和机器学习的经验证据

Research on the Analysis and Prediction of Financial Irregularities of Listed Companies——Empirical evidence based on methods of enterprise portraits and machine learning

为了在接受业务委托、计划审计工作等前期阶段,能够恰当识别由上市公司财务违规带来的审计风险,审计师可选择指标变量、构建预测模型来识别存在违规的上市公司。本文构建了涵盖公司治理、财务状况、经营状况和情感态度的综合指标体系,通过标签建模构造财务违规公司画像,并利用朴素贝叶斯、决策树和随机森林等机器学习算法进行违规识别和预测。结果表明,存在财务违规的公司呈现出审计费用少、股利分配率低、每股收益较小等特征,并在情感态度上存在负面和自我夸大倾向;基于随机森林的预测模型准确率为92.91%,预测效果较好。研究结论表明,通过考虑管理层情感态度、建立可视化画像、应用机器学习预测模型有助于审计师更好地进行审计风险评估。

In order to properly identify audit risks caused by financial irregularities of listed companies in the early stages of audit work,auditors can select variables and build predictive models to identify listed companies with irregularities.This paper constructs a comprehensive indicator system covering corporate governance,financial status,operating status and sentiment attitude,construct the portrait of listed companies with financial irregularities,and achieve anomaly detection through machine learning such as Naive Bayes model,Decision Tree and Random Forest.The results show that companies with financial irregularities exhibit the characteristics of low audit fees,low dividend distribution rates and low earnings per share,and have negative and self-exaggeration tendencies in the sentiment attitude of annual reports;the accuracy rate of the prediction model based on Random Forest is 92.91%,showing a satisfactory prediction effect.By considering the sentiment attitude of management,establishing enterprise portraits,and applying machine learning predictive models,auditors can better conduct audit risk assessment.