基于Logistic模型的上市公司财务造假识别研究

Research on recognition of financial fraud of listed companies based on Logistic model

利用沪深两市受证监会处罚、具有财务造假行为的103家上市公司,以1∶2的比例匹配规模、行业相当的正常上市公司,选取各公司连续3年的财务数据进行研究。先用主成分分析法(PCA)进行特征提取,再进行Logistic回归建模。前两年的样本数据集用于训练模型,最后一年的样本数据集用于检验模型。研究表明:综合偿债因子、综合盈利投资因子、资产运营效益因子、综合现金流量因子、资产销售配比因子5个主成分变量可作为识别财务造假的预警指标,所构建的模型具有较好的预测效果和稳健性,可对上市公司财务造假发挥监测防范作用。

Using 103 listed companies that are punished by the CSRC and have financial fraud behaviors in Shanghai and Shenzhen stock markets,and matching the normal listed companies of the same scale industry with the proportion of 1∶2,this paper selects and studies the financial data of each company for three consecutive years.First,PCA is used for feature extraction,then Logistic regression is used for modeling.The model is trained by the sample data set of the first two years and tested in the sample data set of the next year.The research shows that the five principal component variables of comprehensive debt service factor,comprehensive profit investment factor,asset operation benefit factor,comprehensive cash flow factor and asset sales matching factor can be used as early warning indicators to identify financial fraud,and the model has good prediction effect and robustness,which will play a monitoring and preventive role in financial fraud of listed companies.