Identifying Key Factors in Financial Statement Fraud Detection in China’s Stock Market through Logistic Regression and SVM

Abstract

In the complex capital market, since false disclosures can undermine investors’ confidence, the detection of financial statement fraud becomes extremely important. False disclosures can weaken investors’ confidence and affect the operation of the market. Research on the detection of financial fraud in China’s A-share market is relatively weak, especially in terms of regulatory enforcement and market behavior response. Therefore, this study aims to utilize machine learning to investigate the fraudulent behaviors in the financial reports of Chinese listed companies. This paper uses logistic regression (LR) and support vector machine (SVM) to classify and distinguish financial statements. The data comes from CSMAR and Wind databases, while the penalty records are from the China Securities Regulatory Commission. The final research results show that LR outperforms SVM in terms of accuracy and interpretability. The accuracy rate is 96. 8% and the F1 value is 80%. The market reaction analysis further reveals that after the penalty announcement is released, there will be a short-term upward trend in the company’s stock price, but a long-term poor performance. This actually reflects the excessive reaction of investors and that it will take some time for the stock price to correct in the later stage. The results of this study provide empirical evidence for future applications of artificial intelligence in detecting stock market fraud and also contribute to the understanding of the Chinese capital market. They offer some references for regulatory authorities, investors, and enterprises.