机器学习与会计舞弊治理:基于非谜选因子的预测视角

Machine Learning and Accounting Fraud:A Predictive Perspective Based on Non-selection Factors

本文研究将机器学习引入资本市场监管,优化会计舞弊治理的理论和现实问题。使用1998-2021年会计舞弊数据,采用8种主流机器学习模型,从非述选因子视角评估机器学习遇制会计舞弊的优势和潜力。研究发现,使用机器学习预测会计舞弊,并不依赖事前因子述选,预测效果超过了事前述选会计指标的各种组合,预测指标AUC平均提升12.22%。分析表明这与事前述选指标更容易受到市场针对性的规避行为有关。在此基础上,本文进一步讨论了机器学习优化中国资本市场“双随机、一公开”抽检政策的潜力,认为引入机器学习能够大幅提高会计舞弊检出数量,降低重大舞弊案例的发现阈值,缩短会计舞弊的处罚时滞,从而遇制会计舞弊的扩张。

This paper explores the theoretical and practical implications of applying machine learning to capital market regulation,aiming to enhance the governance of accounting fraud.Using accounting fraud data from China's capital market between 1998 and 2021,the study employs eight mainstream machine learning models to assess the strengths and potential of machine learning in curbing accounting fraud,focusing on a non-selection factor perspective.The study finds that using machine learning to predict accounting fraud does not rely on the pre-selection of factors,with prediction performance surpassing various combinations of pre-selected accounting indicators,showing an average AUC improvement of 12.22%.The analysis suggests that this is related to the fact that pre-selected indicators are more susceptible to targeted avoidance behaviors by the market.A subsequent study discusses the potential of machine learning for optimising the"Dual random selections plus timely release of results"inspection policy in China's capital market.The introduction of machine learning can significantly increase the detection of accounting fraud,lower the threshold for discovering major fraud cases,and shorten the time delay in penalty,thus curbing the spread of accounting fraud.