文章摘要
Liu Yunjing,Wu Bin,Zhang Min.Financial Fraud Recognition Model and Application[J].The Journal of quantitative and technical economics,2022,(7):152-175
上市公司财务舞弊识别模型设计及其应用研究
Financial Fraud Recognition Model and Application
  
DOI:
中文关键词: 财务舞弊  机器学习  非平衡样本  应用分析
英文关键词: Financial Fraud  Machine Learning  Unbalanced Samples  Application Analysis
基金项目:本文获得国家自然科学基金资助项目(72172149)、湖南省哲学社会科学基金资助项目(20JD010)的资助。
Author NameAffiliation
Liu Yunjing School of Business, Renmin University of China 
Wu Bin School of Accountancy, Hunan University of Finance and Economics 
Zhang Min School of Business, Renmin University of China 
Hits:
Download times:
中文摘要:
      研究目标:运用新兴机器学习的方法预测公司财务舞弊。研究方法:选取11类财务比率指标与文本信息、公司治理、内部控制等非财务指标作为初始输入变量,采用欠采样方法处理训练集样本非平衡的问题,选择轻量梯度提升机算法对公司是否发生舞弊建立分类模型。研究发现:采用轻量梯度提升机算法极大地提升了预测准确性;相对于逻辑回归、支持向量机、随机森林、梯度提升决策树,轻量梯度提升机算法的预测效果最好;使用全部输入变量比仅仅使用有限传统变量的预测能力更强;预测模型在案例分析、行业分析和股价崩盘检测中也展现出很好的预测效果。研究创新:引入新的机器学习算法识别财务舞弊,采用欠采样的方法对训练集样本进行平衡处理,从多个角度进行应用分析。研究价值:有助于实时高效地识别舞弊并及时进行监管,实现对经济运行更为准确的监测、分析、预测、预警,从而提升资本市场的治理效能,促进经济平稳运行。
英文摘要:
      Research Objectives: Use emerging machine learning methods to predict corporate financial fraud Research Methods: Select 11 types of financial ratio indicators and non-financial indicators including text information, corporate governance, internal control and other non-financial indicators as the initial input variables, use the under-sampling method to deal with the problem of unbalanced training set samples, and select the lightweight gradient boosting machine algorithm (LightGBM) to establish a classification model to identify fraudulent companies Research Findings:The use of the LightGBM algorithm has greatly improved the accuracy of prediction Compared with logistic regression, support vector machines, random forests, and gradient boosting decision trees, the LightGBM algorithm has the best prediction effect The use of all input variables is more predictive than just limited traditional variables The financial fraud recognition model also shows good predictive performance in case analysis, industry analysis, and stock price crash detection Research Innovations: Introduce new machine learning algorithms to identify financial fraud, use undersampling to balance the training set samples, and conduct application analysises from multiple perspectives Research Value: The paper is helpful to identify financial fraud efficiently and conduct timely supervision, to achieve more accurate monitoring, analysis, forecasting, and early warning of economic operation, thereby enhancing the governance efficiency of the capital market and promoting the smooth operation of the economy
View Full Text   View/Add Comment  Download reader
Close