基于Logistic回归模型的上市公司财务报告舞弊识别研究

Research on the Financial Report Fraud Detection of Listed Companies based on Logistic Regression Model

上市公司财务报告舞弊阻碍了资本市场的健康发展,破坏了资本市场的资源配置,也损害了投资者的利益,一直是困扰国内外资本市场发展的一个难题。如何有效识别上市公司财务报告舞弊行为成为广大投资者和证券监管部门重点关注的课题。本文为了能够发现财务报告舞弊的规律,选取了2004至2013年因财务报告舞弊被证监会公开行政处罚的44家上市公司作为样本,分析可能对财务报告舞弊敏感的27个指标。然后对应44家舞弊公司选择44家财务报表正常的公司作为对比,对两组样本的指标数据进行显著性检验,并建立向后逐步法的逻辑回归模型作为财务报告舞弊识别模型。研究结果表明,27个指标中11个指标存在显著差异,得出具有折旧率变高、股权比较分散、应收账款周转天数变长等特征的上市公司更有可能进行财务舞弊。最后,本文再选取2004年到2008年间的15家舞弊上市公司和30家对比上市公司作为检验样本,检验模型的识别正确率。

Listed companies' financial reporting fraud has been a difficult problem plagued domestic capital market development,which not only hinder the healthy development of the capital market,undermining the resource allocation of capital markets,but also harm the interests of investors.How to effectively identify financial fraud of listed companies becomes the subject that investors and securities regulators focus on.44 listed companies are chosen in this paper which are punished by China Security Regulatory Commission from 2004 to 2013 for their financial fraud to find their principles.27 financial indicators which are sensitive to financial fraud are selected.To research the impact of the 27 indicators 44 normal listed companies never occurred fraud are selected as a comparison sample.After carry significant test and non-parametric tests in two samples of indicator data and backward stepwise logistic regression model we found that11 indicators have significant differences in normal movements of the index's fraud companies and depreciation rates,changes in earnings quality index ownership concentration and other indicators.Finally,the 15 fraud listed companies and 15 normal listed companies from 2004 to 2008 are selected as recognition accuracy test samples to test the model.