上市公司财务报表欺诈鉴别
发布时间：2019-01-11 11:34
【摘要】：中国大陆上市公司的财务报表的欺诈行为由来已久,对投资人、债权人以及整个国民经济环境的危害十分严重,但同时对于注册会计师、审计师来说,对欺诈财务报表的鉴别却一直是难题。本文首先根据公开信息选择出财务报表欺诈的风险因子(red flags),建立起财务欺诈合理怀疑指标体系。然后利用中国沪市上市公司的财务报表历史数据训练出财务报表欺诈的预测模型,并对模型的预测效果做出评估。由于欺诈财务报表在总体中的比例很少,所以我们采用不等概率概率抽样,即在欺诈类别样本的抽样概率大于在非欺诈类别样本的抽样概率,在这种情况下传统的参数估计方法需要修正。本文列举了logistic回归的在不等概率抽样条件下进行参数估计的方法,证明了神经网络模型在不等概率抽样条件下修正输出的方法。另外,由于论文的目的是估计财务报表欺诈的可能性,本文还分析的神经网络输出贝叶斯后验概率所需的条件。
[Abstract]:There is a long history of fraud in the financial statements of listed companies in mainland China, which is very harmful to investors, creditors and the national economy as a whole. But at the same time, for certified public accountants and auditors, The identification of fraudulent financial statements has been a problem. This paper firstly selects the risk factor of financial statement fraud according to the public information, (red flags), and establishes the reasonable suspect index system of financial fraud. Then the forecasting model of financial statement fraud is trained by using the historical data of financial statements of listed companies in Shanghai Stock Exchange of China and the forecasting effect of the model is evaluated. Because of the small proportion of fraudulent financial statements in the total, we use unequal probability sampling, that is, the sampling probability in the fraud category is greater than that in the non-fraudulent sample. In this case, the traditional parameter estimation method needs to be modified. In this paper, the method of parameter estimation based on logistic regression under unequal probability sampling condition is listed, and the method of modifying the output of neural network model under unequal probability sampling condition is proved. In addition, because the purpose of this paper is to estimate the possibility of financial statement fraud, the conditions necessary for the neural network to output Bayesian posteriori probability are analyzed.
【学位授予单位】：北方工业大学
【学位级别】：硕士
【学位授予年份】：2006
【分类号】：F239.4;F224
本文编号：2407086
[Abstract]:There is a long history of fraud in the financial statements of listed companies in mainland China, which is very harmful to investors, creditors and the national economy as a whole. But at the same time, for certified public accountants and auditors, The identification of fraudulent financial statements has been a problem. This paper firstly selects the risk factor of financial statement fraud according to the public information, (red flags), and establishes the reasonable suspect index system of financial fraud. Then the forecasting model of financial statement fraud is trained by using the historical data of financial statements of listed companies in Shanghai Stock Exchange of China and the forecasting effect of the model is evaluated. Because of the small proportion of fraudulent financial statements in the total, we use unequal probability sampling, that is, the sampling probability in the fraud category is greater than that in the non-fraudulent sample. In this case, the traditional parameter estimation method needs to be modified. In this paper, the method of parameter estimation based on logistic regression under unequal probability sampling condition is listed, and the method of modifying the output of neural network model under unequal probability sampling condition is proved. In addition, because the purpose of this paper is to estimate the possibility of financial statement fraud, the conditions necessary for the neural network to output Bayesian posteriori probability are analyzed.
【学位授予单位】：北方工业大学
【学位级别】：硕士
【学位授予年份】：2006
【分类号】：F239.4;F224
【引证文献】
相关期刊论文 前1条
1 段玉峰;孟菲;;我国上市公司会计舞弊及其治理[J];合作经济与科技;2011年21期
相关博士学位论文 前1条
1 岳殿民;中国上市公司会计舞弊模式特征及识别研究[D];天津财经大学;2008年
，本文编号：2407086
本文链接：https://www.wllwen.com/guanlilunwen/shenjigli/2407086.html
