摘要
中国新型加转轨经济的特征,监管制度等方面的不完善,决定了根据发达国家相对成熟资本市场上市公司财务数据建立的会计舞弊检测模型的不适用。通过采用会计舞弊检测中应用普遍并具有较高预测正确率的logistic回归方法,在对现有文献中预测效果较好的财务指标进行方差分析基础上,选择具有显著性的变量建立舞弊检测模型。基于SPSS13.0平台,选择2002～2006年间被中国证监会出具处罚公告的舞弊上市公司及其与之匹配的非舞弊公司控制样本数据,完成了确定样本规模、回归模型与回归参数选择等实验。结果发现,样本规模对舞弊检测正确率有显著影响,而参数中分类点的变化对正确率无明显影响,参数选择对混合逐步回归模型具有显著影响。最终通过比较实验获得具有八个指标的最佳拟合数据检测模型,该模型与现有会计舞弊检测模型相比具有较高的判定率。
In China, the characteristics of transitional and emerging economies result in the inefficacy of accounting fraud detection models derived from the mature security market of developed countries because of the imperfection of regulations. The logistic regression model, which is shown popular and high - quality performance in reference literatures is a- dopted in this paper. The variables are selected based on variance analysis. We carry out the experiments to determine the sample sizes, regression models and related parameters on SPSS 13.0 system. All the fraud samples are from the punish- ment releases of China Security Regulatory Commission during 2002 - 2006. And the matched samples are also chosen from that period. The results show that the sample sizes will affect the model apparently and the clustering point doesn＇t work. Meanwhile, the parameters selection is one of the important factors which affect the step - regression model. The best model which includes eight variables for accounting fraud detection is obtained finally. And it does lead to the higher accuracy of detection comparing with the current models.
出处
《经济与管理研究》
CSSCI
北大核心
2012年第2期88-95,共8页
Research on Economics and Management
基金
国家自然科学基金"基于数据挖掘的上市公司会计舞弊识别问题研究"(70872082)
河北省自然科学基金"上市公司会计舞弊模式识别研究"(F2009000111)
河北省高等学校自然科学重点项目"基于无导师学习算法的上市公司会计舞弊识别模型研究"(ZH2011114)
