基于阈值检验的财务造假预测——基于中国上市公司的数据分析

Prediction of Financial Fraud Based on Threshold Test——A Case Study of Listing Corporation in China

在大量参考国内外学者对造假公司预测研究的基础上,本文以1993-2012年沪深交易所上市并被发现舞弊的191家公司为样本,并选取61家正常公司作为对比分析.根据GONE理论来挑选具有中国上市公司独特发展背景的11个新指标,并以Benish的M模型为基础建立了更适合中国市场的新模型.通过二元逻辑回归法进行筛选,发现公司规模、经营活动现金流比例、资产负债率、是否亏损等4个指标显著性最强,对舞弊公司的影响较大.借助阈值和ROC检验工具可知本文建立的新模型不仅只是在指标上适用性强,在辨别造假公司上确实也要优于M和F模型,对于多个样本进行模型检验时发现第一类错误率更低,总体准确率有所提升,更适用于在中国资本市场下的应用.

On the basis of referring to a large number of research about financial fraud prediction mforeign scholars, this paper takes the listed and found to be fraudulent companies which are in Shanghai and ShenzhenStock Exchange during 1993-2012 as a sample, selects 11 new indicators witli Chinese listed companies developmentbackground according to the theory 〇 f GONE, and bases on M model by Benish as thWe screen indexes by logistic regression and finally find that the four indexes had the higher significance. They are company size, Operating cash fow ra t io , asset liability ratio and whether losses. With the help of threshold and ROC test tools, it can be known that the new model not only is stronger in the inthan M and F model. The simulation test of many samples show that type I error rates are lower, that an overall accuracyhas improved, and that it is more suitable for application in China’s capital market.