【作者】 连竑彬；

【导师】 黄世忠；

【作者基本信息】 厦门大学 ， 会计学， 2008， 博士

【摘要】 本文以我国1994年至2006年间137家涉嫌财务报表舞弊的上市公司为样本,详细描述了我国舞弊公司在规模、业绩、行业、公司治理、舞弊行为、外部审计和舞弊后果等方面的特征。研究发现:(1)我国舞弊公司规模较小,且舞弊前盈利状况欠佳;制造业、农业和信息技术业是舞弊公司相对较为集中的行业;经济较为落后的地区舞弊公司比例较高。(2)我国舞弊公司流通股比例较高;股东大会出席率较低;董事会次数较多;股权集中度较低;第一大股东控制度较低;高管人员持股数较高;高管人员平均年龄较低;高管人员中男性成员比例较高。(3)收入确认问题是我国财务报表舞弊的最大“重灾区”,而利用关联方交易和资产重组进行舞弊是我国的一大特色。(4)我国上市公司舞弊金额巨大,持续时间较长,且相当大比例的舞弊公司在上市前或上市后短期内开始实施舞弊。(5)我国舞弊公司被注册会计师发现的比例极低;相当高比例的舞弊公司在舞弊期间曾经更换会计师事务所。(6)我国受监管部门处罚的公司数量呈现出明显的逐年递增趋势,但处罚周期较长。(7)舞弊公司被ST、暂停上市或终止上市的比例极高。(8)我国监管部门对舞弊公司的处罚主要以“批评教育”为主,受到实质性罚款或市场禁入处罚的舞弊公司较少,且罚款金额极低。在对我国上市公司财务报表舞弊现状进行深入分析之后,本文使用分层逻辑回归分析法来构建舞弊甄别模型。首先,本文利用上市公司舞弊动机、舞弊机会和舞弊借口的替代变量构建出我国上市公司舞弊事前预警模型。然后,本文通过设置六方面财务指标来捕捉舞弊性财务报表数据的异常表现,并利用存在显著差异的财务指标构建出我国上市公司舞弊事后侦查模型。事前预警模型和事后侦查模型构成了分层回归分析中的第一层次模型,并产生相应的舞弊风险因子。在此基础上,本文将事前舞弊预警风险因子和事后舞弊侦查风险因子结合在一起,构建出我国上市公司舞弊综合甄别模型,其总体判别准确率为80.2%。最后,本文通过对样本的适当筛选,将时间因素纳入分析范围,在逻辑回归模型中加入了各种财务比率纵比指标,构建出我国上市公司舞弊及时甄别模型,其总体判别准确率进一步提高到83.2%。更多还原

【Abstract】 In this paper, we selected 137 listed companies in China that had been subject to accounting enforcement action during 1994-2006 as the fraud sample, and then collected data about this fraud sample to depict detailedly the basic character, corporate government character, behavior character, independent audit character and sequel character of fraud companies. We found that (1) Relative to non-fraud companies, Chinese fraud companies were relatively small and endure more financial pressure; the industries affected most frequently included manufacturing, agriculture and IT industry; The area where most of the fraud companies were located are underdeveloped provinces. (2) Relative to non-fraud companies, Chinese fraud companies had relatively high percentage of current stock, low attendance of shareholders meeting, low frequency of directors meetings, low stock equity concentration degree, low largest stockholder control degree, high directors and officers’ stock ownership, low directors and officers’ average ages, high percentage of male directors and officers. (3) The most common techniques used to fraudulently misstate financial statement information is improper revenue recognition, and related party transaction and asset reorganization are peculiar techniques used by Chinese fraud companies. (4) Cumulative amounts of frauds were relatively large in light of the relatively small sizes of the companies involved. The financial statement frauds generally involved multiple fiscal periods. (5)Few fraud companies were detected by external auditor, many companies changed auditors during the fraud period. (6) The numbers of companies subject to accounting enforcement action had shown an increasing trend. (7)A large proportion of fraud companies had suffered special treatment, suspending or terminating the listing of the companies’ shares. (8)Few fraud companies suffered substantiality punishment.Based on the status quo of financial statements fraud in Chinese capital market, we attempt to use cascaded logit analysis to establish an integrative fraud-detecting model. Firstly, under the "fraud triangle" framework, we put the proxy variables of fraud incentive, fraud opportunity and fraud rationalization into logit regression model to establish the ante-fraud warning model. Afterwards, we selected six categories financial ratios to catch the abnormal fluctuate of fraudulent financial statements, and put these financial ratios into logit regression model to establish the post-fraud detecting model. The ante-fraud warning model and post-fraud detecting model represent the first-tier models in cascaded logit analysis. Each of these first-tier models generates an index, ranging from zero to one, of risk of fraud based on the variables included in the model. Then we combined these two indices into the second-tier logit model to establish the integrative fraud-detecting model. Finally, we bring the time factor into our analysis and put the financial trend ratios into the cascaded logit regression to establish the in time fraud-detecting model. We found the concordance and classification accuracy of the in-time fraud-detecting model are higher than the integrative fraud-detecting model.更多还原