Research On The Detection And Governance Of Accounting Fraud Of Listed Companies In China

Authenticity of accounting information of listed companies is an important guarantee for the effective operation of capital markets, while the accounting fraud is to tamper and distort the information disclosed publicly. The ongoing fraud case is a heavy blow to investor confidence, while shaking the foundation of credit capital markets, hindering the healthy and stable development of capital markets. Therefore, Research on the detection and governance of accounting fraud has important theoretical and practical significance. Over the years, scholars have done rich and fruitful research and exploration. From the early theoretical analysis and questionnaire survey for deepening the awareness and understanding the reason of fraud, to identify a single factor fraud, and then to build multidimensional models to identify fraud, the related research on the detection and governance of accounting fraud has yielded fruitful results. However, with the accumulation of research and the deepening understanding of the essential characteristics of accounting fraud and increasingly complexity about accounting fraud, there are limitations on fraud detection have become increasingly prominent, including:(1) Multicollinearity of high-dimensional variables to identify fraud,(2) Nonlinear curve fitting problem of fraud, as well as(3) Uncertainty of non-fraud companies. This article will attempt to address them. Because of the complexity and multi-objective, in order to improve the effect of fraud detection models, the dimensions of the index system increasing, and multicollinearity along with growing, will lead to the model parameter estimation unreliable, so consider removing multicollinearity to enhance the performance of the model is necessary. Secondly, since the detection of accounting fraud is experience-dependent work, of nonlinear characteristics obviously, it is necessary to consider to build the nonlinear model to fit the nature of the problem better. Moreover, previous studies based on the dichotomous detection model, however, in the case of non-fraud company is difficult to determine, the dichotomous fraud detection model is applicable no longer. It is necessary to consider new ideas to construct the model.Around solving the above problems, the following work has done in the paper: First, based on reviewing the domestic and foreign literature on the detection and governance of accounting fraud of listed companies, clarifies the development of research, and learn from the institutional economics and behavioral economics theory and iceberg theory, triangle theory, four-factor theory, fraud risk factors theory and the fraud diamond theory to analysis the causes of accounting fraud. Secondly, Guided by the fraud diamond theory, it selects and builds an index system to identify accounting fraud from the pressure, opportunity, rationalization and capacity, and then Logistic regression models are built based on statistical methods and case-based reasoning models based on artificial intelligence. Furthermore, It discusses the uncertainty of non-fraud companies and the limitations applicable about fraud detection model based traditional assumptions. It builds a comprehensive accounting fraud risk index. Finally, It examines the statistical characteristics of accounting fraud risk index. It reveals the high-risk company from the perspective of the listed sector, industry and administrative areas. It chooses some variables about accounting fraud governance, combined with accounting fraud governance-related literature and our capital market special institutional background, including the company features, characteristics of equity, senior echelon features, feature governance institutions, executive incentive situation and characteristics of external audit. It tests the relevance of each variable and fraud risk index and the level of significance of the difference of mean of the variables by different fraud risks. It puts forward some countermeasures to fraud governance at last accordingly.By the above research, here are the basic conclusions of this paper:(1) By building accounting fraud detection model, It concludes that the correct rate of nonlinear-principal component Logistic regression taking into account nonlinear characteristics of fraud and explanatory variables multicollinearity is higher than the linear Logistic regression model and principal component Logistic regression model. The test set recognition accuracy of the case-based reasoning model using different formulas and different values of K is better than Logistic regression model on the whole. It corroborates the effectiveness of using case-based reasoning technology to identify accounting fraud. It provides a good accounting fraud identification tool for investors, creditors and other stakeholders.(2) By constructing accounting fraud risk index, it characterizes accounting fraud risk of listed companies in detail. First, unlike previous studies, accounting fraud risk index summarize all the information of indicators by a quantitative composite index. Indexation of fraud risk is the advanced stage to detect accounting fraud, reflecting the research trends from radiation to integrate. Secondly, The constructs basis of the index is different. Previous studies begins by collecting fraud and non-fraud binary sample, and then builds pattern recognition classifier based on characteristic index, and then determines the new sample whether fraud. Due to the concealment of fraud, it often can not be discovered to be found even a few years before, which makes the selection of "Non-fraud companies" be lack of objectivity. Different from the simple inertia of dividing the company into fraud and non-fraud categories, the index evaluation method does not require collect samples of known class beforehand, but assuming all of the listed companies are encountering the accounting fraud risk, but with different degrees. Therefore, the index is a more detailed evaluation of the risk instead of classification. Third, the index abandons the original ideas of the accounting fraud identification model(such as Logistic Regression) in the evaluation of fraud risk by summing indicators for fraud identification linearly, which is called "and" method, instead of scoring fraud according to the degree of risk, which is called "or" method. As long as there is an indicator exceeds the threshold value, fraud warning will be carried out. So, the model is more sensitive, more realistic, and of higher value. The index focuses on the pre-warning rather than afterthought, which is more conducive to reducing investor losses, and protecting investor confidence.(3) It analysis the statistical characteristics of accounting fraud risk index, and lists the high risk companies from the perspective of the listed sector, industry and administrative areas. It selects and tests the effect of fraud governance variables, and put forward countermeasures. First, from the results of accounting fraud risk index statistical analysis, the risk level of listed companies on Growth Enterprise Market is higher than other companies Listed. The level of accounting fraud risk in the Hainan Province, Inner Mongolia Autonomous Region and Ningxia Hui Autonomous Region are higher. The level of accounting fraud risk in the West and Northeast region are higher than the East and Middle region. Secondly, it warns the high risk companies from the view of listed sector, industry and the provinces, autonomous regions, municipalities. It serves to remind investors and creditors to avoid investment risks, easy regulators to concentrate its limited resources on high-risk company fraud key supervision. Finally, By analyzing and studying the effect of accounting fraud risk governance index and different dimensions of fraud governance variables empirically, it finds that non-cross-listing, smaller, non-state-owned holding, younger of directors, supervisors and senior management, low level education of directors, supervisors and senior management, chairman or general manager and other important personnel changes, frequently meetings of shareholders and board of directors are often linked to high accounting fraud risk. Low level of ownership concentration, low proportion of the supervisory board holding, low proportion of tradable shares, low proportion of male members in the senior team, short-term of directors, supervisors and senior management, less moderate scale of the Board and the Supervisory Board or other governing bodies, lack of equity incentive executives, large-scale audit firms, high audit fees are linked to low accounting fraud risk. Regulators and listed companies could develop fraud countermeasures accordingly.In order to govern accounting fraud more effectively, from the perspective of regulators, they should focus on increasing efforts to control the GEM listed companies, strengthening the supervision on the listed companies at the Hainan Province, Inner Mongolia Autonomous Region and Ningxia Hui Autonomous Region, strengthening the supervision on the listed companies which are non-cross-listing, smaller, non-state-owned holding, of younger directors, supervisors and senior management, of lower level education of directors, supervisors and senior management, undergoing the chairman or general manager and other important personnel changes, or having frequent meetings of shareholders and board of directors. From the perspective of listed companies, they can govern accounting fraud more effectively by raising equity concentration, improving the equity ratio held by Supervisory Board and the proportion of tradable shares, maintaining a high proportion of male members of directors, supervisors and senior executives, extending the term of directors, supervisors and senior executives, maintaining an appropriate size of the board of directors and supervisors, improving the equity incentive to executives, especially restricted stock incentive, employing larger audit firm, and improving audit fee and so on.