Study On The Characteristics And Detection Model Of Financial Reporting Fraud:empirical Evidence From Chinese Listed Companies

Accounting is a universal Commercial language. Objective and impartial accounting information can provide an effective basis for economic decisions, and it is also useful in maintaining the capital market stability. The financial reporting fraud inflicts on investors(especially small investors) seriously, disrupts the normal order of capital market. Study on the characteristics and detection model of financial reporting fraud is necessary, which can improve the efficiency of supervision, reduce the incidence of financial reporting fraud and protect the investors.This paper analyzes the motive of financial reporting fraud, using the data of the Chinese listed companies penalized by SEC in 2004-2009. Then, all the fraud companies are divided into two categories according to the motive of fraud, in order to study the specific characteristics of various fraudulent practices in financial reporting, and build targeted fraud detectiing models.Based on the data of 56 fraudulent financial reportings and 56 non-fraudulent financial reportings, characteristics of the companies implemented FFR to regulate profits is illustrated. The results of the research show that the companies implemented FFR to regulate profits have a higher proportion of accounts receivable and other receivables, a lower proportion of inventories and fixed assets, a higher asset-liability ratio and a lower total asset turnover. When faced with the same profit pressures, nature of the largest shareholders and the proportion of shares owned by the largest shareholders have major impact on the occurrence of FFR; while the inhibition of internal corporate governance mechanisms have no affect on the financial reporting fraud. Compared with the normal company, the board size, supervisors board size and proportion of shares owned by board, supervisors boardand and executives is significantly less than the non- fraudulent company. And in both cases, CPA can identify FFR.Based on the data of 79 fraudulent financial reportings and 79 non-fraudulent financial reportings, characteristics of the companies implemented FFR not to regulate profits is illustrated. It is concluded that the companies implemented FFR not to regulate profits have a higher proportion of accounts receivable and other receivables, a lower proportion of inventories and fixed assets, a higher asset-liability ratio, a higher operating expenses ratio and financial expense ratio; a lower total asset turnover and receivables turnover. There are four variables to be possible indicators of non- regulating profits FFR.These included audit opinion, accounting firms, frequency of supervisors board meetings, the proportion of shares owned by the executives.Finally, all the samples are divided into two groups, one group is the training sample set, the other is test sample. With the data of training sample set, the detection models of FFR are constructed using LibSVM algorithm of RBF kernel and linear kernel. Then, the prediction accuracy of the model is tested with the data of test sample set. Finally, the paper compares the prediction accuracy of the LicSVM models and the Logistic regression models.It turns out that the prediction accuracy of LibSVM algorithm for RBF kernel function model to detect regulating-profits FFR is 86.67%, the overall accuracy is 87.5%. Corresponding to the model of LibSVM linear kernel function algorithm, the prediction accuracy is 83.33%, the overall accuracy rate of 86.61%. the accuracy of the Logistic regression model is 80% and 83.04%.Corresponding to the model to detect non-regulating-profits FFR, the prediction accuracy of LibSVM algorithm for RBF kernel function model is 83.33%, the overall accuracy is 87.97%. The prediction accuracy of the model of LibSVM linear kernel function algorithm is 87.5%, the overall accuracy rate is 83.54%. The accuracy of the Logistic regression model is 83.33% and 80.38%. In general, the model of the RBF kernel function with SVM algorithm is more effective for detecting regulating-profits FFR, and the model of the linear kernel function with SVM algorithm is more effective for detecting non-regulating-profits FFR.On the basis of above research, this paper put forward the corresponding policy recommendations. There should be adequately game in order to reduce the financial reporting fraud. Decision-making institution of Board should be perfect. There should be a Employee Stock Ownership Institution to improve the proportion of shares owned by the executives, board and supervisors. And independent director system should be improved, To ensure that independent directors can really play a role in the listed companies.