Financial Statement Fraud Detection - Applicable of Dechow F-score in China

: This study focuses on the effectiveness of the Dechow F-SCORE model in monitoring financial fraud in the Chinese financial market. At the same time, the research also aims to thoroughly evaluate the advantages and deficiencies of the F-SCORE model. To implement research, six pairs of listed companies have been selected from different industries, and each company includes a company that has been involved in financial fraud cases and a company that has not had financial fraud. The annual report data of these companies conducted a comprehensive analysis in the vertical and horizontal directions. However, the results of the study show that when the F-SCORE model is applied to compare the financial fraud of these companies, the results are not noticeable. The results of the data analysis show that the F-SCORE model does not seem to detect financial fraud in Chinese-listed companies effectively. The conclusion of this study pointed out that the application of the F-SCORE model in Chinese listed companies is limited, and it may require more improvement and customization to adapt to the exceptional circumstances of the Chinese financial market. This also emphasizes that financial fraud monitoring involves various methods and tools to meet the needs of different regions and markets.


Background
Does the F-score model apply to monitoring financial fraud in Chinese companies?Financial fraud seems to be increasingly moving away from the fringe market activities of the financial sector and becoming a pervasive type of behavior across the industry.The aftermath of the financial crisis of 2007-2008 revealed many scandals in which financial market participants infected the markets with fraudulent information to gain personal advantage [1].
In recent years, the increasing prevalence of financial fraud has emerged as a significant threat to companies operating in a challenging external environment.According to statistics, the number of financial fraud cases disclosed on the official website of the China Securities Regulatory Commission (CSRC) has been escalating since 2000, with both the frequency and magnitude of fraudulent activities showing a marked increase.By April 2023, the total number of CSRC penalties in 2023 has reached 34 [2].Figure 1 below shows how the number of CSRC punishments has changed from 2010-2020.
Figure 1: 2010-2020 number of financial fraud cases punished by CSRC [3] The frequent occurrence of financial fraud hinders the healthy and sustainable development of companies, seriously harms the interests of investors, and affects the orderly operation of the market.Therefore, it is imperative to identify and prevent financial fraud effectively.Patricia Dechow et al. (2011) developed the F-score model to detect whether a firm has committed financial fraud.However, only some empirical articles in China demonstrate the applicability of this model in China.Therefore, this study is innovative in landing the F-score in China.

Research Design
In this paper, the F-score model is used to detect the company's economic operation to determine whether there is financial fraud.Data collection is conducted by selecting six pairs of public companies; one company in each team has committed financial fraud while the other has not.The selection criteria are based on the companies' size, growth, and industry.The pairs include Yihua Lifestyle and Sleemon, Kangmei Pharmaceutical and Yunnan Baiyao, ZONECO and Guolian Aquatic Products, Lonkey Industrial and Shenma Industrial, Tempus Hold and China Tourism Group Duty-Free Corporation Limited, as well as Kangde Xin Composite Material Group and Huadian Energy.
The F-score model is applied to calculate the financial metrics for each pair of companies to identify companies likely to manipulate earnings, and the data are from the disclosures released by the companies -annual reports.The model variables measure evidence of manipulation and managers' predisposition/opportunity to manipulate.
There are two types of analysis of data: vertical and horizontal comparisons.The vertical comparison involves comparing a company's financial metrics before and after the occurrence of fraud.The horizontal comparison consists of comparing the companies with and without financial fraud to identify common factors that may contribute to fraudulent behavior.
The research aims to answer the question of whether the F-score model is effective in identifying financial fraud in Chinese public companies.A conclusion will be drawn based on the results of the analysis.

2.
Data Collection and Analysis

Data Collection
In this paper, a total of 12 companies from 6 different industries are selected.Two companies from the same industry, one a fraud company and the other a non-fraud company are compared to conclude.
Table 1 shows the basic data of these 12 companies for calculation.In the selection of data, this paper takes the fraud year of the respective industry's fraud company as the primary line selects the financial information of the three years before the fraud and the two years at the time of the fraud, and uses this as the basis for calculating the seven variables and the Fscore required for Dechow F-score.
The seven variables and their formulas required for this paper are as follows:  2, all the variables for calculating the predicted value as well as the F-score and the formulae for the variables, are shown.
In the Dechow F-score study, it mentions that F > 1 indicates "above normal" risk, F > 1.85 indicates "substantial" risk and F > 2.45 indicates "high" risk [5].If this study needs to prove that the Dechow F-score applies to China, it needs to show that the F-score of fraud companies in all industries is driven to increase before and after the fraud year, while the change in the F-score of non-fraud companies is not significant.Conversely, the Dechow F-score is not applicable to China.From Table 3, it can be seen that there is no obvious difference between the F-score of the fraud year and the F-score of the non-fraud year in the fraud company, and a suitable basic figure cannot be found to judge the severity of the risk.After comparing the F-score of the F-score of the fraud year and the F-score of the non-fraud year in the non-fraud company, it is not consistent with the expectation that the F-score is smoother.4, by describing the median and mean of the F-score for all misstatement years and the F-score for all non-misstatement years, leads to the conclusion that the mean and median of the F-score for all non-misstatement years will be greater than that for all This is contrary to Dechow's findings [4].This also suggests that the Dechow F-score may not be applicable to China.

Regression Analysis
In order to further verify the accuracy of the results obtained from the descriptive analysis, this study utilises regression analysis to determine the relationship between the F-score and whether the company is a fraud company as well as the fraud year.This study was designed with the formula: −   = 0 +  1 •   +  2 •   +  3 •  •   + ｡Where Post indicates whether the fraud company is in the fraud year or not, Post=0 means that the fraud company is in the pre-fraud year, Post=1 means that the fraud company is in the fraud year.Fraud indicates whether the company is a fraud company or not.Fraud=0 means the company is non-fraud company, Fraud=1 means the company is fraud company.t in the formula represents the year.Table 5, is the database for the regression analysis.The Post value, Fraud value, and F-score of the fraud and non-fraud companies are filled in and brought into the formula according to the rules.
After regression analysis, the following values are obtained.6 shows that the fraud*post corresponding F-score is -0.00162, which is relatively close to 0, indicating that F-score is not significant.Further analysis shows that the change in the F-score before and after the fraud is insignificant.

Conclusion
To sum up, this article introduces the F-score model, the F-score model applied to companies with financial fraud in China, and the horizontal and vertical comparison of the F-score model (vertical comparison involves comparing the company's financial performance before and after the fraud occurs.Indicators and horizontal comparisons involve comparing firms with and without financial fraud).
From the Table 4 data analysis, it can be concluded that there is no significant difference between the F-score of the fraudulent company in the fraudulent year and the F-score in the non-fraudulent year.Therefore, the F-score cannot directly conclude whether the company has blatant fraud.
From the data analysis in Tables 3 and 6, it can be inferred that the value of the F-score is opposite to the data of the F-score, so there is a bold inference that the F-score is unsuitable for data analysis of Chinese companies.
Although the data analysis in this article believes that the F-score is unsuitable for financial or non-financial fraud analysis by Chinese companies, does the F-score have strong accuracy?
First, the F-score model is often more useful than accuracy, especially if the class is not evenly distributed.Accuracy works best if false positives and false negatives have similar costs.If the cost of false positives and false negatives differs, it is better to consider precision and recall.
Plus, systematic bias makes high precision and low accuracy possible.One of the problems with recall, precision, F-measure, and accuracy used in information retrieval is that they are prone to bias.
What are the advantages of the F-score?Microscopic precision or recall will result in a lower overall score.Therefore, it helps to balance the two metrics.If you choose the positive class as the class with fewer samples, the F-score can help to balance the metrics between positive/negative models.
In conclusion, the F-score has its advantages and disadvantages.However, it may not be suitable for analyzing Chinese companies regarding financial fraud.Therefore, it is challenging to determine if the F-score has fully utilized its potential in aiding the analysis of Chinese companies' financial fraud.

Table 1 :
The basic data of 12 companies for calculation

Table 2 :
Variables and formula required

Table 3 :
F-scores for 12 companies in different years

Table 4 :
F-scores for 12 companies in different years

Table 5 :
Data required for regression analysis

Table 6 :
Result of regression analysis