Research on Financial Fraud Identification Mechanisms Incorporating ESG Information

In recent years, although China’s capital market has experienced rapid growth, companies in all industries have continued to develop and expand their operations with great momentum. However, the number of corporate financial fraud incidents has not decreased but continues to grow. In China, due to historical reasons and institutional deficiencies, many companies have a large number of financial fraud problems that have not been controlled and are still fermenting, which not only brings great damage to the interests of investors but also endangers the entire social and economic system and affects the stable operation of the national economy.

In the face of the frequent occurrence of financial fraud financial auditing under the framework of hard indicators of financial evaluation and identification system has become increasingly difficult to cope with the growing complexity of financial fraud [1-2]. First of all, the traditional financial audit is only for listed companies that have disclosed the true, complete, objective and fair opinion of financial information. The audit process is more focused on the enterprise’s current operating conditions and related financial indicators but ignores the enterprise’s mode of operation, behavioural patterns and the future of the ability to carry out an in-depth analysis of the sustainable development of the [3-5]. Secondly, some special industries have high professional knowledge barriers, and the auditors of accounting firms are only professionals in the field of accounting and finance, and sometimes they are not able to make appropriate identification and judgement on some complex matters involving industry professional knowledge [6-9]. Again, due to the nature of financial auditing, it is difficult to examine and evaluate the audited entity comprehensively and thoroughly with limited human and material resources, and therefore, may not be able to understand the whole picture of some complex economic interests [10-13]. Finally, due to the fierce market in the auditing industry, accounting firms employed by audited entities may have difficulty in fully safeguarding their independence and objectivity in order to compete for client resources or retain high audit service revenues [14-16]. The emerging ESG evaluation system is precisely able to break through the limitations of the rigid indicator framework of financial auditing, comprehensively and holistically assess the overall credit quality of enterprises, and play an important role in the early warning of corporate credit risk [17-21].

The article first argues the harm of corporate financial fraud from three aspects: corporate self-interest, corporate social reputation and market environment, explains the necessity of financial fraud identification, and then discusses the importance of ESG information for financial fraud identification. XGBoost is used as a classification model, and financial fraud cases have been selected as samples to construct a financial fraud identification model since 1998. The model introduces the information value to construct the indicator screening model, extracts 18 financial variables and 4 non-financial variables, and applies the XGBoost algorithm to classify the sample data after cleaning and normalising the sample data. The identification accuracy of fraudulent and non-fraudulent enterprises is tested separately, and different models are selected as comparisons to test the robustness of this paper’s model. Incorporating ESG information to identify and analyze the fraud mechanisms of K enterprises is necessary to achieve further in-depth analysis of the causal chain of corporate financial fraud.

2

The need for ESG integration for financial fraud identification

2.1

The dangers of corporate financial fraud

Financial malpractice will not only bring serious consequences to the enterprise itself, but also, as an important part of the securities market, will inevitably jeopardize other stakeholders and the entire securities market.

Although in the short term, enterprises can gain many benefits from financial fraud, in the long term, financial fraud can cause harm but not benefit. First of all, once financial fraud tastes sweetness, the management will continue to carry out financial fraud and long term, will eventually have the day of exposure. Once discovered by the market, it will face a series of serious consequences such as penalties, share price decline, or even suspension of trading. Secondly, when the management is accustomed to financial fraud to whitewash the financial statements, the enterprise will no longer be so focused on the development of the main business and ultimately will only put the cart before the horse. In the long run, this type of enterprise cannot develop.

In addition, enterprises do not exist in isolation in the market environment. Each enterprise has many stakeholders. For the creditors of the enterprise, if the financial fraud of the enterprise is discovered, it is uncertain whether the enterprise itself can continue to develop, and it is even more uncertain as to the repayment of debts. For the public who have purchased the company’s shares, they are even more innocent. All along, the public shareholders had been deceived by a false financial report of the enterprise, and eventually, the enterprise fraud was discovered, and the shareholders had to pay for the wrongdoings of the enterprise because of the plummeting of the share price.

Furthermore, as there are a large number of companies in the securities market, if the number of companies engaging in financial fraud increases, the securities market will inevitably be affected. Specifically, once the implementation of financial fraud enterprises is exposed, the securities market regulators need to investigate and verify the company, which will occupy a large number of related resources. Secondly, once the financial fraud, the interests of investors will be harmed if the number of companies carrying out financial fraud increases, as the place of securities issuance and trading of the securities market, will naturally be substantial turbulence, which will, in turn, Reduce the confidence of investors in the securities market for investment, once a large number of investors choose to withdraw from the securities market investment, the securities market want to long-term development will be more difficult.

2.2

ESG concepts

As we can see from the above, financial fraud has multiple hazards, so it is important to identify financial fraud as early as possible. The judgment information for identification is one of the important conditions to achieve accurate identification. Therefore, this paper introduces ESG information as the auxiliary information for financial fraud identification.

ESG is a value concept that focuses on the comprehensive performance of enterprises in the three aspects of environment, social responsibility, and corporate governance rather than only focusing on financial data [22]. ESG performance, on the other hand, is an indicator that measures the fulfilment of enterprises in the three aspects of environment (E), society (S), and corporate governance (G), which implies various types of information about enterprise operation and management, and is able to send good signals to the market, and some of the Investors will use ESG performance to further judge the enterprise’s future sustainable development ability, the level of corporate management and other information, has become an important indicator for external stakeholders to assess the enterprise’s sustainable development ability. For example, good environmental performance means that enterprises will actively support the development of environmental protection, pollution prevention and energy utilisation, and will pay more attention to the future sustainable development of the enterprise. Good social responsibility performance indicates that the enterprise will pay attention to the management and maintenance of employee rights and interests, product quality, supply chain management, social investment, investor protection, etc., and will have goodwill, but also can side show the enterprise’s good management and risk resistance ability. Good corporate governance performance is even more linked to the enterprise’s governance structure and potential risks. The better the performance of the enterprise, its management system is standardised, the solvency is better, and the lower the risk that may exist.ESG ratings contain a huge amount of information, which not only contains multi-dimensional indicators of the environment, society, and corporate governance, but also involves indicators that may affect the future operation of the enterprise, such as. Environmental pollution penalties, administrative penalties, lawsuits, data breaches, etc.

3

Financial fraud identification modelling

3.1

XGBoost Algorithm

The XGBoost algorithm is an optimisation algorithm obtained by enhancing the GBDT (Gradient Enhanced Regression Tree) algorithm based on the Boosting algorithm strategy [23]. The GBDT was developed by Friedman (2001) and has two components: the regression tree and the gradient boosting [24]. Compared with linear regression, a regression tree can sequentially select the variables that best explain the results during the fitting process, place the most important variables in the tree for initial segmentation, and then gradually construct the interaction process with other variables to improve the efficiency of the model from the level of variable selection.

The XGBoost algorithm is an additive model consisting of multiple decision trees, and the algorithmic process is shown below:

Step 1: Let training set T = (x₁, y₁), (x₂, y₂), …, (x_n, y_n), loss function $l (y_{i}, {\hat{y}}_{i})$ , regularisation τ(f_k), k represent the k th decision tree, then the objective function of XGBoost is defined as: (1) $L (ϕ) = \sum_{i} (l (y_{i}, {\hat{y}}_{i})) + \sum_{k} τ (f_{k})$ (2) $τ (f_{k}) = r T + \frac{1}{2} λ \sum_{j = 1}^{T} ω_{j}^{2}$

i is the i nd sample, i = 1, 2, 3, …, n, ${\hat{y}}_{i}$ are the predicted values for the i th sample x_i: (3) $x = {\hat{y}}_{i}^{(t - 1)} + f_{t} (x_{i})$ (4) ${\hat{y}}_{i} = \sum_{k = 1}^{K} f_{k} (x_{i}) = {\hat{y}}_{i}^{(t - 1)} + f_{t} (x_{i})$

XGBoost expressions can be updated: (5) $L^{(t)} = \underset{i = 1}{\sum^{n}} [l (y_{i}, {\hat{y}}_{i}^{(t - 1)} + f_{t} (x_{i}))] + \sum_{k} τ (f_{k})$

Step 2: Perform a binomial Taylor expansion of the loss function, assuming that both the first and second order derivatives of $l (y_{i}, {\hat{y}}_{i})$ at y_o exist: (6) $l (y_{i}, y) \approx l (y_{i}, y_{0}) + l^{'} (y_{i}, y_{0}) (y - y_{0}) + \frac{l^{″} (y_{i} {\hat{y}}_{i}^{(t - 1)})}{2} {(y - y_{0})}^{2}$

l(y_i, y) the second order Taylor expansion at ${\hat{y}}_{i}^{(t - 1)}$ : (7) $l (y_{i}, x) \approx l (y_{i}, {\hat{y}}_{i}^{(t - 1)}) + l^{'} (y_{i}, {\hat{y}}_{i}^{(t - 1)}) (x - {\hat{y}}_{i}^{(t - 1)}) + \frac{l^{''} (y_{i}, {\hat{y}}_{i}^{(t - 1)})}{2} {(x - {\hat{y}}_{i}^{(t - 1)})}^{2}$ (8) $g_{i} = l^{'} (y_{i}, {\hat{y}}_{i}^{(t - 1)})$ (9) $h_{i} = l^{″} (y_{i}, {\hat{y}}_{i}^{(t - 1)})$

Delete the constant term $l (y_{i}, {\hat{y}}_{i}^{(t - 1)})$ to simplify the objective function: (10) $L^{(t)} = \underset{i = 1}{\sum^{n}} [g_{i} f_{t} (x_{i}) + \frac{h_{i}}{2} f_{t}^{2} (x_{i})] + \sum_{k} τ (f_{k})$ (11) $\sum_{k} τ (f_{k}) = \sum_{k = 1}^{t - 1} τ (f_{k}) + τ (f_{k}) = τ (f_{k}) + constant$

Assumption I_j = {i|q(x_i) = j}, i.e. all the samples belonging to the j nd leaf node x_i are divided into the same node sample set. Training all samples can be obtained eventually: (12) $L^{(t)} = \underset{j = 1}{\sum^{T}} [(\sum_{i} g_{i}) ω_{j} + \frac{1}{2} (\sum_{i} h_{i} + λ) ω_{j}^{2}] + λ T, i \in I_{j}$

Step 3: The functions at each leaf node of l(y_i, y) are independent of each other, and the model yields an optimal solution when the functions at all nodes are optimal, at which point each leaf node has weight $w_{j}^{*}$ : (13) $w_{j}^{*} = \frac{Σ_{i} g_{i}}{\sum_{i} h_{i} + λ}$

Optimal objective function: (14) $o b j = - \frac{1}{2} \sum_{j = 1}^{T} \frac{{(\sum_{i} g_{i})}^{2}}{\sum_{i} h_{i} + λ} + λ T$

When selecting features to divide a node into left and right nodes in the actual tree model, the variable that makes the objective function decrease the most, i.e., the variable with the largest gain, should be preferred for tree splitting, and the gain is expressed as: (15) $G a i n = \frac{1}{2} [\frac{{(\sum_{L} g_{L})}^{2}}{\sum_{L} h_{L} + λ} + \frac{{(\sum_{R} g_{R})}^{2}}{\sum_{R} h_{R} + λ} - \frac{{({(\sum_{L} g_{L})}^{2} + {(\sum_{R} g_{R})}^{2})}^{2}}{\sum_{L} h_{L} + \sum_{R} h_{R} + λ}] - λ$

Where the subscript L represents the split left node and R represents the split right node.

Above is the decision tree splitting principle of the XGBoost model in a certain iterative bargaining process. The XGBoost algorithm introduces a greedy strategy in the multi-layer iterative process to achieve the goal of maximising gain gain and appropriately sets the gain threshold to avoid overfitting the model. The feature selection function that comes with the XGBoost algorithm is mainly achieved by calculating the gain Gain of each variable and ranking the importance of each variable from high to low according to the gain Gain to achieve the interpretability of the model. Compared to the GBDT algorithm, the XGBoost algorithm has a second-order Taylor expansion of the loss function, which makes the algorithm more accurate. On this basis, the XGBoost algorithm comes with a regularization condition, which improves model accuracy while preventing overfitting of the learner. During operation, the XGBoost algorithm can fully utilize the CPU for multi-threaded parallel computation to improve the operating rate. In addition, the XGBoost algorithm can automatically learn the missing values in the input data, which reduces the data preprocessing process.

3.2

Financial fraud identification model construction

Based on the superior performance of the XGBoost model, this paper constructs a financial fraud recognition model on its basis. In this paper, financial fraud cases since 1998 are selected as model training and testing samples. By checking the penalty announcements issued by the regulators as of 31 December 2021 one by one, after excluding the violations only for delayed disclosure, the samples of companies involved in falsification of financial statement accounts in the first category of violations are marked as financial fraud samples, and finally, 1441 financial fraud samples are obtained, which are collated to obtain the distribution table of the number of financial fraud companies per year as shown in Table 1. The training set and test set are divided according to the ratio of 6:4, i.e., the size of the training set is 865 samples, and the size of the test set is 576 samples.

Table 1.

The distribution of financial fraud companies

Year	Number of listed companies	The number of corrupt companies	The proportion of fraudulent companies among the listed companies in that year(%)
1998	829	27	3.26%
1999	927	40	4.31%
2000	1060	51	4.81%
2001	1141	34	2.98%
2002	1201	41	3.41%
2003	1265	53	4.19%
2004	1355	48	3.54%
2005	1352	31	2.29%
2006	1430	29	2.03%
2007	1547	28	1.81%
2008	1604	25	1.56%
2009	1753	30	1.71%
2010	2102	37	1.76%
2011	2344	50	2.13%
2012	2471	70	2.83%
2013	2518	64	2.54%
2014	2634	93	3.53%
2015	2822	124	4.39%
2016	3123	171	5.48%
2017	3495	208	5.95%
2018	3588	187	5.21%
Sum	40561	1441	3.55%

Twenty-two indicators were identified through the indicator screening model, including 18 financial variables and 4 non-financial variables, namely: current ratio (X01), quick ratio (X02), inventory turnover (X05), accounts payable turnover (X06), accounts receivable turnover (X07), accounts receivable-to-revenue ratio (X08), total asset turnover (X09), inventory-to-revenue ratio ( X10), shareholders’ equity turnover (X11), return on assets (X12), return on invested capital (X13), net profit margin on total assets (X15), return on long-term capital (X17), growth rate of total assets (X18), proportion of independent directors (X19), proportion of shares held by the Board of Directors (X20), proportion of shares held by the Supervisory Board (X21), proportion of shares held by the ten largest shareholders ( X22). After data normalization, the sample data was divided into training and testing sets using the five-fold cross-validation method. The XGBoost algorithm was used as a classifier to construct the financial statement fraud recognition model. The flow of the financial fraud recognition model is shown in Figure 1.

4

Performance analysis and application study of financial fraud identification models

4.1

Model Recognition Performance Testing

4.1.1

Financial fraud enterprise identification

The primary goal of the financial fraud identification model is to accurately detect the financial fraud of a company in a timely manner, to combat illegal behaviours, and to maintain the benign development of the capital market, so we are more concerned about the performance of the model in identifying the financially fraudulent enterprises. As shown in Figure 1, the accuracy of the model declines as the coverage of the model increases because in order to improve the coverage, the learner determines as many samples as possible to be in the category of financial malpractice, thus leading to a decline in the accuracy. The accuracy versus coverage curve for the financial fraud samples is depicted in Figure 2. As can be seen from the figure, the accuracy of the tested model in discriminating out-of-sample data ranges from 18.56 per cent to 100 per cent, and accordingly, the percentage of identified coverage ranges from 100 per cent to 3 per cent.

In practice, regulators can trade off accuracy against coverage depending on the objective. If it wants to find financially fraudulent companies with 100 per cent accuracy, it can find 3 per cent of the typical financially fraudulent companies from the sample of selected companies, and then combat these companies as typical. Suppose the regulator wants to conduct a comprehensive screening of companies. In that case, the parameters can be adjusted to find nearly half of the fraudulent companies with an accuracy rate of about 34 per cent, thus narrowing down the scope to be screened and replacing the initial screening with an identification model.

4.1.2

Identification of non-fraudulent financial firms

Investors do not have enforcement powers and do not care about whether a firm is financially fraudulent; by contrast, investors are more concerned about which firms are not financially fraudulent, and thus, their greatest need is to find enough firms in which they can safely invest. This requires investors to weigh the risk they are willing to take against the size of the stock they expect to expand and if they are willing to tolerate a higher rate of error, the number of non-financially fraudulent firms that can be added to the investment pool will be greater. On the contrary, if investors cannot bear the risk of misjudgement at all, the model identifies fewer unfinancially fraudulent firms but is better able to ensure that each judgment is accurate. In this paper, the accuracy versus coverage curve for the unfinancial fraud sample is generated by adjusting the sample threshold, as shown in Figure 3. The model still guarantees about 88% accuracy when it identifies 75% of the unfinancial fraud firms within the sample, and about 84% accuracy even when it identifies all of the unfinancial fraud firms.

Bringing the model into practical application. There are about 3,000 companies in the A-share market at the end of 2023, and according to statistics, the average default rate among listed companies in China is 13%, which means that about 390 companies have carried out the operation of financial fraud. If the trained machine learning model for judgement, it can be from the 13% default probability of the sample, screening out the error rate of 8.5%, covering 75% of the non-financial fraud enterprises of the stock investment pool, which greatly broadens the scope of the capital market investor’s selection of stocks, which proves that from the point of view of the selection of stocks for the investor, the model’s effect has been very significant. Moreover, if the trained model is outsourced and a visualisation window is added, investors are free to choose from 88% to 100% accuracy in order to acquire different numbers of non-financial fraudulent company stocks, depending on the trade-off between the level of risk they are comfortable with and the pool of stocks they wish to broaden.

4.1.3

Robustness Tests

In order to confirm the good nature of the financial fraud identification model constructed in this paper, this paper selects representative algorithms among the three major classes of machine learning algorithms, i.e., traditional classification algorithms, neural network algorithms and integrated learning algorithms, in which the support vector machine model is selected for traditional classification algorithms, the BP neural network model for neural network algorithms, and the random forest algorithm adopting Bagging strategy and the GBDT algorithm adopting Boost strategy for integrated learning algorithms. Random Forest algorithm with Bagging strategy and GBDT algorithm with Boost strategy is selected among the integrated learning algorithms, and the above four algorithms are used for debugging the parameters respectively, and the algorithm of this paper is launched to predict for the same sample respectively, and its performance is evaluated through the indexes of accuracy rate, precision rate, recall rate and F1 value. The evaluation results of the model performance are shown in Table 2.

Table 2.

Performance evaluation of different models

Model	Accuracy(%)	Precision(%)	Recall(%)	F1(%)	Time(s)
SVM	79.48	69.27	77.19	73.01	1585
BP	85.28	81.07	72.84	76.73	2582
RF	90.49	76.66	74.76	75.70	2150
GBDT	91.52	85.21	68.85	76.16	620
XGBoost	94.11	90.68	68.78	78.23	50

As can be seen from the table, the excellent performance of the XGBoost algorithm in this paper is verified. Among them, the accuracy rate reaches 94.11%, the F1 value reaches 78.23%, the time consumed is only 50 seconds, and the performance of all evaluation indexes is better than the comparison algorithm. In addition, regardless of the machine learning algorithm verified, it can have an accuracy rate of nearly 80% or more and an F1 score of 70% or more, which demonstrates that the recognition model constructed in this paper has a consistently good performance regardless of the training strategy used, and thus the robustness of the model can be tested.

4.2

Analysis of Financial Fraud Mechanisms Incorporating ESG Information - The Case of Enterprise K

4.2.1

Fraud identification analysis of the environment (E) dimension

In this section, Enterprise K in the pharmaceutical industry is selected as the research object, and the performance of the three ESG dimensions is analysed in turn. Firstly, the scores and ratings of the ESG performance of Enterprise K are used as auxiliary references, and then ESG information derived from various public channels such as annual reports, responsibility reports, government information and the media is used as the main object of analysis, and 12 ESG items of Enterprise K are analysed for the identification of financial malpractice. The scores for the environment (E) dimension of Enterprise K are shown in Table 3.

Table 3.

Environmental dimension score

Rating time	Best peer score	Worse peer score	Average peer score	K score	Rank
2015	23.797	14.563	16.516	18.407	10/22
2016	24.839	15.875	17.183	15.759	7/17
2017	23.404	14.274	19.199	17.456	8/16
2018	28.434	16.184	17.786	23.066	6/21

The environmental (E) dimension can be broadly divided into environmental management (environmental management system, objectives, employee environmental awareness, green policies, etc.), environmental disclosure and negative environmental information. From Table 3, it can be seen that Company K’s score in the environmental dimension is not high, with a four-year average score of 18.672, but it is still in the upper middle of the industry, indicating that the pharmaceutical industry as a whole performs poorly in the environmental dimension, and Company K’s relative performance in the industry is still good.

1)

Environmental disclosure

In terms of energy consumption and energy saving, the total energy consumption of the pharmaceutical manufacturing industry is much lower than that of the traditional high-energy-consuming industrial manufacturing industry. However, its manufacturing and operating processes are mainly based on the consumption of electricity.K Enterprises needs to have a large amount of energy to support the integrated operation of the whole industrial chain and the aggressive expansion of the business scale. However, K Enterprises has not disclosed enough energy-saving and emission-reduction indicators. In terms of pollutant emissions disclosure, K Enterprise has better transparency, and its official website publishes quarterly environmental monitoring reports for both its traditional Chinese medicine and Western medicine bases.

In addition to the two environmental factors of energy saving and emission reduction and pollutant discharge, many other indicators can be used to measure the impact of enterprises on the environment, such as energy-saving technological means in production and operation, certification of corporate environmental management, and the establishment of environmental protection systems. However, Enterprise K’s social responsibility report focuses too much on industrial layout, future planning, and corporate image promotion, with less disclosure of relevant environmental aspects.

Overall, in the environmental dimension, Enterprise K has insufficient disclosure of energy consumption and energy saving, pollutant emissions, and lack of quantitative indicators, and there is an obvious tendency of selective disclosure and whitewashing of negative environmental information, which shows that Enterprise K has not taken the initiative to assume the social responsibility of the leading enterprises in the industry, which implies that there is a certain degree of moral risk. This tendency implies that there is a certain degree of ethical risk and that the enterprise may falsify information during a crisis.

2)

Negative environmental information

Enterprise K was issued a Decision on Ordering the Correction of Illegal Acts in August 2016 because the company had committed a serious waste liquid treatment violation, causing environmental pollution. In addition, the subsidiaries of K Enterprises were entered into the List of Priority Discharge List for Water Environment in 2016 and 2017. Despite the fact that Enterprise K has had fewer environmental violations, it is one of the few companies in its industry that has been penalized. Overall, Enterprise K has more negative information related to environmental pollution than the peer average. In addition, as the public attaches greater importance to environmental protection and the level of regulation on environmental aspects is strengthened, Enterprise K’s failure to meet environmental standards may also incur reputational risk and legal, and regulatory risk, which in turn may affect the sustainable operation of the enterprise. Thus, the moral hazard propensity and sustainability risk of enterprise K may form the pretext and pressure motivation for financial fraud.

4.2.2

Social Responsibility (S) Dimension Fraud Identification Analysis

Table 4 shows the social responsibility (S) dimension scores of K Enterprises. Table 4 shows that Enterprise K also performs poorly in the social responsibility (S) dimension, with scores of 15.253, 16.048, 17.964, and 17.587 from 2015-2018, scoring below average among its peers year after year. Its performance in public welfare and donations is better, but problems in product management and employee management are more prominent.

1)

Employee Management

The data shows that the company’s employee composition is becoming more diverse, which is a reflection of the company’s openness and inclusiveness. However, in terms of educational attainment, Enterprise K’s average of 18.7% bachelor’s degree and above is far lower than the 32.47% compared to the average in the same industry, and there is no clear disclosure on the percentage of master’s degree and doctorate, which is undoubtedly a short board for Enterprise K, exposing the risk of shortage of middle and senior talents.

The company does not disclose employee organizations or congresses, which is not conducive to the expression of employees’ opinions in defense of their rights and interests. In addition, the proportion of administrative staff is as high as 26.7%, which is more than double the average percentage of companies in the same industry. This proportion of administrative staff may be too high, which is likely to lead to problems such as unclear division of authority and unclear management. The proportion of technical staff is 12.45%, which is also lower than the industry average.

Enterprise K, as a pharmaceutical company, should be based on scientific and technological levels as the core competitiveness, and its low percentage of highly educated personnel and scientific research into the staff investment may not be able to provide a guarantee for the subsequent sustainable development, which is likely to make the K enterprise in the peer competition at a disadvantage, unable to obtain excess earnings, the lack of such core competitiveness will ultimately make the enterprise in the normal operation of the enterprise can not achieve the performance objectives, thus forming financial fraud This lack of core competence will ultimately prevent the enterprise from achieving its performance targets in normal operations, thus creating a pressure motive for financial fraud.

2)

Supply Chain Management

Enterprise K’s lack of clear disclosure of upstream and downstream relationships, its lack of after-sales service and good communication mechanisms with consumers and suppliers, and its failure to establish appropriate procedures to prevent commercial bribery create a risk that the company may be involved in corruption. Furthermore, Enterprise K does not have clear procedures for restraining its suppliers, nor does it further clarify the specifics of its upstream and downstream suppliers and counterparties, and clearly defines whether there are related party transactions, which may not fully ensure the level of compliance in corporate transactions, and may even give rise to financial malpractice.

Table 4.

Society responsibility dimension score

Rating time	Best peer score	Worse peer score	Average peer score	K score	Rank
2015	25.920	14.705	17.081	15.253	14/22
2016	25.043	15.785	17.552	16.048	11/17
2017	22.274	13.019	17.965	17.964	4/16
2018	21.957	11.818	17.755	17.587	12/21

4.2.3

Fraud identification analysis of corporate governance (G) dimensions

Table 5 shows the scores for the corporate governance (G) dimension of firm K. From the table. It can be seen that Enterprise K scores extremely low on the corporate governance dimension, with an average score of only 10.41 for the four years, a relatively weak level of corporate governance, and worrisome disclosure of information in key parts of the organisational structure, internal control mechanisms, compliance management and risk management. With irrational shareholding structure, imperfect governance mechanism, a large number of functional departments of the board of directors and the lack of supervision regulations, obvious deficiencies in the company’s internal control system, and major problems in risk monitoring and constraints on major controlling shareholders and the protection of the rights of small and medium-sized shareholders, K Enterprises are exposed to the greatest risk of fraud in the corporate governance dimension compared to the environmental and social dimensions.

1)

Corporate Governance Structure

Good corporate governance not only benefits the company’s management but also prevents financial fraud. From the information presented by several channels, the governance structure of K enterprise is unreasonable, and the imperfection of the governance mechanism creates the possibility for the occurrence of financial malpractice.

Firstly, the general meeting of shareholders failed to effectively protect the interests of small and medium-sized investors. The small proportion of public shareholders and the arbitrariness of the controlling shareholders resulted in the inability of the general meeting and the board of directors to effectively protect the interests of small and medium-sized investors in decisions related to financial fraud. Second, the overlap of corporate governance roles. Multiple positions and roles are highly overlapped in one person, which violates the original checks and balances and undermines the hierarchical constraints of “shareholders” meetings, board of directors, and management. Thirdly, the supervisory board did not play a supervisory role; the proportion of external supervisors on the supervisory board of Enterprise K was zero, and its members also held other positions in the company, which together with the controlling shareholders, formed a huge community of interests, and were unable to effectively supervise the company’s operation. Finally, the responsibilities of the shareholders’ meeting, board of directors and supervisory board of Enterprise K are confused with those of the finance, business and other related departments, which leads to the failure of the whole company’s internal control system. Taken together, Enterprise K performs very poorly in its corporate governance structure, exposing a huge risk of financial fraud.

2)

Anti-Corruption Bribery and Oversight Systems

Company K’s annual report, internal control system report and social responsibility report did not cover anti-corruption-related laws and regulations, preventive measures and supervision and implementation methods, and given Company K’s business model, the company has certain regulatory risks and corruption and bribery risks in the supply, drug trading and retailing sectors. In fact, the top management of Company K started to pay bribes several times in 2003, and the targets of the bribes involved the local drug regulatory authorities, the issuance regulatory authorities of the Securities and Futures Commission, and other relevant personnel, and as of 2015, according to the announcement of the relevant authorities, the cumulative amount of bribes involved in the bribery behaviour of Company K that had been revealed had exceeded RMB 16.8 million.

However, in the face of so many incidents involving corruption, Company K still has not developed a relevant anti-corruption system as well as a monitoring system, exposing the fact that its management and governance may have overridden the company’s internal controls, and at the same time, such governance initiatives have to be questioned as to their motives.

Table 5.

Govern dimension score

Rating time	Best peer score	Worse peer score	Average peer score	K score	Rank
2015	14.077	9.433	11.407	10.449	17/22
2016	16.639	10.795	13.249	9.947	17/17
2017	15.160	10.409	13.327	10.794	13/16
2018	17.923	8.882	14.234	10.445	18/21

5

Conclusion

In this paper, XGBoost is used as a classification algorithm to construct a model for recognizing financial fraud. The identification test found that the model’s discriminative accuracy for out-of-sample data varied from 18.56% to 100%, and the percentage of identification ranged from 100% to 3%, while the model still ensured an accuracy of about 88% when 75% of non-financial malpractice companies were identified within the sample. The results of the robustness test show that the identification using the XGBoost algorithm achieves an accuracy of 94.11%, F1=78.23%, and takes 50 seconds, all of which are better than the comparative models such as BP neural networks and random forests, indicating that the financial fraud identification model based on XGBoost has excellent performance.

Combined with ESG information, the financial fraud mechanism is analysed using Enterprise K as an example. It is found that there are loopholes in the dimensions of environmental disclosure, negative environmental information, employee management, supply chain management, corporate governance structure, and anti-corruption and bribery and monitoring systems, which have the risk of financial fraud. This shows that the integration of ESG information can better analyze the causal mechanism of corporate financial fraud, so as to more accurately identify enterprises with financial fraud risk.

Langue:: Anglais

Périodicité:: 1 fois par an
Sujets de la revue:: Sciences de la vie, Sciences de la vie, autres, Mathématiques, Mathématiques appliquées, Mathématiques générales, Physique, Physique, autres

RSS Feed de la revue

Research on Financial Fraud Identification Mechanisms Incorporating ESG Information

Xiaoyu Fu

Publié en ligne: 05 févr. 2025

Reçu: 05 sept. 2024

Accepté: 17 déc. 2024

DOI: https://doi.org/10.2478/amns-2025-0058

Mots clésESG, Xgboost, Financial fraud, Identification models

© 2025 Xiaoyu Fu, published by Sciendo

This work is licensed under the Creative Commons Attribution 4.0 International License.

Mots clés
ESG, Xgboost, Financial fraud, Identification models