Do Directors’ Network Positions Affect Corporate Fraud?

Abstract

Corporate fraud poses a significant obstacle for sustainable business development. Drawing on social network analysis, this paper used data originated from Chinese-listed companies from 2009 to 2022 and found that directors’ network position significantly mitigates corporate fraud. Mechanism tests indicated that the quality of external auditors and internal control play a mediating role in this relationship. Further analysis showed that the network positions of independent directors, non-independent directors, and female directors individually inhibit the inclination of corporate fraud when considering various types of directors. Of note, the busy director hypothesis was not applicable in explaining the impact of directors’ network position on corporate fraud. This study provides a new approach to improving the sustainability of enterprises in newly emerging markets via the analysis of director networks. It is also beneficial to the research on director networks and corporate fraud in companies, offering insights for corporate governance and fraud prevention in companies and regulatory agencies.

1. Introduction

In recent years, fraud by listed companies has occurred frequently due to high economic benefits and low violation costs, and it has not been effectively curbed despite repeated prohibitions. This widespread phenomenon of corporate fraud has attracted close attention from regulatory bodies worldwide. In response, the Chinese government has recognized the dangers of corporate fraud, and the State Council has issued “Several Opinions on Strengthening Supervision, Preventing Risks, and Promoting High-Quality Development of the Capital Market”, with Article 3 specifically emphasizing strict regulation of key issues such as financial fraud and misuse of funds. Against this backdrop, using the fundamentals of corporate governance to explore strategies for preventing corporate fraud is particularly important for improving governance levels, strengthening investor protection, and promoting the sustainable development of capital markets [1].

Corporate fraud not only harms investors’ interests but also disrupts the normal market order severely, hindering the quality improvement of the capital market. Thus, the scholars have performed substantial studies on the factors influencing corporate fraud, concentrating on internal governance factors, external environmental factors, and firm characteristics. From the perspective of internal governance, studies found that board independence [2], board gender diversity [3], and audit committee independence and frequency of activities [4] reduced corporate fraud. Research regarding the external environment showed that legal framework [5], market competition [6], and business cycles [7] can influence corporate fraud. In addition, a study related with firm characteristics indicated that factors such as management shareholding [8], institutional investors [9], CEO characteristics [10], firm size [7], and financial status [11] were significantly associated with corporate fraud. On the other hand, the directors’ network position refers to a director-held position within a network structure formed by simultaneous employment of directors in multiple companies. Current research on directors’ network positions involved in company value [12], corporate innovation [13], corporate social responsibility [14], and management forecasting [15] focus on internal governance and business operations. Generally, these studies on directors’ network position were carried out based on the advantage of information and resource provided by the director networks.

Although numerous studies have been performed separately on director networks and corporate fraud, there is little research paying attention to the effect of informal institutions on corporate fraud. Director networks belong to one of the typical forms of informal institutions. According to the social network embeddedness theory and social capital theory, a director’s network position may generate more reputation, heterogeneous information and resources, enhance the director’s monitoring motivation and monitoring ability, and inhibit the occurrence of corporate fraud [16]. However, the “busy director” hypothesis suggests that directors with more responsibilities might be busier and more prone to distraction, making it difficult for them to allocate sufficient time and attention to supervising the companies they serve [17,18]. Studies by Kuang and Lee found that well-connected independent directors do not reduce the likelihood of corporate misconduct [19]. However, research by Xing et al. showed that companies with well-connected independent directors are less likely to engage in fraudulent activities [16]. These studies focus on the networks of independent directors without addressing how the entire director network and different types of director networks affect corporate fraud. This study aims to answer three crucial questions: (1) Does the entire directors’ network position affect corporate fraud? (2) What are the mechanisms behind this effect? (3) Do various types of director networks cause the discrepancy on this effect?

Based on a sample of Chinese A-share-listed companies from 2009 to 2022, this study investigates the impact of directors’ network positions on corporate fraud. It then reveals a negative correlation between directors’ network positions and the likelihood of corporate fraud. Internal controls and external audits serve as mediating factors in this relationship. Compared to companies with directors in marginal positions, those with directors occupying central network positions are less likely to engage in corporate fraud. Well-connected directors restrain corporate fraud by increasing the internal control quality and hiring top-tier auditors. Additionally, further analysis indicates that the network positions of independent directors, non-independent directors, and female directors, as well as busy directors, significantly reduce the probability of corporate fraud. These findings have important theoretical and practical implications for the prevention of corporate fraud. Theoretically, this study enriches the literature on corporate governance and social network theory by highlighting the vital role of directors’ network positions in corporate governance. Practically, it provides valuable recommendations on corporate management for policymakers and regulatory authorities. Overall, companies can effectively reduce corporate fraud by optimizing director networks, strengthening internal controls and improving the quality of external audits.

The main contributions of this study can be summarized as follows:

(1)

From the perspective of theoretical value, this paper contributes to the study of director networks in the field of corporate governance. Although the literature has extensively discussed director networks, views on their effect in internal governance and business management remain divided, with significant disagreements. The empirical findings of this study show that directors’ network positions help suppress corporate fraud, and the “busy director hypothesis” does not hold in this field. Moreover, while much of the current literature focuses on the causes and consequences of corporate fraud from the perspectives of formal institutional arrangements, such as internal corporate governance and compensation design, this paper takes the social relations of informal institutional arrangements as an entry point and confirms that the external social relations of the company, represented by the director’s network, influence the company’s economic behavioral decision-making.

(2)

From a research methodology perspective, this paper utilizes social network analysis techniques to create indicators of director network centrality. Unlike prior empirical studies that converted director networks from two-mode to one-mode models using projection techniques, this study assesses directors’ network positions based on semi-local centrality within the bipartite graph structure of the director network. The proposed method reduces the information loss and redundancy caused by projection techniques. Additionally, it allows for the separate assessment of independent directors, non-independent directors, and female directors within the entire director network. This paper provides a new perspective for the analysis of director network.

(3)

The empirical results of this paper show that the network positions of directors (independent directors, non-independent directors, and female directors) play a role in suppressing corporate fraud. The study also reveals how director network positions suppress the inclination towards corporate fraud through specific mechanisms (including the quality of internal control and external auditors).

The paper is organized as follows: Section 2 outlines the origin of the hypotheses, while Section 3 presents the data and methodology used in this study. Then, Section 4 discusses the empirical results, and Section 5 provides an in-depth analysis of director network types and the busy director hypothesis. Finally, Section 6 briefly summarizes the paper’s findings, offers policy proposals, and discusses its limitations.

2. Theoretical Analysis and Hypotheses

2.1. Directors’ Network Positions and Corporate Fraud

Coles et al. proposed that external tournament incentives significantly promote manager effort and company performance [20]. According to the director incentive theory, central network positions provided directors with higher visibility and greater career opportunities, motivating them to work harder on corporate governance in order to outperform their peers. Central network positions can incentivize directors through several mechanisms. First, directors in central network positions have greater access to industry information and resources, allowing them to better monitor and guide company operations. They utilize the information and resources to implement the best practices, thereby improving the internal control quality and corporate governance. Second, owing to directors’ central network positions, their actions and decisions easily attract the attention of their peers and the public. Consequently, they place a higher value on personal reputations and supervisory roles to avoid any corporate fraud that could damage their reputations. Third, the central network positions of directors were able to enhance their professional development and future career opportunities through corporate governance performance. It further motivated them to work harder on corporate governance for ensuring transparency and compliance in company operations [21]. Through the above mechanisms, director networks regulate the motivation and ability of directors to monitor corporate fraud effectively.

A director’s monitoring motivation is largely influenced by their reputation in the director network. As noted by Fama, the reputation is crucial in motivating directors to effectively monitor in order to avoid legal risks and reputational damage [21]. The reputation holds an important economic value in the corporate social environment [22,23,24]. Social network theory posits that an actor’s network position directly determines its reputation acquisition [25]. Effective corporate monitoring not only enhances directors’ reputations in the human capital market, but also helps them avoid legal risks and reputational losses due to malfeasance, which in turn affects their career prospects. Wang and Zhang [26] demonstrated that the characteristics of a director ‘s network position directly affects its own reputation acquisition, with directors located in the center of the network enjoying a higher reputational value. The more central a director’s position in the network is, the greater its ability to access and use resource information, thus providing more effective guidance and supervision for the company, and thus earning a higher reputation among the elite circle of directors [27]. In addition, directors at the center of the network are more likely to be offered potential directorships, which enhances their independence in the performance of their duties and reduces the risk of losing their positions in conflicts with major shareholders and management, thereby enhancing the objectivity and proactivity of monitoring. A study by Cashman et al. [28] found that chain directors with higher network centrality also have a higher likelihood of obtaining additional directorships in the future. Thus, directors at the center of the director network have access to a wealth of social capital, including board seats in other firms, due to their strong social network connections. This position allows them to perform their monitoring duties more independently and effectively without fear of compromising their existing positions by offending large shareholders or management.

A director’s supervisory capacity is affected by their access to information. It has been shown that the quality of information is a key determinant of the effectiveness of directors’ governance function. Through director networks, operators and investors can obtain information from both internal and external sources, effectively reducing the information asymmetry between stakeholders and the firm [29,30]. Dachin et al. [29] pointed out that the more adequate information directors have, the more they are able to respond to the challenges faced by the company in a timely manner, thus improving the effectiveness of internal governance. Simultaneously, directors’ timely access to corporate information directly reduces the likelihood of corporate fraud. Due to the existence of a network of affiliations, directors have strong access to information, enabling them to implement timely and effective supervision of the company, thereby preventing problems before they occur [31]. Bakke et al. found that the unexpected death of a well-connected director leads to a significant decline in the firm value of interlocked companies, highlighting the importance of director networks in maintaining corporate performance [13]. This study suggests that well-connected directors improve the governance and monitoring capabilities of the overall board due to accessible and valuable information and resources. Therefore, the more well-connected directors in the company, the more the information and resources brought about by the network of directors can curb the occurrence of corporate fraud in a timely manner.

However, based on the “busy director hypothesis”, directors serve in multiple companies at the same time, which inevitably leads to their energy being scattered and them having less time to invest in a single company. Therefore, well-connected directors cannot allocate enough time and attention to supervising and advising all the boards they serve. Research by Coles et al. [32] shows that busy outside directors tend to overpay CEOs, which in turn leads to poor firm performance. Based on this, the American Institute of Boards of Directors (AIBD) and the Investor Association (IA) have established regulations that limit the number of concurrent directorships each director can have in a public company. Ferris et al. [33] proposed the “busy director hypothesis”, suggesting that if a member of a board of directors within a company concurrently serves as a director of more than one company, this will weaken the director’s ability to regulate the individual company and create a weak internal corporate governance environment. Building on this theory, Fich and Shivadasani [18] found that busy directors are associated with weaker corporate governance, and that when directors hold concurrent positions in three or more firms, these firms typically have lower market-to-book ratios and operating profits. Furthermore, when these firms are underperforming, the CEOs of busy directors’ firms leave at a lower rate than those of firms with non-busy directors. Beasley [2] found that chain directors with positions in multiple companies on average experience a higher frequency of breaches in their companies. Andres et al. [34] found that directors’ “busyness” can be explained by several factors, including their position in the director’s network, in addition to the number of companies they work for. Directors at the center of the network need to invest more time in maintaining relationships, which may affect their supervisory efficiency. Faleye [35] argues that directors usually take on multiple roles in order to increase their personal career value or to gain wealth, which does not lead to an increase in boardroom efficiency. Ahn et al. [36], in a more in-depth study, suggest that a high proportion of interlocking directors on the board may significantly reduce boardroom efficiency, which in turn negatively affects the company’s share price. In addition, if a company has a richer network of director relationships, the level of management compensation within the company is significantly higher, and the company’s Tobin’s Q is lower. This suggests that director moonlighting can have a greater impact on the effectiveness of corporate governance. Kuang and Lee [19] investigated two potential processes of corporate violations and found that interlocking director relationships may reduce the probability of a company being audited and reduce the number of penalties imposed.

Based on the above analyses, this study points out that busy directors are often perceived as having limited monitoring capabilities because they serve on multiple boards at the same time. However, well-connected directors, even if they are busy, are able to compensate their limited time and energy with critical information and enriched experience derived from director networks. In addition, busy directors typically have stronger incentives to safeguard their reputations. Considering the reputation effect and information resource advantage, well-connected directors have a stronger ability and motivation to monitor corporate behaviors for the potential decline of corporate fraud. This study proposes the following hypotheses:

Hypothesis 1 (H1).

Directors’ network position is negatively correlated with corporate fraud.

2.2. The Mediating Role of Corporate Internal Control

In the resource dependence theory, resources are key elements for the survival and development of enterprises. They are essential for establishing competitive advantages and enhancing sustainable development. Director networks serve as an important channel that enables enterprises to access external information resources, thus improving internal control quality and preventing corporate fraud. Well-connected directors who value their personal reputation, aim to maintain network centrality and social capital and strengthen corporate governance and internal control quality. Moreover, they are more inclined to take advantage of network information to enhance the quality of internal controls. The research showed that well-connected directors positively influence internal control systems by performing the best legal practices and rigorous monitoring measures [19]. Director networks generated spillover effects that can improve the quality of internal control in affiliated companies [37,38].

Following that, an improved quality of corporate internal control can suppress corporate fraud by strengthening risk identification capabilities and continuous monitoring functions. Specifically, a high-quality internal control system tends to effectively identify and assess various risks faced by enterprises, including financial, operational, and compliance risks. Through timely risk identification and assessment, enterprises were capable of formulating and implementing appropriate control measures to prevent corporate fraud [39]. Activities within the internal control system, such as audits and compliance checks, are used to detect and correct unusual activities in routine operations, preventing the occurrence and spread of corporate fraud [40]. Continuous monitoring functions of the internal control system regularly assesses its effectiveness and adaptability through internal auditing, supervision, and inspection. The enterprise continuously optimizing the internal control system can maintain a high-pressure situation against corporate fraud and create a deterrent effect on company management, which further reduces the incidence of corporate fraud [41].

Based on the above analysis, we formulate Hypothesis H2:

Hypothesis 2 (H2).

Internal control quality mediates the relationship between directors’ network position and corporate fraud.

2.3. The Mediating Role of External Auditing

External auditing can fulfill its external governance function by addressing deficiencies in a company’s internal governance system [42,43]. From the perspective of the directors’ network, the closer a director is to the center of the network, the higher their reputation and the greater the loss in case of reputational damage. Therefore, well-connected directors are less likely to want corporate fraud to occur in the company they work for and will take active steps to reduce the probability of major problems, thereby lowering the likelihood of damage to their reputations. In addition to the negative response of “voting with their feet” (i.e., resigning from the board), directors can also influence corporate decisions by playing an active role on the board of directors. One of the more critical decisions is the hiring of an accounting firm.

On the one hand, high-quality auditors can serve an external governance role by providing high-quality accounting information to assist the board of directors in monitoring management and constraining large shareholders. On the other hand, the engagement of high-quality auditors can also send positive messages to the market [44], reflecting the effectiveness of directors’ governance and further enhancing directors’ personal reputations. Therefore, well-connected directors can reduce the probability of corporate fraud by improving the quality of external audit [45]. This leads to Hypothesis 3:

Hypothesis 3 (H3).

External audit quality mediates the relationship between directors’ network position and corporate fraud.

In summation, the theoretical model of this study is shown in Figure 1.

3. Data and Methodology

3.1. Data and Sample Selection

Drawing on Xing et al. [16], our initial sample consists of all firms listed in China’s A-share market from 2009 to 2022. All data are obtained from the CSMAR database and the WIND database, which are widely used by scholars. Specifically, data about directors are obtained from the listed company character profile series of the CSMAR database and the executive profiles of the WIND database. By leveraging the listed company codes and personal director IDs, the director network constructed in this paper includes 5097 listed companies with 74,052 directors. The corporate fraud data are gathered from the CSMAR database, focusing on public condemnation, public criticism, or public punishment announcements of listed companies by the China Securities Regulatory Commission (CSRC), Shanghai and Shenzhen Stock Exchanges and the Ministry of Finance (MOF). Data on company characteristics and governance measures were obtained from the CSMAR database and the WIND database.

To ensure accuracy and consistency in the data analysis, the data were processed according to the following criteria: (1) Data from the financial sector were excluded due to the unique structural characteristics of the industry, which could introduce analytical bias. (2) Observations with missing data were removed to avoid potential bias and misleading inferences. (3) To mitigate the effects of outliers, all continuous variables were trimmed within the 1% and 99% quantile ranges. The final dataset consists of 43,439 observations from 4849 companies. The software employed for statistical testing in this paper includes Stata 16.0, NetworkX for Python 3.5.

3.2. Variable Measurement

3.2.1. Dependent Variable

The year of fraud is defined as the year in which the fraud exists, not the year in which it is detected. When the fraud event spans many years, we identify the firm as fraudulent when it persists for multiple annual observations [16]. We identify the years in which firms exhibit fraudulent behavior by obtaining documents from the listed firms’ breach database in the CSMAR database pertaining to the handling of regulatory announcements of breaches. The types of corporate fraud specifically include fictitious profits; misrepresentation of assets; false records (misleading statements); delayed disclosure; material omissions; inaccurate disclosure (other); fraudulent listings; contribution violations; unauthorized changes in the use of funds; appropriation of company assets; insider trading; illegal trading of shares; stock price manipulation; illegal guarantees; and general accounting mishandling.

3.2.2. Independent Variable

This study constructs a two-dimensional “company–director” matrix by year based on the comprehensive information of individual directors obtained from the CSMAR database. According to graph theory and data structure, the director network is an undirected, unweighted bipartite graph. Undirected networks imply that influence and information flow in both directions between companies and directors. In the unweighted network, each link between the company and the director is assigned equal importance. Considering the essential difference between listed companies (legal persons) and directors (natural persons) in the directors’ network, and the correlation between the listed company and the director, the criterion for the bipartite graph is satisfied. Therefore, the directors’ network is a bipartite graph $G:G=\left(X,Y,E\right)$, where the set of dots $X\left(G\right)$ as top nodes represent listed companies, the set of dots $Y\left(G\right)$ as bottom nodes represent directors, and the set of edges $E\left(G\right)$ represent company–director linkage relationships. The edge sets $E\left(G\right)$ arise due to the linkages formed by natural persons as members of the board of directors of one or more listed companies.

Previous studies usually employ the projection technique of social network analysis to convert the two-dimensional matrix of “company–director” into a one-dimensional matrix of “director–director” (1 if a director and b director serve on the board of directors of at least one company at the same time, and 0 otherwise) and then calculate the network centrality index at the individual director level of the sample company. The mean, maximum, or minimum value of the network characteristic variable for all directors of the company is used to reflect the director network position at the company level. However, using projection techniques to convert a two-mode model into a one-mode model (as shown in Figure 2) will cause the number of edge sets $E\left(G\right)$ to expand, making the director network information redundant. Newman et al. [46] and Guillaume and Latapy [47] show that projecting onto a bipartite graph leads to very dense networks and extremely high clustering coefficients.

To evaluate the efficiency and effectiveness of the network analysis, this paper does not use the projection technique and instead retains the bipartite graph structure (company–director) of the director network. It adopts the semi-local centrality proposed by Chen et al. [48] to access the director network position. This measure considers that the influence of a node is not only related to its own degree value, but also to the influence of nodes in its neighborhoods, taking into account the fourth-order neighborhoods information of the node, number of neighbors reachable within two steps from the node. It is a compromise between median centrality and proximity centrality, which reflecting the position of the node in the network and reducing the computational complexity.

The formula for calculating the semi-local centrality of a node is as follows:

$$C{L}_i={\displaystyle \sum _{j\in N\left(i\right)}{Q}_j}$$

(1)

where $C{L}_i$ denotes the semi-local centrality of node $i$; $N\left(i\right)$ denotes the set of first-order neighbors nodes of node $i$; and $Q_j$ denotes the second-order neighborhoods degree of node $j$, the number of neighbors reachable within one step from node $j$.

By calculating the semi-local centrality of all nodes in the director network, we can use two metrics to measure the location of the director network at the company level. The first is a direct measure using the semi-local centrality of the top node (company), indicating the number of companies that a listed company can be directly associated with through its board members (as shown in Figure 3). The second is measured using the mean value of the semi-local centrality of all the bottom nodes (directors) connected to the top node, indicating the number of other directors that each director of a listed company on average can directly associate with through the network. We utilize the first one as an explanatory variable and the second one as a substitution variable for the robustness test.

3.2.3. Mediating Variable

The first mediating variable in this study is internal control. Following the methodologies of Feng, H. et al. [49], We use the DIBO Internal Control Index from the DIBO database to evaluate the level of internal control in companies. The DIBO Internal Control Index is a key indicator for measuring the annual efficiency of a company’s internal control. It assesses a range of factors, including company and industry risks, internal audits, internal control deficiencies, misconduct, related-party transactions, and legal proceedings. A higher internal control index indicates better internal control quality within the company.

The second mediating variable selected in this study is whether the audit is conducted by one of the Big Four auditing firms. We use this variable to assess the impact of external audit quality on corporate fraud. The presence of a Big Four firm typically indicates a higher level of audit quality and rigor. This variable helps to evaluate the effect of high-quality auditing on its subsequent impact on corporate fraud.

3.2.4. Control Variable

Drawing on the existing literature, the control variables selected in this paper specifically include leverage ratio (Lev), enterprise value (Tobin’s Q), ownership concentration (Top 1), return on net assets (ROE), company size (Size), time to market (Age), board size (Board), nature of property rights (SOE), shareholding ratio of institutional investors (Indsh), and CEO duality (Dual). To account for unobservable time and industry-specific effects, the paper also incorporates year (Year) and industry (Ind) dummy variables.

Table 1. Definition of variables. The measurement methods for each statistical variable.

Type	Name	Symbol	Definitions
Dependent Variable	Corporate Fraud	Fraud	Fraud is a dummy variable that equals one if the firm commits fraud, and zero otherwise.
Independent Variable	Directors’ Network Position	Centrality	The semi-local centrality of the top node.
Mediating Variable	Internal Control	DBI	DIBO Internal Control Index
	External Audit	Big 4	Audit by a Big 4 international firm takes the value of 1, otherwise 0.
Control Variable	Leverage Ratio	Lev	Total liabilities at the end of the year divided by total assets at the end of the year.
	Enterprise Value	Tobin’s Q	Ratio of the market value of an enterprise’s assets to their replacement cost.
	Ownership Concentration	Top 1	Shareholding of the first largest shareholder among all shareholders.
	Return on Net Assets	ROE	The ratio of a firm’s net profit to its average net worth, reflecting the level of compensation received by owners’ equity.
	Company Size	Size	The natural logarithm of the company’s total assets.
	Time to Market	Age	Company listing age.
	Board Size	Board	Number of board members.
	Nature of Property Rights	SOE	Depending on the nature of the company’s beneficial owner, the variable takes the value of 1 for state-owned firms and 0 for non-state-owned firms.
	Shareholding Ratio of Institutional Investors	Indsh	Institutional investor shareholding as a percentage of total equity.
	CEO duality	Dual	The dummy variable equals 1 if the Chairman and CEO are the same person, 0 otherwise.
	Industry	Ind	Industry dummy variables are assigned as binary indicators based on the standard industry classification, with a value of 1 for companies in a specific industry and 0 for all others.
	Year	Year	Year dummy variables are assigned a value of 1 for the relevant. year and 0 for all other years.

provides detailed information on these variables.

3.3. Model Setting

3.3.1. Basic Regression Model

To test Hypothesis 1, the following model was constructed in this study:

$$Frau{d}_{i,t}={\alpha }_0+{\alpha }_{1}Centralit{y}_{i,t}+{\alpha }_k{\displaystyle \sum Control{s}_{i,t}}+{\displaystyle \sum Year}+{\displaystyle \sum Industry}+{\epsilon }_{i,t}$$

(2)

In the equation, the firm is indexed by i and the year by t. Controls are control variables, and ε is a random perturbation term.

3.3.2. Modeling the Mediating Effects of Internal Controls

To verify whether a director’s network position influences corporate fraud by strengthening internal controls, this study employs stepwise regression to construct models 3 and 4 to test hypothesis H2. In models 3 and 4, “DBI” represents internal control, and the other variables remain consistent with those previously mentioned.

$$DB{I}_{i,t}={\alpha }_0+{\alpha }_{1}Centralit{y}_{i,t}+{\alpha }_k{\displaystyle \sum Control{s}_{i,t}}+{\displaystyle \sum Year}+{\displaystyle \sum Industry}+{\epsilon }_{i,t}$$

(3)

$$Frau{d}_{i,t}={\alpha }_0+{\alpha }_{1}Centralit{y}_{i,t}+{\alpha }_{2}DB{I}_{i,t}+{\alpha }_k{\displaystyle \sum Control{s}_{i,t}}+{\displaystyle \sum Year}+{\displaystyle \sum Industry}+{\epsilon }_{i,t}$$

(4)

3.3.3. Modeling the Mediating Effect of External Audit

To verify whether director’s network position influences corporate fraud through the Big Four auditing firms, this study employs stepwise regression to construct models 5 and 6 to test hypothesis H3. In models 5 and 6, “Big 4” indicates whether the audit is conducted by one of the Big Four auditing firms, and the other variables remain consistent with those previously mentioned.

$$Big{4}_{i,t}={\alpha }_0+{\alpha }_{1}Centralit{y}_{i,t}+{\alpha }_k{\displaystyle \sum Control{s}_{i,t}}+{\displaystyle \sum Year}+{\displaystyle \sum Industry}+{\epsilon }_{i,t}$$

(5)

$$Frau{d}_{i,t}={\alpha }_0+{\alpha }_{1}Centralit{y}_{i,t}+{\alpha }_{2}Big{4}_{i,t}+{\alpha }_k{\displaystyle \sum Control{s}_{i,t}}+{\displaystyle \sum Year}+{\displaystyle \sum Industry}+{\epsilon }_{i,t}$$

(6)

4. Empirical Analysis

4.1. Descriptive Statistics

From the descriptive statistical analysis of the variables in

Table 2. Results of descriptive statistics. Observing the observations, mean, standard deviation, median, minimum value, maximum value, and VIF of each variable provides an understanding of the dataset’s characteristics.

Variant	N	Mean	Std. Dev.	Median	Min	Max	VIF
Fraud	43439	0.363	0.481	0.000	0.000	1.000	-
Centrality	43439	3.530	1.254	3.850	0.000	5.236	1.13
Lev	43439	0.455	1.295	0.455	−0.195	178.345	1.40
Tobin’s Q	43439	2.551	71.783	2.055	0.000	14,810.306	1.39
Top1	43439	34.162	15.22	34.162	0.000	100.000	1.24
ROE	43439	0.031	4.405	0.031	−207.397	713.204	1.00
Size	43439	22.079	1.373	17.560	11.348	28.636	1.05
Age	43439	13.93	8.502	13.000	0.000	32.000	1.31
Boardsize	43439	10.446	3.696	10.000	4.000	58.000	1.12
SOE	43439	0.34	0.474	0.000	0.000	1.000	1.44
Indsh	43439	38.913	25.54	38.913	0.000	144.675	1.31
Dual	43439	0.286	0.452	0.000	0.000	1.000	1.00

, it is observed that the mean value of Fraud is 0.363, indicating that corporate fraud occurs in about 36.3% of the overall sample of company–year samples, and that the observed sample is reasonably distributed. The mean value of Centrality is 3.53, and the median is 3.85, which indicates a more reasonable distribution of values. However, the highest value of Centrality is 5.236, and the lowest value is 0, indicating that there are significant differences in the network location of directors of different listed companies. Furthermore, the Variance Inflation Factor (VIF) constructed using all variables is less than 5, so there is no problem of multicollinearity.

We grouped the annual samples of listed companies according to the presence of corporate fraud and conducted t-tests on the means of the variables, with the results presented in

Table 3. Results of t-test for variable means.

Variant	Non-Corporate Fraud Companies	Corporate Fraud Companies	Test of Difference
	N = 27,652	N = 15,787
	Average Value	Average Value
Centrality	54.037	52.289	4.25 ***
Lev	0.435	0.490	−4.35 ***
Tobin’s Q	2.646	2.120	0.60
Top1	35.446	31.910	23.45 ***
ROE	0.099	−0.087	4.20 ***
Size	22.129	21.991	10.15 ***
Age	13.639	14.440	−9.45 ***
Board	9.920	11.369	−40.05 ***
SOE	0.369	0.289	17.15 ***
Indsh	40.252	36.565	14.50 ***
Dual	0.290	0.279	2.50 **

Note: *** p < 0.01; ** p < 0.05.

. The mean value of Centrality of the corporate fraud group is lower than that of the non-corporate fraud group, and there is a significant difference between the two, which preliminarily confirms that the director network has a certain impact on the company’s corporate fraud behavior. The results for the remaining variables are generally consistent with the findings of the existing literature. Compared to companies without corporate fraud, companies with corporate fraud have higher Lev, Age, Board. But lower Top1, ROE, Size, SOE, Indsh, and Dual.

4.2. Basic Regression Results

Table 4. The regression results of directors’ network position on corporate fraud.

Variable	(1)	(2)
Centrality	−0.030 ***	−0.056 ***
	(−3.23)	(−6.41)
Lev		0.113 ***
		(3.85)
Tobin’s Q		−0.001 ***
		(−3.22)
Top 1		−0.008 ***
		(−9.58)
ROE		−0.031 ***
		(−3.06)
Size		−0.000 ***
		(−6.71)
Age		0.017 ***
		(10.21)
Boardsize		0.119 ***
		(35.47)
SOE		−0.482 ***
		(−17.08)
Indsh		−0.004 ***
		(−6.92)
Dual		−0.031
		(−1.28)
Year	Control	Control
Industry	Control	Control
Constant	−1.081	−0.918
	(−6.66)	(−5.72)
R2	0.067	0.028
Observations	43439	43439

Note: z-statistics are shown in brackets. *** p < 0.01.

presents the regression results for the relationship between Fraud and Centrality. Column (1) of the table displays the results of the univariate logistic regression. The coefficient of Centrality is significantly negative at the 1% level (coefficient = −0.03, p < 0.01). Column (2) shows the regression results after adding control variables. It can be observed that the coefficient of Centrality is still significantly negative at the 1% level (coefficient = −0.056, p < 0.01) after all control variables are added, which implies that Fraud is significantly negatively related to Centrality. The regression results are consistent with Hypothesis 1.

4.3. Mechanism Test and Results

Table 5. Test results of the mechanism of directors’ network position and corporate fraud.

Variable	(1)	(2)	(3)	(4)
	DBI	Fraud	Big4	Fraud
Big4				−0.239 ***
				(−4.20)
DBI		−0.128 ***
		(−4.2)
Centrality	0.174 ***	−0.041 ***	0.074 ***	−0.055 ***
	(3.17)	(−6.36)	(3.17)	(−6.36)
Lev	−3.386	0.631 ***	0.030	0.113 ***
	(0.85)	(3.85)	(0.85)	(3.85)
Tobin’s Q	−0.154 ***	0.022 ***	−0.088 ***	−0.001 ***
	(−4.27)	(−3.22)	(−4.27)	(−3.22)
Top 1	0.018 *	−0.006 ***	−0.003 *	−0.008 ***
	(−1.69)	(−9.59)	(−1.69)	(−9.59)
ROE	1.308	−0.796 ***	0.004	−0.031 ***
	(0.48)	(−3.06)	(0.48)	(−3.06)
Size	0.000 ***	−0.000 ***	0.000 ***	−0.000 ***
	(18.64)	(−5.83)	(18.64)	(−5.83)
Age	−0.081	0.008 ***	−0.002	0.017 ***
	(−0.51)	(10.11)	(−0.51)	(10.11)
Boardsize	−0.124	0.110 ***	0.001	0.119 ***
	(0.10)	(35.47)	(0.10)	(35.47)
SOE	0.501 ***	−0.485 ***	0.190 ***	−0.481 ***
	(3.08)	(−17.04)	(3.08)	(−17.04)
Indsh	0.000 ***	−0.003 ***	0.033 ***	−0.003 ***
	(25.24)	(−6.40)	(25.24)	(−6.40)
Dual	−0.003	−0.032	0.071	−0.031
	(1.26)	(−1.27)	(1.26)	(−1.27)
Year	Control	Control	Control	Control
Industry	Control	Control	Control	Control
Constant	8.083	−0.428	−4.183	−1.077
	(20.78)	(−2.43)	(−9.70)	(−6.64)
R2	0.194	0.067	0.194	0.067
Observations	43,439	43,439	43,439	43,439

Note: z-statistics are shown in brackets. *** p < 0.01; * p < 0.1.

presents the results of the mechanism analyses in this study. In order to test the mediating effect of internal control quality, the internal control index (DBI) of listed companies issued by Dibble Company is used to measure the quality of corporate internal control. The specific path of the role of listed company directors’ network positions and corporate fraud tendency is tested, and the regression results are shown in Column (1). The regression coefficient of Centrality and DBI is 0.174, and it is significant at the 1% level, which indicates that the network location of directors of listed companies increases the level of internal control of the company. According to Column (2), the coefficient of Centrality and corporate fraud is −0.041 and significantly negatively correlated at 1%, and the coefficient of DBI and corporate fraud is −0.128 and significantly negatively correlated at 1%, indicating that corporate fraud is inhibited by improving the quality of internal control. The regression results are consistent with Hypothesis 2.

To detect the mediating effect of external audit quality, this paper measures external audit quality in terms of the Big 4 international accounting firms (Big4). According to the results reported in Column (3), the coefficient between director network position and external audit quality is 0.074, which is significantly positive at the 1% level, indicating that firms with the advantage of director network position have higher external audit quality. Based on the regression results shown in Column (4), the coefficient of Centrality with corporate irregularities is −0.055 and significantly negative at the 1% level, and the coefficient of Big4 with corporate fraud is −0.239 and significantly negative at the 1% level, indicating that directors’ network position inhibits corporate fraud by improving the quality of external audit. The regression results are consistent with Hypothesis 3.

4.4. Endogeneity Test

(1) The Heckman two-stage method is used to eliminate the endogeneity problem caused by selection bias in the sample. In the first stage, the dependent variable is the dummy variable Centrality_dum, which equals 1 if the director’s network position (Centrality) is greater than the annual median of the indicator, and 0 otherwise. The Probit model is used for regression analysis to calculate the Ni–Mills Ratio (Imr). In the second stage, Imr is added to the model as a control variable for regression analysis. As shown in column 1 of

Table 6. Endogeneity test results.

Variable	Fraud		Fraud		Fraud	Fraud	Fraud
	Heckman		Instrumental Variable		Lag One Phase	Fixed Effect	Small Company
	(1)		(2)	(3)	(4)	(5)	(6)
Centrality		−0.056 ***	−1.394 ***	−0.024 ***	−0.058 ***	−0.012 ***	−0.026 *
		(−6.41)	(−17.56)	(−3.48)	(−6.26)	(−6.55)	(−1.65)
Lev	0.009 *	0.108 ***	−0.010	0.005 **	0.379 ***	0.005 **	0.009
	(1.77)	(3.66)	(−1.33)	(2.28)	(8.11)	(2.42)	(0.57)
Tobin’s Q	0.005	−0.004 ***	0.000	−0.000	−0.002	−0.000 *	−0.000
	(1.50)	(−6.83)	(0.48)	(−1.60)	(−1.29)	(−1.67)	(−0.66)
Top 1	0.001	−0.008 ***	−0.002 ***	−0.002 ***	−0.008 ***	−0.002 ***	−0.008 ***
	(1.26)	(−9.90)	(−3.60)	(−9.86)	(−8.66)	(−9.74)	(−4.35)
ROE	−0.002	−0.030 ***	0.004 *	−0.001 **	−0.024 **	−0.001 ***	−0.013
	(−1.19)	(−2.93)	(1.88)	(−2.57)	(−2.56)	(−2.69)	(−1.62)
Size	−0.000 ***	−0.000 ***	0.000 ***	−0.000 ***	−0.000 ***	−0.000 ***	−0.000
	(−3.21)	(−5.94)	(2.99)	(−5.48)	(−6.68)	(−5.79)	(−0.22)
Age	−0.010 ***	0.023 ***	0.015 ***	0.004 ***	0.015 ***	0.004 ***	0.037 ***
	(−10.16)	(12.93)	(10.91)	(10.71)	(8.72)	(10.73)	(10.23)
Boardsize	−0.059 ***	0.154 ***	0.110 ***	0.026 ***	0.091 ***	0.025 ***	0.118 ***
	(−27.56)	(23.03)	(21.55)	(34.14)	(27.81)	(39.29)	(16.16)
SOE	−0.215 ***	−0.355 ***	0.223 ***	−0.099 ***	−0.430 ***	−0.103 ***	−0.388 ***
	(−12.89)	(−10.39)	(7.63)	(−15.96)	(−14.78)	(−17.35)	(−6.00)
Indsh	−0.003 ***	−0.002 **	0.004 ***	−0.001 ***	−0.004 ***	−0.001 ***	−0.001
	(−11.34)	(−2.37)	(7.93)	(−6.44)	(−6.76)	(−7.10)	(−1.46)
Dual	0.019	−0.042 *	−0.022	−0.008	−0.009	−0.007	−0.118 **
	(1.36)	(−1.76)	(−1.16)	(−1.61)	(−0.38)	(−1.39)	(−2.44)
IMR		0.587 ***
		(6.04)
Year	Control	Control	Control	Control	Control	Control	Control
Industry	Control	Control	Control	Control	Control	Control	Control
Constant	1.518	−1.972	3.022	0.307	−0.889	0.338	−2.159
	(15.20)	(−10.72)	(14.25)	(8.02)	(−5.27)	(10.11)	(−6.55)
R2	0.067	0.067	0.000	0.081	0.048	0.064	0.066
Observations	43,439	43,439	43,439	43,439	38,244	43,345	10,824

Note: z-statistics are shown in brackets. *** p < 0.01; ** p < 0.05; * p < 0.1.

, the coefficient of Imr is significant at the 1% level, indicating a problem of selection bias in the sample. It is necessary to consider the estimation bias that may be caused by this problem, and the regression results show that Centrality is still significantly negatively correlated with corporate fraud at the 1% level, suggesting that the director’s network position inhibits the occurrence of corporate fraud in listed companies.

(2) To solve the endogeneity problem caused by omitted variables, we use instrumental variables for the test and conducted weak instrumental variable tests for methodological verification. Considering the difficulty of determining appropriate instrumental variables (IVs) in social network analysis, we adopt the method proposed by Hu et al. [50] and Hu et al. [51]. This method utilizes the prediction residuals of director network positions, and selected corporate fraud as the explanatory variable. These residuals are unrelated to corporate fraud and other control variables but are highly associated with directors’ network positions. Therefore, they are used as the instrumental variables in this study. The results presented in column (2) of Table 6 are consistent with those of the baseline regression. Besides, the mean value of the directors’ network position of companies in the same industry and the same year, is employed as another instrumental variable. As shown in column (3) of Table 6, Centrality is significantly negatively correlated with Fraud at the 1% level, which is consistent with the results of basic regression.

(3) To mitigate the possibility of reverse causality, a lag is imposed on both the independent and control variables. As shown in column (4) of Table 6, Centrality exhibits a significant negative correlation at the 1% level after one lag period, consistent with the results observed in the basic regression.

(4) To excluding the problem of firm-level omitted variables, this paper controls the individual fixed effect at the company level. As shown in column (5) of Table 6, after controlling for individual fixed effects, the regression coefficient of Centrality is −0.012, which is significant at the 1% level. This indicates that after controlling for individual fixed effects, the inhibitory effect of listed company directors’ network position on financial fraud remains significant, supporting Hypothesis 1.

(5) To further address the endogeneity problem caused by sample self-selection bias, we use the previous methodology in the research conducted by Intintoli et al. [52] and Hu et al. [51]. Small firms are defined as companies in the bottom quartile by firm size, and their samples are analyzed using the benchmark regression model. The results in column (6) of Table 6 indicate that the relationship between directors’ network positions and corporate fraud is not driven by directors’ self-selection.

4.5. Robustness Test

(1) Replacement of Measurement of Explained Variables: This study re-measures corporate fraud by taking the logarithm of the number of corporate fraud events plus one. The regression results are shown in column 5 of

Table 7. Robustness test results.

Variable	Fraud				Fraud Event	Fraud	Fraud
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Centrality					−0.021 ***	−0.051 ***	−0.053 ***
					(−15.89)	(−4.85)	(−5.56)
Centrality-b	−0.010 ***
	(−2.62)
Degree		−0.165 ***
		(−6.23)
Closeness			−1.568 ***
			(−5.24)
Betweenness				−119.265 ***
				(−5.19)
Lev	0.759 ***	0.188 ***	0.190 ***	0.193 ***	0.008 ***	0.181 ***	0.057 **
	(12.84)	(5.37)	(5.40)	(5.45)	(5.27)	(3.94)	(2.53)
Tobin’s Q	0.033 ***	−0.002 ***	−0.002 ***	−0.002 ***	−0.000 ***	0.003	−0.001 **
	(4.26)	(−4.68)	(−4.69)	(−4.75)	(−2.70)	(0.68)	(−1.99)
Top 1	−0.006 ***	−0.012 ***	−0.012 ***	−0.012 ***	−0.001 ***	−0.009 ***	−0.007 ***
	(−7.42)	(−12.49)	(−12.28)	(−12.19)	(−6.07)	(−8.54)	(−7.29)
ROE	−0.992 ***	−0.027 ***	−0.027 ***	−0.027 ***	−0.002 ***	−0.017 *	−0.019 **
	(−14.95)	(−2.87)	(−2.93)	(−2.95)	(−5.23)	(−1.95)	(−2.10)
Size	−0.000 ***	−0.000 ***	−0.000 ***	−0.000 ***	−0.000 ***	−0.000 ***	−0.000 ***
	(−9.22)	(−7.76)	(−7.82)	(−7.73)	(−6.04)	(−5.61)	(−6.18)
Age	0.001 ***	0.027 ***	0.026 ***	0.026 ***	0.004 ***	0.015 ***	0.015 ***
	(6.11)	(13.75)	(13.39)	(13.35)	(15.09)	(6.92)	(7.90)
Boardsize	0.113 ***	0.173 ***	0.168 ***	0.167 ***	0.036 ***	0.121 ***	0.125 ***
	(32.77)	(44.90)	(45.47)	(45.37)	(79.48)	(27.88)	(33.81)
SOE	−0.521 ***	−0.719 ***	−0.722 ***	−0.727 ***	−0.104 ***	−0.427 ***	−0.487 ***
	(−18.32)	(−21.55)	(−21.67)	(−21.85)	(−24.30)	(−11.79)	(−15.79)
Indsh	−0.003 ***	−0.005 ***	−0.005 ***	−0.005 ***	−0.001 ***	−0.003 ***	−0.003 ***
	(−6.37)	(−8.27)	(−8.15)	(−8.04)	(−7.96)	(−4.32)	(−5.99)
Dual	−0.032 **	−0.023	−0.022	−0.022	−0.003	−0.028	−0.031
	(−1.33)	(−0.81)	(−0.79)	(−0.79)	(−0.07)	(−0.96)	(−1.15)
Year	Control	Control	Control	Control	Control	Control	Control
Industry	Control	Control	Control	Control	Control	Control	Control
Constant	−1.421	−2.014	−2.238	−2.287	−0.127	−1.462	−1.067
	(−8.60)	(−10.74)	(−12.32)	(−12.62)	(−5.01)	(−11.14)	(−6.07)
R2	0.073	0.155	0.155	0.155	0.179	0.061	0.058
Observations	43,338	43,439	43,439	43,439	43,345	28,297	34,004

Note: z-statistics are shown in brackets. *** p < 0.01; ** p < 0.05; * p < 0.1.

, where the regression coefficient for the director’s network position is significantly negative, consistent with the previous basic regression results.

(2) Replacement of Measurement of Explanatory Variables: Replacement explanatory variables including the mean semi-local centrality of the bottom nodes of the directors’ network (Centrality-b) and the three most common centrality indexes proposed by Freeman [40] are used respectively. These three centrality indexes are degree centrality (Degree), closeness centrality (Closeness), and betweenness centrality (Betweenness), which are calculated by transforming the two-mode model of director network into one-mode model (director–director) based on projection technology. As shown in columns 1–4 of Table 7, the regression coefficients of centrality after replacing the explanatory variables are all significantly negatively correlated, and all are consistent with the results of the previous basic regression.

(3) To further eliminate the influence of industry factors on the regression results, we retain only the annual sample of manufacturing listed companies and re-examine them. The test results, as shown in column (6) of Table 7, indicate that directors’ network position is significantly negatively correlated with corporate fraud, which is basically consistent with the previous conclusion.

(4) To exclude the impact of the COVID-19 pandemic on this study, we exclude the sample data of companies from 2020 to 2022, change the sample interval to 2009–2019, and re-examine the tests. The regression results, shown in column (7) of Table 7, are basically consistent with the previous basic regression results.

5. Further Analysis

The previous study explored the impact of independent directors’ network positions on corporate fraud, without considering the entire director network and distinguishing between director types. It mainly examined the impact of independent directors’ network positions on corporate fraudulent behavior. Directors include both non-independent and independent directors. Non-independent directors usually come from within the company, mostly corporate executives, who meet more often, interact more frequently, and form strong intra-corporate relationships with each other. In contrast, independent directors usually originate from universities, accounting and legal intermediaries, industry associations, etc. They meet and interact less frequently, and the relationships between inside directors and independent directors, as well as among independent directors, are usually weakly linked [53]. Due to the differences in the characteristics and relationships of different types of directors in the director network, their influence on fraudulent behavior in listed companies may vary.

Drawing on the studies of Liang et al. [54] and Lin et al. [55], this study constructs the independent director network and non-independent director network of listed companies, respectively. It uses the semi-local centrality of the top nodes in the independent director network and the non-independent director network as measures of network position and places them in the full sample for testing. The results in column (1) of

Table 8. Results of further analyses of director network types.

Variable	Fraud	Fraud	Fraud
	Non-Independent Directors Network (1)	Independent Directors Network (2)	Women Directors Network (3)
Centrality	−1.161 ***	−0.955 ***	−0.049 ***
	(−21.05)	(−19.01)	(−2.60)
Lev	0.202 ***	0.231 ***	0.191 ***
	(5.57)	(6.10)	(5.40)
Tobin’s Q	−0.002 ***	−0.002 ***	−0.002 ***
	(−4.86)	(−5.41)	(−4.70)
Top 1	−0.013 ***	−0.014 ***	−0.012 ***
	(−13.08)	(−14.20)	(−12.11)
ROE	−0.030 ***	−0.030 ***	−0.027 ***
	(−3.15)	(−3.15)	(−2.95)
Size	−0.000 ***	−0.000 ***	−0.000 ***
	(−6.20)	(−6.57)	(−8.02)
Age	0.033 ***	0.025 ***	0.025 ***
	(16.43)	(13.45)	(13.06)
Boardsize	0.209 ***	0.202 ***	0.167 ***
	(46.09)	(46.47)	(45.32)
SOE	−0.608 ***	−0.598 ***	−0.729 ***
	(−17.83)	(−17.90)	(−21.88)
Indsh	−0.004 ***	−0.004 ***	−0.005 ***
	(−5.80)	(−6.04)	(−8.16)
Dual	−0.026	−0.023	−0.020
	(−0.94)	(−0.82)	(−0.71)
Year	Control	Control	Control
Industry	Control	Control	Control
Constant	0.627	0.248	−2.329
	(2.73)	(1.11)	(−12.87)
R2	0.116	0.115	0.115
Observations	42,630	42,627	43,312

Note: z-statistics are shown in brackets. *** p < 0.01.

show that the coefficient of the non-independent directors’ network position (Centrality) is −1.161 and is significant at the 5% level. The coefficient of the independent director network location (Centrality) in column (2) is −0.955 and is significant at the 5% level. These results indicate that both independent directors’ and non-independent directors’ network locations have inhibitory effects on corporate fraud in listed companies. Independent and non-independent directors with extensive connections are affected by the reputation effect and actively supervise corporate behaviors to avoid fraudulent activities in the companies they serve.

The diversity of director gender is highly associated with corporate governance and disclosure [56,57]. Abdullah et al. found the importance of female directors on the board for promoting environmental sustainability disclosure [58]. The presence of female directors may reduce the probability of corporate fraud occurring [53,59,60,61]. Nekhili et al. show that a higher proportion of female independent directors leads to greater board oversight, which further reduces audit risk [62]. Eugster et al. find that the corporate female director ratio can reduce certain types of corporate fraud [1]. Women are more risk-averse than men in financial decision-making and are less tolerant of litigation and reputational damage. Thus, we expect female director network positions to similarly inhibit corporate fraud.

We construct a female director network based on the entire director network (firm–director), retaining only the bottom node as a female director and using the semi-local centrality of the top node as a measure of network location. This metric reflects the number of firms to which the firm is directly associated through female directors. The results in column (3) of Table 8 show that the coefficient of the network position of female directors (Centrality) is −0.049 and is significant at the 5% level, which is consistent with Hypothesis 1. This indicates that the closer the network location of female directors is to the center, the lower the probability of corporate fraud occurring.

Considering that the busy director hypothesis is not applicable to this study, further analyses are conducted in this paper. Drawing on Ferris et al. [33], who define a director with three or more directorships as a busy director, we counted the number of bottom nodes (directors) connected to each top node (company) with a degree of node greater than or equal to three as the number of busy directors of the company based on the bipartite graph structure of the director network. The logarithmic form of the number of busy directors of the company is used as an alternative explanatory variable for the benchmark regression analysis.

Table 9. Results of further analyses of the busy director hypothesis.

Variable	Fraud	Fraud	Fraud
	(1)	(2)	(3)
Centrality	−0.019 **	−0.003	−0.070 ***
	(−2.21)	(−0.29)	(−3.90)
Lev	0.114 ***	0.114 ***	0.113 ***
	(3.86)	(3.86)	(3.85)
Tobin’s Q	−0.001 ***	−0.001 ***	−0.001 ***
	(−3.25)	(−3.25)	(−3.23)
Top 1	−0.008 ***	−0.008 ***	−0.008 ***
	(−9.57)	(−9.55)	(−9.63)
ROE	−0.031 ***	−0.032 ***	−0.032 ***
	(−3.09)	(−3.10)	(−3.10)
Size	−0.000 ***	−0.000 ***	−0.000 ***
	(−6.85)	(−6.90)	(−6.76)
Age	0.016 ***	0.016 ***	0.016 ***
	(10.01)	(9.95)	(9.99)
Boardsize	0.117 ***	0.116 ***	0.117 ***
	(34.46)	(34.05)	(35.23)
SOE	−0.493 ***	−0.494 ***	−0.491 ***
	(−17.53)	(−17.57)	(−17.42)
Indsh	−0.004 ***	−0.004 ***	−0.004 ***
	(−7.14)	(−7.25)	(−6.94)
Dual	−0.030	−0.030	−0.030
	(−1.26)	(−1.25)	(−1.26)
Year	Control	Control	Control
Industry	Control	Control	Control
Constant	−1.201	−1.196	−1.196
	(−7.44)	(−7.41)	(−7.41)
R2	0.066	0.066	0.066
Observations	43,338	43,338	43,338

Note: z-statistics are shown in brackets. *** p < 0.01; ** p < 0.05.

presents the results of the regression analysis of busy directors affecting corporate fraud for the entire director network, independent director network, and non-independent director network. Column 1 of Table 9 reflects the regression for all busy directors with a Centrality coefficient of −0.019, which is significantly negatively correlated at the 5% level. Column 2 reflects the regression for independent busy directors with a Centrality coefficient of −0.003, which is not significant. Column 3 reflects the regression of non-independent busy directors with a Centrality coefficient of −0.070, which is significantly negatively related at the 1% level. These results indicate that neither directors, independent directors, nor non-independent directors lead to an increase in the probability of corporate fraud occurring due to an increase in busyness.

6. Discussion and Conclusions

Preventing corporate fraud is key to the pursuit of sustainable development. Existing studies have explored the impact of independent directors’ network positions on corporate fraud, mainly through independent director networks. This paper expands the scope of the study by constructing entire director networks, independent director networks, non-independent director networks, and female director networks to provide empirical evidence on the impact of different types of director network positions on corporate fraud. Additionally, it highlights the mediating roles played by the quality of external audits and the quality of internal controls.

This paper validates the effect of directors’ network position on corporate fraud behavior using directors’ social networks as an entry point. It is found that director networks significantly reduce the probability of corporate fraud. This confirms that social networks play a greater role in the governance effect of corporate directors, as these network relationships can help directors perform their duties better by improving their motivation and ability to monitor through the reputation mechanism and the advantage of information resources. Mechanism analysis shows that internal control and external audit play a mediating role in the effect of directors’ network on corporate fraud. This suggests that well-connected directors are able to increase their monitoring role on corporate behavior by upgrading internal controls and hiring higher quality auditors. Further analysis reveals that the network positions of independent, non-independent, and female directors all significantly reduce the probability of corporate fraud. This suggests that well-connected independent, non-independent, and female directors will actively fulfill their monitoring role to avoid corporate fraud in the companies they work for. In addition, the regression analyses conducted specifically for the “busy director hypothesis” do not support the hypothesis, suggesting that directors do not reduce the quality of their oversight of the firm by serving in multiple firms, which would lead to an increase in the probability of corporate fraud. The above conclusions indicate that director networks exert complementary effects on board independence, female directors, and busy directors. Firstly, director independence is often recognized as an important mechanism in promoting the capability of board monitoring. However, independent directors may lack sufficient internal information and resources, limiting their monitoring effectiveness. In contrast, well-connected independent directors can access more information and resources through their extensive networks, thus enhancing their monitoring capabilities. Hence, director networks and director independence may have complementary rather than substitutive effects. Secondly, female directors play an important role in enhancing corporate governance diversity and decision-making quality. The central position of a female director in networks further increases her influence and accessible resource, which effectively facilitates the inhibition of corporate fraud. Thus, director networks and female directors may also have complementary effects. Thirdly, busy directors are often perceived to have limited monitoring capabilities due to serving on multiple boards. However, highly connected directors, even if they are busy, can compensate for their lack of time and energy with critical information and experience provided by their extensive networks. Thus, there may be a complementary relationship between director networks and busy directors. In conclusion, the results in this study suggest that directors’ network positions enhance the monitoring effect of the board on corporate fraud.

In an economic context dominated by social relations, the director network, as an informal institutional arrangement, is one of the forms of management organizations spontaneously formed by the company. It can help the company enhance its earning capacity and prompt the rational allocation of resources, and this positive effect has been confirmed in much of the literature. Based on the micro level, this paper explores the impact of directors’ network position on corporate fraud in order to provide certain insights for government regulators and enterprises.

Firstly, it is crucial to correctly understand the governance role of director networks. The director network is not only the main external channel for inter-firm information transfer but also an effective external governance mechanism. The company should optimize the appointment mechanism of board members and properly establish a well-connected director network so that enterprises can access quality resources and information, which is conducive to strengthening corporate governance.

Secondly, establishing a sound internal control system is essential. To a certain extent, the external governance system needs to influence corporate governance through the internal governance system. Therefore, the combination of external and internal governance can effectively enhance the governance role. The company should improve the internal control system and guarantee the quality of internal control to promote the active role of the director network, thereby improving the effect of corporate governance and regulating corporate behavior.

Thirdly, it is crucial to improve the quality of external audits. High-quality external audits enhance the transparency and reliability of financial reporting through independent and objective audit procedures. With the assistance of the reputation and social capital of well-connected directors in their networks, enterprises can select an external auditor with a high degree of credibility and experience, which greatly reduces the incidence of corporate fraud.

Lastly, the government should strengthen legislative construction and create a fair business environment. A good legal environment and a fair business environment can guide the cooperative subjects in the social relationship network towards the goal of promoting the legality of regulating corporate behavior, giving full play to the role of the directors’ network in improving the quality of the company.

This paper has limitations that future research should address and overcome. First, the scope of this study is limited to the Chinese market due to data collection limitations. However, future research should look beyond China and compare the impact of director network positions on corporate fraud in different countries to gain a broader understanding. Secondly, this study focuses on measuring director network characteristics from a static perspective, which is useful for understanding the structure of the director network in a particular year but does not consider the fact that the director network is in a constant process of change over time. This limits the ability to analyze the impact of the director network on corporate fraud from a dynamic perspective. Finally, this study discusses corporate fraud but does not delve into the specific differences between disclosure fraud and non-disclosure fraud. Corporate fraud is a vast area, and future research should explore in more depth how directors’ network positions play a role in different types of corporate fraud.