Personal Networks, Board Structures and Corporate Fraud in Japan

Abstract

We examine the impact of corporate governance and personal networks on corporate fraud in Japanese companies, using panel logit and Cox proportional hazard models to analyze fraud occurrence and detection. This study focuses on the effects of Japan’s recent corporate governance reform and explores the unique influence of personal networks. Our key findings indicate that recent changes in corporate governance in Japan have been effective in preventing the occurrence of fraud and accelerating its detection. Additionally, stronger personal networks among board members help prevent fraud concealment, highlighting cultural differences in the effectiveness of personal networks in corporate governance compared to findings from Europe and the US.

1. Introduction

In recent years, empirical research on corporate fraud has increased substantially due to the accumulation of diverse data and heightened interest in corporate fraud as a socio-economic problem. Notable scandals, such as the Enron scandal in the United States (October 2001)1 and the Toshiba accounting scandal in Japan (May 2015)2, have drawn significant attention worldwide. Corporate fraud extends beyond accounting manipulations like window dressing, encompassing a wide range of deceptive practices. Examples include the LIBOR scandal involving Barclays Bank in the United Kingdom (2012), the Volkswagen emissions scandal in Germany (September 2015), and the vehicle test fraud at Japanese major car companies Toyota, Mazda, Yamaha, Honda and Suzuki (June 2024). These frauds have had far-reaching impacts, often affecting the global economy.

Corporate fraud is a significant social problem and an important research subject in economics. Corporate wrongdoing undermines investor confidence, reduces shareholder value, misallocates capital and destabilizes financial markets (Khanna et al. 2015). These consequences have driven economists to investigate the causes of corporate fraud.

Using data from Japanese listed companies, this paper empirically examines how the structures of boards of directors and personal networks among board members affect the occurrence and detection of corporate fraud. Specifically, we use regression analysis to analyze the relationship between corporate governance indicators and personal network indices based on board members’ connections and the occurrence and detection of corporate fraud.

This study has three unique characteristics. First, we focus on two distinct stages of corporate fraud: occurrence and detection (or concealment). While most previous studies have concentrated only on the occurrence of corporate fraud, it is equally important to identify governance structures that enable the prompt detection of fraud, thereby preventing its concealment.3 For example, the top management of Volkswagen AG was blamed for concealing the company’s emissions test irregularities.

Secondly, we focus on a unique Japanese personal network: alma mater. The empirical research on the impact of personal networks among board members on corporate behavior is a relatively new field, with the literature accumulating mainly in major financial journals since the 2010s (Cohen et al. 2008; Fracassi and Tate 2012; Kramarz and Thesmar 2013; Khanna et al. 2015; El-Khatib et al. 2015; Fracassi 2017; Schoenherr 2019). Various studies have examined the effects of personal networks on financial and economic behavior, identifying both positive and negative impacts. However, further empirical evidence is needed, especially regarding the impact of Japanese personal networks on financial and economic activities.4

Various factors generate personal networks in our economic society, such as place of birth, alma mater, previous workplace and hobbies.5 As Amano (2005) pointed out, educational background (alma mater) can affect labor market outcomes such as employment and promotion in Japanese society, but there is little empirical evidence to support these predictions. Kawaguchi and Ma (2008), using data from natural experiments, is one of the few empirical studies showing that educational background affects promotion in public offices. In this study, we estimated sets of personal network indicators based on board members’ alma maters and analyzed the impact of these networks on the occurrence and detection (or concealment) of corporate fraud. Our empirical analysis of financial and economic behavior using Japanese data provides valuable evidence for understanding the economic role of alumni networks in societies.6

Thirdly, using eight different indicators of corporate governance, we examined the impact of the corporate governance reform implemented by the Japanese government in recent years on corporate fraud. For example, the amendment of the Companies Act in 2015 is one of the major reforms. Under this reform, “Company with Audit & Supervisory Committee (監査等委員会設置会社: kansa tou iinkai sechigaisha)” was introduced as a new form of governance which listed companies can elect.7 As a result, there are now three different corporate governance forms from which listed companies can choose: “Company with Audit & Supervisory Board (監査役会設置会社: kansayakukai secchigaisha)” introduced in 1993, “Company with a Nominating Committee, etc. (指名委員会等設置会社: shimei iinnkai tou secchigaisha)” introduced in 2003,8 and “Company with Audit & Supervisory Committee.” There have been active discussions on transitioning from “Company with Audit & Supervisory Board”, commonly adopted by traditional Japanese companies, to the other forms.9 This paper examines the differences in the impact of the three governance forms on fraud.

Government and institutional investors have also demanded an increase in the ratio of outside and independent directors. The Japanese government asserts that boards of directors will better fulfill their functions by appointing outside and independent directors. However, there has been no quantitative research on the relationship between these corporate governance reforms and corporate fraud in Japan. We analyze whether these measures are effective in preventing the occurrence and detection (or concealment) of fraud.

Our results are as follows: Recent changes in corporate governance reform in Japan have generally been meaningful in terms of preventing the occurrence of fraud and accelerating the detection (and preventing the concealment) of fraud. The probability of fraud occurring can be reduced by decreasing the number of directors and by changing the form of a board of directors to a “Company with a Nominating Committee, etc.” Regarding detection, companies with higher rates of independent directors are more likely to detect fraud; in other words, they do not tend to conceal fraud. Under “Corporate Governance Reform”, the Japanese government has urged companies to reform their boards by reducing board size to improve their efficiency, separating management supervision and execution and increasing the ratio of independent directors. These reforms have improved corporate governance in Japan in terms of preventing fraud occurrence and preventing its concealment.

These findings also suggest that the mechanisms for fraud occurrence and detection differ, implying that corporate governance measures to prevent fraud occurrence may differ from those needed to prevent fraud concealment. This highlights the need for researchers to re-examine the impact of corporate governance on the two distinct aspects of corporate fraud: occurrence and detection (or concealment), which have not been clearly distinguished and discussed previously.

Regarding the analysis of personal networks, our results show that stronger alumni networks among board members are more likely to prevent fraud concealment. A board of directors with a strong personal network has the potential to mitigate information asymmetry and prevent concealment compared to a board without such a network. This finding contrasts with previous studies in Europe and the United States, which have empirically demonstrated that strong personal networks reduce the expected costs of corporate fraud and increase its probability. This implies that the impact of personal networks on corporate fraud may differ among countries, societies, and cultures.

Our paper is organized as follows. In Section 2, we explain the characteristics of the data used in this analysis and review previous research, comparing their data with ours. Section 3 describes how we quantified personal network indicators and the empirical models for fraud occurrence and detection. Section 4 presents the results for personal network indicators and corporate governance indicators in the fraud occurrence and detection models. Finally, Section 5 concludes the paper.

2. Data

In light of the objective of this paper to quantitatively analyze the impact of the structure of boards of directors and personal networks on the occurrence and detection of fraud, this section describes the characteristics of our data (fraud data, personal network indicators and corporate governance indicators) in comparison with previous studies.

2.1. Fraud Data

There is no universal definition of corporate “fraud”.10 Previous research has used various definitions depending on the research objectives. For example, some studies follow conceptual definitions, while empirical studies typically use keyword-based definitions.11 Specifically, quantitative analyses often rely on fraud datasets containing fraudulent corporate cases identified by specific keywords from published articles in newspapers, etc.12 Previous research utilizes different keywords. While most U.S. studies post-Enron focus on financial reporting irregularities, Japanese studies cover a broader range of frauds. Beasley (1996) extracted cases from “financial statement frauds” publicly reported by the SEC (Securities and Exchange Commission) and those with the headline “Crime-White Collar Crime” in the WSJ index (Wall Street Journal Index).13 Nakamura (2001) and Kobayashi et al. (2010) are two previous empirical studies on corporate fraud in Japan. Nakamura’s study includes cases not necessarily regarded as fraud, such as “environmental pollution” and “destruction of nature”, while Kobayashi et al. focus on specific keywords, excluding events not necessarily regarded as fraud, such as “corporate ethics.”14

This paper, following Kobayashi et al. (2010), uses the following 15 keywords to extract cases from published articles (from two databases: Nikkei Telecom 21 and FCG Research Institute, Inc.) from January 2014 to August 2017: “Bid-rigging”, “Misrepresentation”, “Accident”, “System trouble”, “Unpaid overtime”, “Violation of the Waste Disposal and Public Cleansing Law”, “Fraudulent accounting”, “Income concealment”, “Benefit provision”, “Cartel”, “Insider”, “Embezzlement”, “Misappropriation”, “Recall” and “Information leak.”15 We construct a fraud dataset that includes 731 cases, covering the period before and after the introduction of the revised Companies Act in 2015.16

Our fraud dataset records the time of the occurrence and detection of each fraud case, which helps to clarify the “latent period” of each case.17 While there is limited empirical research on corporate fraud in Japan, existing studies often neither focus on the detection of fraud nor clearly distinguish between occurrence and detection. Kobayashi et al. (2010), one of the few empirical studies on corporate fraud in Japan, regarded the date of the first media report as the occurrence of fraud. However, this date is not necessarily the same as the date of detection, which is based on the publication date of the news article. Since their research interest lies in the relationship between the reporting of fraud and stock prices, the distinction between fraud occurrence and detection was not considered crucial. Aoki (2015)’s dataset is also unclear about whether it pertains to occurrence or detection, even though the research examines the relationship between corporate fraud occurrence and corporate governance. It may be more effective to clearly distinguish between the occurrence date and the detection date to study fraud occurrence more accurately.18

It is likely that there are two types of detection, depending on who detects the fraud: detection by the media (public) and detection by legal authorities. The timing of these types of detection can vary; both may detect in the same month, the media may detect first followed by legal authorities, or vice versa. In this paper, we define detection as detection by the media (public).

The purpose of this paper is to analyze the effects of boards of directors’ structures and personal networks on the occurrence and detection of fraud. While previous empirical studies on preventing fraud concealment by boards of directors have not clearly distinguished between fraud detection and occurrence, our novel fraud dataset enables us to identify these two events distinctly.19

2.2. Personal Network Indicators

Individuals conduct their economic activities within various personal networks. These personal (or social) networks, which connect individuals to each other, are built upon factors such as place of origin, alma mater, previous workplace and hobbies. Empirical research examining the impact of personal networks among board members on corporate behavior is relatively new and has gained popularity in major financial journals since the 2010s. It has been empirically shown that CEOs with extensive networks are more likely to appoint new board members connected through these networks, which can have a detrimental effect on corporate values (Fracassi and Tate 2012; El-Khatib et al. 2015).

Regarding corporate fraud and CEO networks, Khanna et al. (2015) show that appointment-based CEO connectedness (the connections CEOs build through appointing executives and directors) increases the risk of corporate fraud.20 In other words, board appointments based on connections with CEOs increase the likelihood of fraud occurrence and reduce the likelihood of fraud detection. According to their analysis, personal networks reduce the expected cost of fraud by making it easier to conceal fraud, reducing the likelihood of a CEO being dismissed if fraud is detected and lowering the adjustment costs of executing fraud. As Khanna et al. (2015) state, appointment-based CEO connectedness within boards is noteworthy for regulators, investors and corporate governance professionals. Therefore, further analysis focusing on various personal networks within boards of directors is needed.

Our paper focuses on personal networks of “Graduate University (final academic background)” and “Prefecture of Origin”, based on information recorded in ‘Yakuin Data (Executive Staff Data)’ (Toyo Keizai Inc.). In Japan, the concept of academic cliques (alma mater networks by university clubs) is often discussed in business magazines, but its economic impact has not been academically analyzed. For example, DIAMOND Online (2019),21 a popular Japanese business magazine, provides an intriguing analysis of which universities and high schools foster strong ties among graduates in government and business communities and which academic cliques’ members achieve successful careers. However, this is not an academic analysis.22

According to Amano (2005), academic cliques have been influential in various industries in Japan since the Meiji Restoration in the mid-19th century, driven by social and economic changes. In other words, academic cliques (alma mater personal networks) are a unique aspect of personal networks in Japan.23 Another unique form of personal network we examine is the “prefecture of origin.” In Japan, strong connections often exist between people from the same hometown. For example, “Kenjin-kai” (prefectural associations) members support each other in business and daily life outside their hometown. Additionally, because ‘Yakuin Data (Executive Staff Data)’ does not contain data on the high schools from which board members graduated, we use “prefecture of origin” as a proxy for this information.24 This paper examines the impact of alma mater (based on the university from which the highest degree was obtained) and same hometown (prefecture of origin) personal networks within boards of directors on the occurrence and detection of corporate fraud.

It is an empirical question whether personal networks among board members prevent or encourage fraud occurrence or speed up or delay fraud detection. Strong personal networks can mitigate information asymmetry, prevent fraud occurrence and hasten fraud detection. However, as Khanna et al. (2015) have shown, personal networks can also lead to fraud by lowering the expected costs of fraud and hampering proper decision-making, thereby facilitating its concealment.

2.3. Corporate Governance Indicators

In recent years, the necessity of “Corporate Governance Reform” has been emphasized in Japan. As a measure based on the Corporate Governance Code announced in 2015, it has been advocated that boards of directors should enhance their functions by appointing outside directors and independent directors. The importance of dialogue between institutional investors and companies, as highlighted by the Stewardship Code (established in 2014 and revised in 2017), has also been recognized. Furthermore, the Cabinet Office Ordinance (issued in 2019) has progressively enhanced the disclosure of performance-linked remuneration and the remuneration amounts for executive staff, as specified in the Annual Securities Report.

In addition, due to the “Company with Audit and Supervisory Committee” system introduced in the amended Companies Act of 2015, there are now three forms of governance for listed companies in Japan. There have been active discussions on transitioning from “companies with board of corporate auditors”, commonly adopted by traditional Japanese companies, to the other forms. Furthermore, institutional investors have also demanded that the ratio of outside and independent directors be increased. However, there has been no quantitative research on the relationship between “Corporate Governance Reform” and corporate fraud in Japan, although this has been discussed from various perspectives since the beginning of the 2010s.

In this study, we use eight variables from the “data related to the corporate governance report” of Nikkei NEEDS (

Table 1. Corporate Governance Indicators.

Variables	Explanations
dmn	The number of directors
dmt	Governance system: “Company with Nominating Committee, etc.” = 1, The others = 0
dmt2	Governance system: “Company with Audit and Supervisory Committee” or “Company with Nominating Committee, etc.” = 1, Company with Audit and Supervisory Board = 0
dmc	Chairman of the Board of Directors: Outside Directors = 1, Others = 0
dmte	Term of office of director in the articles of incorporation (year)
dmoutr	Ratio of the number of outside directors to that of directors
dmoutindr	Ratio of the number of independent directors to that of directors
adr	Ratio of the number of Audit and Supervisory Board members and Audit Committee members to that of directors

).25 This study uses more corporate governance variables than previous studies. With some exceptions (Wang et al. 2010; Wang 2013; Khanna et al. 2015), few empirical studies on corporate fraud detection have examined the impact of governance and compensation systems. As evidenced by Volkswagen’s gas emissions scandal, there is a strong relationship between governance, compensation systems and fraud detection. Analyzing this relationship in a multifaceted manner is both academically and practically worthwhile.26

On the other hand, a number of studies examining fraud occurrence use corporate governance indicators (Beasley 1996; Abbott et al. 2000; Uzun et al. 2004; Farber 2005; Krishnan 2005; Khanna et al. 2015; Aoki 2015). Common explanatory variables include the ratio of outside directors, the ratio of outside directors on the audit committee, the number of years since the CEO’s appointment, and whether the CEO also serves as the chairman of the board. These studies report that a higher ratio of outside directors significantly suppresses the occurrence of fraud. Compared to previous studies, this paper analyzes the occurrence of corporate fraud using a larger set of corporate governance indicators.

3. Empirical Models

This section first describes the personal network indicators (quantification of personal networks) used for our analysis, and then describes an empirical model for the occurrence and detection of fraud.

3.1. Quantification of Personal Networks

For our empirical analysis, it is necessary to quantify the degree of personal connections among board members according to alma mater and hometown. In this paper, based on Jackson (2014), we create two indicators, Density and Mean Degree, for alma mater and prefecture networks.

Density is expressed by the following equation and has a value from 0 to 1, and the larger the number, the stronger the personal connections in the network. For example, if all board members are from the same university, then Density = 1; conversely, if all board members are from different universities, then Density = 0.

$$Density={\displaystyle \frac{Numberofactualconnections}{Numberofpotentialconnections}}$$

This concept is illustrated in Figure 1, which is a graphical representation of the personal network of alumni (alma mater) of a listed company’s board of directors.27 In Figure 1, the 14 nodes represent each member of the board of directors, and the links represent connections as alumni. The numbers assigned to the nodes represent the hierarchy within the board, with ① representing the CEO. In this figure, ①, ⑬ and ⑭ are connected by links, and ②, ③, ⑤ and ⑩ are also connected by links. This is an actual graph of a listed company. The former shows the connections between the University of Tokyo and the latter Waseda University. The board of directors has the University of Tokyo clique centered on the CEO and the Waseda University clique including two senior managing directors and one managing director. In this case, the actual number of connections is nine ($={\displaystyle \frac{3\ast (3-1)}{2}}+{\displaystyle \frac{4\ast (4-1)}{2}}$), and the potential number of connections is 91 ($={\displaystyle \frac{14\ast (14-1)}{2}}$), so Density = 0.099. In this way, based on the information about the university from which a director graduated and the prefecture in which a director was born and grew up, ${Density}_{school}$ and ${Density}_{home}$ were calculated, respectively.

The second metric, Mean Degree, is the average number of links a node has in a network.28 It represents the average number of people on the board of directors to which each member connects, and takes a value between 0 and n − 1 (n is the number of nodes). Referring to Figure 1 as an example, since four nodes have three links, three nodes have two links, and the remaining seven nodes do not have links, Mean Degree = 1.29 ($={\displaystyle \frac{3\ast 4+2\ast 3+0\ast 7}{14}}$) can be calculated. In this way, based on the information about the university from which a director graduated and the prefecture in which a director was born and grew up, ${MeanDegree}_{school}$ and ${MeanDegree}_{home}$ can be calculated, respectively. The larger the Mean Degree, the stronger the personal connections within the network. However, unlike Density, Mean Degree is an indicator that depends on the size of the board of directors (the number of directors), and we should keep this in mind when analyzing the estimation results.

3.2. Empirical Model for Fraud Occurrence

A panel logit model was used for the analysis of fraud occurrence. All listed Japanese companies from 2014 to 2017 were examined using following Equation (1):

$$y_{it}=\beta X_{it}+c_t+{\epsilon }_{it}$$

(1)

$$y_{it}=\left\{\begin{array}{c}1:afraudoccursatcompanyiinyeart\\ 0:afrauddid{n}^{’}toccuratcompanyiinyeart\end{array}\right.$$

where i represents the company, and t represents each year. $X_{it}$ includes explanatory variables (“Corporate Governance Indicators” and “Personal Network Indicators”) and their Summary Statistics and Correlation Matrix are shown in

Table 2. Summary Statistics.

Variable	Obs	Mean	Std. Dev.	Min	Max
Dependent Variable
fraud_dummy	32,968	0.013	0.114	0	1
(duration)	731	40.167	60.075	0	624
Variables related to Personal Network
Densityschool	12,126	0.326	0.379	0	1
Meandegreeschool	12,126	3.411	4.298	0	32
Densityhome	12,126	0.485	0.362	0	1
Meandegreehome	12,126	5.171	4.436	0	33
Variables related to the form of governance
dmn	32,968	7.798	3.076	1	30
dmt	32,968	0.054	0.226	0	1
dmt2	32,968	0.072	0.259	0	1
dmc	32,968	0.004	0.063	0	1
dmte	24,542	1.401	0.494	1	10
dmoutr	32,968	0.163	0.159	0	1
dmoutindr	32,968	0.095	0.132	0	1
adr	32,968	0.485	0.220	0	4

and

Table 3. Correlation Matrix.

	Fraud_Dummy	Density School	Meandegree School	Density Home	MEANDEGREE Home	dmn	dmt	dmt2	dmc	dmte	dmoutr	dmoutindr	adr
fraud_dummy	1.000
Densityschool	0.004	1.000
	(0.660)
Meandegreeschool	0.031	0.918	1.000
	(0.001)	(0.000)
Densityhome	0.004	0.659	0.609	1.000
	(0.700)	(0.000)	(0.000)
Meandegreehome	0.045	0.545	0.656	0.868	1.000
	(0.000)	(0.000)	(0.000)	(0.000)
dmn	0.100	−0.039	0.198	0.004	0.375	1.000
	(0.000)	(0.000)	(0.000)	(0.684)	(0.000)
dmt	−0.018	0.014	−0.015	0.013	−0.036	0.120	1.000
	(0.001)	(0.134)	(0.097)	(0.145)	(0.000)	(0.000)
dmt2	0.015	0.032	−0.011	0.030	−0.043	0.130	0.856	1.000
	(0.006)	(0.000)	(0.217)	(0.001)	(0.000)	(0.000)	(0.000)
dmc	0.010	0.047	0.045	0.033	0.033	0.023	0.002	0.078	1.000
	(0.086)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.759)	(0.000)
dmte	−0.033	−0.037	−0.053	−0.014	−0.027	−0.132	−0.225	−0.255	−0.050	1.000
	(0.000)	(0.001)	(0.000)	(0.216)	(0.016)	(0.000)	(0.000)	(0.000)	(0.000)
dmoutr	0.029	0.135	0.088	0.099	0.029	0.001	0.250	0.386	0.122	−0.225	1.000
	(0.000)	(0.000)	(0.000)	(0.000)	(0.001)	(0.823)	(0.000)	(0.000)	(0.000)	(0.000)
dmoutindr	0.032	0.074	0.051	0.059	0.022	0.055	0.342	0.410	0.112	−0.221	0.662	1.000
	(0.000)	(0.000)	(0.000)	(0.000)	(0.016)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)
adr	−0.037	0.003	−0.110	−0.029	−0.188	−0.660	−0.527	−0.616	−0.042	0.210	−0.237	−0.274	1.000
	(0.000)	(0.775)	(0.000)	(0.001)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)

Note: p-values are reported in parentheses.

, respectively. $c_t$ represents the time-fixed effect (dummy variables for each year) and ${\epsilon }_{it}$ represents the error term. As Table 3 shows, ${Density}_{school}$ (${Density}_{home}$) and ${MeanDegree}_{school}$ (${MeanDegree}_{home}$) are highly correlated, so we separately examined the effects of densities and mean degrees.

The variables dmt and dmt2 are also examined separately. These two variables examine the different effects of alternative governance systems. As discussed in the Introduction, there are now three different corporate governance forms in Japan from which listed companies can choose: “Company with Audit & Supervisory Board (監査役会設置会社: kansayakukai secchigaisha)”, introduced in 1993, “Company with a Nominating Committee, etc. (指名委員会等設置会社: shimei iinnkai tou secchigaisha)”, introduced in 2003, and “Company with Audit & Supervisory Committee (監査等委員会設置会社: kansa tou iinkai sechigaisha)”, introduced in 2015. The degree of separation between supervision and business execution is considered to be the highest in “Company with a Nominating Committee, etc.”, second highest in “Company with Audit & Supervisory Committee” and lowest in “Company with Board of Auditors”. From this viewpoint, the highest form, “Company with a Nominating Committee, etc.”, can be more efficient in preventing fraud occurrences than the other two forms, which is examined using dmt. Also, the two recently introduced forms (the two highest forms) can be more efficient than the older form, “Company with Audit & Supervisory Board.” This difference is examined using dmt2.

In our analysis, the dmn number of directors is a proxy for firm size. This measure captures the scale and complexity of the firm, allowing us to account for the potential influence of firm size on both corporate governance indicators and fraud occurrence. By including firm size as a control variable, we aim to isolate the effects of board composition and personal networks from the confounding influence of firm size.

3.3. Empirical Model of Fraud Detection

The Cox proportional hazards model is used to examine the impact of “Corporate Governance Indicators” and “Personal Network Indicators” on the duration between fraud occurrence and detection by using the following Equation (2):

$$\sum _{i=1}^m\sum _{j=1}^3{LRH}_{ij}=\sum _{i=1}^m\sum _{j=1}^3\sum _{k=1}^n{\alpha }_{ijk}{x}_{ijk}$$

(2)

${LRH}_{ij}$ (Logarithmic Relative Hazard) of the i-th of the m fraud cases at the time of j is the explained variable.29 $x_{ijk}$ is the personal network indicators (or corporate governance indicators) at the time j in the i-th fraud case. There are n explanatory variables, of which k indicates the k-th explanatory variable. As in the logit analyses, we separately examined the effects of densities and mean degrees, as well as dmt and dmt2.

In regard to personal networks among board members, these can affect fraud detection through several channels: enhanced oversight and monitoring, as board members with strong personal ties may be more motivated to ensure compliance with ethical standards; improved information sharing, enabling better decision-making and quicker identification of potential fraud due to open communication; and reputation concerns, which are especially significant in Japanese society where maintaining social and professional standing can deter fraudulent behavior.

While our analysis provides valuable insights into the relationships between corporate governance, personal networks and fraud occurrence and detection, it is important to acknowledge the potential endogeneity issues that could affect our results. Endogeneity may arise from unobservable firm and board characteristics that simultaneously influence both the likelihood of fraud and the structure of personal networks.

One possible source of endogeneity is the omission of firm-specific and board-specific factors that could correlate with both fraud and personal networks. For instance, certain alma maters may have a culture or reputation that impacts the behavior of their graduates in a way that influences both their likelihood of engaging in fraud and their network connections. Fixed effects for alma mater schools, in particular, could help to capture the inclination towards fraud and the nature of personal networks. Unfortunately, our current dataset does not include sufficient information to implement these fixed effects. However, we recognize this as a limitation and suggest that subsequent studies collect more detailed data to address this issue.

4. Empirical Results

Table 4. Estimation results of fraud occurrence by logit analysis.

	Single Regressions		Multiple Regressions
			(a)	(a)’	(b)	(b)’	(c)	(c)’	(d)	(d)’
Variables related to Personal Network
Densityschool	0.132	0.143	0.291	0.286	0.304	0.305
	(0.55)	(0.59)	(0.78)	(0.77)	(0.81)	(0.81)
Meandegreeschool	0.054 ***	0.056 ***					0.025	0.025	0.026	0.027
	(2.83)	(2.90)					(0.88)	(0.91)	(0.92)	(0.96)
Densityhome	0.003	−0.007	−0.101	−0.083	−0.107	−0.098
	(0.01)	(−0.03)	(−0.26)	(−0.22)	(−0.27)	(−0.25)
Meandegreehome	0.063 ***	0.065 ***					0.002	0.005	0.001	0.003
	(3.51)	(3.51)					(0.07)	(0.17)	(0.04)	(0.11)
Variables related to the form of governance
dmn	0.188 ***	0.200 ***	0.253 ***	0.287 ***	0.260 ***	0.288 ***	0.242 ***	0.275 ***	0.250 ***	0.277 ***
	(11.59)	(12.03)	(5.98)	(6.58)	(6.08)	(6.46)	(5.42)	(6.00)	(5.51)	(5.91)
dmt	−1.011 ***	−0.690 *	−1.855 **		−1.444 *		−1.828 **		−1.417 *
	(−2.69)	(−1.75)	(−2.28)		(−1.75)		(−2.25)		(−1.72)
dmt2	0.16	0.51 **		0.035		0.165		0.079		0.210
	(0.74)	(2.20)		(0.06)		(0.28)		(0.14)		(0.36)
dmc	−0.149	−0.112	0.182	0.348	0.084	0.183	0.173	0.328	0.074	0.163
	(−0.21)	(−0.16)	(0.19)	(0.36)	(0.08)	(0.19)	(0.18)	(0.34)	(0.08)	(0.17)
dmte	−0.673 ***	−0.780 ***	−0.366	−0.311	−0.329	−0.284	−0.367 *	−0.313 *	−0.330	−0.286
	(−4.26)	(−4.82)	(−1.59)	(−1.34)	(−1.41)	(−1.21)	(−1.60)	(−1.35)	(−1.42)	(−1.22)
dmoutr	0.912 **	1.510 ***	−0.687	−0.679	−0.822	−0.825	−0.716 *	−0.722 *	−0.852	−0.868
	(2.45)	(3.73)	(−0.57)	(−0.56)	(−0.67)	(−0.67)	(−0.59)	(−0.59)	(−0.69)	(−0.70)
dmoutindr	0.924 **	2.067 ***	2.874 **	2.755 **	3.932 ***	3.921 ***	2.886 **	2.761 **	3.949 ***	3.926 **
	(2.40)	(4.03)	(2.26)	(2.14)	(2.91)	(2.88)	(2.26)	(2.14)	(2.92)	(2.88)
adr	−1.056 ***	−1.572 ***	1.710 **	2.859 ***	1.704 **	2.621 ***	1.686 **	2.858 ***	1.677 **	2.623 ***
	(−3.81)	(−5.11)	(2.06)	(3.11)	(2.05)	(2.76)	(2.02)	(3.10)	(2.01)	(2.76)
Constant			−8.126 ***	−9.103 ***	−8.936 ***	−9.803 ***	−8.072 ***	−9.061 ***	−8.880 ***	−9.761 ***
			(−8.83)	(−9.38)	(−9.18)	(−9.62)	(−8.85)	(−9.41)	(−9.20)	(−9.65)
Year FE(dummies)	No	Yes	No	No	Yes	Yes	No	No	Yes	Yes
Obs	-	-	8072	8072	8071	8071	8072	8072	8071	8071
Company	-	-	3026	3026	3025	3025	3026	3026	3025	3025

Note 1: Z-statistics are reported in parentheses. ***, **, * are significant at the 1%, 5%, 10% level respectively. Note 2: “Single regression” represents the estimation result of single regression in which only each variable is used as an explanatory variable.

presents the estimated results of fraud occurrences using logit analysis, while

Table 5. Estimation results of fraud detection by Cox proportional hazards model.

	Single Regressions		Multiple Regressions
			(a)		(b)		(c)		(d)
Variables related to Personal Network
Densityschool	1.433 *		2.139 **		1.980 *
	(1.67)	Obs. 223	(2.06)		(1.80)
Meandegreeschool	1.025						1.050 *		1.045
	(1.56)	Obs. 223					(1.79)		(1.58)
Densityhome	1.238		0.778		0.800
	(1.12)	Obs. 223	(−0.78)		(−0.69)
Meandegreehome	1.017						0.983		0.984
	(1.21)	Obs. 223					(−0.70)		(−0.66)
Variables related to the form of governance
dmn	0.969 **		0.988		0.965		0.982		0.958
	(−2.29)	Obs. 424	(−0.36)		(−0.85)		(−0.52)		(−0.99)
dmt	1.598		1.392				1.377
	(1.46)	Obs. 483	(0.61)				(0.59)
dmt2	0.904				0.627				0.607
	(−0.94)	Obs. 483			(−0.85)				(−0.91)
dmc	1.487		0.867		0.940		0.975		1.047
	(0.79)	Obs. 483	(−0.19)		(−0.08)		(−0.03)		(0.06)
dmte	1.146		1.264		1.231		1.287		1.250
	(1.14)	Obs. 358	(1.23)		(1.08)		(1.31)		(1.16)
dmoutr	.418		0.223		0.204		0.234		0.214
	(2.45)	Obs. 424	(−1.33)		(−1.40)		(−1.29)		(−1.36)
dmoutindr	1.619		8.889 *		11.396 **		8.302 *		10.959 **
	(1.49)	Obs. 424	(1.88)		(2.01)		(1.82)		(1.98)
adr	1.513 *		1.374		0.564		1.310		0.517
	(1.49)	Obs. 424	(0.49)	Obs. 167	(−0.50)	Obs. 167	(0.41)	Obs. 167	(−0.58)	Obs. 167

Note 1: Z-statistics are reported in parentheses. **, * are significant at the 5%, 10% level respectively. Coefficients are Hazard ratios. Note 2: “Single regression” represents the estimation result of single regression in which only each variable is used as an explanatory variable.

shows the estimated results of fraud detection by the Cox proportional hazards model. The analyses for each table are as follows:

4.1. Fraud Occurrence

The estimation results are shown in Table 4, which has ten columns. The first two columns display the results of single regressions where each variable is used individually as an explanatory variable. The remaining columns show the results of multivariate regressions. In the multivariate regressions, the odd numbered columns ((a), (b), (c) and (d)) present the results without year-fixed effects, while the even numbered columns ((a)’, (b)’, (c)’ and (d)’) show the results with year-fixed effects.

There are some corporate governance variables that have significant effects on fraud occurrence in both single regressions and multivariate regressions: dmn and dmoutindr have positive effects, while dmt has a negative effect. dmn represents the number of directors, indicating the size of each board of directors. The coefficients for dmn show a 20.7% increase in the likelihood of fraud occurrence for each additional director in single regressions and a 28.8% to 33.4% increase in multivariate regressions. This result means that the probability of fraud increases as the number of directors increases. As shown in Table 2, board size varies among companies, with a mean of about eight, and a max. of 30. Under “Corporate Governance Reform”, the Japanese government has urged companies to reduce board sizes to improve their efficiency. This result suggests that the government’s reform direction may be effective in preventing fraud. Additionally, this variable can be related to personal networks. In Figure 1, the 14 nodes represent each member of the board of directors. The number of nodes is often called the size of network, and is a basic concept of complex network analysis. From this viewpoint, we may say that a smaller board size, or board personal network, may be more effective in preventing fraud.

The results for dmt also suggest that the Japanese corporate governance reforms have had some positive effect on preventing the occurrence of corporate fraud. In terms of the degree of separation between supervision and business execution in a company, the form “Company with a Nominating Committee, etc.”, is regarded as the best form among three options. This variable, dmt, examines whether “Company with a Nominating Committee, etc.”, is more efficient in preventing fraud occurrences than the other two forms. The results show that the coefficients are significantly negative, indicating that the probability of fraud is significantly lower for companies with a nominating committee, with a reduction ranging from 49.9% to 85.8%. However, the results for dmt2 indicate that the form “Company with Audit & Supervisory Committee” cannot be said to be effective in preventing fraud. The coefficient in the single analysis is significantly positive, and the other coefficients are also positive, suggesting that this form may actually lead to a higher probability of fraud. The “Company with Audit & Supervisory Board” is considered a halfway form. The Japanese government initially tried to promote investment clubs to spread the adoption of “Company with a Nominating Committee, etc.”, similar to those in the United States at the beginning of this century. However, the criteria required for companies were too stringent, leading to limited adoption. As a compromise, the “Company with Audit & Supervisory Committee” was introduced. These second-form companies might be those unable to meet the criteria for the first, best form, “Company with a Nominating Committee, etc.” Consequently, these companies might have higher probability of fraud compared to those following the other forms.30

Unlike the results for dmn and dmt, the results for dmoutindr suggest that the corporate governance reforms implemented by the Japanese government may not be effective in preventing the occurrence of fraud. dmoutindr represents the ratio of the number of independent directors to the total number of directors.31 The coefficients for dmoutindr are significantly positive and larger than those of the other variables, indicating that a higher ratio of independent directors is associated with a substantial 151.9% increase in the likelihood of fraud occurrence in single regressions, and dramatically more in multivariate regressions. Under “Corporate Governance Reform”, Japanese listed companies are required to increase the ratio of independent directors because the government and investors believe that boards of directors better fulfill their functions by appointing independent directors. However, these results seem to suggest the opposite outcome: the higher the ratio of independent directors, the more likely companies are to experience fraud.

How did this unexpected result come about? It is likely to be related to the nature and quality of the fraud data. Our dataset, like those in previous research, exhibits partial observability; only detected cases are recorded in fraud statistics. As we discuss in the next subsection, companies with higher rates of independent directors are more likely to detect fraud; in other words, they do not tend to conceal fraud. Therefore, the results for dmoutindr in our logit analyses could suggest that the positive relationship between fraud occurrence and this variable is a result of those companies being better at detecting fraud.

Regarding personal network variables, the mean degree of school (${MeanDegree}_{school}$) and mean degree of home (${MeanDegree}_{home}$) networks, while significant in single regressions, are not significant in multivariate regressions, suggesting their limited independent influence on fraud occurrence. Similarly, the density of hometown and school networks (${Density}_{home}\mathrm{a}\mathrm{n}\mathrm{d}{Density}_{school}$) do not show significant impacts on fraud likelihood. These findings highlight the critical role of board composition and governance structures in mitigating or exacerbating corporate fraud, while personal networks among board members appear to have a less clear impact.

4.2. Fraud Detection

Next, we discuss our estimates of fraud detections by the Cox hazard model, investigating the association between the concealment period (duration) of frauds and predictor (explanatory) variables. The estimation results are shown in Table 5, which has five columns. The first column presents the results of single regressions where each variable is used individually as an explanatory variable, while the other columns show the results of multivariate regressions.

The duration between fraud occurrence and its detection varies among corporate frauds. As shown in Table 2, duration ranges from 0 to 624 months; while some frauds are detected in the same month they occur (duration = 0), in the most egregious cases, the fraud was concealed for more than half a century (duration = 624). This analysis uses these fraud cases to examine the impact of explanatory variables on the duration between fraud occurrence and detection. Unlike the previous logit analysis, which had a large sample size, this analysis had a limited sample size.32

${Density}_{school}$ is the only variable that has significant effects in both single regressions and multivariate regressions. These results suggest that stronger alumni networks among board members are more likely to prevent fraud concealment. For example, a hazard ratio of 1.433 indicates a 43.3% increase in the hazard of fraud detection for each unit increase in ${Density}_{school}$. A board of directors with strong personal networks has the potential to mitigate information asymmetry and prevent concealment compared to a board without such networks.

This result contrasts with previous studies in Europe and the United States, which have empirically demonstrated that strong personal networks reduce the expected costs of corporate fraud and increase its probability. This implies that the impact of personal networks on corporate fraud may differ among countries, societies and cultures.

It is the estimation results of dmoutindr that we should focus on. Although these do not have a significant result in single regression, they have the largest hazard ratios and significant effects in their multivariate regressions. A board of directors with higher rates of independent directors has greater potential to detect fraud compared to a board with a low rate; in other words, independent directors might reduce a board’s incentive to conceal fraud. For instance, a hazard ratio of 1.619 indicates a 61.9% increase in the hazard of fraud detection for each unit increase in the ratio of independent directors, with extremely high ratios in multivariate regressions ranging from 8.889 (1669.8%) to 11.396 (over 1000%). These results suggest that the recent changes in corporate governance reform in Japan have been meaningful in terms of preventing the concealment of fraud.

4.3. Implications and Contributions

Our findings indicate that recent corporate governance reforms in Japan, particularly the introduction of Nominating Committees and the increase in the ratio of independent directors, are effective in reducing the occurrence and enhancing the detection of corporate fraud. These results underline the importance of separating supervision from business execution and appointing independent directors to strengthen oversight mechanisms.

The implications of this study extend to practical recommendations for both regulatory bodies and corporate management. Policymakers should consider mandating higher ratios of independent directors and promoting governance structures that ensure robust oversight. Companies, on the other hand, should be encouraged to increase the number of independent directors and ensure their empowerment to act effectively, thereby improving transparency and supervision. This alignment between regulation and practice could significantly enhance corporate governance standards and reduce fraud.

Additionally, this research advances the understanding of corporate governance by incorporating the unique aspect of personal networks among board members, particularly in a non-Western context. Our findings suggest that strong personal networks can aid in fraud detection, contrasting with some Western studies and highlighting cultural nuances in governance practices. This study enriches the literature on corporate governance and fraud, providing a foundation for future research to explore these dynamics in different cultural and regulatory environments.

5. Conclusions

In recent years, Japan has recognized the need for “Corporate Governance Reform.” One major reform has been advocating for higher ratios of independent directors on boards of directors. Institutional investors have also pushed for this change. Additionally, there is a movement from the traditional Board of Corporate Auditors to a structure including Nominating Committees, reflecting a shift towards stronger external oversight.

Our findings suggest that these governance changes are generally effective in preventing the occurrence and enhancing the detection of fraud. This is particularly evident through the transition to Boards with Nominating Committees, the reduction in the number of directors, and the appointment of independent directors. Specifically, increasing the proportion of independent directors has notably expedited fraud detection and helped prevent its concealment.

Moreover, our study reveals that strong personal networks, such as those formed through alma maters, play a significant role in preventing corporate fraud concealment. It is possible that boards with robust personal networks are better at mitigating information asymmetry, thus making it easier for them to detect fraud more effectively than those without such networks. This effectiveness may be partly attributed to the role of reputation, which is particularly significant in Japanese society, where social and professional standing can greatly influence business conduct. This finding contrasts with prior research from the US and Europe, suggesting that the influence of personal networks can vary significantly across different cultural contexts.

Looking forward, while our research has provided insights into the dual aspects of fraud occurrence and detection, it has not fully tackled the issue of partial observability inherent in fraud data. Future studies should explore the biases resulting from undetected fraud cases to further refine our understanding of corporate fraud dynamics. Additionally, while our research has shown the potential of personal networks in corporate finance and corporate governance research, this topic could be explored more deeply from multiple perspectives. Research on personal networks has recently been accumulating globally, and a variety of personal networks data is becoming available. Based on new findings and data, future work is required to explore the relationship between corporate behavior and the personal networks of individuals who make up companies.