THE RELATIONSHIP BETWEEN SOCIAL
NETWORKS AND FIRM VALUE
A quantitative study on the Frankfurter Wertpapierbörse
** **
Master Thesis
MSc Business Economics, Finance
** ** ** ** By Danique W.M. Captein 5953588 ** **
** Supervisor prof. J.K. Martin **
** University of Amsterdam **
Statement of originality
This document is written by Student Danique Captein who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Abstract
This study provides new insights into the long-term and short-term relationship between the social networks of board members and firm value.
Related studies on this subject appear to show ambiguous results, varying from a positive to a negative relationship. This might be caused by the potential for endogeneity bias due to self-selection: being part of a social network could be related to board member characteristics or firm characteristics that also influence firm value.
Therefore, in this paper a unique instrument is applied to circumvent the potential for endogeneity due to self-selection. The instrument represents a group of individuals that is born with a strong social network. These are noble-titled board members, which are overrepresented on boards in Germany.
The results of this study show that no relationship exists between the social networks of board members and firm value. This holds for the long-term as well as for the short-term and for all noble-titled board members together as well as for subgroups including only noble-titled supervisors and noble-titled managers.
Acknowledgements
First and foremost, I would like to thank my supervisor prof. J.K. Martin for his help, continued guidance and for providing me his dataset.
Index
1.0 Introduction ………....01
2.0 Related literature ………....04
2.1 Social network theory………..04
2.2 Social networks and managerial decision-making………..05
2.3 Empirical evidence on the direct relationship between social networks and firm value………. 07
3.0 Hypotheses……….. 10
4.0 An introduction to the methodology………. 12
4.1 The setup of my study……… 12
4.2 German nobility ………. 13
4.3 German board structure………. 14
4.4 Indices on the Deutsche Börse……… 15
5.0 Methodology………... 17
5.1 The long-term relationship between board members’ social networks and (the growth in) firm value……… 17
5.2 Event study………. 21
6.0 Data………. 26
6.1 Data and descriptive statistics #1………... 26
6.2 Data and descriptive statistics #2………... 33
7.0 Results……… 37
7.1 Introduction to the long-term relationship between board members’ social networks and firm value……….37
7.2 The announcement effect of appointing board members with a 7.3 strong social network………..43
8.0 Conclusion and discussion………. 50
8.1 Conclusion……….. 50
8.2 Discussion………... 51
APPENDICES……….. 53
1.0 Introduction
“The core idea of social capital is that social networks have value”, Putnam (2001) stated. Social capital is about ‘who people know’ and the norms of reciprocity that arise from those connections. In order to investigate whether there is a relationship between the social networks of executives and managerial decision-making Shue (2013) investigates randomly assigned section peer groups of former Harvard Business School students. She finds that individual executives react to the mean characteristics of their section peer group after reunions take place. This result shows that the social networks of executives influence their managerial decision-making.
Whether the influence on managerial decision-making is advantageous to the company the executive works for is unclear. There are two leading theories: one theory emphasizes the information advantages caused by social networks (Cohen et al. 2008, Ellison and Fudenberg 1995, Burt et al. 2013) and the other points out that social networks are detrimental to good governance (Kramarz and Thesmar 2013, Fracassi and Tate 2012). Following these theories, the relationship between the social networks of board members and firm value could be positive as well as negative. This study investigates that specific relationship by the following research question:
What is the long-term and short-term relationship between the social networks of board members and firm value?
Studying whether a relationship between the social networks of board members and firm value exists improves our understanding of the determinants that might be of interest while searching for a new board member. Insights into this topic would therefore be valuable for top-level recruiters, the leadership of listed companies as well as it will provide interesting information for investors and analysts.
The social networks of board members are defined by their set of social connections with different backgrounds, knowledge and expertise. In line with the unclear relationship between executives’ social networks and managerial decision-making, existing literature shows ambiguous results to this research question. However most studies suggest that a negative relationship exists.
Therefore the hypotheses are stated below:
- H1: there is a negative long-term relationship between the social networks of supervisors and (the growth in) firm value.
- H2: there is a negative long-term relationship between the social networks of managers and (the growth in) firm value.
- H3: the announcement effect of the appointment of supervisory board members with a strong social network is negative.
- H4: the announcement effect of the appointment of management board members with a strong social network is negative.
Hypothesis 1 and hypothesis 2 are tested by a cross-sectional OLS regression. The variable of interest is a dummy variable that equals ‘1’ when there is a noble-titled board member on board of a particular company and ‘0’ otherwise. In order to prevent omitted variable bias, different variables are included in the regression to control for firm specific and board specific differences.
Hypothesis 3 and hypothesis 4 are tested by an event study. The event of interest is the announcement that a noble-titled board member gets appointed to the board of a particular company. The Cumulative Average Abnormal Returns (CAARs) over different event windows are taken to investigate whether the announcement causes a market reaction. Both tests are run for the whole group of noble-titled board members, as well as for noble- titled supervisors and managers apart from each other.
When measuring a board members’ social network, there is a potential for endogeneity bias due to self-selection: being part of a social network could be related to board member characteristics or firm characteristics that also influence firm value (Braggion 2008). Thereby results can be biased. Prior studies faced this challenge as well. In this study, a unique method is applied to identify board members with a strong social network. These are noble-titled board members. Because nobles are born with a strong social network a way is found to circumvent the potential for endogeneity bias due to self-selection. The availability of board composition data and an overrepresentation of noble-titled board members make the Frankfurter Wertpapierbörse the designated research area to study this question. Moreover, Germany has a two-tier board system, which means that a clear separation exists between management board members and supervisory board members. At lasts, in general and in particular with respect to education, nobles and non-nobles are treated
the same in Germany.
This thesis proceeds as follows: chapter 2.0 provides some basic network theories and it discusses the related literature. This theoretical framework provides input to the hypotheses formulated above and in chapter 3.0. Chapter 4.0 provides an introduction to the methodology, including extra information about nobility and the board structure in Germany. A detailed description of the methodology is given in chapter 5.0. Chapter 6.0 provides information on the data used to test the hypotheses and in chapter 7.0 the results are discussed. Chapter 8.0 concludes and discusses.
2.1 Related literature
An extensive literature exists on the relation between social networks and firm value. This chapter addresses the social networks of board members. Insights regarding the relation between the social networks of board members and firm value serve as input to the formulation of my hypotheses.
Before the results of those studies can be well interpreted, section 2.1 discusses what social networks are. In an organizational context, section 2.2 addresses the impact of board members’ social networks on managerial decision-making. The results of studies on the direct relation between board members’ social networks and firm value are provided in section 2.3.
2.2 Social network theory
In order to fully understand what a social network means the social network theory provides clarifying insights. This section provides information on some basic network concepts and answers the following questions: what is a social network? How can we breakdown social networks? And in what kind of settings do social networks arise?
As a starting point, a clear distinction should be made between a ‘network’ and a ‘social network’. A network is a set of relations between objects. Those objects can be people, institutions and nations, but they can be brain cells or electrical transformers as well (Kadushin, 2012). With respect to a network there are no social relationships required among the objects. In contrary, social networks do require the involvement of relationships among people (Perry-Smith, 2011).
Social networks can be analyzed by using a sociogram, which is a graph that depicts relationships between people (Kadushin, 2012). Points in a sociogram represent individuals, groups or institutions and lines show their corresponding relationships. In social network analysis points are better known as nodes and lines are better known as ties (Scott, 2012). In order to speak of a social network, there should be at least two nodes and a relationship between those nodes. The sociogram in Appendix Figure 1 depicts a web of relationships between nodes. In section 2.2 more information will be provided on the interpretation of this sociogram.
Where the concept ‘network’ can be specified in a ‘social network’, a social network can be divided into three different kinds of networks. These are egocentric, socio-centric and open-system networks. Egocentric networks are networks around a single node or individual. This kind of network plays a central role in this study, since I investigate the impact of a
single board member’s social network on firm value. In socio-centric networks the starting point is not an individual, but a group. It always encompasses more than the connections of a single node. Socio-centric networks can be seen as networks in a “box”, where a box could be a classroom, an organization or a society. Sometimes it is not possible to clearly mark out networks. While studying connections between corporations clear boundaries are missing. In that case we speak of open-system networks (Kadushin, 2012).
Social networks can be formed in formal as well as in informal systems. The next chapter addresses the social networks of board members from publicly traded companies. While speaking of social networks in an organizational context we tend to think of a formal system. However, formal and informal systems are not mutually exclusive. Within the formal system of an organization, small groups – like project groups, departments, the management board of directors or the supervisory board members – work closely together. Cooperation requires intensive interaction between the individuals participating in the group and thereby individuals develop feelings for each other. Therefore, the formal system of an organization can give rise to small, informal systems (Kadushin, 2012).
2.3 Social networks and managerial decision-making
Based on the information provided in the previous section, we have a basic understanding of what social networks are and where they emerge. This section addresses how a board members’ social network is related to managerial decision-making, which in turn could influence firm value. There are two main views regarding the relation between board members’ social networks and managerial decision-making. Those views bring to light that the effects of social networks can either be positive or negative.
2.2.1 A positive effect of social networks: information advantages
In line with the positive view, extensive literature exists on social capital. Putnam (2001) stated, “the core idea of the social capital theory is that social networks have value”. Social capital is about who people know and the norms of reciprocity that follow from those connections. In an organizational context, many studies emphasize an information advantage caused by social capital (Cohen et al. 2008, Ellison and Fudenberg 1995, Burt et al. 2013). Figure 1 can illustrate how an information advantage could emerge. The sociogram depicts several social clusters and each cluster provides a flow of information. The participants – represented by nodes – are most concerned about the information flowing in their cluster. They can be aware of other clusters, but they don’t absorb and make use of the information
flows in those clusters. Sometimes there are no ties between the nodes in different clusters; we then speak of a structural hole. However, persons like Robert in Figure 1 who participate in more than one cluster can take these structural holes away. Those persons can fulfill a brokerage function by incorporating more information flows at the same time and being able to control them (Burt, 2001). This could lead to an information advantage.
Haunschild and Beckman (1998) find a positive effect as well of social networks on the collection of information. Additionally, they show that the information advantage is less for large and concentrated companies. Large companies have better access to information compared to small and medium size companies, because at large companies there are more individuals responsible for the collection of information and they are more inclined to employ people who have the task to connect with others outside the organization. This diminishes the chance on structural holes between different clusters.
Shue (2013) also shows that there is a significant relation between executive social interactions and managerial decision-making. To find out, she investigated randomly assigned section peer groups. These are alumni groups that are composed of individuals during their study at the Harvard Business School (HBS). After graduation, the peer groups still meet up once in a while. Shue develops a model to test whether individuals in section peer groups react to the mean characteristics of those groups after reunions take place. She shows that this is the case and that this adaptive behavior is much stronger within section peers – that still meet up once in a while – compared to class peers. However Shue states that more information could be the results of social connections, she finds as well that the connections may lead to weakly less efficient acquisitions in the mergers and acquisitions market.
Moreover, Saloner (1985) emphasizes the role social networks can play in finding new directors. Supervisors tend to use their social network and advise managers when there are job openings for board positions. Saloner shows that this can improve the director- management match since the supervisors have more information about the people they introduce.
2.2.2 A negative effect of social networks: weaker corporate governance
On the other side of the coin, a possible negative outcome of social networks in an organizational context can be the act of weaker corporate governance. Barnea and Guedj (2007) find evidence that better connected board members and board members whose connections are better connected exhibit weaker firm governance. In their study being “better
connected” is described by “having connections to other board members”. Weaker firm governance may be the result of the fact that decision-makers within a company form an innercircle of the most qualified persons in the company. They feel highly committed to each other and as a consequence they may become less critical towards one another.
Fracassi and Tate (2012) find similar results. However, they show that the reason for weak governance in companies where CEOs and outside directors have prior connections could be diverse. Their explanations vary from fear regarding future career opportunities, to groupthink due to common backgrounds. These explanations result from the social network measure applied by Fracassi and Tate. They only included prior education and prior employment in their network measure.
Kramarz and Thesmar (2013) study the relationship between social networks and firm governance as well. They find that social networks are detrimental for good firm governance. Kramarz and Thesmar conclude this based on a study on the networks of CEOs and supervisory board members. They state that social networks appear to be an important determinant of board composition. Moreover, after the board is composed CEOs and supervisors that have prior ties – in this study when they both had a career in civil service – CEO turnover appears to be lower. This is even the case when firm performance is poor. This result suggests a negative relationship between social networks and firm performance. The consequences of a board members’ social networks on managerial decision- making appear to be ambiguous. The possible information advantage would mainly suggest a potential increase in firm value, but weaker firm governance would mainly suggest a potential decrease in firm value. However, these are the main views regarding the relationship between board members’ social networks and managerial decision-making. There are also exceptions. For instance, Benmelech and Frydman (2014), show that those CEOs who served in the military service are less likely to be involved in fraud. The military service network, according to the results of this study, appears to be related to better firm governance.
2.3 Empirical evidence on the direct relationship between social networks and firm value
This section provides detailed information on the direct relation between board members’ social networks and firm value. In line with the findings presented in section 2.2, related literature on the direct relation shows ambiguous results as well. It is important to note that in all studies described in this section, the researchers investigated a long-term relationship
(at least in years) between the above-described variables. In only one study, done by Ishii and Xuan (2011), the short-term relationship is investigated.
2.3.1 Board members’ social networks are related to an increase in firm value
Horton et al. (2012) define the social networks of executives and outside directors by their connectedness. Connectedness is measured by the closeness and the brokerage position of an executive or a director. Closeness provides information about someone’s position within a network; how efficiently and effectively the executives and outside directors can reach others. A brokerage position tells something about an individuals’ ability to act as a broker between two actors. Referring back to section 2.2 and Appendix Figure 1, a brokerage position gives an indication about someone’s ability to take away structural holes. Horton et al. thereby emphasize the importance of looking at indirect connections as well as part of the entire director network. However, they only collect data on observable, formal ties. They show that better - connected companies are positively associated with overall future performance. They measure firm performance by the companies’ market-to-book ratio, return on assets, one-year total stock price return and the one-year sales growth. Moreover, Horton et al. (2012) find that compensation depends on the characteristics of social connections as well. Better connections appear to result in higher compensation for board members.
Moreover, Cohen et al (2008) measure networks across companies. They study the connectedness between managers of available funds and portfolio managers. They measure their connectedness purely based on educational background and distinct between different levels of connectedness. The first level requires that they have attended the same school, the second level requires that they have attended the same school and received the same degree, the third level requires that they have attended the same school at the same time and at lasts, the forth level requires that they have attended the same school at the same time and passed on the same degree. Cohen et all show with an OLS pooled regression that portfolio managers of mutual funds place larger concentrated orders on stocks they are connected to through their network. These portfolio managers appear to perform significantly better on those related holdings.
2.3.2 Board members’ social networks are related to a decrease in firm value
Fracassi and Tate (2012) test the networks of CEOs and investigate whether prior connections between a CEO and its independent directors have an impact on firm value. They construct a social network index based on the CEOs and the independent directors’ current employment, prior employment, education and other activities. In order to circumvent the endogeneity problem of board composition, Fracassi and Tate didn’t investigate what happens when a connected, independent director steps on board, but they tested what happens when a connected director departs from a board due to death or retirement. They find that firm value, measured by Tobin’s Q, increases after an independent director departs from the board while having pre-existing social ties to the CEO.
Ishii and Xuan (2011) use a broader definition of social networks and investigate social connections in the merger market. They focus on social networks between the directors and the senior executives of an acquirer and a target. Since they study the short-term effect of a merger announcement, Ishii and Xuan are interested in the perception of the market regarding this announcement. In order to test this, they apply a Cumulative Abnormal Return (CAR) model. They find that a merger announcement in which prior social relationships exist between the acquirer and the target has a significantly negative impact on the cumulative abnormal returns of the acquirer in a three day window [-1,+1]. The impact of such an announcement appears to be related to positive, but insignificant CARs for the target. Moreover, the impact on the combined company is tested as well. There appears to be a negative impact on firm value when strong social connections exist between directors of the acquirer and the target. This suggests the existence of a negative relationship between the value of the combined company and social ties between directors of the separate companies. Braggion (2011) studies the egocentric networks of CEOs that are part of a broader network: the Freemasonry. Because the Freemasonry is a secret club, people are inclined to feel a stronger connection to their club members, which results in more trust and reliance between its members. Braggion finds that firm performance of large public companies measured by Tobin’s Q worsens when a Freemason manages a firm. This result is statistically significant.
To conclude, the relation between board members’ social networks and firm value appears to be ambiguous. In the next section, I formulate my hypotheses. These are all based on the literature findings described in this chapter.
3.0 Hypotheses
In section 2.2 the most important theories that could indicate a positive or negative relationship between board members’ social networks and firm value are discussed. My study, however, questions whether there ís a relationship between those two variables on the long-term and on the short-term. In this section I therefore summarize the most important literature findings discussed in section 2.3 and formulate my hypotheses. Prior studies on the direct relation between board members’ social networks and firm value show ambiguous results. Appendix Table 1 provides an overview of the findings of those studies. Regarding a strong social network, I focus on individual board members that have a strong social network of connections with different backgrounds, knowledge and expertise. I therefore study egocentric networks; discussed in section 2.1.
This broad measure of social networks is best comparable with the measure applied by Braggion (2011), because the Freemasonry connects people that in fact have different backgrounds, knowledge and expertise. The Freemasonry may be a broad network itself, but Braggion investigated the egocentric networks of CEO’s that belong to this network. So, in this study on the Freemasonry the assumption is that it is their Freemasonry membership that ensures a strong egocentric network. Braggion finds a statistically significant negative relationship between board members’ social networks and firm value of publicly listed companies. Overall, most of the studies show this negative relationship on the long-term. Fracassi and Tate (2012) study egocentric networks as well and show that prior connections between CEOs and directors result in lower firm value.
Moreover, Fracassi and Tate, Saloner (1985) and Kramarz and Thesmar (2013) distinct between management board members (managers) and supervisory board members (supervisors) in their study. Referring back to Chapter 2, they all studied the relationship between directors and supervisors and their impact on firm performance. Where Saloner finds that a priori relations between managers and supervisors lead to a better director-management match, Kramarz & Thesmar and Fracassi & Tate find that those relations could be detrimental to firm governance and that it would decrease firm performance. However, those studies mainly concern about the networks concentrated around CEOs. Since managers and supervisors have completely different tasks, I study the egocentric networks of managers and supervisors apart from each other.
Therefore I formulate the following hypotheses:
- Hypothesis 1: there is a negative long-term relationship between the social networks of supervisors and (the growth in) firm value.
- Hypothesis 2: there is a negative long-term relationship between the social networks of managers and (the growth in) firm value.
As mentioned before, all studies – except the study done by Ishii and Xuan (2011) – investigate a long-term relationship. Following the semi-strong form of the Efficient Market Hypothesis, all public available information will immediately be absorbed by market participants and reflected in a companies’ stock price (Malkiel, 2003). Therefore I expect that market participants anticipate the long-term expectations. Therefore, the hypotheses on the short-term relationship between board members’ social networks and firm value are stated as follows:
- Hypothesis 3: the announcement effect of the appointment of supervisory board members with a strong social network is negative.
- Hypothesis 4: the announcement effect of the appointment of management board members with a strong social network is negative.
4.1 An introduction to the methodology
Some important lessons can be drawn from the literature study described in chapter 2. This chapter brings to light that – in order to fully understand the setup of my study and the interpretation of the results – more information is required. Before the methodology is discussed in Chapter 5, this chapter provides that information. Section 4.1 discusses the setup of my study, based on the lessons learned from the related literature study. Section 4.2. and section 4.3. are provided in order to fully understand the setup of my study.
4.2 The setup of my study
Due to the potential for endogeneity bias it can be challenging to define an appropriate social network measure. Social connections and social clusters as illustrated in Appendix Figure 1 do not arise on itself. When measuring the strength of a board members’ social network, a self-selection problem arises: being part of a social network could be related to some board member characteristics or firm characteristics that influence firm performance as well (Braggion 2008). Findings on the relationship between board members’ social networks and firm performance can therefore easily be biased by the characteristics of the candidates that want to belong to those networks (Borgatti & Halgin, 2011).
Inspired by Shue (2013) and Braggion (2011) who used the conditionally random assignment of peer groups and respectively the Freemasonry to identify social networks and tackle the potential self-selection problem, the setup of this study is described in the following paragraph.
As discussed in the previous chapter, this study is about individual board members with a strong social network. This network is defined as a board members’ social network of connections with different backgrounds, knowledge and expertise. In order to identify those board members and prevent the potential endogeneity bias of self-selection, I use an instrument that represents individuals that are born with a strong social network. These are noble-titled individuals. Because nobles are born with a strong social network, a way is found to circumvent the potential endogeneity bias of self-selection. Because noble-titled individuals are born with a strong social network, the assumption is made that they do not only have a strong social network but they have a stronger social network compared to other board members as well. Moreover, Horton et al. (2012) emphasize the importance of looking at indirect connections besides direct connections. This instrument encompasses both direct and indirect connections, as well as formal and informal connections.
(FWB). The reason to focus on the German market is threefold. First, German elites have the same educational opportunities as other Germans (Hartmann, 2010). This means that there are no special schools for the education of nobles. I therefore expect no difference in the quality of education between those who are titled and those who are not. This diminishes a potential omitted variable biases that could appear when there are – besides their social network – differences between nobles and non-nobles that could influence firm value. Second, in Germany publicly listed companies have a two-tier board system. For listed companies, the separation of management and control is even regulated by law. In this system supervisors don’t take a seat in the management board among their colleagues (Oehmichen et al, 2009). Since a distinction between managers and supervisors is made in this study, a clear distinction in their functions is desirable. At lasts, there is an overrepresentation of nobles on German boards.
Moreover, Haunschild and Beckman (1998) found that information advantages – that suggest a positive relation between board members’ social networks and firm value – appear to be less for large and concentrated companies. The index a companies’ stocks are part of provides some information on the characteristics of a company. Therefore a distinction is made between different indices. The FWB has five main indices, which are the DAX, MDAX, SDAX, TecDAX and the CDAX.
The decisions regarding the setup of my study require some extra information. Therefore the remainder of this chapter improves our understanding of nobility in Germany, provides an extensive explanation of the board structure of German publicly listed companies and informs about the companies listed on each index on the FWB.
4.3 German nobility
From an international perspective, nobles belong to the class of people who historically had a special social and political status. Their corresponding privileges varied from country to country. This is mainly due to differences in local political circumstances; when states became large and strong, governments were not eager to partly ‘transfer’ their own power to those entitled (Dewald, 1996). In Germany, nobles had different privileges as well. They officially lost those privileges after the Weimar Republic was born. Since 1919, nobles were no longer seen as a legal entity. This was even regulated by law in Article 109 of the Weimar Constitution. However, nobles could keep their title, but it became part of their family name. A signal of nobility is the predicate “von” as part of the family name that almost all German nobles carry. In order to acquire a personal German nobility title, one should be born into the
noble class or marry a noble. Nowadays, it is even possible to license a title by a legal title- holder.
While the legal power of German nobility was abolished in 1919, it remains an important social force. This can be noticed by active phenomena like the “Vereinigung der Deutschen Adelsverbande”, the monthly publications of the “Deutsches Adelsblatt” and there are active groups of noble-titled individuals coming together on a regular basis. The existence of title-holders that offer nobility titles emphasize the willingness of people to acquire such a title.
Therefore noble-titled individuals still belong to a social class. Referring back to Appendix Figure 1, nobility provides its members a place in a social cluster by birth. A social cluster composed of nobles. Their background connects them. The social advantages go beyond this specific cluster, because every member within this cluster can fulfill a brokerage function for others as well. Moreover, as shown by Hartmann (2010) individuals from the upper class in Germany are more likely to get a top management position compared to individuals from the working and middle class. This holds when only talented PhD holders are included. The outcome emphasizes the importance of someone’s background.
4.4 German board structure
Publicly listed companies are characterized by the separation of ownership and control. When managers benefit themselves at the expense of its shareholders, this separation results in principal-agent problems (Edgerton, 2012). Initially, Germany had a one-tier board system where there was a difference between executive and non-executive directors. However, these two functions where executed within the same board (Baums, 2003). In fact, this could diminish good monitoring and the separation of ownership and control.
In order to protect shareholders against agency problems, Germany switched from a one-tier board structure to a two-tier board structure in 1870. Since then, all stock corporations and limited liability companies with more than 500 employees were by law obliged to have a two-tier board structure. The Aufsichtrat – better known as the supervisory board members – and the Vorstand – better known as the management board members – compose this board structure. In this system supervisors don’t take a seat in the management board among their colleagues (Oehmichen et al, 2009). The supervisory board functions as a representative body for the companies’ shareholders.
The responsibilities of supervisors can be summarized as follows:
- Approval or dismissal of management board members - Monitoring of the management board members
- Validation of important reports, for instance the annual report - Approval of the annual report
Nowadays, employees are seen as the principals of the corporate employer as well. The supervisory board might therefore also be composed of employees (Baums, 2003). As stated in the Co-Determination Act of 1976 it depends on the size of a company whether there must be employees represented on the supervisory board. Companies with less than 500 employees are not required to have employees on their supervisory board. For companies with more than 500 and less than 2000 employees at least one third of the supervisory board members should be employee representatives. At last, half of the supervisory board members are required to be employee representatives when a company counts more than 2000 employees.
Supervisors can be appointed on an individual basis. Proposed candidates for the supervisory board are announced to the shareholders during a General Meeting. Shareholders decide whether to appoint supervisors as their representatives.
With the two-tier board system a better protection of shareholders is intended. There are, however, doubts about its effectuality. In practice, also the interests of supervisory board members and shareholders can diverge. Moreover, the process of gathering information as a supervisor can be more difficult in a two-tier system (Nietsch, 2005).
Due to the clear distinction in the two-tier board system between managers and supervisors in Germany, the relationship between board members’ social networks and firm value is investigated for both groups apart from each other in this study.
4.3 Indices on the Deutsche Börse
As shown by Haunschild and Beckman (1998) information advantages of social networks appear to be less for large and concentrated companies. In order to know the differences between companies that have their stocks traded on diverse indices, this section provides information on the indices of the Frankfurter Wertpapierbörse.
DAX
The DAX consists of the 30 largest and most actively traded equities in Germany. These are better known as ‘blue chips’. A stock can be part of the DAX when it belongs to the top 30 in terms of trading volume and market capitalization based on their quantity of free float and their sector. Every year in September a review of the composition of the DAX takes place. Despite this weighing date, companies can be removed from the DAX when they no longer belong to the top 45. On the other side of the coin, companies can be added to the DAX when they belong to the top 25.
MDAX
The MDAX consists of the 50 largest companies from the classic sectors. Their weighing is based on the same standards as described under ‘DAX’, namely trading volume and market capitalization. All companies listed on the MDAX rank below those listed in the DAX. A review of the composition of the MDAX takes place twice a year in March and September. There will however be continuous checks and when a company no longer belongs to the top 75, it will be removed from the MDAX. Also, when a company belongs to the top 40 and follows the above-described criteria it will be added to the MDAX.
SDAX
The SDAX consists of the 50 companies from the classic sectors directly following those included in the MDAX. A review of the composition of this index takes place four times a year in March, June, September and December. The criteria for companies in this index equal those described before: trading volume and market capitalization.
TecDAX
The TecDAX consists of the 30 largest companies from technology sectors. Here as well, weighing is based on trading volume and market capitalization. This index includes those companies that are in the technology sector and rank below the DAX. Twice a year, in March and September, a review takes place. In order to keep this index updated, continuous checks take place where the same criteria are maintained as described under DAX.
CDAX
The CDAX consists of all other shares of domestic companies that follow the standards. It thus represents the whole German Equity Market. As stated by Holler (2012) many illiquid stocks are traded on the CDAX.
5.1 Methodology
This chapter addresses the methodologies applied in order to test the hypotheses. First, section 5.1 covers the regression used to find the long-term relationship between board members’ social networks and the (growth in) firm value. Concerning the announcement effect, section 5.2 describes an event study. The event study shows whether there is a positive or negative abnormal return due to the appointment of a noble-titled board member. In effect, this test reflects the perception of market participants regarding the future profitability of the company due to the change in the companies’ board structure.
5.2 The long-term relationship between board members’ social networks and (the growth in firm value
As stated in chapter 3 the long-term relationship between board members’ social networks and firm value will be found by testing the following hypotheses:
- Hypothesis 1: there is a negative long-term relationship between the social networks of supervisors and (the growth in) firm value.
- Hypothesis 2: there is a negative long-term relationship between the social networks of managers and (the growth in) firm value.
A cross-sectional regression is applied to test whether (the growth in) firm value on the long- term is lower when there is at least one noble-titled member on the board of a publicly listed company in Germany compared to boards without noble-titled members. The dependent variable of the regression is (the change in) Tobin’s Q, from now on denoted as Qi. Qi is measured on a yearly basis by the following formula:
[BV of TA +(CSO*Price) –BV of CS] + MV of PS [BV of TA]
With BV (Book Value), TA (Total Assets), CSO (Common Shares Outstanding), CS (Common Shares), MV (Market Value) and PS (Preferred Stock).
Qi measures the market value of a company to its replacement cost (Lang et al, 1989). It is an appropriate measure to determine the long-term relationship between board members’ social networks and firm value, because it is a forward-looking measure under the existing management. When Qi is >1, the market value of a company exceeds its replacement cost
and thereby an increase in firm value is anticipated. When Qi is <1, the market value of a
(Lindenberg and Ross, 1981). Moreover, Qi allows for comparison across companies. The variables in this model are derived from the model applied by Campbell and Vera (2010). They studied the long-term relationship between firm value (measured by Tobin’s Q) and women on boards. However, in contrary to them, I apply a cross-sectional regression and add some control variables. The base year of the cross-sectional regression is 2004. During the course of this study, a noble-titled board member represents an individual with a strong social network. The underlying assumption is that the only difference between noble-titled board members and non-noble-titled board members is that noble-titled board members have a stronger social network. As described in chapter 2, the existing literature on social networks is related to information advantages on the one hand and corporate governance issues on the other. Based on comparable studies the expectation is made that the disadvantages of a stronger social network on firm value outweigh the advantages. In order to tests whether firm value on the long-term decreases when a noble-titled board member has a seat on a board, three regression equations are applied. These regressions will be run for all noble-titled board members together, as well as separately for noble-titled supervisors and noble-titled managers. The regression equations look as follows:
Qit = β1VARIOUSi + ΣβJ CoBji + Σβk FIRMjit=0 + Qi[t-1] + εi (1)
ΔQi[t-1, t+1] = β1VARIOUSi + ΣβJ CoBji + Σβk FIRMjit+1 + Qi[t-1] + εi (2)
ΔQi[t-5, t+5] = β1VARIOUSi + ΣβJ CoBji + Σβk FIRMjit+5 + Qi[t-1] + εi (3)
Here i represents a variable that is related to a certain company, j represents a particular control variable and t represents the measurement year. The variable VARIOUS represents the variable of interest, which can be NOBLEDUM, NOBLESUP or NOBLEMAN. NOBLEDUMi is a dummy variable that equals ‘1’ when at least one board member is noble- titled and ‘0’ otherwise. NOBLESUP and NOBLEMAN replace the variable NOBLEDUM when a separate analysis on noble-titled supervisors and noble-titled managers is run. Different control groups that represent specific control variables are included to prevent omitted variable bias. CoBji is a sum of variables j related to the Characteristics of the Board of company i. These are included to prevent noise and absorb the relationship between the board composition which could have an impact on the presence of a noble-titled board member and its subsequent (change in) Tobin’s Q. The same counts for FIRMji, which represents firm-specific variables. The control variable FIRM is deployed for the last year of each regression period. This means that for regressions (1), (2) and (3) the variables are taken
on time t0=2004, t+1=2005, and t+5=2009 respectively, because for each regression the content of variables on those dates are the most important drivers behind the change in Tobin’s Q. In Table 1 the variables of each group are specified. Moreover, in case Tobin’s Q comes from an extraordinary level this could have an impact on the magnitude of the change in Tobin’s Q after the introduction of a noble on a board. Therefore in order to prevent a misspecification error, the first- lagged value of Tobin’s Q is included in each equation.
Table 1: Variables of the regression equations. This table provides information on the abbreviations given in the regression equations. The dependent variable is (the change in) Tobin’s Q, The variables of interest are NOBLEDUM, NOBLESUP and NOBLEMAN. The control variables CoB, FIRM and the first-lagged value of Tobin’s Q are explained as well.
Variable of interest Meaning
Dependent variable
(change in) Tobin's Q Proxy for (change in) firm value Independent variable
NOBLEDUM Dummy: noble-titled board member. YES=1 NO=0
NOBLESUP Dummy: noble-titled supervisor. YES=1 NO=0
NOBLEMAN Dummy: noble-titled manager. YES=1 NO=0
Control group variables
CoB Number of supervisors
Number of managers
Average compensation per supervisor Average compensation per manager
FIRM Return on assets
Total assets
Number of employees
Qi[t-1] First lagged value of Tobin's Q
Error term
εi Error term
The first regression equation measures the absolute value of Tobin’s Q at time t=2004. The second regression equation (Δ)Qi[t-1, t+1] measures the difference in Tobin’s Q over the period
[Qt-1=2003 – Qt+1=2005], where t is measured in years. However, it might take some time before
the (dis) advantages of a noble-titled board members’ network are reflected in the change in Tobin’s Q. Therefore, in the third equation ΔQi[t-5, t+5] reflects the difference in Tobin’s Q
over the period [Qt-5=1999 – Qt+5=2009].
Some important assumptions underlie this approach. First, there is only information available on the composition of boards in the year 2004. This composition is assumed to remain the same over time with a maximum of five years around t0=2004 regarding the last regression. More general, due to limited information, within-group variations over time are assumed to be non-existent. This encompasses both the quantity of (noble-titled) board
members, as well as the quality of those (noble-titled) board members. Another assumption is that companies that have noble-titled individuals on their board, only structurally differ from companies without noble-titled board members in the sense that the former has noble-titled board members. Besides, I assume that companies from both groups are not fundamentally different or active in a specific industry. In other words, boards with noble-titled individuals are not more likely to be active in industries that experience more uncertainty and thereby heavier extremes or in industries that may have been into a boom or a bust over the investigated time periods. Moreover, there is no information available on the year that a noble-titled board member joins a board. Therefore I assume that any social network (dis) advantages of noble-titled board members related to firm-value remain constant over time, starts when a noble-titled board member joins a board and ends when he or she departs from the board
The appearance of multicollinearity in a regression could cause imprecise measured OLS estimators and variances (Stock and Watson, 2006). Therefore a correlation matrix is made for every regression and the Variance Inflation Factors (VIF) are calculated. The Variance Inflation Factor indicates whether there is serious multicollinearity. When VIF exceeds 10, serious multicollinearity exists.
Heteroscedasticity is a common challenge while facing cross-sectional data. It states that the error term does not have a zero-mean. In other words, different observations have different variances. In that case, the parameters of the regression become inefficient. Therefore, a White test on heteroscedasticity is run. Under the null hypothesis of White, the errors are homoscedastic. When the null hypothesis can be rejected, homoscedastic standard errors can no longer be assumed. In that case, heteroscedasticity-robust standard errors will be applied.
5.3 Event Study
An event study measures the economic impact of an event on the value of a company (MacKinlay, 1997). As discussed in Chapter 3, the semi-strong form of the Efficient Market Hypothesis (EMH) states that all public available information will immediately be absorbed by market participants and reflected in a companies’ stock price. Regarding the long-term effect of a board members’ social network on firm value, a negative relation was expected. Following the EMH, I expect that the market anticipate on this negative relation. This section therefore describes the methodology applied to test the following hypotheses:
- Hypothesis 3: the announcement effect of the appointment of supervisory board members with a strong social network is negative.
- Hypothesis 4: the announcement effect of the appointment of management board members with a strong social network is negative.
The following section provides information on how hypothesis 3 and hypothesis 4 are tested.
5.2.1 Event Study Methodology
Before an event study can be undertaken, some important decisions should be made (MacKinlay, 1997). These decisions involve the following steps:
- Choose an event of interest
- Choose an event window: a period over which an event occurs - Choose which companies to investigate
- Decide which model to use in order to calculate normal returns
- Define the estimation window: a period over which parameters are estimated - Calculate actual returns and normal returns
- Compute the Cumulative (Average) Abnormal Returns (C(A)AR)
Figure 1 (MacKinlay, 1997): Event study timeline. This figure makes a distinction between the windows that are of interest in event studies. It makes a distinction between the estimation window, the event window and the post event window.
In this study, the event of interest is the announcement that a noble-titled individual gets appointed to the board of a company that is listed on the Frankfurter Wertpapierbörse (FWB). The announcement date is the event date, T0. The event window in this study is [-9, +9] and all time intervals following within the event window from 9, +9] to [-1, +1] will be observed. The most common investigated event window is [-[-1, +1], but due to the possibility that information is leaked prior to the official announcement other event windows can be valuable as well. However, the longer the event window, the less reliable results become. Through time the chance increases that other events – like earnings announcements – influence the returns of a particular company. The underlying assumption of this model therefore is that during the event window no other events have taken place that could influence returns.
All companies listed on the FWB are taken into account, but only those companies that announced a noble to their board count as an event. From those companies, the actual returns on the intervals [-120, -20] and [-9, +9] are taken.
Actual returns are calculated by the following formula:
, (4)
where Rit is the return on a specific stock i on day t, Pt is the price of a stock i on day t and Pt-1 is the price of stock i on day t-1.
Normal returns are the expected returns in the absence of an event. There are different ways to calculate the expected or normal returns. Examples are the Capital Asset Pricing Model (CAPM), the constant expected returns model and the market model (Baker et al., 2007). The first two models won’t be applied in this study. For the CAPM a risk-free rate is required. The constant expected returns model is the simplest model and assumes that the returns of an asset over time are independent and identically normally distributed with a constant mean and variance. However, in this study the method applied by Campbell and Vera (2010) is chosen and the market model is used. The market model assumes that a stable linear relationship exists between the return of the market and the returns of a particular stock (MacKinlay, 1997). Then the OLS estimators are unbiased and efficient. Following Campbell and Vera the estimation window is [-120, -20]. This means that over the period [-120, -20] the relation between the market return and the return of a particular stock is taken. Similar to the condition set by Ishii and Xuan (2014) a stock should have at least 30 non-missing daily returns in order to be included in the model.
,
The ordinary least-squares (OLS) market model looks as follows:
, (5)
where Rit is the return on a specific stock i on day t, Rmt is the return of the market on day t, αi and βi are the parameters estimated for stock i and εit is the error term for stock i on day t. The market return is the return over [-120, -20] of the index where stock i belongs to during the time around an event. All investigated stocks belong to one of the indices described in section 4.3.
Because the model assumes that there is a stable linear relationship between the return of the market and the return of a particular stock, expected returns can be derived from the parameters of regression (5). By subtracting the expected returns from the actual returns, the abnormal returns (ARi,t) can be found. The abnormal returns are calculated as follows:
, (6)
where ARit is the abnormal return on a specific stock i on day t, Rit is the return on a specific stock i on day t, Rmt is the return of the market on day t and αi and βi are parameters estimated for stock i.
The following step is taken to calculate the average abnormal return on day t. The average abnormal return can be found by the following formula:
(7) where AARt is the Average Abnormal Return on day t, N is the sample size and ARit is het abnormal return of stock i on day t.
The Cumulative Average Abnormal Return, CAAR(T1,T2), where T1 and T2 are two specific days within the event period is calculated as follows:
, (8)
The outcomes of equations (7) and (8) reflect whether there is a positive or negative relationship in returns between T days around an event. To investigate whether the results are statistically significant, a test statistic is applied. This is done for all companies listed on the FWB that announced the appointment of a noble on their board, as well as for subsamples including only the announcements of supervisory board member appointments and management board member appointments.
5.2.1 Testing for significance
In order to apply the appropriate test statistic a test on normality of the CAARs is executed first. The reason is that parametric tests – like standard t-tests – on the CAAR rely on the assumption that the abnormal returns of securities are jointly normally distributed and independent across securities.
Therefore, before being able to test a mean with a parametric test two conditions should be met. First the sample must be random and independent. I assume that this is the case. The second condition that should be met is that N must be at least 30 or that the sample comes from a normally distributed population. The former follows from the Central Limit Theorem, which states that the sample distribution of a standardized sample average becomes approximately normal when sample size N is large (Stock and Watson, 2006). When normality can be assumed, the following cross-sectional t-test is applied.
, (9)
where √N is the square root of the quantity of events, CAAR(T1, T2) is the cross-sectional Cumulative Average Abnormal Return over the corresponding time spans varying from [-9, +9] to [-1, +1]. SCAAR is the standard deviation of the CAAR(T1, T2) and is calculated by taking the square root of the following formula:
, (10)
where the square root of s2CAAR is the standard deviation of CAAR(T1, T2), N equals the number of events and CARi is calculated the same way as the CAARs are calculated, but now the sum of the abnormal returns (instead of average abnormal returns) is taken.
Under the null hypothesis of the cross-sectional t-test, the CAAR(T1, T2) is not significantly different from zero. When H0 can be rejected, there is sufficient proof that
CAAR(T1, T2) is unequal to 0 can be assumed.
In case N is smaller than 30, a normality test on the abnormal returns is applied. In order to test whether the sample comes from is normally distributed population a Shapiro – Wilk (SW) test is run on the abnormal returns in the event window of every company.
The SW test detects whether a distribution deviates from normality in terms of skewness, kurtosis or both and looks as follows:
, (11)
where yi is the ith order statistic, y-hat is the sample mean and ai represents a measure of the order statistics of independent and identically distributed (i.i.d.) random samples from a standard normal distribution and a matrix of the covariances of the order statistics. This test appears to perform relatively well for small samples (Razali and Wah, 2011). The null hypothesis of SW assumes that the abnormal returns are normally distributed. In case the null hypothesis is rejected, there is sufficient proof to assume non- normality. The SW is run on the abnormal returns of all companies separately. In case the null hypothesis of the SW test can be rejected at least one time a non-parametric test is more appropriate. This will be the Wilcoxon Signed-Ranks test. The Wilcoxon Signed-Ranks test is calculated as follows (Serra, 2004).
, (12)
Where SUM r+ is the sum of positive differences from the median. The test takes into account the signs and the magnitude of the differences. Moreover, it assumes that the quantity of negative numbers equals the quantity of positive numbers. In other words, the Wilcoxon Signed-Rank test investigates whether the CAAR over the period T1, T2 deviates from showing an equal number of positive and negative CAARs over the above-stated time spans. Zero-values are excluded from the sample.
All test statistics discussed in this section will be tested at a 1%, 5% and 10% significance level. Moreover, the tests will be run for all noble-titled individuals appointed to the boards of companies that are listed on the FWB, to a subset of only noble-titled supervisor appointments and to a subset of only noble-titled manager appointments. In order to get more specific insights, the CAARs are tested on a pre- and post-announcement event window as well for all (sub)groups.
6.1 Data
This chapter addresses the collection of the data required to test the hypotheses. First, section 6.2 describes the collection of data for the long-term cross-sectional regression. Secondly, section 6.2 provides the collection of the data for the short-term event study.
6.1 Data and descriptive statistics #1
6.1.1 Data input in order to run the long-term regression
Table 1 in section 5.1 provides an overview of the variables of interest. The basis of the data on companies listed on the FWB is provided by prof. J.K. Martin. During the year 2004 the composition of all boards from companies listed on the Frankfurter Wertpapierbörse (FWB) is investigated and documented. Of all those companies, information is provided on the quantity of supervisors, the quantity of managers, the average compensation of supervisors, the average compensation of managers, whether there are noble-titled individuals on a board and if so, the specific function of a noble-titled board member is given. Prof. Martin found the information by searching on the prefix ‘Von’ at Hoppenstedt and in the annual reports of the corresponding companies. The regression is solely based on the 2004 dataset. All variables that belong to the variable group ‘CoB’ are provided by the dataset, as well as the variables NOBLEDUM, NOBLESUP and NOBLEMAN. All variables that belong to the variable groups CoB, FIRM, and the (lagged) values of Tobin’s Q are taken from Datastream. For Tobin’s Q, the following equation is used: [(Equity Market Value + Preferred Stock Market Value + Liabilities Book Value)/(Equity Book Value + Liabilities Book Value)].
6.1.2 Descriptive statistics
As discussed in section 5.1 three cross-sectional regression equations are run in order to test whether companies with noble-titled board members show a lower (growth in) Tobin’s Q compared to companies without noble-titled board members. The first regression (1) looks at Tobin’s Q at the year 2004. The second regression (2) investigates the change in Tobin’s Q over a time span of two years and the third regression (3) studies the change in Tobin’s Q over a time span of ten years. The variable of interest is the dummy variable NOBLEDUM that equals ‘1’ when there is a noble-titled board member and ‘0’ when there is no noble- titled board member on the board of a particular company. Variations of this dummy are NOBLESUP, which equals ‘1’ when at least one supervisor is noble-titled and NOBLEMAN,
which equals ‘1’ when at least one manager is noble-titled. The underlying assumption is that noble-titled board members have a stronger social network, which in turn could influence (the change in) firm value. This section provides a detailed overview of the specifics of the data used to run those regressions.
All companies listed on the Frankfurter Wertpapierbörse in 2004 are investigated. As discussed in section 4.3 all indices are reviewed at least once a year based on trading volume and market capitalization. Therefore some companies could be excluded from the regressions. The database consists of 544 companies. For each separate regression, different observations are missing. Companies with missing observations are excluded from the regression. Regarding regression (1) 144 companies, regression (2) 151 and regression (3) 307 companies are excluded due to missing values. This means that the regressions are based on observations of 400, 393 and respectively 237 companies. Appendix Table 2 provides an overview of the data filtering process.
The first regression (1) is based on 400 companies, of which 76 companies have at least one noble-titled board member. Since supervisors and managers fulfill different tasks and responsibilities and thus could influence firm value in a different way, Figure 2 provides information on the noble-titled board members’ positions.
- Regression 1 -
Figure 2: Companies with noble-titled board members in regression (1) data. This figure depicts the total number of companies with noble-titled board members included in regression (1). Top left the output variable of the regression is given, top right the total number of companies in the regression. A distinction is made between companies with only noble-titled supervisors, noble-titled managers and companies that have both - supervisors and managers - with a noble-title on their board.
The figure shows that there are more noble-titled supervisors than there are noble-titled managers. Also, when there is a noble-titled manager on a board, in half of the cases there is a noble-titled supervisor on that board as well. The same pattern is shown in the regressions (2) and (3) where the dependent variable is the change in Tobin’s Q over a particular period.
Table 2 summarizes the data required to run the first regression (1). The mean value of NOBLEDUM, NOBLESUP and NOBLEMAN informs that 18,7%, 17,1% and 3,5% of the total number of companies in the first regression have noble-titled board members, supervisors and respectively managers on their board. Most variables appear to be highly skewed to the right. Of those variables the natural logarithm is taken to make the variables better interpretable. The variables Tobin’s Q and Return on assets are winsorized at a 5 percent level. Since a difference is expected between companies with at least one noble-titled board member and companies without noble-titled board members, these data preparations took place for both company groups apart from each other.
Table 2: Descriptive statistics of regression (1) data. This table sums up the variables of interest, provides an explanation of those variables, informs whether the data is winsorized or whether the logarithm of variables is taken and lists the mean, standard deviation, min. value and max. value.
Variable Explanation Data Mean St. Dev. Min Max
Qt=2004 Qtw Winsorized 1.43 0.660 0.764 3.545
NOBLEDUM Noble dummy - 0.187 0.390 0 1
NOBLESUP Supervisor noble dummy - 0.171 0.377 0 1
NOBLEMAN Manager noble dummy - 0.035 0.184 0 1
Qsuper Number of supervisors Logarithm 1.764 0.650 0.693 3.091
Qman Number of managers Logarithm 1.091 0.525 0 2.565
Csuper Compensation supervisor Logarithm 9.632 0.961 3.025 12.360
Cman Compensation manager Logarithm 12.718 0.938 6.979 15.762
ROAt Return on assets Winsorized 1.341 11.704 (35.85) 20.54
Tat Total assets Logarithm 12.208 2.535 6.957 20.681
nrEMt Number of employees Logarithm 6.650 2.385 0 12.941
Qmin1 First lag Tobin's Q Logarithm 0.280 0.449 (1.109) 2.003
Figure 3 shows the number of companies per index that have noble-titled board members. None of the noble-titled board members work at companies that are listed on the TecDAX. Most of the companies that have noble-titled board members are listed on the CDAX.
However, based on the total number of companies listed on the DAX (30), there are relatively many companies with noble-titled board members (12).
Figure 3: Companies with noble-titled board members per index in regression (1). This figure shows the number of companies per index that
have noble-titled board
