Anders Sørgård Per Skøien - CEO’s Dilemma - Can CEO minority ownership increase the power of the minority group? BI Norwegian Business School – Thesis GRA 19003 Supervisor: Bogdan Stacescu Submission date: 03.09.2012 Campus: BI Oslo Programme: MSc in Financial Economics MSc in Business and Economics – Major in Finance This thesis is a part of the MSc programme at BI Norwegian Business School. The school takes no responsibility for the methods used, results found and conclusions drawn.
66
Embed
CEO’s Dilemma · 2017. 1. 5. · S DILEMMA WITH FAMILY TIES ... In our master thesis we intend to investigate the impact on majority owned firms’ dividend policy of having the
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Anders Sørgård
Per Skøien
- CEO’s Dilemma - Can CEO minority ownership increase the power of the minority group?
BI Norwegian Business School – Thesis
GRA 19003
Supervisor: Bogdan Stacescu
Submission date: 03.09.2012
Campus: BI Oslo
Programme: MSc in Financial Economics
MSc in Business and Economics – Major in Finance
This thesis is a part of the MSc programme at BI Norwegian Business School. The school takes no
responsibility for the methods used, results found and conclusions drawn.
Master Thesis GRA 19003 03.09.2012
Page i
Acknowledgements
This thesis is the result of many people’s effort, and we want to express our
gratitude to all those who have assisted us in various ways.
We would like to thank our supervisor, Associate Professor Bogdan Stacescu, for
his persistent assistance, guidance and encouragement throughout this process.
Without his help the final product could never have been what it is today. We
would also like to thank the Centre for Corporate Governance Research for
providing us with access to the data.
Finally we would like to thank BI Norwegian Business School in general and the
staff of the financial faculty especially for their assistance and availability.
_______________ _______________
Anders Sørgård Per Skøien
Master Thesis GRA 19003 03.09.2012
Page ii
Abstract
The potential conflict of interest between majority and minority shareholders is a
highly relevant topic for many firms. Using a large sample of Norwegian non-
listed companies, this paper investigates the potential effects on this conflict of
having the CEO as one of the minority owners. We find a significant, positive
relationship between the dividend payout ratio and having the CEO as a minority
owner, conditional on the CEO not being in the majority owner’s family. When
the CEO is in the majority owner’s family, we find no such relationship. These
results suggest that in companies with a high potential for majority-minority
conflicts, having an independent CEO as one of the minority owners may help
reduce the potential for the second agency problem.
Master Thesis GRA 19003 03.09.2012
Page iii
Table of Contents
ACKNOWLEDGEMENTS............................................................................................................. I
ABSTRACT .................................................................................................................................... II
TABLE OF CONTENTS.............................................................................................................. III
In our master thesis we intend to investigate the impact on majority owned firms’
dividend policy of having the CEO as a minority shareholder. We find this
interesting because most of the existing literature on potential agency conflicts has
focused on the conflict between owners and management (Demsetz and Lehn
1985; Shleifer and Vishny 1997; Demsetz and Villalonga 2001), and literature
with focus on the conflict between majority and minority owners have mainly
focused on the concentration of minority ownership and family ties between
owners (Faccio, Lang and Young 2001; Berzins, Bøhren and Stacescu 2011).
In majority owned firms the controlling shareholder has both the incentives and
the ability to monitor the management’s actions, thus reducing the potential
conflict between owners and management (Shleifer and Vishny 1986; Tirole
2006; Bøhren 2011). However, in such firms there is a potential conflict of
interest between the majority and the minority owners, due to the incentive of the
controlling owner to extract private benefits (enjoyed only by the majority owner)
at the expense of security benefits (enjoyed pro rata by all security holders). This
problem is referred to as the second agency problem (Villalonga and Amit 2006;
Bøhren 2011) or the majority-minority problem (Shleifer and Vishny 1997). We
will use these terms interchangeably throughout this paper. Given CEO’s position
inside the firm she1 can monitor the actions of the majority owner directly, a
capability that other minority shareholders may not have, especially if the
concentration of minority shareholders is low (Berzins, Bøhren and Stacescu
2011). The goal of this thesis is to investigate if having the CEO as a minority
shareholder – thus possibly increasing the monitoring ability of the minority
group – will effect firms’ dividend payments. Furthermore, we will control for
potential family ties between the majority owner and the CEO as we believe such
ties may have an impact on the CEO’s behavior.
When addressing agency problems with respect to management it may be
counterintuitive to think about the second agency problem. The first agency
problem, between owners and management, is widely researched, and anyone well
read within the topic of corporate governance will likely have touched upon the
1 The reference to the CEO as a female in this paper is made for simplistic reasons and does not necessarily reflect the actual gender of the CEO
Master Thesis GRA 19003 03.09.2012
Page 2
subject. Therefore it is imperative to clarify that even though this paper addresses
agency problems linked to the CEO, we are in fact discussing the second agency
problem. The data sample used in this thesis consists only of observations where
the CEO has an ownership share, majority or minority. Thus a pure owner-
management conflict is not possible with our sample, because the management is
always an owner.
Existing literature primarily focus on public firms, even though a majority portion
of the economy consists of private firms. In Norway 99,9 % of all limited liability
firms are private, (based on average numbers from the period 1994-2008, not
counting subsidiaries or financial firms) and they account for 74 % of aggregate
assets (Bøhren 2011). In this respect Norway is unique in its high data quality and
availability, because mandatory information disclosure is largely the same for
public and private firms (Bøhren 2011).
The outline of this thesis is as follows: In chapter 2 we perform an extensive
literature review presenting relevant theory and empirical research. In chapter 3
we present our data and sample collection, while chapter 4 presents our
hypotheses, methodology and descriptive statistics. Chapter 5 presents the results
of our analysis and provides a series of robustness tests. Finally, in chapter 6 we
present our conclusions, final remarks and suggestions for future research. This
paper also have two appendices, Appendix 1 contains tables 1-27, which report
detailed information regarding our data and results. Appendix 2 contains figures
referred to in the text.
Master Thesis GRA 19003 03.09.2012
Page 3
2. Literature Review
2.1 Dividend theories
In 1956 John Lintner made the argument that a change in a firm’s dividend policy
is a result of management’s believes that earnings have permanently changed. He
stated that “…most managers sought to avoid making changes in their dividend
rates that might have to be reversed within a year or so.” (Lintner 1956, 99). His
article supports the findings of Graham and Dodd (1951) who argued that
dividend policy has an impact on firm value.
This is at odds with the dividend irrelevance theorem first formulated by Miller
and Modigliani (1961). They claimed that in a market without imperfections, firm
value is not affected by its dividend policy. Despite this, researchers have
observed that investors reward firms that pay dividends through higher stock
prices (Johnson, O’Meara and Shapiro 1951; Fisher 1961). This anomaly is
known as the dividend puzzle (Black 1976).
Research on the relationship between dividend policy and firm value is still
inconclusive (Black 1976). Relaxing the assumptions made by Miller and
Modigliani in 1961 may render the dividend irrelevance theorem mute, which has
lead to the development of several explanatory models. But also within the
framework described by Miller and Modigliani in 1961 the dividend irrelevance
theorem is challenged (DeAngelo and DeAngelo 2006).
One assumption Miller and Modigliani make in their paper is that all parties have
the same information. Relaxing this assumption has lead to the development of a
dividend policy model known as the signaling model, (Bhattacharya 1979; Miller
and Rock 1985). This model takes into account the information asymmetry that
exists between outside investors and insiders in the firm. It states that firms use
dividends as a way to signal outside investors about their expected future cash
flows, thus reducing informational asymmetries. Within the signaling model
framework, an increase in dividend payments is taken as a sign of increased
expected future cash flows, which again leads to increased firm value. Benartzi,
Michaely and Thaler (1997) find that stock prices tend to increase following a
positive dividend announcement and decrease following a negative
announcement, but they find no evidence of successive changes in earnings
Master Thesis GRA 19003 03.09.2012
Page 4
following a positive dividend change. However, they do find that following a
negative dividend change, firms experience a significant earnings increase in the
two following years. Also, Benartzi et al. (2005) find that dividend changes are
uncorrelated with future earnings. These findings contradict the signaling model.
Grullon, Michaely and Swaminathan (2002) propose an alternative hypothesis to
the signaling model called the maturity hypothesis. They claim that if the dividend
change is not a signal of changes in future earnings, but still leads to an increase in
the current stock price, then it must be because of a reduction in the firm’s
systematic risk. They explain; “…we should expect dividend increases to be
associated with subsequent declines in profitability and risk.” (Grullon, Michaely
and Swaminathan 2002, 388). As a firm matures, its investment opportunity set
shrinks. This may leave the firm with higher free cash flow, which according to
the maturity hypothesis can explain increased dividends.
Jensen (1986) links this free cash flow to agency theory by claiming that free cash
flow should be paid out as dividends to avoid the potential for agency conflicts.
He defines free cash flow as “…cash flow in excess of that required to fund all
projects that have positive net present values when discounted at relevant cost of
capital.” (Jensen 1986, 2). This is supported by Allen and Michaely (2003) who
states that dividend payouts are used to avoid overinvestment. According to Tirole
(2006), informational asymmetries may prevent outsiders form stopping insiders’
opportunistic behavior, and thus it may be important for companies wanting to
attract outside investors to reduce these asymmetries. Paying out excess cash as
dividends is one way management can reduce the potential dangers linked to such
asymmetries. La Porta et al. (2000) also claim that a firm’s investment policy
cannot be viewed independently from its dividend policy, and that dividends may
be viewed as a way to ensure that all shareholders are paid on a pro rata basis for
their investments.
2.2 Agency Theory
2.2.1 Introduction to Agency Theory
The ideas behind agency theory can be traced back at least to 1776, when Adam
Smith wrote that “…directors of such companies […] being the managers rather
of other people’s money than of their own, it cannot well be expected, that they
Master Thesis GRA 19003 03.09.2012
Page 5
should watch over it with the same anxious vigilance with which the partners in a
private copartnery frequently watch over their own”, (Smith 1776, 606-607). Thus
agency conflicts can be said to come as a result of the separation of ownership and
control. When a company has a single owner, who is also the only employee and
has financed the company using 100 % equity, there are no agency conflicts
(Bøhren 2011). But if one or more of these conditions are not met, ownership
becomes separate from control and the potential for agency conflicts occurs.
In 1976 Jensen and Meckling further formalized a definition of the agency
relationship as “…a contract under which one or more persons (the principal(s))
engage another person (the agent) to perform some service on their behalf which
involves delegating some decision making authority to the agent.” (Jensen and
Mekling 1976, 5). In this definition the authors use the word ‘contract’, which
may have several interpretations. Schanze (1987) points out that even though one
may experience overlap in the real world, it is important to distinguish between
principal-agent relationships and relationships based on formalized contracts. In
its nature a relationship between a principal and an agent is more complex, and
because of this it becomes impractical, if not impossible, to construct an adequate
contract covering all possible actions, rights and responsibilities of the agent in all
possible future states of the world.
Jensen and Meckling (1976) go on to define the agency problem as a result of
diverging interests between the principal and the agent, assuming they are both
utility maximizers. This divergence creates the potential for agency costs. The
authors divide agency costs into three separate sources:
- Expenditures by the principal. In order for the principal to ensure that the
agent acts in her best interest, the principal has to use some resources. This
can be to pay for costly monitoring of the agent or to pay for incentive
alignment programs.
- Expenditures by the agent. It may be necessary for the agent to commit
resources to guaranteeing that she will not take any actions that may harm
the principal’s interests. Such costs are also referred to as bond costs.
- Residual loss. Given that expenditures by both the principal and the agent
are costly, it may be suboptimal to spend all resources necessary to ensure
100 % alignment of incentives between the principal and the agent. Any
Master Thesis GRA 19003 03.09.2012
Page 6
divergence still present after the expenditures of both the principal and the
agent is called the residual loss.
Given that the potential for agency costs is the price of separating ownership from
control, investors need a reason not to simply keep the control for themselves.
Fama and Jensen (1983a; 1983b) look at the fundamental structure of firms, and
point at several potential benefits that can come from this separation. Firstly they
point to the benefits of specialization. A professional manager may be more suited
to manage a company than the company’s investors. In a specialized society,
everyone focuses one their area of expertise, and everyone benefits from it.
Secondly they point out that there are several effective ways of controlling an
agency problem. For example by aligning incentives or by dispersing the decision
making function among several agents. Thirdly the authors discuss the possibility
of controlling managers’ decision making through formal decision hierarchies and
through incentives to ensure mutual monitoring.
Tirole (2006) states that sources of potential agency conflicts occur between firm
insiders and outsiders. And they revolve around three main issues:
- The first is related to informational asymmetries, and it is called adverse
selection. It refers to the fact that company insiders may have private
knowledge regarding the firm’s technology or environment.
- The second, called hidden knowledge, is closely related to the first. The
difference is that while adverse selection refers to information the insider
possesses before entering into a contract with the outsider, hidden
knowledge refers to private information acquired after the date of
contracting.
- The third issue concerns the problem that outsiders have limited ability to
observe the insider’s actions regarding project selection, risk level and
effort exerted. This issue is commonly referred to as the problem of moral
hazard.
2.2.2 The Second Agency Problem
As mentioned in the introduction, the focus of this thesis is on the second agency
problem, i.e. potential conflicts between majority and minority owners. If
investors are not paid on a pro rata basis, this is a sign of majority-minority
Master Thesis GRA 19003 03.09.2012
Page 7
conflicts (Bøhren 2011). As argued above, in firms with a majority owner
conflicts between owners and management become less severe. Instead, the
majority-minority problem becomes an important area of attention. If the majority
owner’s incentives are not aligned with that of the minority owners’, the potential
for the second agency problem arises (Shleifer and Vishny 1997). The potential
for such agency conflicts becomes higher if the majority owner owns close to 50
% of the company’s shares (Berzins, Bøhren and Stacescu 2011). In this case the
majority owner has control of the company, thus enjoying 100 % of any private
benefit extracted, while only enjoying close to 50 % of any dividends. Hence, the
majority owner has an incentive to avoid dividend payments and instead extract
value from the company through various private benefits.
A higher ownership share for the CEO may reduce the potential for the second
agency problem because it will increased minority ownership concentration and
monitoring capability. When the minority ownership is dispersed, it becomes
more costly for these owners to coordinate their efforts to monitor and control the
majority owner (Berzins, Bøhren and Stacescu 2011).
If the CEO herself is the majority owner, it creates a situation where the CEO has
ample opportunities to extract private benefits at the expense of the minority
owners, because now the majority owner monitors herself. Thus this is a situation
with a high potential for second agency problems.
2.3 Theoretical Link between Dividends and Agency Theory
As can be seen in Jensen (1986), La Porta et al. (2000) and Berzins, Bøhren and
Stacescu (2011) dividend policy is commonly used as a measure for agency
conflicts. However, in what way this relationship should manifest itself is subject
to different theories.
Two models connecting dividend policies and potential agency conflicts are the
outcome and substitution model (La Porta et al. 2000). The authors explain that in
firms with high potential for majority-minority conflicts, the outcome model
argues that the majority owner will use his control to extract private benefits at the
cost of the minority owners. As a result, dividends will be low. The substitution
model on the other hand, argues that the majority owner values a good reputation
among the minority shareholders. Given that low dividends are taken as a sign of
Master Thesis GRA 19003 03.09.2012
Page 8
expropriation by the controlling owner, dividend payouts will be higher in firms
where the potential for majority-minority conflicts is higher.
In the same paper, La Porta et al. (2000) find support for the outcome model by
showing that stronger legal protection of minority shareholders is associated with
higher dividend payouts. When looking at majority owned, Norwegian non-listed
firms, Berzins, Bøhren and Stacescu (2011) find support for the substitution
model. Furthermore, they find that the concentration of minority owners is
negatively correlated with dividend payments, which also supports the
substitution model.
One could argue that the link between dividends and agency costs remains
unclear, but Ang, Cole and Lin (2000) claim that measuring absolute agency costs
is problematic because one needs a base case company to compare against, one
with no agency problems2. Even though such companies do exist they are not
publicly traded, which have made acquiring data for empirical research difficult.
However, using a dataset consisting of privately held companies the authors are
able to find such a group of base case firms. They also introduce another
alternative measure for agency costs, the ratio of annual sales to total assets. This
they argue is a proxy for the loss in revenues as a result of inefficient asset
utilization.
In this paper we will use dividends to measure agency costs. We choose this
because this paper is closely linked to the work done by Berzins, Bøhren and
Stacescu (2011), who in their paper also used dividends. We also believe that
despite the criticism it remains the main empirical tool for measuring agency
costs.
The research into agency conflicts in firms falls under the canopy corporate
governance. In economic literature corporate governance is a relatively new field,
but it is rapidly growing, producing around 1000 new research articles every year
(Bøhren 2011). A majority of these papers focus on listed firms, owing to the
limited data availability on non-listed firms (Damodaran 2002). Because dispersed
ownership is more common in listed firms, conflicts between managers and
2 As defined by Bøhren (2011), see chapter 2.2.1 for details
Master Thesis GRA 19003 03.09.2012
Page 9
owners are more researched than conflicts among majority and minority owners
(Berzins, Bøhren and Stacescu 2011).
2.4 Incentive theory
There are two sides of agency theory. One concerns preventing the agent from
opportunistic behavior. The other is about incentivizing the agent to exert the
desired behavior, thus closely linking agency theory and incentive theory. This
need to incentivize the agent also comes as a result of the separation of ownership
and control.
David Sappington begins his paper Incentives in Principal-Agent Relationships
with the quote “If you want something done right, do it yourself”, (Sappington
1991, 1). This quote hints to the problems that may occur from separating
ownership and control. Incentive theory focuses on tasks that the owner finds
either too complicated or too costly to perform herself. In these cases the owner3
will hire someone else to perform the tasks, preferably someone more suited.
The author shows that the moment the two sides of this relationship have
diverging interests, friction occurs4. As mentioned above, Jensen and Meckling
(1976) describe diverging interests as a source of agency conflicts. Sappington
(1991) proposes that in this case it becomes optimal for the principal to align the
incentives of the agent with his own. One approach is for the principal to engage
in monitoring of the agent. Another is to introduce some performance dependent
reward system, which will make the agent’s wealth more dependent on the same
factors as the principal’s wealth.
Holström and Milgrom (1991) show that monitoring and performance incentives
can be both complements and substitutes. They state that when an agent has more
of her own wealth tied to the company, she should be subject to less monitoring.
This is especially true when monitoring is costly. The link between monitoring
and incentive schemes should also lead to lower incentive schemes when
monitoring becomes less costly (Eisenhardt 1989).
3 In economics ’the owner’ commonly refers to the owner of the capital 4 Assuming they are utility maximizing
Master Thesis GRA 19003 03.09.2012
Page 10
Most research done on monitoring and incentive schemes in the agency theory
context have focused on the first agency problem, between the CEO and the
owners, (Eisenhardt 1989; Holström and Milgrom 1991; Core and Guay 1999;
Prendergast 1999; Bebchuk and Fried 2003). They focus on monitoring and
incentive schemes as ways of aligning the incentives of owners and management.
In our paper the main focus is the second agency problem, between majority and
minority owners. Berzins, Bøhren and Stacescu (2011) propose that a majority
owner is concerned with her reputation among existing and potential minority
owners, and thus try to establish a reputation of fair treatment. A potential
economic rationale for this can be that the majority owner predicts a sustained
need for new equity investments in the future, thus increasing her need for a good
report with the minority owners. Thus the reputation effect gives the majority
owner an incentive not to exploit the minority owners.
Furthermore, given the CEO’s unique position as a central person in any
company’s organizational structure, she will know about – if not actively
participate in – any form of private benefit extraction done by the majority owner.
Therefore, as stated by Professor Øyvind Bøhren5, the CEO is in a superior
position to monitor the majority owner. And given that any private benefit
extracted by the majority owner will come at the expense of the minority owners;
an increased ownership share in the company for the CEO will lead to increased
incentives to monitor.
Seward and Walsh (1990) point out that “Top managers are aware of their
precarious employment situation […] they know that they are at risk of being
dismissed for suboptimal organizational performance even if they did not
contribute to the problem (or worse, even if they kept a problem from becoming
as bad as it could have been).” (Seward and Walsh 1990, 430-431). In this paper
the authors focus on the first agency problem. Still, we believe that the argument
that the CEO has a strong incentive to avoid any conflicts with the majority owner
is transferable to the discussion of the second agency problem. Using a dataset
consisting of all public companies in Norway from the period 1989-2001, Bøhren,
Sharma and Vegarud (2004) find that 8 % of CEOs that are forced to quit, resign
as a result of a direct conflict with the company’s owners. Furthermore, 32 %
5 Personal correspondence with authors, 31. May 2012
Master Thesis GRA 19003 03.09.2012
Page 11
resigned because of conflicts with the Board of Directors. Given that the largest
owner has a significant influence over the company’s board, it may be difficult to
accurately separate the causes behind the CEOs’ resignations. Regardless, this
shows that as far as job security counts as an incentive for the CEO, avoiding
conflicts with the majority owner is important.
How Norwegian CEOs value job security can be related to their level of risk
aversion. To provide an accurate measure of risk aversion is outside the scope of
this thesis, but according to the cultural dimensions developed by Hofstede (1980)
Norwegians score 0,56 on the uncertainty avoidance dimension (Geert-Hofstede
2012). Hofstede claims that this would indicate that people are fairly relaxed and
not averse to taking risks. Using this as a proxy for risk aversion among
Norwegian CEOs may indicate that they are willing to engage in some conflicts
with the majority owner, even if this means increased uncertainty regarding their
job security. A second factor that could impact Norwegian CEOs’ level of risk
aversion is that the CEO compensation level in Norway is relatively small
compared to most European countries and the USA, both in absolute numbers and
relative to the average employee (Nielsen and Randøy 2002).
Another, and much more researched, incentive for the CEO is the economic return
in the form of salary and stocks or stock options, (Jensen and Murphy 1990b;
Bergstresser 2006; Bøhren 2011). The CEO is more incentivized to exert effort
when her expected return is dependent on the level of her effort. And equity-based
compensation is an effective way of linking the CEO’s wealth to that of the
owners. The level of compensation is determined by the Board of Directors
(Bøhren 2011), and as mentioned above, the majority owner has severe influence
over the board.
As the CEO’s ownership share increases, the capital returns will become a more
dominant part of her total wealth. Her wealth will thus vary more with that of the
other shareholders (Jensen and Murphy 1990a). Assuming that the CEO is utility
maximizing, this will increase her incentives to actively monitor the majority
owner and engage in conflicts in order to reduce the second agency problem, even
if this may have a negative effect on her job security.
6 On a scale from 0 to 1
Master Thesis GRA 19003 03.09.2012
Page 12
In this thesis we refer to the conflicting incentives of the CEO as the CEO’s
dilemma. On one side, the CEO has economic incentives to reduce the second
agency problem through active monitoring. On the other side, engaging in
conflicts with the majority owner may conflict with the incentive of job security.
This places the CEO in a situation where she has to weigh the conflicting
incentives against each other. Further complicating the issue of CEO’s dilemma is
potential family ties between the CEO and the majority owner, presented in the
following chapter.
2.5 Family Firms
A majority of the firms in our sample are family firms, and the CEO is also often
in the majority owner’s family. When referring to the term ‘family’ in this thesis,
we use a wide definition referring to relationships of both blood and marriage up
to the fourth level of kinship.
From a resource-based perspective Huybrechts et al. (2011) point out that family
firms have a special set of intangible resources compared to non-family firms.
Researchers have classified these resources together into a concept called
‘familiness’ (Cabrera-Suárez, Saá-Pérez and García-Almeida 2001; Sirmon and
Hitt 2003; Chrisman, Chua and Steiner 2005). The concept refers to a unique set
of resources that occurs as a result of the interaction between family and business.
One such intangible resource is that of the organizational culture, (Hofstede 1980;
Peters and Waterman 1982). According to Zahra, Hayton and Salvato (2004)
family firms develop a more group oriented organizational culture, where
members of the family involved with the business collaborate more and tend to
behave altruistically towards each other (Lubatkin et al. 2005). Sirmon and Hitt
(2003) claim that such altruistic behavior within the family results in family firms
having lower agency costs than other firms, especially because members of the
same family tend to have mutually shared interests. However, according to
Berzins, Bøhren and Stacescu (2011) an assumption that family members always
have aligned interests may not be correct, as family members can have different
opinions regarding business matters.
Another intangible resource covered by the familiness concept is human capital.
Human capital is knowledge, skills and capabilities people acquire in order to act
Master Thesis GRA 19003 03.09.2012
Page 13
in new ways (Coleman 1988). Schulze et al. (2001) find that family firms are
more likely to hire directors from within the family, even though outsiders may be
more qualified. This adverse selection problem may lead to a suboptimal human
capital within family firms.
However, both internal and external directors hired into family firms experience a
low turnover rate compared to other firms (Miller and Le Breton-Miller 2003). In
their paper, Miller and Le Breton-Miller look at this lower turnover rate as a
positive human capital resource because the know-how and experience is
preserved within the firm for a longer period of time. We believe that it can also
be looked at as a sign of higher job security within family firms.
If the CEO is a minority owner with family ties to the majority owner, it is
possible that this connection can have an impact on the behavior of the CEO.
Hence it is important to control for family ties when looking at the CEO’s
dilemma, as such ties can complicate the dilemma further. A graphical illustration
of CEO’s dilemma is provided in figures 1 and 2 of Appendix 2.
Master Thesis GRA 19003 03.09.2012
Page 14
3. Data and Sample Selection
The dataset used in this thesis is extracted from BI’s Center for Corporate
Governance Research’s (CCGR) database. The database contains highly detailed
accounting and ownership data for all Norwegian companies with limited liability
(AS and ASA). The accounting dataset is available from 1994-2009, and the
governance dataset, containing information such as ownership ranks, Herfindahl
indexes and family ties between owners, is available from 2000-2009. This high
level of details is made possible by Norwegian law’s strict accounting statement
requirements.
Our population consists of all private, limited liability firms in Norway. In order
to obtain the sample used to estimate our models we apply a series of filters,
reducing the dataset to our final sample, see Table 1 for details. The rationale
behind these filters follows closely that of Berzins, Bøhren and Stacescu (2011).
1. First we exclude financial firms. Such firms operate under special
regulatory capital requirements and special accounting rules.
2. We exclude firms without positive sales and assets. The reason is twofold;
first we want to exclude firms without activity. Second we want to avoid
including shell corporations.
3. We have removed companies where the accounting or ownership data has
inconsistent values. Inconsistent values include largest owner or CEO
ownership outside the range 0-100 percent and negative revenues, equity
or dividends.
4. Given our research question, only companies with a majority owner are
included. This is done to ensure that the agency conflicts in the firms are
majority-minority related, not conflicts between managers and the
controlling shareholder. Also, we exclude companies with only one owner
because such companies are not subject to the second agency problem.
5. Firms with negative earnings are taken out to avoid any problems arising
due to negative payout ratios.
6. Finally the smallest 5% of the firms measured by revenue are excluded
because it can be argued that such firms do not have any agency conflicts.
As outlined in chapter 2.3, a common measure for agency conflicts is a company’s
dividend policy. Still it is important to acknowledge that dividend policy may be
affected by other factors. A firm’s profitability, liquidity and financial constraints
Master Thesis GRA 19003 03.09.2012
Page 15
may affect its ability to pay out dividends. In our thesis we have taken such
possibilities into account by adding control variables. Also, a country’s tax laws
may have an impact on dividend policy. In Norway the taxation on dividends
changed in 2006 with the introduction of the 2006 Norwegian shareholder income
tax, (Alstadsæter and Fjærli 2009). The tax reform resulted in similar effective tax
rates on labor income, capital gains, interest income and dividend income.
Therefore we limit our data sample to include only the years after the reform took
place, in order to avoid tax effects having an impact on dividend policy.
After applying these filters we are left with a sample of 59 535 observations
across four years, 2006-2009. These are again distributed between 11 671 and
16 575 observations per year, see Table 1 in Appendix 1 for details. An important
note is that we have no observations where the CEO is without an ownership
share. This means that the CEO is either the majority or one of the minority
owners in all of our observations. This curiosity is unintended by the authors, and
we will address it again in chapter 6.
As in the papers by Berzins, Bøhren and Stacescu (2011) and Michaely and
Roberts (2012) all firms in our sample are located within the same legal regime,
where minority stockholders are well protected. This is rare in the empirical
literature, as most research has been done using samples reaching across several
different legal regimes (see for example La Porta et al. (2000)).
Master Thesis GRA 19003 03.09.2012
Page 16
4. Methodology, Models and Summary Statistics
4.1 Research question and hypotheses
This thesis places itself under the canopy of CEO ownership and corporate
governance, where its contribution lies in investigating the potential links between
the CEO and the second agency problem. More specifically, if having the CEO as
a minority owner can help reduce the potential for the second agency problem.
Following the argumentation in chapter 2, we have developed the following
general research question:
Is there a relationship between CEO ownership and a company’s dividend payout
policy?
Given our sample, there are three possible cases of CEO ownership: The CEO can
be the majority owner, the CEO can be a minority owner with family ties to the
majority owner, or the CEO can be a minority owner with no family ties to the
majority owner. Together these cases are the foundation for our thesis, as we are
looking at potential differences among them.
As argued in chapter 2 we suspect that family ties may have specific implications
on how the CEO behaves as a minority owner. Therefore we find it necessary to
separate CEOs with and without family ties to the majority owner. To analyze if
there are differences between these three cases of CEO ownership, we have
developed four specific hypotheses:
Hypothesis 1: Having the CEO as a minority shareholder versus a majority
shareholder makes no difference on the company’s dividend payout policy
The alternative hypothesis is that there is a difference in a company’s dividend
payout policy when the CEO is a minority versus a majority shareholder.
Hypothesis 2: Having a minority owning CEO with family ties to the majority
owner has no impact on the company’s dividend payout policy
The alternative hypothesis is that there is a difference in a company’s dividend
payout policy when the CEO is a minority shareholder with family ties to the
majority shareholder versus the two other cases presented above.
Master Thesis GRA 19003 03.09.2012
Page 17
Hypothesis 3: Having a minority owning CEO with no family ties to the majority
owner has no impact on the company’s dividend payout policy
The alternative hypothesis is that there is a difference in a company’s dividend
payout policy when the CEO is a minority shareholder without family ties to the
majority shareholder versus the two other cases presented above.
Hypothesis 4: Having a minority owning CEO has no impact on a company’s
dividend payout policy after controlling for family ties to the majority owner
The alternative hypothesis will be that when including all CEO ownership cases in
the same model, thus controlling for family ties, CEO minority ownership has an
impact on the dividend payout policy.
We will run four models to test our hypotheses; Model 1 is designed to test
hypothesis 1, Model 2 is design to test hypothesis 2, etc. To see if we can falsify
one or more of our four hypotheses, we have developed the following research
model7:
The model follows relatively closely to the model used by Berzins, Bøhren and
Stacescu (2011). The link between their paper and ours was presented in chapter
2. Each model will include different specifications of the CEO ownership
variable, in line with the three cases presented above; CEO as majority, CEO as a
minority with family ties to majority owner and CEO as a minority with no family
ties to the majority owner.
4.2 Dependent Variables
In this thesis our base case will be to measure the dividend payout ratio as
dividends divided by operating income. But, because it is possible for companies
to manipulate earnings numbers and in this way artificially manipulate their
dividend payout ratio, we perform a similar control as La Porta et al. (2000) and
Berzins, Bøhren and Stacescu (2011); we also use alternative measures for the
7 The notation ”ij” refers to company i, year j
Master Thesis GRA 19003 03.09.2012
Page 18
dividend payout ratio. Our two alternate dividend payout ratios are dividends to
sales and dividends to total assets;
In the robustness section of this paper, presented in chapter 5.2.2, we will also
perform a logistical regression where the dependent variable will be a dummy
equal to 1 if the company pays dividends.
4.3 Regression models
CEO ownership is our key independent variable. We run the regression four
times, each time with a new specification of the CEO ownership variable, each
specifically designed to test hypotheses 1 to 4. In Model 1, CEO ownership is a
dummy variable equal to 1 when the CEO is a minority owner in the company. In
Model 2, CEO ownership is a dummy variable equal to 1 if the CEO is a minority
owner and is in the majority owner’s family. In Model 3, CEO ownership is a
dummy variable equal to 1 if the CEO is a minority owner with no family ties to
the majority owner. Model 4 includes both the CEO ownership dummy variables
from models 2 and 3. If both dummies are 0, it means that the CEO is the majority
owner.
According to the competing models presented by La Porta et al. (2000) a
reduction in the second agency problem should lead us to predict a positive in
models 1-3 according to the outcome model, and a non-positive or negative
according to the substitution model. Having the CEO as a minority owner will
potentially increase the power of the minority owners and the outcome of this
will be higher dividend payments. The substitution model expectations will be the
opposite in every model, increased minority power will lead to lower dividend
payments. In the following sections we will for simplicity only present our
outcome model expectations. The expectations for Model 4 will be clarified in
chapter 4.3.4.
Master Thesis GRA 19003 03.09.2012
Page 19
4.3.1 Model 1
In Model 1 we only look at whether the CEO is a minority owner or not,
disregarding any potential family ties between the CEO and the majority owner.
As mentioned in chapter 3, the CEO has an ownership share – minority or
majority – in all our observations, hence this model looks into if there is any
difference between having the CEO as a majority or a minority owner. Given that
the potential for second agency problems is highest when the CEO is herself a
majority owner, we would predict higher dividends with the CEO as a minority
owner. Thus the outcome model predicts a positive coefficient in Model 1.
4.3.2 Model 2
In Model 2 the dummy variable obtains the value 1 if the CEO is a minority
owner with family ties to the majority owner. The effects of such a family
relationship between the CEO and the majority owner can be ambiguous. As
argued in chapter 2.5 family members often have mutually shared interests. This
would lead us to believe that any decision made with respect to the firm’s
dividend policy will not be affected by whether or not the CEO is a minority
owner, because the CEO’s interests are the same as those of the majority owner
anyway. In this case the prediction would be an insignificant coefficient. If
their interests are not aligned, the family ties should not impact the dividend
payout ratio. In chapter 2.4 it was further argued that the CEO faces a dilemma as
a minority owner when her interests are not aligned with the majority owner’s. On
one hand she has incentives to monitor the majority owner and potentially reduce
the second agency problem. On the other hand she has incentives to avoid any
conflicts with the majority owner due to job security considerations. If the job
security incentives outweigh the monitoring incentives we believe that the CEO
will take no action to reduce the second agency problem. Thus having the CEO as
a minority owner should not affect the company’s dividend payments and the
coefficient should be insignificant also here, (having a CEO with aligned interests
or having a CEO with conflicting interests, but who does nothing about it should
provide the same result; a passive CEO). If the monitoring incentives outweigh
the job security incentives the CEO should engage in monitoring, thus reducing
the potential for the second agency problem. If the monitoring incentives
Master Thesis GRA 19003 03.09.2012
Page 20
dominate, the outcome model prediction is a positive coefficient. The above
discussion is visualized in figures 1 and 2.
4.3.3 Model 3
In Model 3 the dummy variable obtains the value 1 if the CEO is a minority
owner and has no family ties to the majority owner. Given that there are no family
ties, this is a case where we expect having the CEO as a minority owner more
clearly will increase the power of the minority group. Again the CEO’s incentives
will be important, and the outcome model prediction follows closely that of
Model 2; if the job security incentives are strongest, should be insignificant. If
the monitoring incentives are the strongest, should be positive.
4.3.4 Model 4
Finally we run Model 4, where we include the two dummy variables from models
2 and 3 at the same time. The first dummy equals 1 if the CEO is a minority
owner with family ties to the majority owner. The second dummy equals 1 if the
CEO is a minority owner with no family ties to the majority owner. If both
dummies equal 0, it implies that the CEO is the majority owner. The motivation
behind Model 4 is that when looking at the three cases of CEO ownership in
isolation, which is done in models 1-3, we could end up with an incomplete
picture of the CEO’s dilemma. Hence we need to control for all family ties in the
same model, and this is done in Model 4. Furthermore, it may give us an
indication as to which of the two possible explanations presented under Model 2
that can explain an insignificant coefficient for the first dummy variable. If the
CEO is unwilling to monitor the majority owner because the job security
incentives outweigh the monitoring incentives, both and are expected to be
insignificant. If, however, turns out to be insignificant and significant, this
could be because CEOs that are related to the majority owner remain passive even
though the monitoring incentives outweigh the job security incentives, simply
because they have aligned interests with the majority owner. Such a result will be
in accordance with Sirmon and Hitt (2003), who claim that members of the same
family tend to have mutually shared objectives. Again, we invite the interested
reader to see figures 1 and 2 for a detailed visualization and clarification of the
CEO’s dilemma, both with and without family ties.
Master Thesis GRA 19003 03.09.2012
Page 21
4.4 Control Variables
This chapter presents and explains the control variables used in our regressions.
The control variable Liq is included because a firm holding a lot of easily
liquidated assets or available cash is more likely to pay out dividends than other
firms, (DeAngelo, DeAngelo and Stulz 2006). Liquidity is calculated as the ratio
of Cash and Cash equivalents to Total assets; and we predict that 8 is positive;
Somewhat similar to the rationale behind the liquidity variable above, we expect
firms with high profitability to pay out more in dividends than other firms (Fama
and French 2001). Profitability is calculated as Return on Assets (ROA), and the
prediction is that the coefficient is positive;
Financially constrained companies may have trouble financing dividend
payments. Companies experiencing high growth may be financially constrained
due to a need to finance this growth (Fama and French 2001; DeAngelo,
DeAngelo and Stulz 2009). The variable Growth is a measure of this, and it is
calculated as sales’ Compounded Annual Growth Rate (CAGR) over the last three
years. When interpreting this variable it is important to note that it is the effect of
sales growth after controlling for profitability. Based on this we predict a negative
coefficient.
The Risk variable is a measure of the volatility of a company’s sales9. If a
company experience high sales volatility, they may be reluctant to pay out the
same proportion of their earnings as dividends as they would with less volatile
sales. This is in line with Lintner’s (1956) explanation that managers increase
8 Models 1-3 have one key independent variable, thus the control variables start from β2. In . Model 4, with its two key independent variables, the control variables are shifted upwards 9 Measured using standard deviation
Master Thesis GRA 19003 03.09.2012
Page 22
dividends only if they are confident that the increase does not have to be reversed
in the near future. The link between dividends and risk can also be explained by
the maturity hypothesis proposed by Grullon, Michaely and Swaminathan (2002),
finding a significant relationship between dividend changes and changes in
systematic risk. Our prediction of the coefficient is thus that it is negative.
Fama and French (2001) found that the dividend paying firms in their sample
were on average 13 times larger than the non-paying firms. Grullon, Michaely and
Swaminathan (2002) explain through their maturity hypothesis that when a firm
matures, its growth opportunities shrink, leaving it with a higher free cash flow.
By linking this to the free cash flow hypothesis by Jensen (1986), the authors
predict that as a firm’s size and age increases, its dividends should also increase.
This leads us to predict positive coefficients for both, and ;
4.5 Summary Statistics
Table 2 in Appendix 1 contains summary statistics. On average about 14 % of the
firms in our sample pay out dividends in a given year, and the average payout
ratio is about 7 % for the entire sample and 54 % for the payers. As in the research
papers of Berzins, Bøhren and Stacescu (2011) and Fama and French (2001), the
median firm in our sample does not pay dividends. The largest owner holds on
average 60 % of the company, and in about 85 % of the cases the largest family
also has the CEO. All the ownership variables are stable over time. Note that in
order to handle outliers, we have winsorized some of the variables, see Table 2 for
details.
Master Thesis GRA 19003 03.09.2012
Page 23
5. Results
In this chapter we will present the results from our regression analysis. Section 5.1
reports our base case findings for the four models presented in chapter 4, while
section 5.2 presents the effect of alternative dividend payout measures and a series
of other robustness tests. In section 5.3 we perform a logistical regression and a
tobit analysis.
Table 3 shows that the correlations between the variables are generally low, and
therefore we do not expect a serious problem with multicollinearity (Gujarati
2003). Furthermore, the correlations between the different dividend payout
measures are fairly high, thus we expect stable results across alternate payout ratio
measures.
5.1 Base Case
In tables 4-7 we report the findings from models 1, 2, 3 and 4 respectively. Each
table presents the findings separately for each year and for the pooled sample.
Also, each table includes results created following the Fama-MacBeth (FMB)
approach (Fama and MacBeth 1973) calculated from the year-by-year estimates.
These are included because they are helpful in order to test the stability of the
parameters. However, when interpreting these results it is important to be aware
of one potential weakness; the FMB findings are only affected by the stability of
the parameters, not whether or not said parameters are significant. Ergo these
results should be handled with some care, as stable, insignificant parameters
across years may result in a significant FMB coefficient.
5.1.1 Model 1
Model 1 looks at the impact on company dividend policy of having the CEO as a
minority versus a majority owner, without controlling for family ties between
owners.
The individual years, the pooled sample and the FMB results all reflect a stable,
positive and significant dummy regressor. Thus, firms tend to pay higher
dividends when the CEO is a minority owner compared to if she is a majority
owner. These findings are in line with the outcome model since the dividend
payout ratio tends to increase when the CEO is a minority owner. One possible
Master Thesis GRA 19003 03.09.2012
Page 24
explanation for this can be that the power of the minority group increases.
Another explanation can be that the majority owner has less power when she is
not the CEO. An important note here is that these explanations are not mutually
exclusive, but can both lead to the findings presented.
The control variables are generally as predicted in chapter 4 both in terms of sign
and significance. For a given ownership structure the liquidity and profitability
variables are significant and positive, implicating that companies with more liquid
assets and higher profitability have a higher dividend payout ratio. Also, as in the
paper by Fama and French (2001), the size variable is positive and significant,
indicating that large companies pay more in dividends than small. As expected,
both the growth variable, which provides an indication of how financially
constrained a company is, and the risk variable have a negative impact on
dividend payments. This implies that firms experiencing high growth and high
sales volatility tend to have lower dividend payout ratios. The age variable is
insignificant in two out of four years, but when it is significant it has a positive
impact on dividend policy, as predicted.
The control variables are stable across all models, hence we will not repeat the
above discussion in the following text.
5.1.2 Model 2
Model 2 looks at the impact on firm dividend policy of having the CEO as a
minority owner when she is related to the majority owner.
The regression yields an insignificant relationship between the dummy variable
and the dividend payout ratio. Hence, the results argue that having the CEO as a
minority owner has no effect on a firm’s dividend payout policy if the CEO is
related to the majority owner. These results can be in line with the arguments of
Sirmon and Hitt (2003), who claim that family members have mutually shared
interests. But they can also be a result of the job security incentives outweighing
the monitoring incentives. To determine the reason behind the CEOs’ behavior,
we need to determine how CEOs that are unrelated to the majority owner behaves.
This is done in models 3 and 4, presented below.
Master Thesis GRA 19003 03.09.2012
Page 25
5.1.3 Model 3
Model 3 is similar to Model 2, but here we examine the impact on a firm’s
dividend policy of having the CEO as a minority owner when the CEO is not
related to the majority owner.
The dummy variable is now significant in three out of four years. Hence, having
the CEO as a minority owner has a positive impact on a firm dividends policy
when she is unrelated to the majority owner. This indicates that the monitoring
incentives outweigh the job security incentives, and it also indicates that the
findings from Model 2 come as a result of the family ties between the CEO and
the majority owner. Again, these results are in line with the outcome model.
5.1.4 Model 4
In Model 4 we combine models 2 and 3 and run the two dummy variables
together. This will as mentioned in chapter 4 give us one model that covers all the
three possible cases of CEO ownership. If both dummies are 0, the CEO is the
majority owner. If dummy 2 (3) is equal to 1, then the CEO is a minority with
(no) family ties to the majority owner. Note that the two dummies are mutually
exclusive, hence they cannot both equal 1 at the same time.
The dummy variable presented in Model 2 remains insignificant and the one in
Model 3 is now significant every year. These findings support that having the
CEO as a minority owner tends to increase a firm’s dividend payout ratio, after
controlling for family ties between the CEO and the majority owner. As argued in
chapter 4 there are at least two explanations for the first dummy variable to be
insignificant. Using the results from this model we argue that minority CEOs are
potentially suited monitors and that the insignificant effect of CEO ownership
when the CEO has family ties to the largest owner comes as a result of her having
aligned interests with the majority owner, not because job security incentives
outweigh monitoring incentives. Again, these results are in line with the outcome
model.
5.1.5 Implications for Hypotheses
Our results have the following implication for our hypotheses stated in chapter 4:
At a first glance one could make the mistake of thinking that one could
unconditionally falsify our first hypotheses. However, when testing hypothesis 2,
Master Thesis GRA 19003 03.09.2012
Page 26
3 and 4 it becomes apparent that whether or not having the CEO as a minority
owner affects a firm’s dividend policy is dependent on her family ties with the
majority owner.
5.2 Robustness Tests
In this chapter we perform a series of robustness tests. This is done to test the
stability of our findings and to look for other possible explanations for the
observed differences in dividend payout policy between the three CEO ownership
cases. First we run models 1-4 using alternative dividend payout measures as our
dependent variable. Second, we include different control variables to see if they
impact our results. Note that these second robustness tests are done only to Model
4, using the original dividend payout ratio, because as argued above Model 4
represents our key findings.
5.2.1 Alternative Dividend Payout Measures
As pointed out in chapter 4 measuring dividends over operating income may in
some cases yield a misleading number. Firms may for example conduct earnings
management, attempting to manipulate their reported earnings thus falsely alter
their dividend payout ratio. To control for this, we rerun models 1-4, this time
using alternative measures for the dividend payout ratio. This follows the
approach of La Porta et al. (2000), Faccio, Lang and Young (2001) and Berzins,
Bøhren and Stacescu (2011).
The first alternative measure is dividends to sales. As can be seen in tables 8-11 in
Appendix 1, the key findings from the base case models are largely reproduced.
One exception is Model 1, where the key independent variable becomes
insignificant in three out of four years. The findings in models 2, 3 and 4 are
stable and in support of the outcome model.
However, some of the control variables are more unstable when applying this
dividend payout measure. In line with the findings of Berzins, Bøhren and
Stacescu (2011), the control variables liquidity and profitability always have
positive and significant coefficients. The growth control variable also remains
stable for all the models. The rest of the control variables are more unstable, but as
argued above they do not affect our findings.
Master Thesis GRA 19003 03.09.2012
Page 27
The other alternative measure of the dividend payout ratio applied is dividends to
total assets. The results of using this payout ratio as the dependent variable can be
found in tables 12-15 in Appendix 1. Here the key findings are the same as in the
original models from chapter 5.1. The relationship between a firm’s dividend
payout policy and the CEO’s ownership share and family ties are stable and the
findings support the outcome model.
Again, the liquidity, profitability and growth variables are stable across the
models. Furthermore, the control variable size is also stable, as opposed to the
original models.
5.2.2 Other Robustness Tests
So far our findings have been in line with the outcome model and stable across
different measures of the dividend payout ratio. In the following we will perform
a series of other robustness tests to control for the sensitiveness of our findings to
changes in the model. The tests have been performed on Model 4, as this model
represents our key findings, and the dependent variable is dividends to earnings.
In the following we will present the key conclusions from these tests. For further
details, please see tables 16-25 in Appendix 1.
First we include a dummy control variable equal to 1 if the CEO has a seat on the
board. Given the board’s role in determining the company’s dividend payout
policy, having the CEO on the board may help explain the CEO’s impact on
dividend policy which we currently attribute to her ownership and family ties.
However, we find no significant relationship between this dummy variable and
the dependent dividend variable, see table 16 for details.
Secondly we include a margin variable, calculated the following way;
Our current measure of profitability is ROA, which by definition punishes firms
that are asset heavy compared to those that are not. Hence, it becomes interesting
to use an alternative measure of profitability to see if it affects our results. Margin
may be one such measure, assuming that companies with higher margins tend to
be more profitable. When applying this alternative measure of profitability, the
Master Thesis GRA 19003 03.09.2012
Page 28
margin variable also turns out positive and significant. However, the dummy
variables do not change, hence our results are robust to this different profitability
measure. If both margin and ROA are included simultaneously, the margin
variable becomes unstable and negative, while the profitability remains positive
and significant. This implies that after controlling for return on assets, having a
high margin has no clear impact on dividend policy. None of these findings have
any impact on our main findings. For further details see tables 17 and 22.
There are several ways of calculating a company’s liquidity. In our original model
we used cash and cash equivalents divided by total assets, but one can argue that
this measure may appear more liquid than it is. For example if the ‘cash
equivalents’ refers to stock holdings which are less liquid, it can be difficult to
quickly exchange these holdings into cash. Thus we also apply an alternative
measure; the company’s cash flow10. The alternative measure is significant and
positive when applied alone. When controlling for the original liquidity measure it
becomes more unstable but remains positive. Neither one has an impact on our
key findings. The models are reported in tables 18 and 23.
In the aftermath of the Norwegian tax reform of 2006 outlined in chapter 3, a
discrepancy occurred between personal taxation and company taxation on
dividends, leaving personal owners subject to tax on dividends while exempting
company owners. Thus it is possible that individual stockowners transferred their
ownership shares to holding companies in order to avoid dividend taxation. We
want to control for this possibility by including the ownership fraction held by
personal owners. If the ownership structures of firms affect the dividend
payments, a lower personal ownership share may lead to increased dividend
payments. However, at some point the personal owners will want to extract their
funds from these holding companies as well, and this transfer will normally come
in the form of taxable dividends. Thus this trend will not in the end exempt the
personal owners from taxation. Our tests show that the ownership fraction held by
personal owners variable is not significant, and has, as expected, no impact on our
findings. Tables 19 and 24 show detailed test results.
As argued in chapter 2, the potential for the second agency problem is reduced as
the ownership share of the majority owner increases. The discrepancy between 10 Cash flow is calculated using the CCGR formula (in millions) and normalized by the log of sales
Master Thesis GRA 19003 03.09.2012
Page 29
private and security benefits are reduced as the majority ownership share
approaches 100 %. Hence the majority owner’s motivation for extracting private
benefits at the cost of dividends are reduced as her ownership share increases. To
see how this effect impacts our findings we include the ownership of the largest
owner as a control variable. However, as can be seen in table 24 this variable is
insignificant.
When ownership becomes more dispersed, there is a greater chance that one or a
few of the owners experience an external liquidity shock while the other owners
do not. Such asymmetric liquidity shocks can be even more plausible if the
owners are not from the same family. On the other hand, if the ownership is
highly concentrated such external liquidity shocks are more likely to apply to
everyone at the same time. When owners face liquidity shocks, they will want
dividend payments in order to acquire the necessary liquid assets. This can lead us
to believe that higher dividend payments may be driven by higher ownership
dispersion, and not the CEO being a minority owner. However, as can be seen in
tables 20 and 25, the control variable included to account for ownership dispersion
– the Herfindahl index – is insignificant and has no impact on our findings.
Table 21 includes all control variables in the same model, and as can be seen this
has no impact on our findings; a stable, significant and positive relationship
between a company’s dividend payout ratio and the CEO being a minority owner,
not related to the majority owner.
5.3 Logistic Regression and Tobit Analysis
As discussed in chapter 4 our median firm does not pay dividends. In fact, only
around 14 % of the companies in our sample are dividend payers. However,
stockholder conflicts may be just as related to dividend payments for companies
that do not pay dividends as for those that do. Therefore it can be interesting to see
if CEO minority ownership actually affects the probability of a company paying
dividends. Also, our findings can be affected by the fact that our dependent
variable has a lower bound of 0, (a company can never pay negative dividends). In
order to investigate these points, we perform a logistic regression and a tobit
analysis.
Master Thesis GRA 19003 03.09.2012
Page 30
5.3.1 Logistic Regression
A logistical (‘logit’) regression can be useful if one wants to look at the
probability of an event occurring, e.g. the probability that a company pays
dividends. The logit model estimates a regression model using maximum
likelihood, because the probabilities are non-linear and cannot be made linear
through transformation. The logit model overcomes the limitation of linear
models that rely on OLS, which can result in fitted probabilities that are negative
or greater than 1, (Brooks 2008). Figure 3 in Appendix 2 illustrates this point.
The probability of a company paying dividends is calculated as follows:
∑
βiXi refers to the independent variables in the model, which in our case are
identical to those used in Model 4 from chapter 5.1.
The results obtained from this model, presented in table 26, show that similar to
our main findings the relationship between dividend propensity and CEO minority
ownership is positive when the CEO is unrelated to the majority owner and
insignificant when the CEO is a minority owner with family ties to the majority
owner. Hence the probability of a company paying dividends is higher if the CEO
is a minority owner unrelated to the majority owner.
5.3.2 Tobit Analysis
Because it is not possible to pay out negative dividends, our dependent variable
has a lower bound of 0. However, it is likely that some companies in our dataset
would in fact prefer to pay negative dividends (i.e. to receive money from their
stockholders rather than paying money to them). Such companies will in our
dataset have observed values of 0, because the dependent variable is censored,
(Brooks 2008; Dougherty 2011).
One indication that we have a censored variable is that a lot of observations are
clustered around the value of 0 (McDonald and Moffitt 1980). If the dependent
variable is in fact censored, this could mean that our slope coefficients are biased
Master Thesis GRA 19003 03.09.2012
Page 31
downwards and our intercept coefficient is biased upwards. The usual way to deal
with the problem of censored dependent variables is by performing a tobit
analysis, (Brooks 2008). The difference between a normal OLS regression and a
tobit analysis is illustrated in figure 4 and 5 in Appendix 2.
Table 27 presents the results of our tobit analysis. Our findings are still the same
as they were in Model 4, supporting the outcome model. Having the CEO as a
minority owner will increase the dividend payout as long as the CEO is unrelated
to the majority owner.
Master Thesis GRA 19003 03.09.2012
Page 32
6. Conclusions and Final Remarks
After presenting our findings and subjecting them to a series of robustness tests in
chapter 5, we will now turn to concluding remarks and final comments. This
chapter is structured as follows; chapter 6.1 provides our conclusions, while
Table 1 shows the effect of applying different filters to produce our final sample (sample all) which we use in the regressions. Thepopulation consists of all Norwegian non-listed, limited liability firms. We remove financial firms such as banks and insurancecompanies due to their special regulatory capital requirements and special accounting rules. Furthermore we remove firms withno activity measured as zero total assets. We also remove spurious data which we define as percent ownership to largest owneror CEO ownership outside the range 0-100 percent and negative revenue, equity or dividends. The alert reader will notice that oursample significantly decreased when we remove spurious data, the reason is mostly due to missing CEO ownership data, whichis a key variable in our thesis. To make sure that conflicts between majority and minority owners is the agency problem weremove all firms that have no majority owner. Similarly we remove single owned firms due to their lack of such conflicts. To avoidincluding shell corporations we remove any company with zero revenues. We delete companies with negative earnings to avoidpotential problems with negative dividend payout ratios. Then we remove the smallest 5 percent of the companies measured bysales since they have no stockholder conflicts. Finally we show the number of firms with positive dividends.
Table 2 shows the mean and (median) of the variables used in our thesis. Dividend propensity is the fraction of firms thatpay dividends, while dividends to earnings, sales and assets are the three ways we measure the dividend payout ratio.Liquidity and profitability is respectively cash and cash equivalents and operating profit to total assets. Growth is measuredas sales' CAGR over the past 3 (minimum 2) years. Risk is the standard deviation of sales divided by the average sales forthe last 7 (minimum 3) years while size and age are the log of sales and the log of number of years since the firm wasfounded respectively. Holding of largest owner is the ownership share of the largest owner while holding of CEO is theCEOs ownership share. Largest family has CEO is a dummy equal to one if the CEO is, or is related to (up to and includingthe fourth level of kinship) the largest owner. CEO minority is a dummy variable that takes the value 1 if the CEO is aminority owner regardless of any potential familie ties to the majority owner. CEO minority inside familes equal to 1 if theCEO is a minority owner and has family ties (up to and including the fourth level of kinship) to the majority owner. CEOminority outside family is equal to 1 if the CEO is a minority owner with no family ties to the majority owner. Note that CEO isregarded as a minority owner if CEO ownership is equal to or below 50% in the previous three variables. Margin is the ratioof operating income to sales and Cash flow is calculated using the standard CCGR formula (reported in millions) andnormalized by dividing by the log of sales. CEO is on the board is a dummy equal to 1 if the CEO is represented on theboard of directors. Herfindahl is a measure of ownership concentration while fraction held by personal owners is thecumulative ownership share held by personal owners. Note that dividends to earnings is winsorized such that the maximum value is set equal to 1 and that dividends to sales and assets are winsorized at the 0 and 99% level. Further is Liquidity,Profitability, Growth, Risk and Margin winsorized at the 0,5 and 99,5% level.
Table 4-7 report the results for the base-case OLS regressions as specified in chapter 5. The p-values are shown in parentheses. Thedependent variable is dividends payable divided by that year's earnings. Further, the dependent variable is winsorized such that themaximal value is set equal to 1. CEO minority is a dummy variable that takes the value 1 if the CEO is a minority owner regardless of anypotential familie ties to the majority owner. Further, CEO minority inside family is equal to 1 if the CEO is a minority owner and has familyties (up to and including the fourth level of kinship) to the majority owner. CEO minority outside family is equal to 1 if the CEO is a minorityowner with no family ties to the majority owner. Note that CEO is regarded as a minority owner if CEO ownership is equal to or below50% in the previous three variables. Liquidity is cash and cash equivalents holdings to total assets and profitability is operating profit tototal assets. Growth is measured as sales' CAGR over the past 3 (minimum 2) years. Risk is the standard deviation of sales divided bythe average sales for the last 7 (minimum 3) years while size and age are the log of sales and the log of number of years since the firmwas founded respectively. Liquidity, Profitability, Growth and Risk are winsorized at the 0,5 and 99,5% level. We report the adjusted R2
for each year and for the pooled sample. Note that for FMB the reported R2 is the average for the individual years and N is set equal to thepooled sample. .
Table 7 - Model 42006 2007 2008 2009 Pooled Sample FMB
Table 6 - Model 32006 2007 2008 2009 Pooled Sample FMB
Table 5 - Model 2FMB
Table 4 - Model 12006 2007 2008 2009 Pooled Sample FMB
Table 8-15 report the results for models 1-4 when we apply different ways to measure the dividend payout ratio. In table 8-11 and 12-15we regress on dividends divided by sales and assets respectively. All independent variables are as explained in table 4.
.
Table 11 - Model 4 with sales2006 2007 2008 2009 All FMB
Table 10 - Model 3 with sales2006 2007 2008 2009 All FMB
Table 9 - Model 2 with salesFMB
Table 8 - Model 1 with sales2006 2007 2008 2009 All FMB
Table 16-25 report the results for various robustness tests we have performed by including and excluding various variables from our base case models. The variables are as explained in Table 2