Corporate Finance in Family Business Groups

Jongsub Lee
November, 2009

Abstract

I present a theory of the optimal capital structure and dividend policy for expanding family business groups
vertically or horizontally. When private control benefits are substantial, takeover threats impose a constraint
on external equity financing. Debt can overcome this restriction but introduces the possibility of bankruptcy
where control benefits are also lost. Relative to a horizontal structure, a vertical pyramid enhances internal
capital financing, but the family has to share more of the profit from the new firm with its existing shareholders, implying that a pyramid is more likely when external financing constraints are more severe, or the
new firm is less profitable but capital intensive. In equilibrium, subsidiaries are less leveraged than horizontal
entities directly controlled by the family, because the parent firm supports subsidiaries with greater amounts
of internal capital. Within a pyramid, the leverage ratio should decrease from top to bottom because the
parent firm has a larger collateralized debt capacity. At the same time, dividend payout should increase
from top to bottom because this is how the family transfers wealth out of the subsidiaries, without selling
control shares to ensure its control over the parent firm against default. Therefore, the theory predicts a
decreasing leverage ratio from top to bottom of the pyramid, supported by a dividend policy where the
parent firm pays out less to maximize group internal capital, while subsidiaries pay out more to service the
parent firms debt. I confirm these predictions using a unique data set on Korean business groups. Together,
the empirical results and theory suggest that the structure of a business group is strategically designed to
maximize control.
Key words: Business group structure, private control benefit, internal capital, capital structure, dividend
policy
JEL classification: G32, G34

PhD Candidate at the Stern School of Business, 44 West Fourth St, Suite 9-197, New York University,
New York, NY 10012, phone: 1-212-998-0438, email: jongsub.lee@stern.nyu.edu.
*Acknowledgements: I am in debt to Marti Subrahmanyam, Kose John, Daniel Wolfenzon, and Raghu
Sundaram for their unconditional support and guidance. I wish to thank Viral Acharya, Sanjiv Das, Yakov
Amihud, Samuel Lee, Joel Hasbrouck, Stijn Van Nieuwerburgh, Jennifer Carpenter, Anthony Lynch, Michael
Lemmon, Bryan Kelly, Xavier Giroud and seminar participants at New York University for their helpful comments. Sooyeon Lee, Farhang Farazmand, and Robert Tumarkin provided excellent support in preparation
of this manuscript. All errors are my own.

Introduction

A business group is a set of multiple legal entities controlled by a single party. The controlling
party can be a family, a state, or sometimes a government. Most corporate finance theories have
been developed to understand how a stand-alone firm operates. However, firms in a business group
tend to cooperate to achieve the specific goal of the controlling party. Thus, to understand how
a firm in a business group makes corporate financing decisions, one needs to take into account
an additional important dimension: the business group structure. This paper presents a theory
addressing the relation between the capital structure decision and the business group structure.
The capital structure should be supported by a dividend policy that is also linked to the business
group structure, and the dividend policy highlights the role of dividends as a means to transfer
capital within a business group (Goplan et al., 2006).
Business groups are prevalent in many countries (La Porta et al., 1999, Claessens et al., 2000,
and Faccio and Lang, 2002). La Porta et al. (1999) document that, of large public firms in
27 countries, more than 30% belong to business groups. Based on that survey, business groups
are typically structured using a vertical pyramidal structure, a horizontal spin-off structure, or
some combination of the two. When a business group expands vertically, the controlling party
attaches a new member firm to the group as a subsidiary of an existing firm in the group. Under
this structure, there are two types of firms: 1) a parent firm located at the top of the pyramid
controlling a number of subsidiaries (hereafter, a central firm) and 2) a subsidiary controlled by a
central firm. In contrast, in horizontal growth, the controlling party sets up a new firm independent
from existing firms in the group and directly controls the firm by using its own shares in the firm.
Under this structure, the new firm operates as a newly spun-off, stand-alone entity (hereafter, a
stand-alone entity).
Based on the choice of business group structure, the controlling party makes appropriate financing
decisions, indicating that the capital structure decisions and dividend policies of the three types of
firms differ according to how the firms are structured within the business group. However, under
what circumstances the controlling party prefers a vertical pyramid over the horizontal structure
has not been studied. The existing literature has not delved into this issue, either theoretically or
empirically.
Most studies focus on the equity ownership structure of the controlling party in a business group,
provided that the business group structure is already in place (Jensen and Meckling, 1976, La Porta
et al., 1997, and Bebchuck et al., 2000). They neither justify the business group structure in the
data nor explore the capital structure of the business group. Few papers directly investigate the
optimal capital structure for leveraged pyramidal business groups (Bianco and Nicodano, 2006,
and Luciano and Nicodano, 2009), but none of them explain how the business group structure,
external equity and debt capital allocation, and a dividend policy should be jointly determined in
equilibrium. To address this void, I develop a model that explains the equilibrium condition in a
unified framework.
La Porta et al. (1999) document that pyramidal structures are popular in countries with weak
protection rights for minorities. They measure the strength of these protection rights using the
anti-director index developed in La Porta et al. (1988) and report that in countries with poor
investor protection rights, 31% of large public firms belong to a pyramid, whereas only 18% of large
2

public firms are pyramidal in countries with good protection for minorities. These studies suggest
that the degree of protection for minority shareholders is related to preference for the pyramidal
structure.
Another stream of studies documents that a corporations control premium is also closely related to
the degree of investor protection rights. Nenova (2003) and Dyck and Zingales (2004) find that the
premium of control rights is substantial when protection rights are weak. Nenova (2003) reports that
the control premium amounts to 50% of a firms total value when the degree of investor protection
rights is weakest. Their findings indicate that the controlling party of a business group generates
substantial side benefits to running a corporation when the protection rights for minorities are
weak; thus substantial private control benefits become a first-order concern of the controlling party
in such an environment. Given this concern, the controlling party has an incentive to enhance
its controlling influence over the business group, which makes a group structure that helps the
controlling party achieve this goal more popular. This suggests that the pyramidal structure can
be viewed as a control-enhancing business group structure, but the exact mechanism by which the
pyramidal structure enhances the controlling partys control over the business group is not clear.
I examine a condition where the pyramidal structure is favored over the horizontal structure as a
control-enhancing mechanism in equilibrium. To this end, I model the decision problem of a family
entrepreneur who wants to expand a business group with a newly arisen project, either vertically
or horizontally, while ensuring its control over the firms in the business group. I start with the
major premise that a corporations control rights are valuable and thus the new project delivers a
substantial private control benefit to the family, in addition to security cash flow benefit. These
benefits induce the family to never give up control of the corporation. Under these circumstances,
I show that the internal financing advantage of the pyramidal structure is useful for the family
to maximize its control benefits over the group firms. I restrict the type of controlling party in
the model to a family rather than a state or government, since a state or government may not
try to maximize private control benefits in addition to the security benefit due to reasons of social
responsibility, even though the private benefit is substantial.
Given the objective of the controlling family, I analyze the following research questions:
When does the controlling family prefer the pyramidal structure over the horizontal structure?
How are different business group structure choices related to the familys control-enhancing
incentive?
What is the optimal capital structure for a subsidiary under a pyramid and for a stand-alone
entity directly controlled by the controlling party under a horizontal structure? Under the
pyramidal structure, should a central firm be more leveraged than its subsidiaries? If so,
what drives the leverage decision of the two types of firms to differ?
How is the dividend policy related to the choice of business group structure? In a pyramid,
are the dividend policies for a central firm and its subsidiaries different? What is the role
of dividends as a means of transferring capital within a business group, and how are they
related to the capital structure in equilibrium?
I answer these questions from a control efficiency perspective, and the efficient decision on group
structure choice, financing decisions, and dividend policy is defined as the one that delivers maximal
3

sum of security cash flow benefit and the expected control benefit. I refer to the optimal decision
as a control-efficient one since the private control benefit is as important as the security cash flow
benefit in this model.
To link financing decisions and dividend policy to the control-maximizing familys problem, I introduce the rule of one share, one vote and a take-over threat into the model. These two elements
implicitly impose a minimum threshold on the controlling partys ownership in a given firm. The
level of the threshold is endogenized as a function of the relative managerial ability of the family
compared to that of a rival who contends against the family to take over control of the given firm.
I assume that when a corporate election is called due to the takeover threat, outside shareholders
will vote for the family or the rival, based on who they believe is best able to run the firm. Because
of this voting behavior of the outside shareholders, the relative managerial ability of the family
determines how much capital the family can raise through external equity financing. A relatively
poorly skilled family faces a very limited equity financing capacity since it cannot sell many shares
to outside capital markets to maintain control of the firm against the rival. When options for
raising capital through external equity are limited, the family will seek other financing methods to
extend the financing capacity of the firm and will thus consider using internal cash resources or
external debt.
A recent study by Almeida and Wolfenzon (2006) proposes a way in which the availability of internal
capital differs between a pyramidal structure and a horizontal structure. The authors show that a
pyramidal structure is useful to help the family extend the availability of internal capital beyond
that under the horizontal structure. They argue that under the pyramid, the family can use the
entire free cash in an existing firm to infuse equity into a new subsidiary investment, whereas under
the horizontal structure, the family can only use its own cash flow stake in the existing firm due
to the independence of the new stand-alone entity from the firm. This argument suggests that the
business group structure determines the level of available internal capital.
I show that this internal capital advantage of the pyramidal structure leads to greater financing
capacity in terms of cash and external equity. Under the pyramid, the family establishes control over
the existing firm, which in turn controls the subsidiary. This indirect control link in the pyramid
implies that for a given level of the familys relative managerial ability, the minimal required
ownership in the new member firm is the same, regardless of the business group structure choice.
Hence, the family can raise exactly the same amount of capital through external equity under
both the pyramidal and horizontal structures; however, due to the greater availability of internal
capital under the pyramid, the maximal amount of cash and external equity financing capacity
under the pyramid is greater than that under the horizontal structure. Therefore, when there is
an ownership limit due to control constraints, the family can implement more capital-intensive
projects by adopting the pyramidal structure without using external debt.
The financing advantage of the pyramid in cash and external equity makes the capital structure
of the subsidiary different from that of the stand-alone entity under the horizontal structure. The
family can also raise capital by issuing external debt in addition to using cash and external equity.
But issuing external debt is costly, since the family loses private control benefits when the firm
defaults and control of a firm can be taken over by creditors under the absolute priority rule
(AP)(Harris and Raviv, 1988, Israel, 1991, and Aghion and Bolton, 1992). Hence, the family will
try to issue the external debt at the minimal required level, and the financing advantage in cash

and external equity in the pyramid implies that the family can issue less external debt under the
pyramidal structure than under the horizontal structure. Since less external debt results in a low
default probability, there is a relative control benefit advantage under the pyramidal structure
compared to the horizontal structure.
But greater availability of internal capital under the pyramidal structure does not come with zero
cost. The internal capital advantage in the pyramid is achieved by fully utilizing the entire free
cash in the firmthe sum of its own money and the money entitled to the minority shareholders
to subsidize the subsidiary, without paying out earnings of the existing firm as dividends. This
implicit dividend policy of the existing central firm under the pyramid implies that the family
needs to share the profits from the subsidiary with the minority shareholders of the central firm.
But if the family chooses the horizontal structure, it can receive full profits from the stand-alone
entity, since it uses its own money for the firms investment. Therefore, the cost of excessive internal
capital available under the pyramid is the partial loss in rent from the subsidiary, and thus the
pyramid is characterized as a relative project rent disadvantage
The trade-off of excessive internal cash in pyramidrelative control advantage and relative rent
disadvantageimplies that the family prefers the pyramidal structure over the horizontal structure
if a new investment is more capital intensive and less profitable. This condition for the pyramid
is consistent with the internal capital market theory that has been applied to business groups
to explain their investment decisions, but the condition derived in this paper is the equilibrium
outcome of the business groups pyramidal structure, which has not been emphasized in the existing
internal capital market theories (Jensen, 1986, and Scharfstein and Stein, 2000).
A less leveraged firm that satisfies the low profitability and capital intensity conditions for the
pyramid emerges as a pyramidal subsidiary. A lower leverage ratio implies higher expected control
benefit, which satisfies the condition where the pyramidal structure is more attractive than the
horizontal structure. Therefore, a firms leverage ratio functions as an important determinant of
business group structure in the model.
In addition, I examine the capital structure within the pyramid more closely and analyze whether
external debt should be issued more by a central firm or by a subsidiary within the pyramid. Under
the pyramid, the additional debt capital raised by the existing central firm can be transferred to
the subsidiary by equity infusion, whereas direct capital transfer is not feasible under the horizontal
structure due to the new stand-alone entitys independence from the existing firm. I characterize the
capital structure for the two types of firms in the pyramidthe central firm and the subsidiary
by proposing two novel structural properties of the pyramid: 1) There is a collateral advantage to
issuing debt from the central firm, since the family can augment the firms own operating cash flow
with the dividends paid by the subsidiary (co-insurance benefit), and 2) control over the subsidiary
is conditional, depending upon the solvency of the central firm (control optionality).
These two structural properties of the pyramid imply that the family will implement an aggressive
payout policy for the subsidiary. Because of its goal to consolidate control over the subsidiary, the
family cannot liquidate the control shares that the central firm holds in the subsidiary, and thus
monetizing cash flow entitled to this position involves using dividend channels to transfer capital
from the subsidiary to the central firm. By imposing the ex post optimal full payout policy for the
subsidiary, it is efficient to issue most of the debt from the central firm and minimally from the
subsidiary, thereby minimizing the probability of losing control benefits from both firms. Therefore,
5

when a fully leveraged pyramidal structure is observed in equilibrium, the leverage ratio tends to
decrease from the top to the bottom of the structure. Moreover, a fully leveraged pyramid with a
decreasing leverage ratio from top to bottom is more likely to be observed if the family effectively
minimizes the groups default probability, which is more likely to be satisfied for a more profitable,
less capital-intensive, and larger central firm than the subsidiary. Therefore, the equilibrium condition of the fully leveraged pyramid suggests that a pyramidal business group pursues profit stability
for survival rather than profit maximization, and this tendency of a pyramidal business group is
documented in many existing empirical studies in the business group literature (Nakatani, 1984,
Prowse, 1992, Ferris et al., 2003, and Shin and Park, 1999).
In sum, the model proposes that a control-efficient, fully leveraged pyramid should exhibit a decreasing leverage ratio1 from top to bottom, suggesting that the leverage ratio is closely related
to the position and centrality of a firm in the business group.2 Moreover, this capital structure
is justified by an increasing dividend payout ratio from the top to the bottom of the pyramid in
equilibrium; central firms tend to pay out less to keep internal capital at maximal availability while
subsidiaries tend to pay out more to extend the central firms debt capacity.
I test the implications for capital structure and dividend policy using a unique comprehensive data
set of Korean family business groups. The government-collected data set provides ownership and
financial information of all firms in Korean business groups in great detail. The data period, from
1998 to 2004, is ideal to test these implications, since it immediately follows the 1997 Asian financial
crisis, during which almost every business group in Korea underwent massive group restructuring.
Using the group structure metrics developed by Almeida et al. (2009), I identify three types of
firms in each business group in the data and document empirical evidence that is consistent with
the theoretical predictions.
This paper is organized as follows: Section 2 provides a baseline model that highlights how the
difference in the availability of internal capital under the pyramid and horizontal structures is related
to the capital structure decisions for a pyramidal subsidiary and a stand-alone group entity. That
section proposes a condition for the pyramid to be optimal in equilibrium. Section 3 analyzes the
capital structure decision of two firms in the pyramid, a central firm and a subsidiary, and justifies a
decreasing leverage ratio from the top to the bottom of the pyramid. To complete the rationale for
how a business group expands, Section 4 discusses the optimality of the multidivisional conglomerate
structure, the third possible structural choice other than the pyramidal and horizontal structures.
Section 5 develops and tests empirical implications from the theory. That section describes the
data of Korean business groups, known as chaebol s, and explains the group structure metrics used
to test the predictions. That section will also provide the main empirical results and the results of
robustness checks. Section 6 presents the studys conclusions.
1

A firms leverage ratio is measured as the ratio of total debt to the firms own operating assets.
The definition of position and centrality will be introduced in Section 5. Explaining the definitions here briefly,
the position of a given firm is roughly the same as the distance of the firm from the family, and a firm controlling a
number of subsidiaries has high centrality.
2

Baseline Model: Capital Structures for a Pyramidal Subsidiary

and a Stand-Alone Entity Under the Horizontal Structure

This section formally presents the family entrepreneurs decision problem. When the family already
owns firm A and tries to set up a new group member, firm B, based on a new investment opportunity,
it needs to make a business group structure choice and financing decision accordingly.

2.1

Model Setting

Economic Environment
The models economic environment is characterized by the following: 1) weak investor protection
rights, 2) a one-share one-vote rule, and 3) the AP in bankruptcy. Hence, a dual class share
structure is excluded from possible ownership structure choices.3 With AP, corporate control is
taken over by creditors when the firm defaults. For simplicity, the market risk-free rate is assumed
to be zero in this economy, and corporate securities are priced based upon the market investors
distributional belief on the cash flow from securities.
Agents and Preferences
The model consists of three types of agents, all of which are risk neutral: 1) a family entrepreneur,
2) a rival management that contends for corporate control against the incumbent family, and 3)
a large number of atomistic passive investors. The passive investors purchase corporate securities
and vote in the corporate election. Hence, they are passive in the sense that they do not contend
for corporate control. All the agents are payoff maximizers due to the risk neutrality of their
preferences.
Managerial Ability and Information Structure
The family and the rival each have their own managerial abilities, which are assumed to be projectspecific, in the new project and thus do not depend upon the incorporation choice or the financing
decision of the new firm. I denote the managerial ability of the family by p, where p is assumed
to be uniformly distributed on the interval [0, 1]. The rivals managerial ability is also uniformly
distributed on [0, 1] and is assumed be randomly drawn from the distribution. The familys managerial ability level p and the distributional assumption on the managerial abilities of both the
family and a rival are assumed to be common knowledge in this economy. I assume that any agent
is limited to this knowledge, and thus the information structure on managerial ability is symmetric.
Technology
3
Dual class structures cannot be optimal, even under the settings of the model in this paper, since there is a cost
to issue unlimited amounts of the dual class shares that comes from legal restrictions on the use of the shares or share
price discounts due to liquidity issues. I do not discuss this in this paper.

A project available to the family is summarized by a triple, (Y, i, K), where i is the amount of
investment outlay and Y = Y (p, i) is the stochastic cash flow of the project that depends on p
and i. The mean cash flow of the project, Y , is Y = (1 + pr(i))i, where r(i) is the net mean
return of the project cash flow when p = 1, and thus the mean cash flow increases in p. 4 The
NPV of the project, Y i, is assumed to be non-negative i, p [0, 1]. Moreover, I assume
that managerial ability only affects Y but not the cash flows risk profile. 5 Thus, denoting the
stochastic component of the project cash flow by , Y = Y + , where  follows a cumulative
distribution function (c.d.f.) of F () and F () is assumed to be in the increasing hazard rate (IHR)
family. Hence, H() f ()/(1 F ()) increases in .6 Lastly, K = K(i) denotes a deterministic
amount of control benefit from the new project, exclusively available to the corporate controller at
the end of the project. I assume that K(i) does not depend on managerial ability but only on i.7
I assume K
i > 0, which implies that the absolute amount of private control benefit for a larger
project is greater than that for a smaller one.
Dates of Actions
The model has three dates: date 0, date 1, and date 2. At date 0, a family entrepreneur with zero
wealth starts up a firm A based on a positive NPV project, (YA , iA , KA ), with pA and c.d.f. of
FA (.), by selling a fraction (1) of the firms shares8 and the firm generates cash flows and private
control benefits at both date 1 and date 2. Suppose that at date 1, firm A generates cash flow
C and private control benefit KA . The family consumes KA and a new investment opportunity,
(YB , iB , KB ) with pB and FB (.), arises. Given the established ownership, , in firm A in addition
to the firms retained earnings of C, the family tries to set up a new group firm B based on this new
project. The new group member firm B will be structured as either a subsidiary of the existing firm
A (in a pyramidal structure) or a stand-alone entity that is independent from firm A and controlled
directly by the family (in a horizontal structure).9 . The family can finance the new firm with any
mix of cash, zero-coupon debt, and equity under each business group structure. There is a market
for corporate control that is assumed to be open immediately after the family makes the business
group structure and financing decisions. In the market for corporate control, a rival management
emerges and contends against the family for control over firm B. The rival is assumed to initially
have no equity in firm B and emerges with zero wealth for simplicity. These assumptions imply
4
WOLG, I normalize the net present value (NPV) of the project cash flow when p = 0 to be zero, and thus

Y (p = 0, i) i = 0.
5
A project with a higher p first-order stochastically dominates a project with a lower p.
6
This assumption is made to guarantee unique interior optimality of the debt financing decision of a given firm
and, in fact, is not restrictive, since it is valid for many probability distributions, such as the (truncated) normal,
(truncated) exponential, uniform, and Laplace distributions. Gamma and Weibull distributions with more than one
degree of freedom also satisfy this assumption. For more details on the IHR family, see Barlow and Proschan (1975).
This assumption is also used in Chemmanur and John (1996) and G
arleanu and Zwiebel (2009).
7
Clearly, the technology of K depends on the strength of investor protection rights. If I denote the strength by
s, where 0 s 1, then the higher s, the more minority investors in the economy are protected legally, and thus
K(i;s)
< 0 holds i.
s
8
This date 0 all equity financing assumption for firm A is taken to simplify the analysis of the model, and can be
relaxed.
9
In addition to these two business group structure choices, the family can adopt the new project as a new division
of the existing firm A (hereafter, conglomerate structure), but for the first part of the model, I focus on the pyramidal
and horizontal structure choices. Section 4 later compares the two structures to the conglomerate structure.

that the rival contends for corporate control over firm B solely by relying on the proxy votes it can
win from passive investors in the corporate election. Once the corporate election results are in, the
winner of the election will control firm B and implement the new project.10 At date 2, the project
cash flows in both firms A and B will be realized and the payoff to the family will be determined.
The payoff to the family at date 2 under each of the two business group structurespyramidal and
horizontalconsists of the sum of the cash flow and private control benefits from both firms A and
B.
Valuable Corporate Control and Agent Voting Behavior
Assumption 1 (Valuable corporate control). The right to control a corporation is valuable and thus
the incumbent management is not willing to give up corporate control, even if a rival management
offers better security benefits.
The deterministic amount of private control benefit, K, possibly from managerial perquisites, any
diverted corporate resources through accounting fraud, or reputation effects, cannot be contracted
away in a cost-free manner. Since I consider an environment with weak investor protection rights,
the amount of benefit is expected to be substantial.
Assumption 1 induces the family and a rival to be control driven.11 Hence, in the corporate control
contest, the family will vote for itself even if it knows its managerial skills are inferior to those of
any rival. Since the rival emerges with zero wealth and no equities in a given firm, it contends for
control over the firm solely to receive the private control benefit and also votes for itself.
The passive investors, who hold the floating shares of outside equities, vote for either the family
or the rival, based on who they believe is best able to run a corporation, and their decision is not
affected by any financing or dividend policies.12 The fraction of votes that the family is expected
to win from the passive investors is denoted by (p), and I assume that the family can win more
votes from the passive investors when its managerial ability level is higher (i.e., a higher p). Hence,
(p)
p > 0 holds.
At date 1, the family compares financing decisions under each of the two business group structures
and chooses to set up the new firm B under the structure that provides a higher expected payoff at
date 1. The joint c.d.f. of the two firms projects is denoted by FA,B (., .). The probability density
functions are denoted by lowercase f (.).
10

In this model, the actual rival takeover does not occur in equilibrium, since the family strategically chooses its
ownership level to ensure its control over firm B.
11
This behavior of the family and the rival can be formally stated as follows: Denote the value of the equity of
a given firm with a debt of face value D, the incumbents managerial ability level of p [0, 1] by S(p, D), and the
expected control benefit with a debt of face value D by k(D). Then, both the family and the rival are control driven
if S(p = 0, D) + k(D) S(p = 1, D) D 0. This implies that even if the family is more poorly skilled than any
possible rival, it is not likely to give up control to the rival to receive a higher security benefit from its equity stake.
I implicitly exclude the possibility that the rival offers a strategic tender to the family at the bid price of the sum of
its security and control benefits (Grossman and Hart, 1987).
12
When a firm is highly leveraged or pays out substantial amounts of dividends, it is more vulnerable to a rivals
strategic tender offer. I exclude this possibility in this model. Thus, dividend policy and capital structure decision
do not affect the outcome of the corporate control contest.

2.2

Horizontal Structure

Under the horizontal structure choice, the family has a personal wealth of C to spend for the
new firm B, since the firm is independently incorporated of firm A. Hence, the family implements
a full payout policy for firm A and receives C as dividends. Let the amount of cash financing,
the fraction of equities held by the family, and the amount of zero-coupon debt for firm B under
H , respectively, and 0 tH C. The total amount of
this structure be denoted as tH , H , and DB
capital raised for firm B by internal and external financing under this structure is denoted by MBH
H ) + B (D H ) where S (D H ) and B (D H ) respectively denote the
and MBH = tH + (1 H )SB (DB
B
B
B
B
B
B
H . I define a stochastic total cash
date 1 values of firm Bs equity and debt with face value of DB
H , where X H = M H i + Y . Then, the
flow in firm B at date 2 before any debt repayment, XB
B
B
B
B
H , is
familys stochastic payoff at date 2 under the horizontal structure,


H = C + YA + tH + H (XB DH )+ + KA + KB 1 H H

{X D }
B
B

When the family sells (1 H ) shares of firm B, it wants to ensure control over firm B with a
fraction H of equities. In the corporate election, a fraction (1 H ) of equities held by the passive
investors will be won by either the family or the rival. By the definition of (p), a fraction (pB )
of the passive investors votes is expected to be won by the family in the corporate election. This
implies that the family can maintain control rights over firm B if
H + (pB )(1 H ) (1 H ) (1 (pB ))
Therefore, the familys equity financing capacity is
H
M ax( 12 12
1 , 0) 1

(pB ) H 1
where


(pB ) = M ax

1 1 2(pB )
,0
2 1 (pB )

B)
The lower bound of the equity financing capacity, (pB ), satisfies (p
0, since (p)
pB
p > 0.
Hence, for a higher pB , the family can sell more shares of firm B to the external capital market.
This lower limit of the equity financing parameter, (pB ), is denoted hereafter as B for notational
simplicity.

Because investors are risk neutral and break even in equilibrium, the following should hold:
H ) + B (D H )
MBH = tH + (1 H )SB (DB
B
B
H D H ]+ = Y
B iB
tH + H E FB [XB
B

Using this condition, the date 1 expected payoff to the family under the horizontal structure,
H ], can be written as
H E FA,B [
10

H )]
= C + YA + YB iB + KA + KB [1 FB (XB < DB

The familys date 1 program under the horizontal structure is

H)
M ax(tH , H ,DH ) H (tH , H , DB
B

B H 1

s.t.

MBH iB
0 tH C
Lemma 1. It is optimal for the family to set MBH = iB .
Proof. See Appendix B.1.
When the investment outlay can be supported only by cash and equity financing, any surplus in
the raised funds does not help increase the familys payoff. When debt should be issued, a marginal
dollar increase in the face value of debt provides additional capital strictly less than a dollar, since
the debt is risky. Therefore, any fund surplus in this case reduces the familys payoff, and thus is
not optimal. This tight financing condition can be applied to any choice of business group structure
and is therefore applied throughout the rest of Section 2.
Imposing the tight financing condition from Lemma 1, the date 1 program under the horizontal
structure is
H )]
M ax(tH , H ,DH ) C + YA + YB iB + KA + KB [1 FB (DB
B

s.t.

B H 1
(1)
tH

H) = i
+ YB H SB (DB
B

0 tH C
where
H
H +
SB (DB
) = E FB [YB DB
]

The first constraint is the equity financing constraint, and the second is the budget constraint. The
third is the cash financing constraint.

2.3

Pyramidal Structure

When the pyramidal structure is chosen, the availability of the familys internal cash resources is
extended to the entire retained earnings, C, in firm A, since the family allows firm A to invest in
the new subsidiary B. This is achieved by implementing a zero-payout policy for firm A. Let one
11

denote the amount of cash financing, the fraction of equities held by the family, and the amount of
P , respectively, and 0 tP C.13
zero-coupon debt for firm B under this structure by tP , P , and DB
By imposing tight financing conditions, the familys date 2 stochastic payoff under the pyramidal
P , is given by
structure,


P = C tP + YA + P (YB DP )+ + KA + KB 1

B
{YB DP }
B

Since the managerial ability of the family is assumed to not depend on the choice of business group
structure, the familys equity financing capacity under the pyramidal structure is the same as that
under the horizontal structure. This implies
B P 1
By imposing the breakeven condition, the familys date 1 program under the pyramidal structure
is
P )]
M ax(tP , P ,DP ) C + YA + (YB iB ) + KA + KB [1 FB (DB
B

s.t.

B P 1
(2)
P) = i
tP + YB P SB (DB
B

0 tP C
where
P
P +
SB (DB
) = E FB [YB DB
]

The first constraint is the equity financing constraint, and the second is the budget constraint. The
third is the cash financing constraint.14

2.4

Results

Under both the pyramidal and horizontal structures, risky debt is issued if the investment outlay,
iB , cannot be supported by only cash and external equity.
13

In this baseline model, I exclude the possibility that the family can raise debt through firm A and subsidize firm
Bs investment through this extra debt capital. This possibility will be analyzed in Section 3.
14
I assume that under the pyramidal structure, with fraction of ownership in firm A, the familys control of firm
A at date 1 is secured. This assumption does not affect the main results in this paper and can thus be relaxed if
necessary.

12

Lemma 2. Under the horizontal structure,

H
H
(tH
, , DB )
(t, , 0) if iB C + (1 B )YB

where t [0, C], [B , 1] s.t. t + (1 )YB = iB

=

H ) if iB > C + (1 B )YB

(C, B , D

where S (D
H ) = C+YB iB
B
B
B

Under the pyramidal structure,

, P , DP )
(tP
B

(t, , 0) if iB C + (1 B )YB

where t [0, C], [B , 1] s.t. t + (1 )YB = iB

=

P ) if iB > C + (1 B )YB

(C, B , D

where S (D
P ) = C+YB iB
B
B
B

Proof. In Appendix B.2.

Lemma 3. For a given iB , the leverage ratio for firm B, measured as the ratio of the book value of
the total debt to the book value of stand-alone assets, under the pyramidal structure is lower than
that under the horizontal structure.
Proof. See Appendix B.3.
When the family chooses the pyramidal structure, the leverage ratio of a subsidiary B is less
than that of a stand-alone entity B under the horizontal structure due to the greater availability of
internal capital. The internal financing advantage under the pyramid enables the family to minimize
the use of debt and can thus lead to a greater expected control benefit. Therefore, the pyramid
conveys a relative control benefit advantage, but this advantage does not come up at zero cost in
equilibrium. The following proposition describes the condition where the pyramidal structure is
optimal in equilibrium.
Proposition 1. The pyramidal structure is chosen as the optimal business group structure in
equilibrium only if




H ) FB (DP ) (1 ) YB iB
iB C + (1 B )YB and KB FB (D
(3)
B
B
Proof. See Appendix B.4.


On the right-hand side (RHS) of the second condition, (1 ) YB iB shows that the cost of the
excessive internal capital availableunder the pyramid is the division of rent from the new project,
H ) FB (DP ) , on the RHS of the second condition dewhereas the positive amount, KB FB (D
B
B
notes the benefit of excessive internal capital under the pyramid. Under the pyramidal structure,

13

only a fraction of rent is available to the family, whereas full rent is available to it under the horizontal structure. Hence, there is a trade-off for excessive internal cash under the pyramid between
the relative expected control benefit advantage and the relative rent disadvantage. This trade-off
implies that the pyramidal structure becomes optimal in equilibrium only if the relative control
advantage dominates the relative rent disadvantage.
Equation(3) in Proposition 1 implies that the pyramidal structure is more likely to be optimally
chosen in equilibrium if 1) the new project is capital intensive (i.e., iB is large) and 2) the project
delivers lower rent (i.e., lower YB iB ) and higher control benefits (i.e., higher KB ) and thus the
cash capital becomes useful.
For a small investment, the family can support the required investment outlay without hurting
the expected control benefit, even under the horizontal structure. Therefore, for a small project,
there is no benefit to adopting the pyramid because of the relative rent disadvantage. Hence, for a
large-scale project, the family will favor the pyramidal structure over the horizontal structure.
Due to the relative rent disadvantage, the family will not prefer a profitable investment under the
pyramidal structure. The more profitable it is, the more costly it is to use the excessive internal
capital under the pyramid. This preference for the less profitable and capital-intensive project
choice under the pyramid is consistent with existing theories, such as those of Bebchuck et al.
(2002) and Almeida and Wolfenzon (2006).
Lastly, Proposition 1 suggests that the amount of debt in firm B is an important determinant of the
optimal business group structure, since the debt amount determines the expected control benefit.
This suggests that for a given amount of control benefit KB , the lower a firms leverage, the more
likely the firm is to be adopted as a pyramidal subsidiary in equilibrium. Hence, in equilibrium, the
expected control premium of a pyramidal subsidiary should be higher than that of a firm controlled
directly by the family.
Corollary 1. A relatively poorly skilled family will find a pyramidal structure to be optimal in
equilibrium. If the pyramidal structure is chosen, the familys managerial ability, pB , is more likely
to be below the unconditional average, 0.5. This relation implies that in equilibrium, the ownership
of a pyramidal subsidiary tends to be concentrated.
Proof. See Appendix B.5.
Corollary 1 indicates that the internal capital advantage is especially useful when the family has
inferior equity financing capacity due to poor managerial skill. This implies that the pyramidal
structure is favored over the horizontal structure if the family is relatively less able than a rival.
The familys relatively poor managerial skills for a pyramidal subsidiary imply that, in equilibrium,
the ownership of a pyramidal subsidiary is more concentrated than a stand-alone entity directly
controlled by the family, since (p)
p 0. This result resolves puzzling empirical evidence that the
separation between cash flow and control rights under the pyramid is not as extreme as documented
by Franks and Mayer, 2000.

14

Extension: Capital Structures for a Central Firm and a Subsidiary in a Fully Leveraged Pyramid

Under the horizontal structure, any new debt issuance from firm A at date 1 is not desirable since
1) the capital raised from the new debt cannot be used to support firm Bs investment due to firm
Bs independence from firm A15 and 2) any increase in the probability of firm A defaulting reduces
the expected control benefit from the firm, which is ensured without the debt.
Under the pyramidal structure, however, determining whether issuing debt from firm A at date 1
is optimal it is not a straightforward matter, since direct capital transfer from firm A to firm B
is feasible through the equity infusion. When the family adopts the pyramidal structure, central
firm A invests in the subsidiarys equities. This implies that the firm acquires a new asset, control
shares in firm B. Hence, under the pyramidal structure, central firm A has two types of assets: 1)
its own existing operating assets and 2) the newly acquired control shares in firm B. Therefore,
if the control shares can be used as additional collateral for firm As debt, the family can further
reduce the debt amount that firm B should issue for the new investment. Given these conditions,
I analyze the optimality of firm As recapitalization under the pyramid.
To simplify the analysis, I assume that at the current ownership level of the family in firm A, ,
the familys control over the firm is tightly secured, and thus there is no further equity financing in
firm A at date 1.16

3.1

Pyramidal Structure: Date-1 Program Reformulation

At date 1, the family considers raising capital for firm B by issuing zero-coupon debt with face
P from firm A.
value DA
P , D P , ). The
Let the amount of additional capital raised by firm As debt be denoted by BA (DA
B
P
stochastic total cash flow in firm B before debt repayment at date 2 is denoted by XB = MBP iB +
YB , where MBP is the total capital raised by internal and external financing for firm B. Here is a
P D P ]+ ), paid out as dividends. When
fraction of the cash flow claim of firm Bs equity (i.e., [XB
B
P D P ]+ as a non-operating profit. Therefore,
firm B pays out dividends, firm A receives P [XB
B
P (), is written as
the total cash flow in firm A before any debt repayment at date 2, XA
P
P
P
P
P +
XA
() = C + BA (DA
, DB
, ) tP + YA + P [XB
DB
]
P at date 2 by only using X P (), since X P () is the actual amount of
The family needs to repay DA
A
A
cash the family has access to for debt repayment. Firm As default is then defined by 1{X P ()<DP } ,
A
A
with 0 1. With this default trigger for firm A, firm As equity holder has a claim on
15
I implicitly exclude the possibility that the family will raise capital by issuing debt from firm A and pays out the
capital as dividends to transfer it from firm A to firm B. This would be equivalent to prematurely cashing out the
future cash flow from firm A. This possibility is excluded by assuming that external capital can be raised only after
firm As dividend policy is implemented. However, allowing this possibility does not affect the main results in this
section; it may, however, conditionally affect the capital intensity prediction of the central firm in Proposition 3.
16
The mathematical definition of tightly secured is following: (pA ) = . This assumption can be relaxed, and
does not affect the main implication of the model in this section.

15

P ( = 1) D P ]+ 1
[XA
P ()D P } . Note that the default trigger is defined with 0 1, but the
{XA
A
A
claim amount in solvency is defined with the maximum of = 1.

Therefore, the familys date 2 stochastic payoff under the pyramid is

P () = [X P ( = 1) DP ]+ 1 P

{X ()DP }
A
A
A

+KA 1{X P ()DP }

A

(4)

+KB 1{X P ()DP ,

A

P D P }
XB
B

Equation (4) highlights an intrinsic property of the pyramidal structure. The third line in Equation
(4) indicates that the familys control over firm B is subject to firm As solvency. When firm A
defaults, its control rights are transferred to its creditors, which implies that the creditors of firm
A become the effective controller of firm B in such an event. Hence, there is intrinsic control
optionality under a pyramid.
P > 0, D P 0, and tP , if X P ( = 0) < D P X P ( = 1), then WLOG,
Lemma 4. For a given DA
A
A
A
B

it is optimal to set = 1

Proof. See Appendix B.6.

Lemma 4 says that when the family cannot repay firm As debt with the firms own operating cash
flow, it has an incentive to transfer capital from firm B to firm A. Due to the non-marketability
of firm As control stakes in firm B, the capital should be legally transferred through dividends.
Lemma 4 introduces another intrinsic property of the pyramid, called co-insurance benefit. The
asset composition of central firm A and the optimal dividend policy for subsidiary B imply that
there is a co-insurance benefit by augmenting cash flows from the two assets to service the central
firms debt.
By imposing Lemma 4, the familys date 2 stochastic payoff under the pyramid is
P = [X P ( = 1) DP ]+

A
A
+KA 1{X P ( =1)DP }
A

+KB 1{X P ( =1)DP ,

A

P D P }
XB
B

Because investors are risk neutral and break even in equilibrium,

P ( = 1) D P ]+ = C + Y
A + YB iB
E FA,B [XA
A

Using this condition, the familys date 1 expected payoff under the pyramidal structure with = 1,
P ], is
P = E FA,B [

16

= C + YA + (YB iB )
P ( = 1) D P }
+KA P rob{XA
A
P ( = 1) D P , X P D P }
+KB P rob{XA
A
B
B

The familys date 1 program under the pyramidal structure with = 1 is finally written
P , D P ; = 1)
M ax(tP , P ,DP ,DP ) P (tP , P , DA
B
A

B P 1

s.t.

MBP iB
P , D P ; = 1)
0 tP C + BA (DA
B

where

P)
MBP = tP + YB P SB (DB




P , D P ; = 1) = E FA,B M in D P , X P ( = 1)
BA (DA
A
A
B
The first two constraints are the equity financing constraint and the budget constraint, respectively,
and, the third is the cash financing constraint.
P > 0), it is optimal to set tP = C + BA (DP , DP ; =
Lemma 5. For iB > C + (1 B )YB (i.e., D
B
A
B
P , D P ; = 1) + Y
B P SB (DP )
1) and MBP = iB where MBP = C + BA (DA
B
B
Proof. See Appendix B.7.
Imposing Lemma 5 yields, with = 1,
tP



P +
P +
P +
] DA
= C + YA + P E FB [YB DB
] E FA,B YA + P [YB DB
|
{z
}
P ,D P ; =1)
BA (DA
B

P ) = i gives a new group budget constraint with = 1

Plugging this into tP + YB P SB (DB
B



P ]+ D P + = i
C + YA + YB E FA,B YA + P [YB DB
B
A
P > 0), the familys date 1 program under
Focusing on the case where iB > C + (1 B )YB (i.e., D
B
the pyramidal structure with Lemma 5 and the optimal ex post dividend policy, = 1, is written
P , DP )
M ax(DP ,DP ) P (DA
B
A

s.t.

(5)
P , DP ) = i
C + YA + YB SA (DA
B
B

17

where

= C + YA + (YB iB )
P D P } + K P rob{X P D P , Y D P }
+KA P rob{XA
B
B
A
A
A
B

(6)
P = Y + [Y D P ]+
XA
A
B B
B



P , D P ) = E FA,B Y + [Y D P ]+ D P +
SA (DA
A
B B
B
B
A
P > 0, P = B . Here = 1 is applied and is assumed hereafter in this
Since one assumes D
B
analysis. The constraint is the new group budget constraint.

3.2

Results

The solution of Program (5) depends on the correlation between YA and YB . A limitation on
P , D P ), can not be expressed in a
mathematical tractability is that firm As equity value, SA (DA
B
P
+
simple closed form, since the term, [YB DB ] is truncated at zero. But, if the correlation between
YA and YB is restricted to 1, then optimality can be characterized in a straightforward, intuitive
fashion. Hence, to illustrate the models main economic intuition, in this Section I restrict the
correlation between the two cash flows to be 1.
I will relax this correlation restriction and show that the results derived in this section are not
changed for the general correlation cases in Section 3.3. The same intuition will be carried over to
the general correlation cases.
When the cash flows from both firm A and B are perfectly correlated, YA = YA +  and YB = YB + 
where  F (). For notational simplicity, I denote c.d.f. of YB by F (.) in this section.
Hence, I can write YA = YB +Q where Q = YA YB . Let the security benefit, C +YA +(YB iB ),
be denoted by Z. Then, from Equation (6), one has
P

P Q) + D P , Y D P }
= Z + (KA + KB ) P rob{YB (1 + B ) (DA
B B
B
B
P Q), Y < D P }
+KA P rob{YB (DA
B
B

= Z + (KA + KB ) P rob{YB M ax

P Q)+ D P
(DA
B B
P
, DB
1+B

o
}

(7)

P Q) Y , Y < D P }
+KA P rob{(DA
B
B
B

which can be summarized as

P =

P )]
Z + (KA + KB ) [1 F (DG

0P D P
if DA
B

0P ) K F (D P ) if D 0P < D P
Z + (KA + KB ) KA F (DA
B
B
A
B

18

(8)

where

0P = D P Q
DA
A
P =
DG

0P + D P
DA
B B
1+B

P is an effective group debt amount under the pyramid, lying between D 0P and D P by definition.
DG
A
B
0P D P , then the familys control default on the entire business group is synchronized at the
If DA
B
P , which is strictly less than D
P , the amount
central firms default, but the event is defined at DG
B
of debt issued from firm B in the baseline pyramid case.
P <D
P , consider the following two cases, where two projects are identical (i.e.,
To illustrate why DG
B
P =D
P , DP = 0), which is the
Q = 0 YA = YB = Y and FA (.) = FB (.) = F (.)). One case is (DA
B
B
P

case where the family issues DB amount of debt only from firm A and zero debt from firm B, and
P = 0, D P = D
P ). The second case is the solution of the baseline model.
the other is (DA
B
B

The new group budget constraint is compared to the baseline one as follows:
P ]+
C + Y + Y E F [(1 + B )Y D
B


  

 
P + + Y B E F Y D
P +
C + Y E F Y (1 B )D
B
B
P ]+
> C + Y B E F [Y D
B
Since it is known that the value of the sum of two call options is always greater than the value of a
call-on-call option, the inequality in the second line of the above algebra holds for any correlation
value between -1 and +1. The above comparison tells us that if the family issues debt only from
P , satisfies D P < D
P . And this implies that
firm A, then the amount of debt issued by the firm, DA
A
B
P
P = DA is far smaller than D
P . Therefore, the family can
the effective group debt amount, DG
B
1+B
dramatically reduce the probability of control default on the entire business group by shifting debt
from firm B to firm A.
As a first step to solve the optimality of recapitalization, the following lemma summarizes when
0P D P or D 0P < D P is an optimal recapitalization of the pyramid.
either DA
B
A
B
Lemma 6. When the cash flows of firm A and B are perfectly correlated, in an interior equilibrium
0P , D P > 0),
(i.e.,DA
B
0P and D P satisfy
1. The optimal debt amounts DA
B

0P
P

DA DB if KA

KB
B

D0P < DP if K >

A
A
B

KB
B

19

and are characterized as follows:

0P D P D P
B

DA
if KA K

G
B
B

P ]+ = C+YA +YB iB

where E F [YB DG

1+B

0P < D P
B
if KA > K
DA
B
B

P 0 ) = KB H(D P ) and

where KA H(DA

B
B

F
0P ]+ + E F [Y D 0P ]+ = C + Y
A + YB iB
E [YB DA
B
B
B
2. The optimal payoff to the family, P , satisfies

KB
P
P

if KA B , = Z + (KA + KB ) [1 F (DG )]

if K >
A

KB
B ,

= Z + (KA + KB )

0P )
KA F (DA

(9)

(10)

0P )
KB F (DB

Proof. See Appendix B.8.

0

P ), is the cost-to-benefit ratio of firm As debt, which summaThe hazard rate function, KA H(DA
rizes the marginal cost of reducing the control benefit by issuing debt from firm A to the marginal
benefit of financing by issuing debt from firm A. The same intuition holds for the scaled hazard rate
P
B
function, K
B H(DB ), which is the cost-to-benefit ratio of firm Bs debt. Lemma 6 says that in an
interior equilibrium, the cost-to-benefit ratios of the optimal debt amounts for both firms should
0P < D P . Otherwise,
B
be equal. Therefore, if K
B
B < KA , the IHR property of H(.) implies DA
0P D P is optimal.
DA
B
0P D P is more
Proposition 2.
In equilibrium, when a fully leveraged pyramid emerges, DA
B
P

0P

likely to be observed than DA < DB .

Therefore, the leverage ratio for central firm A, measured as a ratio of the book value of the
total debt to the book value of stand-alone assets, tends to be greater than that of subsidiary
B in equilibrium when the project profiles of the two firms are identical.
Proof. See Appendix B.9.
P )],
Given that the optimal payoff under the pyramid in the baseline case is Z + KA + KB [1 F (D
B
there is a trade-off
in the recapitalization
between 1) the extra gain in the expected control benefit of


P ) F (DP ) , and 2) the extra loss in firm As expected control benefit, KA F (DP ).
firm B, KB F (D
B
G
G
Therefore, if the extra gain in firm Bs expected control benefit exceeds firm As extra loss, the
family has an incentive to recapitalize the pyramid. When KA is much greater than KB , the
recapitalization gain is likely to be dominated by the recapitalization loss, and the optimal solution
degenerates to that of the baseline case. This implies that if a fully leveraged pyramid is observed
0P D P is more likely.
in equilibrium, DA
B
20

Proposition 3. A fully leveraged pyramid with a decreasing leverage ratio from the top to the
bottom is favored over the horizontal structure in equilibrium if


H ) F (DP ) (1 )(YB iB ) + KA F (DP ) (11)
iB C + (1 B )YB and KB F (D
B
G
G
Proof. See Appendix B.10.
In equilibrium, the fully leveraged pyramidal structure is chosen over the horizontal structure only
if recapitalization of the pyramid enhances relative control advantage. This condition is more likely
to be met, for a given iB , KB = K(iB ), C, and , if 1) the lower KA = K(iA ) is and 2) the smaller
P is. The first condition implies that the fully leveraged pyramid emerges in equilibrium only
DG
if firm A delivers a smaller private control benefit. Therefore, the investment of firm A at date 0
tends to be smaller when one observes the fully leveraged pyramid at date 1 in equilibrium.
The second condition implies that a fully leveraged pyramid is optimal only if the family successfully
minimizes the probability of group default.17 This intuition is consistent with the empirical finding
in the conglomerate literature (Nakatani, 1984, Prowse, 1992, and Ferris et al., 2003). These
studies argue that cross-entity subsidization within a business group, which is motivated from the
profit stability tendency, makes the business groups possess greater debt capacity. Consistently
with these studies, I propose that the structural properties of the pyramidcontrol optionality and
co-insurance benefitgenerate extended debt financing capacity of a pyramidal business group.
The following lemma summarizes the condition where the family can successfully minimize the
effective group debt amount.
Corollary 2. A fully leveraged pyramid is more likely to be observed in equilibrium if central firm
A is large. Since the central firms investment tends to be small, this condition implies that the
familys managerial ability for central firm A is better than that for subsidiary B under the fully
leveraged pyramid.
Proof. See Appendix B.11.
For the family to favor the fully leveraged pyramid over the horizontal structure, it needs to
P . For a given project
effectively minimize the probability of groupwise control default, and thus DG
profile of firm B, this is more likely when the existing central firm is profitable (i.e., YA is large and
iA is small). Hence, a fully leveraged pyramid is more likely to emerge when the familys managerial
ability in firm A is proficient.

3.3

Robustness of Results: General Correlations

This section numerically analyzes the optimal leverage ratios for general correlation cases. The
numerical results in this section suggest that a general correlation value does not change the main
P
A small value of DG
does not imply that the leverage ratio of a pyramidal business group should be lower than
P
that of a horizontal business group. Central firm As debt amount, DA
, under a fully leveraged pyramid can exceed
H

DB , which is the optimal debt amount of firm B under the horizontal structure.
17

21

results in Section 3.2 for a reasonable parameter space, which, in turn, implies that the correlation
effects on the main results in the previous section are not first order.
I now provide a numerical example with bivariate normally distributed cash flows for the two firms.
To highlight the effect of correlation on the optimal leverage ratios, I assume that the two projects
are identical in mean and standard deviation of cash flow and investment size. I vary the correlation
from -0.9 to 0.9.18 A bivariate normal distribution for Yj for j = A, B implies
(YA , YB )T N [YA , YB ]T ,
where =

2
A
A B
2
A B B

Since I assume that the project profiles are identical across the two projects, YA = YB = Y and
A = B = . The main numerical objects are defined in Appendix A.
The parameters used in the numerical analysis are given in Table 1.
Table 1: Parameters
Variable
Y

iB
KA
KB

B
C

Parameter Value
102
30.6
100
30.6
30.6
0.25
0.4
5

The time period between date 1 and date 2 is set to be 1 year. Hence, in Table 1, projects have
a return of 2% per annum, and risks are 30% per annum of the cash flows. The control benefit
amount for both firms is 30% of the firm value for both projects. This value of private control
benefit is roughly accurate for countries such as Mexico, Italy, and Argentina, according to Dyck
and Zingales, 2004. I set B = 0.4, = 0.25, and C = 5 to ensure an interior solution, which can
be relaxed.
With these parameter values, the optimization program (5) is solved for a general correlation,
[0.9, +0.9], by increasing the correlation by 0.3 for each optimization. The results are given
in Appendix A.
First, to illustrate the extent to which the required debt amount is sensitive to a correlation, I set
P = 0 and compute D P that satisfies the new group budget constraint. Hence, this is the upper
DB
A
bound of firm As debt amount that the family needs to issue to recapitalize firm A. Panel (a) of
P (labeled
Figure 1 shows the results. The panel tells us that the maximal required debt amount, DA
18

The two extreme correlation values, -1 and 1, are excluded in this analysis due to singularities.

22

as Recap A. Max), does not vary greatly across correlations. This implies that the correlation does
P is smallest, with a
not affect the debt amount at first order. When the correlation is -0.9, DA
value of 33.80, and gradually increases as the correlation increases to +0.9. When the correlation
P has a maximal value of 33.87. Therefore, from = 1 to = 1, the required debt
is +0.9, DA
amount of firm A is increased minimally by just 0.07. I also report the required debt amount
P , labeled
issued by firm B without recapitalizing firm A (i.e., the solution of the baseline case, D
B
as No-recap. B) to demonstrate how the family can minimize the required amount of debt through
P is much smaller than D
P and DP is almost 36.5% of D
P.
the recapitalization. In the panel, DA
B
A
B
Hence, one can see that recapitalization is financially very advantageous.
P = 0 and solve the optimal recapitalization program. Panel (b) of
Now, I relax the restriction DB
Figure 1 shows the results. The leverage ratio in this panel is defined as the amount of debt of each
firm divided by Y , the value of each firms own operating assets. The leverage ratios of the two
firmslabeled as Recap. A and Recap. Bdo not vary significantly across different correlations.
As the correlation varies from -0.9 to +0.9, the leverage ratio of firm A increases from 29.14 to
29.20%, and firm Bs increases from 11.41 to 11.45%. The ratio of firm As leverage to firm Bs is
1
= 2.5 in this numerical
roughly 2.5010 on average. This is close to KKABB , which has a value of 0.4
analysis. This implies that the optimal debt amounts for firms A and B are determined mostly by
KB
KA B , consistent with the results in Lemma 6, where I assumed perfect correlation. This suggests
that the intuition in the previous section carries over to the general correlation cases.

Panel (c) of Figure 1 shows how optimal payoff varies with correlations and how much the family
gains from the recapitalization. Recap. refers to the payoff with the recapitalization, and No-recap.
refers to the one of the baseline case. The results show that the family receives a greater expected
payoff with the recapitalization. Moreover, the expected payoff tends to increase as the correlation
value decreases. This is because the family can reduce the group default probability through
diversification due to low correlation. The expected payoff decreases from 61.1517 to 60.8657 as
the correlation increases from -0.9 to +0.9. But this diversification effect on the expected payoff to
the family is not at first order, either. The effect of recapitalization is clearly at the first-order, and
the difference between the two payoffs, with and without recapitalization, is 11.4623 on average,
which is equivalent to 23.11% of the expected payoff that the family receives in the baseline case.
This payoff enhancement is substantial.

Discussion: Why Not Conglomerate?

So far I assume that the family should structure the new group member firm B by either establishing
it as a pyramidal subsidiary or a stand-alone entity under the horizontal structure. But, clearly,
the family can take the new project as a new business unit in firm A (i.e., conglomerate structure
hereafter) and this conglomerate structure may be advantageous over both the pyramid and the
horizontal structures.
It is important to show that under certain circumstances, the pyramidal structure strictly dominates
the conglomerate structure to justify the existence of the pyramid, given that the family can also
choose the conglomerate structure. To this end, I compare the conglomerate structure with the
pyramidal business group structure and examine when the pyramidal structure dominates the

23

conglomerate structure in equilibrium. The end of this section will provide a short discussion on
the comparison between the conglomerate and horizontal structures.

4.1

The Date-1 Program of the Family under the Conglomerate Structure

Suppose that at date 1, the family adopts the new project, (YB , iB , KB ), as a new division of firm A.
The familys ownership in firm A, , and a free cash flow of C in the firm at date 1 are given. Since
the dividends policy does not affect the familys monetary part of the payoff, WLOG, I assume that
the family does not pay out any dividends and the maximum internal slack, C, is used. Hence, the
internal financing amount under the conglomerate, tC , satisfies 0 tC C. In addition to this
internal slack, the family issues a fraction (1 0 ) of new shares and raises zero-coupon debt with
C . After the new share issuance, the familys ownership in the conglomerate is
a face value of DA
diluted to 0 .
Suppose that the familys managerial ability in the combined businesses in the conglomerate, pC ,
which can be different from pA and pB . In this case, due to the takeover
threat
on the conglomerate,


(pC )
0
the family has an equity financing constraint that satisfies M in
, 1 1. For notational


simplicity, I denote M in (pC ) , 1 by V (pC , ) and express the new variable by VC hereafter.
Let the total capital raised by the internal cash, external debt, and equities be denoted by MBC .
C ) + B (D C ), where S (D C ) and B (D C ) are the date 1
This satisfies MBC = tC + (1 0 )SA (DA
A
A
A
A
A
A
values of firm As equity and debt, respectively. Let the stochastic total cash flow in firm A at date
C , where X C = Y + M C i + Y .
2 before any debt repayment be denoted by XA
A
B
B
A
B
With these assumptions, the date 2 stochastic payoff to the family under the conglomerate structure,
C , is



C
C +
C = C tC + 0 [XA

DA
] + (KA + KB )1{X C DC }
A

Breakeven and tight financing conditions imply that the date 1 expected payoff to the family under
C ], is
this structure, C = E FA,B [
C
C = C + YA + (YB iB ) + (KA + KB )[1 FA,B (YA + YB < DA
)]

The familys date 1 program under the conglomerate structure is

C )]
M ax(tC ,0 ,DC ) C = Z + (KA + KB )[1 FA,B (YA + YB < DA
A

s.t.

VC 0 1
(12)
tC

C) = i
+ YA + YB 0 SA (DA
B

0 tC C

24

where

Z = C + YA + (YB iB )
C ) = E FA,B [Y + Y D C ]+
SA (DA
A
B
A

The first constraint is an equity financing constraint, and the second is a budget constraint. The
third is a cash financing constraint.
This program resembles the pyramidal structure problem but differs chiefly in its equity financing
constraint and equity value. The lower limit of equity ownership under the conglomerate, VC , is
higher than that under the pyramid, B , if the managerial ability of the family over the conglomerate, pC , stays the same as that over the single new project, pB (i.e., pC = pB ). This is because
the family needs to secure control over the conglomerate solely by using its own equity stake. HowC ), is greater than S (D P , D P )
ever, the value of equity under the conglomerate structure, SA (DA
A
A
B
in Equation (6).

4.2

The Optimal Solution of the Conglomerate Structure Program

Lemma 7. The optimal financing decision for the conglomerate is as follows:

0
C
(tC
A, , DA )

(t, w, 0)

where
=

C)

(C, VC , D

where

Proof.

C =i ,tC =C)
C (MA
B
C
DA

if iB C + (1 VC )(YA + YB )
t [0, C], w [VC , 1]
if iB > C + (1 VC )(YA + YB )
C ]+ = C+YA +YB iB
E FA,B [YA + YB DA
VC

C > 0, and zero otherwise.

< 0 for DA

With this financing decision, the optimal date 1 expected payoff to the family under the conglomerate structure is

if iB C + (1 VC )(YA + YB )
Z + KA + KB
C
=
(13)

Z + (KA + KB )[1 FA,B (YA + YB < DA )] if iB > C + (1 VC )(YA + YB )

4.3

Discussion: Pyramids Versus Conglomerates

Suppose that the familys control over the existing firm A is tightly secured at (i.e., = (pA ))
and the family cannot improve its managerial ability by combining the two businesses, YA and YB .
In this case, pC pA holds. This implies (pC ) = (pA ), and thus the family cannot issue any
new shares from the conglomerate (i.e., VC = 1), since (pC ) 1.
Proposition 4. A pyramid strictly dominates a conglomerate structure for the new project, (YB , iB , KB )
with iB > C, if there is no managerial synergy for the family to run both projects, (YA , iA , KA ) and
(YB , iB , KB ), under the conglomerate structure and the two projects are perfectly correlated.
25

Proof. See Appendix B.12.

To finance the new project, the pyramidal structure requires a smaller amount of debt than the
conglomerate. This can be seen in the following scenario: Under the pyramid, the family can sell
an additional fraction (1 B ) of firm Bs shares to finance the new project, whereas under the
conglomerate structure, it cannot sell any shares without managerial synergy. This extra equity
financing capacity under the pyramid is obtained through indirect control of the subsidiary. Unless
the family loses control over central firm A, it can sell a fraction (1 B ) of shares of firm B.
However, under the conglomerate structure, the family needs to secure control over both projects by
using its own equity stake and thus faces additional limitations in equity financing capacity. Therefore, through effectively reducing the required debt amount by selling equities of the subsidiary, the
pyramid has a relative control advantage over the conglomerate. Since both structures are limited
to receive rent from the new project, the relative control advantage under the pyramid directly
leads to the pyramids optimality over the conglomerate in equilibrium. Unless the new investment
is small and supportable by internal capital, C, the pyramidal structure is strictly preferred over
the conglomerate if the family cannot improve its managerial ability on the conglomerate through
synergy. This implies that the pyramid dominates the conglomerate for a capital intensive project.
If the family is better able to run the combined businesses, YA and YB , than the existing business in
firm A, YA , then the preference of the pyramid over the conglomerate structure becomes conditional
on VC . In this case, managerial synergy under the conglomerate structure implies (pC ) < , and
thus VC < 1. Then, the family can sell (1 VC ) more shares of the conglomerate. Since the value
of the unleveraged equity on the combined businesses, YA + YB , is greater than that on the single
business, YB , the family may be able to raise enough capital by selling the larger-valued equities
of the conglomerate (i.e., using the large size of a firm to maximize control, hereafter called the
size effect). Hence, if the maximal amount of equity financing under the conglomerate without
debt, (1 VC )(YA + YB ), is greater than that under the unleveraged pyramid, (1 B )YB , then
the conglomerate is favored over the pyramid.
But if the investment outlay, iB , is sufficiently large that the family needs to issue debt under
the conglomerate, then the analysis becomes complicated and the optimality of the conglomerate
structure over the pyramid depends on whether the diversification benefit of the combined cash
flows under the conglomerate is greater than that under the pyramid. Here, the magnitude of
the conglomerates diversification benefit depends on the size of the investment and the level of
correlation. I do not discuss this case in detail in this paper.
Based on the discussion so far, it is known that the pyramidal structure is strictly preferred over
the conglomerate if the family cannot improve its managerial ability through synergy, and the new
project is highly correlated to the project in the existing central firm. The first condition of no
managerial synergy implies that a pyramidal structure is more likely to emerge in equilibrium when
the familys ownership in the existing firm A is significantly diluted such that the family cannot
raise capital through any new share offerings. Therefore, the family has a strong incentive to adopt
the pyramidal structure when the size of the existing firm A becomes sufficiently large, and the size
effect of the firm is not effective in deterring a takeover threat. This suggests that the central firm
under the pyramid tends to be large, and this intuition is consistent with the result in Corollary 2.
The second condition for the correlation between the central firm and the new subsidiary opens
26

a discussion on whether the firms in the pyramidal ownership chain should operate in similar
industries or not. Jang et al. (2005) find that Korean business groups tend to pursue higher
operational efficiency through a vertical integration of their business lines where central firms
depend on subsidiaries in a related business for their materials and intermediary goods. However,
to generalize this condition, we need more international evidence.

4.4

Discussion: Horizontal Spin-Offs Versus Conglomerates

The previous section reveals that under a conglomerate structure, the family needs to share rent
from the new project with the existing firms old shareholders. This sharing rule of rent implies that
the family does not maximize profitability under the conglomerate, in contrast to the horizontal
structure, under which it maximizes profitability. Therefore, a stand-alone entity directly controlled
by the family under the horizontal structure should be more profitable than the conglomerate.
Since the pyramidal structure emerges when the family cannot improve its managerial skill by
combining multiple businesses, the central firm under the pyramid resembles a conglomerate when
the pyramidal structure is observed.
The expansion of a family business group can be summarized as follows: The family expands the
existing firm under a conglomerate structure until the firm becomes large and profitable and then
it starts expanding the business group 1) by spinning off a stand-alone entity, if the scale of the new
project is small and the project is very profitable, or 2) by forming a pyramid, if the new project
is capital intensive and less profitable.

5
5.1

Empirical Implications and Results

Testing Hypotheses

The model predicts the following testable hypotheses.

H1 The more capital intensive, with a larger private control benefit, and less profitable a firm is,
the more likely it is to be a pyramidal subsidiary in equilibrium. The subsidiary tends to
be less leveraged than a stand-alone entity directly controlled by the family and tends to be
younger than a central firm.
The more capital intensive a firm is, the more severe the financial constraint the family faces
when it chooses the horizontal group structure. A greater amount of internal capital under
the pyramidal structure enables the family to minimize reliance on external debt financing
and thus the family can receive a higher expected control benefit. This relative advantage
in the expected control benefit under the pyramid is optimal when the extra internal capital
carries a low cost, which is implied by low rent from the subsidiary. Therefore, when the
project delivers a large private control benefit and is less profitable, the pyramidal structure
emerges as the optimal business group structure in equilibrium and the subsidiary should be
less leveraged. The pyramidal subsidiary tends to be younger than the central firm since it
can emerge only if there is an established central firm in the business group.

27

H2 The less capital intensive, with a smaller private control benefit, and more profitable a firm is,
the more likely it is to become a central firm under the pyramid in equilibrium. The central
firm tends to be more leveraged, older, and larger than the subsidiary.
The central firm can extend its debt financing capacity by combining its own operating cash
flow with the non-operating cash flowthe dividends paid by its subsidiaries. To utilize
the co-insurance benefit of the dividends, the family optimally leverages up the central firm.
By doing so, the family can effectively minimize the groupwise default probability while
extending the central firms debt capacity. The default probability of the central firm is
effectively minimized if the firm operates on debt-favored projects. The central firms cost of
debt financing is low when the firms private control benefit is low. This condition is more
likely to be satisfied when the firm is less capital intensive, more profitable, and larger than
the subsidiary.
H3 Central firms tend to pay out less to ensure the greater availability of group internal capital
to support capital-intensive projects taken on by their subsidiaries.
When the pyramidal structure is chosen, the choice is made to minimize the amount of
external debt. To ensure the greater availability of group internal capital, the family decides
not to pay out any dividends from the central firm and allocates the internal capital to group
firms. Thus, a firm with many subsidiaries is more likely to keep its retained earnings as an
internal financing resource and work as the headquarters of an internal capital market.
H4 Pyramidal subsidiaries tend to pay out more to support the debt of highly leveraged central
firms.
The family pushes the subsidiaries to pay out as much as possible to service the central
firms debt. Since the controlling shares in the subsidiaries cannot be monetized due to nonmarketability, the only way to utilize this position is to receive dividends from subsidiaries.
Thus, the optimal payout policy for subsidiaries is to fully pay out their earnings whenever
the dividends can help rescue the central firm from default.

5.2

Data and Variable Descriptions

The data set I use is from Almeida et al., 2009. The data set consists of the ownership and financial
information for all nonfinancial group firms in Korean family business groups, chaebol s, between
1998 and 2004. The data do not include any chaebol s controlled by the Korean government and
thus consist entirely of family-controlled Korean business groups. In 1981 the Korean government
established the Korean Fair Trade Commission (KFTC) to regulate market competition. Each year
the KFTC defines a list of chaebol s and collects complete ownership data for all group firms for
each chaebol as part of its regulatory purpose.
The ownership information is detailed so as to show what fraction of shares is owned by the family
owner, the owners relatives, and group affiliates. I define the fraction of common shares owned
by the family owner and any relatives as the ownership of the controlling family. The KFTCs
definition of family relative is broad and includes large shareholders who might cooperate with the
owner family to exercise control. Therefore, the definition of the controlling familys ownership used
here can be interpreted as the maximal stake of control shares that the owner family can utilize
28

when exercising control. Moreover, when I compute the ownership fraction of the controlling family,
I adjust for outstanding treasury stocks. Ownership information is collected as of April 30th each
year and mapped to financial information at the end of the fiscal year that is closest and prior to
the reporting date of the ownership information.
The KFTC also collects these firms financial information, making it publicly available upon request.
The KFTC financial information is augmented by the financial data from two other databases
collected by 1) the Korea Listed Companies Association and 2) the Korea Investors Service. These
two databases contain information for both listed public firms and private firms in each chaebol
that are subject to external audit. Hence, the data used in this paper provide a comprehensive
dynamic panel of Korean chaebol, including equity ownership structure and financial information.
The data period, 19982004, is ideal for testing the models predictions. The Korean economy
experienced the 1997 Asian financial crisis and, during this studys data period, many business
groups underwent active restructuring. Because of this restructuring across all chaebol s, massive
mergers and acquisitions were initiated by both domestic competitors, comprising other, competing
chaebol s, and foreign investors, such as private equities and leveraged buyouts, making this a period
when chaebol s were concerned with retaining control over group firms.
One difficulty with the empirical analysis results from the complex ownership structure of Korean
chaebol. Almost every chaebol uses a circular ownership chain on top of a vertical pyramid. Complex
ownership links between group firms make it difficult to define whether any given firm in a chaebol
is a pyramidal subsidiary, a group central firm, or a stand-alone firm, since each group firm belongs
to multiple pyramidal ownership chains. To identify a firms type, I use the group metrics developed
by Almeida et al. (2009), who use the dividend algorithm to compute the position and centrality of
a group firm in each chaebol. The authors define three main group structure metrics: 1) P osition,
2) Centrality, and 3) Loop. The definitions of the three metrics are briefly discussed below and
the mathematical definitions are given in Appendix C.
5.2.1

Group Structure Metrics

Position
One of the main group structure metrics is P ositioni,t , which measures the distance between the
family owner and a given firm i in the group at year t. P ositioni,t is defined as the cash-flow
weighted distances of a given firm i from the controlling family. Its mathematical definition is given
in Definition 2 in Appendix C.
A simple example is useful in explaining how this metric works. When a business group consists
of two firms and the two firms belong to a single pyramidal ownership chain, then the central firm
at the top of the pyramid is one distance away from the family and the subsidiary is two distances
away from the family. Since the two firms belong to a single vertical ownership chain, the top
firm has a P ositioni,t value of 1 and the bottom firm has a P ositioni,t value of 2. In this simple
example, the weight assigned to the single pyramidal ownership chain is 1. Therefore, in this case
the distance of a given firm i at year t from a controlling family is equivalent to P ositioni,t .
To illustrate how P ositioni,t is defined for a more complex case, suppose that a given subsidiary
29

i belongs to multiple pyramidal ownership chains. In this case, the distance of a given firm i
from the family depends upon which pyramidal ownership chain one takes into account to measure
the distance of the firm. To define a firms position in such multipyramidal chains, Almeida et
al. (2009) propose a method to quantify the weight assigned to each pyramidal chain. They first
compute the dividends the family can receive from firm i in each of the pyramidal ownership chains
if the firm pays a dollar dividend. To compute the amount of dividends the family receives from
firm i, the authors assume that any intermediate firms between the family and firm i pay out the
full dividends they receive from firm i. Because of the possible circular ownership chain, a given
firm can belong to a number of pyramidal ownership chains.
To deal with ownership chain circularity, Almeida et al. (2009) introduce two ownership matrices:
1) the intercorporate holding matrix (A in Appendix C) and 2) the familys direct equity stake
in each group firm (f in Appendix C). Using these two ownership matrices, the authors compute
the familys cash flow from firm i for each possible pyramidal ownership chain to which the firm
belongs, and weight-average the distances of firm i using the cash flows.
Centrality
The other group structure metric is Centralityi,t , which measures how important a given firm i is
in ensuring control over an entire groups firms at year t. A high value of Centralityi,t implies that
firm i at year t is central in control to other member firms in the same business group, which, in
turn, implies that the firm has a number of subsidiaries. To define this group metric, Almeida et
al. (2009) define a variable named Critical control threshold (CCi,t hereafter) for firm i at year t.
By its name, CCi,t is a critical value of the control threshold.
The definition of a control threshold, T , is necessary to define CCi,t . Here T determines whether or
not a given firm is under the familys control, which is determined if the family has a voting share
fraction greater than T , either directly or indirectly, in a given firm. This definition of the familys
control over a firm with T is described in Assumption 2 in Appendix C. Hence, for a given value
of T , a set of firms is defined as being under the familys control (see Definition 3 in Appendix C).
As one increases the control threshold value, T , the more parsimoniously the set of firms under
the familys control are defined. This implies that there is a maximal value of T for each group
firm such that for a given value of T , firm i is treated as being under the familys control. Hence,
the maximum value of T defines the familys control over firm i, and thus CCi,t is set to be the
maximum value of T . By its definition, CCi,t can be interpreted as an upper bound of the minimum
required fraction of shares that qualify firm i as being under the familys control at year t.
Once CCi,t for firm i at year t is defined for all group firms (i.e., i = 1, 2, ..., N ), a centrality of the
firm, Centralityi,t , is defined based upon CCi,t i. Almeida et al. (2009) define Centralityi,t as
the average decrease in CCi,t across all firms at year t if a given firm i is excluded from the group.
For example, consider a business group structured with two firms in a two-step, single pyramidal
ownership chain. Suppose the family owns 40% of the top firms shares and the top firm owns 50%
of the bottom firm. With this structure, both firms are defined as being under the familys control
with any T 40%, since the familys control over the top firm is established with T 40% and,
unless the familys control over the top firm is lost, the family has indirect votes of 50% in the
bottom firm through control over the top firm. For T > 40%, neither of the two firms are treated
30

as being under the familys control, since the familys control over the top firm is not defined with
T > 40%, and thus the family has no voting shares in the bottom firm. According to the definition
of CC, both firms have a CC of 40% in this case.
If one now excludes the top firm from the group, the familys control over the bottom firm is only
established for a T of zero, since the family will no longer have an ownership link to the bottom
firm after it loses its ownership link to the top firm. Hence, CCi,t for the bottom firm, a subsidiary,
drops to zero from 40% when one excludes the top firm. From Definition (5) in Appendix C, in
this example the value of 40% is assigned to the Centrality of the top firm.
If one now supposes that the bottom firm is excluded from the business group. Since the familys
total voting fraction in the top firm does not change, even after eliminating the bottom firm, CCi,t
for the top firm stays at the same level of 40%, which, in turn, implies that the Centrality of the
subsidiary is zero.
Loop
The third group structure metric is Loopi,t . The idea defining this metric is that if a firm belongs
to a circular ownership chain, then any dollar dividend paid by the firm will eventually be received
by itself in a finite number of steps. For example, suppose that firm A owns firm B, which owns
firm C, which owns firm A. Then, any dividend paid by firm A will be received by itself in three
steps, and the firm has a Loop value of 1.
5.2.2

Two Main Dummy Variables, Pyramid and Central : Identifying the Three
Types of Firms in a Business Group

Table 2 summarizes the statistics of the group structure metrics and several characteristics of the
firms in the data. The data consist of 3442 firmyear observations, with 1072 firms for 47 chaebol
groups for seven years, from 1998 to 2004. The average P osition value of firms in the data is
2.11 and its standard deviation is 0.816. These statistics of P osition imply that the depth of the
pyramidal structure for each chaebol is shallow.
The average Centrality value is 0.150, with a standard deviation of 0.049. Using the average value
of Centralityi,t , 0.015, I define a dummy variable, Centrali,t , that denotes whether or not a given
firm i at year t is a central firm. The definition of Centrali,t is

Centrali,t =

1 if Centralityi,t 0.015
0 if Centralityi,t < 0.015

The dummy variable Centrali,t will be used as a dependent variable in testing the two empirical implications, 1) H2 and 2) H3. The main reasons for using this dummy variable as a main regressand
in the regression analysis, instead of the continuous variable, Centralityi,t , are 1) Centralityi,t is
too right skewed and 2) more than 50% of Centralityi,t takes on the value of zero. This extremely
right-skewed variable might underestimate the correlation between leverage ratio and Centralityi,t .
Cautious about this possibility, I split the firms in the sample into two groups, using the average
value of Centrality.
31

The average P osition for central firms (i.e., Centrali,t = 1) is 1.610, with a standard deviation of
0.600. The 25th and 75th percentiles of P osition for central firms are 1.10 and 2.01, respectively.
This implies that roughly 75% of the central firms in the sample are positioned more than 1.10
distance away from the family and half of them have a position value between 1.10 and 2.01.
Moreover, the average P osition for firms belonging to a circular ownership chain (i.e., Loop = 1)
is 1.859, with a standard deviation of 0.764. The median P osition of firms belonging to a circular
ownership chain is 1.710. Hence, roughly 50% of central firms positioned between 1.10 and 2.01
distances away from the family owner tend to belong to a circular ownership link.
Table 2 summarizes the statistics of Top Firms Centrality. The variable Top Firms is defined
as firms whose P osition is less than 1.1. The median Centrality of Top Firms is 0.002 and the
75th percentile is 0.040. These statistics suggest that more than 50% of Top Firms are not central
firms, and thus more than half of top firms are stand-alone entities directly controlled by the family.
These firms are least likely to belong to a circular ownership chain.
With the conditional statistics, I define another dummy variable, P yramidi,t , identifying whether
a given firm i is a pyramidal subsidiary or a stand-alone group entity. The definition of the dummy
variable is

P yramidi,t =

1 if P ositioni,t 2.0
0 if P ositioni,t 1.1

This variable will be used to test H1 and H4. Here P yramidi,t = 1 implies that the firm is
located more than two distances away from the family and is a subsidiary. In addition, firms with
P yramidi,t = 0 are stand-alone firms directly controlled by the family.
Less than 50% of firms with P yramidi,t = 0 may be also central firms, but I do not exclude them
from the testing sample to make sure that subsidiaries do not greatly outnumber stand-alone firms.
Furthermore, the theory suggests that both central firms and stand-alone entities should be more
leveraged and pay out less than subsidiaries. Therefore, the results of testing H1 and H4 using
this dummy variable are consistent with theoretical predictions.
Table 2 reports the statistics of three additional variables: Public, Employees, and FirmAge. The
variable Public is a dummy that takes on a value of 1 if the firm is publicly listed, and zero otherwise.
Based on the mean value of Public, 26% of the firms in the data are public, accounting for 279
firms. Employees is the number of employees in a firm, and Firm Age is measured in years since the
year of foundation. The average firm in the sample has 1204 employees and has been in business
for 17 years.
5.2.3

Financial Variables

Since the theoretical predictions for leverage ratio are derived for the ratios based on each firms
own operating assets, I first quantify the book value of a firms operating assets.
The book value of a firms Total Assets consists of two components: 1) Operating Assets and
2) Non-operating Assets (i.e., the firms control stakes in the other group member firms). The
accounting information in the data allows one to identify each of the two types of assets in a firm.
32

The accounting information in the data is constructed by an accounting method called the equity
method. The equity method is used by firms to assess the profits earned by their investments in
other companies. The firm reports the income earned on the investments on its income statement
based on its share of the company assets. The reported profit is proportional to the size of the
equity investment. This is the standard technique used when one company has significant influence
over another.
Based on this accounting method, there are three important variables that enable me to quantify
the assets and profits from a firms own operation and those from a firms investment in the other
member firms equities: 1) Equity Method Stock, 2) Gains from Equity Method, and 3) Losses from
Equity Method.
Equity Method Stock refers to the aggregate book value of equity stakes held in the other group
member firms, and thus this variable quantifies the book value of Non-operating Assets. The
variable Gains (Losses) from Equity Method refers to how much profit (loss) a given firm makes
from its equity investments in the other group firms. Clearly, Gains (or Losses) from Equity Method
is reported based on their book values. Using these three items, I quantify two financial variables,
Operating Assets and Operating Profits, as
Operating Assets = Total Assets Equity Method Stock
Operating Profits = Total Profits Gains from Equity Method
+Losses from Equity Method
Using these items, I define the book leverage ratio of firm i at year t, LEVi,t , as the ratio of the
book value of the debt to the book value of Operating Assets. The book value of debt is computed
as the sum of the following three variables: Short-term debt, Long-term debt in current liability,
and Long-term debt. Similar to the definition of LEVi,t , I define a measure of capital expenditure,
CAP EXAi,t , as the ratio of the book value of total capital expenditures to Operating Assets. Using
Operating Profits, I define a measure of a firms operating profitability, OPROA, as the ratio of
Operating Profits to Operating Assets. I also define two measures for the dividends payout ratio,
DIV A and DIV U N A, as the ratios of the book value of total dividends paid out to Operating
Assets and to Total Assets, respectively.
The summary statistics of these financial variables and the correlations between them are reported
in Tables 4 and 5, respectively. From Table 4, the average leverage ratio is 20.3% on the basis of
operating assets, with a standard deviation 38.2%. Firms in the data have a 2.50% rate of return
on assets, on average. Capital expenditures are around 5.60% of the book value of operating assets,
on average, and total dividends paid out are 80 basis points(bps) of the book value of operating
assets.
The leverage ratio is negatively correlated with P yramid, which implies that subsidiaries tend to
have lower leverage ratios than stand-alone firms. The leverage ratio is also positively correlated
with Central, which implies that central firms are more leveraged than non-central firms. Capital
expenditures tend to be larger for subsidiaries than stand-alone firms, and smaller for central firms
than non-central firms. Relatively smaller dividends are paid out from central firms than from noncentral firms, whereas pyramidal subsidiaries tend to pay out more than stand-alone entities. The
signs of the correlations reported in Table 5 are consistent with the models theoretical predictions.
33

The last variable I introduce is for industry code. In the regression analysis provided in the
following sections, I control for industry fixed effects. To this end, I use industry classification
codes corresponding roughly to the U.S. two-digit SIC codes. There are 45 different industries in
the sample.

5.3
5.3.1

Main Empirical Results

H1: Position and Leverage

To test hypothesis H1, I estimate the following model:

P yramidi,t = a0 + a1 LEVi,t1 + a2 OP ROAi,t1 + a3 CAP EXAi,t1
+1 Controls + i,t

(14)

Equation (14) is a direct implication of the hypothesis, where as a firm becomes less leveraged,
more capital intensive, and less profitable, the more likely it will emerges as a pyramidal subsidiary.
Hence, one expects to see a1 , a2 < 0, and a3 > 0.
The theory predicts a contemporaneous relation between P yramidi,t and these three RHS variables.
But corporate governance literature suggests that a pyramidal subsidiary is likely to suffer from
low profitability due to a severe agency problem caused by the separation between cash flow and
control rights, given that the pyramidal structure is already in place. Moreover, the entrenched
family is likely to have an empire-building incentive when the agency problem is extreme under
the pyramid. Therefore, a reverse causality problem can arise in Equation (14) when I use the
contemporaneous regressors on the RHS. I try to control for the problem by using lagged variables
as main regressors on the RHS. I also run the same regression with the contemporaneous variables
and report the results. A more detailed discussion on the issue of reverse causality will be provided
in Section 5.4.
I control for F irmAge, P ublic, and Ln(Assets), where Ln(Assets) is defined as the natural log of
the book value of T otal Assets. Since pyramidal subsidiaries tend to be younger, less established,
and smaller, the expected signs of these three control variables are negative. I use F irmAget1
scaled by 1000 in the regression to adjust the magnitude of the variables coefficient. The usual
controls for Year and Industry fixed effects are included in the regression. In addition to these two,
I also control for group fixed effects. Standard errors are clustered at each firm level to correct for
any possible within-firm correlation.
The results are reported in Table 6. The first three columns in Table 6 report the results when
the lagged variables are used for all the regressors on the RHS of Equation (14). Column (4)
shows the results when the contemporaneous regressors are used. The coefficients of the leverage
ratio, LEVi,t1 , in the first three columns are negative, and they are statistically significant at less
than 1% in the first two and at 5% in the third. This result confirms that a less leveraged firm is
more likely to be a pyramidal subsidiary. The negative and significant coefficient of OP ROAi,t1
confirms the theoretical prediction that a less profitable firm is more likely to be a subsidiary. The
coefficient of CAP EXAi,t1 is positive and statistically significant in columns (2) and (3) at less
than 5 and 10%, respectively, implying that the more capital intensive a firm is, the more likely it
becomes a subsidiary.
34

The coefficients of the control variables F irmAge, P ublic, and Ln(Assets) have the right signs and
all of them are strongly significant at less than 1%. This suggests that business groups grow over
time, and a young and less established smaller firm tends to be a pyramidal subsidiary.
Column (4) reports the regression results when I change the variables from lagged to contemporaneous. The significantly negative coefficient of LEV confirms that the leverage ratio of a firm
plays an important role in determining the firms location in the business group.
In column (5), the dummy left-hand side (LHS) variable, P yramidi,t , is replaced with a continuous
counterpart, P ositioni,t . One might argue that the results in the first four columns are possibly
driven by the generic definition of the dummy variable, P yramidi,t . To show the robustness of the
results, I run the regression with P ositioni,t as a dependent variable.
The results in column (5) show that LEVi,t1 and F irmAget1 are still significant. The signs of
the coefficients of these two variables are consistent with the results in the other four columns, but
the statistical significance of the coefficient of LEVi,t1 is reduced to the 5% level from the 1%
level.
However, this is expected, since I also include firms with P ositioni,t between 1.1 and 2.0, and these
are the intermediate central firms that tend to belong to circular ownership links. These firms tend
to have mixed characteristics, between those of a central firm and those of a subsidiary. If these
firms act like central firms, they tend to have a higher leverage ratio, which possibly reduces the
explanatory power of LEVi,t1 . However, the negative and significant coefficient of LEVi,t1 , even
with the new continuous dependent variable, suggests that the results in the first four columns of
Table 6 are not likely to be driven by the definition of P yramidi,t .
Column (6) runs the same regression as in column (5), but only with firms that do not belong to
any circular ownership links (i.e., Loop = 1). In this case, the coefficient of LEVi,t1 is significant
at less than 1%. Therefore, the mixed firm characteristics of an intermediate central firm belonging
to a circular ownership link are confirmed.
5.3.2

H2: Centrality and Leverage

This section tests the hypothesis H2 by estimating the following model:

Centrali,t = b0 + b1 LEVi,t1 + b2 OP ROAi,t1 + b3 CAP EXAi,t1
+2 Controls + i,t

(15)

The test specification is almost the same as in Equation(14) in Section 5.3.1 and the same controls
as in Equation(14) are included in the regression. Similar to the procedure in the previous section,
F irmAge, P ublic, and Ln(Assets) are controlled for. Also Year, Industry, and group fixed effects
are controlled for in the regression. The standard errors are clustered at each firm level to correct
for within-firm correlations.
The results are reported in Table 7. From the theory, I expect the sign of the coefficient of LEVi,t1 ,
b1 , to be positive. The expected sign of b3 , the coefficient of CAP EXAi,t1 , is negative, whereas
the sign of b2 , the coefficient of OP ROAi,t1 , is positive.
The first three columns show the results when I regress the LHS variable, Centrali,t , on the lagged
regressors. In all three columns, b1 turns out to be positive and is strongly significant at less than
35

1% level in the first two and 5% level in the third. Hence, a higher leverage ratio for a central firm is
confirmed empirically. This result, together with the result from the previous Section 5.3.1, implies
that the leverage ratio from the top to the bottom of the pyramid in Korean chaebol tends to be
decreasing. Moreover, this decreasing leverage ratio from the top to the bottom of the pyramid is
found in Italian family business groups (Bianco and Nicodano, 2006).
The coefficient of OP ROAi,t1 , b2 , is positive in the first three columns, although it is not statistically significant. The sign confirms that a firm is more likely to adopt a new subsidiary when
the firm realizes a profitable return from its existing business. In all three columns, the coefficient
of CAP EXAi,t1 , b3 , confirms a negative expected sign and is significant, especially in columns
(2) and (3), at less than 5%. The theory says that the family has an incentive to recapitalize the
central firm only if the cost of the firms debt is not high. The cost of recapitalization is likely to be
high when the central firm generates a substantial private control benefit, which is expected to be
higher for a large project. Therefore, a firm tends to leverage itself up and serves as a central firm
when the firm does not invest in a large project. The negative sign and significance of b3 confirm
this prediction.
The positive and significant coefficients of F irmAgei,t1 and P ublici,t1 confirm that chaebol grow
over time and that an older, established firm is more likely to be a central firm. In column (3), the
firm size variable, Ln(Assets)i,t1 , turns out to be significant at less than 1%. This is consistent
with the theoretical prediction that an established, large firm is more likely to serve as a central
firm, since its large size helps minimize the probability of group default.
Column (4) reports the results when the lagged regressors are replaced with the contemporaneous
ones. The significant and positive coefficient of LEVi,t is still present, and the magnitude of the
coefficient is more or less the same as that in column (1).
In column (5), to show that the results in the first four columns are not driven by the generic
definition of Centrali,t , I use a continuous variable, Centralityi,t , as a dependent variable. The
significance of LEVi,t1 is greatly reduced, to the 10% level, but the positive sign of its coefficient
is confirmed. One possible reason for the lower significance of LEVi,t1 is that there are many zero
values below the median Centralityi,t .
In column (6), I run the same regression as in column (5), only with firms that do not belong to
a circular ownership link. The coefficient of LEVi,t1 becomes strongly significant at less than the
1% level, suggesting that the firms belonging to a circular ownership link tend to have the mixed
characteristics between those of a central firm and those of a subsidiary.
5.3.3

H3: Centrality and Dividend Payouts

Another important feature of the pyramidal structure is the greater availability of internal capital.
To test whether central firms serve as an internal capital market in the business group, the following
regression is run:
DIV Ai,t (or DIV U N Ai,t ) = c0 + c1 Centrali,t + c2 OP ROAi,t + c3 CAP EXAi,t + c4 LEVi,t
+c5 P ublici,t + 3 Controls + i,t
(16)
36

The causal link implied in Equation (16) is that when a firm becomes a central firm, it tends to
pay out fewer dividends. This conservative dividend policy of a central firm is more likely to be a
consequence of the business group structure in place, but less likely to cause the centrality of the
firm in the business group next year. The expected sign of c1 is negative.
I include the measures of profitability, OP ROAi,t , capital expenditure, CAP EXAi,t , and leverage
ratio, LEVi,t , on the RHS, since all these variables are perceived to be closely related to the
amount of dividends paid out by a firm. The more profitable a firm is, the more dividends the
firm is expected to pay out, which implies that the expected sign of c2 is positive. The variables
CAP EXAi,t and LEVi,t are related to the cash outflows from a firm. Both c3 and c4 are expected
to be negative. I also control for P ublic, since a publicly listed firm tends to pay out more dividends
to appreciate its share price. I expect to see a positive sign on the coefficient of P ublic. The usual
Year and Industry fixed effects are also controlled for in the regression, and the standard errors are
clustered at each firm level to correct for within-firm correlations.19
The results are reported in Table 8. In the first two columns, I use DIV Ai,t as a dependent variable,
and in the last two columns, I use DIV U N Ai,t as a dependent variable. In all four columns, one
finds a strong tendency for central firms to pay out fewer dividends. The significance of c1 in all
four columns is at less than 1% and the magnitude of the coefficient is roughly -30 bps. This tells
us that a central firm pays out less by around 30 bps of the book value of its operating assets.
The variables c2 , c3 , and c4 come up with their expected signs. The variables OP ROAi,t and
LEVi,t seem to be significantly correlated with the amount of dividends paid out by a firm, and
the significance is at less than 1% for OP ROAi,t and less than 10% for LEVi,t . Here c3 shows a
negative expected sign but is not statistically significant.
The variables DIV Ai,t and DIV U N Ai,t have a book value of operating assets and total assets as
their denominators. Hence, the positive correlation between Centrali,t and LEVi,t can cause the
payout ratios to be lower for the central firms. To demonstrate the robustness of the results, in
column (5) I use Dividendsi,t , the book value of total dividends paid out, as a dependent variable.
In this column, I use OrdIncomei,t , the book value of operating profits, Capexi,t , the book value
of total capital expenditures, P ublici,t , and Liabilitiesi,t , the book value of total liabilities, as
independent variables. To control for the size effect, I control for Assetsi,t , the book value of total
assets, and Assets2i,t , the square of Assetsi,t . The results are mostly the same as in the other four
columns.
The results in Table 8 support that an internal capital market exists in Korean chaebol, and this
results support the empirical findings in Table 6, where I report that a low leveraged firm tends
to be a pyramidal subsidiary. Hence, the results in Tables 6 and 8 are consistent with the models
internal capital market theory.
19

In the regression, I do not control for group fixed effects, since the group-specific structure variable, Central, is
on the RHS of the regression equation. Controlling for this additional fixed effect does not change the results, and
the magnitude and significance of the point estimates are almost the same as reported in the paper.

37

5.3.4

H4: Pyramid and Dividend Payouts

The theory predicts that a central firm can be highly leveraged due to control optionality and
the co-insurance benefit of the dividends from subsidiaries. Hence, one expects to see extensive
dividend payouts from subsidiaries to central firms in a business group. To test this prediction, I
run the following regression:

DIV Ai,t (or DIV U N Ai,t ) = d0 + d1 P yramidi,t + d2 OP ROAi,t + d3 CAP EXAi,t + d4 LEVi,t
+d5 P ublici,t + 4 Controls + i,t
(17)
I use a dummy variable, P yramidi,t in the main specification, but to show the robustness of the
results, I also run the regression with P ositioni,t .
The specification is similar to Equation (16). I control for P ublic and the usual Year and Industry
fixed effects. Standard errors are clustered at each firm level.20
According to theory, the sign of d1 should be positive. The expected signs of the coefficients of
OP ROAi,t , CAP EXAi,t , and LEVi,t should be the same as in Equation (16), and d2 should be
positive, and both d3 and d4 negative.
The results are reported in Table 9. The first two columns show the results when I use DIV Ai,t , the
book value of total dividends divided by the book value of operating assets, as a dependent variable.
The other two columns show the regression results with DIV U N Ai,t as a dependent variable, which
is the book value of total dividends divided by the book value of total assets. The sign of d1 , the
coefficient of P yramidi,t , is positive and statistically significant at less than 1% for column (1), and
at less than 5% for columns (2) through (4). The average magnitude of the coefficients reveals that
a pyramidal subsidiary pays out more than the other group firms by around 28 bps of the book
value of its operating assets, and by around 29 bps of book value of its total assets. A significantly
positive d2 and a negative d4 confirm that the more profitable and less leveraged a firm is, the more
it pays out.
In the last column, column (5), I use P ositioni,t as a dependent variable. I use the same independent
variables as in Equation (16) from the previous section. The coefficient of P ositioni,t is significant
at less than 5%. Hence, neither the generic definition of P yramidi,t nor the possible issue in the
denominators of DIV Ai,t and DIV IN Ai,t drive the results.
In summary, the results in Table 9 imply that subsidiaries tend to pay out more, and the dividends
paid out will be transferred to the central firm as the firms non-operating profit, which helps
increase the firms leverage ratios, as shown in Table 7. De Jong et al. (2009) report empirical
evidence on the dividends policy of a French family business group that is similar to the results
in this paper. The authors document that in French pyramids, more dividends are paid out from
subsidiaries to service central firm debt, and the central firms tend to be highly leveraged.
Benevolent dividend payouts in pyramidal subsidiaries are also documented in Faccio et al., 2001.
20

Similar to the specification used in Section 5.3.3, I do not control for group fixed effects in this regression either.
Controlling for this additional groups fixed effects, however, does not change the results, and almost the same
economic magnitude and significance of the point estimates are carried over to the case where I control for group
fixed effects.

38

Using family business group data in Western Europe and Asia, their findings support the substitution hypothesis(La Porta et al., 2000), that by committing to the benevolent dividend payouts,
the controlling family can achieve better firm valuation of group firms. The substitution hypothesis
is grounded in a corporate governance perspective and fails to establish the joint relation between
dividend policy and capital structure decisions. Hence, the theory and the empirical evidence in
this paper can provide a more comprehensive picture of how the two policies are linked together to
generate a control-efficient family business group.

5.4

Robustness Checks: Is the Leverage Ratio a Determinant of business Group

Structure?

Group structure metrics such as P ositioni,t and Centralityi,t are relatively persistent. Moreover,
the leverage ratio is known to be persistent. This implies that the empirical results in Tables 6
and 7 may be subject to the reverse causality problem due to the persistence of the dependent and
independent variables. To provide further evidence that a firm with low leverage is more likely to
be a pyramidal subsidiary and a highly leveraged firm is more likely to be a group central firm, I run
two additional tests utilizing my datas dynamic properties. A number of firms in the sample show
significant change in their P osition/Centrality in two consecutive years, and these firms enable me
to further investigate the causal relation between the leverage ratio and group structure metrics.
5.4.1

Robustness Checks: Does a Low Leverage Ratio Predict a Large Increase in

Position?

First, by using a threshold value for change in P osition, I identify firms that become pyramidal
subsidiaries by increasing their P osition values above the threshold. Similarly, I also identify firms
that become stand-alone entities by decreasing their P osition values below the threshold. To define
a threshold value of large change in P osition, I define a new variable,
DP ositioni,t = P ositioni,t P ositioni,t1

(18)

where DP ositioni,t measures the change in the position of firm i from year t1 to year t. Details of
the summary statistics of DP ositioni,t are given in Table 10. The average and standard deviation
of DP osition are 0.027 and 0.035, respectively. The 25th and 75th percentiles of the variable
are -0.024 and 0.047, respectively, and the median value is zero. The zero median of the variable
suggests that most firms in the sample tend to stay at the same P osition year to year.
I choose two different sets of threshold values, Tkj , j = U p, Down, for k = 1, 2, and define an event
where a given firm i experiences a significant change in P ositioni,t from year t 1 to year t by
using these variables. The main reason to focus only on a significant change in P osition in this
test is that leverage ratios are persistent, and thus using a small increase/decrease (i.e., Up/Down)
in P ositioni,t will not help resolve the issue of reverse causality.
Using a pair of threshold values for up and down movement in P osition, (TkU p , TkDown ) for k = 1, 2,

39

I define a dummy variable, BCP yramidki,t , as follows:

BCP yramidki,t

if DP ositioni,t TkU p , P yramidi,t = 1

1
=
0
if DP ositioni,t TkDown , P yramidi,t = 0

N ot Assigned O.W.

(19)

The dummy variable BCP yramidki,t identifies two events: 1) Firm i becomes a subsidiary located
at the bottom of the pyramid at year t by increasing its position value from year t 1 by more than
TkU p (i.e., BCP yramidi,t = 1) and 2) firm i becomes a stand-alone entity located at the top of the
group by decreasing its position value from year t1 by more than TkDown (i.e., BCP yramidi,t = 0).
I do not impose the restriction that the type of firm i should also change in two consecutive years
(i.e., P yramidi,t 6= P yramidi,t1 ). Hence, several firms in the testing sample can preserve their
type from year t 1 to year t, even after experiencing a large increase/decrease in P osition. To
ensure a reasonable testing sample size, I do not impose the restriction and focus instead on large
changes in each firms P ositioni,t .
To complete the definition for BCP yramidki,t , I define the threshold values of DP ositioni,t for each
k = 1, 2 as
T1U p = 0.027, T1Down = 0.027 for k = 1
T2U p = 0.047, T2Down = 0.024 for k = 2

(20)

For k = 1, I use the average value of DP osition, 0.027, as a threshold for both increases (i.e., Up)
and decreases (i.e., Down) in P osition. I call this the symmetric thresholds case. Firms with an
absolute value of position increase/decrease from year t1 to year t greater than 0.027 are included
in the testing sample. Among these firms, only a firm with P yramidi,t = 1 after P osition increase
takes BCP yramid1i,t of 1. Similarly, only a firm with P yramidi,t = 0 after P osition decrease takes
BCP yramid1i,t of 0.
For k = 2, I consider the asymmetric nature of the distribution of DP osition and use two different
threshold values for the up and down movements of P osition. I call this the asymmetric thresholds
case. I use 0.047 as the threshold value for increases in P osition, which is the 75th percentile of
DP osition. I use -0.024, the 25th percentile of DP osition, as the threshold value for decreases
in P osition. I assign BCP yramid2i,t = 1 only to those firms that satisfy P yramidi,t = 1, and
BCP yramid2i,t = 0 to those satisfying P yramidi,t = 0.
With BCP yramidki,t for k = 1, 2, I run the following causal regression:
BCP yramidki,t = f0 + f1 LEVi,t1 + f2 OP ROAi,t1 + f3 CAP EXAi,t1
+5 Controls + i,t
f or k = 1, 2

(21)

Equation (21) should be much less restrictive in the reverse causality problem, since it is least likely
that any large change in P osition from year t 1 to year t reverse-causes the leverage ratio decision
made at year t 1. The same argument applies to the other regressors on the RHS of Equation
40

(21), operating profitability and capital expenditures. I control for Ln(Assets)i,t1 but I do not
control for the other two variables, F irmAge and P ublic, in this regression, since they are both
highly correlated with Ln(Assets)i,t1 . The usual Industry and Year fixed effects are controlled
for, and the standard errors are clustered at each firm level.
The results are reported in Table 11. The first two columns show the results when I use the average
value of DP osition, 0.027, as a threshold value for up/down position movements. Here f1 has a
negative value of -0.053 in column (1) and -0.058 in column (2), and both are statistically significant
at less than 1%. Hence, a low leverage ratio results in a large P osition value increase, and a firm
with a low leverage ratio is likely to be a pyramidal subsidiary in the following year.
The negative value of f2 in columns (1) and (2) confirms the theoretical prediction that as the
profitability of a firm decreases, the likelihood of it being a pyramidal subsidiary increases. In
column (1), the positive and significant coefficient of CAP EXAi,t1 confirms the capital-intensive
result in Table 6. In column (2), the significance of the coefficient is not present, but the sign is still
confirmed. The control variable, Ln(Assets)i,t1 , is still a significant determinant of a pyramidal
subsidiary, implying that a smaller firm will tend to be a pyramidal subsidiary.
In columns (3) and (4), I use asymmetric threshold values for DP ositioni,t . Even with this more
stringent definition of BCP yramid2i,t , the leverage ratio remains an important determinant of
business group structure.
5.4.2

Robustness Checks: Does a Highly Leveraged Firm Become a Group Central

Firm?

Similar to what was done with P ositioni,t , I define two events for firm i at year t: 1) It becomes
a group central firm at year t from a non-central firm at year t 1 by increasing its centrality by
more than a threshold value and 2) it remains a non-central firm from year t 1 to year t. In
this analysis, I restrict the P osition of firms to less than 2.0 to exclude cases where a pyramidal
subsidiary becomes an intermediate central firm located at the bottom of the pyramid.
I compute the change in Centralityi,t and denote it by a new variable, DCentralityi,t , which
measures the increase/decrease in Centrality of firm i from year t 1 to year t. It is defined as
DCentralityi,t = Centralityi,t Centralityi,t1

(22)

The summary statistics of DCentralityi,t are given in Table 10. The average and standard deviation
of DCentrality are 0.002 and 0.024, respectively. The median and 25th and 75th percentiles of
the variable are all close to zero, which implies that the change in Centrality is very small and the
distribution of the variable has a right fat tail. To define a significant change in Centrality, I use
one standard deviation of DCentrality, 0.024, as the threshold value. By using this threshold for
DCentralityi,t , I define a dummy variable, BCCentrali,t , as


Centrali,t = 1,
Centrali,t1 = 0,

if
1
DCentralityi,t 0.024, P ositioni,t 2.0
BCCentrali,t =
0
if Centrali,t = 0, Centrali,t1 = 0, P ositioni,t 2.0

N ot Assigned O.W.
(23)
41

The new dummy variable, BCCentrali,t , takes on a value of 1 if a non-central firm i at year t 1
(i.e., Centrali,t1 = 0) becomes a central firm at year t (i.e., Centrali,t = 1) by increasing its
centrality by more than 0.024. The zero value of this variable indicates that firm i does not change
its Centrality much from year t 1 to year t and remains a non-central firm in the two consecutive
years. With this dummy variable, I run the following regression:
BCCentrali,t = g0 + g1 LEVi,t1 + g2 OP ROAi,t1 + g3 CAP EXAi,t1
+6 Controls + i,t

(24)

Equation (24) directly tests whether the leverage ratio, operating profitability, and capital expenditure are the determinants of a central firm. One expects g1 > 0, g2 > 0, g3 < 0. I control for
F irmAge, P ublic, and Ln(Assets), and the usual Industry and Year fixed effects are also controlled
for. The standard errors are clustered at each firm level.
The results of Equation (24) are reported in Table 12. In all four columns, a high leverage ratio
predicts that the firm is more likely to be a new central firm in the following year. The sign of g1
is positive and statistically significant at less than 5% at least, except in column (4), where it is
significant at the 5% level. A positive and statistically significant g2 indicates that a firm is more
likely to be a new central firm when it realizes a profitable return from its existing business.
The signs of the coefficients of F irmAgei,t1 and P ublici,t1 suggest that an old public firm tends
to be a new central firm. The coefficients are significant at the 5% level at least. Lastly, a larger
firm seems more likely to be a new central firm, and the fact that the coefficient of Ln(Assets)i,t1
is positive and significant at the 1% level supports this.

Conclusions

I propose a theory of how a control-driven family designs a business group structure and makes
debt financing decisions to achieve efficient control of the group. By combining the settings in
two well-known papers Harris and Raviv (1988) and Almeida and Wolfenzon (2006)I propose
several novel predictions about pyramidal family business groups, decreasing leverage ratios and
increasing dividend payouts from the top to the bottom of the pyramid. The internal capital market
formed at the top of the pyramid and the pyramids two intrinsic propertiescontrol optionality
and co-insurance benefitsdrive the decreasing leverage ratios from the top to the bottom of the
pyramid to be optimal. These decreasing leverage ratios are supported by the increasing dividend
payouts from the top to the bottom of the pyramid.
Using unique Korean chaebol s 7-years panel data, I provide empirical evidence consistent with
these model predictions. The leverage ratio seems to be closely related to the choice of business
group structure. Dividends are used as the means to transfer resources from one group firm to
another to financially support each other.
The theory and empirical evidence support that controlling families of Korean chaebol strategically
design business group structure and make financing decisions to maximize control benefits.
The theory also proposes the following additional testable hypotheses: 1) the familys ownership
in a pyramidal subsidiary tends to be concentrated, whereas its ownership in a central firm tends
42

to be low, 2) lower fraction of inside equity in a group firm implies that the firm tends to be highly
valuated and profitable, 3) group firms controlled through the pyramidal ownership chain tend to
operate in similar industries rather than diversified ones. Testing these implications is remained as
a subsequent research agenda of this paper.

43

References
Aghion, Philippe and Bolton, Patrick, 1992, An Incomplete Contracts Approach to Financial
Contracting, Review of Economic Studies 59, 473-494.
Almeida, Heitor, and Wolfenzon, Daniel, 2006, A theory of pyramidal ownership and family business groups, Journal of Finance 61, 2637-2680.
Almeida, Heitor, Park, Sang Y., Subrahmanyam, Marti G., and Wolfenzon, Daniel, 2009, The
structure and formation of business groups: Evidence from Korean chaebols, Working paper,
New York University.
Barlow, R., and Proschan, P., 1975, Statistical Theory of Reliability and Life Testing, New York,
Holt, Rinehart, and Winston.
Bebchuck, A. Lucian, Kraakman, Reinier, and Triantis, George, 2000, Stock pyramids, crossOwnership, and dual class equity: the creation and agency Costs of separating control from
cash flow rights, In: Morck, R. (Ed.), Concentrated Corporate Ownership, 445-460.
Bianco, Magda, and Nicodano, Giovanna, 2006, Pyramidal groups and debt, European Economic
Review 50, 937-961.
Chemmanur, Thomas, and John, Kose, 1991, Optimal Incorporation, Structure of Debt Contracts
, and Limited-recourse Project Financing, Journal of Financial Intermediation 5, 372-408.
Claessens, Stijn, Simeon Djankov, and Larry H.P. Lang, 2000, The separation of ownership and
control in East Asian Corporations, Journal of Financial Economics 57, 2741-2771.
De Jong, Abe, Dejong, V. Douglas, Hege, Ulrich, and Mertens, Gerard, 2009, Debt and dividends
in pyramidal structures: evidence from France, Working paper, University of Iowa
Dyck, I.J. Alexander and Zingales, Luigi, 2004, Private Benefits of Control: An International
Comparison, Journal of Finance 59, 537-600
Faccio, Mara, Larry H.P. Lang, and Young, Leslie, 2001, Dividends and Expropriation, American
Economic Review 91, 54-78.

44

Faccio, Mara, and Larry H.P. Lang, 2002, The ultimate ownership of western European corporations, Journal of Financial Economics 65, 365-395.
Ferris, P. Stephen, Kim, A. Kenneth, and Kitsabunnarat, Pattanaporn, 2003, The cost(and benefits?) of diversified business groups: The case of Korean chaebols, Journal of Banking and
Finance 27, 251-273.
Franks, Julian, and Mayer, Colin, 2001, Ownership and Control of German Corporations, The
Review of Financial Studies 14, 943-977.
G
arleanu, Nicolae, and Zwiebel, Jeffrey, 2009, Design and Renegotiation of Debt Covenants, The
Review of Financial Studies 22, 749-781.
Goplan, Radhakrishnan, Nanda, Vikram, and Seru Amit, 2006, Do business groups use dividends
to fund investment?, Working paper, Washington University.
Grossman, Sanford J., and Hart, Oliver D., 1987, One share/one vote and the market for corporate
control, NBER Working paper No. 2347.
Harris, M., and Raviv, A.,1988, Corporate control and capital structure, Journal of Financial
Economics 20, 175-202.
Israel, Ronen, 1991, Capital Structure and the Market for Corporate Control: The Defensive Role
of Debt Financing, Journal of Finance 46, 1391-1409.
Jang, Hasung, Kang, Hyung Cheol, and Park, Kyung Suh, 2005, The Choice of Group Structure:
Divide and Rule, CEI Working Paper Series, No. 2005-6
Jensen, Michael, 1986, Agency costs of free cash flow, corporate finance and takeovers, American
Economic Review 76, 323-329.
Jensen, Michael, and Meckling, William, 1976, Theory of the firm: managerial behavior, agency
costs and ownership structure, Journal of Financial Economics 3, 305-360.
La Porta, F. Lopez-de-Silanes, A. Shleifer and R. Vishny, 1997, Legal determinants of external
finance, Journal of Finance 52, 11311150.

45

La Porta, F. Lopez-de-Silanes, A. Shleifer and R. Vishny, 1998, Law and Finance, Journal of
Political Economy 106, 11131155.
La Porta, Rafael, Florencio Lopez-de-Silanes, and Andrei Shleifer, 1999, Corporate ownership
around the world, Journal of Finance 54, 471-517.
La Porta, F. Lopez-de-Silanes, A. Shleifer and R. Vishny, 2000, Agency problems and dividend
policies around the world, Journal of Finance 55, 1-33.
La Porta, Rafael, Florencio Lopez-de-Silanes, and Andrei Shleifer, 2002, Investor Protection and
Corporate Valuation, Journal of Finance 57, 1147-1170.
Luciano, Elisa, and Nicodano, Giovanna, 2008, Leverage and firm scope, SSRN Working Paper
Series.
Nakatani, I., 1984, The economic role of corporate financial groupings, In: Aoki, M. (Ed.), Economic Analysis of Japanese Firm, Elsevier, New York.
Nenova, Tatiana, 2003, The value of corporate voting rights and control: A cross-country analysis,
Journal of Financial Economics 68, 325351
Prowse, S., 1992, The structure of corporate ownership in Japan, Journal of Finance 42, 1121-1140.
Scharfstein, S. David, and Stein, C. Jeremy, 2000, The Dark side of internal capital markets:
divisional rent-seeking and inefficient investment, Journal of Finance 55, 2537-2564.
Shin, H., Park, Y.S., 1999, Financing constraints and internal capital markets: evidence from
Korean chaebols, Journal of Corporate Finance 5, 169-191.

46

Numerical Analysis: Quadrature Definitions

The joint p.d.f of the cash flows for firm A and B is denoted by f (YA , YB ). The payoff to the
family under the pyramidal structure, P , is written as
R + R +
P = Z + (KA + KB ) DP DP P (yDP ) f (x, y)dxdy
B

+KA

R DP R +
B

P
DA

f (x, y)dxdy

And, the equity values of firm A and firm B are

P , DP ) =
SA (DA
B

R DP R +
B

P P (yD P )
+ DA
B

+
P)
SB (DB

R DP R +
B

P
DA

R +
P
DB



P ) D P f (x, y)dxdy
x + P (y DB
A

P )f (x, y)dxdy
(x DA

P )f (y)dy
(y DB

The payoff and equity value functions are expressed with the double-integration on a non-square
integration region. To compute these quantities, I use a numerical quadrature. Each doubleintegration is computed for [5, 5] range of Y , and the computation delivers a precision level of
109 .

47

Figure 1: Symmetric Projects, KA = KB , 30% of a Firm Value

Panel (a)
100

Recap. A Max
Norecap. B

Debt Amount

80

60

40

20
1

0.5

0
Correlation

0.5

Panel (b)
1

Recap. A
Recap. B
Norecap. B

Leverage Ratio

0.8
0.6
0.4
0.2
0
1

0.5

0
Correlation

0.5

Panel (c)

Expected Payoff

65

Recap.
Norecap.

60

55

50

45
1

0.5

0
Correlation

0.5

48

B
B.1

Proofs
Proof of Lemma 1

As long as iB C +(1B )YB , H (tH , H , 0) = C +YA + YB iB +KA +KB , which is constant

for tH C, H B . Hence, raising an excessive fund by cash and equity financing beyond iB
H =i )
H (MB
B
does not increase the payoff, H (tH , H , 0). But, using risky debt is costly since
<0
DH
B

H > 0.
for DB

H i +Y D H ) (M H =i )
(P rob(MB
B
B
B
B )
B
H
DB
H M H +i ) (M H =i )
(1FB (DB
B )
B
B
B
=
H
DB
H
H
H M H + i )(M H = i ) (DB MB +iB ) (M H = i )
= fB (DB
B
B
B
H
B
B
B
DB




F
H
+
H
F
H +

B
(C+YB E [YB DB ] )
H)
H ) = 1 + H E B [YB DB ]
fB (DB
f
(D
= 1
B
B
DH
DH
B

= 1

H
DB

fB (YB )dYB

H)
fB (DB

0
A marginal dollar face value increase in the risky debt returns capital, which is strictly less than a
H ) + B (D H ) should exactly equate i for i > C + (1
dollar. Therefore, tH + (1 H )SB (DB
B
B
B
B
B )YB . 

B.2

Proof of Lemma 2

From Lemma 1, we know

j
j (MB
=iB )
j
DB

j
> 0 and j = P, H. This implies that risky debt
< 0 for DB

is issued only if iB exceeds the maximal cash and equity financing amount, and the issued amount
of risky debt should be the minimum that satisfies MBj = iB j = P, H. 

B.3

Proof of Lemma 3

H) =
From Lemma 2, we have SB (D
B
H D
P . Hence, for iB 0,
D
B
B

B.4

H
DB
YB

>

C+YB iB
B
P
DB
. 
YB

P) =
and SB (D
B

C+YB iB
.
B

This implies that

Proof of Proposition 1

By imposing the optimal financing policy from Lemma 2, we have the optimal payoff to the family
under the horizontal structure given as:


C + YA + (YB iB ) + KA + KB 
H
 if iB C + (1 B )YB
=
H
) if iB > C + (1 B )YB
C + YA + (YB iB ) + KA + KB 1 FB (D
B

49

Similarly, the optimal payoff to the family under the pyramidal structure is


C + YA + (YB iB ) + KA + KB 
P

 if iB C + (1 B )YB
=
P
) if iB > C + (1 B )YB
C + YA + (YB iB ) + KA + KB 1 FB (D
B
(25)
H > D
P from Lemma 3. Since YB iB 0, the horizontal structure is always chosen if iB <
D
B
B
C + (1 B )YB . For iB C + (1 B )YB , the pyramidal structure is chosen if
P H

H )) FB (DP ) (1 )(YB iB )
KB FB (D
B
B


B.5

Proof of Corollary 1





H ) FB (DP ) (1) YB iB decrease in pB for iB 0.
iB C (1B )YB and KB FB (D
B
B
(1(pB ))YB
YB
B)

0 since (p
pB
pB 0 and pB > 0. This implies that iB C (1 B )YB decreases
in pB .




H ) FB (DP ) (1 ) YB iB is clearly met at pB = 0. This is because
Moreover,
KB FB (D
B
B


H ) FB (DP ) 0 due to D
H > DP for iB and YB (pB = 0, iB ) iB = 0.
KB FB (D
B
B
B
B





KB /YB
H ) FB (DP ) (1 ) YB iB decrease
< 0 and (YB ipBB )/YB > 0 imply that KB FB (D
B
B
pB
in pB . At pB = 1, iB < C + Y (pB = 1, iB ) for
iB 0 since (pB ) = 0 and Y (pB = 1, iB )iB > 0.

H ) FB (DP ) = 0 at pB = 1. Since Y (pB = 1, iB ) iB > 0,
This implies that KB FB (D
B
B


H
P

KB FB (DB ) FB (DB ) (1 ) YB iB < 0 at pB = 1.

(pB = 0) = 0.5 implies that for iB > 2C, there exists 0 < p < 1 such that the two conditions
in Proposition 1 hold at strict equality. This implies that when the pyramidal structure is chosen,
0 pB p holds for iB > 2C.
Moreover, for iB 2C, the pyramidal structure is chosen only if pB = 0. Therefore, for iB 0,
the likelihood of pB 0.5 is greater than that of pB > 0.5 given that the pyramidal structure
B)
0
is chosen since p is uniformly distributed over [0, 1]. This condition in addition to (p
pB
completes the proof. 

B.6

Proof of Lemma 4

P ()
= DP . Then,
From the conditions given in Lemma 4, we know that there is 0 < 1 s.t. XA
A
the date-2 stochastic payoff to the family under the pyramid is summarized as


0
if <
 P

=
+
P
XA ( = 1) DA
+ KA + KB if
 P

P + + K + K > 0, it is optimal to set
This implies that WLOG,
Since XA
( = 1) DA
A
B
= 1 is optimal. 
50

B.7

Proof of Lemma 5

I need to show

P =i ,tP =C+B(D P ,D P )
P (MB
B
A
B )
P
DA

<0,

P =i ,tP =C+B(D P ,D P )
P (MB
B
A
B )
P
DB

P , D P > 0.
< 0 for DA
B

Lets denote C + YA + (YB iB ) by Z, and by using a joint p.d.f of YA and YB , define P r1 and
P r2 as following:
P ]+ D P }
P r1 = P rob{YA + P [YB DB
A
P
P} + Pr
= P rob{YA DA , YB < DB
2
R DP R
B
=
P fA,B (x, y)dxdy + P r2
DA
P
P
P
P
P r2 = RP rob{Y
R A + [YB DB ] DA , YB DB }
= DP DP P (yDP ) fA,B (x, y)dxdy
B

Then,


P
P
P MBP = iB , t = C + B(DA
, DA
) = Z + KA P r1 + KB P r2
Z DP
B
P r2
P r1
P
=

fA,B (DA
, y)dy +
P
P
DA
DA
R
P r2
P
P
P
=

P
DP fA,B (DA (y DB ), y)dy
D
B

This implies
0 and

P r2
P
DA

P r2
P
DA

P , D P > 0. Moreover,
< 0 for DA
B

P r1
P
DA

P , D P > 0 since
< 0 for DA
B

R DP

P , y)dy <
fA,B (DA

P , D P > 0.
< 0 for DA
B
P r2
P
DB

R
P P (y D P ), y)dy
= P DP fA,B (DA
B
B
R
P )dx
DP fA,B (x, DB
A

This implies

P r2
P
DB

< 0 for

P , DP
DA
B

> 0.

Lastly,
P r1
P
DB

=
=

P r2
P
P fA,B (x, DB )dx +
P
DA
DB
R
P
P
P
DP fA,B (DA (y

And thus,

P r1
P
DB

P ), y)dy
DB

P , D P > 0.
< 0 for DA
B

All these quantities imply the following:

P
P
DA
P
P
DB

R DP

P , y)dy
fA,B (DA
R
P P (y D P ), y)dy
[KA + KB ] DP fA,B (DA
B
B
R
P
= KB DP fA,B (x, DB )dx
A
R
P P (y D P ), y)dy
[KA + KB ] DP fA,B (DA
B

= KA

P , D P > 0. 
And both are negative for DA
B

51

(26)

B.8

Proof of Lemma 6

0P , D P > 0, P = . First, suppose D 0P < D P . Then

For an interior equilibrium such that DA
B
B
A
B
P , D P ) = E F [Y D 0P ]+ + E F [Y D P ]+
SA (DA
B
B
B
B
A
B

Therefore, the following holds:

SA
0P
DA
SA
P
DB

0P )]
= [1 F (DA
P )]
= B [1 F (DB

Set the Lagrangian for the program (5) with Equation (8) in the paper.
0P , D P , ) = Z + (K + K ) K F (D 0P ) K F (D P )
L(DA
B
A
B  B
B
A
 A
P , DP ) i
+ C + YA + YB SA (DA
B
B

is a non-negative multiplier for the budget constraint in the program. The constraint qualification
P , D P , is clearly met. Moreover, the convexity of S (D P , D P ) in
for the two control variables, DA
A
B
A
B
P
P
DA and DB , in addition to IHR assumption on F (.), imply Kuhn and Tucker(KKT) condition is a
sufficient and necessary for the optimality. Hence, the optimality is characterized by the following
two first-order conditions:


L
0P ) + 1 F (D 0P ) = 0
= KA f (DA
0P
A
DA


L
P ) + 1 F (D P ) = 0
=
K
f
(D
B
B
P

B
B
D
B

These two conditions imply

f (D0P )

0P
A
KA 1F (D
0P ) = KA H(DA ) = =
A

P
KB f (DB )
P )
B 1F (DB

KB
P
B H(DB )

0P < D P at optimality, and they satisfy the

B
Hence, if KA > K
B
B , then by IHR assumption, DA
0P ]+ + E F [Y D 0P ]+ = C + Y
A +
budget constraint at strict equality. And thus, E F [YB DA
B
B
B
KB
P

0P

YA iB . If KA B , then DA < DB can not be optimal. Therefore, only if KA > K

B ,
0P < D P is optimal. At this optimality, the payoff to the family under the pyramidal structure
DA
B
is given as
0P
P
P = Z + KA + KB KA F (DA
) KB F (DB
)
0P D P > 0, the following holds:
Now, for DA
B
0P , D P ) = [1 + ]E F [Y D P ]+
SA (DA
B
B
B
G
P)
= SA (DG

This implies
P)


SA (DG
P
= [1 + B ] 1 F (DG
)
P
DG
0P D P > 0 gives
Setting the Lagrangian for DA
B




P
P
P
L(DG , ) = Z + (KA + KB ) 1 F (DG
) + C + YA + YB SA (DG
) iB

52

First-order sufficient and necessary condition by KKT is

L
P
DG

P ) + [1 + ][1 F (D P )] = 0
= (KA + KB )f (DG
B
G
(KA +KB )
P
1+B H(DG )

P , the budget constraint is tight since > 0. Thus we have

And, at this DG
P +
[1 + B ]E F [YB DG
] = C + YA + YB iB
P , D P < D P < D 0P . At this optimality, the payoff to the family under
From the definition of DG
B
G
A
the pyramidal structure is given as
P
P = Z + (KA + KB )[1 F (DG
)]

B.9

Proof of Proposition 2

B 1
Lets denote K
KA B by . From Equation (25) in Appendix B.4, the optimal date-1 expected payoff
P > 0, is
to the family under the pyramid without recapitalization of firm A, when DB


= Z + (Y
B iB ) + KA + KB 1 F (D
P)
PBM
B

P ]+ = iB .
where Z = C + YA + (YB iB ) and C + YB B E F [YB D
B
The recapitalization is done only if the optimal date-1 expected payoff to the family under the
, satisfies P > P .
pyramid with recapitalization of firm A, PEM
EM
BM


KB
P

P
P

) F (DP ) > KA F (DP ). ThereWhen 1 (i.e., KA B ), EM > BM only if KB F (D

G
G
B
B
fore, the larger K
,
the
more
likely
the
recapitalization
of
firm
A
is
optimal.
Hence,
as

increases,
KA
firm A is more likely to be re-capitalized.


P
P
P
0P
B
P
When < 1 (i.e., K
KA < B ), EM > BM only if KB F (DB ) F (DB ) > KA F (DA ). This
B
condition is also more likely to be met for a larger K
KA .


P
0P
B
P
But since K
KA increases in , with < 1, KB F (DB ) F (DB ) > KA F (DA ) is not likely to be
met.
0P , D P > 0 along the pyramid, D 0P D P is more likely to be
Therefore, when we observe DA
B
A
B
observed in equilibrium.

This implies that when the two project profiles are identical, and thus Q = YA YB = 0,


53

P
D
A
YA

P
D
B
.
YB

B.10

Proof of Proposition 3

Comparing the two optimal payoffs, P under a fully leveraged pyramid and H under the
horizontal structure. For iB C + (1 B )YB ,
P H 0
P )] (Y
B iB ) KA KB [1 F (D
H )] 0
(YB iB ) + (KA + KB )[1 F (DG
B
H
P

KB [F (DB ) F (DG )] (1 )(YB iB ) KA F (DG ) 0

H ) F (DP )] (1 )(YB iB ) + KA F (DP )
KB [F (D
B
G
G
H = 0, and thus a fully leveraged pyramid is never chosen in
For iB < C + (1 B )YB , D
B
equilibrium. 

B.11

Proof of Corollary 2

P ]+ = C+YA +YB iB from Lemma 6. This implies that for a given project profile of
E F [YB DG
1+B
P decrease in Y
A . Larger YA and smaller iA predicted from Proposition 3 imply
(YB , iB , KB ), DG
that NPV of firm As project, YA iA tends to be high if we observe a fully leveraged pyramid
with decreasing leverage ratio from top to bottom. Therefore, pA should be high under the fully
leveraged pyramid. 

B.12

Proof of Proposition 4

P = 0. Then, the family receives the optimal payoff, P (D P = 0), from a fully
Suppose that DB
B
P = 0) = Z + (K + K ) [1 F
P
leveraged pyramid such that P (DB
A
A
A,B (YA + B YB < DA )] where
+
P
F
E A,B [YA + B YB DA ] = C + YA + YB iB .

Perfect correlation assumption gives YA = YB + Q and YB = YB +  where Q = YA YB and

 F (.). Hence, in this perfect correlation case, we have FA,B (., .) = F (.).
C ]+ = i .
From Lemma 7, by plugging in VC = 1, we have C + YA + YB E FA,B [YA + YB DA
h
i+ B
P
C
P
DA Q
DA Q
DA
F
This implies that for the perfectly correlated projects, 1+B
since E YB 1+B
=
2
h
i

DC +
C+YA +YB iB
> C+YA +2YB iB = E F YB 2A
.
1+B
h
 P i
P = 0) = Z + (K + K ) 1 F DA Q
Therefore, for the perfectly correlated projects, P (DB
A
A
1+B
h
 C i
DA Q
C
C
P
P
is greater than (DA ) = Z + (KA + KA ) 1 F
. And we know (DB = 0) may or
2

may not be optimal solution. Hence, the optimal payoff under the fully leveraged pyramid, P ,
P = 0). We have P C , which completes the proof. 
satisfies P P (DB

54

Definitions of group metrics

The description of the group metrics in this section is from Almeida et al.(2009). Consider a
business groups with N firms. We define the matrix of inter-corporate holdings A as follows:

0 s12
...
s1N
s21
0
...
s2N

A= .
.
.
..
.
.
.
.
.
.
.
sN 1

...

sN

N 1

where sij is the stake of firm i in firm j. In other words, column j contains the stakes of the
corporate direct owners of firm j in all other firms, 1, 2...i..N, i 6= j We also define a vector with
the direct stakes of the family in each of the N firms:

f1
f2

f = .
.
.
fN
Definition 1 (Ultimate Ownership). The ultimate ownership of the family in each of the n firms
is given by u = [u1 u2 ... uN ]0 :
u0 = f 0 (IN A)1
where IN is the N N identity matrix.
Definition 2 (Position). The position of firm i in the group is defined as:
P ositioni =

X
f 0 An1 di
n
ui

n=1

Almeida et al.(2009) assume the following two for defining Centrality:

Assumption 2 (Control Threshold). A family controls a firm if and only if it holds more than T
votes in it.
Assumption 3. The votes that a family holds in a firm is the sum of its direct votes plus all the
direct votes of firms under family control, where control is defined in Assumption 2.
Definition 3 (Set of firms under the familys control). For a given threshold T , the set of firms
controlled by the family is given by:
X
C(T ) = {i N : fi +
sji T }.
jC(T ), j6=i

Definition 4 (Critical Control Threshold). For any firm i N , the critical control threshold is
given by
CCi = max{T | i C(T )}
55

Definition 5 (Centrality). We define the centrality of a firm i as:

P
P
CCji
CCj
Centralityi =

j6=i

j6=i

]N 1

where CCji is the critical control threshold of firm j, computed as if firm i held no shares in the
other group firms.
Definition 6 (Loop). Let
loopi = min{n n 1 and d0i An di > 0},
then firm i is in a loop if and only if loopi < . Hence, define Loopi = 1{loopi <} .
I define Loopi,t = 0 if any dividends paid by firm i do not return to the firm itself within N steps,
where N denotes the total number of group firms in a business group.

56

Tables
Table 2: Summary Statistics of Group Structure Metrics
Variables
Position
Centrality
Loop
Central Firms Position
Circular Firms Position
Top Firms Centrality
Public
Employees
Firm Age
No. of Firms
No. of Groups

Mean
2.110
0.150
0.253
1.610
1.859
0.046
0.259
1204
17
1072
47

Stdev
0.816
0.049
0.435
0.600
0.764
0.093
0.438
3770
14

Median
2.057
0.000
0.000
1.365
1.710
0.002
0.000
198
13

25%
1.400
0.000
0.000
1.100
1.220
0.000
0.000
45
4

75%
2.547
0.003
1.000
2.010
2.288
0.040
1.000
846
26

Obs.
3442
3418
3442
556
872
457
3442
3442
3442

Table 3: Correlations between variables in Table 2

Variables
Position
Centrality
Loop
Public
Employees
Firm Age

Position
1.00
-0.25
-0.18
-0.24
-0.17
-0.30

Centrality

Loop

Public

Employees

Frim Age

1.00
0.21
0.38
0.25
0.41

1.00
0.42
0.30
0.46

1.00
0.35
0.60

1.00
0.32

1.00

57

Table 4: Summary Statistics of Financial Variables

Variables
Total Assets (in mil. USD)
Operating Assets (in mil. USD)
LEV
OPROA
CAPEXA
DIVA
DIVUNA

Mean
905
831
0.203
0.024
0.056
0.008
0.006

Stdev
3090
2870
0.382
0.113
0.148
0.099
0.021

Median
88.3
85.3
0.120
0.027
0.029
0.000
0.000

25%
20.1
19.3
0.000
-0.009
0.008
0.000
0.000

75%
509
469
0.315
0.085
0.072
0.006
0.006

Obs.
3442
3442
3442
2974
2590
2975
2975

Table 5: Correlations between Pyramid, Central and main financial variables in Table 4
Variables
Pyramid
Central
LEV
OPROA
CAPEXA
DIVA

Pyramid
1.000
-0.400
-0.10
0.002
0.085
0.025

Central

LEV

OPROA

CAPEXA

DIVA

1.000
0.100
-0.006
-0.070
-0.032

1.000
-0.211
-0.050
-0.082

1.000
0.040
0.262

1.000
-0.021

1.000

58

Table 6: Pyramid vs. Leverage

LEV
OP ROA
CAP EXA
F irmAge

(1)

(2)

(3)

(4)

(5)

(6)

P yramidi,t

P yramidi,t

P yramidi,t

P yramidi,t

P ositioni,t

P ositioni,t

-0.056***
(-3.59)
-0.217
(-1.47)
0.077
(1.10)
-9.23***
(-5.09)

-0.049***
(-3.06)
-0.262*
(-1.83)
0.151**
(2.00)

-0.040**
(-2.15)
-0.254*
(-1.82)
0.130*
(1.76)

-0.060***
(-2.66)
-0.169
(-1.61)
-0.019
(-0.39)
-9.19***
(-6.48)

-0.074**
(-2.01)
-0.317
(-0.94)
0.219
(0.89)
-15.75***
(-6.25)

-0.097***
(-2.92)
-0.355
(-1.35)
0.197
(1.44)
-13.94***
(-5.74)

Yes
Yes
Yes
1856
0.28
t
No

Yes
Yes
Yes
1452
0.43
(t 1)
No

Yes
Yes
Yes
974
0.50
(t 1)
Yes

-0.189***
(-4.13)

P ublic
Ln(Assets)
Industry
Y ear
Group
Obs
Adj R2
RHSV ar
Loop = 0

Yes
Yes
Yes
1019
0.30
(t 1)
No

Yes
Yes
Yes
1019
0.26
(t 1)
No

-0.074***
(-5.78)
Yes
Yes
Yes
1019
0.30
(t 1)
No

***0.01, **0.05, *0.1, t-stats are shown in parentheses

Standard errors are clustered at each firm level

This Table shows results of the following regression:

P yramidi,t = a0 + a1 LEVi,t1 + a2 OP ROAi,t1 + a3 CAP EXAi,t1 + 1 Controls + i,t
Columns (1)-(3) report the results when the dummy variable, P yramidi,t , is a dependent variable, and all
independent variables are lagged. The results when the dummy variable, P yramidi,t , is a dependent variable,
and all independent variables are contemporaneous are shown in column (4). Columns (5) and (6) report
the results when the continuous counterpart of P yramidi,t , P ositioni,t , is used as a dependent variable, and
all independent variables are lagged. In column (6), I exclude firms in circular ownership link (i.e., Loop = 0
is imposed), whereas I do not impose the condition in column (5).

59

Table 7: Central vs. Leverage

LEV
OP ROA
CAP EXA
F irmAge

(1)

(2)

(3)

(4)

(5)

(6)

Centrali,t

Centrali,t

Centrali,t

Centrali,t

Centralityi,t

Centralityi,t

0.047***
(2.71)
0.015
(0.13)
-0.082
(-1.04)
12.26***
(9.18)

0.046***
(2.88)
0.050
(0.44)
-0.175**
(-2.11)

0.049**
(2.19)
0.039
(0.34)
-0.175**
(-2.40)

0.058***
(2.98)
0.010
(0.12)
0.007
(0.15)
10.77***
(9.46)

0.004*
(1.93)
-0.022
(-1.55)
-0.011
(-1.52)

0.006***
(2.66)
-0.028*
(-1.70)
-0.003
(-0.39)

0.017***
(5.11)
Yes
Yes
Yes
1437
0.36
(t 1)
No

0.016***
(6.17)
Yes
Yes
Yes
965
0.36
(t 1)
Yes

0.335***
(8.76)

P ublic
Ln(Assets)
Industry
Y ear
Group
Obs
Adj R2
RHS V ar
Loop = 0

Yes
Yes
Yes
1452
0.32
(t 1)
No

Yes
Yes
Yes
1452
0.30
(t 1)
No

0.123***
(12.85)
Yes
Yes
Yes
1452
0.38
(t 1)
No

Yes
Yes
Yes
2590
0.27
t
No

***0.01, **0.05, *0.1, t-stats are shown in parentheses

Standard errors are clustered at each firm level

This Table shows the results of the following regression:

Centrali,t = b0 + b1 LEVi,t1 + b2 OP ROAi,t1 + b3 CAP EXAi,t1 + 2 Controls + i,t
Columns (1)-(3) report the results when the dummy variable, Centrali,t , is a dependent variable, and all
independent variables are lagged. The results when the dummy variable, Centrali,t , is a dependent variable,
and all independent variables are contemporaneous are shown in column (4). Column (5) and (6) report the
results when the continuous counterpart of Centrali,t , Centralityi,t , is used as a dependent variable, and all
independent variables are lagged. In column (6), I exclude firms in circular ownership link (i.e., Loop = 0 is
imposed), whereas I do not impose the condition in column (5).

60

Table 8: Payout vs Central

Industry
Y ear
Obs
Adj R2

Yes
Yes
2590
0.12

Yes
Yes
2590
0.12

Yes
Yes
2590
0.12

Yes
Yes
2590
0.12

(5)
Dividendst
-3.72**
(-2.51)
0.036***
(4.60)
-0.004
(-0.40)
-0.018***
(-4.58)
0.036
(0.28)
0.019***
(5.38)
-9.07
(-0.92)
Yes
Yes
2567
0.81

RHS V ar

Central
OP ROA
CAP EXA

(1)
DIV At
-0.003***
(-3.00)
0.049***
(5.50)
-0.003
(-1.07)

LEV
P ublic

0.001
(0.92)

(2)
DIV At
-0.003***
(-2.90)
0.047***
(5.38)
-0.003
(-1.17)
-0.002*
(-1.72)
0.001
(0.80)

(3)
DIV U N At
-0.004***
(-3.74)
0.047***
(5.29)
-0.003
(-1.14)

0.001
(0.59)

(4)
DIV U N At
-0.003***
(-3.66)
0.045***
(5.14)
-0.003
(-1.24)
-0.002*
(-1.66)
0.001
(0.48)

Centrality(107 )
OrdIncome
Capex
Liabilities
P ublic(107 )
Assets
Assets2 (1014 )

***0.01, **0.05, *0.1, t-stats are shown in parentheses

Standard errors are clustered at each firm level

This Table shows the results of the following regression:

DIV Ai,t (or DIV U N Ai,t ) = c0 + c1 Centrali,t + c2 OP ROAi,t + c3 CAP EXAi,t + c4 LEVi,t + c5 P ublici,t +
3 Controls + i,t
All RHS variables are contemporaneous. In columns (1) and (2), I use DIV At , book value of total dividends
divided by book value of operating assets, as a dependent variable. In columns (3) and (4), I use DIV U N At ,
book value of total dividends divided by book value of total assets. In column (5), I use book value of
total dividends, Dividends, as a dependent variable, and use a continuous variable, Centrality, as a main
independent variable. OrdIncome, Capex, Liabilities, and Assets represent the book value of operating
profits, total capital expenditures, total liabilities, and total assets, respectively.

61

Table 9: Payout vs Pyramid

Industry
Y ear
Obs
Adj R2

Yes
Yes
1856
0.12

Yes
Yes
1856
0.12

Yes
Yes
1856
0.11

Yes
Yes
1856
0.12

(5)
Dividendst
0.118**
(2.31)
0.036***
(3.90)
-0.003
(-0.27)
-0.017***
(-5.40)
-0.028
(-0.28)
0.018***
(7.44)
-7.37
(-0.77)
Yes
Yes
2590
0.80

RHS V ar

P yramid
OP ROA
CAP EXA

(1)
DIV At
0.0028***
(2.22)
0.052***
(4.82)
-0.0023
(-1.02)

LEV
P ublic

0.0001
(0.07)

(2)
DIV At
0.0027**
(2.16)
0.050***
(4.71)
-0.003
(-1.11)
-0.002*
(-1.66)
-0.0002
(-0.12)

(3)
DIV U N At
0.0030**
(2.33)
0.050***
(4.66)
-0.003
(-1.09)

-0.0003
(-0.21)

(4)
DIV U N At
0.0028**
(2.27)
0.048***
(4.53)
-0.003
(-1.19)
-0.003
(-1.64)
-0.0005
(-0.28)

P osition(107 )
OrdIncome
Capex
Liabilities
P ublic(107 )
Assets
Assets2 (1014 )

***0.01, **0.05, *0.1, t-stats are shown in parentheses

Standard errors are clustered at each firm level

This Table shows the results of the following regression:

DIV Ai,t (or DIV U N Ai,t ) = d0 + d1 P yramidi,t + d2 OP ROAi,t + d3 CAP EXAi,t + d4 LEVi,t + d5 P ublici,t +
4 Controls + i,t
All RHS variables are contemporaneous. In columns (1) and (2), I use DIV At , book value of total dividends
divided by book value of operating assets, as a dependent variable. In columns (3) and (4), I use DIV U N At ,
book value of total dividends divided by book value of total assets. In column (5), I use book value of
total dividends, Dividends, as a dependent variable, and use a continuous variable, P osition, as a main
independent variable. OrdIncome, Capex, Liabilities, and Assets represent the book value of operating
profits, total capital expenditures, total liabilities, and total assets, respectively.

Table 10: Summary Statistics of Change in Group Structure Metrics

Variables
DPosition
DCentrality

Mean
0.027
0.002

Stdev
0.350
0.024

Median
0
0

62

25%
-0.023
-0.000

75%
0.047
0.000

Obs.
2370
2350

Table 11: Robustness Check: Pyramid vs. Leverage

Adj R2

Dependent Variable: BCP yramidk

k = 1: Symmetric Thresholds k = 2: Asymmetric Thresholds
(1)
(2)
(3)
(4)
-0.053*** -0.058***
-0.052*** -0.057***
(-5.48)
(-7.01)
(-5.04)
(-6.64)
-0.260
-0.275*
-0.281
-0.305*
(-1.61)
(-1.66)
(-1.60)
(-1.69)
0.230*
0.166
0.245*
0.189
(1.96)
(1.35)
(1.96)
(1.46)
-0.034***
-0.039***
(-2.62)
(-2.74)
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
313
313
282
282
0.21
0.23
0.20
0.23

Cluster Level

Firm

LEVt1
OP ROAt1
CAP EXAt1
Ln(Assets)
Industry
Y ear
Obs

Firm

Firm

Firm

***0.01, **0.05, *0.1, t-stats are shown in parentheses

Standard errors are clustered at each firm level

This Table shows the results of the following regression:

BCP yramidki,t = f0 + f1 LEVi,t1 + f2 OP ROAi,t1 + f3 CAP EXAi,t1 + 5 Controls + i,t
The dependent variable, BCP yramidki,t , is a dummy variable taking value of 1 if a given firm i
increases its P osition value from year t 1 to year t more than a given threshold level and becomes
a pyramidal subsidiary at year t. It takes a value of 0 if a given firm i decreases its P osition value
from year t 1 to year t more than a given threshold level and becomes a stand-alone entity at
year t. See Section 5.4.1 for more details of the definition of the threshold values that are used
to define BCP yramidi,t . Columns (1) and (2) show the results when I use the average P osition
change as thresholds for both increase and decrease of P osition. Columns (3) and (4) show the
results when I use the 25 percentile of P osition change as a threshold for P osition decrease, and
the 75 percentile of P osition change for P osition increase.

63

Table 12: Robustness Check: Central vs. Leverage

LEVt1
OP ROAt1
CAP EXAt1
F irmAget1
P ublict1
Ln(Assets)t1
Industry
Y ear
Obs
Adj R2

Dependent Variable: BCCentrali,t

(1)
(2)
(3)
(4)
0.174** 0.167** 0.185**
0.156*
(2.05)
(1.91)
(2.15)
(1.82)
0.173
0.204*
0.202*
0.194*
(1.58)
(1.85)
(1.81)
(1.66)
0.001
0.018
-0.006
-0.037
(0.01)
(0.20)
(0.06)
(-0.36)
2.54**
(2.05)
0.097***
(3.22)
0.032***
(3.39)
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
369
369
369
369
0.13
0.14
0.15
0.16

***0.01, **0.05, *0.1, t-stats are shown in parentheses

Standard errors are clustered at each firm level

This Table shows the results of the following regression:

BCCentrali,t = g0 + g1 LEVi,t1 + g2 OP ROAi,t1 + g3 CAP EXAi,t1 + 6 Controls + i,t
The dependent variable, BCCentrali,t , is a dummy variable taking value of 1 if a given firm i
becomes a new central firm at year t by increasing its centrality more than a given threshold level
from year t1, and 0 otherwise. See Section 5.4.2 for more details of the definition of BCCentrali,t .
Test samples are restricted to a condition that P osition of a firm should be less than equal to 2.0.
The regression focuses on the firms within the top-tier of the business group.

64