Family Business Groups around the World: Costs and Benefits of Pyramids*

Ronald W. Masulis

Owen Graduate School of Management Vanderbilt University

Phone: +1 615-322-3687 ronald.masulis@owen.vanderbilt.edu

Peter Kien Pham

Faculty of Economics and Business University of Sydney

Phone: +61 2-9036-9103 p.pham@econ.usyd.edu.au

Jason Zein

Australian School of Business University of New South Wales

Phone: +61 2-9385-5875 j.zein@unsw.edu.au

September 16, 2009

* The authors would like to thank Heitor Almeida, Bill Christie, Espen Eckbo, Joseph Fan, Jie Gan, Ernst Maug, Randall Morck, Vikram Nanda, Chander Shekhar, Linus Siming, Gloria Tian, Zhaoxia Xu, conference participants at the 2008 Australian National University Summer Research Camp, the 2008 Duke-UNC Corporate Finance Conference, the 2008 HKUST Summer Symposium of Family Business Research, the 2008 Monash University Executive Compensation and Corporate Governance Symposium, and the 2008 Washington University at St. Louis Fifth Annual Corporate Finance Conference, the 2009 European Financial Management Symposium on Corporate Governance and Control at the University of Cambridge, the 2009 Financial Management Association European conference, and seminar participants at the Hong Kong Polytechnic University, the New University of Lisbon, and the Stockholm School of Economics for valuable suggestions. We are also very grateful to Karl Lins and Yupana Wiwattanakantang for sharing their emerging market and Thai group data for verification purposes, and to Suk-won Kim, Daejin Kim, Gloria Tian, and Xiaowei Xu for the help in tracking down information on Korean and Canadian groups. All errors remain our own.

Family Business Groups around the World: Costs and Benefits of Pyramids

Abstract: Using a comprehensive dataset of 28,039 firms in 45 countries, we investigate the motivations for family-controlled business groups. Contrary to expropriation theories, we find that access to capital plays a critical role in explaining group prevalence at the country level, and that within individual groups, internal equity funding, investment intensity, and firm performance all increase down a pyramidal chain, reflecting the funding advantages that pyramids offer. Controlling for endogeneity in group affiliation shows that certain firm types realize significant benefits from business group membership. Our results highlight important funding and certification support that family groups provide to their members.

1

Concentrated ownership in the hands of wealthy families or individuals is a common feature

among listed firms in a large number of economies around the world. To expand their businesses

while maintaining control, these families or individuals often form business groups where

multiple member firms are connected through direct, pyramidal, and/or cross-holding ownership

ties to a common ultimate owner.1 Such conglomerate structures have been an important and

persistent phenomenon associated with capital market development in much of the world.2 Yet,

despite their prevalence and longevity, international evidence on the functions and characteristics

of family business groups remains sparse.

This study is the first systematic examination of family-controlled business groups around

the globe.3 Utilizing a new hand-collected dataset of 875 family groups controlling 2,763 publicly

listed firms in 45 countries, we first characterize country-level environments that are conducive

to group formation. We then map out the organizational structure of each family group in our

sample. This allows us to analyze whether decisions related to the placement of firms within a

pyramidal group are more consistent with facilitating shareholder expropriation or access to

capital. Finally, we explain aggregate performance differences between group and non-group

firms, taking into account potential endogeneity in group affiliation. Across the various lines of

analysis, the empirical evidence consistently indicates that the continuing economic importance

of family groups stems from the valuable funding and certification support that they provide to

member firms.

1 Pyramids refer to hierarchical structures whereby a group uses one of its listed subsidiaries to hold a controlling ownership position in another listed subsidiary. Cross-holdings refer to arrangements whereby firms in the same group hold large reciprocal ownership stakes in one another. 2 In the US, pyramidal groups were popular until the introduction of double-taxation of inter-corporate dividends ended their expansion in the 1930s (Morck (2005)). Around the world, some of the largest firms in many countries are controlled by family business groups, such as Fiat (Italy), Ford (US), Hutchison Whampoa (Hong Kong), News Corp (Australia), Overseas Chinese Banking Corp (Singapore) and Samsung (South Korea). Further, many family groups have a long history. The Jardine Matheson group (Hong Kong/Singapore) and the Bolloré family group (France), for example, were established more than 150 years ago. 3 For brevity, we use ‘family business groups’, ‘family groups’ or simply ‘groups’ to refer to ‘family-controlled business groups’.

2

Our findings offer new empirical perspectives on the relative balance between the

hypothesized costs and benefits of business groups (see Morck, Wolfenzon, and Yeung (2005)

and Khanna and Yafeh (2007) for a review). Thus far, the international corporate governance

literature has disproportionately focused on control motivations behind business groups. In

particular, the pyramid structure that groups often employ is viewed as a mechanism to preserve

private benefits of control for the ultimate controlling shareholders without necessitating a

commensurate capital contribution, which results in a separation of cash-flow rights from control

rights (La Porta, Lopez-de-Silanes, and Shleifer (1999)). Johnson, La Porta, Lopez-de-Silanes,

and Shleifer (2000), Bae, Kang, and Kim (2002), Bertrand, Mehta, and Mullainathan (2002), Joh

(2003), and Baek, Kang, and Lee (2006) suggest that this separation can encourage and facilitate

expropriation of minority shareholders through various tunnelling activities. At a broader level,

Morck, Wolfenzon, and Yeung (2005), Stulz (2005), and Fogel (2006) argue that groups can

leverage the disproportionately large amount of assets under their control to cultivate powerful

political influence. This can enable groups to insulate themselves from external market discipline

and to acquire economic dominance at the expense of institutional and capital market

development.

Beyond control and expropriation motives, a few studies have also highlighted important

benefits of business groups. Khanna and Palepu (2000) suggest that group reputation substitutes

for underdeveloped legal and regulatory mechanisms that leave minority investors vulnerable to

expropriation risks and information asymmetries. Almeida and Wolfenzon (2006a) argue that

business groups, especially those structured as pyramids, possess a financing advantage created

by their internal capital markets, which are better able to support the funding requirements of new

firms operating in underdeveloped external capital markets. According to Hoshi, Kashyap, and

Scharfstein (1991), Khanna and Yafeh (2005), and Gopalan, Nanda, and Seru (2007), such

3

internal capital markets also facilitate risk-sharing and intra-group financial support as a way to

overcome external capital constraints and the associated risk of financial distress.

While the overall impact of family business groups remains a matter of academic debate,

it is important to note that despite expropriation risks, minority shareholders continue to co-invest

alongside controlling families. For instance, we find that on average, 16 percent of listed firms in

each sample country belong to a family group, rising to over 40 percent for emerging markets

such as Sri Lanka, Chile, the Philippines, and Turkey. Further, group firms are on average 12

years older than their unaffiliated peers. In addition to the contrasting arguments in the literature,

these simple observations also motivate us to investigate whether business groups survive and

expand because they offer sufficient funding and certification support to compensate for the

perceived costs associated with private benefits of control.

Utilizing our broad sample of 45 countries, we first identify country-level characteristics

associated with the prevalence of family groups across markets. While groups can emerge in

different countries as a result of unique historical circumstances (Morck and Steier (2005)), this

analysis allows us to gauge the relative importance for group development of a common set of

country-level factors previously discussed in the literature, categorized into those related to

private benefits of control, to poor access to external capital, and to other regulatory constraints.

Overall, we find that measures of external capital availability show stronger negative associations

with the prevalence of family groups than the strength of mechanisms aimed at curbing private

benefits of control. This finding highlights the importance of the funding advantage created by a

group’s internal capital, especially when access to external capital is limited.

Our main line of analysis is conducted at the firm level. Utilizing a detailed procedure to

identify group membership, structure, and ownership characteristics, we are able to examine how

firm attributes vary with their position within a group and with group affiliation in general. This

4

analysis allows us to assess the relative importance of control motivations and access to capital

concerns when structuring a group. The first observation from a general comparison of group and

non-group firms is that group firms do not exhibit characteristics that would identify them as

being potential targets for shareholder expropriation by controlling families. In particular, group

firms tend to be well-established, pay more dividends, and receive greater analyst coverage than

their non-group counterparts.

One of our most striking findings is that within each group, firms at the bottom of a

pyramidal chain appear to be the least likely targets of expropriation, despite predictions that the

controlling shareholder’s low cash-flow rights and high control rights in these firms leave them

vulnerable. Instead, these firms exhibit higher growth prospects and capital requirements than

those at the top or in the middle of business groups. This indicates that the bottom tier of a

pyramid is populated by capital-intensive, low free-cash-flow firms, where the scope for

consumption of private benefits of control is more limited. Equally striking is our finding that

firm performance measures increase down a pyramidal chain, where the divergence of cash-flow

from control rights is actually more pronounced. These findings are consistent with Almeida and

Wolfenzon’s (2006) proposition that firms in a pyramidal group possess a financing advantage

over those in a horizontal structure because the former can access the retained earnings of their

parents. In contrast, other control-enhancing mechanisms, such as shares with differential voting

rights, do not offer the same financing advantage. We find that after controlling for pyramid layer

positions, group firms using dual-class shares indeed suffer from a valuation discount.

In further investigating the funding and certification roles of groups, we find that they

retain larger direct ownership stakes in bottom-tier firms than in other group firms (such as

holding companies at the apex of the group). For a member firm, we show that it is the direct

shareholding (and voting rights) by its immediate parent(s), rather than the cash-flow rights of the

5

family, that is positively related to firm value. In interpreting these findings, we argue that when a

new group member is established, the upper echelons of the ownership chain(s) connecting it to

the family are already configured in an optimal manner, and thus, are difficult to alter. This

leaves the link(s) between the new member and its immediate parent(s) as the group’s primary

ownership decision. Thus, a large direct ownership position reflects the group’s financial

commitment to its new member and reduces the costs of raising external capital to satisfy the

latter’s substantial funding needs. Such funding also allows the group to obtain significant control

rights, so as to generate a certification effect that could potentially offset the large moral hazard

risk faced by outside investors in markets with limited shareholder protection. This is analogous

to the rationale for giving superior control rights to venture capitalists.4

Our findings regarding the importance of internal group capital to member firms motivate

a final empirical question: does group membership in aggregate raise the value of a firm vis-à-vis

other firms in the market? Based on ordinary least squares (OLS) regressions, group firms on

average have lower Tobin’s Q and return on assets (ROA) than non-group firms. At first glance,

this finding is consistent with a negative impact of private benefits of control documented in

some prior studies (Claessens, Djankov, Fan, and Lang (2002)). We observe that this negative

relation may also arise due to an endogenous selection bias, as firms with high-value projects

may be able to overcome external funding constraints on their own, enabling them to remain

independent. In contrast, firms with lower-value projects (or those facing high information

asymmetry) can find it impossible to obtain external funding independently, and thus, are more

likely to be funded within a group structure.

After taking into account endogenous selection, we find that the valuation discount

associated with group firms is indeed partly due to endogeneity. For certain group firms, group

4 See Gompers and Lerner (1996) and Kaplan and Stromberg (2003) for discussions of control rights held by venture capital investors.

6

membership actually has a positive effect on their value (Tobin’s Q), but not on their cash-flow

generating abilities (ROA). Consistent with our earlier findings, this evidence suggests that group

structures help alleviate financing constraints for those firms possessing traits that make them

originally more likely to be part of a group. In other words, these firms would experience an even

larger valuation discount if they remained independent. This finding does not imply that minority

shareholders are unconcerned about potential losses from consumption of private benefits, but

rather that for certain types of member firms, such concerns can be outweighed by a group’s

funding and certification benefits.

Overall, our findings contribute to a growing body of evidence on the importance of a

business group’s internal capital market. Gopalan, Nanda, and Seru (2008) find that dividend

payments are used as a means of internally redistributing funds within business groups in Asia

and Europe. Almeida, Park, Subrahmanyam, and Wolfenzon (2008) report that in Korean

business groups, firms with high investment requirements are more likely to be in pyramids. In

terms of existing theories, we find support for Almeida and Wolfsenzon’s (2006) proposition that

beyond facilitating control, placing firms in pyramids provides a financing advantage, because

these group firms can tap into a larger pool of retained earnings. Our results are also consistent

with the views of Gomes (2000), Khanna and Palepu (2000) and Gopalan, Nanda, and Seru

(2007) that reputation is a key element in attracting funding in the presence of weak investor

protection. More broadly, our evidence emphasizing the importance of internal capital to family

business groups echoes Stein’s (1997) explanation for why large conglomerates exist, namely

that they play a vital role in allocating scarce capital in the presence of information asymmetries.

The rest of the paper is structured as follows. Section I describes our ownership data and

group construction procedures. Section II discusses the association between various country-level

factors and group prevalence. Section III examines group firm characteristics and performance

7

differences between group and non-group firms. Finally, section IV summarizes our conclusions.

I. Ownership Data and Group Construction

A. Data Sources

To construct business groups, we begin by collecting ownership data for a sample of firms

in 45 different countries from two primary data sources, namely the Osiris database from Bureau

Van Dijk and the Worldscope database from Thomson Reuters. Despite the use of multiple

databases, ownership information is missing for a number of firms. This introduces biases if a

country only has a small number of listed firms and if firms with missing data are potentially

important links within some groups. To minimize such biases, we manually search for and collect

ownership data on these firms from Dunn and Bradstreet’s Who Owns Whom database, Thomson

Reuters’ OneSource database, and from various data providers in LexisNexis (for example, The

Major Companies Database) and Factiva (for example, the Taiwan Economic Journal database

of Asian companies). In addition, ownership information is obtained from stock exchange and

securities regulator websites (for example, Belgium, Chile, Colombia, India, Indonesia, and Italy)

and directly from company annual reports extracted from the Standard and Poors’ Mergent

Online database, stock exchange websites, and company websites (this constitutes a large portion

of ownership information for firms in Argentina, Israel, Malaysia, Mexico, Pakistan, Sri Lanka,

Singapore, and Thailand). This array of data sources allow us to fill in holes in the ownership

structure of firms and check for inconsistencies. Due to the large sample size and the complexity

of ownership identification, we focus on ownership data for 2002. However, for a small number

of firms without 2002 ownership data, we use the earliest ownership information available in the

2003-2006 period. In total, ownership information is available for 28,039 sample firms.

8

B. Group Construction Procedures

Based on the above ownership data, the first stage of the group construction process is to

distinguish between widely held firms and controlled firms. As a starting point, we identify the

controlling shareholder of a firm as the largest shareholder, who effectively controls (through

personal holdings or those of affiliates) at least 20 percent of a firm’s voting rights. We lower this

threshold to 10 percent if the largest shareholder also has other forms of control such as being a

firm’s founder, CEO, or chairman of the board.5

In a majority of firms, the largest controlling shareholder is clearly visible. However,

complications arise when shareholdings of the controller spread out across a number of entities

such as private holding companies, family members, trusts, and other group firms, none of which

constitute a controlling stake on their own. Therefore, we carefully examine shareholder data of

firms with no clear controlling stakes, but with a large number of share blocks, to identify if these

blocks actually belong to or are controlled by the same owner.6 This manual search is limited to

markets where beneficial ownership disclosure is not mandatory. In particular, some company

annual reports and stock exchange (and regulator) websites provide information on whether

fragmented block holdings actually belong to the same controlling shareholder. We also search

for the names of common block holders in shareholder lists of other firms in the same country

that have a clearly visible controlling shareholder, as this often allows us to trace their commonly

used holding vehicles. Finally, we painstakingly search for information to identify holders of the

remaining unknown fragmented blocks of at least five percent of shares, using LexisNexis,

Factiva and the Google search engine.

Once control is established, the second stage seeks to find the ultimate controlling

5 There are a few cases where a sample firm is reported to be ‘effectively’ controlled by the founding family through executive and board positions, but the family has divested its interests to below 10 percent of voting rights (for example, the Banco Santander group in Spain). For consistency, these firms are still considered as widely held. 6 In many cases the procedure is straightforward as individual large block shareholders have the same surname, or can be quickly identified as the spouse of the controller.

9

shareholder and determine whether they are a family (defined as both families and individual

entrepreneurs) or a non-family entity (defined as governments, other widely-held firms, and

widely-held financial institutions). For each sample firm, if the controlling shareholder is a public

company, we proceed to investigate its ownership to determine the company’s controlling

shareholder (if there is one), and so on, until we reach the ultimate owner at the top of the chain

of control. If at any stage, the controlling shareholder is a private company, we also investigate

the identity of the owner of this entity, primarily through the Who Owns Whom database, the list

of subsidiaries of other public companies (available in Osiris), the annual report (or website) of

the sample firm being investigated, and related media articles and analyst reports accessed via the

Google search engine or Factiva, LexisNexis searches.7,8 In some cases, we find that a firm is

ultimately owned by an alliance of businessmen (such as the GEA Group of Colombia and the

Shrem Fudim Group of Israel) or by a coalition of families (such as the Boel, Solvay, and Janssen

families in Belgium). In both cases, we classify these firms as family-controlled. When two or

more firms in the same market share the same ultimate controlling family, they are defined as

belonging to a family-controlled business group.9,10

Once a family group is identified, the third stage involves re-examining the ownership of

each member firm to identify whether other non-controlling stakes of the firm also belong to the

ultimate controlling family and other firms in the same group. The identification procedure is the

same as the one used in the second stage and is applied to all ownership stakes in excess of five

7 In many cases, we can identify the ownership structure of the private company (such as when it is reported that the private company is a 50/50 partnership). Otherwise, we assume that the private company is 100 percent owned. 8 In a small number of cases where we cannot identify the individuals behind a private company, but if this company controls multiple firms, we assume that it is also controlled by a family or an individual. 9 A few studies examine broader categories of business groups (Hoshi, Kashyap, and Scharfstein (1991), Perotti and Gelfer (2001)). We focus on family-controlled groups not just because they are the most popular, but because other groups may not possess the same reputation incentives and control motivations. For example, Japanese Keiretsus have often been criticised as being “headless bodies” because there is no dominant shareholder who bears responsibility for the actions of the group members and who values the private benefits derived from their control. 10 If two firms with the same controlling family are listed in different national markets, we do not consider them to be part of a business group. In our sample 87 firms are not counted as group firms because of this restriction. However, most of our results remain the same when these firms are included in the sample.

10

percent of issued shares. Shares with differential voting rights are also identified using share class

information from Worldscope, Datastream, Mergent Online, and company annual reports.

The final stage of the group construction procedure is to ensure that the identified group

structures are as complete and accurate as possible by verifying our sample of groups using

independent sources. In particular, we check the richest people lists from Forbes Magazine (and

various country-specific sources) and the world’s largest family firms list from Family Business

Magazine to ensure that wealthy individuals and families, who often own a business group, are

accounted for. Our group information is also verified against (public and proprietary) data from

various country-specific studies/sources: Rabelo and Vasconcelos (2002) for Brazil, Morck,

Percy, Tian, and Yeung (2005) and the Inter-Corporate Ownership publication for Canada,

Majluf, Abarca, Rodriguez, and Fuenes (1998) for Chile, Fisman (2000) and Sato (2004) for

Indonesia, Kosenko (2007) for Israel, business group data from the Korean Fair Trading

Commission website for Korea, Shimizu (2004) for Peru, the Taiwan Economic Journal database

for Taiwan, Polsiri and Wiwattanakantang (2006) for Thailand, Lins (2003) for emerging

markets, Claessens, Djankov and Lang (2000) for East Asian countries, and Faccio and Lang

(2002) and UBS Investment Research (2006) for Western European countries. Finally, our

verification process is aided by the fact that the entire structures of many groups are disclosed on

their websites or through media articles. In total, we identify 875 family-controlled groups,

comprising of 2,763 publicly listed firms from 45 countries.

Our study utilizes a more comprehensive dataset than most previous research into cross-

country ownership structures. To our knowledge, this is the first study of the prevalence and

structures of family groups across five continents, and including both developed and emerging

markets. For example, Claessens et al. (2000) and Faccio and Lang (2002) focus on nine East

Asian and 13 Western European countries, respectively. La Porta et al. (1999) examine the 27

11

richest economies, but cover only the 20 largest firms in each country. In addition to examining

more countries, we also have more extensive coverage of firms in most of these countries.

C. Group Placement and Ownership Structure Measures

For each group in the final sample, we map out its structure and calculate the position,

ownership concentration, and the ultimate cash-flow rights and control rights of the controlling

family for all member firms. Figure 1 illustrates the calculation of these measures using a typical

family business group in the sample: the Arab Malaysian group of Malaysia, ultimately

controlled by Azman Hashim through a pyramid structure.

[INSERT FIGURE 1 HERE]

The first measure is the layer position of each firm in a group (Pyramid Layer). In

particular, we count how many listed firms exist above each group firm. When a firm is

controlled through multiple ownership chains, Pyramid Layer is determined based on the chain

associated with the largest shareholding. In the above example, the group position measure is

three for Global Carriers (by following the chain associated with the largest shareholding, which

is the 37 percent stake held by AMFB Holdings), two for AMFB Holdings, one for AMMB

Holdings, and zero for Arab Malaysian Corporation.11

The second measure is the total direct percentage ownership (Direct Ownership) that a

group holds in a member firm. For example, Azman Hashim controls Global Carriers through

two chains: (i) the 37 percent stake held by AMFB Holdings, and (ii) the 12 percent stake held by

11 The case of Global Carriers illustrates a complication that arises when a firm is controlled by two firms at different positions in the pyramid. For robustness purposes, we adopt an alternative weighted average positioning measure developed by Almeida et al. (2008). This measure adjusts for multiple chains of control by weighting the Pyramid Layer value of each chain by the ultimate cash-flow rights that the chain provides to the ultimate owner. These values are then summed to obtain the final weighted position. In our example, Global Carriers is at layer three (underneath AMFB Holdings) and provides 2.8 percent (0.35×0.34×0.64×0.37) of its cash-flow rights to the ultimate owner through this chain. Global Carriers is also at layer two (underneath AMMB Holdings), where it provides another 1.4 percent (0.35×0.34×0.12) of its cash-flow rights to the ultimate owner. Its weighted average position is therefore (2.8/4.2)×3 + (1.4/4.2)×2 = 2.66. However, we do not report the results related to this formulation as they are similar to those obtained from the Pyramid Layer variable.

12

AMMB Holdings. The Direct Ownership measure for Global Carriers is thus 37 + 12 = 49

percent. Under a one-share-one-vote regime, this measure reflects both the degree of control of a

group over a member firm and the equity contribution from its internal capital market. However,

as group firms may employ shares with differential voting rights, we also calculate an alternative

direct ownership measure based on percentage of voting rights (Direct Control Rights).12

The third measure is the ultimate cash-flow rights of the controlling family (Ultimate CF

Rights) in each group firm, defined as the former’s total claim on each dollar of earnings

generated by the latter. This is calculated by aggregating the cash-flow rights across all of the

ownership chains through which a family maintains its control. The cash-flow rights of each

chain are the product of all percentage shareholdings connecting the firms along the chain. In the

Arab Malaysian group, the ultimate cash-flow rights of Azman Hashim in Global Carriers are

(0.35×0.34×0.64×0.37) + (0.35×0.34×0.12) = 0.042 or 4.2 percent.

Cross shareholdings present a complication in the calculation of these measures. Consider

the following example. Ultimate owner A owns 40 percent of Firm B, B in turns owns 30 percent

of Firm C and 20 percent of Firm D. In addition, C and D own 20 percent in each other, creating

a cross-holding layer. In such a case, the cross shareholding between two group firms means that

each effectively owns some shares in itself. To exclude the effect of such circular ownership, we

rescale our ownership and ultimate cash-flow rights measures by the percentage of shares not

self-owned by cross-holding firms.13 It should be noted that cross shareholdings are even more

12 As a robustness check, we also measure the strength of control that the ultimate owner retains along and across various ownership chains, following La Porta et al. (1999) and Claessens et al. (2000). Along each chain, we take the smallest observed percentage voting interests (the weakest link) as the control rights of the chain. Where multiple control chains exist, the total control rights are the sum of the weakest links across all chains. This definition reflects the risk that the ultimate owner may lose control of a shareholding chain at the point where their voting rights are the lowest. In the Arab Malaysian example, the weakest-link control rights of Global Carriers are equal to: 34 (weakest link along the first chain) + 12 (weakest link along the second chain) = 46 percent. However, we do not report the results related to this formulation as they are similar to those obtained from the Direct Control Rights variable. 13 For example, Firm C itself has an interest in 0.2×0.2 = 0.04 or 4 cents of every one dollar of earnings that it generates. This effectively means that each shareholder of Firm C has a share of cash-flow rights that is 1/(1-0.04) times more than the level implied by his or her nominal percentage shareholding.

13

relevant to the control rights calculation as they give the ultimate owner increased voting rights in

both firms through the exchange of reciprocal voting blocks. Table I summarizes the definitions

for our main ownership and control measures, as well as other variables used in our analysis.

[INSERT TABLE I HERE]

D. Country-Level Descriptive Statistics

We employ several country-level measures of the economic importance of business

groups. The first is the percentage of all listed firms in a national market that belong to family

business groups (% Group). The second measure is the percentage of all listed firms in pyramid-

controlled family groups (% Pyramid), which focuses attention on the use of pyramids as a

means of structuring family business groups. We also calculate % Group MC and % Pyramid

MC, which are value-weighted averages of % Group and % Pyramid, respectively, based on their

percentages of aggregate stock market capitalization in a country.14

Table II presents our country-level group statistics. Not surprisingly, they reveal that

family business groups are more important in emerging markets. For example, the proportion of

listed firms belonging to family business groups (% Group) is at least 30 percent in Chile,

Colombia, Israel, Philippines, Sri Lanka, and Turkey, with Sri Lanka being the largest at 64

percent. The trend is similar for group firms controlled through pyramids (% Pyramid). However,

there are some important disparities between these two statistics, indicating that pyramids are a

less popular means of structuring business groups in some markets than others. For example, the

proportion of listed firms controlled by groups is 29 percent in India and 15 percent in Pakistan,

but only about one third of these firms are structured as pyramids. In contrast, the proportion of

pyramid-controlled group firms relative to all group firms is much larger in Colombia and Sri

Lanka (29 percent out of 39 percent and 50 percent out of 64 percent, respectively).

14 We are unable to obtain market capitalization information for approximately 1.8 percent of our group firms.

14

[INSERT TABLE II HERE]

While family business groups may only control a few firms in some markets, they can

often be very large firms that play a key role in the economy, as captured by the market

capitalization based measures (% Group MC and % Pyramid MC). Most notably, East Asian

countries display significant disparities in terms of the relative number of group firms and their

importance as a fraction of aggregate market value. For example, only 11 percent of listed firms

in Singapore belong to family groups, but they represent 43 percent of total market capitalization.

It is also important to note that in some markets, a single family can be in control of a

significant fraction of aggregate stock market capitalization. For example, Chile’s Angelini

family, Colombia’s Grupo Empresarial Antioqueño (GEA), Denmark’s A.P. Moeller-Maersk

group, Italy’s Tronchetti Provera family, Korea’s Lee Gun-Hee, Mexico’s Carlos Slim Helú, the

Philippine’s de Ayala family, Thailand’s Thaksin Shinawatra and Turkey’s Sabanci family all

control over 10 percent of their respective country’s stock market capitalization. In contrast, in

more developed markets such as the US, UK and Japan, the largest family business group owns

less than 2 percent of stock market capitalization.

While the above measures describe the frequency (breadth) of group and pyramid firms,

they do not indicate the extensiveness (depth) of individual pyramid groups in a country. Thus,

we also report the average number of layers (measured by the firm-level Pyramid Layer variable)

observed across all group firms in a country. This statistic shows that the depth and complexity of

pyramids can vary substantially across countries (countries with the highest average Pyramid

Layer are Colombia, Israel, France and Sri Lanka). It is noteworthy that the depth of pyramid

groups can also differ greatly across countries. Pyramid firms make up approximately 20 percent

of market capitalization in Colombia, Indonesia, and Malaysia, but the average Pyramid Layer

differs substantially across these countries (1.55, 0.87, and 0.68, respectively).

15

II. Country-level Analysis

The business group literature suggests that country-level environments contribute

significantly to business group development around the world. However, Morck, Wolfenzon, and

Yeung (2005) state that “…empirical work is needed to solidify hypotheses in this area and to

distinguish presumptions from facts”. In this section, we utilize cross-country variations in our

sample to examine the association between country-level factors and the prevalence of family

business groups. It should be noted that because of the longevity and potential influence of

groups over time, it is difficult to make causal inferences on the roles of specific country-level

macroeconomic and institutional variables. We are however able to document the strength of

association between certain country-level factors and the importance of business groups. The

absence of any significant associations with particular hypothesized factors would cast doubt on

their importance.

A. Selection of Country-Level Variables

A.1. Private Benefits of Control

Perhaps the most cited explanation for the prevalence of family business groups is that

they facilitate the extraction of private benefits of control. In its most serious form, group control

enables the ultimate owner to tunnel resources out of one firm and into another at the expense of

minority shareholders of the former.15 We measure private benefits of control within a country by

its average block premium (Block Premium) paid for purchases of large share blocks as estimated

by Dyck and Zingales (2004). Since block premia are only available for 33 of our 45 sample

countries, we also use individual country-level determinants of private benefits of control

identified by Dyck and Zingales (2004). These determinants include legal mechanisms such as

15 Morck, Wolfenzon, and Yeung, (2005) suggest that group control can also increase private benefits because controlling families can acquire significant political power, which can be used to entrench or strengthen their control over corporate assets or to solicit political favors.

16

the extent of shareholder rights, legal enforcement of rights, and the quality of corporate

disclosure (these are aggregated into a single variable, Governance Index, using weights from a

principal component analysis), and extra-legal mechanisms such as product market competition

(Competition) and reputation penalties, proxied by newspaper circulation (Newspaper).16 Table I

provides the data sources as well as more detailed descriptions of these variables.

A.2. Access to Capital

In addition to control motivations, we examine whether family business groups are more

popular in markets where access to external capital is restricted. Some aspects of the accessibility

of capital, however, are linked to the average level of private benefits of control (Dyck and

Zingales (2004)). In particular, we avoid using direct measures of equity market development as

proxies for access to external capital (for example, the relative size of the stock market or the rate

of initial public offerings), as these variables are found in prior studies to be related to the scope

for private benefits of control and shareholder rights protection (La Porta, Lopez-de-Silanes,

Shleifer, and Vishny (1997)). Rather, we consider measures that capture the pool of potentially

available external capital funding in a country, which are less likely to be determined by the level

of private benefits of control. First of all, we use the natural logarithm of gross domestic product

(GDP) per capita (Log GDP per Capita) as a summary measure of access to capital. Poorer

countries, characterized by low accumulated wealth and underdeveloped financial sectors, have

more restricted pools of funds available for corporate investment. Alternatively, we use several

more narrowly defined measures of capital availability. The first is savings scaled by GDP

(Savings to GDP), as saving intensity is viewed as an important factor that promotes the efficient 16 The extensiveness of shareholder rights protection is proxied by the Anti-director Rights index (as constructed by La Porta et al. (1997) and updated by Pagano and Volpin (2005)). We also use both the Anti-self-dealing index from Djankov et al. (2008), and a revised Anti-director Rights index constructed by Spamann (2008) as alternative shareholder rights measures. Similarly, we also use several alternative measures for legal enforcement, such as the Property Rights Protection index (from the Heritage Foundation and the Wall Street Journal) and Judicial Efficiency (from the International Country Risk Guide), and for financial disclosure, the Accounting Disclosure index from Bushman, Piotroski, and Smith (2004). As an alternative construction of Newspaper, we use the Press Freedom index from Reporters without Borders. All alternatives provide qualitatively similar results.

17

supply of capital (see Pagano (1993)). The second is the extent of capital supplied by the

financial system. We measure this using the amount of equity capital from institutional investors,

calculated as total equity invested (both locally and internationally) by domestic banks, insurance

companies, pension funds and mutual funds scaled by domestic stock market capitalization

(Institutional Funds) obtained from Li, Moshirian, Pham, and Zein (2006). Finally, we consider

exogenous shocks that threaten capital supply caused by a nation’s political environment.

Political instability can increase sovereign risk, which can restrict the amount of capital provided

by both domestic and international investors. We obtain a measure of political stability from the

World Bank’s Governance Matters database (Political Stability).

A.3. Other Regulatory Factors

Business group development can be restricted by several other forms of regulation.

Among these, we consider accounting consolidation rules that govern how a parent can treat its

subsidiaries for financial reporting and tax purposes. Consolidation rules could discourage group

formation by treating member firms as separate taxable entities, thereby restricting tax-

minimizing offsets achieved through the consolidation of profits and losses within a group. We

measure this effect by an indicator variable (Consolidation), which equals one if a country allows

a parent firm to consolidate a subsidiary in which it has an ownership stake of less than 90

percent, and zero if consolidation is only allowed for ownership levels of 90 percent or above.17

In the absence of consolidation, an important attraction for organizing firms as business

groups is the ability to minimize overall tax liabilities by shifting profits, assets, and liabilities

within a group through various intra-group transactions. We measure the extent to which a tax

system controls and monitors such transactions through the use of an index (Intra-Group Tax)

17 Morck (2005) points out that the double-taxation of inter-corporate dividends, introduced by the U.S. government in the 1930s, was explicitly aimed at repressing business groups. However, based on information from the Deloitte International Taxation Guide, there are only three countries in the sample that have some form of double taxation of inter-corporate dividends. Our empirical analysis thus excludes this factor due to insufficient cross-country variation.

18

that aggregates four key aspects of inter-company tax regulations in each country as reported in

the Deloitte International Taxation Guide (see more detailed descriptions in Table I).

Finally, we examine restrictions on one important mechanism through which business

groups are formed, namely through a partial acquisition. Such transactions allow a group to gain

control of another firm, without having to conduct a full takeover. However, when takeover law

dictates that all shareholders of a target firm must be treated fairly and equally, group expansion

may be discouraged as it is costly and difficult to execute a partial acquisition. Such law may also

introduce complications in negotiating a block purchase from a controlling shareholder, making

the deal less likely. For example, it may be difficult to gain a controlling, but partial ownership

stake if a country’s takeover law specifies a low ownership threshold beyond which a tender offer

to all shareholders is required. We use an index of takeover rules (Takeover Index) constructed by

Nenova (2006) for 50 countries as a proxy for restrictions on conducting partial acquisitions.18,19

B. Country-Level Results

Table III reports regression estimates using % Group and % Pyramid as alternative

dependent variables. For the dependent variable % Group, Model (1) reports the regression

estimates involving the summary measures of private benefits of control (Block Premium) and

access to capital (Log GDP per Capita). The coefficient of Log GDP per Capita dominates that

of Block Premium in the regression, indicating that the prevalence of groups is associated more

with a “poor-country” effect than the scope for extracting private benefits of control. In Models

(2) to (4), for both the % Group and % Pyramid regressions, we replace the summary measures

(Block Premium and Log GDP per Capita) with more specific factors that determine the scope 18 Due to the index being unavailable for Venezuela, the sample size for our country-level regression drops to 44 countries, but this omission does not influence the results with respect to other country-level variables. 19 We also include the natural logarithm of GDP (LogGDP) in the regression to control for size differences across market that may create biases when aggregating firm-level data. Holderness (2008) argues that to account for such biases and other missing variable concerns when aggregating firm-level ownership measures to the country level, regression analysis should also be conducted at the firm level, adding unstacked country-level factors as explanatory variables. We address this issue further in Section III.E.

19

for private benefits and access to capital as explanatory variables. However, among the

determinants of private benefits (Governance Index, Competition, and Newspaper), we cannot

include the first two jointly in the same regression since they are highly correlated (correlation of

-0.822).20 The measures of capital availability (Savings to GDP, Institutional Funds, and Political

Stability) are also strongly correlated and they are tested in separate models.

The most notable result in Table III is the strong association between our access to capital

variables and the prevalence of family groups. The coefficients of Savings to GDP, Institutional

Funds, and Political Stability are all negative and significant in both the % Group and % Pyramid

regressions. In contrast, Governance Index and Newspaper, which should restrict business group

expansion by curbing private benefits of control, do not have any significant explanatory power.21

In interpreting the Institutional Funds result, it should be recognized that as a measure of

access to capital, this variable may also reflect aspects of a country’s corporate governance

environment. For example, Khorana, Servaes, and Tufano (2005) find that higher quality legal

systems and greater corporate transparency promote equity investments by mutual funds.22 Thus,

corporate governance can affect group prevalence through two channels: directly by facilitating

extraction of private benefits and indirectly by restricting the amount of external capital provided

by financial institutions to competing independent firms. Our results suggest that if governance

environments do have an impact on group development, this impact is more likely to be indirect.

[INSERT TABLE III HERE]

20 In unreported regressions, we replace the Governance Index with Competition and find that the coefficient of Competition is also insignificant. 21 A potential concern is that this result can be confounded by reverse causality. For example, groups can wield significant political power (Morck, Wolfenzon, and Yeung (2005)) that can stymie development of corporate governance standards. To address this, we use an instrumental variable (IV) regression, and employ the legal origins of our sample countries as a historically predetermined instrument. The results remain quantitatively similar. 22 Although Institutional Funds reflects the pool of equity investments (international and domestic) of all institutional investors, it is possible that business groups crowd out institutional investors’ investment in domestic equity. We use an alternative construction of Institutional Funds based on total assets under management of a country’s investment funds industry (obtained from Khorana et al. (2005)), which reflects funds inflows rather than asset allocations. This alternative measure is also related to group prevalence, although at a weaker significance level of 10 percent.

20

Among the remaining explanatory variables, the coefficient of Intra-Group Tax is

significant across all models. This suggests that the ability to minimise tax liabilities through

asset transfers across firms creates an incentive for business group formation. The Takeover

Index is also consistently significant, indicating that when acquirers are not restricted in their

ability to purchase a controlling stake through a partial acquisition, group formation and external

expansion is made significantly easier.23

For robustness purposes, we utilize two modified measures of group prevalence as

dependent variables in our regressions. This allows us to further assess the impact of access to

capital, while minimizing the cross-country variations in the level of private benefits of control.

For each country, the modified measures are (1) the number of family group firms as a proportion

of all firms with a controlling shareholder (Group Firms/Controlled Firms), and (2) the number

of family group firms as a proportion of all firms controlled by a family (Group Firms/Family

Firms). We argue that since concentrated ownership is a product of a business environment that

values corporate control, using these scaled variables allows us to measure group popularity,

conditional on the aggregate scope for private benefits of control in a market. If the roles of

private benefits of control are important, but difficult to observe in our previously reported

regression specifications due to the lack of a direct measure, then we would expect a weaker

relation between access to capital variables and group prevalence in these alternative models.

All the specifications involving the modified dependent variables in Table III indicate

that, consistent with our earlier results, Institutional Funds, Savings to GDP, and Political

Stability continue to have strong associations with the prevalence of family groups across

countries. We find that weak governance environments are not associated with the extent to

23 We also test one important component of the Takeover Index by itself, that is, the ownership threshold that triggers a formal bid to buy all outstanding shares. We find that like Takeover Index, the indicator variable based on the median of these thresholds is also significantly related to our country-level business group variables.

21

which family-controlled firms belong to business groups. In fact, Governance Index is positive

and significant, suggesting that controlling for other factors, strong governance may actually

promote group formation as controlling families are able to allay expropriation fears of minority

shareholders, while still utilizing the competitive advantage of an internal capital market.

In unreported regressions, we also test whether group firms remain under the control of

founding families and their heirs due to a shortage of well-trained professional managers (Khanna

and Palepu (1997)). We measure the quality and capacity of a country’s business education using

a survey variable from the World Economic Forum’s Global Competitiveness Report. We also

use inheritance laws information from Ellul, Pagano, and Pannunzi (2009) to examine the

possibility that restrictive inheritance laws limiting the amount of assets that can be bequeathed

by a group founder to a single heir can lead to break-ups of business groups. However, neither of

the above variables is significantly related to the proportion of listed firms belonging to a group.

III. Firm-Level Empirical Analysis

While the country-level results offer some preliminary information on the market

environments in which business groups are popular, the firm-level evidence presented below

provides us with a more rigorous and in-depth examination of the motivations behind group

development and architecture. Our extensive dataset and detailed group construction facilitate

several lines of inquiry. First, we identify the main differences between group and non-group

firms, focusing on characteristics related to their potential expropriation risks as well as their

financing needs and investment activities. Second, within each group, we analyze how member

firm traits change with their layer placement in a pyramid group. Third, we investigate whether

this placement and other within-group ownership linkages influence firm performance. Finally,

we document the overall impact of group affiliation by re-examining performance differences

22

between group and non-group firms in a multivariate setting, taking into account the role of

endogeneity in group membership choice in generating any differences.

A. Selection of Firm-Level Variables

Our firm-level analysis focuses on firm characteristics that are related to the importance

of private benefits of control, reputation, and internal capital within family groups. Another key

consideration in our selection is whether there is sufficient data coverage across sample countries

for a particular variable. To minimize the number of missing observations, firm-level data from

Worldscope and Datastream are supplemented with data from Osiris and Bloomberg. We also

perform extensive cross-checking across databases to ensure data accuracy.

As it is difficult to directly measure private benefits of control, we focus on expropriation

risk to minority shareholders. In particular, large, established, widely monitored firms are perhaps

less susceptible to expropriation. Thus, we compare firms along these dimensions by measuring

firm size, as market capitalization (total assets are also reported), firm age, as the number of years

since incorporation, and the number of analysts providing coverage. Further, when firms

aggressively pursue corporate investments or pay high dividends, there is less scope for

expropriation. We measure investment intensity by size of contemporaneous capital expenditures

scaled by total assets and the dividend rate by the ex-ante five-year average of dividend yields.

To analyze the roles of internal group capital, we examine a firm’s growth characteristics.

While investment intensity captures future growth opportunities, we also include historical

growth measures based on five-year average growth rates of total assets (sales growth is also

reported). To show how controlling shareholders rely on internal group capital to limit their

exposure to risky firms, while leveraging its use to fund firms with high information asymmetry,

we construct measures of a firm’s idiosyncratic risk (standard error from a one factor market

model estimated from five years of monthly stock returns prior to 2002) and asset tangibility

23

(total intangible assets divided by total assets). In addition, the functioning of a group’s internal

capital market can be further analysed by tracing where debt funding (total interest-bearing debt

scaled by total assets) is raised within the pyramid, relative to where capital needs are greatest.

Finally, we assess the aggregate impact of group affiliation and certain group structures

through two standard firm performance variables. The first measure is Tobin’s Q (proxied by

market value of assets divided by book value of assets) and the second is return on assets

(earnings before interests, tax, depreciation and amortization scaled by total assets). Panel C of

Table I summarizes the sources and construction of these firm-level variables.24

B. Univariate Analysis of Group Firm Characteristics

Table IV reports median differences for selected firm-level variables across group firms

and a set of matched firms. For each group firm, matched firms are either those having the same

two-digit Standard Industry Classification (SIC) codes or those with similar market capitalization

(ranging from 90 to 110 percent of the group firm’s market capitalization).25 The Wilcoxon

Signed-Rank test is used to assess whether differences are significantly different from zero.26

We begin by comparing the performance of group and non-group firms. Table IV reports

that based on size matching, group firms have significantly lower Q and ROA than non-group

firms. This is consistent with the finding of Claessens et al. (2002), who conclude that separation

of cash-flow rights and control rights (achieved through pyramids) reduces firm value.27 At this

point, however, it is unclear whether the inferior performance of group firms is a result of

24 To limit the number of outliers, our firm-level analysis excludes firms that are not traded for more than six months in 2002 and those that are six months away from being delisted. A few firms also have extremely high leverage (and hence, high Q), indicating likely cases of financial distress or reporting/data errors. We therefore also exclude firms in the top 2.3 percentile (two standard deviations from the mean of a normal distribution) of sample leverage ratios. 25 In smaller markets, finding a non-group match at the two-digit SIC level is not possible, so we match at the one-digit SIC level. When this is not possible, we use the country median. 26 For robustness purposes, we restrict the matching sample to cover only (i) stand-alone companies that do not belong to other types of groups, such as government-controlled and other non-family corporate groups, and (ii) firms that are family-controlled, but unaffiliated to any group. These alternative procedures generate very similar results. 27 Masulis, Wang and Xie (2009) find similar results in dual-class firms that raise control relative to cash-flow rights.

24

observable differences between group and non-group firms (addressed later in a multivariate

regression analysis), or endogenous selection effects caused by unobservable differences (also

addressed later in models accounting for potential endogeneity in group membership choice).

[INSERT TABLE IV HERE]

There are indeed many observable differences between group and non-group firms. In

particular, group firms are typically larger than their industry peers (by $70.89 million in total

assets and $19.91 million in market capitalization, both measured in U.S. dollars), perhaps

reflecting group firms’ ability to leverage their internal capital and reputation advantages to grow

and take on larger projects (Khanna and Palepu (1997)). Compared to matched non-group

industry peers, group firms have higher historical asset growth (Asset Growth) and historical

sales growth (Sales Growth). Group firms are also older, reflecting their ability to survive periods

of financial distress through intra-group support.28 The fact that group firms are larger and more

established raises doubt as to whether they would risk destroying the valuable reputational capital

(and corporate value) of the entire group by engaging in expropriation of a single member firm’s

shareholders. The costs of such actions can be further amplified by the greater level of analyst

coverage (Analyst) of group firms relative to non-group firms, which makes any expropriation

more detectable. The importance of group reputation is also evident in their significantly higher

levels of intangible assets (Intangibles).29 This could reflect groups’ investments in brand names

and recognition of the value of family reputation, or it may reflect the fact that firms with large

intangible assets face greater information asymmetry and often require group financing support.

The lack of dividend payments is often interpreted as an indication that minority

28 In an unreported result based on a sub-sample of firms with listing date information, we also find that group firms are significantly older at the time of going public. This suggests that internal group capital enables member firms to delay tapping into the external capital market until the cost of doing so is sufficiently low. 29 Belenzon, Berkovitz, and Bolton (2009) also report that a substantial share of innovation (measured by the number of patents) across 15 European countries is concentrated among business groups.

25

shareholders find it difficult to enforce their rights or that they are being expropriated (La Porta,

Lopez-de-Silanes, and Shleifer (2000)). However, we find that group firms pay higher dividends

relative to both industry- and size-matched non-group firms, which is more consistent with

reputation and internal capital market arguments. Similar to Faccio, Lang, and Young (2001), our

evidence suggests that groups pay out returns to shareholders to allay fears of expropriation and

to strengthen their reputation.30 Further, we note that dividends are an important channel for

redistributing funds within a family group (Gopalan, Nanda, and Seru (2008)).

Despite their higher level of intangible assets, group firms are also able to borrow more

than their non-group peers. This greater borrowing capacity is likely to be the result of intra-

group mutual support, which reduces bankruptcy risk. Groups may actively seek to use more debt

because issuing equity can risk a loss of control (Faccio and Masulis (2005)). Debt financing of

one group firm can also be used to support the funding requirements of other rapidly growing

member firms, again consistent with an internal capital market argument. This conjecture is

supported by the significantly higher levels of capital expenditures (CAPEX) observed in group

firms compared to their non-group peers (although this difference is only consistent across all

layers of a group when group firms are matched to industry peers). Overall, these differences

suggest that great care must be taken when assessing the effects of group affiliation.

C. Group Firm Characteristics and Pyramid Layer Placement

C.1. Univariate Results

We next document differences in firm characteristics among member firms within the

same business group. To examine within-group heterogeneity, we classify group firms into three

categories, namely Top, Middle, and Bottom, based on their layer positions (Pyramid Layer)

30 On a related note, we find that group firms retain significantly less cash holdings (relative to total assets) than non-group firms. This provides further evidence against expropriation arguments for the existence of business groups. Unfortunately, due to poor coverage of cash holdings data, we do not include this variable in our main analysis.

26

within the group structure. For each group, firms are placed into one of these three categories

such that the distance between the Top and Bottom firms is maximised. For example, in Figure 1,

the Arab Malaysian Corporation, situated in the direct control layer, is classified as the Top firm.

Global Carriers, situated at pyramid layer three, is classified as the Bottom firm. All the firms in

between (at layers one and two) are classified as Middle firms. Alternatively, we examine firms at

the bottom of a pyramidal ownership chain (that is, those firms in a pyramid that do not hold a

controlling ownership stake in any other group firm). In Figure 1, Arab Malaysian Property,

South Peninsular Industries and Global Carriers are all classified as Bottom of Chain firms.

In Table IV, we compare characteristics of group firms at different layers in the pyramid

to their non-group peers (Group vs. Non-Group Tests) and also examine differences among firms

at different layers in the same pyramidal business group (Within-Group Tests). For brevity, we

only report comparisons of Top and Bottom firm categories in the Group vs. Non-Group Tests.

For Within-Group Tests, we exclude strictly horizontal groups, in which all member firms are

directly connected to a controlling family, so that we can draw comparisons between Bottom and

Top (and Middle) categories, as well as between Bottom of Chain firms and Top firms.

Table IV shows that Q actually rises significantly as we go down a pyramid to such an

extent that Bottom firms no longer underperform their size-matched peers in the manner that

other group firms do. Within a group, Q is also higher for Bottom of Chain firms compared to

Top firms, and for Bottom firms compared to Middle firms. Operating performance (ROA) also

appears to be significantly higher in Bottom firms compared to Middle firms within the same

group. This suggests that for some pyramidal groups in our sample, the potentially high-growth

Bottom firms (high Q) begin to realize their growth potential by displaying superior operating

performance. If group support is valuable to the growth of bottom-tier firms, then this is bound to

translate into superior operating returns at later stages of their development.

27

The level of capital expenditures (CAPEX) also rises significantly down a pyramidal

chain, consistent with the findings of Almeida et al. (2008) for Korea. Bottom firms also have

higher capital expenditures than their industry and size-matched peers. In contrast, Asset Growth

decreases down the pyramidal chain, and Top firms also have higher Sales Growth than Bottom

of Chain firms. While these results may seem contradictory, it is important to recognize that

measures of historical growth need not be consistent with measures of expected future growth.

This evidence suggests that family groups place firms with higher growth prospects at lower

pyramid layers and that these firms are funded (established) by parent firms that have

experienced strong historical growth and accumulated substantial retained earnings. Further, most

borrowings appear to be strategically undertaken by Top firms despite their lower investment

needs, indicating that debt channelled to new members is an important source of internal capital.

Comparisons of Idiosyncratic Risk indicate that both Bottom and Bottom of Chain firms

are generally riskier than Top firms in the same group. Bottom and Bottom of Chain firms are also

significantly smaller and younger, and receive less analyst coverage.31 These findings suggest

that groups are carefully structured to limit investment and risk exposure of ultimate controlling

families, while facilitating the pursuit of promising, but risky investment opportunities (similar to

a venture capitalist limiting their risk taking, while exercising substantial control).32

Young, risky, high-growth firms with high information asymmetry have difficulty raising

capital, particularly in environments where capital is scarce and investor protection is weak.

Family groups are in a position to fund such firms at the bottom of their pyramidal group,

because it is precisely at this point where a group can most effectively leverage its internal

31 The size results hold by construction, since the parent’s investment in a subsidiary will be recorded in the parent’s balance sheet. While accounting consolidation can affect absolute firm characteristics (such as size), it should not affect other relative measures. In fact, any biases created would understate our results since each relative measure for a Top firm is partially made up of the weighted average of corresponding measures of their subsidiaries. 32 An alternative explanation is that groups expropriate bottom-tier firms by forcing them to take unnecessary risk. However, this is inconsistent with our evidence that bottom-tier firms have higher firm value than other group firms.

28

capital. This means that such young firms can rely less on expensive external capital as well as

reduce the cost of the residual external capital required, as they are backed both implicitly by a

group’s reputation, and explicitly through its internal capital investments.33 This conjecture is

supported by within-pyramid comparisons of Direct Ownership, which show that the aggregate

size of a group’s direct shareholdings (and voting rights) in a member firm is significantly greater

in Bottom firms than in Top firms (a similar but unreported result is found for Direct Control

Rights). If control and expropriation were the only motives of expanding a pyramidal group, then

the group should actually maximize its direct shareholdings/voting rights in Top firms and

minimize its cash-flow rights in Bottom firms.34 Overall, our univariate findings regarding the

nature of group firms and the strategies with which they are funded highlight striking similarities

between the functions of such groups and those of private equity funds in developed markets.35

C.2. Multivariate Regression Results

To provide a more comprehensive analysis of within-group heterogeneity, we examine

the relation between group firm characteristics and layer positions in a multivariate regression

framework. We report the regression results using the Pyramid Layer measure and the Bottom of

Chain indicator as alternative dependent variables.36 The explanatory variables are firm

characteristics employed in Table IV.37 We also use group-specific fixed effects to control for the

potential influence of differences in group-level characteristics (for example, group size,

33 Evidence in Gopalan, Nanda, and Seru (2007) on Indian family groups supports the proposition that member firms benefit from the financial strength of the group and can receive additional capital when they are in financial distress. 34 This is because voting rights of the apex firm generate control of an entire pyramid, and hence, are more valuable than voting rights of individual lower-tier firms. In addition, if bottom firms are to be expropriated, it may be optimal for the controlling family to limit exposure to these firms by retaining just the minimum necessary controlling stake. 35 A case-by-case analysis by UBS Investment Research (2006) of 26 European holding companies, most of which are part of family groups in our sample, shows that many of these companies adopt strategies akin to those of a private equity fund. Private equity and unlisted assets on average make up 47 percent of their portfolios. 36 The results are similar when the Bottom indicator is used as the dependent variable. 37 Among the size measures, we use the natural logarithm of market capitalization (Log Size) in this and subsequent regression analyses, as equity value perhaps better reflects the control and funding motivations of the ultimate owner with respect to a group firm. Among the historical growth measures, we use Asset Growth rather than Sales Growth, largely because the former is less susceptible to extreme observations.

29

reputation, entrepreneurial skills of controlling families, scope of private benefits of control, and

so on). In regressions where the dependent variable is the binary variable Bottom of Chain, we

employ a conditional fixed-effects logit estimation to incorporate the large number of covariates

generated by group fixed effects. Both regression models (and those in our subsequent analysis)

include sector fixed effects, constructed using the first digit of a firm’s primary SIC code.

The results reported in Table V confirm most of our univariate findings. Firms with larger

capital requirements (CAPEX) tend to be placed closer to the bottom of a group. In contrast, firms

with large borrowing capacities and strong historical asset growth are more likely to be near the

apex of the pyramid, reflecting the roles these firms play in establishing, funding, or acquiring

new subsidiaries. We further divide the sample firms into those from emerging markets and those

from developed markets (following the classification in Standard and Poors’ Emerging Market

Database). It appears that the previous results related to capital expenditures and historical asset

growth are mainly driven by business groups in emerging markets, where access to external

capital is more critical. This again reinforces our earlier findings about the importance of internal

group capital to member firm development, especially in less developed capital markets.

[INSERT TABLE V HERE]

D. Firm Performance and Group Members’ Placement and Ownership Structures

We next examine whether variations in firm performance within a group can be explained

by the position and ownership characteristics of member firms. La Porta, Lopez-de-Silanes, and

Shleifer (2002) find that the cash-flow rights of ultimate owners are positively related to firm

value for a sample of large firms in 27 countries. Claessens et al. (2002) not only document a

similar result for cash-flow rights for East Asian firms, but also find that ultimate owners’ control

rights are negatively related to firm value. However, these results are based on samples of both

group and non-group firms and consequently may reflect systematic differences in the properties

30

of the two samples. The aggregate impact of group affiliation and potential endogeneity in group

membership choices make it difficult to isolate the impact of a pyramidal structure (and the

separation of cash-flow rights from control rights) on firm performance.

To reduce the confounding effect of such systematic differences, the following analysis is

restricted to only group firms. In particular, we estimate alternative regressions of Q (or ROA) on

various pyramidal position and ownership measures. Since firms from different groups are likely

to differ significantly from one another because of unique group-level qualities (for example,

group reputation, connections, and entrepreneurial skills of the controlling family), we use group

fixed effects to control for unobservable group-level differences. As documented in the univariate

results, there are significant within-group differences in firm size, firm age, historical asset

growth rate, leverage, dividend yield, and capital expenditures, all of which can influence

performance. These factors form part of our set of control variables. In addition, firm

performance can be influenced by a firm’s level of information asymmetry, which reduces the

effectiveness of shareholder monitoring. We therefore also include intangible assets to total assets

(Intangibles) and the logarithm of analyst coverage (Log Analyst) to proxy for firm-level

information asymmetry. Following Villalonga and Amit (2006), we control for a firm’s beta in

the Q regression, since asset-pricing theory predicts that systematic risk has an important role in

determining firm valuation. Finally, to deal with the possibility that intra-group transfers can

cause variations in performance such that regression standard errors differ systematically across

groups, we use cluster-adjusted standard errors, with a cluster defined as a business group.

[INSERT TABLE VI HERE]

Based on OLS estimations, Table VI documents that lower cash-flow rights accruing to

controlling families are actually associated with member firms having higher Q and ROA. Q is

also positively related to a family’s control rights. Thus, when analyzing within-group

31

differences, we do not find that the separation of cash-flow rights from control rights, which is

hypothesized to capture expropriation incentives, results in inferior firm performance. Our results

favor a different interpretation of group structure dynamics. The negative correlation between

ultimate cash-flow rights and Q indicates that firms established at or close to the bottom layer of

a business group (those giving low cash-flow rights to controlling families) may actually benefit

the most from being part of a group. In fact, there is a positive and significant relation between Q

and the layer position of member firms. We obtain the same finding regardless of whether a

firm’s position is defined as the number of group firms above it (Pyramid Layer) or as an

indicator for firms at the bottom of a pyramidal chain (Bottom of Chain). This confirms earlier

univariate results that the benefits of a group’s capital and certification support are greatest at the

bottom of the pyramid.

The results show that ROA is also greater for firms close to the bottom of a pyramid.38

This result raises the question of why supposedly junior group members can deliver both high

value (Q) and high operating returns (ROA) at the same time. We investigate this issue in more

detail by categorizing firms at the bottom of pyramid chains into newly established firms (those

listed within the last five years) and seasoned firms (those listed for more than five years). We

find that most of the superior operating performance of bottom-of-chain firms relative to other

group firms is driven by the sub-sample of seasoned firms, while the Q result remains consistent

across both sub-samples (this analysis is not reported for brevity). This suggests that for firms at

the bottom of a pyramid, group structure enhances firm value not only when they are first

established, but internal group capital aids them to realize the cash-flow benefits from their

ongoing investment opportunities.

38 We also consider potential endogeneity in the relation between Q (and ROA) and Pyramid Layer, that is, there may be unobserved growth-related factors that influence both pyramid placement and firm performance. In particular, we re-estimate the regression between Pyramid Layer and Q (or ROA) using a 2SLS model, where the first-stage regression determines Pyramid Layer, as reported in Table V. The results (unreported) are consistent with the earlier OLS regression estimates.

32

We also find that Q is significantly and positively related to the Direct Ownership (or

Direct Control Rights) measure. When a new firm is added to a group, the ultimate cash-flow

rights of the controlling family in the firm is not an active ownership decision variable as the

remainder of the ownership chain(s) connecting the immediate parent(s) of a new firm to the

family is well established and likely to be optimally configured. Instead, the main ownership

decision is the size of the last link in the chain(s), that is, the direct shareholdings of the

immediate parent(s) of a new firm. This direct ownership link held by the parent(s) one layer up

in the pyramid not only signals a group’s commitment to a new firm, which can help alleviate its

information asymmetry with outside investors, but also reflects the extent to which the firm is

funded by less expensive internal capital. Both of these effects are likely to explain the higher

market valuation of group firms with greater Direct Ownership.

Past empirical research into the separation of cash-flow rights and control rights

(sometimes referred to as the ownership-control wedge) often does not distinguish between a

wedge created by a pyramid structure and that created by dual-class shares with differential

voting rights. Almeida and Wolfenzon (2006a) suggest that the use of these two mechanisms may

have different motivations. Our results thus far suggest that a wedge created through pyramiding

is actually associated with higher firm valuation due to its internal capital market benefits.

However, it is unlikely that a wedge created through dual-class shares would have the same

impact given that it does not enhance a firm’s access to capital. To examine this issue, we first

construct the conventional Wedge variable, defined as the difference between an ultimate owner’s

cash-flow and control rights, and include it in the performance regression together with Pyramid

Layer shown in Model (5). We do not detect any marginal effect of Wedge on either Q or ROA,

after controlling for Pyramid Layer (note that Pyramid Layer and Wedge are highly correlated).

In Model (6), we replace Wedge with Dual-Class Wedge, which measures the wedge created

33

solely by dual-class shares. The resulting coefficient is significant, reflecting a negative impact of

dual-class shares on Q, after accounting for a firm’s position in the pyramid. These results

indicate that control enhancing devices can have heterogeneous effects.39

E. Firm Performance and Group Affiliation

Our final analysis seeks to document the aggregate impact of group structures by

comparing performance of group and non-group firms in a multivariate regression setting. Most

importantly, we raise and address the issue of endogeneity. In reviewing the divergent evidence

on this issue, Khanna and Yafeh (2007) point out that “comparisons of group versus non-group

firms are plagued with selection issues, the most obvious one being the assumption that group

affiliation is exogenous”. Indeed, our previous univariate results documenting systematic

differences between group and non-group firms and among firms in the same group indicate that

group membership choices may not be exogenous. In this section, we investigate whether

endogeneity can partly explain the relative performance of group over non-group firms.

E.1. OLS Regression Results

To provide comparability to past results, we first investigate the relation between Q (and

ROA) and group affiliation using a simple multivariate OLS regression. Our main explanatory

variable is Group Indicator, which takes a value of one if a firm belongs to a family group, and

zero otherwise. The control variables are similar to those used in Table VI. Table VII first reports

baseline OLS estimates of the Q and ROA regressions. Similar to the univariate comparisons, we

find that both Q and ROA are significantly lower for family-group firms than for other firms. 40

39 In a robustness check, we exclude countries that allow consolidation of financial statements of parent and subsidiary at ownership thresholds below 75 percent, as this may create (double counting) biases in the book-value based variables of group firms. The 75 percent threshold is used because only a small proportion (under 10 percent) of within-group shareholdings in our sample exceeds 75 percent. The results are similar to those reported in Table V. 40 To minimize the possibility that the impact of family-group affiliation on firm performance is a result of the generic benefits and/or costs of a concentrated ownership structure or commitment of a controlling family (or individual), the performance regression models are alternatively estimated on a sub-sample of all firms controlled by families (or individuals). The results remain qualitatively similar.

34

This result is comparable to those reported by Claessens et al. (2002) and Lemmon and Lins

(2003). If group affiliation can be assumed to be exogenous, then our evidence would be

consistent with their conclusions that the lower firm values and operating returns of group firms

are likely to reflect expected future (or even actual) expropriation of minority shareholder

interests by controlling families.41

[INSERT TABLE VII HERE]

E.2. Treatment-Effects Regression Results

We next examine the possibility that the negative coefficient of Group Indicator in the

OLS regression may partly reflect an endogenous selection bias. If group affiliation alleviates

external capital constraints associated with high information asymmetry, then Group Indicator

may be correlated with the unexplained components of Q. For example, firms engaging in

projects that are capital intensive and lack immediate pledgeable cash flows may be unable to

raise funds in external capital markets by themselves due to high information asymmetries, and

may require family group support. Conversely, firms that are able to overcome information

asymmetry because they have high-value projects can opt to remain independent. Both scenarios

can result in group firms possessing lower current market valuation than their non-group peers.

To address this endogenous selection problem, we employ a treatment-effects model in

which we estimate the performance regression (the outcome equation) simultaneously with the

group-selection regression (the treatment equation) using maximum likelihood estimation.42 The

self-selection issue is addressed by explaining group membership with the prior set of

explanatory variables from the performance regression and a new set of identifying instruments. 41 However, in an unreported test, we find that the valuation discount of group firms is actually lower in emerging markets than in developed markets. This indicates that expropriation risk, which is expected to be higher in emerging markets, is unlikely to be the only explanation for the performance differences between group and non-group firms. 42 This estimation requires a strong distributional assumption of bivariate normality, but provides cluster-adjusted standard errors. We also estimate the model using both a two-stage instrumental variable estimation (with the first stage involving a probit regression) as suggested by Maddala (1983) and the Heckman two-step consistent estimator, which do not impose the same assumption. Both generate similar results to the maximum likelihood estimation.

35

We employ three identification strategies in selecting a set of instruments that we argue

are correlated with group affiliation, but not with the unexplained components of performance.

First, group membership can be identified using cross-country differences. Among the previously

discussed country-level variables, we select takeover regulations and intra-group tax regulations

as instruments. As shown in Table III, these factors are significantly correlated with the

prevalence of group firms in a country. Takeover and intra-group tax regulations, however, are

unlikely to generate differences in firm performance across countries as they control wealth

transfers between firms, rather than value creation by firms in the same jurisdiction.

Second, we use three separate firm-level instruments, namely Idiosyncratic Risk and two

indicator variables for firms operating in media and real-estate related industries.43 Idiosyncratic

risk is frequently used as an instrument for ownership structure when examining its relation to Q

(Himmelberg, Hubbard, and Palia (1999) and Villalonga and Amit (2006)). In the context of our

study, idiosyncratic risk is related to the likelihood of group membership (as group structures

help to diversify a controlling family’s exposure to firms with high firm-specific risk), but should

be unrelated to Q (which in theory is a function of market risk, rather than firm-specific risk).44

Our industry instruments are based on the proposition that business groups can overcome co-

ordination failures and hold-up problems by controlling firms that can provide important

spillover benefits to other member firms (see Morck and Nakamura (2007)). For instance, a

group can control complimentary businesses in the sense that one firm can provide the

infrastructure or reduce the costs involved in operating the second firm. For this reason, media

and real estate indicators are potentially suitable instruments because firms in these industries can

generate synergistic benefits for other group firms (e.g. publicity, marketing, access to prime real

43 Similar results are obtained if we use the standard error obtained from the three-factor model as the alternative measure for Idiosyncratic Risk. 44 Idiosyncratic risk may be empirically correlated with Q due to the possibility that both are associated to a firm’s growth and information asymmetry. However, as we already control for growth and information asymmetry, we argue (and later test) that idiosyncratic risk is unrelated to the unexplained component of Q in the outcome equation.

36

estate, and so on). Moreover, firms in these industries can provide amenity value to controlling

shareholders.45 Unlike outright expropriation, however, both amenity and synergistic spillover

benefits can be realized without minority investors necessarily bearing a cost for their creation.46

Also, there is no clear evidence that media and real estate firms systematically under- or over-

perform those in other industries.

Finally, group membership can be identified based on historical market conditions around

a firm’s listing date. We argue that independent (non-group) firms are more likely to go public

when capital market conditions are favourable (to take advantage of “windows of opportunities”

and overcome their access to external capital constraints) than group firms, which can rely on

group funding that is largely independent of external capital market environments. We use the

cumulative return of the domestic stock market index in the listing year of each firm as an

instrument (Index Return at Listing).47 As this variable is based on historical market conditions, it

is unlikely to have a direct and persistent influence on current performance of individual firms.48

We document the strength of our instruments at the individual firm level by first running a

probit model of Group Indicator on firm-level controls (same as those used in Tables V, VI, and

VII), country-level controls (governance and access to capital factors as used in Table III),49 and

the instruments (note that two specifications arise due to the incomplete coverage of Index Return

at Listing). The results are reported in the first two columns of Table VIII. The relation between

45 Djankov, McLeish, Nenova and Shliefer (2003) show that families often control media firms due to their potential amenities. Bunkanwanicha, Fan and Wiwattanakantang (2008) suggest that family-controlled property development and construction firms build and rely on large information networks and political connections because they are heavily dependent on government regulation and proprietary information. In an unreported industry analysis, we find that firms in media and real estate industries are indeed significantly over-represented in the group firm sample. 46 For example, media owners are likely to have close relationships to sports, media and movie stars. 47 Information on the year of listing is obtained from Osiris, or else we use the first trading date listed in Datastream. Coverage in both data sources for listing dates prior to 1990 is sporadic and not consistent across countries, so we construct this variable only for a subset of firms listed from 1990 onwards. 48 It is possible that firms taking advantages of “windows of opportunities” to go public may exhibit greater post-listing underperformance than others. However, our results based on the instrument Index Return at Listing remain unchanged even after excluding newly listed firms (those having been listed for less than five years). 49 Due to the large number of observations in the firm-level analysis, we are able to use all governance and access to capital measures jointly rather than individually as in the specifications in Table III.

37

group membership and firm characteristics generally follows our univariate findings reported in

Table IV. For brevity, we do not report the coefficients of individual country-level controls, but

they confirm the country-level results reported in Table III, that is, the likelihood of a firm being

part of a business group is negatively related to access to capital factors and positively related to

Governance Index. Most importantly, we show that in most cases, the selected instruments have a

significant influence (in the expected direction) on the likelihood of group membership.

To check that the instruments are not correlated with the error terms in the outcome

equation, we test for the exclusivity condition of all instruments individually, using a generalized

Anderson-Rubin test developed by Stock and Wright (2000), and collectively using the Hansen J

test. The test statistics (not reported for brevity) are not significant. Thus, we cannot reject the

null hypothesis that the instruments can be excluded from the performance equation.

[INSERT TABLE VIII HERE]

We next estimate the treatment-effects model containing both the group membership and

performance equations simultaneously. The former (selection) equation uses either of the probit

specifications discussed above, and the latter equation regresses either Q or ROA on Group

Indicator and the same set of controls as those used in the probit models, but excluding the

instruments. We find that across both treatment-effects regressions involving Q, the coefficient of

Group Indicator becomes significantly positive. There are also strong statistical reasons not to

assume that the group-affiliation choice is entirely exogenous, as the null hypothesis of zero

correlation between the residuals from the two equations in our treatment-effects model is

consistently rejected by the Wald test. Thus, there are unobserved factors that influence both the

likelihood of group affiliation and firm value. The treatment-effects model also indicates that the

instruments still have a significant impact on group membership in most cases. These findings

remain the same in unreported tests where we estimate the model using only family-controlled

38

firms, or where we split the sample into emerging and developed markets.

The positive coefficient Group Indicator in the treatment-effects models involving Q

generates an important implication that is again consistent with an internal capital market

explanation for business groups. The fact that group firms on average underperform non-group

firms may to some extent be due to the negative impact of private benefits of control, but it may

also reflect potential endogenous selection effects. After controlling for endogeneity, we find that

groups help to enhance firm value for member firms with particular characteristics captured by

our control variables and instruments.50 Further, this is not inconsistent with the results from the

ROA regression in Table VIII. Unlike the Q regressions, group affiliation continues to be

negatively related to ROA after accounting for endogenous selection. This is because ROA does

not incorporate the market’s assessment of the value of a group firm, but instead reflects the

contemporaneous operating conditions of the firm. In our selection arguments, group firms

exhibit inferior operating performance to non-group firms precisely because business groups have

an advantage in financing firms with less attractive security attributes, that is, those that find it

difficult to obtain external funding because of their lower pledgeable cash flows.51

IV. Conclusion

This study examines the motivation behind the existence of family-controlled business

groups. At the country level, we find that family groups are less prevalent where access to

external capital is unrestricted, where there are stringent regulations related to taxation of intra-

50 We also employ propensity score matching (PSM) to minimize the roles of selection biases (due to observable factors) in creating performance differences between group and non-group firms (using either Model (1) or Model (2) of Table VIII to generate propensity scores). By simply matching based on observables, the treatment effect on the treated (the effect of group membership on group firms) is already substantially smaller than the unmatched performance difference between group and non-group firms, and is even insignificant in some PSM specifications. 51 Systematic differences between group and non-group firms can also create a selection bias that affects the within-group estimation reported in Table VI. We revisit the within-group analysis, this time employing the Heckman two-stage procedure, in which the first stage is a probit selection model of group affiliation identical to Model (1) of Table VIII. Essentially, we find very similar results to the OLS regressions in Table VI.

39

group transactions, and where merger and acquisition rules make it difficult for business groups

to expand by partially acquiring firms. While investor protection is an important factor that may

influence the availability of external capital, it does not appear to provide much additional

explanatory power in cross-country models of business group prevalence. Although it is difficult

to establish causality and to account for unique historical circumstances leading to the emergence

of groups, our country-level results favor an access to capital explanation for their popularity.

The significant negative coefficient of our access to capital factors suggest that a group’s ability

to support member firms with internal capital make them more popular (or perhaps more

entrenched) in markets where external capital is limited.

Our firm-level analysis provides the clearest contrast between funding and control

motives in group development. Within each group, we find that young, risky, and capital

intensive firms are placed at the bottom of a pyramidal chain. Family groups finance such firms

at the bottom of their pyramid structures because it is precisely at this point where groups obtain

the greatest leverage from their own invested capital and reputation, while minimizing the risk

borne by controlling families. More importantly, we find that this placement is in turn positively

associated with firm value and that firms with greater proportions of equity investment by their

own group are also likely to have higher market values. Thus, family groups appear to provide

substantial benefits to member firms by reducing the need to rely on expensive external capital.

This result differs from existing research that emphasizes the increased expropriation risk to

minority shareholders when cash-flow rights and control rights diverge for firms at the bottom of

a pyramidal group. However, consistent with expropriation based theories, we find that group

firms in which shares with differential voting rights are employed suffer a valuation discount.

In relation to the aggregate impact of group structures, we find that group firms appear to

underperform their non-group counterparts. While this may reflect the higher expropriation risk

40

associated with groups, we examine an additional possibility. That is, firms with high-value

projects can overcome information asymmetries by themselves to obtain external funding,

whereas those with lower immediate pledgeable cash flows may find it difficult to do so and may

require business group support. Accounting for this endogeneity, we find that group structures

help improve the valuation of certain member firms.

Overall, our findings are consistent with the importance of internal capital market benefits

that groups can provide. We conclude that preserving private benefits of control is not the sole

motive, or perhaps not even the primary motive, behind the creation and expansion of pyramidal

groups. Rather, such structures also serve a critical function of leveraging a group’s internal

capital and reputation. However, Almeida and Wolfenzon (2006b) argue that such internal capital

advantages, while beneficial for member firms, do not always translate into economy-wide

allocative efficiency because in certain instances, a group may choose to internally fund

enterprises that are inefficient relative to those competing for funds outside a group. Therefore,

an important implication of our analysis is that improvements in access to external capital can

weaken the economic role of family business groups. For emerging markets in particular,

openness to foreign investment and the development of a venture capital industry can provide

new competition to business groups by offering alternative funding sources for capital-intensive

and high-growth ventures. To the extent that other capital providers can also bring experience,

advanced technologies, and other strategic advantages to such firms, the competitive advantage

of family business groups can be significantly diminished, thereby seriously eroding the value of

their support to member firms and their overall economic dominance.

41

References

Almeida, Heitor, Sang-Yong Park, Marti Subrahmanyam and Daniel Wolfenzon, 2008. The structure and formation of business groups: Evidence from Korean Chaebols, SSRN Working Paper: http://ssrn.com/abstract=1354500 Almeida, Heitor, and Daniel Wolfenzon, 2006a. A Theory of Pyramidal Ownership and Family Business Groups, Journal of Finance 61, 2637-2680. Almeida, Heitor and Daniel Wolfenzon, 2006b. Should Business Groups Be Dismantled? The Equilibrium Costs of Efficient Internal Capital Markets, Journal of Financial Economics 79, 99–144. Bae, Kee-Hong, Jun-Koo Kang, and Jin-Mo Kim, 2002. Tunneling or Value Added? Evidence from Mergers by Korean Business Groups, Journal of Finance 57, 2695-2740. Baek, Jae-Seung, Jun-Koo Kang, and Inmoo Lee, 2006. Business Groups and Tunneling: Evidence from Private Securities Offerings by Korean Chaebols, Journal of Finance 61, 2415-2449. Belenzon, Sharon, Tomer Berkovitz and Patrick Bolton, 2009. Intracompany Governance and Innovation, NBER Working Paper: http://www.nber.org/papers/w15304. Bertrand, Marianne, Paras Mehta, and Sendhil Mullainathan, 2002. Ferreting Out Tunneling: An Application to Indian Business Groups, Quarterly Journal of Economics 117, 121-148. Bunkanwanicha, Pramuan, Joseph P. H. Fan, and Yupana Wiwattanakantang, 2008. Why do shareholders value marriage? ECGI - Finance Working Paper: http://ssrn.com/abstract=1108642 Bushman, Robert M., Joseph D. Piotroski, and Abbie J. Smith, 2004. What determines corporate transparency? Journal of Accounting Research 42, 207-252. Claessens, Stijn, Simeon Djankov, and Larry H.P. Lang, 2000. The separation of ownership and control in East Asian corporations, Journal of Financial Economics 58, 81-112. Claessens, Stijn, Simeon Djankov, Joseph P. H. Fan, and Larry H. P. Lang, 2002. Disentangling the Incentive and Entrenchment Effects of Large Shareholdings, Journal of Finance 57, 2741-2771. Djankov, Simeon, Rafael La Porta, Florencio Lopez-de-Silanes and Andrei Shleifer, 2008. The law and economics of self-dealing, Journal of Financial Economics, 88, 430-465. Djankov, Simeon, Caralee McLiesh, Tatiana Nenova, and Andrei Shleifer, 2003. Who owns the media? Journal of Law and Economics, 46, 341-381. Dyck, Alexander, and Luigi Zingales, 2004. Private Benefits of Control: An International Comparison, Journal of Finance 59, 537-600.

42

Ellul, Andrew, Marco Pagano, and Fausto Panunzi, 2009. Inheritance Law and Investment in Family Firms, SSRN Working Paper: http://ssrn.com/abstract=1331368 Faccio, Mara, and Larry H. P. Lang, 2002. The Ultimate Ownership of Western European Corporations, Journal of Financial Economics 65, 365-395. Faccio, Mara, Larry H. P. Lang, and Leslie Young, 2001. Dividends and Expropriation, American Economic Review 91, 54–78. Faccio, Mara, and Ronald Masulis, 2005. The Choice of Financing Methods in European Mergers & Acquisitions, Journal of Finance 60, 1345-1388. Fisman, Raymond, 2001. Estimating the Value of Political Connections, American Economic Review 91, 1095-1102. Fogel, Kathy, 2006. Oligarchic Family Control, Social Economic Outcomes, and the Quality of Government, Journal of International Business Studies 37, 603-622. Gomes, Armando, 2000. Going Public without Governance: Managerial Reputation Effects, Journal of Finance 55, 615-646. Gompers, Paul and Josh Lerner, 1996. The Use of Covenants: An Empirical Analysis of Venture Partnership Agreements, Journal of Law and Economics 39, 463-498. Gopalan, Radhakrishnan, Vikram K. Nanda, and Amit Seru, 2007. Affiliated Firms and Financial Support: Evidence from Indian Business Groups, Journal of Financial Economics 86, 759-795 Gopalan, Radhakrishnan, Vikram K. Nanda, and Amit Seru, 2008. Do Business Groups Use Dividends to Fund Investments? SSRN Working Paper: http://ssrn.com/abstract=639184 Himmelberg Charles P., Glen R. Hubbard, and Darius Palia, 1999. Understanding the Determinants of Managerial Ownership and the Link between Ownership and Performance, Journal of Financial Economics 53, 353-384. Holderness, Clifford, 2008. Do Differences in Legal Protections for Investors Explain Differences in Ownership Concentration, SSRN Working Paper: http://ssrn.com/abstract=1104678 Hoshi, Takeo, Anil Kashyap, and David Scharfstein, 1991. Corporate structure, liquidity and investment: evidence from Japanese industrial groups, Quarterly Journal of Economics 106, 33-60. Joh, Sung Wook, 2003. Corporate Governance and Firm Profitability: Evidence from Korea before the Economic Crisis, Journal of Financial Economics 68, 287-322. Johnson, Simon, Rafael La Porta, Florencio Lopez-de-Silanes and Andrei Shleifer, 2000. Tunneling, American Economic Review 90, 22-27.

43

Kaufmann, Daniel, Aart Kraay, and Massimo Mastruzzi, 2003. Governance Matters III: Governance Indicators for 1996–2002, The World Bank Working Paper. Kaplan, S., and P. Stromberg, 2003. Financial Contracting Theory Meets the Real World: An Empirical Analysis of Venture Capital Contracts, Review of Economic Studies 70, 281-315. Khanna, Tarun, and Krishna Palepu, 1997. Why Focused Strategies May Be Wrong for Emerging Markets, Harvard Business Review 75, 41-51. Khanna, Tarun, and Krishna Palepu, 2000. Is Group Affiliation Profitable in Emerging Markets? An Analysis of Diversified Indian Business Groups, Journal of Finance 55, 867-891. Khanna, Tarun, and Yishay Yafeh, 2005. Business Groups and Risk Sharing around the World, Journal of Business, 78, 301-340. Khanna, Tarun, and Yishay Yafeh, 2007. Business Groups in Emerging Markets: Paragons or Parasites? Journal of Economic Literature 45, 331-372. Khorana, Ajay, Henri Servaes, and Peter Tufano, 2005. Explaining the size of the mutual fund industry around the world. Journal of Financial Economics 78, 145-185. Kosenko, Konstantin, 2007. Evolution of Business Groups in Israel: Their Impact at the Level of the Firm and the Economy, Israel Economic Review 5, 55-93. La Porta, Rafael, Florencio Lopez-de-Silanes and Andrei Shleifer, 1999. Corporate Ownership around the World, Journal of Finance 54, 471-517. La Porta, Rafael, Florencio Lopez-de-Silanes and Andrei Shleifer, 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 La Porta, Rafael, Florencio Lopez-de-Silanes, Andrei Shleifer, and Robert Vishny, 1997. Legal Determinants of External Finance, Journal of Finance 52, 1131-1150 Lemmon, Michael L, and Karl V. Lins, 2003. Ownership Structure, Corporate Governance, and Firm Value: Evidence from the East Asian Financial Crisis, Journal of Finance 58, 1445-1468 Li, Donghui, Fariborz Moshirian, Peter Kien Pham, and Jason Zein, 2006. When Financial Institutions Are Large Shareholders: The Role of Macro Corporate Governance Environments, Journal of Finance 61, 2975-3007. Lins, Karl V., 2003. Equity Ownership and Firm Value in Emerging Markets, Journal of Financial and Quantitative Analysis 38, 159-184. Maddala, Gangadharrao S., 1983. Limited Dependent and Qualitative Variables in Econometrics. New York: Cambridge University Press.

44

Majluf, Nicolas, Nureya Abarca, Dario Rodriguez and Luis A. Fuentes, 1998. Governance and Ownership Structure in Chilean Economic Groups, Revista ABANTE 1, 111-139. Masulis, Ronald W., Cong Wang and Fei Xie, 2009. Agency Problems at Dual Class Companies, Journal of Finance 64, 1697-1727. Morck, Randall, 2005. How to Eliminate Pyramidal Business Groups: The Double Taxation of Intercorporate Dividends and Other Incisive Uses of Tax Policy. In J. Poterba (ed.), Tax Policy and the Economy 19, MIT Press, 135-179. Morck, Randall, and Masao Nakamura, 2007. Business Groups and the Big Push: Meiji Japan's Mass Privatization and Subsequent Growth, NBER Working Paper: http://www.nber.org/papers/w13171 Morck, Randall, Michael Percy, Gloria Tian and Bernard Yeung, 2005. The Rise and Fall of Widely Held Firms: the History of Corporate Ownership in Canada, in R. Morck (editor), The History of Corporate Governance around the World: Family Business Groups to Professional Manager, NBER, the University of Chicago Press, 65-140 Morck, Randall, and Lloyd Steier, 2005. The Global History of Corporate Governance: An Introduction, in R. Morck (editor) The History of Corporate Governance around the World: Family Business Groups to Professional Manager, NBER, the University of Chicago Press, 1-64. Morck, Randall, Daniel Wolfenzon, and Bernard Yeung, 2005. Corporate Governance, Economic Entrenchment, and Growth, Journal of Economic Literature 43, 655–720. Nenova, Tatiana, 2006. Takeover Laws and Financial Development, World Bank Policy Research Working Paper: http://ssrn.com/abstract=935491 Pagano, Marco, 1993. Financial Markets and Growth: An overview, European Economic Review 37, 613-622. Pagano, Marco, and Paolo Volpin, 2005. Political Economy of Corporate Governance, American Economic Review 95, 1005-1030. Perotti, Enrico C., and Stanislav Gelfer, 2001. Red barons or robber barons? Governance and investment in Russian financial–industrial groups, European Economic Review 45, 1601-1617 Polsiri, Piruna, and Yupana Wiwattanakantang, 2006. Thai Business Groups: Crisis and Restructuring, in Sea-Jin Chang (ed.), Business Groups in East Asia: Financial Crisis, Restructuring, and New Growth, Oxford University Press, Oxford. Rabelo, Flavio, and Flavio C. Vasconcelos, 2002. Corporate Governance in Brazil, Journal of Business Ethics 37, 321-335. Sato, Yuri, 2004. Corporate Governance in Indonesia: a Study on Governance of Business Groups, Asian Development Experience Vol. 2: The Role of Governance in Asia, edited by Yasutami Shimomura, Institute of Asian Studies, Singapore.

45

Shimizu, Tatsuya, 2004. Family Business in Peru: Survival and Expansion under the Liberalization, Institute of Developing Economies Working Paper. Spamann, Holger, 2008. ‘Law and Finance’ Revisited, SSRN Working Paper: http://ssrn.com/abstract=109552

Stein, Jeremy C., 1997. Internal Capital Markets and the Competition for Corporate Resources, Journal of Finance 52, 111-133. Stock, James H. and Jonathan H. Wright, 2000. GMM with Weak Identification, Econometrica 68, 1055-1096. Stulz, Rene M., 2005. The Limits of Financial Globalization, Journal of Finance 60, 1595-1638. UBS Investment Research, 2006. European Holding Companies, UBS Limited. Villalonga, Belen and Raphael Amit, 2006. How Do Family Ownership, Control and Management Affect Firm Value, Journal of Financial Economics 80, 385-417.

46

Table I. Description of Main Variables

Panel A: Country-Level Measures of Prevalence of Family Business Groups

% Group (% Pyramid)

The percentage of all listed firms in a market that belong to a family business group (that belong to a family business group and are controlled through a pyramid structure).

% Group MC (% Pyramid MC)

The percentage of total market capitalization held by firms that belong to a family business group (that belong to a group and are controlled through a pyramid structure).

Family Group Firms / Controlled Firms

The percentage of all controlled firms (that is, firms with a controlling shareholder) in a market that belong to a family business group.

Family Group Firms / Family-controlled Firms

The percentage of all family-controlled firms in a market that belong to a family business group.

Panel B: Country-Level Variables explaining the Prevalence of Family Business Groups

Block Premium The country average difference between the negotiated price for a controlling block of shares and the prevailing market price. Source: Dyck and Zingales (2004).

Governance Index A principal components aggregation of i) a measure of the minority shareholder rights (the Anti-director Rights index) from La Porta et al. (1997) and updated by Pagano and Volpin (2005), ii) a legal enforcement index incorporating the strength of the rule of law, regulatory quality and control of corruption from Kaufmann, Kraay, and Mastruzzi (2003), and iii) a survey variable measuring disclosure standards from the World Economic Forum’s Global Competitiveness Report 2003.

Newspaper The total average circulation (or copies printed) of daily newspapers per 1000 inhabitants as measured in the year 2000. Source: World Association of Newspapers.

Intra-Group Tax The stringency of a country’s tax law related to intra-group transactions, equal to the sum of four indicator variables. The first three respectively indicate whether the law regulates the following three intra-group tax minimization practices: (i) transfer pricing, (ii) the use of thin-capitalization companies, and (iii) registration of holding companies in tax havens. The last variable indicates if there is explicit reporting requirement for companies engaging in the above transactions. Source: Deloitte International Taxation Guide.

Log GDP (Log GDP Capita)

The natural logarithm of a country’s gross domestic products (gross domestic products scaled by total population) as of 2001. Source: World Bank.

Institutional Funds Total equity investments of banks, insurance companies, pension and mutual funds in a country scaled by domestic stock market’s capitalization, collected at various points during 2001-2002. Sources: OECD publications, websites of national regulators and international associations (European Federation for Retirement Provision, International Federation of Pension Funds Administrators, and Fédération Européenne des Fonds et Sociétés d'Investissement).

Political Stability Perceptions of the likelihood that the government will be destabilized or overthrown by unconstitutional or violent means. Source: World Bank.

Savings to GDP Total domestic savings scaled by a country’s gross domestic products as of 2001. Source: World Bank.

Consolidation An indicator variable equal to one if a parent firm can consolidate a subsidiary in which it has an ownership stake of less than 90 percent, and zero otherwise. Source: Price Waterhouse Coopers’ Corporate Taxes: A Worldwide Summary.

Takeover Index An index measuring the extent of fair and equitable treatment of all shareholders in the takeover process and the transparency of the process. Source: Nenova (2006).

Panel C: Firm-Level Variables

Group Indicator An indicator variable which equals one if a firm belongs to a family business group.

Direct Ownership (Direct Control Rights)

The percentage of issued shares (voting rights) of a group firm held directly by its parent firm(s) in the same group.

Ultimate CF Rights The ultimate cash-flow rights of a group firm attributable to the controlling shareholder.

47

Pyramid Layer An integer variable that counts the number of layers of listed firms that exist between a group firm and its ultimate controlling shareholder.

Bottom of Chain An indicator variable which equals one if a firm is at the bottom of a pyramid chain.

Q A proxy for Tobin’s Q, calculated as the sum of balance-date market value of equity, book value of preference shares, minority interests, and liabilities, divided by the book value of total assets as of 2002. Sources: Worldscope, Datastream, and Osiris.

ROA Earnings before interests, tax, depreciation, and amortization in 2002 scaled by the average of year-beginning and year-end total assets. Sources: Worldscope, Osiris.

Log Size The log of balance-date market capitalization. Sources: Datastream, Worldscope, and Osiris.

Asset Growth The pre-2002 five-year average of total assets growth rate. Sources: Worldscope and Osiris.

Debt Total interest-paying debt divided by total assets. Sources Worldscope and Osiris.

Dividend Yield The pre-2002 five-year average of dividend per share divided by share price. Source: Datastream and Osiris.

CAPEX Total capital expenditures divided by total assets. Sources: Worldscope and Osiris.

Log Analyst The natural logarithm of one plus the number of analysts covering a firm. Source: I/B/E/S.

Log Age The natural logarithm of the age (in years) of a firm from its incorporation date to 2002. Source: Worldscope and Osiris.

Intangibles Total intangible assets divided by total assets. Sources: Worldscope and Osiris.

Beta The systematic risk of a firm obtained from estimating the one-factor market model on the firm’s monthly stock returns in the five years prior to 2002. Sources: Datastream and Bloomberg.

Idiosyncratic Risk The standard error from estimating the one-factor market model on each firm’s monthly stock returns in the five years prior to 2002. Sources: Datastream and Bloomberg.

Media Indicator An indicator for firms that primarily operate in a media related industries (that is, those with three digit-SIC codes in the following ranges: 271-273, 482-484 and 781-782).

Real Estate Indicator An indicator for firms that primarily operate in the real estate/construction industries (that is, those with three-digit SIC codes in the following ranges: 152-154 and 651-655).

Index Return at Listing The annual market index return in the year of a firm’s listing. Source: Datastream.

48

Table II. Country-level Descriptive Statistics on Family Business Groups The table reports family business group statistics for each of the 45 sample countries. % Group is the percentage of listed firms that belong to a group. % Pyramid is the percentage of firms that belong to a group and are controlled through a pyramid. % Group MC (and % Pyramid MC) is the percentage of market capitalization held by group-controlled (and pyramid-controlled) firms. Pyramid Layer is the country average of the number of listed firms that separate group firms from their ultimate owner.

Country N No. of

Groups No. of

Group Firms% Group % Pyramid

% Group MC

% PyramidMC

Pyramid Layer

Argentina 57 2 6 10.53 3.51 7.03 1.03 0.33 Australia 1405 33 95 6.76 3.35 8.08 1.58 0.55 Austria 103 3 7 6.80 3.88 8.98 7.83 0.86 Belgium 164 14 38 23.17 11.59 26.58 19.80 0.71 Brazil 341 16 57 16.72 10.26 13.17 6.37 0.84 Canada 1220 18 55 4.51 2.46 12.90 6.71 0.62 Chile 170 17 71 41.76 22.94 42.18 22.26 0.79 Colombia 56 4 22 39.29 28.57 44.27 21.10 1.55 Czech 73 2 4 5.48 2.74 1.50 0.47 0.50 Denmark 184 6 13 7.07 1.63 19.44 0.88 0.23 Finland 169 7 19 11.24 5.33 3.10 0.92 0.58 France 801 29 83 10.36 5.87 9.02 2.85 0.94 Germany 816 30 73 8.95 4.04 5.23 0.88 0.53 Greece 263 16 50 19.01 11.79 18.81 3.72 0.64 Hong Kong 519 32 105 20.23 11.56 25.96 11.84 0.83 Hungary 41 2 4 9.76 4.88 1.50 0.03 0.50 India 659 56 189 28.68 10.02 20.89 4.10 0.36 Indonesia 330 25 83 25.15 13.94 48.20 18.63 0.87 Ireland 73 3 6 8.22 1.37 2.03 0.00 0.17Israel 226 19 90 39.82 26.11 23.25 11.96 1.27 Italy 291 16 53 18.21 8.59 25.86 17.10 0.85 Japan 3233 36 82 2.54 1.05 2.89 1.23 0.43 Korea 1355 85 277 20.44 10.63 56.64 48.76 0.64 Malaysia 998 52 167 16.73 8.92 38.40 19.82 0.68 Mexico 123 9 26 21.14 10.57 34.97 22.15 0.62 Netherlands 183 5 10 5.46 2.19 4.69 3.36 0.50 New Zealand 96 3 7 7.29 6.25 10.99 10.09 1.86 Norway 178 7 16 8.99 3.93 5.17 0.97 0.50 Pakistan 221 14 34 15.38 3.62 6.68 3.19 0.26 Peru 143 8 31 21.68 11.19 43.09 11.38 0.55 Philippines 221 30 98 44.34 16.74 30.08 8.96 0.45 Poland 137 5 10 7.30 2.92 3.44 0.50 0.50 Portugal 78 6 19 24.36 12.82 10.23 6.66 0.58 Singapore 627 19 68 10.85 7.02 42.53 16.82 0.84 South Africa 298 7 18 6.04 3.02 1.31 0.92 0.61 Spain 163 7 20 12.27 7.36 4.66 2.15 0.70 Sri Lanka 117 15 75 64.10 49.57 42.96 19.47 0.91 Sweden 294 13 56 19.05 9.52 25.61 16.61 0.52 Switzerland 298 5 11 3.69 1.01 0.78 0.06 0.27 Taiwan 637 41 147 23.08 14.44 41.33 20.67 0.66 Thailand 465 30 100 21.51 10.97 47.05 17.71 0.53 Turkey 250 25 102 40.80 18.80 42.20 16.90 0.53 UK 2369 15 37 1.56 0.76 1.48 1.04 0.49 USA 7562 87 227 3.00 0.90 3.04 0.69 0.31 Venezuela 32 1 2 6.25 3.13 0.38 0.00 0.50

49

Table III. Multivariate Regression Analysis of Cross-Country Variations in Family Business Groups The table reports OLS regression results for a sample of 44 countries. % Group is the percentage of listed firms in the market that belong to a group. % Pyramid is the percentage of listed firms that belong to a group and are controlled through a pyramid. Group Firms/Controlled Firms is the percentage of controlled firms that belong to a family business group. Group Firms/Family Firms is the percentage of all family-controlled firms that belong to a family business group. Block Premium is calculated by Dyck and Zingales (2004) based on the average difference between the negotiated price for a controlling block of shares and the prevailing market price. Governance Index is an index based on principle components weights of anti-director rights, an enforcement index and corporate disclosure. Newspaper is the average newspaper circulation per 1000 inhabitants. Log GDP per Capita is the natural logarithm of GDP per capita. Log GDP is the natural logarithm of GDP. Institutional Funds is total institutional equity investments scaled by stock market capitalization. Savings to GDP is total domestic savings scaled by GDP. Consolidation is an indicator variable constructed based on the ownership threshold at which subsidiaries can be consolidated into the parent for taxation purposes. Intra-Group Tax measures the stringency of tax laws related to intra-group transactions. Takeover Index is from Nenova (2006) and measures the extent to which target shareholders are treated fairly and equally in the takeover process. Heteroskedasticity-consistent standard errors are reported in parentheses.

% Group %Pyramid Group Firms/Controlled Firms Group Firms/Family Firms

(1) (2) (3) (4) (1) (2) (3) (1) (2) (3) (1) (2) (3) Intercept 0.668a 0.688a 0.730a 0.395b 0.446a 0.472a 0.242b 0.918a 0.959a 0.551a 1.043a 1.093a 0.682a

(0.156) (0.157) (0.163) (0.157) (0.131) (0.137) (0.100) (0.143) (0.163) (0.142) (0.146) (0.151) (0.151) Block Premium -0.001

(0.001) Governance Index 0.030 0.006 0.023 0.024c 0.008 0.020 0.051b 0.028 0.048b 0.042 0.020 0.040c

(0.023) (0.023) (0.022) (0.014) (0.015) (0.014) (0.024) (0.024) (0.019) (0.026) (0.021) (0.022) Newspaper -0.032 -0.002 0.023 -0.024 -0.004 0.015 -0.040 -0.011 0.029 -0.042 -0.011 0.027

(0.022) (0.026) (0.023) (0.016) (0.017) (0.015) (0.027) (0.034) (0.031) (0.027) (0.033) (0.030) Log GDP Per Capita -0.032c

(0.019) Log GDP -0.013 -0.024 -0.018 -0.013 -0.020c -0.017c -0.021 -0.032c -0.026c -0.032c -0.042b -0.037a

(0.015) (0.014) (0.012) (0.012) (0.011) (0.010) (0.019) (0.018) (0.013) (0.018) (0.017) (0.014) Institutional Funds -0.126a -0.086a -0.119b -0.116c

(0.034) (0.023) (0.058) (0.064) Savings to GDP -0.008a -0.005b -0.008b -0.009b

(0.003) (0.002) (0.004) (0.004) Political Stability -0.108a -0.076b -0.142a -0.140a

(0.036) (0.031) (0.042) (0.039) Consolidation 0.003 0.004 -0.038 0.005 0.009 -0.019 0.010 0.027 -0.013 0.035 0.025 -0.017 0.033

(0.032) (0.030) (0.029) (0.028) (0.020) (0.019) (0.020) (0.044) (0.039) (0.037) (0.047) (0.045) (0.040) Intra-Group Tax -0.031b -0.040a -0.050a -0.031b -0.022b -0.028a -0.016c -0.052a -0.062a -0.044a -0.055a -0.066a -0.047b

(0.012) (0.013) (0.015) (0.012) (0.008) (0.010) (0.008) (0.019) (0.020) (0.016) (0.020) (0.021) (0.017) Takeover Index -0.204a -0.250a -0.148 -0.188b -0.147b -0.081 -0.103 -0.338a -0.240b -0.250b -0.332a -0.224b -0.245b

(0.074) (0.083) (0.098) (0.092) (0.057) (0.070) (0.067) (0.105) (0.104) (0.096) (0.113) (0.100) (0.112) Adjusted R2 0.405 0.450 0.504 0.536 0.353 0.397 0.451 0.428 0.467 0.588 0.479 0.527 0.608

No. of observations 37 44 44 44 44 44 44 44 44 44 44 44 44 a, b, and c denote significance at the 1, 5, and 10 percent levels.

50

Table IV. Differences in Firm Characteristics between Group and Non-Group Firms and between Firms within the Same Group The table reports i) group vs. non-group comparisons, which are made between group firms and their matched industry peers as well as peers matched by market capitalization, and ii) within-group comparisons, which are made between firms at different layers of the same business groups. For each group, firms are classified into those that are directly controlled (Top), those that are at the absolute bottom of the pyramid (Bottom) and those that lie in between (Middle). Bottom of Chain firms are those held in a pyramid but do not control any other group firms. Q is the ratio of market value of total assets over book value of total assets. ROA is EBITDA divided by the average of year-beginning and year-end total assets. Total Assets and Market Cap is the $US value (in millions) of total assets and market capitalization. Asset Growth is the average annual growth rate of total assets based on the five years previous to 2002. Sales Growth is the average growth rate of net sales based on the five years previous to 2002. Debt is total interest-paying debt divided by total assets. Beta is the estimate of beta obtained from estimating the market model on a firm’s monthly stock returns over the five-year period before 2002. Idiosyncratic Risk is the regression standard error from the same market model. Age is firm age from incorporation. Analyst is the number of analysts covering a firm. Intangibles is the ratio of intangible assets to total assets. CAPEX is the ratio of capital expenditures over total assets. Dividend Yield is the ratio of dividend-per-share over share price. For these variables, Panel A reports their median differences and whether they are significantly different from zero, using the Wilcoxon Signed-Rank test. For ease of interpretation, Panel B also reports the mean differences (and whether they differ significantly from zero based on a t-test) for the variables where the median differences are predominantly zero but are significant. Panel C reports the mean differences for group placement and the following ownership variables. Direct Ownership is the percentage of issued shares of a group firm held directly by its parent firm(s) in the same group. Ultimate CF Rights is the ultimate cash-flow rights attributable to the controlling shareholder. Wedge is the difference between the percentage of voting rights held by group parent(s) and Ultimate CF Rights.

Group vs. Non-group Tests Within-Group Tests Industry Matched Size Matched

Group vs. Non- Group

Bottom vs. Non Group

Top vs. Non Group

Group vs. Non Group

Bottom vs. Non Group

Top vs. Non Group

Bottom

vs. Top

Bottom of Chain

vs. Top

Bottom vs.

Middle

Panel A: Median Differences for All Firm Characteristics

Q -0.019 -0.024 -0.026 -0.063a -0.033 -0.076a 0.024 0.032c 0.062b

ROA 0.268 0.634 0.477 -1.226a -0.209 -0.963a 0.508 0.406 1.184b

Total Assets 70.887a 33.185a 99.028a - - - -207.326a -209.012a -182.787a

Market Cap 19.912a 12.345a 21.800a - - - -40.411a -41.264a -24.832b

Asset Growth 0.096c -1.079 0.695a -1.616 -2.218b -1.201 -1.698a -1.606a -3.079b

Sales Growth 0.415a 0.096 1.340a -1.125 -0.956 -0.371c -0.589 -0.927b -0.227

Age 3.000a 0.000a 4.500a 2.000a 0.000 4.000a -6.750a -6.000a -7.000a

Debt 2.156a -0.040 3.764a 2.247a -0.658 3.922a -2.357a -2.431a -0.928

Beta 0.030a 0.030c 0.029b 1.2×10-5a -0.002 -0.001b -0.049b -0.048a -0.063

Idiosyncratic Risk -0.786a -0.680c -0.857a 0.053a -0.091 0.115b 0.796a 0.909a 1.468b

CAPEX 0.015a 0.099a 0.000a -0.180a 0.000a -0.159a 0.208b 0.116c 0.335b

Dividend Yield 0.000a 0.000a 0.000a 0.000a 0.000a 0.000a 0.000 0.000 0.000

Analyst 0.000a 0.000a 0.000a 0.000a 0.000a 0.000a 0.000a 0.000a 0.000a

Intangibles 0.000a 0.000a 0.000a 0.000a 0.000a 0.000a 0.000 0.000 0.000

Panel B: Mean Differences for Firm Characteristics with Predominantly Zero Median Differences

CAPEX 1.651a 2.443a 1.358a 1.693a 2.570a 1.454a 0.902c 0.725 0.515

Dividend Yield 0.748a 0.929a 0.628a 0.701a 0.891a 0.604a -0.031 -0.088 0.024

Analyst 2.583a 1.807a 2.714a 0.266a 0.377a 0.168 -2.016a -1.928a -1.188b

Intangibles 2.615a 2.407a 2.879a 3.115a 3.111a 3.387a 0.868 0.007 0.579

Panel C: Mean Differences for Selected Within-Group Ownership Characteristics Pyramid Layer 1.195a 1.103a 1.087a

Direct Ownership 8.790a 8.586a 4.379b

Ultimate CF Rights -18.754a -18.692a -6.212a

Wedge 26.378a 25.990a 10.621a

a, b, and c denote significance at the 1, 5, and 10 percent levels.

51

Table V. Characteristics of Firms at the Bottom of Family Business Groups The sample is composed of only group firms. The dependent is either the number of layers of listed firms that exist between a group firm and its ultimate controlling shareholder (Pyramid layer), or the indicator variable for whether a firm is at the bottom of a pyramid chain (Bottom of Chain). Log Size is the log of US-dollar market capitalization. Beta is the estimate of beta obtained from estimating the market model on a firm’s monthly stock returns over the five-year period before 2002. Idiosyncratic Risk is the regression standard error from the same market model. Debt is total interest-paying debt divided by total assets. Asset Growth is the ex-ante five-year average annual growth rate in asset. CAPEX is the ratio of capital expenditures over total assets. Dividend Yield is the ex-ante five-year average of the ratio of dividend-per-share over share price. Intangibles is the ratio of intangible assets to total assets. Log Age is the logarithm of firm age. Log Analyst is the logarithm of the number of analysts covering a firm plus one. All regressions include group-specific fixed effects and indicator variables for broad industry sectors (based on their first SIC digit). The Pyramid Layer models are fitted using the within regression estimator, and the Bottom of Chain models are fitted using conditional logit fixed-effects estimator. The standard errors (in parentheses) are corrected for cluster-specific heteroskedasticity, with a cluster being defined as a country in the sample.

Pyramid Layer Pr(Bottom of Chain=1)

Full Sample Emerging markets

Developed Markets

Full Sample Emerging markets

Developed Markets

Log Size -0.082a -0.078a -0.089a -0.496a -0.472a -0.532a

(0.015) (0.021) (0.021) (0.059) (0.072) (0.105) Beta -0.062c 0.021 -0.114a -0.294b -0.087 -0.485a

(0.034) (0.058) (0.042) (0.127) (0.184) (0.185) Idiosyncratic Risk 0.813a 0.499c 0.879b 2.096a 1.958b 2.148 (0.253) (0.268) (0.423) (0.781) (0.915) (1.487) Debt -0.305a -0.259b -0.320a -1.185a -1.176a -1.400b

(0.073) (0.109) (0.100) (0.309) (0.381) (0.558) Asset Growth -0.098b -0.187b -0.031 -0.115 -0.051 -0.157 (0.045) (0.085) (0.045) -0.122 -0.209 -0.157 CAPEX 0.226c 0.337b 0.049 1.701b 2.486b 0.231 (0.121) (0.152) (0.247) (0.848) (1.103) (2.098) Dividend Yield 0.464 0.514 1.386 -0.217 1.205 -4.733 (0.509) (0.530) (1.408) (2.253) (2.508) (5.053) Intangibles 0.231 0.239 0.375b 1.226c 0.882 1.353 (0.150) (0.260) (0.187) (0.635) (0.960) (0.843) Log Age -0.143a -0.167a -0.085a -0.508a -0.560a -0.437a

(0.030) (0.044) (0.031) (0.093) (0.132) (0.138) Log Analyst -0.013 -0.024 0.011 0.021 -0.060 0.115 (0.026) (0.035) (0.043) (0.097) (0.118) (0.174) Adjusted/Pseudo R2 0.550 0.562 0.589 0.202 0.170 0.255No. of observations 2454 1419 1035 2454 1419 1035 a, b, and c denote significance at the 1, 5, and 10 percent levels.

52

Table VI. OLS Regression of Within-Group Firm Differences and Performance Measures The sample is composed of only group firms. The dependent variable is the ratio of market value of total assets over book value of total assets (Q) or EBITDA scaled by the average of year-beginning and year-end total assets (ROA). For each group firm, Direct Ownership is the aggregate percentage of issued shares held directly by its immediate parent(s), that is, other group firms and/or the ultimate owner. Direct Control Rights is proportion of voting rights held directly by its immediate parent(s). Pyramid layer is an integer variable that counts the number of layers of listed firms that exist between a group firm and its ultimate controlling shareholder. Bottom of Chain is the indicator variable for whether a firm is at the bottom of a pyramid chain. Ultimate CF Rights is the ultimate cash-flow rights of a group firm attributable to the controlling shareholder. A logit transformation is applied to Direct Ownership, Direct Control Rights, and Ultimate CF Rights. Wedge is Direct Control Rights minus Ultimate CF Rights, measuring the aggregate separation of cash-flow rights from control rights. Dual-Class Wedge is Direct Control Rights minus Direct Ownership, reflecting separation of cash-flow rights and control rights due to the use of dual-class (or multiple-class) shares. Log Size is the log of US-dollar market capitalization. Beta is the estimate of beta obtained from estimating the market model on a firm’s monthly stock returns over the five-year period before 2002. Asset Growth is the ex-ante five-year average annual growth rate in total assets. Debt is total interest-paying debt divided by total assets. CAPEX is the ratio of capital expenditures over total assets. Dividend Yield is the ex-ante five-year average of the ratio of dividend-per-share over share price. Intangibles is the ratio of intangible assets to total assets. Log Age is the logarithm of firm age. Log Analyst is the logarithm of the number of analysts covering a firm plus one. Firm-level controls are lagged by one period in the ROA regression. All models include group-specific fixed effects and indicator variables for broad industry sectors (based on their first SIC digit), and are fitted using the within regression estimator. The standard errors (in parentheses) are corrected for cluster-specific heteroskedasticity, with a cluster defined as a group in the sample.

Q ROA

(1) (2) (3) (4) (5) (6) (1) (2) (3) (4) (5) (6)

Direct Ownership 0.037b 0.031c 0.030c -0.003 -0.005 -0.008 (0.018) (0.018) (0.018) (0.008) (0.008) (0.008) Pyramid Layer 0.059b 0.073b 0.064a 0.015b 0.015c 0.014b (0.024) (0.035) (0.023) (0.006) (0.008) (0.006) Bottom of Chain 0.092a 0.023b (0.031) (0.012) Ultimate CF Rights -0.032c -0.012a (0.019) (0.005) Direct Control Rights 0.041b 0.003 (0.016) (0.008) Wedge -0.049 -0.003 (0.111) (0.044) Dual-Class Wedge -0.234c 0.021 (0.135) (0.064) Log Size 0.096a 0.101a 0.105a 0.099a 0.101a 0.102a 0.011 0.012c 0.012c 0.012c 0.012c 0.012c (0.018) (0.017) (0.018) (0.017) (0.017) (0.017) (0.007) (0.007) (0.007) (0.007) (0.007) (0.007) Beta -0.044 -0.043 -0.040 -0.045 -0.044 -0.043 (0.036) (0.036) (0.036) (0.036) (0.036) (0.036) (continued next page)

53

Debt 0.356a 0.373a 0.375a 0.370a 0.379a 0.379a -0.024 -0.022 -0.022 -0.023 -0.022 -0.022 (0.104) (0.104) (0.104) (0.104) (0.104) (0.104) (0.043) (0.043) (0.043) (0.043) (0.043) (0.043) Asset Growth 0.049 0.055 0.051 0.053 0.054 0.053 -0.011 -0.009 -0.010 -0.010 -0.009 -0.009 (0.050) (0.050) (0.049) (0.050) (0.050) (0.050) (0.014) (0.014) (0.014) (0.014) (0.014) (0.014) CAPEX 0.579c 0.566c 0.558c 0.563c 0.568c 0.568c 0.442 0.436 0.433 0.437 0.436 0.436 (0.341) (0.343) (0.338) (0.340) (0.341) (0.342) (0.269) (0.270) (0.266) (0.269) (0.270) (0.270) Dividend Yield -0.658 -0.666 -0.626 -0.673 -0.615 -0.629 0.347b 0.341b 0.349b 0.333b 0.334b 0.334b (0.614) (0.620) (0.610) (0.619) (0.619) (0.624) (0.155) (0.156) (0.157) (0.155) (0.157) (0.156) Intangibles 0.013 -0.002 -0.004 0.005 -0.002 -0.006 -0.037 -0.041 -0.043 -0.040 -0.040 -0.040 (0.228) (0.224) (0.226) (0.225) (0.220) (0.221) (0.051) (0.050) (0.052) (0.051) (0.050) (0.050) Log Age -0.050c -0.043 -0.041 -0.048 -0.046 -0.046 -0.003 -0.001 -0.001 -0.002 -0.001 -0.001 -0.029 -0.029 -0.03 -0.029 -0.029 -0.029 -0.008 -0.008 -0.008 -0.008 -0.008 -0.008 Log Analyst -0.035 -0.035 -0.035 -0.038 -0.041 -0.041 0.012 0.012 0.012 0.012 0.013 0.013 (0.028) (0.028) (0.028) (0.028) (0.028) (0.028) (0.010) (0.011) (0.010) (0.011) (0.011) (0.011) Adjusted R2 0.493 0.495 0.495 0.495 0.494 0.494 0.541 0.542 0.542 0.543 0.541 0.541 No. of observations 2454 2454 2454 2454 2454 2454 2409 2409 2409 2409 2409 2409 a, b, and c denote significance at the 1, 5, and 10 percent levels.

54

Table VII. Family Group Membership and Firm Performance Measures – OLS regression The sample is composed of both group and non-group firms. The dependent variable is the ratio of market value of total assets over book value of total assets (Q) or EBITDA scaled by the average of year-beginning and year-end total assets (ROA). Group Indicator is an indicator variable that equals one if a firm belongs to a family-controlled group and zero otherwise. Log Size is the log of US-dollar market capitalization. Beta is the estimate of beta obtained from running the market model on a firm’s monthly stock returns over 5-year period to 2002. Asset Growth is the five-year average annual growth rate in asset. Debt is total interest-paying debt divided by total assets. CAPEX is the ratio of capital expenditures over total assets. Dividend Yield is the ex-ante five-year average of the ratio of dividend-per-share over share price. Intangibles is the ratio of intangible assets to total assets. Log Age is the logarithm of firm age. Log Analyst is the logarithm of the number of analysts covering a firm plus one. All regression models include indicator variables for industry sectors (based on their first SIC digit) and country fixed effects. Firm-level controls are lagged by one period in the ROA regression. Each of the specifications is estimated first on the full sample, and then on the sample of family-controlled firms. The standard errors (in parentheses) are corrected for cluster-specific heteroskedasticity, with a cluster being defined as a country in the sample.

Q ROA

Full sample Family-firm sample

Full sample Family-firm sample

Group Indicator -0.151a -0.153a -0.225a -0.011b -0.010c -0.029a

(0.041) (0.044) (0.042) (0.005) (0.006) (0.006)

Log Size 0.138a 0.123a 0.166a 0.025a 0.026a 0.025a

(0.020) (0.019) (0.016) (0.003) (0.004) (0.004)

Beta 0.048c 0.051b 0.035 0.010 0.010 0.008

(0.026) (0.025) (0.036) (0.013) (0.014) (0.008)

Debt 0.559b 0.578b 0.641b 0.010 0.014 0.024

(0.259) (0.258) (0.265) (0.020) (0.021) (0.021)

Asset Growth 0.117b 0.107c 0.116 0.089c 0.104c 0.114c

(0.053) (0.054) (0.100) (0.049) (0.052) (0.057)

CAPEX 0.829a 0.842a 0.924a 0.799a 0.894a 0.556a

(0.250) (0.236) (0.263) (0.182) (0.181) (0.146)

Dividend Yield -3.492a -3.834a -2.409a 0.032b 0.031b 0.026c

(0.891) (0.881) (0.727) (0.014) (0.014) (0.014)

Intangibles -0.117c -0.045 -0.025 -0.127a -0.127a -0.148a

(0.065) (0.087) (0.100) (0.032) (0.034) (0.021)

Log Age -0.207a -0.211a -0.226a 0.010 0.010c 0.014c

(0.053) (0.051) (0.053) (0.006) (0.006) (0.008)

Log Analyst -0.211c -0.188c -0.185b -0.011b -0.010c -0.029a

(0.110) (0.100) (0.081) (0.005) (0.006) (0.006)

Country fixed effects YES NO YES YES NO YES

Country-level factors NO YES NO NO YES NO

Adjusted R2 0.133 0.119 0.147 0.183 0.171 0.193

No. of observations 22194 22154 9508 22070 22070 9530 a, b, and c denote significance at the 1, 5, and 10 percent levels.

55

Table VIII. Group Membership and Firm Performance – Treatment-Effect Regression The sample contains both group and non-group firms. Q is the ratio of market value of assets over book value of assets. ROA is EBITDA scaled by the average of year-beginning and year-end total assets. Group Indicator is equal to one if a firm belongs to a group, and zero otherwise. Log Size is the log of US-dollar market capitalization. Beta is the beta estimate from running the market model on a firm’s monthly stock returns over the five-year period before 2002, and Idiosyncratic Risk is the regression standard error. Asset Growth is the ex-ante five-year average annual asset growth rate. Debt, CAPEX, and Intangibles are total debt, capital expenditures, and intangible assets divided by total assets. Dividend Yield is the ex-ante five-year average of dividend-per-share over share price. Log Age is the log of firm age. Log Analyst is the log of the number of analysts plus one. Media (Real Estate) Indicator is the indicator for media (real estate) related industries. Intra-Group Tax is the stringency of tax laws related to intra-group transactions. Takeover Index is the extent to which regulations dictate fair and equitable treatment of shareholders in the takeover process. Index Return at Listing is the index return during the listing year of a firm. All models include sector indicator variables. Firm-level controls are lagged by one period in the ROA regression. Country-cluster standard errors are reported in parentheses. Treatment-effects models (last four columns) are estimated using maximum likelihood.

Pr(Group Indicator=1) Q

ROA

Model (1)

Model (2) Model (1) as

treatment eq. Model (2) as treatment eq.

Model (1) as treatment eq.

Model (2) as treatment eq.

Estimates from the outcome equation Group Indicator 1.796a 1.959a -0.352a -0.399a (0.279) (0.313) (0.058) (0.062) Log Size 0.076a 0.077a 0.107a 0.128a 0.024a 0.028a (0.022) (0.021) (0.020) (0.023) (0.004) (0.003) Beta 0.000 0.009 0.051c 0.071b (0.028) (0.026) (0.027) (0.029) Debt 0.176a 0.189a 0.499c 0.585b 0.055b 0.061b (0.053) (0.045) (0.256) (0.290) (0.026) (0.027) Asset Growth -0.020 -0.001 0.127c 0.167b 0.118b 0.086c (0.031) (0.029) (0.066) (0.075) (0.053) (0.045) CAPEX -0.050 -0.003 0.824a 0.919a 0.018 0.019 (0.200) (0.213) (0.221) (0.293) (0.021) (0.021) Dividend Yield -0.133 -0.421 -3.678a -3.005a 0.872a 0.802a (0.570) (0.562) (0.801) (0.839) (0.173) (0.183) Intangibles 0.048 -0.063 -0.134c -0.192 -0.129a -0.133a (0.124) (0.115) (0.081) (0.134) (0.027) (0.025) Log Age 0.060c 0.019 -0.219a -0.181a 0.020a 0.029a (0.032) (0.035) (0.039) (0.036) (0.007) (0.009) Log Analyst 0.001 0.007 -0.203b -0.284b 0.035b 0.038b (0.045) (0.040) (0.090) (0.122) (0.014) (0.016) Country-level factors YES YES YES YES YES YES

Instruments Estimates of instruments from treatment equation Idiosyncratic Risk 0.668c 0.594c 0.611a 0.466a 0.450c 0.387c (0.357) (0.339) (0.165) (0.145) (0.255) (0.221) Media Indicator 0.429a 0.434a 0.237a 0.254a 0.204b 0.189c (0.109) (0.140) (0.056) (0.056) (0.082) (0.110) Real Estate Indicator 0.166c 0.186c 0.007 0.027 0.064 0.108 (0.094) (0.109) (0.052) (0.045) (0.069) (0.074) Intra-Group Tax -0.104c -0.110c -0.024 -0.023 -0.072c -0.070c (0.059) (0.061) (0.019) (0.018) (0.040) (0.040) Takeover Index -1.745a -1.524a -0.330b -0.133 -0.596b -0.552b (0.411) (0.346) (0.136) (0.119) (0.266) (0.223) Index Return at Listing -0.011b -0.010a -0.018a (0.006) (0.002) (0.006) Pseudo R2 0.133 0.123 Wald test of indep. eqs 188.003a 253.954a 233.658a 410.143a No. of observations 22163 14375 22163 14375 22005 14059 a, b, and c denote significance at the 1, 5, and 10 percent levels.

56

Figure 1. The Arab Malaysian Group, controlled by Azman Hashim.