User Tools

Site Tools


Swiss Study Shows 147 “Super Entities” Rule the World


By Susanne Posel Occupy Corporatism

The Swiss Federal Institute (SFI) in Zurich released a study entitled “The Network of Global Corporate Control” that proves a small consortium of corporations – mainly banks – run the world. Overlooking Milintel a mere 147 corporations which form a “super entity” have control 40% of the world’s wealth; which is the real economy. These mega-corporations are at the center of the global economy. The banks found to be most influential include:

  • Barclays
  • Goldman Sachs
  • JPMorgan Chase & Co
  • Vanguard Group
  • UBS
  • Deutsche Bank
  • Bank of New York Melon Corp
  • Morgan Stanley
  • Bank of America Corp
  • Société Générale

However as the connections to the controlling groups are networked throughout the world, they become the catalyst for global financial collapse.

James Glattfelder, complex systems theorist at the SFI explains: “In effect, less than one per cent of the companies were able to control 40 per cent of the entire network.”

Using mathematic models normally applied to natural systems, the researchers analyzed the world’s economy. Their data was taken from Orbis 2007, a database which lists 37 million corporations and investors. The evidence showed that the world’s largest corporations are interconnected to all other companies and their professional decisions affect all markets across the globe.

George Sudihara, complex systems expert for SFI claims that this phenomenon is a common structure that could be found in nature. Comparing the manufactured reality of the financial markets to the ecosystems of the planet, Sudihara says that although the 147 corporations that rule the world through influence and interconnectedness are no more harmful than the natural cycles of our weather or animal kingdoms.

Yet because of the facts presented in the study, the financial crashes of the last century can be traced back to these tightly-knit networks. Future disasters can also be projected based on this analysis because of the “connectedness” of these influential entities which are only 147 corporations.

It is often promoted that global capitalism could be useful for making the markets more stable by simply acquiescing more control to these super-entities, while their may be an intentional methodology of boom and busts to maintain prominance. Many do and many do not recognize that there is a complex architecture of global economic power that causes financial systems to ebb and flow and crash and burn.

As the banking cartels force countries in the EuroZone into sovereign debt, there is a weakening of the entire business community globally and of course less jobs and cash in pocket of the free man. Big business, Wells Fargo, JPMorgan Chase and others as well as the aristocratic factions will financially gain and hoard their extortions in tax exempt havens and will include land and asset grabs on the cheap!

The sovereign land-grab by the central banking cartels across Europe is mirrored in a recent Goldman Sachs report: “The more the Spanish administration indulges domestic political interests … the more explicit conditionality is likely to be demanded.” in the form of interest payable bail-outs payable by workers in taxes to the coffers of a handful of players. The European Central Bankers agreed to give any nation in the Euro-Zone a bailout if they agreed to hand over the country to them under the guise of “new rules and conditions when applying for assistance.”

As America drifts downstream toward economic implosion, the Federal Reserve another Central Bank unveiled QE3 last week as a pump and dump scheme to prop up the US dollar by printing cash and sending it to European Central Banks as taxable bail-out money. As well as purchasing all of the mortgage-backed securities from the same banks that created the economic collapse and acquiring land in massive land-grabs; the likes of which have never been seen in the US.

In this scenario, “The Central Banks” are really a front to an ultra-clandestine clique of individuals including aristocratic elements who are re-sorting the landscape.

The Network Of Global Corporate Control

Stefania Vitali1, James B. Glattfelder[1], and Stefano Battiston[1]

1: Chair of Systems Design, ETH Zurich, Kreuzplatz 5, 8032 Zurich, Switzerland, corresponding author, email:


The structure of the control network of transnational corporations affects global market competition and financial stability. So far, only small national samples were studied and there was no appropriate methodology to assess control globally. We present the first investigation of the architecture of the international ownership network, along with the computation of the control held by each global player. We find that transnational corporations form a giant bow-tie structure and that a large portion of control flows to a small tightly-knit core of financial institutions. This core can be seen as an economic “super-entity” that raises new important issues both for researchers and policy makers.


A common intuition among scholars and in the media sees the global economy as being dominated by a handful of powerful transnational corporations (TNCs). However, this has not been confirmed or rejected with explicit numbers. A quantitative investigation is not a trivial task because firms may exert control over other firms via a web of direct and indirect ownership relations which extends over many countries. Therefore, a complex network analysis [1] is needed in order to uncover the structure of control and its implications. Recently, economic networks have attracted growing attention [2], e.g., networks of trade [3], products [4], credit [5, 6], stock prices [7] and boards of directors [8, 9]. This literature has also analyzed ownership networks [10, 11], but has neglected the structure of control at a global level. Even the corporate governance literature has only studied small national business groups [12]. Certainly, it is intuitive that every large corporation has a pyramid of subsidiaries below and a number of shareholders above. However, economic theory does not offer models that predict how TNCs globally connect to each other. Three alternative hypotheses can be formulated. TNCs may remain isolated, cluster in separated coalitions, or form a giant connected component, possibly with a core-periphery structure. So far, this issue has remained unaddressed, notwithstanding its important implications for policy making. Indeed, mutual ownership relations among firms within the same sector can, in some cases, jeopardize market competition [13, 14]. Moreover, linkages among financial institutions have been recognized to have ambiguous effects on their financial fragility [15, 16]. Verifying to

Figure 1 CGC

Figure 1: Ownership and Control. (A&B) Direct and indirect ownership. (A) Firm i has Wij percent of direct ownership in firm j. Through j, it has also an indirect ownership in k and l. (B) With cycles one has to take into account the recursive paths, see SI Appendix, Sec. 3.1. (C&D) Threshold model. (C) Percentages of ownership are indicated along the links. (D) If a shareholder has ownership exceeding a threshold (e.g. 50%), it has full control (100%) and the others have none (0%). More conservative model of control are also considered see SI Appendix, Sec. 3.1.

what extent these implications hold true in the global economy is per se an unexplored field of research and is beyond the scope of this article. However, a necessary precondition to such investigations is to uncover the worldwide structure of corporate control. This was never performed before and it is the aim of the present work.


Ownership refers to a person or a firm owning another firm entirely or partially. Let W denote the ownership matrix, where the component Wij ∈ [0, 1] is the percentage of ownership that the owner (or shareholder ) i holds in firm j. This corresponds to a directed weighted graph with firms represented as nodes and ownership ties as links. If, in turn, firm j owns Wjl shares of firm l, then firm i has an indirect ownership of firm l (Fig. 1 A). In the simplest case, this amounts trivially to the product of the shares of direct ownership Wij Wjl . If we now consider the economic value v of firms (e.g., operating revenue in USD), an amount Wijvj is associated to i in the direct case, and WijWjlvl in the indirect case. This computation can be extended to a generic graph, with some important caveats [17, SI Appendix, Secs. 3.1 and 3.2].

Each shareholder has the right to a fraction of the firm revenue (dividend) and to a voice in the decision making process (e.g., voting rights at the shareholder meetings). Thus the larger the ownership share Wij in a firm, the larger is the associated control over it, denoted as Cij .Intuitively, control corresponds to the chances of seeing one’s own interest prevailing in the business strategy of the firm. Control Cij is usually computed from ownership Wij with a simple threshold rule: the majority shareholder has full control. In the example of Fig. 1 C, D, this yields Cijvj = 1vj in the direct case and Cij Cjl vl = 0 in the indirect case. As a robustness check, we tested also more conservative models where minorities keep some control (see SI Appendix, Sec. 3.1). In analogy to ownership, the extension to a generic graph is the notion of network control : cneti = ∑j Cijvj + ∑j Cij cnet. This sums up the value controlled by i through its shares in j, plus the value controlled indirectly via the network control of j. Thus, network control has the meaning of the total amount of economic value over which i has an influence (e.g. cinet = vj + vk in Fig. 1D).

Because of indirect links, control flows upstream from many firms and can result in some shareholders becoming very powerful. However, especially in graphs with many cycles (see Figs. 1B and S4), the computation of cnet , in the basic formulation detailed above, severely overestimates the control assigned to actors in two cases: firms that are part of cycles (or cross-shareholding structures), and shareholders that are upstream of these structures. An illustration of the problem on a simple network example, together with the details of the method are provided in SI Appendix, Secs. 3.2 – 3.4. A partial solution for small networks was provided in [18]. Previous work on large control networks used a different network construction method and neglected this issue entirely [11, SI Appendix, Secs. 2 and 3.5]. In this paper, by building on [11], we develop a new methodology to overcome the problem of control overestimation, which can be employed to compute control in large networks.


We start from a list of 43060 TNCs identified according to the OECD definition, taken from a sample of about 30 million economic actors contained in the Orbis 2007 database (see SI Appendix, Sec. 2). We then apply a recursive search (Fig. S1 and SI Appendix, Sec. 2) which singles out, for the first time to our knowledge, the network of all the ownership pathways originating from and pointing to TNCs (Fig. S2). The resulting TNC network includes 600508 nodes and 1006987 ownership ties.

Figure 2

Figure 2: Network topology. (A) A bow-tie consists of in-section (IN), out-section (OUT), strongly connected component or core (SCC), and tubes and tendrils (T&T). (B) Bow-tie structure of the largest connected component (LCC) and other connected components (OCC). Each section volume scales logarithmically with the share of its TNCs operating revenue. In parenthesis, percentage of operating revenue and number of TNCs, cfr. Table 1. (C) SCC layout of the SCC (1318 nodes and 12191 links). Node size scales logarithmically with operation revenue, node color with network control (from yellow to red). Link color scales with weight. (D) Zoom on some major TNCs in the financial sector. Some cycles are highlighted.

Notice that this data set fundamentally differs from the ones analysed in [11] (which considered only listed companies in separate countries and their direct shareholders). Here we are interested

Table 1: Bow-tie statistics. Percentage of total TNC operating revenue (OR) and number (#) of nodes in the sections of the bow-tie (acronyms are in Fig. 2). Economic actors types are: shareholders (SH), participated companies (PC).

in the true global ownership network and many TNCs are not listed companies (see also SI Appendix, Sec. 2).

Network Topology

The computation of control requires a prior analysis of the topology. In terms of connectivity, the network consists of many small connected components, but the largest one (3/4 of all nodes) contains all the top TNCs by economic value, accounting for 94.2% of the total TNC operating revenue (Tbl. 1). Besides the usual network statistics (Figs. S5, S6), two topological properties are the most relevant to the focus of this work. The first is the abundance of cycles of length two (mutual cross-shareholdings) or greater (Fig. S7 and SI Appendix, Sec. 7), which are well studied motifs in corporate governance [19]. A generalization is a strongly connected component (SCC), i.e., a set of firms in which every member owns directly and/or indirectly shares in every other member. This kind of structures, so far observed only in small samples, has explanations such as anti-takeover strategies, reduction of transaction costs, risk sharing, increasing trust and groups of interest [20]. No matter its origin, however, it weakens market competition [13, 14]. The second characteristics is that the largest connect component contains only one dominant strongly connected component (1347 nodes). Thus, similar to the WWW, the TNC network has a bow-tie structure [21] (see Fig. 2 A and SI Appendix, Sec. 6). Its peculiarity is that the strongly connected component, or core, is very small compared to the other sections of the bow-tie, and that the out-section is significantly larger than the in-section and the tubes and tendrils (Fig. 2 B and Tbl. 1). The core is also very densely connected, with members having, on average, ties to 20 other members (Fig. 2 C, D). As a result, about 3/4 of the ownership of firms in the core remains in the hands of firms of the core itself. In other words, this is a tightly-knit group of corporations that cumulatively hold the majority share of each other. Notice that the cross-country analysis of [11] found that only a few of the national ownership networks are bow-ties, and, importantly, for the Anglo-Saxon countries, the main strongly connected components are big compared to the network size.

Concentration of Control

The topological analysis carried out so far does not consider the diverse economic value of firms. We thus compute the network control that economic actors (including TNCs) gain over the TNCs’ value (operating revenue) and we address the question of how much this control is concentrated and who are the top control holders. See Fig. S3 for the distribution of control and operating revenue.

It should be noticed that, although scholars have long measured the concentration of wealth and income [22], there is no prior quantitative estimation for control. Constructing a Lorenz-like curve (Fig. 3) allows one to identify the fraction η ∗ of top holders holding cumulatively 80% of the total network control. Thus, the smaller this fraction, the higher the concentration. In principle, one could expect inequality of control to be comparable to inequality of income across households and firms, since shares of most corporations are publicly accessible in stock markets. In contrast, we find that only 737 top holders accumulate 80% of the control over the value of all TNCs (see also the list of the top 50 holders in Tbl. S1 of SI Appendix, Sec. 8.3). The corresponding level of concentration is η*1 = 0.61%, to be compared with η*2 = 4.35% for operating revenue. Other sensible comparisons include: income distribution in developed countries with η*3 ∼ 5%−10% [22] and corporate revenue in Fortune1000 (η*4 ∼ 30% in 2009). This means that network control is much more unequally distributed than wealth. In particular, the top ranked actors hold a control ten times bigger than what could be expected based on their wealth. The results are robust with respect to the models used to estimate control, see Fig. 3 and Tbls. S2, S3.


The fact that control is highly concentrated in the hands of few top holders does not determine if and how they are interconnected. It is only by combining topology with control ranking that we obtain a full characterization of the structure of control. A first question we are now able to answer is where the top actors are located in the bow-tie. As the reader may by now suspect, powerful actors tend to belong to the core. In fact, the location of a TNC in the network does matter. For instance, a randomly chosen TNC in the core has about 50% chance of also being among the top holders, compared to, e.g., 6% for the in-section (Tbl. S4). A second question concerns what share of total control each component of the bow-tie holds. We find that, despite its small size, the core holds collectively a large fraction of the total network control. In detail, nearly 4/10 of the control over the economic value of TNCs in the world is held, via a complicated web of ownership relations, by a group of 147 TNCs in the core, which has almost full control over itself. The top holders within the core can thus be thought of as an economic “super-entity” in the global network of corporations. A relevant additional fact at this point is that 3/4 of the core are financial intermediaries. Fig. 2 D shows a small subset of well-known financial players and their links, providing an idea of the level of entanglement of the entire core.

Figure 3

Figure 3: Concentration of network control and operating revenue. Economic actors (TNCs and shareholders) are sorted by descending importance, as given by cnet . A data point located at (η, θ) corresponds to a fraction η of top economic actors cumulatively holding the fraction θ of network control, value or operating revenue. The different curves refer to network control computed with three models (LM, TM, RM), see SI Appendix, Sec. 3.1, and operating revenue. The horizontal line denotes a value of θ equal to 80%. The level of concentration is determined by the η value of the intersection between each curve and the horizontal line. The scale is semi-log.

This remarkable finding raises at least two questions that are fundamental to the understanding of the functioning of the global economy. Firstly, what are the implication for global financial stability? It is known that financial institutions establish financial contracts, such as lending or credit derivatives, with several other institutions. This allows them to diversify risk, but, at the same time, it also exposes them to contagion [15]. Unfortunately, information on these contracts is usually not disclosed due to strategic reasons. However, in various countries, the existence of such financial ties is correlated with the existence of ownership relations [23]. Thus, in the hypothesis that the structure of the ownership network is a good proxy for that of the financial network, this implies that the global financial network is also very intricate. Recent works have shown that when a financial network is very densely connected it is prone to systemic risk [24, 16]. Indeed, while in good times the network is seemingly robust, in bad times firms go into distress simultaneously. This knife-edge property [25, 26] was witnessed during the recent financial turmoil.

Secondly, what are the implications for market competition? Since many TNCs in the core have overlapping domains of activity, the fact that they are connected by ownership relations could facilitate the formation of blocs, which would hamper market competition [14]. Remarkably, the existence of such a core in the global market was never documented before and thus, so far, no scientific study demonstrates or excludes that this international “super-entity” has ever acted as a bloc. However, some examples suggest that this is not an unlikely scenario. For instance, previous studies have shown how even small cross-shareholding structures, at a national level, can affect market competition in sectors such as airline, automobile and steel, as well as the financial one [14, 13]. At the same time, antitrust institutions around the world (e.g., the UK Office of Fair Trade) closely monitor complex ownership structures within their national borders. The fact that international data sets as well as methods to handle large networks became available only very recently, may explain how this finding could go unnoticed for so long.

Two issues are worth being addressed here. One may question the idea of putting together data of ownership across countries with diverse legal settings. However, previous empirical work shows that of all possible determinants affecting ownership relations in different countries (e.g., tax rules, level of corruption, institutional settings, etc.), only the level of investor protection is statistically relevant [27]. In any case, it is remarkable that our results on concentration are robust with respect to three very different models used to infer control from ownership. The second issue concerns the control that financial institutions effectively exert. According to some theoretical arguments, in general, financial institutions do not invest in equity shares in order to exert control. However, there is also empirical evidence of the opposite [23, SI Appendix, Sec. 8.1]. Our results show that, globally, top holders are at least in the position to exert considerable control, either formally (e.g., voting in shareholder and board meetings) or via informal negotiations.

Beyond the relevance of these results for economics and policy making, our methodology can be applied to identify key nodes in any real-world network in which a scalar quantity (e.g., resources or energy) flows along directed weighted links. From an empirical point of view, a bow-tie structure with a very small and influential core is a new observation in the study of complex networks. We conjecture that it may be present in other types of networks where “rich-get-richer” mechanisms are at work (although a degree preferential-attachment [1] alone does not produce a bow-tie). However, the fact that the core is so densely connected could be seen as a generalization of the “rich-club phenomenon” with control in the role of degree [28, 3, SI Appendix, Sec. 8.2]. These related open issues could be possibly understood by introducing control in a “fitness model” [29] of network evolution.


We acknowledge F. Schweitzer and C. Tessone for valuable feedback, D. Garcia for generating the 3D figures, and the program Cuttlefish used for networks layout.

Authors acknowledge the financial support from: the ETH Competence Center “Coping with Crises in Complex Socio-Economic Systems” (CCSS) through ETH Research Grant CH1-01-08-2; the European Commission FP7 FET Open Project “FOC” No. 255987.


  1. Barabási A, Albert R (1999) Emergence of scaling in random networks. Science 286: 509.
  2. Schweitzer F, Fagiolo G, Sornette D, Vega-Redondo F, Vespignani A, et al. (2009) Economic networks: The new challenges. Science 325: 422-425.
  3. Fagiolo G, Reyes J, Schiavo S (2009) World-trade web: Topological properties, dynamics, and evolution. Phys Rev E 79: 36115.
  4. Hidalgo C, Hausmann R (2009) The building blocks of economic complexity. Proc Natl Acad Sci 106: 10570.
  5. Boss M, Elsinger H, Summer M, Thurner S (2004) Network topology of the interbank market. Quant Financ 4: 677–684.
  6. Iori G, De Masi G, Precup O, Gabbi G, Caldarelli G (2008) A network analysis of the Italian overnight money market. J Econ Dyn Control 32: 259–278.
  7. Bonanno G, Caldarelli G, Lillo F, Mantegna RN (2003) Topology of correlation-based minimal spanning trees in real and model markets. Phys Rev E 68: 046130.
  8. Strogatz S (2001) Exploring complex networks. Nature 410: 268–276.
  9. Battiston S, Catanzaro M (2004) Statistical properties of corporate board and director networks. Eur Phys J B 38: 345–352.
  10. Kogut B, Walker G (2001) The small world of germany and the durability of national networks. Amer Sociol Rev 66: 317–335.
  11. Glattfelder JB, Battiston S (2009) Backbone of complex networks of corporations: The flow of control. Phys Rev E 80.
  12. Granovetter M (1995) Ind. Corp. Change, Oxford University Press, chapter Coase Revisited: Business Groups in the Modern Economy.
  13. O’Brien D, Salop S (1999) Competitive Effects of Partial Ownership: Financial Interest and Corporate Control. Antitrust Law J 67: 559.
  14. Gilo D, Moshe Y, Spiegel Y (2006) Partial cross ownership and tacit collusion. RAND J Econ 37: 81–99.
  15. Allen F, Gale D (2000) Financial contagion. J Polit Econ 108: 1–33.
  16. Stiglitz JE (2010) Risk and global economic architecture: Why full financial integration may be undesirable. Working Paper 15718, NBER.
  17. Brioschi F, Buzzacchi L, Colombo M (1989) Risk capital financing and the separation of ownership and control in business groups. J Bank Financ 13: 747–772.
  18. Baldone S, Brioschi F, Paleari S (1998) Ownership Measures Among Firms Connected by Cross-Shareholdings and a Further Analogy with Input-Output Theory. 4th JAFEE Inter-national Conference on Investment and Derivatives.
  19. Dietzenbacher E, Temurshoev U (2008) Ownership relations in the presence of cross-shareholding. J Econ 95: 189–212.
  20. Williamson O (1975) Markets and hierarchies, analysis and antitrust implications: a study in the economics of internal organization. Free Press, New York.
  21. Broder A, Kumar R, Maghoul F, Raghavan P, Rajagopalan S, et al. (2000) Graph structure in the Web. Comput Netw 33: 309–320.
  22. Atkinson A, Bourguignon F (2000) Handbook of income distribution. Elsevier.
  23. Santos J, Rumble A (2006) The American keiretsu and universal banks: Investing, voting and sitting on nonfinancials’ corporate boards. J Finan Econ 80: 419–454.
  24. Battiston S, Delli Gatti D, Gallegati M, Greenwald B, Stiglitz J (2009) Liaisons dangereuses: Increasing connectivity, risk sharing and systemic risk. Working Paper 15611, NBER.
  25. Alesandri P, Haldane A (2009). Banking on the state. Speech given at the Bank of England. Available at
  26. May R, Levin S, Sugihara G (2008) Ecology for bankers. Nature 451: 893–895.
  27. La Porta R, de Silanes FL, Shleifer A (1999) Corporate ownership around the world. J Finance 54: 471–517.
  28. Colizza V, Flammini A, Serrano M, Vespignani A (2006) Detecting rich-club ordering in complex networks. Nat Phy 2: 110–115.
  29. Garlaschelli D, Capocci A, Caldarelli G (2007) Self-organized network evolution coupled to extremal dynamics. Nat Phys 3: 813–817.

Supporting Information:

The Network of Global Corporate Control

Stefania Vitali, James B. Glattfelder and Stefano Battiston Chair of Systems Design, ETH Zurich, Kreuzplatz 5, 8032 Zurich, Switzerland


  • 1 Acronyms and Abbreviations ………………………………………… 12
  • 2 Data and TNC Network Detection …………………………………….. 13
  • 3 Network Control ………………………………………………….. 16
  • 3.1 The Existing Methodology ………………………………………… 16
  • 3.2 The Algorithm: Computing Control While Remedying the Problems ……….. 18
  • 3.3 Proving the BFS Methodology Corrects for Cycles ……………………. 20
  • 3.4 An Illustrated Example ………………………………………….. 22
  • 3.5 Relations To Previous Work ………………………………………. 24
  • 4 Degree and Strength Distribution Analysis …………………………… 25
  • 5 Connected Component Analysis ………………………………………. 26
  • 6 Bow-Tie Component Sizes …………………………………………… 28
  • 7 Strongly Connected Component Analysis ………………………………. 29
  • 8 Network Control Concentration ……………………………………… 31
  • 8.1 Control of Financial Institutions ………………………………… 31
  • 8.2 Relation to the Rich Club Phenomenon ……………………………… 32
  • 8.3 Top Control-Holders Ranking ……………………………………… 32
  • 9 Additional Tables

1. Acronyms and Abbreviations

The list of acronyms and abbreviations used in the main text and this Supporting Online Material:

  • BFS: breadth-first search (search algorithm)
  • CC: (weakly) connected component
  • FS: financial sector
  • IN: in-section of a bow-tie
  • LCC: largest CC
  • LM: linear model (for estimating control from ownership; see also RM and TM)
  • NACE: (industry standard classification system )
  • OCC: other connected components (everything outside the LCC)
  • OECD: Organization for Economic Co-operation and Development
  • OR: operating revenue
  • OUT: out-section of a bow-tie
  • PC: participated company
  • RM: relative model (for estimating control from ownership; see also LM and TM)
  • SCC: strongly connected component (in the main text, this is synonymous with the core of the bow-tie in the LCC)
  • SH: shareholder (economic actors holding shares in TNCs)
  • TCH: top control-holder (list of TNCs and SHs that together hold 80% of the network control)
  • TM: threshold model (for estimating control from ownership; see also LM and RM)
  • TNC: transnational corporation (OECD definition)
  • T&T: tubes and tendrils (sections in a bow-tie that either connect IN and OUT, are outgoing from IN, or ingoing to OUT, respectively)

2. Data and TNC Network Detection

The Orbis 2007 marketing database∗ comprises about 37 million economic actors, both physical persons and firms located in 194 countries, and roughly 13 million directed and weighted ownership links (equity relations). Among many others, information on the industrial classification, geographical position and operating revenue of the actors are provided. This data set is intended to track control relationships rather than patrimonial relationships. Whenever available, the percentage of ownership refers to shares associated with voting rights. The definition of TNCs given by the OECD[1] states that they […] comprise companies and other entities established in more than one country and so linked that they may coordinate their operations in various ways, while one or more of these entities may be able to exercise a significant influence over the activities of others, their degree of autonomy within the enterprise may vary widely from one multinational enterprise to another. Ownership may be private, state or mixed.

Accordingly, we select those companies which hold at least 10% of shares in companies located in more than one country. However, many subsidiaries of large TNCs fulfill themselves this definition of TNCs (e.g. The Coca-Cola Company owns Coca-Cola Hellenic Bottling Company which in turn owns Coca-Cola Beverages Austria). Since for each multinational group we are interested in retaining only one representative, we exclude from the selection the companies for which the so-called ultimate owner (i.e., the owner with the highest share at each degree of ownership upstream of a company∗ ) is quoted in a the stock market. In substitution, we add the quoted ultimate owner to the list (if not already included). In the example above, this procedure identifies only the Coca-Cola Company as a TNC. Overall we obtain a list of 43060 TNCs located in 116 different countries, with 5675 TNCs quoted in stock markets.

Starting from the list of TNCs, we explore recursively the neighborhood of companies in the whole database. First, we proceed downstream of the TNCs (see Fig. S1) with a breadth-first search (BFS) and we identify all companies participated directly and indirectly by the TNCs. We then proceed in a similar way upstream identifying all direct and indirect shareholders of the TNCs. The resulting network can be divided into three classes of nodes, TNC, SH and PC, as shown in Fig. S2. The TNC network constructed in this way consists of 600508 economic entities and 1006987 corporate relations. Notice that it may be possible to reach a PC from several TNCs, or to reach a TNC from several SHs. In other words, paths proceeding downstream or upstream of the TNCs may overlap, giving rise to CCs of various sizes. It is worthwhile to distinguish the data set constructed here from the one analysed in [5], which was not obtained using a recursive search, but with the simple method of collecting only listed


Figure S1

Figure S1: Illustration of the first two steps in the recursive exploration downstream of a TNC. Starting from “Benetton Group” the BFS explores all the direct neighbors (A), and then the neighbors’ neighbors (B).

companies and their direct shareholders. This method neglects all indirect paths involving non-listed companies, so that the true ownership network was only approximated. Moreover, 48 countries were analysed separately, ignoring all cross-country links, an approach which inevitably leaves out entirely the global structure of ownership. The aim there was to construct disjoint national stock market networks, from which the backbones were extracted and analyzed. Here, however, we focus on the entire global topology.

Figure S2

Figure S2: General structure of the TNC network. Three types of economic actors appear: 77456 SHs, 43060 TNCs and 479992 PCs. The network contains in total 600508 nodes, and 1006987 links. Links are mainly from the TNCs to the PCs and amongst the PCs themselves.

3. Network Control

Figure S3: Cumulative distribution function of network control and operating revenue. The network control (TM) in the LCC and the operating revenue of the TNCs in the LCC, from which it is computed, is shown.

In this section, we first recapitulate in detail the existing method for computing the value or control in a network. In a second step, we highlight two problems that plague this approach, especially in networks with bow-tie topology (see main text, Sec. Network Topology). The first is that the control assigned to firms that are part of cross-shareholding structures is overestimated. The second is a similar overestimation of the control of the shareholders who are themselves not owned by others. These two problems require independent solutions. In particular, the second problem was never raised before in the literature. We provide a novel algorithm that, for the first time, solves both problems and allows the computation of control also for large networks. This method represents a fundamental improvement to previous works, including our own one [5], as explained below in details. Finally, we illustrate the problem and the corrections introduced by the algorithm using a representative example of a small bow-tie network.

3.1. The Existing Methodology

While ownership is an objective quantity given by the percentage of shares owned in a company, control, reflected in voting rights, can only be estimated using a model. There are two steps involved in the derivation of the notion of control we use in this work. Firstly, direct control is estimated from the direct ownership relations. Network control is then computed on the basis of direct control considering all paths in the network. For the computation of the direct control, we use three models: the linear model, applying the one-share-one-vote rule [2, 3], the threshold model [4] and the relative control model [5]. In the main part of the text, we denote these three models as LM, TM and RM, respectively.

According to the LM, there is no deviation between ownership and control, thus the direct control matrix coincides with the ownership matrix, Lij = Wij . In the TM, full control over a company is assigned to the actor holding a number of shares higher than a predefined threshold (50% in our case), while the other holders are assigned zero control. The control matrix for the threshold model is denoted as Tij . Finally, the RM assigns control based on the relative fraction of ownership shares that each shareholder has (using a Herfindhal-like concentration index). The control matrix is defined as . In particular, the RM assigns high control to a shareholder with a small share in absolute terms, if this share is significantly bigger than the shares of all the other shareholders. For each of these three control matrices, network control is computed with the same procedure. In the main text we use the TM as our main measure, and compare all the results with the LM and the RM. It should be stressed that the global findings are insensitive to the chosen model of direct control.

As explained in the main text, the value of the portfolio of firms owned directly by i should be computed taking into account the value of the firms owned by the firms in the portfolio and so on. Thus, the network portfolio value pinet consists of the value gained indirectly plus the value of the direct portfolio: pinet = ∑j Wij vj + ∑j Wij pjnet. The vector v represents the intrinsic value of the firms (e.g., operating revenue, total assets or market capitalization). Here we use operating revenue, because it is readily available for the economic actors under investigation and it is comparable across sectors (this is not true for total assets). In analogy to the definition above, we introduce the network control (value) [5]. This quantity measures the value controlled by a shareholder taking into account the network of firms in which it has direct or indirect shares. In matrix notation,

c<sup>net</sup> = Cc<sup>net</sup> + Cv

where C ∈ {L, T, R} is one of the three direct control matrices. The solution to Eq. (1) is given by

c<sup>net</sup> = (I − C)<sup>−1</sup> Cv =: Čv

For the matrix (I − C) to be non-negative and non-singular, a sufficient condition is that the Frobenius root of C is smaller than one, λ(C) < 1. This is ensured by the following requirement: in each strongly connected component S there exists at least one node j such that ∑i∈S Cij < 1. This means that there exists no subset of k firms (k = 1, . . . , n) that are entirely controlled by the k firms themselves, a condition which is always fulfilled. By taking the series expansion of (I − C)−1 , it can be proven that: C(I − C)−1 = (I − C)−1 C. As a consequence, Č in Eq. (2) coincides with the solution to the equation

Equation 5

This is corresponds to the definition of integrated ownership given in [6]. Hence, as in [5], we can interpret cnet as the value of control an economic actor gains from all its direct and indirect paths in the network.

Notice that Eq. (1) is related to the notion of eigenvector centrality used to investigate power and influence both in social and economic networks [7, 8]. There is also an additional interpretation of network control in terms a physical system in which a quantity is flowing along the links of the network [5]. In this picture, nodes associated with a value vj produce vj units of the quantity at time t = 1. The weight of a link ij, given by the adjacency matrix entry Aij , determines the fraction of vj that flows through it. Then the inflow, i.e. the flow φi entering the node i from each node j at time t is the fraction Aij of the quantity produced by j plus the same fraction of the inflow of j:

Equation 4

In matrix notation, at the steady state, this yields

φ = Aφ + Av Equation 5

which is formally identical to Eq. (1). Thus if v corresponds to an intrinsic economic value of the nodes, then the network control corresponds to the inflow of control over this value. The network portfolio value of a node is determined by the total inflow of value entering the node.

Next to network control, a related quantity is the so-called network value

Equation 6

which is akin to a Hubbell index centrality measure [9]. This measure is well-established in the literature [6]. The solution is vnet = (I − C)−1 v. By noting that

Equation 7

we find

Equation 10

In other words, the network value of an economic actor is given by its intrinsic value plus the value gained from network control. It is an estimate of the overall value a corporation has in an ownership network. Notice that network value and network control of a company can differ considerably. As an example, Wall Mart is in top rank by operating revenue but it has no equity shares in other TNCs and thus its network control is zero. In contrast, a small firm can acquire enormous network control via shares in corporations with large operating revenue.

From Eq. (7), where cnet = Cv = Cvnet, network control can either be understood as the value of control gained from the intrinsic value reachable by all direct and indirect paths or the value of control given by the network value of directly controlled firms.

3.2. The Algorithm: Computing Control While Remedying the Problems

Unfortunately, the equations defining network control and network value suffer from three drawbacks. Firstly, the computation overestimates control when there are cycles in the network (for example in an SCC† ), i.e., when the number of inter-firm cross-shareholdings grows [10]. Secondly, as we have discovered, it also leads to paradoxical situations. Consider for instance an SCC that is reachable from a single root-node r that owns an arbitrarily small share in one of the firms in the SCC. The above definition assigns to such a node the sum of the intrinsic value of all the nodes in the SCC. This is obviously not a correct estimate of the control of the node r. These two issues are best understood in the flow analogy. Indeed, in a dense SCC control flows through the nodes many times. The smaller the incoming links from the IN are the longer it takes until the flow stops, as, in the steady state, everything ultimately flows to and accumulates in the root-nodes. However, since control corresponds to the total inflow over an infinite time this exaggerates the control of the nodes in the SCC and all the control ultimately flows to the root-nodes. Thirdly, for large networks, the computation of the inverse matrix can be intractable. Here, for the first time, we overcome the aforementioned problems and propose a new methodology that consists of applying an algorithm to compute network control by treating different components of the network separately.

We first illustrate the algorithm for the computation of vnet. Then cnet = vnet − v. In order to calculate the network value for any specific node i, we extract the whole subnetwork that is downstream of a node i, including i. For this purpose, a breadth-first-search (BFS) returns the set of all nodes reachable from i, going in the direction of the links. Then, all the links among these nodes are obtained from the control matrix of the entire network, except for the links pointing to i which are removed. This ensures that there are no cycles involving i present in the subnetwork. Let B(i) denote the adjacency matrix of such a subnetwork, including i, extracted from the control matrix C = (L, T, R). Without loss of generality, we can relabel the nodes so that i = 1. Since node 1 has now no incoming links, we can decompose B = B(1) as follows:

Equation 11

where d is the row-vector of all links originating from node 1, and Bsub is associated with the subgraph of the nodes downstream of i. The value of these nodes is given by the column-vector vsub. By replacing the the matrix B in the expression vnet = Čv + v = C(I − C)−1 v + v and taking the first component we obtain:

Equation 12

where now Equation 13

Notice that if node i has zero in-degree, this procedure yields the same result as the previous formula: B(i,∗) = (0, đ) = Ĉ(i,∗). The notation A(i,∗) for a matrix is understood as taking its i-th row. In the next section it is shown that our calculation is in fact equivalent to the correction proposed by [10] to address the problems of the overestimation of network value in the case of ownership due to the presence of cycles.

However, both methods still suffer from the problem of root nodes accumulating all the control. This issue was previously overlooked because the cases analysed did not have a bow-tie structure and because the focus was not on the empirical analysis of control. To solve this issue, we adjust our algorithm to pay special attention to the IN-nodes of an SCC. We partition the bow-tie associated with this SCC into its components: the IN (to which we also add the T&T), the SCC itself, and the OUT. Then, we proceed in multiple steps to compute the network value for all parts in sequence. In this way, the control flows from the OUT, via the SCC to the IN.

Finally, the network control is computed from the network value as cnet = vnetv. In detail, our algorithm works as follows:

1. OUT: Compute the network value v net (i) for all the nodes in the OUT using Eq. (10).

2. OUT → SCC: Identify the subset S1 of nodes in the SCC pointing to nodes in the OUT, the latter subset denoted as O. To account for the control entering the SCC from the OUT, compute the network value of these selected nodes by applying vnet (s) = ∑o Cso vnet (o) + vs to them. This is an adaptation of Eq. 8, where s and o are labels of nodes in S1 and O, respectively. Note that we only needed to consider the direct links for this. This computation is also equivalent to applying Eq. (10), which considers the downstream subnetworks of S1, i.e., the whole OUT.

3. SCC: Employ Eq. (10) to the SCC-nodes restricting the BFS to retrieve only nodes in the SCC itself. Note that for those SCC-nodes that were already considered in step 2, their network value is now taken as the intrinsic value in the computation. This means one first needs to assign vi → vnet (i) + vi.

4. SCC → IN: In this step we solve the problem of the root-nodes acquiring an exaggerated fraction of the network value. For the subset of IN-nodes I directly connected to some SCC-nodes S2, we again apply vnet (i) = ∑sCis vnet (s) + vi , where i and s are labels of nodes in I and S2, respectively. However, note that due to the cycles present in the SCC, this computation is not equivalent to Eq. (10). In this way only the share of network value over the SCC which is not owned by other SCC-nodes is transferred to the IN-nodes.

5. IN: Finally, use Eq. (10) for assigning the network value to the nodes in the IN-subnetwork. In this case the BFS should not consider the SCC-nodes since their value has been already transfer-ed to their first neighbors in the IN. However, it should retrieve the T&T departing from the IN. Again, for the IN-nodes treated in step 4, first assign vi → vnet (i) + vi.

Notice that if any part of the bow-tie structure contains additional smaller SCCs, these should be treated first, by applying steps two to four.

This dissection of the network into its bow-tie components also reduces the computational problems. Although we perform a BFS for each node and compute the inverse of the resulting adjacency matrix of the subnetwork as seen in Eq. (10), the smaller sizes of the subnetworks allow faster computations.

To summarize, using one of the three adjacency matrices estimating direct control, C ∈ {L, T, R}, net we can compute the corresponding network value for a corporation: vinet . By deducting the net . Operating revenue is taken for the value operating revenue, we retrieve the network control: cinet of the TNCs (vi). Fig. S3 shows the distribution of the operating revenue of the TNCs and the resulting network value.

3.3 Proving the BFS Methodology Corrects for Cycles

Here we show that the BFS algorithm presented in the last section yields an equivalent computation proposed in the literature to address the problems of the presence of cycles leading to exaggerated network value.

Figure S3

Figure S3: Cumulative distribution function of network control and operating revenue. The network control (TM) in the LCC and the operating revenue of the TNCs in the LCC, from which it is computed, is shown.

In [6] the notion of network value was introduced based on ownership which corresponds, in the case of control, to

Figure S4

which in [10] was identified as being problematic. The authors hence introduced a new model which overcomes this problem of exaggerated indirect value in presence of cycles by introducing

Figure S5

This means that the original matrix C defined in Eq. (3) is corrected by removing all indirect self-loops of any node i. If the network has no cycles, then Eqs. (3) and (12) yield identical solutions.

We introduce here for the first time a correction operator, that incorporates this modification and makes the associated computations clearer

Figure S6

where diag(A) is the matrix of the diagonal of the matrix A. It can be shown that

Figure S7

The associated corrected network value can be identified as

Figure S8

Figure S8

Figure S4: Simple bow-tie network topology. Example with a high degree of interconnectedness of the firms in the strongly connected component (SCC).

Our proposed methodology also corrects for cycles in an equivalent way. This can be seen as follows. By applying the BFS algorithm to node i, we extract the adjacency matrix B(i) of the subnetwork of nodes downstream. From Eq. (12) it holds by construction that

Figure S9

where B(i) is defined equivalently to Eq. (2). In a more compact notation

Figure S10

This concludes that our BFS method and the results in [10] are identical: vinet = vnet(i).

3.4 An Illustrated Example

Consider the network illustrated in Figure S4. It is an example of a simple bow-tie network topology. The SCC is constructed in a way to highlight the problem of cross-shareholdings. Hence there are many cycles of indirect ownership originating and ending in each firm in the core of the bow-tie.

We assume the underlying value of each firm to be one, i.e., v = (1, 1, 1, 1, 1, 1)t , where t denotes the transposition operation. Moreover, we will employ the TM, hence Cij = Wij. This results in the network value and the integrated value to be

Figure S11

using Eq. (8).

So although the total value present in the network is 6 = ∑i vi , firm 5 has an disproportionately large network control of vnet5 = 54, highlighting the problem of overestimating the control in the presence of cycles.

Employing the corrections proposed in [10], i.e. by computing the correction operator defined in Eq. (13), one finds

Figure S12

From this, the corrected values can be computed from Eq. (15)

Figure S13

The correction reduces the values of the firms in the core of the bow-tie by approximately one order of magnitude. This confirms that ˆvnet and ˆcnet are indeed the right measures to consider in the presence of SCCs in the network.

Unfortunately, this example also highlights the second problem of the methodology. It is clear, that root nodes accumulating all the control. As mentioned, our proposed algorithm remedies this problem while still correcting for the overestimation in cycles. One finds from Eq. (10) that

Figure S14

illustrating the change from v1net = ˆv1net = 6 ≥ vnet (1) = 1.5.

To summarize, employing vnet for the computation of control in networks with bow-tie topology overestimates the level of control in the SCC by construction. Using vnet on the other hand always assigns the root nodes the highest control. Only the measure vnet(.) puts root and SCC-nodes on par with each other and the leaf-nodes, allowing for the first time an accurate analysis of the control of each node in the network.

3.5. Relations To Previous Work

To summarize, the relation the existing work is as follows. The notion of network value‡ was introduced in [6], in addition to the integrated ownership matrix. This matrix was later corrected in [10].

The notion of network control was first defined in [5] without any of the corrections described above. Because the networks analysed there comprised only listed companies and their direct shareholders, it was sufficient to apply the uncorrected methodology due to the absence of long indirect paths, see SI Sec. 3. In contrast, in the present work, the full-fledged methodology with all the corrections is required in order to consistently compute the flow of control. This resulted in the introduction of the correction operator and its application to the network value and network control. As mentioned, this allowed us to identify a second problem with the methodology. Subsequently, we have incorporated these insights into an algorithm that is suitable for large networks, correcting all potential problems with computing control. Finally, we also uncover the relationship between network control and network value.

‡Although the authors only considered the case of ownership and not that of control, their methods are equivalent to the definition of control employing the LM.

4. Degree and Strength Distribution Analysis

The study of the node degree refers to the distribution of the number of in-going and out-going relations. The number of outgoing links of a node corresponds to the number of firms in which a shareholder owns shares. It is a rough measure of the portfolio diversification. The in-degree corresponds to the number of shareholders owning shares in a given firm. It can be thought of as a proxy for control fragmentation. In the TNC network, the out-degree can be approximated by a power law distribution with the exponent -2.15 (see Fig. S5A). The majority of the economic actors points to few others resulting in a low out-degree. At the same time, there are a few nodes with a very high out-degree (the maximum number of companies owned by a single economic actor exceeds 5000 for some financial companies). On the other hand, the in-degree distribution, i.e., the number of shareholders of a company, behaves differently: the frequency of nodes with high in-degree decreases very fast. This is due to the fact that the database cannot provide all the shareholders of a company, especially those that hold only very small shares.

Next to the study of the node degree, we also investigate the strength which is defined as ∑j Wij, that is, the sum of all the weighed participations a company i has in other companies j (see Fig. S5B). It is a measure of the weight connectivity and gives information on how strong the ownership relationships of each node are.

Figure S14

Figure S5: Various distribution functions. (A) Cumulative distribution function of the in- and out-degree of the nodes in the LCC (log-log scale). The power-law exponent for the corresponding probability density function of the out-degree is estimated to be -2.15. (B) Cumulative distribution function of the node strength in the LCC (log-log scale). As a reference, a power-law with an exponent of −1.62 is displayed.

5. Connected Component Analysis

Ownership relations between companies create formal ties among them. In a strongly connected component (SCC, see SI Sec. 7), all firms reach via an ownership pathway all other firms, thus owning each other indirectly to some extent. In contrast, in a weakly CC firms can reach each other only if one ignores the direction of the ownership links. This is still a situation of interest from an economic point of view because the flow of knowledge and information is not restricted by the direction of the link. The number and the size distribution of the CC provide a measure of the fragmentation of the market. We find that the TNC network consists of 23825 CC. A majority of the nodes (77%) belong to the LCC (largest connected component) with 463006 economic actors and 889601 relations. The remaining nodes belong to CCs with sizes at least 2000 times smaller. The second largest CC contains 230 nodes and 90% of the CC have less than 10 nodes (see Fig. S6).

From a geographical point of view, the LCC includes companies from 191 countries. Of these, 15491 are TNCs (about 36% of all TNCs but accounting for 94.2% of the total operating revenue) from 83 different countries. The firms that are PCs are much more numerous (399696) and are located in only 38 countries. Finally, there are 47819 SHs from 190 countries. This means that shareholders from all around the world hold shares in TNCs located in a more restricted number of countries, which, in turn, further concentrates their ownership shares of PCs in an even smaller number of countries, mainly Europe and the US.

In addition, a sector analysis of the LCC shows that the most represented industries are the

Figure S15

Figure S6: Cumulative distribution function of the size of the connected components. The data point representing the LCC is not shown, as it is three orders of magnitude larger than second largest (with 230 nodes) and completely offset. As a comparison, a power-law with exponent −3.13 (= α − 1) is shown.

business activities sector, with 130587 companies, followed by the services sector with 99839 companies and the manufacturing sector with 66212 companies. On the other hand, surprisingly, the financial intermediaries sector counts only 46632 companies. However, if we distinguish between in-going and out-going relations, the financial intermediaries hold the largest number of shares (341363). Instead, the manufacturing and services sectors, with respectively 182699 and 170397 companies, have the companies with the most shareholders.

6. Bow-Tie Component Sizes

Does a bow-tie structure and the relative size of its IN, OUT and core result from specific economic mechanisms, or could it be explained by a random network formation process? For correlated networks, as in our case, there is no suitable theoretical prediction [11]. Heuristically, one could address the issue by performing a random reshuffling of links. However, this would violate economic constraints. For instance, exchanging a 10% ownership share in a small company with 10% in a big one requires the modification of the budget of the owner. In addition, the procedure is computationally cumbersome for large data sets.

7. Strongly Connected Component Analysis

Cross-shareholdings, or strongly connected components (SCCs) in graph theory, are sub-network structures where companies own each other directly or indirectly through a chain of links (see Fig. S7). Graphically speaking, this means that they form cycles and are all reachable by every other firm in the SCC.

In economics, this kind of ownership relation has raised the attention of different economic institutions, such as the antitrust regulators (which have to guarantee competition in the markets), as well as that of the companies themselves. They can set up cross-shareholdings for coping with possible takeovers, directly sharing information, monitoring and strategies reducing market competition.

In our sample we observe 2219 direct cross-shareholdings (4438 ownership relations), in which 2303 companies are involved and represent 0.44% of all the ownership relations (see Fig. S7A).

These direct cross-shareholdings are divided among the different network actors as follow:

  • 861 between TNCs;
  • 563 between TNCs and PCs;
  • 717 between PCs;
  • 78 between SHs.

When there is a cross-shareholding involving three companies (see an example in Fig. S7B), many combinations of indirect paths are possible. In our network we observe the following ones:

  • 829 of the type: A → B → C → A;
  • 4.395 of the type: A ↔ B → C → A;
  • 8.963 of the type: A ↔ B ↔ C → A;
  • 3.129 of the type: A ↔ B ↔ C ↔ A.

Figure S16

Figure S7: Examples of existing cross-shareholdings. (A) Mutual cross-shareholding. (B) Possible cross-shareholding with three nodes. (C) Cross-shareholding of higher degree.

Next to these simple examples, we also find many SCCs with bigger sizes. Note that smaller SCCs can be embedded in bigger ones. For instance, in the SCC in Fig. S7C there is also one cross-shareholding between the nodes CI and CG . In total there are 915 unique SCCs, of which almost all (83.7%) are located in the LCC. Focusing only on the LCC, there is one dominant SCC: it is comprised of 1318 companies in 26 countries. We define the bow-tie structure in the LCC by taking this SCC as its core (in the main text, we only refer to this SCC). The next smallest SCC contains 286 companies. This is a group of Taiwanese firms located in the OUT of the bow-tie. The remaining 99.7% of SCCs in the LCC have sizes between two and 21. The biggest SCC outside the LCC contains 19 firms.

8. Network Control Concentration

8.1. Control of Financial Institutions

One meaning of control in the corporate finance literature is the frequency by which a shareholder is able to influence the firm’ strategic decision during the official voting [12]. Differently, in this work, by control we mean how much economic value of companies a shareholder is able to influence. Moreover, we did not limit our focus on the control of a shareholder of a single firm. Instead, we look at the control each shareholder has over its whole portfolio of directly and indirectly owned firms. As a result, the shareholders with a high level of control are those potentially able to impose their decision on many high-value firms. The higher a shareholder’s control is, the higher its power to influence the final decision. In this sense, our notion of control can be related to Weber’s definition of “power”, i.e. the probability of an individual to be able to impose their will despite the opposition of the others [13].

In the literature on corporate control there is a debate on weather financial institutions really exert the control associated with their ownership shares. On the one hand, they are not supposed to seek an active involvement in the companies’ strategies. However, some works argue that institutional investors, including banks and mutual funds, do exert control to some extent [14, 15, 16, 17]. In particular, the outcome of votes can be influenced by means of informal discussions, in which pro-management votes are used as a bargaining chip (e.g., in exchange of business related “favors” or in negotiating the extension of credit)§ . On the contrary, [18] and [19] find that mutual funds, which typically hold large blocks of shares, vote against the management (i.e., in favor of corporate governance proposals) only 33% of the times (in the case of Fidelity Fund). However, they do so in more than 60%, on average, in other 11 cases analysed. These results are suggested to originate mainly from a conflict of interest, where the benefits of providing pension plan management to client corporations outweighs the possible benefits gained from increased shareholder value. However, while some mutual funds are reticent to exercise their power during voting mainly in the US, an activist stance is observed for some smaller funds and when operating outside the US [19]. In any case, in our study US mutual funds represent only a small fraction of all global financial institutions. In general, 49 mutual funds, identified by the NACE code 6714, are among the 737 top power-holders (see main text, Sec. Concentration of Control).

§ For example, a mutual fund owning some percent of a large corporation may try to impose job cuts because of a weak economic situation. This can happen: (i) without voting and (ii) although the fund does not plan to keep these shares for many years. In this case, the influence of the mutual fund has a direct impact on the company and its employees. Furthermore, mutual funds with shares in many corporations may try to pursue similar strategies across their entire portfolio.

8.2. Relation to the Rich Club Phenomenon

The so-called rich club phenomenon [20, 21] refers to the fact that in some complex networks the nodes with the highest degree tend to be connected among each other. Being based solely on node degree, rich club indices are not suitable for ownership networks, in which indirect and weighted paths matter. Moreover, in order to benchmark the resulting value of rich club indices, it is usually necessary to reshuffle the links in the network. This would be a problem in our network because it would lead to economically unviable ownership networks. Notice, however, that the core of the TNC network could be seen as a generalization of the rich club phenomenon with control in the role of degree. Thus, future work should look into this issue more in depth.

8.3. Top Control-Holders Ranking

This is the first time a ranking of economic actors by global control is presented. Notice that many actors belong to the financial sector (NACE codes starting with 65,66,67) and many of the names are well-known global players. The interest of this ranking is not that it exposes unsuspected powerful players. Instead, it shows that many of the top actors belong to the core. This means that they do not carry out their business in isolation but, on the contrary, they are tied together in an extremely entangled web of control. This finding is extremely important since there was no prior economic theory or empirical evidence regarding whether and how top players are connected. Finally, it should be noted that governments and natural persons are only featured further down in the list.

Table S1: Top 50 control-holders. Shareholders are ranked by network control (according to the threshold model, TM). Column indicate country, NACE industrial sector code, actor’s position in the bow-tie sections, cumulative network control. Notice that NACE code starting with 65,66,67 belong to the financial sector.

Figure S17

9. Additional Tables

Table S2: Number of top control-holders (TCHs) located in the SCC and being members of the financial sector (FS). Various intersections thereof. The columns refer to the three models of network control and the TM of network value.

Figure S18

Table S3: Concentration of 80% of network control (LM, TM, RM) and network value (TM). The percentages refer to the network controlvalue held by the TCHs according to their location in the SCC and their possible belonging to the FS, and various intersections thereof.

Figure S19

Table S4: Probability that a randomly chosen economic actor (TNC or SH) belongs to the group of top control-holders with respect to its position in the network structure. The first column refers to all top control-holders (TCHs), the second column to the first 50 TCH.

Figure S20


  1. OECD (2000) The OECD Guidelins for Multinational Enterprises (
  2. Goergen, M, Martynova, M, Renneboog, L (2005) Corporate governance convergence: evidence from takeover regulation reforms in Europe. Oxford Rev. Econ.Pol. 21:243–268.
  3. The Deminor Group (2005) Application of the one share - one vote principle in europe., (, Technical report.
  4. La Porta, R, de Silanes, FL, Shleifer, A (1999) Corporate ownership around the world. J. Finance 54:471–517.
  5. Glattfelder, JB, Battiston, S (2009) Backbone of complex networks of corporations: The flow of control. Phys. Rev. E 80.
  6. Brioschi, F, Buzzacchi, L, Colombo, M (1989) Risk capital financing and the separation of ownership and control in business groups. J. Bank. Financ. 13:747–772.
  7. Bonacich, P (1987) Power and centrality: A family of measures. Amer. J. Sociol. pp 1170–1182.
  8. Ballester, C, Calvo-Armengol, A, Zenou, Y (2006) Who’s who in networks. wanted: the key player. Econometrica pp 1403–1417.
  9. Hubbell, C (1965) An input-output approach to clique identification. Sociometry pp 377–399.
  10. Baldone, S, Brioschi, F, Paleari, S (1998) Ownership Measures Among Firms Connected by Cross-Shareholdings and a Further Analogy with Input-Output Theory. 4th JAFEE International Conference on Investment and Derivatives.
  11. Dorogovtsev, S, Mendes, J, Samukhin, A (2001) Giant strongly connected component of directed networks. Phys. Rev. E 64:25101.
  12. Davis, G (2008) A new finance capitalism? Mutual funds and ownership re-concentration in the United States. Europ. Manage. Rev. 5:11–21.
  13. Weber, M (1922) Wirtschaft und Gesellschaft, Tübingen. Max Weber im Kontext. InfoS- oftWare Karsten Worm, Berlin 2.
  14. Santos, J, Rumble, A (2006) The American keiretsu and universal banks: Investing, voting and sitting on nonfinancials’ corporate boards. J. Finan. Econ. 80:419–454.
  15. Becht, M, Bolton, P, Röell, A, Roosevelt, A (2005) Corporate governance and control. NBER.
  16. Gillan, S, Starks, L (2000) Corporate governance proposals and shareholder activism: The role of institutional investors. J. Finan. Econ. 57:275–305.
  17. Davis, G, Thompson, T (1994) A social movement perspective on corporate control. Admin. Sci. Quart. 39:141–173.
  18. Davis, G, Kim, E (2007) Business ties and proxy voting by mutual funds. J. Finan. Econ. 85:552–570.
  19. Davis, G (2008) A new finance capitalism? Mutual funds and ownership re-concentration in the United States. Euro. Manage. Rev. 5:11–21.
  20. Colizza, V, Flammini, A, Serrano, M, Vespignani, A (2006) Detecting rich-club ordering in complex networks. Nat. Phy. 2:110–115.
  21. Fagiolo, G, Reyes, J, Schiavo, S (2009) World-trade web: Topological properties, dynamics, and evolution. Phys. Rev. E 79:36115.
swiss_study_shows_147_technocratic_super_entities_rule_the_world.txt · Last modified: 2020/02/17 21:01 (external edit)