Source-linked AI summary
The network of global corporate control
Stefania Vitali, James B. Glattfelder, Stefano Battiston
TL;DR
The paper addresses the lack of a global method and evidence for measuring control across international corporate ownership networks. It analyzes ownership pathways among transnational corporations and computes network control, finding a giant bow-tie structure with concentrated control in a small, tightly connected core.
Problem
Global corporate control had not been quantitatively assessed worldwide, although firms can control one another through complex direct and indirect ownership networks.
Method
The paper reconstructs ownership pathways connected to transnational corporations and computes network control from direct and indirect ownership using a cycle-remedying algorithm.
Results
The network forms a bow-tie structure, and control is highly concentrated, with the core containing many powerful actors.
Takeaways & Limitations
The small, tightly connected core constitutes an economic super-entity that raises questions for global financial stability, researchers, and policy makers.
Takeaways & Limitations
The implications for global financial stability remain conditional because the ownership network is treated as a possible proxy for the financial network.
Abstract
from arXiv · showhide
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.
Introduction
The paper develops a global analysis of corporate control from direct and indirect ownership, then applies it to a worldwide transnational-corporation network. It finds concentrated control within a bow-tie topology, especially among a small core.
- Motivation: Global control cannot be assessed from direct ownership alone because firms exert influence through mult country ownership networks.The paper therefore uses complex network analysis to investigate control globally.
- Method: Ownership links represent direct stakes, while indirect ownership propagates through chains of firms and can be extended across generic graphs.For a two-step chain, indirect ownership is the product WijWjl; associated economic value is computed correspondingly.
- Method: Control is estimated from ownership using a threshold rule in which a majority shareholder receives full control and others receive none.Network control aggregates the economic value influenced directly and indirectly through controlled firms.
- Method: The basic control computation can severely overestimate control for firms in cycles and shareholders upstream of strongly connected structures.The paper introduces an algorithm that removes cycles involving the focal node when computing downstream network value.
- Global network: 600508 nodes and 1006987 ownership ties form the resulting transnational-corporation network, extracted from ownership pathways connected to 43060 TNCs.The largest connected component contains the top TNCs by economic value and 94.2% of total TNC operating revenue.
- Concentration of control: 737 top holders accumulate 80% of total network control, with η∗1 = 0.61% compared with η∗2 = 4.35% for operating revenue.The paper reports that network control is more unequally distributed than wealth and that the result is robust across control models.
Discussion
The global control network concentrates control in a small, densely interconnected core, raising questions about financial stability and market competition. The methodology also offers a possible framework for identifying influential nodes in other directed, weighted flow networks.
- Financial stability: If ownership structure proxies financial ties, the intricate global network may also expose institutions to contagion and systemic risk.The paper presents this implication conditionally because financial contracts are usually undisclosed.
- Market competition: Core TNCs with overlapping business domains are connected by ownership relations that could facilitate blocs and hamper market competition.The paper does not establish whether this international core has acted as a bloc.
- Robustness and scope: Results on concentration are robust across three different models used to infer control from ownership.The authors also discuss cross-country legal settings and uncertainty about whether financial institutions invest in equity to exert control.
- Broader significance: The methodology can identify key nodes in real-world networks where scalar quantities such as resources or energy flow along directed weighted links.The paper presents the small, influential core as a new observation in complex networks and relates it to the rich-club phenomenon.
Supporting Information: The Network of Global Corporate Control
The supporting information identifies the paper’s authors and their institutional affiliation at ETH Zurich.
- Stefania Vitali is listed among the paper’s authors.
- James B. Glattfelder is listed among the paper’s authors.
- Stefano Battiston is listed among the paper’s authors and affiliated with ETH Zurich.
1 Acronyms and Abbreviations
The abbreviations list defines terminology used for network structure, control models, data sources, and economic actors.
- Network terminology: BFS denotes breadth-first search, the search algorithm used in the supporting material.
- Control models: LM, TM, and RM denote the linear, threshold, and relative models for estimating control from ownership.
- Economic actors: TNC, SH, and PC denote transnational corporation, shareholder, and participated company.
2 Data and TNC Network Detection
The study constructs a global TNC ownership network from Orbis data by recursively tracing ownership pathways upstream and downstream. The resulting network includes hundreds of thousands of entities and over one million corporate relations.
- Data source: The Orbis 2007 database contains about 37 million economic actors in 194 countries and roughly 13 million directed, weighted ownership links.It also provides industrial classification, geographic position, and operating revenue information.
- TNC identification: TNCs are selected as companies holding at least 10% of shares in companies located in more than one country.Companies whose ultimate owner is another selected company are excluded so each multinational group retains one representative.
- Network construction: A recursive breadth-first search identifies companies directly and indirectly participated by TNCs, then traces their direct and indirect shareholders upstream.
- Network size: The constructed TNC network contains 600508 economic entities and 1006987 corporate relations.Its node classes are TNCs, shareholders, and participated companies.
- Global scope: Unlike a prior national approach based on listed companies and direct shareholders, this study recursively captures indirect paths and focuses on the entire global topology.The earlier method analyzed 48 countries separately and omitted cross-country links.
3 Network Control
The paper develops a corrected, scalable method for computing network control in corporate ownership networks, addressing overestimation caused by cycles and upstream shareholders. It combines direct-control models with downstream-network analysis and validates the correction against prior results.
- Problems with existing control measures: The method identifies two overestimation problems: cross-shareholding firms receive excessive control, and upstream shareholders can be assigned excessive control.The second problem had not previously been raised in the literature.
- Existing methodology: Direct control is estimated using linear, threshold, or relative-control models, after which network control incorporates all direct and indirect ownership paths.The threshold model is used as the main measure, with comparisons to the linear and relative models.
- Existing methodology: The network value of an actor combines intrinsic firm value with value gained through network control, so network value and network control can differ substantially.Operating revenue is used as intrinsic value because it is available and comparable across sectors.
- Corrected algorithm: The BFS algorithm extracts each node’s downstream subnetwork and removes links pointing back to that node, preventing cycles involving the focal node.The resulting computation is equivalent to a correction proposed in earlier literature.
- Corrected algorithm: The correction operator is introduced to clarify the modification, while the full algorithm is designed to compute control in large networks and correct both identified problems.The corrected approach extends earlier work that was sufficient only for networks without long indirect paths.
- Illustrated consequences: In bow-tie networks, the uncorrected measure overestimates control in the strongly connected component, whereas the corrected measure assigns the highest control to root nodes.The correction reduces core-firm values by approximately one order of magnitude and enables more accurate node-level analysis.
4 Degree and Strength Distribution Analysis
The network’s degree and strength distributions reveal highly uneven ownership connectivity: most actors connect to few firms, while a small number of financial companies hold exceptionally diversified portfolios.
- Degree distribution: The out-degree distribution approximately follows a power law with exponent -2.15, indicating that most actors point to few firms.Some financial companies own shares in more than 5000 firms.
- Degree distribution: Out-degree measures portfolio diversification, whereas in-degree approximates the fragmentation of control among a firm’s shareholders.The database’s incomplete coverage of small shareholders makes high in-degree frequencies decline rapidly.
- Strength distribution: Node strength is the sum of weighted ownership participations and measures the weight connectivity of each company.It provides information about how strong each node’s ownership relationships are.
- Strength distribution: The cumulative node-strength distribution is compared with a reference power law having exponent −1.62.The comparison is shown on a log-log scale.
5 Connected Component Analysis
The ownership network is fragmented into many connected components, but a giant component contains most actors and nearly all transnational-corporation revenue, spanning firms and shareholders across many countries.
- Network connectivity: 77% of nodes belong to the largest connected component, which contains 463006 economic actors and 889601 ownership relations.The network contains 23825 connected components in total.
- Component sizes: 90% of connected components have fewer than 10 nodes, while the second largest contains 230 nodes.The component-size distribution is shown against a power law with exponent −3.13.
- Geographical composition: The largest connected component includes companies from 191 countries and 15491 TNCs from 83 countries, representing 94.2% of total TNC operating revenue.It also includes 399696 private companies and 47819 shareholders.
- Sector composition: Business activities, services, and manufacturing are the most represented sectors in the largest component.Financial intermediaries are fewer in number but hold the largest number of shares, with 341363 outgoing relations.
6 Bow-Tie Component Sizes
The paper questions whether the bow-tie’s IN, OUT, and core sizes reflect specific economic mechanisms or could instead arise from random network formation.
- Bow-tie component sizes: The analysis asks whether the bow-tie structure and the relative sizes of its IN, OUT, and core result from economic mechanisms or random network formation.For correlated networks, the paper states that no suitable theoretical prediction is available.
- Bow-tie component sizes: Randomly reshuffling ownership links would violate economic constraints, such as exchanging a 10% stake in a small company with 10% in a large one.The passage presents this as a limitation of a heuristic randomization approach.
7 Strongly Connected Component Analysis
Strongly connected components capture cross-shareholding cycles in which firms are mutually reachable through ownership paths. The network contains many such structures, including one dominant component defining the bow-tie core.
- Strongly connected components: Strongly connected components are sub-networks in which companies own one another directly or indirectly and every firm is reachable from every other.These structures correspond to cycles in the directed ownership graph.
- Economic significance: Cross-shareholdings can protect firms from takeovers while enabling information sharing, monitoring, and strategies that reduce market competition.These ownership structures therefore matter to antitrust regulators and companies.
- Direct cross-shareholdings: 2219 direct cross-shareholdings involve 2303 companies and account for 0.44% of all ownership relations.They include 563 links between TNCs and private companies and 78 links between shareholders.
- Strongly connected components: The network contains 915 unique strongly connected components, with 83.7% located in the largest connected component.Smaller components can be embedded within larger ones.
- Bow-tie core: A dominant strongly connected component of 1318 companies across 26 countries defines the core of the largest component’s bow-tie structure.The next smallest strongly connected component contains 286 companies, while 99.7% of the remaining components range from two to 21 firms.
8 Network Control Concentration
The paper distinguishes portfolio-wide economic control from influence over a single firm and examines whether financial institutions exercise such control. Its ranking finds that many top control-holders are financial actors located in an entangled core rather than operating in isolation.
- Control concept: Global control measures the economic value a shareholder can influence across its directly and indirectly owned portfolio.This differs from measuring influence over one firm during official voting.
- Financial institutions: Financial institutions’ control remains debated because they may avoid active involvement, yet informal discussions can influence corporate voting and strategy.Some studies report mutual funds opposing management in 33% of votes in one case and over 60% on average across 11 others.
- Top control-holders ranking: The first ranking of economic actors by global control shows that many top actors belong to the financial sector and to the network core.The ranking uses network control, with actors identified by country, industrial sector, and bow-tie position.
- Top control-holders ranking: Top actors in the core are tied together in an extremely entangled web of control rather than conducting business in isolation.The paper notes that prior economic theory and empirical evidence had not established whether or how top players were connected.
9 Additional Tables
The additional tables organize top control-holders by financial-sector membership and network position, and report concentration and positional probabilities. They provide comparisons across network-control models and network-value measures.
- Table S2: Table S2 counts top control-holders in the strongly connected component and financial sector, including their intersections across control models.Its columns refer to three network-control models and the threshold model of network value.
- Table S3: Table S3 reports the concentration of 80% of network control and network value by strongly connected-component location and financial-sector membership.The table compares the LM, TM, and RM control models with threshold-model network value.
- Table S4: Table S4 gives the probability that a randomly chosen transnational corporation or shareholder belongs to the top control-holder group based on network position.It distinguishes all top control-holders from the first 50.