We spend billions of dollars trying to understand the origins of the universe, while we still don't understand the conditions for a stable society, a functioning economy, or peace.
(D. Helbing quoted from here)
The publication The Network of Global Corporate Control --- watch a TEDx talk about it --- has gained quite some attention in the news (for instance, sciencenews.org and newscientist.com, which went viral) and the blogosphere (for instance, planetsave.com, physorg.com and johncarlosbaez.wordpress.com). Because some reactions have been particularly hostile, for instance Ms. Yves Smith from Naked Capitalism (see also our responses here and in their comment section), or have inspired the conspiracy theory camp, please let me recapitulate what our paper is and isn't and address some of the voiced concerns, in order to avoid misconceptions. (If you are only interested in our responses to critiques - for instance, computing control from ownership and the notion of control in the financial sector - please see Section 8 below.)
Ownership network, more info and 3d video on youtube.
In a nutshell, it's an empirical and interdisciplinary effort to understand a vast economic dataset using methods from the study of complex networks. The analysis focuses, on the one hand on ownership, and on the other hand on control, and reports novel findings for both. It should be noted that ownership is an objective, measurable quantity and that control has to be estimated from it. As in all fields of science, classification and quantification are the first necessary steps in the process of in-depth understanding of novel phenomena.
You are also cordially invited to join the discussion at PLoS ONE, where our paper is "available for commenting and debate by the readers, making [it] the start of a scientific conversation".
1.) The dataset
It is comprised of ownership relations of 37 million economic agents (natural persons, families, foundations, government agencies, listed and unlisted companies, etc.) from early 2007, located in 194 countries. We identify and focus on about 43k transnational corporations (TNCs) in 116 countries, defining a network of 600k nodes.
It is Bureau van Dijk's commercial Orbis database, used, among others, by reinsurance companies, banks and government agencies.
We started in 2007 with our first cross-country study and then did this current global analysis. Yes, it would be very interesting to analyze a more recent network snapshot. However, it is also interesting to see the network shortly before the financial collapse in 2008. Moreover, other national ownership network studies have shown that clusters of powerful agents were very resilient and unaffected by external forces ([3] and [4]). So perhaps the observed power-structure in this 2007 dataset is still surviving the turmoil today...
2.) The research questions
a.) What is the global architecture of ownership (e.g., TNCs may remain isolated, cluster in separated coalitions, or form a large connected component)?
b.) How is control distributed globally?
c.) Who are the key economic actors?
 |
| Sample of the international financial network, where the nodes represent major financial institutions belonging to the core and the links give the strongest existing relations among them; node colors indicate different geographical areas: EU (red), US (blue), other countries (green); the width and the darkness of the links show their weight; only the most prominent links are shown; the network shows a high connectivity, with many mutual cross-shareholdings as well as longer cycles; this indicates that the financial sector is strongly interdependent, which make the network vulnerable to instability, see [1] or [2]. |
3.) Methods
a.) Our novel methodology extends the known methods for computing control from ownership, and remedies their shortcomings.
b.) Control is proxied by the potential control over the TNCs operating revenue, referred to as network control in the paper. This estimate the percentage of control from the network of direct and indirect ownership relations (using three different models) and multiples this level of control with the operating revenue of the firms a shareholder is connected to (directly and indirectly).
c.) This measure of control can be understood as a quantity flowing through the network, as the methodology is highly attuned to indirect relations between nodes.
d.) Our study is also relevant to the field of complex networks in general, since the methodology can be applied to discover influential nodes in any network where resources are flowing along weighted directed edges.
4.) Novel findings
a.) The network is seen to organize as follows: there are many (~23k) small clusters of connected nodes (called weakly connected components) and then there is one huge connected component, where 77% of all the nodes are located. See the scaling-law distribution of the sizes of the weakly connected components in Figure S6 in the Supporting Information. Interestingly, although 64% of the TNCs are scattered among the many small connected components, they only account for 6% of the operating revenue. In other words, 1/3 of the TNCs are found in the huge connected component and represent nearly all the value. See Figure 2B in the main text.
b.) The topology of this largest weakly connected component has a tiny, but dominant core (in network speak: the largest strongly connected component defining a bow-tie topology) of about 1300 mostly US and GB financial intermediaries. This observation has possible implications for global systemic risk and global market competition (see more in item 7.).
c.) The economics literature would have us think that ownership relations of US and GB financial institutions should not organize as a tightly-knit group owning the majority in each other (i.e., cross-shareholdings and business groups, the paradigm of widely held firms, and the Atlantic or stock market or arm's-length "type" of capitalism, see references in Section 1.2.2 in [1])
d.) The distribution of control (i.e., the potential control over the TNCs value) is unprecedentedly skewed. It takes the shape of a log-normal distribution (see Figure S3 of the Supporting Information) and is roughly an order of magnitude more unequally distributed than wealth (in developed countries). To give some numbers:
i.) 737 top economic agents in the network of 600508 nodes control 80% of the value (operating revenue) of the 43060 TNCs.
ii.) The core, comprised of 1318 nodes, holds 39% of the control.
iii.) There are 147 top agents in the core, controlling 38.4%.
e.) The top actors are in the core and hence are interconnected and do not carry out their business in isolation.
Observe, that findings a.) to c.) are based solely on the objective ownership data, and require no computation of control. Additional information here and here.
5.) Bottom line
a.) Economists are only slowly starting to turn their attention to economic networks. This said, there is an interesting gap in the literature regarding the tools and models used to do so (see Section 1.2.3 in [1]). This is the first analysis of a global dataset on ownership, using a complex networks approach.
b.) It is a starting point for future research. Determining the true implications is up to other scholars to debate on.
c.) Open question: is the core of top actors an emergent property or the result of planned coordination? We tend to think it is an emergent property of the network dynamics.
6.) What the paper isn't
a.) Pushing an economics or socio-political agenda.
b.) Promoting conspiracy theories.
c.) An exact, unambiguous measurement of real-world control.
"Because interpreting and analyzing these kinds of data is difficult, [Davis] says, the analysis serves more as 'an impression of the moon's surface you get with a telescope. It's not a street map.'" Gerald Davis, economist at the University of Michigan in Ann Arbor, quoted from here.
d.) Alleging that the top agents are colluding.
7.) Why study ownership networks?
a.) Previous studies looked at the impact of globalization forces and corporate governance reforms over time on the network topology and found that there was an unexpected resilient structure of powerful agents which was unaffected by these external forces [3,4].
b.) A cross-country analysis revealed that in markets with many widely held corporations (mostly in Anglo-Saxon countries), this local distribution of ownership actually goes hand in hand with a global concentration of ownership (and control), only visible from the bird's-eye view given by the network perspective [5].
c.) Complex ownership patterns, such as cross-shareholdings (e.g., when firms mutually hold shares of each other), are extensively studied motifs in the corporate governance literature and correspond to strongly connected components in networks. Previous studies (on small samples) have shown that cross-shareholdings significantly reduce competition. Accordingly, antitrust institutions all around the world take the existence of complex cross-shareholding structures very seriously. However, they lack the analytical and quantitative tools to deal with large networks.
d.) For some time economic theory has supported the idea that more connected networks are more stable. In contrast, recently, the work of some scholars as well as the view of some authoritative policy makers predict that a higher level of interconnections among financial institutions can lead to higher systemic risk [6].
e.) The network structure matters: the presence of indirect links can significantly amplify the control (or ownership) held by certain economic agents. If one defines the leverage as the ratio between the control we compute, which takes into account the whole network, and the direct control of agents, the resulting probability distribution is scale free (see Figure 4.10 and Section 4.3.3 in [1]). This means that some agents are able to gain a high level of control because of the presence of indirect ownership relations. Usually, this happens by shareholders having control over small firms, which in turn control larger corporations and so forth. Such ownership structures are also called pyramids but were never studied at a global level. Recall the pyramidal group of indirect ownership around Marco Tronchetti Provera, allowing him to control Telecom Italia, one of the world's largest telecom companies, with a disproportional small amount of equity.
f.) General network measures have meaning in ownership networks: the (scaling-law) distribution of the out-degree gives a measure of portfolio diversification (see Figure S5A in the Supporting Information); the distribution of the sizes of the weakly connected components gives a measure of how integrated or fragmented a market is (see the scaling-law distribution in Figure S6 in the Supporting information); the position of an agent in the network can be indicative of its importance.
8.) Problems'n'Answers:
P1: "OMG they discovered financial institutions hold shares! What a scandal!"
A1: It is obvious that large mutual funds own many shares. However, what we are saying is that the potential control they have is surprisingly high. A portfolio manager would arguably go for a diversification strategy and not a controlling one.
P2: "Everybody knows this. It is all so obvious."
A2: There is a big difference between suspecting the existence of a fact and in empirically demonstrating it. We are not aware of an existing study which gives actual empirical evidence of these findings, which some common wisdom indeed anticipated.
"'This is empirical evidence of what’s been understood anecdotally for years,' says information theorist Brandy Aven of the Tepper School of Business at Carnegie Mellon in Pittsburgh." (quote from here).
P3: "You cannot really infer control from ownership."
A3: The separation of ownership and control has been debated in the scholarly literature for decades (see references in Section 6.2.1 in [1]). Although there are many aspects which make the estimation of control from ownership hard to asses (nonvoting shares, dual classes of shares, multiple voting rights, golden shares, voting right ceilings, proxy votes, etc.), researchers have provided, and extensively used, different simple models believed to proxy the control gained from ownership, such as the linear model, assuming the one-share-one-vote principle, or a threshold model, assigning unequivocal control if the percentage of ownership exceeds the threshold. Quite a lot of attention has been devoted to the analysis of ownership and control of individual corporations and of small groups of firms, but never has a large ownership network been investigated. In [5] we added a novel model, that looks at the relative distribution of shares. Even with a small shareholding, if all others hold significantly smaller shares, this model assigns a high level of control.
We computed control using all three fundamentally different models. Neither the topology of the network, nor the distribution of control, nor the identity of the top holders significantly changes in these different scenarios. This unexpected robustness, or invariance, of such global control measures is encouraging, because it means that neither are the models computing arbitrary numbers, nor can the three models have the same bias in computing control the wrong way. So, although for individual agents the computation of direct control can appear arbitrary, the existence of this "aggregated" agreement in the computation is indeed a positive indication that the results are in fact not just spurious or an artifact of the model details. But, as we caution, these numbers are, in any case, an approximation.
P4: "But funds don't exert control."
A4: The question, if funds do or do not exert control, is still being debated in the scholarly literature. It is known that US mutual funds do not always seek to exert control. However, this applies only to them operating in the US. Indeed, the same funds have been shown to exert their power when operating in Europe. We are not aware of a systematic global study about the control financial institutions wield over the companies they have ownership in world-wide. Moreover, control is defined in these studies in a very specific way (for instance the propensity to vote against the management when it comes to issues of corporate governance). In any case, there are 49 mutual funds among the top 737 top holders.
Control can be exerted in ways which are not always visible. We state: "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." Moreover, studies on the network of directors, the boards they serve on, and the interlocks, have revealed these networks to be scale free [7]. This means a few directors are on very many boards. These ties could also be seen as potential channels, through which control can be exerted covertly, when the directors meet in person.
In being cautious, we cannot exclude that the top holders we identify do not globally exert their power in some way. Especially, if we allow for the fact, that control can also be exerted covertly. This is why we talk about potential power in the context of the control numbers we compute. Even if some top holders indeed do assume a passive role, this could be by their own choosing and change anytime.
For references, consult Section 6.2.5 in [1]. and generally Section 6.2 for more concerns that have been addressed.
9.) And finally...
As the world around us is becoming more complex and interconnected, and indeed so at an increasing speed, we believe in the following, in order to try and gain more understanding:
a.) Keep an open mind and question all assumptions, also supposedly established ones.
b.) Let the data speak, not dogmas.
c.) Take the interconnectedness serious: look for and try and understand the network behind complex phenomena.
d.) Use formal tools that can deal with huge amounts of data and complexity in general.
10.) Sources
[1] Ownership Networks and Corporate Control: Mapping Economic Power in a Globalized World, J.B. Glattfelder, 2010
[2] Industrial organization from a geographical and network perspective, S. Vitali, 2010
[3] (Kogut and Walker, 2001)
[4] (Corrado and Zollo, 2006)
[5] (Glattfelder and Battiston, 2009)
[6] (Battiston, Delli Gatti, Gallegati, Greenwald and Stiglitz; 2009)
[7] (Battiston and Catanzaro; 2004)
12 comments:
very well done interesting piece of work!
well done
Wonderful research and article. Thank you! A couple questions: You provide your list of the 'top 50', but is there someplace where can we see the list of all 150 or so that constitute your super-connected core? Second, you've shown control network identified by ownership links, but it's funny, we associate finance entities with lending. What about the lending links? Are firms within the SCC effectively lending to 'related parties'?
Thx:-) We will be providing the list of the top 737 associated with 80% control (of which 147 are in the core with 40%) soon...
We write in the article:
"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]."
So the honest answer answer is, we don't know. In any case, the control we compute should be understood as a proxy. "[...] 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]." (from the Supporting Information).
Your research is fascinating and very important. Could you explain in layman's terms the "Top 50 control-holders" list at the end of your paper? E.g., does No. 1, Barclays PLC, hold more control (in the definition by Weber you've given here) than No. 2, Capital Group Companies Inc., and so on down the list? I'm confused by the "Cumul. network control" percentages at the far right of the table, which get larger as you move down the list, not smaller as I would expect.
Thx:-)
The column showing cumulative control means that the percentages of control of the individual holders appearing in the list are added.
For instance, this means that Barclays has 4.05, but Barclays *plus* Capital have together 6.66. In other words, Capital alone has 2.61 (=6.66-4.05).
That's why, although the control each single holder has does actually decrease down the list, their cumulative control increases.
On the topic encapsulated in the following quote, "c.) Open question: is the core of top actors an emergent property or the result of planned coordination?", I highly recommend to the interested reader the book, "The Art of Cloaking Ownership: The Secret Collaboration and Protection of the German War Industry by the Neutrals: The Case of Sweden," by Gerard Aalders and Cees Wiebes.
The book is exceedingly hard to find (I found one copy available online, as of 17 Nov 2011, for US$189), but available at the U.S. Library of Congress and many university libraries.
This book, and the book "All Honorable Men," by James Stuart Martin (also hard to come by) together suggest that there may be something more to this than spontaneous self-organization, and that there may be a collaboration, to some extent anyway, stretching back at least to the 1930s. The Aalders/Wiebes book is full of many excellent references that may illuminate this question further.
There is no conspiracy theorizing in either of these two books. The first is an academic book in all respects; the second is a firsthand account of a U.S. professor turned Treasury Dept. investigator during and in the aftermath of WW II.
Is there any demographics data available, so that we could visualize how many people, workers and managers, are sustaining, or participating, in this network?
What I'm trying to see is how many people are involved in this network.
By giving it a human face, maybe we could find some interesting conclusions
Oops, forgot to post the link to the full list:
http://ethz.focproject.net/viewer/tnc
I command your willingness to understand how finance works. Unfortunately, it is much harder than physics and than what people typically think!
I was a disappointed that you did NOT mention in your talk that what you have found is just the results of mutual funds owning shares in other banks and that the portfolio managers of these mutual may or may not exert power on these firms. Having worked in Finance for more than 20 years, I suspect the reality is that they do not. In fact, they own these shares because the benchmark against which they are evaluated does! They may from time to time overweight vs that benchmark or they may not but in most cases, they are not activist investors.
Its actually simple to find out. There are only 49 calls to make since they are only 49 funds. Why not start? I am sure by the 10th call, you will already have a trend :-)
It would actually be nice to do it, so that people would stop making shortcuts about Finance. Finance is far from perfect but it is also far from being that simple!!
ckoko@mac.com
"I command your willingness to understand how finance works. Unfortunately, it is much harder than physics and than what people typically think!"
Yes, I believe this is true for most complex systems we see and interact with. However, I also believe that, regarding finance and economics, we could understand a lot more by paying attention to the empirical data and the networks of interaction. As mentioned above, our study is an approximation, a first sketch of a global emergent property. I would truly love to see more detail being filled in and thus probably new insight being gained.
"... you have found is just the results of mutual funds owning shares in other banks and that the portfolio managers of these mutual may or may not exert power on these firms."
It is mentioned above that we started our analysis by asking, what is the global architecture of control? As, to our knowledge, there have been no previous studies addressing this question and we didn't find much guiding help in the theoretical literature, the findings listed under section 4 above are novel. Sorry they disappointed you;)
And what we found is not just what you claim (see 4a to 4e above).
"... but in most cases, they are not activist investors."
This is an observation you gained by experience, but there is conflicting or missing research on the topic. For instance there are studies which claim that financial corporations which are not activist in the US, are activist when operating in Europe. As our study is global, we stick with our claim, that we cannot prove, but also cannot exclude that the top holders we identify do not globally exert their power in some way. See P4 in section 8 above.
"Its actually simple to find out. There are only 49 calls to make since they are only 49 funds. Why not start?"
We actually did talk to a major corporate shareholder. On the one hand they confirmed that they were not activist, but on the other hand said they would be calling up a large firm in their portfolio and requesting management to act in a certain way they wanted.
It probably comes down to semantics and definitions. What we called "control" is a degree of potential influence. This is not what most people understand when they hear the term. In any case, this anecdote highlights our suspicion, that control can also be exerted covertly or informally, without leaving any traces...
I have only recently come across your work which I believe is a very valuable contribution to our understanding.
Two questions immediately arise:
1) How often is the ORBIS database updated? Obviously the 2007 dataset is considerably dated especially as it preceded the huge economic upheaval of 2008-9 which has appeared to accelerated the concentration of ownership and control.
2) What effect do the 60 or more tax havens have on your data given the secrecy of ownership that are a central feature of these tax havens? The sheer scale of the company and wealth holdings and subsidiaries based in the tax havens and their secrecy as revealed in Nicholas Shaxson's ‘Treasure Islands’ must be an important limitation to your research.
Patrick
Dear Patrick,
Thanks for the kind words and interesting questions.
1) Yes, our data is quite old by now. We started our first cross-country analysis in 2007 (http://arxiv.org/abs/0902.0878) before we did the global one. It takes quite some time to get past peer review and have the study go viral;)
In any case it was simply a coincidence that our dataset was a snapshot of things just before the crisis hit. Indeed, it would be great to do a followup study with current data. However, I'm not in this line of research anymore and my collaborators have also moved on:
academiccommons.columbia.edu/download/fedora_content/download/ac:158304/CONTENT/10253.pdf
http://arxiv.org/pdf/1301.2363.pdf
http://physicsoffinance.blogspot.ch/2012/08/debtrank.html
And to answer your question, at the time (and I believe this is still the case) Bureau van Dijk had either an online subscription or you could get a snapshot of the data. They constantly update their database. For instance, ORBIS curently covers over 100 million companies. Our data had about 30 million.
2) Our concern in general was regarding the variation in disclosure rules for ownership data on a global scale. Luckily there appears to be a trend that disclosure thresholds have been lowered:
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1434144
Regarding tax havens in general, our cross-country study did analyze places like the Virgin Islands, Bermudas, Cayman Islands, Singapore, United Arab Emirates, Switzerland, Austria, etc. Of the 48 countries looked at, we did not see any statistical anomalies for the tax havens:
https://www1.ethz.ch/sg/people/formercoll/jglattfelder/online/supp
However, having said this, people have found that the best determinant for concentrated ownership (and bad disclosure practices) in countries is a low level of protection of minority shareholders:
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=103130
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1434144
So one would expect tax havens to have a very high concentration of control.
To summarize, our study is a first approximation of global control structures. The highly skewed distribution of control we found is perhaps even an underestimation, as the example of tax havens would suggest.
Kind regards,
james
Post a Comment