Emergent Centralization, a motivated history of


Hello all, I’m an Econ PhD-turned-founder and I worked on behavioral theories of centralization and division of labor. I’m especially interested in the history of production in market economics. Is there interest around here for a write-up on this sort of thing? I’m using it guide my thinking on my own project.

My guiding question is: how and why did centralized firms and factories emerge out of a decentralized market system? What mental models does this provide for thinking about crypto? Here are some behavioral angles:

  • Cognitive costs. Should we imagine a ‘sphere of indifference’ where it’s too costly/annoying for people to contract? Relevant for DAOs and DPoS, probably for prediction markets and futarchy as well.

  • Culture/Values. Groups can forms around shared values and some centralization emerges for coordination, messaging, reinforcement etc.

  • Power/authority. Once centralization exists incentives can change, analogous to Chris Dixon’s graph.

  • Commitment, teamwork, and agency problems. Effective division of labor starts requiring a number of workers that can’t be found in a family or even community, trust becomes a scarce resource.

Happy to discuss/collaborate.

Edit: Attached a first essay Introducing the problem of authority-based centralization (“Management”) in an economic context. What’s a Boss Final.pdf (385.9 KB)

Alert! Will Fomo3D destroy Ethereum?

“Centralize for efficiency, decentralize for autonomy”

Systems centralize for efficiency reasons. Since they can provide goods and services cheaper, they outcompete decentralized systems. Decentralized systems will stay niche until their efficiency improves to close the cost gap.


counterpoint: decentralized capital allocation seems to be more efficient at allocating capital than centralized capital allocation


You’re definitely right to a degree, bharathrao, but I think this invites the question “why and when is centralization efficient”? I assume most of us think that centralization becomes inefficient at some point-- for instance, command economies.

In economics this question is typically formulated as “what are the boundaries of the firm”


I think a consideration of some political-economy aspects would also be key to this kind of research. This could also allow for more global applicability to the research, given nations like China and India which contain massive portions of the worlds populations and only began to liberalize their markets much more recently than the U.S. and others.

Personally, I’d argue that, there has never been a true global market (yet), since so much of the world for so much of history lived under strong, economically illiberal, regimes. I’d argue that colonialism prevented a true market system from emerging, and though the world has moved in that direction, as a whole we aren’t there yet.

Basically, my question might be “how and why did a decentralized market system originate out of centralized firms (and governments)? Is the market trending towards centralization or decentralization? What role can Ethereum (and other cryptos) play in this changing dynamic?”


That’s markets, not organization.

Markets work best on autonomy and therefore planned markets usually are disasters. Any co-ordinated effort works most efficiently centralized.


I would definitely be interested.


Would absolutely be interested. The Master Switch (Tim Wu) made some interesting arguments around the cycle of centralization/decentralization in information industries, might be useful to check out if you hadn’t already


I would be interested in having a standards document by which the conventions of organizational literature informing the “boundaries of the firm” construct could be harmonized with, or translated back and forth between, the notation in specs like the Yellow Paper. [Edited to note @jonchoi 2017’s CAPM / fixed income modeling, add section 0.1 after looking at notation in some recent posts, retract my cheap talk about cheap talk about “centralization,” and ask please excuse the overwrought distraction, my bad.]

As “extremely not an econ PhD,” personally, I’d find it absurd to be pedantic about playing the hits below, as it were. They are the product of [e:] effort processing [/e:] your characterization of efficient centralization as a boundaries of the firm problem. As a noncontributing member of the peanut gallery, I wonder whether and how building on, or mapping to, items 2 and 3 (universality & modularity) in the whitepaper design principles might (or should, as a matter of economic historiography) instrumentalize a model of industrial development.

  1. [Structures/Setups] (JEL: D21, D80, D86, L15, L2_, L69)
    Arrow 1973
    Milgrom & Roberts 1990
    Fortunato & Hric 2016
    Aghion & Holden 2011
    Jones & Romer 2010
    Schrijvers, Bonneau, Boneh, Roughgarden 2016
    Milgrom 2000
    Black 1976
    Jackson 2010 // [WorldCat]
  2. [Cognitive Costs/Market Frictions] (JEL: D8_, D91, O17)
    Holmstrom & Milgrom 1991
    Greif & Mokyr 2017
    Aoki 2009
    Romer 2015
    Aghion & Bolton 1987
  3. [Culture/Values] (JEL: E14, E22, E26, L14, Z00, Z1_?)
    Lazear 1995
    Jackson 2018
    Pedraza-Farina 2017
    Granovetter 1985
  4. [Power/Authority] (JEL: D02, G32, H32, P26, P48)
    Coase 1937
    Jensen & Meckling 1976
    Demsetz 1983
    Aghion & Tirole 1997
    Grossman & Hart 1986
  5. [Commitment/Teams/Agency] (L2, O31, D23)
    Holmstrom 1982
    Kandel & Lazear 1992
    Klein, Crawford, Alchian 1978


@phillip that’s a good collection of papers, thanks for posting. I also like your Yellow Paper suggestion, as far as I understand it.

While I’m not sure I follow what you meant about playing the hits, I hope the first essay I wrote up (attached) isn’t guilty of what you had in mind. I did attempt a narrow summary of some of the existing research, but hopefully I at least reduced the barriers to entry for people reading by sequestering and consolidating what little math there is, and trying to maintain a conversational tone.

@SamuelTroper I think your point about political economy is a good one and I didn’t address it directly (despite citing Seeing Like a State.) It’s probably reasonable to assume that the model in the paper has the scope conditions that economic exit is possible and that markets are competitive enough to provide alternatives.

Thanks for the suggestion-- Wu’s work on networks and regulation is fantastic and I had not seen it. I will be using it in part 2 of the essay/summary, which has now has two sections in part due to discovering work like Wu’s.

@vbuterin Thank you and I hope it’s worthwhile.


Brand value is one of decisive factors imho (70/20/10 Coke/Pepsi/NoNameCola).

For example, if you have a Lightning Network it is theoretically a network of equal peers, but in reality, every user will want to connect to the hub that has the strongest security/trust brand.

I do not really think efficiency plays much of role. Windows is a hugely inefficient operating system, but people still use it. If you buy a can of Coke, you are potentially 1000 times less efficient as compared to buying instant soda powder.

The attention span of everyone starting from consumers to CEOs of huge companies is tiny and people do not care about efficiency. What people care about is not having to make decisions, thats where the brand value comes into play. People buy things based on a distorted brand reality created by marketing.

A mechanism that encourages decentralization is mitigating security risks.

If I run a Casper validator and it is compromised by an attacker, my deposit is slashed. So it makes more sense for me to participate in a decentralized Casper validator, where many independent servers make a decision by a majority vote.

In the same way, running a decentralized Plasma operator helps mitigate risks - if I am running a Plasma operator on a single server and it is hacked, I am going to lose lots of money, so it is better for me to cooperate with other people and run my Plasma on a cluster of independent nodes.

So arguably, is not efficiency vs autonomy, it is brand value vs security …


@kladkogex Totally agree on market power. The puzzle I wrote up is something closer to “Why do the things with market power have the structure they do?” i.e. if Coke loses brand share and gets replaced, as happened with Kodak and Ford before them, nobody needs to tell us that the replacement is a hierarchical organization, with managers and divisions and salaries, because we’ll just assume it.

To be fair, for some goods, like art, there is substantial market power situated in individuals. But even in these cases there is often scaling, with artists like Chihuly employing managed, salaried employees to create the art.

As for security, it’s a fascinating point-- personally, I got interested in blockchain again after reading Vitalik’s security through coordination costs piece about client side verification establishing trust by committing to better ex post incentives. That post taught me something deep about security, but I didn’t take from it that decentralization is, by itself, a net gain for security. It strikes me as tradeoff since, for instance, security sometimes requires coordination costs to be low which centralization seems to achieve more easily. Seems true of the Putney Debates of 1647 and just as it’s true of the Visa hack. Of course, ex ante coordination on rules around, say, auditable fault tolerance can help, if not solve, the issues for certain threat models.


I read this a lot, especially in crypto spaces, and am not convinced. The reason why is because centralized firms are often deemed ‘efficient’ mostly in narrow terms of how they use scarce natural and human resources for their own objectives, but these objectives are usually not coordinated with what is best for broader society (not Pareto optimal), hence my use of quotes around the word efficiency. In a system/economy-wide sense, the central planners and managers under the Soviet system, for example, would collect information about inputs and outputs that firms were capable of to generate their five-year plans. They could provide goods and services cheaper, sure, but these planners were allocating resources according to what they thought was best for society and consequentially warped social and material behaviors and outcomes, for example by engaging in an arms race with the West, producing uneven development between rural and urban areas, forced industrialization, all which shaped people’s preferences for goods and services. In this wider societal sense, centralization is immensely inefficient as it is a gross misuse of resources (another stark example is Cuba developing monocrops of sugar for trade with the Soviet Block, which accounted for 70% of Cuba’s export revenue in 1989). Decentralized firms, on the other hand, emphasize letting people and other firms decide what is best for themselves on the assumption that they will arrive at the best use of society’s scarce resources and be efficient. However, in the case of decentralization, these firms/systems pay a heavy price for lack of coordination in that there is a very good chance that optimizations for allocations of inputs and outputs for goods and services could be made, affecting people’s preferences and societal development, and which could stunt society, for instance by mispricing healthcare or education. This is compounded by externalities which markets do not fully account for, global greenhouse gases for example. Incidentally, much of the theory underlying the critiques of both centralization and markets (decentralization)) were made in the 1930s in the debate between Ludwig von Mises and Oscar Lange (‘The Socialist Calculation Debates’). I’m not agnostic in these debates, but I think the ‘Big Data revolution’, although much hyped, does shift the debate away from calculation/coordination toward making models of decentralized democratic planning more possible. I think these are more interesting in so far as efficiency is defined as not wasting scarce human and natural resources and iterate for always shifting/evolving optimizations for socially responsible preferences in production and consumption and development of society’s potentials.


I do not think this was ever true. These guys were psychopaths interested in power only and not in efficiently planning anything. They pretended to efficiently plan things because they though projecting an image of efficient planners would help them to stay in power.

The problem with total centralization is that the guys that are at the top of the pyramid are always psychopaths, If the pyramid is high enough, you cant climb it unless you are totally crazy. Because of that, imho an efficient centralized system of a large size is almost impossible - the guys at the top are just too crazy )


You might really want to check out these papers by David Chaum. He conceptualised decentralised computing services and did a lot of research during the 80s about the need for decentralisation

You can find these papers and more on his website chaum dot com - Can’t post more than 2 links


For me, the ideology and systems in place are just as important to consider as the individuals, psychopaths or not. While some apparatchiks no doubt pretended to act ‘in the people’s interest’ many also believed ideologically in what they were doing as evidenced during the period of de-Stalinization which arrived at rejuvenated commitment to central planning over markets however without the fear of Stalinist punishment to enforce it.

Even on a smaller scale, centralized firms operating in markets like auto factories, banks or consulting firms are run by management that makes decisions which baffle many of the people who actually do the work. I have experienced this when, without going into detail, my former management team formed another enterprise that was completely superfluous and actually wasteful. I’ve heard many similar stories, including of consulting firms charging millions to hire out their staff, many of whom they don’t even know if they are working. My experience in decentralized institutions was similarly problematic and had its own coordination problems where autonomous teams often did redundant work or did things that undermined our own productivity. Lean and Agile methodologies seem to be striking some balance between autonomy, self-management and decentralized coordination that look like progress.