(2200 words, very sorry, but I'm afraid that if I write a new shorter version I'll get distracted before I send it out)
Zooko told me a little story at lunch one day. As I recall, it went something like this: You have two magic boxes. Each of them, when you put fuel into it, consumes the fuel and produces some valuable commodity, and it produces N times as much of the valuable commodity if you put N times as much fuel in. But one of the boxes produces an excludable commodity --- the box's possessor gets to choose who benefits from the commodity --- and the other produces a nonexcludable commodity, where everyone benefits from it. Which one produces more value? I answered that it depends on a lot of things, such as the respective ratios of fuel in to valuable commodity out, and the social arrangements surrounding my possession of these boxes. Zooko agreed that it depended on the efficiency issues, but argued that the excludable box is self-reinforcing in a way that the non-excludable box is not --- its possessor can trade its valuable output for more fuel, which means that it's likely that a "machine that runs of itself" will arise around it --- something that he argued couldn't happen with the other box. I have two rebuttals to this argument that I thought of at that lunch, and a third rebuttal that only occurred to me more recently. The first is that if the nonexcludable box is sufficiently efficient, then you can set up the same kind of trading system around it. If you are a potential customer, and you will receive value X in return for paying Y, where Y is less than X, it is self-interestedly rational for you to do so --- regardless of whether other people will additionally receive some value from the transaction. (This works best if the good is nonrival.) There is potentially a prisoner's-dilemma situation, of course, where every potential customer has an incentive to sit back and hope others will pay for all of the nonexcludable commodity that they need. There are several ways out of this: 1. If the commodity has increasing returns over the range expected to be produced, rather than decreasing, then knowing that other people may contribute as well is an incentive rather than a disincentive to your own contribution. (That is, if it's unlikely that the amount produced will be close to "all you need".) 2. If the commodity is fairly customizable, such that it is more likely to fit your particular needs if you pay for its production than if someone else does. 3. If the returns are sufficiently high that the amount you pay is small in proportion to the value you receive. All of these are frequently the case with free software: there are increasing returns as software is more valuable when it can connect to more things; its customizability is enormous; and the private ROI is often enormous. And of course people aren't purely self-interestedly rational, but usually some part of their behavior is driven by such motives. The second rebuttal is that the dependencies on the social arrangements are quite extreme. If you are part of a clan with no private property within the clan, for example, you won't be able to trade commodities for fuel with other people inside the clan. Instead, the people who are most considered most deserving within the clan will get the benefits of the excludable box, and the fuel will come from the clan's common supply. Or, if you are in an environment where commerce is punished, or where swindlers are common, you may not be able to set up the business relationships needed to get much of a supply of fuel for the excludable box. Or, if I can use force to take the box from you after you put fuel in it and get the resulting commodity, or if I can merely take fuel from you, the excludable box will still have little or no advantage. If dominant assurance contracts work in reality, they represent another set of social arrangements that could comfortably fuel the nonexcludable box. In short, the second rebuttal is that the social arrangements that make the excludable box self-sufficient are fairly contingent --- a system of self-interested rational agents who are nevertheless bound to engage in only consensual exchanges based on well-defined property rights --- and not that widespread in the real world. These two suggest the third rebuttal, which is perhaps the most interesting. If the box is possessed by a firm, the entrepreneur who directs the resources of the firm will take into account the benefit to all the firm's members --- not just those directly to the entrepreneur. This provides the non-excludable box, as long as the good it produces is also non-rival (i.e. it's a public good) with a huge boost in its effective efficiency; perhaps 100, 1000, or 10 000 people's benefits are weighed against the cost of the fuel, rather than just one. This allows the public-good box to become self-sufficient at much lower efficiencies. If we look at the existing world, this might explain why historically most of the basic research done in the US has been done by government-funded grants --- the US government is one of the world's largest firms, particularly if you consider all of the citizens as its employees. Much of the applied research has historically been done by very large companies, although with the increasing power of the patent system in pharmaceuticals and genetics, more and more of it is being done by small startups instead. Similarly, much of the work on large free-software projects has been contributed by employees of very large companies like IBM and Intel. If this principle holds water, it might be a second explanation for the existence of firms, next to Coase's argument about the "cost of using the price mechanism" to organize production, more recently called "transaction costs". The problem here is not that the selling of public goods is attended by unusually large marketing costs, risks, or taxes --- in short, it is not a transaction cost at all, or even a cost; it's that it's often in nobody's self-interest to pay the cost to produce those goods, because they receive only a billionth of the benefits, even when producing those goods would produce more benefits overall than it cost. It's one thing to say that a particular arrangement (in this case, large firms) is more economically efficient under certain conditions (in this case, the need for lots of public goods); it's something else to show that people's individual incentives drive them in the economically-efficient direction. For example, in the original situation I was considering, of many independent selfish agents and an efficient public-good box, it's economically efficient for all of the agents to contribute fuel to the box; but only under some circumstances do they have a selfish incentive to do so. I've established that larger firms have more of an incentive to increase production of public goods. But does the production of public goods inside a firm create an incentive to make firms larger, as do Coase's costs of using the price mechanism? I'm not sure it does. Consider the case of an independent consultant who uses Apache but does not contribute to it. Everyone else will benefit if IBM hires him and siphons off some of the value he produces to fund more Apache development. But will IBM? Will the consultant? Does the consultant benefit if she still produces the same value for her clients as before? The clients pay IBM the amount they previously paid her, and then IBM gives her, say, 50% of it. IBM's contributions to Apache increase slightly, and this benefits the consultant. But, by hypothesis, the benefit to her is smaller than the tax on her paycheck --- or there's no need to invoke large firms, is there? She could have just given IBM the money, or billed fewer hours and fixed some Apache bugs in her newly spare time, without becoming an employee. (I know it's not always realistic to tell your client you want to work fewer hours, but sometimes it is.) So it doesn't appear that, in the absence of increasing returns, customizability, extremely high returns, transaction costs, or some similar factor, there's a selfish incentive to go work for a large firm on this basis. Similarly, does IBM benefit? In the above scenario, they probably do, since they're getting paid a lot more than they pay the consultant. But they probably have to offer the consultant more money than she's been making to come work there, and in that case, they're losing money on her. However, they get to increase their Apache contributions a bit, so that now the thousands of other Apache consultants they already had on staff are producing a bit more value --- and there are many thousands of them. So the value they capture by hiring this new employee, even if she's not directly profitable, can be substantial. This is a little dodgy --- I'm saying that by losing more money on a new employee, they can afford to spend more on R&D. I think it's actually correct, though. Consider the case where IBM already had 1000 consultants and, say, 50 full-time Apache contributors on staff. Assume nobody else contribute to Apache, and that the value of an Apache consultant is directly proportional to the number of current Apache contributors, which assumptions are pessimistic --- I think the real world can only make this more profitable, not less. If all these people are paid 1 FTE, then presumably the value-to-customers of each of the 1000 consultants is something around 2 FTE. Increasing the number of contributors by 1 increases the value-to-customers of each of the 1000 consultants by 2%, to 2.04 FTE, for a total revenue of 2040 FTE. So IBM gets back a 40x return on that expense of one more contributor. (Maybe they give some or 39/40 of that to the consultants.) Suppose IBM hires 100 more consultants and 5 more contributors. Now each consultant produces 10% more value (2200 FTE in total for the old consultants, 220 for the new, total 2420 --- 21% more in total, or 42% more profit for IBM if nobody's salary goes up) but IBM is still taxing each contributor's salary with the same 5% R&D. But suppose IBM was previously paying (and/or spending on office space and sales and so on) all 1000 consultants 2*FTE (since that's what they could get if they were independent) and therefore making a loss of 50 FTE. Now if they hire the 100 more consultants and 5 more contributors, they get 220 more FTE of net revenue but increase their R&D expenses by only 5 FTE, so they're suddenly profitable by 220 - 5 - 50 = 165 FTE. They could split this profit with their consultants (say, 164 / 1100 = 0.149 FTE each, for a new IBM consultant salary of 2.149 FTE), who are now making more than they could have made on the open market beforehand. Notice that these profit numbers are proportional to the square of the number of consultants. However, any one of the consultants can now quit and get 2.2 FTE instead of 2.15 for their skills with the newly more powerful Apache. So IBM has a very strong interest in getting more consultants, an interest as strong as Metcalfe's Law would be if it were real, but once they're there, each consultant has a small selfish incentive to leave. (This still works if you pay the contributors well; in fact, you can afford to pay them more than the consultants as long as there are fewer of them. If the contributors were also getting 2*FTE before, then IBM was previously losing 100 FTE and afterwards is only profitable by 220 - 10 - 100 = 110 FTE.) So I think this effect is not strong enough to produce firms in the absence of transaction costs, but it is strong enough to make them much larger than a pure Coasian transaction-costs versus "costs of organizing additional transactions within the firm" and "loss through the waste of resources" analysis would suggest. So to Coase's three factors, "a firm will tend to be larger:" a. the less the costs of organizing and the slower these costs rise with an increase in the transactions organized. b. the less likely the entrepreneur is to make mistakes and the smaller the increase in mistakes with an increase in the transactions organized. c. the greater the lowering (or the less the rise) in the supply price of factors of production to firms of larger size. (I notice he actually left out "the greater are transaction costs", although he certainly mentioned them in the rest of the paper) we can add a fourth: d. the greater the benefit derived from public goods. Probably this is a well-known economic result, but I hadn't heard of it before. It's certainly widely known that governments tend to produce public goods in order to avoid underproduction due to the free-rider problem; probably the extrapolation to large firms is not as surprising for economists as it seemed to me.