Nobel Prize 2014: Jean Tirole

A Nobel Prize for applied theory – now this something I can get behind! Jean Tirole’s prize announcement credits him for his work on market power and regulation, and there is no question that he is among the leaders, if not the world leader, in the application of mechanism design theory to industrial organization; indeed, the idea of doing IO in the absence of this theoretical toolbox seems so strange to me that it’s hard to imagine anyone had ever done it! Economics is sometimes defined by a core principle that agents – people or firms – respond to incentives. Incentives are endogenous; how my bank or my payment processor or my lawyer wants to act depends on how other banks or other processors or other prosecutors act. Regulation is therefore a game. Optimal regulation is therefore a problem of mechanism design, and we now have mathematical tools that allow investigation across the entire space of potential regulating mechanisms, even those that our counterfactual. That is an incredibly powerful methodological advance, so powerful that there will be at least one more Nobel (Milgrom and Holmstrom?) based on this literature.

Because Tirole’s toolbox is theoretical, he has written an enormous amount of “high theory” on the implications of the types of models modern IO economists use. I want to focus in this post on a particular problem where Tirole has stood on both sides of the divide: that of the seemingly obscure question of what can be contracted on.

This literature goes back to a very simple question: what is a firm, and why do they exist? And when they exist, why don’t they grow so large that they become one giant firm a la Schumpeter’s belief in Capitalism, Socialism, and Democracy? One answer is that given by Coase and updated by Williamson, among many others: transaction costs. There are some costs of haggling or similar involved in getting things done with suppliers or independent contractors. When these costs are high, we integrate that factor into the firm. When they are low, we avoid the bureaucratic costs needed to manage all those factors.

For a theorist trained in mechanism design, this is a really strange idea. For one, what exactly are these haggling or transaction costs? Without specifying what precisely is meant, it is very tough to write a model incorporating them and exploring the implications of them. But worse, why would we think these costs are higher outside the firm than inside? A series of papers by Sandy Grossman, Oliver Hart and John Moore point out, quite rightly, that firms cannot make their employees do anything. They can tell them to do something, but the employees will respond to incentives like anyone else. Given that, why would we think the problem of incentivizing employees within an organization is any easier or harder than incentivizing them outside the organization? The solution they propose is the famous Property Rights Theory of the firm (which could fairly be considered the most important paper ever published in the illustrious JPE). This theory says that firms are defined by the assets they control. If we can contract on every future state of the world, then this control shouldn’t matter, but when unforeseen contingencies arise, the firm still has “residual control” of its capital. Therefore, efficiency depends on the allocation of scarce residual control rights, and hence the allocation of these rights inside or outside of a firm are important. Now that is a theory of the firm – one well-specified and based on incentives – that I can understand. (An interesting sidenote: when people think economists don’t really understand the economy because, hey, they’re not rich, we can at least point to Sandy Grossman. Sandy, a very good theorist, left academia to start his own firm, and as far as I know, he is now a billionaire!)

Now you may notice one problem with Grossman, Hart and Moore’s papers. As there was an assumption of nebulous transaction costs in Coase and his followers, there is a nebulous assumption of “incomplete contracts” in GHM. This seems reasonable at first glance: there is no way we could possibly write a contract that covers every possible contingency or future state of the world. I have to imagine everyone that has ever rented an apartment or leased a car or ran a small business has first-hand experience with the nature of residual control rights when some contingency arises. Here is where Tirole comes in. Throughout the 80s and 90s, Tirole wrote many papers using incomplete contracts: his 1994 paper with Aghion on contracts for R&D is right within this literature. In complete contracting, the courts can verify and enforce any contract that relies on observable information, though adverse selection (hidden information by agents) or moral hazard (unverifiable action by agents) may still exist. Incomplete contracting further restricts the set of contracts to a generally simple set of possibilities. In the late 1990s, however, Tirole, along with his fellow Nobel winner Eric Maskin, realized in an absolute blockbuster of a paper that there is a serious problem with these incomplete contracts as usually modeled.

Here is why: even if we can’t ex-ante describe all the future states of the world, we may still ex-post be able to elicit information about the payoffs we each get. As Tirole has noted, firms do not care about indescribable contingencies per se; they only care about how those contingencies affect their payoffs. That means that, at an absolute minimum, the optimal “incomplete contract” better be at least as good as the optimal contract which conditions on elicited payoffs. These payoffs may be stochastic realizations of all of our actions, of course, and hence this insight might not actually mean we can first-best efficiency when the future is really hard to describe. Maskin and Tirole’s 1999 paper shows, incredibly, that indescribability of states is irrelevant, and that even if we can’t write down a contract on states of the world, we can contract on payoff realizations in a way that is just as good as if we could actually write the complete contract.

How could this be? Imagine (here via a simple example of Maskin’s) two firms contracting for R&D. Firm 1 exerts effort e1 and produces a good with value v(e1). Firm 2 invests in some process that will lower the production cost of firm 1’s new good, investing e2 to make production cost equal to c(e2). Payoffs, then, are u1(p-c(e2)-e1) and u2(v(e1)-p-e2). If we knew u1 and u2 and could contract upon it, then the usual Nash implementation literature tells us how to generate efficient levels of e1 and e2 (call them e1*, e2*) by writing a contract: if the product doesn’t have the characteristics of v(e1*) and the production process doesn’t have the characteristics of c(e2*), then we fine the person who cheated. If effort generated stochastic values rather than absolute ones, the standard mechanism design literature tells us exactly when we can still get the first best.

Now, what if v and c are state-dependent, and there are huge number of states of the world? That is, efficient e1* and e2* are now functions of the state of the world realized after we write the initial contract. Incomplete contracting assumed that we cannot foresee all the possible v and c, and hence won’t write a contract incorporating all of them. But, aha!, we can still write a contract that says, look, whatever happens tomorrow, we are going to play a game tomorrow where I say what my v is and you say what your c is. It turns out that there exists such a game which generates truthful revelation of v and c (Maskin and Tirole do this using an idea similar to that of the subgame implementation literature, but the exact features are not terribly important for our purposes). Since the only part of the indescribable state I care about is the part that affects my payoffs, we are essentially done: no matter how many v and c’s there could be in the future, as long as I can write a contract specifying how we each get other to truthfully say what those parameters are, this indescribability doesn’t matter.

Whoa. That is a very, very, very clever insight. Frankly, it is convincing enough that the only role left for property rights theories of the firm are some kind of behavioral theory which restricts even contracts of the Maskin-Tirole sense – and since these contracts are quite a bit simpler in some way than the hundreds of pages of legalese which we see in a lot of real-world contracts on important issues, it’s not clear that bounded rationality or similar theories will get us far.

Where to go from here? Firms, and organizations more generally, exist. I am sure the reason has to do with incentives. But exactly why – well, we still have a lot of work to do in explaining why. And Tirole has played a major role in explaining why.

Tirole’s Walras-Bowley lecture, published in Econometrica in 1999, is a fairly accessible introduction to his current view of incomplete contracts. He has many other fantastic papers, across a wide variety of topics. I particularly like his behavioral theory written mainly with Roland Benabou; see, for instance, their 2003 ReStud on when monetary rewards are bad for incentives.