Gallowglass

Waiting for the Man

Much grumpiness as I wait for my web address to catch up with my Movable Type installation; it's frustrating to have a nice new blog but not to be able really to share it with the world. In the meantime, Stephen Karlson has some interesting thoughts (links bloggered - scroll down to two separate postings) in response to my, and Kieran's posts. He agrees with McArdle to the extent that he believes that economics has more powerful parables than the other social sciences. I'd disagree with him, in a qualified class of a way, but it's not worth getting into an argument over. He also defends game theory as a means of modeling cooperation - and here, I think he overstates the case a bit.

First, I should make clear what we agree upon. Karlson states the main result of the folk theorem precisely, which I didn't do (I didn't want to scare non-game theorists away), but I don't think he disagrees with the point that I was trying to make. There exist vast numbers of alternative equilibria which are just as "good" in game theoretic terms as the one that the modeler has chosen. Some strategy combinations are not equilibria, and so can be eliminated, others may involve implausible assumptions (such as incredible threats) and may therefore be ruled out by more restrictive equilibrium refinements (sub game perfect equilibrium). But multitudes remain, some of which are quite nonsensical, in terms of how "real" agents might behave. So what's to protect us from game theorists who either (a) cook the books by creating models that lead ineluctably to prechosen outcomes*, or (b) who come up with silly and spurious models? As Professor Karlson more or less admits, not much more than the implicit norms governing the field, "the onus is on the researcher to think carefully about the objectives of the agents and how they interact. It is bad form to set up problems in such a way as to force the results." Which may work most of the time in practical terms, but isn't very "scientific." The key point is that there is nothing within game theory itself which distinguishes between equilibria that properly represent how real agents might think and interact, and equilibria that don't. In order to distinguish the one from the other, the game theorist has to rely on his own intuitions as to what is plausible, and what isn't - just like the rest of us common or garden social scientists. And sometimes they get it wrong - I used to go to meetings of the International Society for the New Institutional Economics a lot, and heard some really awful papers being presented - bizarre notions about how society had evolved, without an ounce of empirical research to back them up - but with nice models.

This isn't to say that there's not any value to game theory, not at all. Game theory gives us a powerful way of simplifying and representing certain social relations, that serves as a useful intellectual astringent in itself, and may sometimes lead to non-obvious conclusions. When done properly, it's also damn elegant - some of the work that Avner Greif, say, has done in economic history, or Randy Calvert has done in political science is lovely to behold, if you like that sort of stuff (and I do). And it more or less guarantees consistency between premisses and results. But this comes at a cost; there are certain aspects of social interaction that simply cannot be represented using conventional game theory. And, what's more to the point, a "good" game theoretic model of a social situation relies in the end on the intuition and good judgement of the modeler, which game theory itself cannot substitute for. In this at least, economists aren't that much different from the rest of us social scientists. Karlson has more interesting things to say about transaction cost economics - but I'll save my response to that for later.

* NB - cooking the books in this way is more of a problem with partial equilibrium models or finite games. The folk theorem problem suggests that the problem with infinitely iterated games is quite different - it's a problem of indeterminacy and unboundedness. Not so much that you set up the parameters so that you get the equilibrium outcome you wanted, as that you get to choose from one of many, many different possible equilibria in a potentially rather arbitrary fashion.