There has been a good deal of criticism concerning this tool. The old industrial organisation was a discipline with a primary interest in empirical results that lacked a sound theoretical basis. The new industrial organisation is a discipline with elaborate and elegant models whose application to real-world problems is, however, often very problematic or even outright impossible. Louis Phlips (1995, 11) an advocate of the use of game theory in competition policy is very frank in admitting it: “ … I know that much work remains do be done on practical questions, such as how a given industry can be identified as being in a Nash equilibrium, how it gets into such an equilibrium, how it gets out of it, and how it moves from one such equilibrium to another one.”
Game theoretic models will only help us in coming up with good predictions if its central assumptions are not too far off the mark. The rationality assumptions regularly used in game theory have been met with scepticism. Werner Güth (1992, 272), e.g., believes that the rationality hypothesis is the central weakness of the use of game theory in industrial organisation. If rational behaviour as assumed by the theory cannot generally be taken for granted then game-theoretic predictions will be incorrect even if the model itself is adequately specified (1992, 272).
The summer of 2000 offers a good real-life test of the applicability of gametheoretical models: In Germany, third generation mobile phone (UMTS) licences were auctioned off in a highly regulated process. Some game theorists analysed the rules and made far-reaching predictions concerning the outcome of the auction and were proven wrong (see, e.g., Moldovanu and Jehiel 2001). One could now reply that the participants in the auction were no experts in game theory and the result therefore diverged from the one expected. Yet, all companies participating in the auction heavily relied on experts – even including Nobel laureates. This shows that the predictions derived from game-theoretic models do not seem to be very reliable.
As soon as games are played repeatedly, a very large number of outcomes become possible (this insight is called folk theorem by game theorists because it was common knowledge long before it was formally proven). For the predictive power of game theory, this is a serious problem: one central criterion for judging the quality of theories is the number of outcomes that it predicts will not occur. If a large number of equilibria are possible in repeated games, this is thus a serious problem for the predictive quality of game theory. Attempts to deal with this problem, such as equilibrium selection theories as advocated by Selten (1975) have been only moderately successful.10
In any theory, the outcomes are driven by the assumptions imputed into a theory. This is, of course, also true for game theory. But with regard to game-theoretic models, the sensitivity of the outcomes to minor modifications in the assumptions seems to be very far-reaching. Formulated the other way around: game-theoretic models are not robust.
Some, but not all, game-theoretic models seem to assume a curious asymmetry concerning the information at the disposal of many actors: actors might have incomplete, imperfect or asymmetric information but it is sometimes assumed that the (scientific) observer is not constrained by such problems. Now, if a player is able to fool those he is interacting with it is hard to see why he should not be able to fool scientific observers watching him.
Game theory thus has lots of advantages as well as disadvantages. One problem with game theory that has not been mentioned so far is the assumption of a firm as “given”. Nobody asks for the rationale of its existence because it simply exists. We now turn to a theory in which firms are not assumed to be exogenously given any more but are endogenous to the competitive process. This is the new institutional economics.
Posts
