Monday, August 10, 2009


Solow at the Stiglitz conference:

I also doubt that universal rational expectations provide a useful framework for macroeconomics. One understands the appeal. Think of it this way: Herb Simon was surely right about bounded rationality; no one would deny that most economic agents are actually like that, and natural selection does not work fast enough to eliminate them. Why did the notion of "satisficing" never catch on? I think it is because the assumption of complete rationality tells the modeller what to do, whereas bounded rationality only tells the modeller what not to do. That is not helpful. Something similar is true about rational expectations. If there were a nice parametric family of alternative ways to model expectations, it might catch on. Most of us would happily go along with the notion of expectational equilibrium: if specific underlying expectations generate an outcome in which those expectations are systematically and non-trivially violated, that situation can not be an equilibrium. It is what happens then that needs thought. The situations that agents need to anticipate need not even be probabilistic, surely not stationary. The popular device used to be adaptive expectations; that may have been inadequate. Maybe this is a case for the application of psychological research (and sociological research as well, because the formation of expectations is a social process). Maybe experiments can be designed. Heterogeneity across agents and classes of agents is certainly important precisely here. One would like a simple, definite way to proceed, if that is possible. A good example of the sort of thing I mean is the way the Dixit-Stiglitz model made monopolistic competition easy. (The trouble is that we are dealing with an unobservable.)

No comments: