Skepticism about "rational choice theory" outside of economics is based on the conviction that people make mistakes and expect others to behave in a manner that is not perfectly predictable. Economists have started to take these criticisms seriously, and recent theoretical research has identified numerous situations in which even small departures from perfectly-rational decision making can cause a dramatic change in the landscape of expected payoffs. The effects of "bounded rationality" are particularly important in markets or games where the slopes of expected payoff functions are sensitive to parameters that do not affect the underlying structure of the Nash equilibria. This project uses laboratory experiments to evaluate behavior in a series of situations of wide interest to economists: imperfect price competition, coordination, auctions, and interactive models. In each case, the theoretical predictions of both equilibrium and dynamic learning models are derived in a generalized context that spans perfect rationality in one extreme and perfectly random decision making in the other. Anyone who has looked at noisy laboratory data knows that most human behavior lies somewhere in between these extremes, and this intuition can be evaluated by specifying models in a manner that permits econometric estimation of error and learning parameters from laboratory data.
The experiments are carefully motivated by "noisy" equilibrium and learning models, so that the treatment changes have no effect on the Nash equilibria, but they do alter the predictions based on models of bounded rationality. Initial data patterns reported in this proposal conform nicely to both the dynamic trajectories and eventual equilibrium levels predicted. One goal of this research is to provide a data base that will stimulate further theoretical work on strategic behavior that is guided by carefully documented empirical regularities.
