Abstract:
The introduction of a small amount of bounded rationality into a model sometimes has little effect, and sometimes has a dramatic impact on predicted behavior. We call a model robust to bounded rationality if small deviations from rationality result only in small changes in the equilibrium set. We also say that a model is structurally stable if the equilibrium set (given fully rational agents) varies continuously with the parameter values of the model. Our notions of a model and of rationality are quite broad, allowing us to cover cases in which bounded rationality refers to imperfect optimization, non-rational expectations, or arbitrary behavior by a subset of agents. We show that a model is robust to bounded rationality if and only if it is structurally stable. Thus, we can characterize which models will be robust to bounded rationality and which ones will not, independently of the exact form that bounded rationality takes. In addition, from our characterization it follows that introducing a small amount of bounded rationality will have large effects on predicted outcomes if and only if parameters are near a critical point where the equilibrium set changes in a discontinuous way