Lessons from Learning to Have Rational Expectations

This paper reviews a growing literature investigating how economic agents may learn rational expectations. Fully rational learning requires implausible initial information assumptions, therefore some form of bounded rationality has come into focus. Such learning models often converge to rational expectations equilibria within certain bounds. Convergence analysis has been much simplified by methods from adaptive control theory. Learning stability as a correspondence principle show some promise in common macro models. A new selection problem arises since differences in initial information and learning methods give rise to many different equilibria, making economic modelling sensitive to assumptions on information and information processing.