Adaptive Learning in Games and Dynamic Economies

Game theory and dynamic macroeconomic models typically assume that people are 'ultra-rational,' both in the sense that they know much about the environment and their competitors' behavior, and in the sense that they know how to use that information in the best possible way. This project aims to weaken the assumption that people are fully rational, and to explore what happens when people behave 'adaptively.' Recent results from the literature on artificial intelligence will be used to build a class of adaptive learning schemes that are suitable for studying economic models. These learning schemes will be applied in several economic contexts. For example, the 'coordination failure problems' will be analyzed, that is, how much coordination can eventually be expected from a system of agents, each of whom behaves adaptively. Several applications in game theory and macroeconomics will also be studied. In some settings, the established learning theories will be used as devices to select among alternative equilibria. In other settings, the theories will be used to suggest self- confirming patterns of behavior that correspond to alternative notions of equilibria. Finally, relationships among some existing and new notions of stability, complexity and optimality will also be investigated.