The vast majority of theoretical economic models impose very strong hypotheses as to the rationality of agents. In macroeconomic contexts, the standard assumption is that agents are fully aware of the equilibrium distributions of endogenous variables, i.e., rational expectations prevails. Models in which strategic interactions are important typically impose some strengthened form of Nash equilibrium, in which agents know with complete accuracy the decision plans of their rivals. These hypotheses, although central to the economic predictions of the models, can be criticized on the grounds that they give no account of how the agents acquire such information. Recent research posits that agents engage in passive adaptive learning. For example, the least squares learning approach presumes that the adaptive rule takes the form of a least squares regression, and evolutionary adaptation schemes specific that agents adjust their actions based on the relative fitness of competing strategies. This project seeks to develop a middle ground between the opposing extremes of full rationality and passive adaptation. The overriding goal is to establish a general framework for analyzing expectation formation behavior that allows agents to engage in a process of active cognition with respect to the economic environment, but places inherent limitations on their cognitive abilities in terms of restrictions on the calculation technology. This project develops an approach to bounded rationality in which agents are endowed with a technology for calculating expectations, together with preferences over forecast errors. Agents form expectations by balancing the benefits of imp;roved forecasts against the costs of calculation. This project extends previous work by the investigators to incorporate a much richer variety of economic environments and calculation algorithms. The approach is used to study hyperinflation and to explain puzzling experimental finding by Marimon and Sunder that are difficult to reconcile with either rational expectations or conventional passive adaptive learning approaches. The calculation framework is extended further and applied to problems in macroeconomics, finance and industrial organization.
