Rationalizing Dynamic Choices

Abstract

An analyst observes an agent take a sequence of actions. The analyst does not have access to the agent’s information and ponders whether the sequence of actions observed could be justified through a rational, Bayesian model. Could some gradual release of information have led the agent to optimally take that sequence of actions? We show that a sequence of actions cannot be rationalized if and only if it can be proved to be dominated via a deviation argument. This argument prescribes a way of deviating that would leave the agent better off in any possible scenario, regardless of the information she might have. The criterion of rationalizing one action sequence can be quite permissive, so we consider a data set that records the action sequences chosen by a population of agents. The deviation argument is then further developed to characterize distributions of action sequences that can be rationalized. Finally, three applications are presented: a test of rationality that can reject the Bayesian model, partial identification of preferences without any assumption on information, and a monotonicity result showing the set of action sequences that can be rationalized increases with risk aversion.