Naive calibration

We develop a model of non-Bayesian decision-making in which an agent obtains a signal about a relevant economic fundamental and subsequently takes an action. To interpret the signal, the agent calibrates a simple prediction rule based on a data set that consists of previous signals and state realizations. Her subsequent action affects the probability with which the current signal and the corresponding state realization will be observed and recorded in the data set that will be used in future decisions. We show that this procedure converges to a steady state and that it results in a seemingly pessimistic behavior that is exacerbated by feedback loops. We apply our model to project selection problems and second-price internet protocol version auctions.