Adjustment Dynamics During a Strategic Estimation Task

We use a controlled laboratory experiment to test how decision makers form expectations in a simple strategic estimation task. We consider a probability estimation task in which individual payoffs depend on both a subject's individual forecast and the average forecast of the group. We provide subjects with all the information needed to compute the rational expectations forecast, and we test whether subjects converge to the rational expectations equilibrium. The RE model predicts that when the exogenous state changes, all subjects should adjust immediately to the new equilibrium and that there should be no further adjustment after this initial response. In contrast, in the experimental data, average forecasts are very noisy, frequently biased, and often respond incompletely and with a lag to changes in the exogenous state. Conditional on adjustment, there is a large amount of dispersion in the forecasts recorded by our subjects. We confirm prior findings that strategic considerations result in larger discrepancies between observed collective behavior and equilibrium predictions. Moreover, we find a large degree of heterogeneity in terms of both sensitivity of agents' forecasts to the exogenous state and their strategic sophistication, namely the extent to which they consider the behavior and beliefs of other subjects in their group. We connect our findings to recent work in inattention and level-k models of behavior.