Kalman Filter Learning, Bounded Rationality and Informational Consistency

We develop a behavioural New Keynesian model with rational (RE) and boundedly rational (BR) agents. The latter are anticipated utility learners and use simple heuristic rules to forecast aggregate variables exogenous to their micro‐environment. For the RE agents, we study two information assumptions: standard perfect information and imperfect information (II‐involving Kalman‐filter learning). We show that a consistent set of II for the modeller and the private sector generates endogenous persistence of decisions in response to unobserved exogenous shocks and thus RE under II dominates the BR specifications in empirical comparisons. We reveal substantial evidence of composite RE(II)‐BR structure with estimated proportions of RE and BR agents. Our results highlight the importance of informational consistency for the analysis of behavioural heterogeneity.