Internal Rationality, Learning and Imperfect Information

We construct, estimate and explore the monetary policy consequences of a New Keynesian (NK) behavioural model with bounded-rationality and heterogeneous agents. We radically depart from most existing models of this genre in our treatment of bounded rationality and learning. Instead of the usual Euler learning approach, we assume that agents are internally rational (IR) given their beliefs of aggregate states and prices. The model is inhabited by fully rational (RE) and IR agents where the latter use simple heuristic rules to forecast aggregate variables exogenous to their micro-environment. We find that IR results in an NK model with more persistence and a smaller policy space for rule parameters that induce stability and determinacy. In the most general form of the model, agents learn from their forecasting errors by observing and comparing them with those under RE making the composition of the two types endogenous. In a Bayesian estimation with fixed proportions of RE and IR agents and a general heuristic forecasting rule we find that a pure IR model fits the data better than the pure RE case. However, the latter with imperfect rather than the standard perfect information assumption outperforms IR (easily) and RE-IR composites (slightly), but second moment comparisons suggest that the RE-IR composite can match data better. Our findings suggest that Kalman-filtering learning with RE can match bounded-rationality in matching persistence seen in the data.