Adaptive learning with heterogeneous expectations in an estimated medium-scale New Keynesian model

This paper constructs a medium-scale New Keynesian model with heterogeneous expectations by employing the Euler equation adaptive learning technique. Two agent-types are similar in all ways except in how they form expectations. Agent-type A uses a correctly specified model that is consistent with the minimum state variable (MSV) rational expectations equilibrium solution, while agent-type B uses a misspecified model that omits an important subset of variables in the MSV solution. The model is estimated with Bayesian methods using post-Second World War U.S. data and results show that there is significant expectational heterogeneity in the data and that the heterogeneous expectations model performs better than a homogeneous expectations benchmark model. Furthermore, the model’s dynamics resulting from heterogeneous expectations are analyzed.