Learning dynamics in an estimated medium-sized DSGE model

We analyse how sensitive these results are to the hypothesis about the information set of the agents, for instance in the case where the PLM is based on a limited set of observed or model-based endogenous variables. Preliminary results show that the structural parameter estimates remain highly robust even under these stronger deviations from rational expectations. The model fit is more sensitive to the information set in case a training sample is used to initialize the agents’ beliefs.