Disciplining expectations: adding survey expectations in learning models

Learning estimation of DSGE models is becoming a popular alternative to traditional rational expectation estimation. One of the most cited advantages of the introduction of a learning mechanism in DSGE models is that it generates the observed persistence in economic variables such as inflation and output. The evolution of the expectations obtained by learning is, however, not necessarily compatible with direct measurements of expectations such as the ones reported by surveys. I estimate a DSGE model where the expectations generated by the learning mechanism are restricted to be consistent with the dynamic of the surveys expectations. The results show that the standard learning procedure usually overestimates the persistence of the estimated model. The intuition behind these results is that the parameters and expectations estimates used in the learning procedures are designed to match as closely as possible the dynamic of current and past values of the macroeconomic variables included in the model. Therefore, including the dynamic of the surveys expectations in the objective function implies a restriction in the built-in model expectations and hence, less degree of freedoms to match the persistence observed in the actual data.