Title: DSGE Pileups
Authors: Stephen Morris - Bowdoin College (United States) [presenting]
Abstract: Restrictions from economic theory are usually placed on the parameter space in DSGE models. Such restrictions are also utilized to ensure identifiability. But even when the population parameter value is consistent with theory, the likelihood maximizer need not be. The aim is to document the resulting ``pileup'' phenomenon in the maximum likelihood estimator and to illustrate its consequences for Bayesian inference. The results help to clarify why one frequently observes that the small sample distribution of the MLE and the posterior probability are multimodal, even when regularity conditions including global identifiability are apparently satisfied.