Please use this identifier to cite or link to this item: https://hdl.handle.net/10419/284044 
Year of Publication: 
2024
Series/Report no.: 
Staff Report No. 1084
Publisher: 
Federal Reserve Bank of New York, New York, NY
Abstract: 
This paper studies large sample classical and Bayesian inference in a prototypical linear DSGE model and demonstrates that inference on the structural parameters based on a Gaussian likelihood is unaffected by departures from Gaussianity of the structural shocks. This surprising result is due to a cancellation in the asymptotic variance resulting into a generalized information equality for the block corresponding to the structural parameters. The underlying reason for the cancellation is the certainty equivalence property of the linear rational expectation model. The main implication of this result is that classical and Bayesian Gaussian inference achieve a semiparametric efficiency bound and there is no need for a "sandwich-form" correction of the asymptotic variance of the structural parameters. Consequently, MLE-based confidence intervals and Bayesian credible sets of the deep parameters based on a Gaussian likelihood have correct asymptotic coverage even when the structural shocks are non-Gaussian. On the other hand, inference on the reduced-form parameters characterizing the volatility of the shocks is invalid whenever the structural shocks have a non-Gaussian density and the paper proposes a simple Metropolis-within-Gibbs algorithm that achieves correct large sample inference for the volatility parameters.
Subjects: 
DSGE models
generalized information equality
sandwich form covariance
JEL: 
C11
C12
C22
Persistent Identifier of the first edition: 
Document Type: 
Working Paper

Files in This Item:
File
Size
940.23 kB





Items in EconStor are protected by copyright, with all rights reserved, unless otherwise indicated.