We thank Marco del Negro and Frank Schorfheide for providing access to their data. We thank four anonymous referees, Fabio Canova, Yanqin Fan, Ulrich Müller, Frank Schorfheide, Jim Stock, and numerous seminar and conference participants for helpful comments. The views expressed in this paper are those of the authors and do not necessarily reflect those of the Federal Reserve Bank of Philadelphia or of the Federal Reserve System.
Frequentist inference in weakly identified dynamic stochastic general equilibrium models
Article first published online: 12 JUL 2013
Copyright © 2013 Pablo Guerron-Quintana, Atsushi Inoue, and Lutz Kilian
Volume 4, Issue 2, pages 197–229, July 2013
How to Cite
Guerron-Quintana, P., Inoue, A. and Kilian, L. (2013), Frequentist inference in weakly identified dynamic stochastic general equilibrium models. Quantitative Economics, 4: 197–229. doi: 10.3982/QE306
- Issue published online: 12 JUL 2013
- Article first published online: 12 JUL 2013
- Submitted September, 2012. Final version accepted October, 2012.
- DSGE models;
- confidence sets;
- Bayes factor;
- likelihood ratio
A common problem in estimating dynamic stochastic general equilibrium models is that the structural parameters of economic interest are only weakly identified. As a result, classical confidence sets and Bayesian credible sets will not coincide even asymptotically, and the mean, mode, or median of the posterior distribution of the structural parameters can no longer be viewed as a consistent estimator. We propose two methods of constructing confidence intervals for structural model parameters that are asymptotically valid from a frequentist point of view regardless of the strength of identification. One involves inverting a likelihood ratio test statistic, whereas the other involves inverting a Bayes factor statistic. A simulation study shows that both methods have more accurate coverage than alternative methods of inference. An empirical study of the degree of wage and price rigidities in the U.S. economy illustrates that the data may contain useful information about structural model parameters even when these parameters are only weakly identified.