Abstract
Abstract The authors replicate and extend the Monte Carlo experiment presented in Doz, Giannone and Reichlin (A Quasi-Maximum Likelihood Approach For Large, Approximate Dynamic Factor Models, Review of Economics and Statistics, 2012) on alternative (time-domain based) methods for extracting dynamic factors from large datasets; they employ open source software and consider a larger number of replications and a wider set of scenarios. Their narrow sense replication exercise fully confirms the results in the original article. As for their extended replication experiment, the authors examine the relative performance of competing estimators under a wider array of cases, including richer dynamics, and find that maximum likelihood (ML) is often the dominant method; moreover, the persistence characteristics of the observable series play a crucial role and correct specification of the underlying dynamics is of paramount importance.
Highlights
A central empirical finding from the use of dynamic factor models (DFMs) in applied macroeconomics is that a few factors can explain a large fraction of the total variance of many macroeconomic series
We find that most of the statements in the original article are supported, but the relative quality of full maximum likelihood (ML) estimation, compared to the alternative procedures proposed in the same paper, could be substantial in some cases, whereas in other settings it may not be so decisive to outweigh the extra computational cost
Our narrow sense replication exercise fully confirms the results in the original article
Summary
A central empirical finding from the use of dynamic factor models (DFMs) in applied macroeconomics is that a few factors can explain a large fraction of the total variance of many macroeconomic series. In Doz et al (2011) a solution was put forward to overcome this problem via a two-step hybrid approach that links the simplicity and speed of principal components to the efficiency of the Kalman smoother; later, Doz et al (2012) refined this technique by considering quasi-maximum likelihood estimation where factor estimates are computed iteratively using the Kalman smoother on the state-space representation via the EM algorithm. Both approaches are robust to cross-sectional misspecification, time-series correlation of the idiosyncratic components, and non-Gaussianity.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
