Links
Abstract
I propose a formal method for decomposing frequency-domain information about latent variables in dynamic models. These models describe the joint probability distribution of observed and latent variables. Information transfer from observed to latent variables is quantified as the reduction in uncertainty between the prior and posterior distributions of a given latent variable. By employing frequency-domain techniques, the total information transfer is disaggregated into frequency-specific contributions and the contributions of individual observed variables. This spectral decomposition provides researchers with a tool to trace the origins of information about shocks and other latent variables in structural macroeconomic models, thereby enhancing transparency in model estimation. I demonstrate the method’s utility through applications to three recent studies.