Gene Kindberg-Hanlon, World Bank
Project Description/Abstract
This study addresses the use of variance-maximizing approaches to identify dominant structural shocks at both business cycle frequencies and long horizons. We show that these approaches are potentially vulnerable to biases stemming from confounding shocks. We caution how confounding shocks can lead to incorrect inferences when identifying cyclical drivers at business cycle frequencies. For longer-run inquiries, such as identifying technology shocks, we find that confounding shocks impact macro-econometric models expressed in growth rates, even though they are not influential when we estimate vector autoregressions on the levels of the data. Additionally, confounding shocks are typically not prominent in calibrated DSGE models, which accords with the levels data and explains their success in matching US macroeconomic dynamics. Finally, for instances in which the presence and nature of confounding shocks are uncertain, we suggest the use of a kernel-style spectral method, as it provides \emph{better} empirical identification in a range of simulated data-generating processes.
Co-authors
Alistair Dieppe, European Central Bank
Neville Francis, University of North Carolina, Chapel Hill