Trimmed Whittle estimation of the SVAR vs. filtering low-frequency fluctuations: applications to technology shocks

Abstract

Abstract This paper shows that the trimmed Whittle estimation of the SVAR is superior to filtering (or differencing) undesired, low-frequency fluctuations that may arise in macroeconomic data. Pre-filtering destroys the low-frequency range of the spectrum, thus biasing the estimated parameters and the responses of the variables to shocks. The proposed method, by contrast, accounts for the undesired fluctuations while overcoming these drawbacks. Furthermore, the method remains reliable even when the observed low-frequency variability has been incorrectly considered as external to the SVAR. An empirical application that examines the effect of technology shocks on hours worked is provided to illustrate the results. We find the response of hours positive and similar using both long and short-run identification restrictions, thus providing a solution to a wide debate in the business cycle literature.