Description

The importance of separating secular from seasonal movements in macroeconomic data cannot be understated. For policy purposes, filtering the data is of paramount importance both to analyse macroeconomic fluctuations and to model and quantify the responses of the economy to policy shocks. Despite its importance, seasonality is usually considered at best a nuisance that must be removed from the data before its use. Removing seasonal components is, however, not a trivial task. This paper presents evidence that popular methods to remove seasonality are not harmless procedures and that important information is lost in the filtering of the series. Modern techniques suggest, moreover, that these methods alter our understanding of the relationship among macroeconomic variables and in response to policy shocks. In particular, econometric results suggest that most variables present unit roots not only in their long-run component but also at semiannual and seasonal frequencies. Consequently, the analysis and simulation of policy models undertaken with seasonally adjusted data should be carefully complemented with the analysis of non-filtered data.