This paper investigates the properties of the decomposition of a time series presented in a companion paper (Lacroix, (2008)). The procedure relies upon an extension of Beveridge-Nelson methodology. We focus on its empirical implementation and show the need for additional steps in order to clarify the interpretation of the transitory component. Calendar effects are included in the modelization through a slight extension of the methodology while backward filtering of the cycle provides a smoother picture of its dynamic. In addition, special attention is paid to two drawbacks of any filtering method : revisions of the estimates and desynchronization between the raw series and the seasonal adjusted series. We provide an assessment of these effects through a small simulation experiment. The empirical analysis is devoted to three key indicators, the US GNP, the French IPI and the french contribution to M3 monetary aggregate for the euro zone. A limited comparison with alternative filtering methods shows that the results depend heavily on the method chosen for the decomposition. However, the Beveridge-Nelson decomposition displays nice properties and provides sensible and useful results without excessive expense, thanks to its transparent methodology.
Classification JEL : C14, C22, E32.
Keywords : Beveridge Nelson decomposition, seasonal unit roots, seasonal adjustment, cycle.
Updated on: 06/12/2018 10:59