Unity and Plurality of the European Cycle

Résumé 0

This paper aims to apply the methodology of uni- and multivariate Structural Unobserved ComponentsTime Series Models developed by Andrew Harvey to the study of European growth trends and cycles.The multivariate dimension enables to search similar or, more strongly, common components amongnational series. Common factors models are an interesting way to study the relations between trendsand cycles of several series.The first section presents the general methodology of the paper. The second section presents uni- andmultivariate structural time series models more accurately. It presents explicitly the multivariate formof the models in the trivariate case. The trivariate dimension is convenient to understand the logic ofthese models. The trivariate case remains simple enough to be written explicitly without usingsystematically matrix presentation and it is more general than the bivariate case, which is tooparticular to illustrate the general features of multivariate models.Sections 3 to 5 present and comment the application of this methodology to the study of Europeangrowth trends and cycles. Data covering all the decades 1960 to 1990 are extracted from the quarterlynational accounts collected in the BSDB database of OECD (Business Sector Data Base). Theapplication uses the software STAMP (Koopman, Harvey, Doornik and Shephard, 2000), which wasspecially built to implement the multivariate structural time series models. Three successive ways toexhibit the European cycle are used: the direct split of the European aggregate GDP, compared to theUS split in a bivariate model; the aggregation of the national cycles of the member countries; thesearch for common components between these national cycles. The results of these ways arecompared. Convergence between the results of these approaches is satisfactory.The European aggregate fluctuations reveal two distinct cyclical components, which can beassimilated to the classical Juglar (or decennial) and Kitchin (or triennial) cycles. The European Juglaror decennial cycle exists clearly but it cannot be reduced to a single common component of thenational cycles, which would be generated by a single series of shocks. The European Juglar cycle hasat least the dimension “three”, i.e. it can be understood as the result of the interference of threeelementary and independent sequences of stochastic shocks. These elementary components correspondto current geographical division of Europe. From this point of view, the euro-zone is not yet anoptimal currency area, as the shocks generating the European cycles are not completely symmetrical.The national cycles are not yet reducible to the common symmetrical component which shows throughthe aggregate cycle of the European GDP. This common and symmetrical component contributesapproximately only for one third to the whole national cycles.The sixth section uses the sequences of shocks (or innovations) extracted from the uni– andmultivariate models to build indicators giving information about the evolution of the symmetricalcharacter of shocks hitting euro-zone countries. The vulnerability of the euro-zone to strong shocksand the asymmetry of these shocks show some decreasing trend during the last ten decades but thistrend is neither regular, nor irreversible.The conclusion sums up the main ideas coming from this set of applications. It confirms the practicalinterest of the specific stochastic models used by this paper and the complexity of the European cycle.The definition of a balanced policy mix should take into account the persistent plurality of the a balanced policy mix should take into account the persistent plurality of theEuropean cycles.