Guillermo Carlomagno and Antoni Espasa.
The objective of this paper is to propose a strategy for exploiting short-run common-alities in the sectoral components of macroeconomic variables in order to obtain better models and more accurate forecasts of the components and, hopefully, of the aggregate. Our main contribution concerns cases in which the number of components is large, so that traditional multivariate approaches are not feasible. We show analytically and by Monte Carlo methods that subsets of components in which all the elements share a single common cycle can be discovered by pairwise methods. As the procedure does not rely on any kind of cross-sectional averaging strategy: it does not need to assume pervasive-ness, it can deal with highly correlated diosyncratic components and it does not need to assume that the size of the subsets goes to infinity. Furthermore, the procedure works both with fixed N and T→ ∞, and with [T,N] → ∞. We perform an application tothe US CPI and find good results of our procedure.