Discovering general and sectorial trends in a large set of time series

Abstract

The objective of this research note is to extend the pairwise procedure studied by Car- lomagno and Espasa (ming) to the case of general and sectorial trends. The extension allows to discover subsets of series that share general and/or sectorial stochastic trends between a (possible large) set of time series. This could be useful to model and forecast all of the series under analysis. Our approach does not need to assume pervasiveness of the trends, nor impose special restrictions on the serial or cross-sectional idiosyncratic correlation of the series. Additionally, the asymptotic theory works both, with finite N and T ! 1, and with [T;N] ! 1. In a Monte Carlo experiment we show that the extended procedure can produce reliable results in finite samples.