Computing robust stylized facts on comovement

Résumé

We propose an alternative method of obtaining stylized facts on comovement, based on the cross-correlation function of the prewhitened time series, which only depends on the purely stochastic components of the series and the cross efects between them. This approach has the property of being robust to the filtering procedure and hence to the cicle definition. The usual approach consists of obtaining the cross-correlation function of filtered variables, which reflect a mixture of both the existing cross-correlation between the variables and the autocorrelation structure of each of them. The autocorrelation structure, in turn, crucially depends on the filtering procedure. The relevance of such an approach is tested by revisiting some of the facts reported by Kydland and Prescott (1990).