Time series models of GDP: a reappraisal.

We propose a model diagnostic device to compare different linear and non linear parametric time series models of real GDP business cycle.The comparison appears of remarkable economic importance since different models have very different implications in term of long run persistence of negative shocks on the level of aggregate output.On the basis of the proposed diagnostic six popular models of real GDP are compared in a Monte Carlo simulation.We find that SETAR models and three stages Markov-switching models significantlly overperform the other statistical representation of the series.Since the SETAR form of non linearity is far easier to handle for both estimation and testing we argue in their favour