Non linear dynamics in US macroeconomic time series

Abstract

This paper investigates whether the inherent non stationarity of the US macroeconomic time series may be entirely explained by simple stochastic non linear models (like GARCH). Applying the numerical tools of the analysis of dynamical systems to long time series for the US, we reject the hypothesis that the uncorrelated and homoscedastic residuals of the estimated GARCH models contain no structure. Contrary to the theories that attribute the source of the irregular behaviour of the economic system to erratic factors, we are not able, using GARCH models, to obtain truly random residuals. Given this evidence we put forward the possibility that seemingly but not truly random residuals could be, in principle, better controlled and forecasted in the short run.