Un modello non lineare per la serie storica dell'indice della produzione industriale italiana

Abstract

In this paper we intend to experiment Smooth Threshold Autoregressive model (STAR) in order to capture asymmetric business cycle in the time series of the industrial production index. First we obtain some evidence of non linearity by non parametric lag-regression, then we fit an ESTAR model using the Granger-Terasvirta's procedure. The results are better than the ones available from a linear model. In addition the graph of the transaction function tell us some interesting features on the changing regimes. In the paper we suggest also an informal way to easily obtain a STAR model for various asymmetries in business cycles.