The Forecasting Performance of Seasonal and Nonlinear Models

Abstract

In this paper, we compare the forecasting performance of seasonal and non linear autoregressive models in terms of point, interval, and density forecasts for the growth rates of the Tunisian industrial production, for the period 1976:1- 2006:2. Our results suggest that the point forecasts generated by the linear models perform better than those provided by the nonlinear models at all horizons. By contrast, the analysis of interval and density forecasts at horizons of one and three quarters provide an evident support for the nonlinear models, this result is in line with the literature. Thus, our findings assess the usefulness of nonlinear models to investigate the dynamic behavior of economic systems and to produce accurate forecasts.