Forecasting Using Functional Coefficients Autoregressive Models

The use of linear parametric models for forecasting economic time series\nis widespread among practitioners, in spite of the fact that there is a large\nevidence of the presence of non-linearities in many of such time series.\nHowever, the empirical results stemming from the use of non-linear models are\nnot always as good as expected. This has been sometimes associated to the\ndifficulty in correctly specifying a non-linear parametric model. I this paper I\ncope with this issue by using a more general non-parametric approach, which\ncan be used both as a preliminary tool for aiding in specifying a suitable\nparametric model and as an autonomous modelling strategy. The results are\npromising, in that the non-parametric approach achieve a good forecasting\nrecord for a considerable number of series.