This work focuses on the specification of the threshold autoregressive model and forecasting. We consider U.S. Export is the amount of oil exported from January 1991 to December 2004. We present threshold models that are special cases of the procedure for non-linear models on average above TAR (threshold autoregressive). This means that we start with a simple model and we use a more complicated model if the diagnostic tests indicate that the model obtained is not satisfactory. We will use this procedure to compare an approach of ARMA models and approach the nonlinear threshold for the series. Between the two methods, the prediction threshold autoregressive model is better in the mean square error.
Keyword: threshold, autoregressive model, nonlinearity, test, TAR models, forecasting(10.14266/ajoct19-1)