Forecasting European GDP using self-exciting threshold autoregressive models : a warning

A two-regime self-exciting threshold autoregressive process is estimated for quarterly aggregate GDP of the fifteen countries that compose the European Union, and the forecasts from this nonlinear model are compared, by means of a Monte Carlo simulation, with those from a simple autoregressive model, whose lag length is chosen to minimize Akaike's AIC criterion. The results are very negative for the SETAR model when the Monte Carlo procedure is used to generate multi-step forecasts. When the "naive" procedure of generating forecasts is used, the results are surprisingly better for the SETAR model in long-term predictions. Due to the characteristics of the residuals, a bootstrapping method of forecasting was also used, yielding even poorer results for the nonlinear model. -- nonlinear time series models ; SETAR models ; forecasting