Esmail Amiri

In a comparative review we conduct a detailed analysis of the forecasting performance of linear and nonlinear univariate time series models for GDP growth , the one key variable formacroeconomic analysis, focusing on the US and UK, for them very long time series areavailable. Our main goal is to establish whether simple autoregressive (AR) models canstill be used, or whether they should be substituted for more sophisticated specifications.The model comparison exercise is conducted based on the in-sample and out-of-sample measures.Within an in-sample framework, the models are evaluated on the basis of theirgoodness of fit. Out-of-sample evaluation is performed using three lossfunctions, including the common mean absolute and root mean square forecast error.The evaluation of the forecasting performance of our set of non-linear modelsusing real time data is that the non-linear models are able to capture the underlying processes of GDP rate time series.

nonlinear, GDP, AR, TAR,MSAR,STAR,EXPAR