We consider the usefulness of the two-regime SETAR model for out-of-sample forecasting, and compare it with a linear AR model. A range of newly-developed forecast evaluation techniques are employed. Our simulation results show that time-series data need to exhibit a substantial degree of non-linearity before the SETAR model is favoured on some of these criteria. We find only weak evidence that a SETAR model of US GNP provides more accurate forecasts than a linear AR model.

Additional Metadata
Keywords SETAR model, linear AR model, out-of-sample forecasting
Persistent URL hdl.handle.net/1765/1567
Series Econometric Institute Research Papers
Citation
Clements, M.P, Franses, Ph.H.B.F, & Smith, J. (1999). On SETAR non- linearity and forecasting (No. EI 9914-/A). Econometric Institute Research Papers. Retrieved from http://hdl.handle.net/1765/1567