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...