| Popis: |
We perform a forecast evaluation exercise, where a broad set of linear univari-ate models and autoregressive artificial neural networks are compared against a simple linearbenchmark when predicting Portuguese real GDP growth. The forecasting exercise is performedin a pseudo-real-time framework, meaning that the specification and estimation of each modelare delivered for each quarter of the out-of-sample forecast evaluation interval. The efficacy ofthe models is tested for diverse conceptions of the loss functions, different evaluation samples,and estimation procedures. The empirical results point to the pre-eminence of artificial neuralnetworks comparatively to linear autoregressions. |
