Construction of 1-quarter-ahead forecasts of Real GDP (log-level): yt = log(GDPt) where GDPt is the code of the series in the dataset.

All models are estimated using a rolling window of 100 observations. The first forecast origin is R = 1985 : Q2, and the last forecast origin is T = 2018 : Q3, for a total of P = 134 forecasts. At each forecast origin, we estimated models and save the forecasts from the models.

Forecast evaluation is carried out comparing forecasts for the level with Real GDP GDPt and computing the root mean squared error, RMSE.

Project done with Riccardo Sturla.