model for Switzerland. Forecast uncertainty is evaluated in three different dimensions. First, we investigate the effect on forecasting performance of averaging over forecasts from different models. Second, we look at different estimation windows. We find that averaging over estimation windows is at least as effective as averaging over different models and both complement each other. Third, we explore whether using weighting schemes from the machine learning literature improves the average forecast. Compared to equal weights the effect of the weighting scheme on forecast accuracy is small in our application.