Combining professional and survey forecasts for macroeconomic data

This thesis investigates the combination of survey forecasts and uses data on US GDP growth to determine whether we can benefit from combining forecasts. Two main findings arise from the analysis. First the results show that the sole combination of survey forecasts outperforms the combination of survey forecasts with more conventional time series models forecasts. Second we find that combining the Survey of Professional Forecasters and the Greenbook survey forecasts yields lower RMSE at all but one horizon from nowcasts to four quarters ahead predictions. In particular we show that the Bayesian model averaging combination is preferred for nowcasts. The simple equal-weighted average combination dominates for two and three quarters ahead predictions. Lastly the predictive least square combination is superior for four quarters ahead forecasts.