Forecasting with the help of survey information

In this paper we propose a parsimonious way of combining survey expectations with empirical models to produce forecasts. We do so by augmenting a traditional vector autoregression model with forecasts for different variables and horizons from the ECB Survey of Professional Forecasters. The additional information improves estimation efficiency while maintaining a treatable model. In terms of forecasting performance, the gains from adding survey forecasts are greater at the one and two year ahead horizons, while they are limited at shorter horizons (below one year). Larger gains are found in terms of density performance than in terms of point. Forecasts of real GDP growth benefit the most from survey information, whereas inflation forecasts are improved the least. This latter result is partially driven by the very poor performance of SPF during the 2022 high inflation period. Forecasts for unemployment also benefit from including expectations for GDP and inflation, although not during the COVID pandemic period.