Disciplining Density Forecasts with Survey of Professional Forecasters and Bayesian Quantile Regression

This study proposes an alternative, fully probabilistic approach to combining model-based forecasts with surveys or other judgmental forecasts. This approach allows data to determine how informative survey forecasts are regarding the data-generating process by probabilistically exploiting additional information from surveys. We demonstrate this method by estimating a growth-at-risk model for real GDP growth in the United States. To tilt the density forecasts to survey forecasts, we use an unconventional semiparametric Bayesian quantile regression that jointly estimates the conditional quantiles. We show that additional shrinkage from surveys improves prediction performance, and the information from surveys propagates even to the left tails