What Do We Lose When We Average Expectations?

In this paper, I use the Bloomberg Survey of forecasts to assess if evaluating the distribution of expectations will lead to important additional insights over the evaluation of the simple average. I first introduce new approaches that allow me to assess the forecast accuracy and the information rigidity at the individual level despite a large share of missing data. Applying these new approaches, I find that taking into account the distribution can significantly improve the predictive power of the survey. For example, I find that the part of uncertainty measured by disagreement can improve the prediction of recessions in a dynamic probit model relative to the simple average. On information rigidity, I find that some of the rigidity found at the aggregate level likely stems from the aggregation process. Together, my findings suggest that we should look at individual expectations whenever possible as important insights are lost by just looking at aggregate expectations.