Are Professional Forecasters Bayesian?

Abstract

I evaluate whether expectations of professional forecasters are consistent with the property of Bayesian learning that the expected uncertainty of a fixed target forecast should decline with the horizon. I obtain a measure of individual uncertainty from the density forecasts of the Survey of Professional Forecasters (SPF) and the ECB-SPF and use it to test the prediction of the learning model. Empirically, I find that the prediction is often violated, in particular when forecasters experience unexpected news in the most recent data release, and following quarters in which they produce narrow forecasts. In addition, I find significant heterogeneity in the updating behavior of forecasters in response to changes in these variables.