Title: A Bayesian approach for inference on probabilistic surveys
Authors: Marco Del Negro - Federal Reserve Bank of New York (United States) [presenting]
Abstract: A non-parametric Bayesian approach is proposed to the estimation of forecast densities in probabilistic surveys. We use it to study the evolution of the subjective forecast distribution for the U.S. Survey of Professional Forecasters over the past forty years, focusing especially on second moments. We show that the variance of aggregate forecast distribution fell substantially from the eighties to the nineties (the ``conquest''), and fell again after the Fed announced its long term inflation goal. We also show that disagreement (heterogeneity in the mean forecasts) plays a minor role, but that heterogeneity in uncertainty is very large. The ``conquest'' amounted to convincing high-uncertainty forecasters that inflation is under control. We also find that only a fringe of forecasters place any significant probability of the possibility of a return to the seventies. The likelihood of deflation in the aftermath of the Great Recession was significant (almost ten percent for the average forecaster) but declined to one percent or less for most forecasters thereafter.