Granger Centre Seminar - Raffaella Giacomini (UCL)

Title: Forecasting with Judgment (joint with G. Ragusa)

Abstract: 

We consider a formal framework for incorporating judgmental corrections into a base multivariate density forecast using exponential tilting. We define judgment as a belief about moments of the distribution of a subset of the variables considered by the base multivariate forecast. Common examples are expert judgments about the mean, variance, covariance, quantiles and probability distribution of some macroeconomic variables, such as those contained in the Survey of Professional Forecasters and Blue Chip Analysts forecasts. Although exponential tilting has a long tradition in econometrics, the two main contributions of this paper are: 1) to formalize the method in a classical inferential context with out-of-sample forecast evaluation objectives and parameter estimation uncertainty; 2) to investigate when the incorporation of judgment results in accuracy improvements. In particular, we provide a testable condition that can guide a forecaster in deciding which type of judgment to incorporate and when to incorporate it. An implication of our analysis is that judgment does not have to be correct to be useful for forecasting.