Title: Bayes-Gaussian aggregation of a single set of forecasts
Authors: Ville Satopaa - INSEAD (France) [presenting]
Abstract: A supra-Bayesian aggregator is developed that inputs a decision-maker's (DM) prior distribution of a continuous outcome and then updates that belief based on experts' point predictions. Given that the aggregator only inputs the DM's belief and the experts' predictions, it can be applied to an isolated prediction task without any past data. The underlying probability model is parametric and captures the experts' bias, accuracy, and dependence. An objective prior is developed for these behavioral parameters, and the resulting posterior is shown to be proper. The posterior is estimated with an off-the-shelf numerical procedure that scales well in the number of experts and does not require tuning. Its use is illustrated on real-world point predictions of human body mass and different economic indicators.