Bayesian Aggregation of Experts' Forecasts

Decision makers often consult multiple experts when trying to forecast a future event or quantity. Once the experts' forecasts are obtained, aggregating them into a single forecast can present a significant challenge. The simplest way to aggregate the experts' forecasts is to average them. The Bayesian decision maker, instead, uses the experts' forecasts as data to update prior beliefs about the event or quantity of interest. We examine several prototypical Bayesian aggregation models. We highlight the two main assessment challenges a decision maker faces in constructing such models: (i) judging each expert's discrimination or dependency between the event of interest and the expert's forecast and (ii) judging the dependency between the experts' forecasts, or overlap in their information sources.

Keywords: expert combination; probability reconciliation; decision analysis; resolution; pooling; expert forecasts; expert opinion; expert discrimination, expert dependency; overlapping information