A REEVALUATION OF THE QUASI-BAYES APPROACH TO THE LINEAR COMBINATION OF FORECASTS

The subjective forecasts used in decision analysis should, in principl e, synthesize all available evidence about the subject in analysis. In this manner, when part of the evidence consists of a variety of forec asting models or expert opinions, Decision Theory requires the decisio n maker to formulate a combination of these predictors. This work take s into account the Bayesian methodologies outperformance and quasi-Bay es, as well as the classical model of optimal combination, all applied to the linear combination of petroleum price forecasts, generated by experts from Petrobras-the Brazilian oil company-for several internati onal markets. It presents a theoretical description of the methodologi es followed by a comparative analysis between performances of the best experts' forecasts and combinations. The performances and features of these combinations are also compared.