Non-parametrick Synthesis of Private Probablistic Predictions

This paper describes a new non-parameteric method to synthesize probabilistic predictions form different experts. In contrast to the popular linear pooling method that combines forecasts with the weights that reflect the average performance of individual experts over the entire forecast space, our method exploits the information that is local to each prediction case. A simulation study, replicated from [1], shows that our synthesized forecast is calibrated and whose Brier score is close to the theoretically optimal Brier score. Our robust non-parametric algorithm delivers an excellent performance comparable to the best combination method with parametric recalibration - Ranjan-Gneiting's beta-transformed linear pooling.