Evaluating Probability Forecasts in Terms of Refinement and Strictly Proper Scoring Rules

Abstract

This note gives an easily verified necessary and sufficient condition for one probability forecaster to empirically outperform another one in terms of all strictly proper scoring rules.