Performance assessment of probabilistic forecasts

Jochen Bröcker

MPIPKS, Dresden, Germany

This contribution presents a brief tutorial on probabilistic forecasting. After a few motivating examples, the concepts of reliability and sharpness are discussed. In terms of performance assessment, we will focus on scoring rules, and emphasize why scoring rules should be proper. We show that every proper scoring rule can be separated into terms related to the reliability and the sharpness of the forecast. Eventually, we will discuss non-probabilistic odds as a possible generalization of probability forecasts.