An economic basis for certain methods of evaluating probabilistic forecasts

✍ Scribed by Judea Pearl

• Publisher: Elsevier Science
• Year: 1978
• Weight: 505 KB
• Volume: 10
• Category: Article
• ISSN: 0020-7373
• DOI: 10.1016/s0020-7373(78)80010-8

✦ Synopsis

This paper deals with the question of selecting an appropriate measure of compatibility (also called Scoring Rule) between probabilistic models and empirical data. It is natural to require that if the model predicts the occurrence of an observed event with probability p < 1, then the compatibility measure should reflect the actual economical damage caused to the user who acts as though the event (which is about to happen) has only a p < 1 chance of occurring.

The paper establishes relations between the compatibility measure and the user's distribution of future pay-offs and shows that each of the commonly used measures (e.g. logarithmic, spherical, quadratic) represents a natural payoff emanating from an economic environment of a specific character. Using these relations the problem of selecting an appropriate measure of compatibility reduces to that of characterizing the economic impact of the forecast at hand.