Im Rahmen des Forschungsseminars laden wir recht herzlich ein zum Vortrag:

M.A. Christian Feldbacher (Univ. Düsseldorf) - Probability Aggregation

In this paper we discuss the problem of how to optimize probability aggregation by help of different scoring rules or loss measures. The theory of probability aggregation traces back to Leonard J. Savage's investigation of statistical opinion pooling rules or, as he called it, his ``model of group decision'' (cf. Savage 1954, chpt.10.2) which triggered a vast amount of literature (collected in Genest and Zidek 1986) and which was also expanded to the Bayesian framework (cf., e.g., Mongin 1995). However, in Savage’s account the question of which weights to use remained unsolved. In order to answer this question, in applied statistics the idea was proposed to switch to a dynamic perspective and weight an agent's probabilities according to her past predictive performance. Several suggestions were made for spelling out a forecasters' performance. Highly influential was Glenn Brier's proposal to compare a forecaster's predictions with the true outcome by help of a quadratic loss function, the so-called ``Brier score'' (cf. Brier 1950). We will show that aggregating probabilities by score/loss-based weighting is optimal. By this we intend to combine basic insights of probabilistic opinion pooling with results of formal learning theory.

Ad speaker: Research fellow at the Duesseldorf Center for Logic and Philosophy of Science (DCLPS); PhD thesis on meta-induction and social opinion pooling; publications in the areas of philosophy of science and epistemology.