Title: Combining predictive distributions
Authors: James Taylor - University of Oxford (United Kingdom)
Souhaib Ben Taieb - University of Mons (Belgium)
Xiaochun Meng - University of Sussex (United Kingdom) [presenting]
Abstract: Combining distributional predictions is an important topic in the forecasting literature. Individual distributional predictions are aggregated to reach a consensus distribution that often has better forecasting accuracy. We propose a novel method for estimating the combining weights based on kernel scores. We show that the proposed methods have several appealing properties when compared to the traditional method based on the log score. We use simulation data and the ECB survey of professional forecasters data to support our theoretical results.