A1152
Title: Generalized linear pools for combining probabilistic forecasts
Authors: Xiaochun Meng - University of Sussex (United Kingdom) [presenting]
James Taylor - University of Oxford (United Kingdom)
James Curtis - Solea Energy (United States)
Abstract: For many applications, combining the individual probabilistic forecasts can improve their accuracy. The existing literature has extensively explored linear pools of forecasts of cumulative distribution functions or quantile functions. A general framework of combining methods is proposed, which encompasses the existing linear pools. We analyse the statistical properties of the proposed generalized linear pools. The framework and theoretical findings enable the provision of recommendations regarding the choice of combining methods and scores to use in practice. An empirical illustration is provided on simulated and real data.
