Forecasting Using Time Varying Meta-Elliptical Distributions with a Study of Commodity Futures Prices

Abstract

We propose a methodological approach to the forecast and evaluation of multivariate distributions with time varying parameters. For reasons related to feasible inference attention is restricted to meta-elliptical distributions. We use our approach for the study of a large data set of 16 commodity prices. Our approach leads to a theory for model validation avoiding common problems caused by discontinuities, time variation of parameters and nuisance parameters.