Modelling Non Gaussianity of Oil Price Returns

Abstract

Non Gaussian densities suitable for multivariate generalizations are fitted to daily oil price returns. The absolute and comparative goodness of fit of the several estimated models, is assessed with descriptive and formal methods. A new statistical density forecast test is introduced for that purpose. Extensive descriptive and statistical analysis of the estimated models show that an asymmetric Student' t, with the EGARCH conditional variance model yields a remarkable good fit. The parameters of this density are also stable over several subsamples, while the remaining model parameters are not.