Estimation and Testing of Dynamic Models with Generalized Hyperbolic Innovations

We analyse the Generalised Hyperbolic distribution adequacy to model kurtosis and asymmetries in multivariate conditionally heteroskedastic dynamic regression models. We standardise this distribution, obtain analytical expressions for the log-likelihood score, and explain how to evaluate the information matrix. We also derive tests for the null hypotheses of multivariate normal and Student t innovations, and decompose them into skewness and kurtosis components, from which we obtain more powerful one-sided versions. Finally, we present an empirical application to five NASDAQ sectorial stock returns that indicates that their conditional distribution is asymmetric and leptokurtic, which can be successfully exploited for risk management purposes.