Testing and Valuing Dynamic Correlations for Asset Allocation

A new approach to evaluate different conditional variance information specifications is presented by Engle and Colacito (2006) using an economic loss function. They choose portfolio weights according to the popular minimum variance criterion (see Markowitz (1952)). They test for relative performance of different covariance specifications based on the work of Diebold and Mariano. To extend this research, we look at an alternative weighting criterion which minimizes expected shortfall instead of portfolio variance. We repeat the same tests with the alternative weights and find that for both methods the Diagonal BEKK conditional covariance specification performs best