Assessing The Economic Value of Dynamic Conditional Correlations in a Multi-Asset Framework

In this study the economic value of dynamic conditional correlations in a multivariate framework is evaluated with an economic loss function. Data on 65 assets in seven different asset classes over the period 2003-2018 is employed to construct mean-variance optimised portfolios based on predetermined expected returns and covariance estimates from different models. In addition to traditional multivariate GARCH models used by Engle and Colacito (2006), a Block Dynamic Equicorrelation and a Dynamic Factor GARCH model are used to benefit from high model parsimonity and a Combination model is used to exploit model complementarities. Using the testing procedure of Barendse and Patton (2018) with 1000 expected return simulations, the Scalar BEKK model outperforms all other models. On average, traditional multivariate GARCH models yield required returns that are respectively 2 and 3.3 times higher compared to required returns from portfolios constructed by the Block Dynamic Equicorrelation and the Dynamic Factor GARCH model.