The accuracy of simple copula models in market risk estimates

This study analyses the accuracy of copula-based Value-at-Risk (VaR) models using GARCH volatility-adjusted returns to account for heteroscedasticity. For a mixed 21-dimensional portfolio of five classes of financial assets results reveal that the method-of-moments (MoM) approach significantly reduces the copula calibration time without loss of accuracy in VaR estimates, compared to the conventionally used pseudo-log-likelihood approach. This creates an advantage for practical applications, especially for portfolios with higher dimensions. In comparison to traditional benchmark VaR models, overall results show that a GARCH-meta-Student t model with a MoM copula calibration generates the best accuracy for VaR estimates and is also computationally quite reasonable for higher portfolio dimensions.