Can multivariate GARCH models really improve value-at-risk forecasts?

Abstract

© 2020 Proceedings - 21st International Congress on Modelling and Simulation, MODSIM 2015. All rights reserved. This paper evaluates the performance of multivariate conditional volatility models in forecasting Value-at-Risk (VaR). The paper considers the Constant Conditional Correlation (CCC) model of Bollerslev (1990), and models that allow dynamic conditional correlation such as the Dynamic Conditional Correlation (DCC) model of Engle (2002) and the Time-Varying Conditional Correlation (TVC) model of Tse and Tsui (2002). While the underlying assumptions vary between these models, their common objective is to model volatility for multiple assets by capturing their possible interactions. Thus, they provide more information about the underlying assets that could not be recovered by univariate models. However, the practical usefulness of these models are limited by their complexity as the number of asset increases. The paper aims to examine this trade-off between simplicity and extra information by applying these models to forecast VaR for a portfolio of the Australian dollar with twelve other currencies. This provides some insight into the practical usefulness of the additional information for purposes of risk management.