Abstract

The importance of time varying volatility in securities prices (e.g. GARCH) has by now been amply established in the literature, both in terms of the magnitude and pervasiveness of the phenomenon, and in terms of its significance for risk management in institutional portfolios. Less attention has been devoted to multivariate conditional heteroskedasticity, in spite of the fact that securities are typically held in portfolios rather than in isolation. Recently, Kroner and Ng (1995) have introduced a method for nesting the four most commonly used multivariate GARCH models, allowing for comparative tests of the performance of the models. We propose to apply the Kroner and Ng technique to both daily and intradaily returns on foreign exchange rates, to obtain performance estimates. These conditional covariances will then be used to calculate value at risk (VaR) forecasts for foreign currency portfolios. Daily and intradaily VaR forecasts will be evaluated and compared.