On the Intradaily Performance of GARCH Processes

Abstract

The Auto Regressive Conditional Heteroskedastic (ARCH) model (Engle, 1982) and its Generalised version (GARCH) (Bollerslev, 1986) are now widely used in the foreign exchange literature (Bollerslev et al., 1992) and as a framework for empirical studies of the market microstructure such as the impact of news (Goodhart and Figliuoli, 1991; Goodhart et al., 1993) and government interventions (Goodhart and Hesse, 1993; Peiers, 1997), or inter-and intra-market relationships (Engle et al., 1990; Baillie and Bollerslev, 1990). A main assumption behind this class of models is the relative homogeneity of the price discovery process among market participants at the origin of the volatility process. In other words, the conditional density of one GARCH process can adequately capture the information content of news. In particular, GARCH parameters for the weekly frequency theoretically derived from daily empirical estimates are usually within the confidence interval of weekly empirical estimates (Drost and Nijman, 1993).