On the Intra-Daily Performance of GARCH Processes

In this paper, we show that the use of an alternative time scale can eliminate the inefficiencies in the estimation of a GARCH model caused by intra-daily seasonal patterns. Even so, however, the temporal aggregation properties of the GARCH model do not hold at the intra-daily frequencies, revealing the presence of several time-horizons components. Besides, distinct characteristics were identified in the very short (less than 2 hours) and the very long (several months) run. Finally, the out-of-sample predictive power of GARCH for the volatility was found to be lower than the historical volatility itself implying the presence of other sources of heterogeneity