The Leverage Effect in Reactions of Realized Volatility to Macroeconomic News Surprises

In this paper a HARST and AR(22) model are estimated for realized volatility for several indices and exchange rates. A comparison is made between models where only macroeconomic news announcement date dummies are included and models where actual news surprises are added as explanatory variables. The comparisons are made on both fit and forecast strength. It is shown that the newer HARST model outperforms the HAR and AR models in both forecasting and fitting strength. Further, it is shown that the used (leveraged) news surprises provide better fits but provide no significant increases in forecasting power.