Forecasting gains by using extreme value theory with realised GARCH filter

Abstract

Early empirical evidence suggests that the realised generalised autoregressive conditional heteroskedasticity (GARCH) model provides significant forecasting gains over the standard GARCH models in volatility forecasting. We extend this literature in quantile forecasting by implementing conditional extreme value theory (EVT) framework with realised GARCH. We generate one-step-ahead value-at-risk (VaR) and expected shortfall (ES) forecasts for the S&P CNX NIFTY index using 14 standalone GARCH and GARCH-EVT models. In out-of-sample comparisons, the GARCH-EVT specification generally outperforms the standalone GARCH models. In general, the realised-GARCH EVT models provide the best forecasting performance. This finding is robust to the choice of different realised volatility estimators used to estimate realised GARCH.