This study investigates the daily volatility forecasting for the Standard & Poor’s 100 stock index series from 1997 to 2003 and identifies the essential source of performance improvements between distributional assumption and volatility specification using distribution-type (GARCH-N, GARCH-t, GARCH-HT and GARCH-SGT) and asymmetry-type (GJR-GARCH and EGARCH) volatility models through the superior predictive ability (SPA) test. Empirical results indicate that the GJR-GARCH model achieves the most accurate volatility forecasts, closely followed by the EGARCH model. Such evidence strongly demonstrates that modeling asymmetric components is more important than specifying error distribution for improving volatility forecasts of financial returns in the presence of fat-tails, leptokurtosis, skewness and leverage effects. Furthermore, if asymmetries are neglected, the GARCH model with normal distribution is preferable to those models with more sophisticated error distributions.
Relation:
Expert Systems with Applications 37(7), p.4928-4934
