Improving volatility forecasts of the Nikkei 225 stock index using a realized EGARCH model with realized and realized range-based volatilities

Abstract

This paper applies the realized exponential generalized autoregressive conditional heteroskedasticity (REGARCH) model to analyze the Nikkei 225 index from 2010 to 2017, utilizing realized variance (RV) and realized range-based volatility (RRV) as high-frequency measures of volatility. The findings show that REGARCH models outperform standard GARCH family models in both in-sample fitting and out-of-sample forecasting, driven by the dynamic information embedded in high-frequency realized measures. Incorporating multiple realized measures within a joint REGARCH framework further enhances model performance. Notably, RRV demonstrates superior predictive power compared to RV, as evidenced by improvements in forecast accuracy metrics. Moreover, the forecasting results remain robust under both rolling-window and recursive evaluation schemes.