Assessing Forecasting Performance of Range Volatility Estimators with Linear and Nonlinear Filters: Evidence from KOSPI 200

Assessing Forecasting Performance of Range Volatility Estimators with Linear and Nonlinear Filters: Evidence from KOSPI 200

This study delves on the forecasting performance of three range volatility estimators with ARMA, GARCH, and 2-regime SETAR filters using the KOSPI 200 daily opening, highest, lowest, and closing prices. RMSE has been used as an evaluation criterion. The results of this analysis can be summarized as follows. First, three range volatility estimators showed relatively inferior forecasting performance during the volatile periods but relatively superior forecasting performance throughout the stable periods. Second, Parkinson, and Garman and Klass volatility estimators had much better forecasting performance with a linear ARMA filter while Rogers and Satchell volatility estimator came up with better forecasting performance with nonlinear GARCH and 2-regime SETAR filters. Third, a linear ARMA filter contributed to produce superior forecasting performance with Parkinson, and Garman and Klass volatility estimators while a nonlinear 2-regime SETAR filter helped produce superior forecasting performance with Rogers and Satchell volatility estimator which was designed to reflect the trend into the price process. It is interesting that a GARCH filter always presented a fair forecasting performance regardless of market conditions and the type of range volatility estimators.