Improving GARCH based volatility forecasting using six predictor models

Over the past decades, the worldwide financial markets have been continually evolving. Along with this is a rapidly growing need for an accurate and efficient volatility forecasting method. In this study, the proponents sought to determine if the incorporation of the available volatility estimators would improve the accuracy and efficiency of the conditional variance model of GARCH (1,1) and how intraday prices affect the performance of the GARCH model. The research covered a period of intraday level data from 2008-2016. Within the time period parameter, the proponents gathered a total of 200 prices and observation of the PSEi through a Bloomberg terminal accessed from a local bank in the Philippines. The volatility estimators used in the research were: Parkinson model, Garman-Klass model, Rogers-Satchell model, realized volatility model, realized bipower variation model and, overnight volatility model. The data used in the research study were tested using the MAE, MAPE, and RMSE and, were ran using the EViews program. The results yielded that some of the estimators showed some slight improvement of accuracy based on the standard GARCH model. This meant that the models failed to forecast the volatilities effectively