Volatility forecasting performance of two-scale realized volatility

Abstract

This article examines the forecasting performance of two-scale realized volatility (TSRV) measure in comparison to that of the conventional sparsely sampled realized volatility (SSRV) measure, using selected volatility forecasting models. There is evidence that the forecasts based on TSRV are more efficient and less biased than those based on SSRV, for all the forecasting models employed. This implies that the quality of forecast predominantly depends on the quality of estimate, and not on the forecasting model. With TSRV estimates, the exponentially weighted moving average models for daily forecasts, and the random walk model for weekly and monthly forecasts, marginally dominate the other models on efficiency and bias criteria.