Prediction of Realized Volatility Based on Realized-GARCH-Kernel Model: The Comparison of CHINA and US

Abstract

We propose a Realized-GARCH-Kernel model to predict realized volatilities of 50 ETF in China and S&P500 index in U.S..The Kernel density fitting on disturbance term and semi-parametric method make our model perform well both statistically and economically. First, our model has the lowest in- and out-of-sample prediction errors among five comparable prediction models. The result is robust in eight measures of realized volatility. Second, in both China and U.S. markets, straddle option trading strategies with volatilities predicted with our model generate larger monthly profit and greater Sharpe ratio. Our model is useful in practical investment.