Financial Forecasting: Comparative Performance of Volatility Models in Chinese Stock Markets

Abstract

This paper presents empirical tests and comparisons of GARCH family models and nonparametric models for predicting the volatility of Chinese stock markets. Since the volatility of financial asset returns often exhibits asymmetry, fat-tails and long-range memory property in the stock market, nonparametric models maybe have better performance. By the criteria of mean absolute forecast error (MAE), mean squared error (RMSE) and the hit rate (HR), empirical results show that support vector machine (SVM), a new nonparametric tool for regression estimation, outperforms GARCH family models (GARCH, EGARCH, FIGARCH), moving average and neural network in improving predictive accuracy.