Forecasting Realized Volatility In Financial Market

As the critical variables of financial risk management, derivatives pricing, portfolio construction and so on, how to forecast volatility accurately has been a hot issues in modern finance research field. Forecasting the volatility accurately has a great theoretical and practical significance on managing asset risk for investors and monitoring market risk for regulators. As a considerable amount of research literature documented, the realized volatility model based on high-frequency data has a significant advantage than traditional volatility such as GARCH family model in measuring and forecasting the true market volatility.In this paper, we proposed a new nonparametric realized volatility forecasting model, that is TVF model, which is based on applying and extending the time-varying probability density function theory proposed by Andrew and Vitaliy (2011), and comparing it with the HAR-RV model proposed by Corsi(2004). For the integrated absorbing of characteristics of the nonparametric TVF model and the linear HAR-RV model, this paper originally construct an adaptive time-varing weight mechanism for combination forecast, thus we can get a higher prediction accuracy than the classical arithmetic average combination forecast model. Taking the5minutes high-frequency price data of csi300stock index futures as the empirical research sample, and calculating the realized volatility in sample, this paper comparing the forecasting ability of the realized volatility forecasting model, that is HAR-RV model, TVF model, arithmetic average combination forecast model and adaptive combination forecast model, and the traditional volatility model based on low-frequency data such as GARCH and FIGARCH model. Then, this paper employ the rolling window strategy for out-of-sample forecasting, and use loss function, SPA test and Mincer-Zarnowitz regression to evaluate the predicting power of above model soundly and reliability.The empirical work show that the four realized volatility forecasting model, that is HAR-RV model, TVF model, arithmetic average combination forecast model and adaptive combination forecast model, both have a significantly higher predicting power than GARCH and FIGARCH model. In addition, the nonparametric TVF model forecast more accurately than HAR-RV model. Furthermore, the arithmetic average combination forecast model doesn’t have a significant advantage than any single model, but the adaptive combination forecast model has.