具有时变波动率持续性的已实现EGARCH模型及其实证研究

Realized EGARCH Model with Time-Varying Volatility Persistence and Its Empirical Analysis

研究表明,金融资产收益率的波动率具有时变性、聚集性和非对称性等复杂的特征.此外,波动率还具有很强的持续性,而且这种持续性具有时变特征.与此同时,随着计算机及电子信息技术的快速发展,高频数据越来越容易获得,充分利用高频数据可以提高波动率估计的准确性.Hansen和Huang采用高频数据构建了已实现EGARCH(REGARCH)模型,研究发现其比传统GARCH模型和EGARCH模型具有更好的数据拟合效果.但是,REGARCH模型仍然没有考虑波动率持续性的时变特征.文章在REGARCH模型的基础上,引入随时间变化的GARCH系数,采用混频数据抽样(mixed-data sampling,MIDAS)方法,将其与解释变量(已实现波动率)联系起来,构建了具有时变波动率持续性的REGARCH(TVP-REGARCH-MIDAS)模型.采用上证综合指数数据进行实证研究,结果表明:上证综合指数的波动率具有很高的持续性,而且这种持续性展现明显的时变特征,具体而言,上证综合指数的波动率持续性与已实现波动率呈负相关,即在高(低)的波动率时期会产生低(高)的持续性;TVP-REGARCH-MIDAS模型相比EGARCH模型、REGARCH模型和TVP-GARCH-MIDAS模型具有更好的样本内拟合效果以及样本外波动率预测效果.

It has been well documented that volatility of financial asset returns exhibits complex features,such as time variation,clustering and asymmetry.In addition,the volatility has strong persistence and particularly,this persistence has the characteristic of time variation.Meanwhile,with the rapid development of computer and electronic information technology,the high-frequency data can be obtained more and more easily.Making full use of high-frequency data is useful to improve the accuracy of volatility estimates.Hansen and Huang introduce the realized EGARCH(REGARCH) model to exploit the high-frequency data information,and find that the proposed model has better empirical fit performance than the traditional GARCH model and EGARCH model.However,the REGARCH model still fails to account for the dynamic behaviors of volatility persistence.To address this issue,this paper extends the REGARCH model to introduce a time-varying GARCH coefficient.By using the flexible mixed-data sampling(MIDAS) approach to link the GARCH coefficient to an explanatory variable(realized volatility),this paper introduces the REGARCH model with time-varying volatility persistence(TVP-REGARCH-MIDAS model).Empirical analysis using the Shanghai Stock Exchange Composite Index shows that the volatility of Shanghai Stock Exchange Composite Index has high persistence,and the volatility persistence exhibits obvious time-varying characteristics.Specifically,the volatility persistence of Shanghai Stock Exchange Composite Index is negatively related to the realized volatility,that is,we can observe high(low) persistence during low(high)volatility period.Finally,we observe that the TVP-REGARCH-MIDAS model outperforms the EGARCH model,REGARCH model,and TVP-GARCH-MIDAS model in terms of in-sample fit and out-of-sample volatility forecast.