论文部分内容阅读
本文设波动率的估计误差服从异方差假定,在对已实现波动率进行建模时,根据方差变化来设定模型的自回归系数,构建基于高频数据的HARQ(F)模型。在此基础上,考虑中国股市波动的跳跃行为及杠杆效应,先后构建了HARQ(F)-CJ模型和LHARQ(F)-CJ模型,以改善波动率模型的拟合效果和预测能力。本文假设,当期已实现波动率或其连续成分的估计误差的方差越大,它对未来真实波动率的解释力度则越差,因而其对应系数越小。对上证综合指数近15年的五分钟高频数据进行实证研究发现,基于估计误差异方差假定的动态系数能够提高已实现波动率模型的拟合效果和预测能力。其中,对日回归系数进行基于估计误差方差的动态调整是模型改进的关键。同时考虑中国股市波动的跳跃行为及杠杆效应的LHARQ-CJ模型在所有模型中表现最优。
In this paper, the estimation error of volatility is subject to the assumption of heteroscedasticity. When the realized volatility is modeled, the autoregressive coefficient of the model is set according to the variance variance, and the HARQ (F) model based on high frequency data is constructed. On this basis, taking into account the jump behavior and leverage effect of the volatility in China’s stock market, the HARQ (F) -CJ model and the LHARQ (F) -CJ model were successively constructed to improve the fitting effect and forecasting ability of the volatility model. This paper assumes that the greater the variance of the estimated volatility or its continuous component in the current period, the worse it interprets the future real volatility and the smaller its corresponding coefficient. Empirical research on the five-minute high frequency data of the last 15 years found that the dynamic coefficient based on heteroscedasticity of estimation error can improve the fitting effect and forecasting ability of the realized volatility model. Among them, the daily regression coefficient based on the variance of the estimated error of the dynamic adjustment is the key to model improvement. The LHARQ-CJ model, which considers both the jump behavior and the leverage effect of volatility in China’s stock market, performs best in all models.