基于波动率测量误差的波动率预测模型

本文研究了一种基于波动率测量误差的波动率预测模型,并做了非线性扩展,期望改进预测效果.考虑到文献中关于波动率可能长记忆性和非线性并存的观点,本文以具有长记忆特征的HAR(heterogeneous autoregressive)模型为基础,加入波动率测量误差后模型持续性有所提高,结合非线性的时变参数模型则达到结构变化和减弱异方差的效果.本文用2652天的沪深300高频数据计算的已实现极差波动率来验证模型效果.固定参数下,在HAR型模型中加入测量误差作为调节变量可以较显著地改善样本外预测效果.时变参数下,加入测量误差的HARQ型模型预测效果大多优于对应的HAR型模型.时变参数模型总体上可以改善固定参数模型的预测效果,尤其在预测期较长的情况下改善均是显著的.

Volatility forecasting models based on volatility measurement error

This article studies a volatility forecasting model based on volatility measurement error, and extends it with nonlinear time series model, expecting to improve the forecasting performance. According to the literature, volatility series may have both long memory and nonlinear characteristics. Heterogeneous autoregressive (HAR) model has long memory characteristic. Adjusting HAR model with measurement error promotes model persistence. Further combine the model with time-varying parameter model which is nonlinear, so that structure change and reduction of heteroscedasticity are achieved. Realized range-based volatility is computed based on high-frequency data of CSI300 (2652 trading days) to test the effect of the models. Among the fixed parameter models, the out-of-sample forecasting performance is significantly improved by adding measurement error as adjustment variable. Among the time-varying parameter models, most of the models with measurement error dominate their corresponding models without measurement error. The predictions of time-varying parameter models are in general significantly better than their counterparts with fixed parameters, especially when the prediction period is long.