偏方差波动率预测模型

Partial Variance Volatility Forecasting Model

论文研究了阈值数量和大小对已实现波动率预测的影响,并提出了偏方差已实现波动率预测模型——HAR-PV(G),该模型进一步提高了已实现波动率的预测效果。考虑到不同大小收益对已实现波动率的影响具有非对称性,以HAR-RS模型为基础,选取不同数量和不同大小的阈值组合对日内收益进行分割,并计算对应的偏方差,从而构建HAR-PV(G)模型。论文以沪深300指数为研究对象,比较了不同HAR-PV(G)模型的样本外预测能力。样本外分析表明,阈值数量为3的平分偏方差模型具有比传统HAR、HAR-RS以及其他阈值组合的偏方差模型更好的预测能力。全样本的参数分析也显示阈值数量为3的平分偏方差模型对数据的拟合效果更出众。

This paper proposes a partial variance prediction model,HAR-PV(G),which can test the effects of different quantities and different sizes of thresholds on the prediction of realized volatility,aiming at improving the predicting effect of realized volatility.Due to the asymmetric effects of different intraday returns on the realized volatility,this paper selects threshold combinations of different quantities and sizes to segment intraday returns based on HAR-RS model.Then,this paper calculates the corresponding partial variance to construct HAR-PV(G)model.This paper uses 5-minute high-frequency trading data of CSI 300 index to compare in-sample and out-of-sample performances of HAR-PV(G)model.The results show an equal division of partial variance model with 3 thresholds could achieve a better out-of-sample performance compared with traditional HAR,HAR-RS and other HAR-PV(G)models with different threshold combinations.Meanwhile,this equal division of partial variance model with 3 thresholds also has an excellent fitting effect in the in-sample analysis.