HAR族模型与GARCH族模型对不同期限波动率的预测精度比较——基于沪深300指数高频价格的实证分析

Comparison of HAR-family Models' and GARCH-family Models' Forecasting Performances for Volatilities of Different Terms——An Empirical Study using CSI300 index's High-frequency Prices

以对收益条件方差建模的GARCH族波动率模型,及对高频已实现波动建模的HAR族波动率模型为考察对象,采用滚动时间窗的样本外预测,结合具有Bootstrap特性的SPA检验法,分别使用4种常用的损失函数作为评价指标,比较了13种常用波动率模型对沪深300指数短期、中期、长期波动率的预测精度。实证结果表明,对数形式的HAR-RV-CJ模型对短期、中期、长期波动率的样本外预测都具有最高的精度,且对长期波动率预测的优势最为显著;对收益条件方差建模的GARCH族模型即便加入已实现波动作为附加解释变量,其预测精度仍无法超越对已实现波动直接建模的HAR族模型。

This paper takes GARCH-family volatility models which model asset returns' conditional variance and HARfamily volatility models which model high-frequency realized volatility into consideration,comparing 13 commonly used volatility models' out-of-sample forecasting performances for short-term,middle-term and long-term volatilities through moving windows and a bootstrap SPA test under 4commonly used loss functions.Empirical results indicate that,the logarithmic transformed HAR-RV-CJ model has the best out-of-sample forecasting performance for short-term,middle-term and long-term volatilities and the advantage for long-term volatility forecasting is the most significant;while the GARCHfamily models' forecasting performances are normally worse than the HAR-family models even if the lagged realized volatility is included as explanatory variable.