基于跳跃、好坏波动率和马尔科夫状态转换的高频波动率模型预测

基于跳跃、好坏波动率的视角,采用比ABD检测更稳健的ADS检测法进行甄别跳跃,提出HAR改进模型,进一步考虑到实际波动率的非线性和高持续性动态,文章引入马尔科夫状态转换机制以构建对应的MRS-HAR族模型,推导其参数估计方法,并运用滚动时间窗预测技术和MCS检验评估预测模型结果,并采取不同的窗口期进行稳健性检验.以上海期货交易所的黄金连续(AU0)期货合约为研究对象,实证研究表明:结合马尔科夫状态转换机制,跳跃波动在上涨行情时会抑制未来波动性;结合马尔科夫状态转换机制,好坏波动率在上涨行情时正负冲击相对平衡,而在下跌行情时好(坏)波动率抑制(加剧)未来波动性;MCS检验证实,结合马尔科夫状态转换的MRS-HAR族模型相比于HAR族模型具有更优的预测精度,进一步考虑由ADS检测修正的好坏波动率和符号跳跃能够改善波动率模型的预测能力,其中基于符号跳跃和马尔科夫状态转换的MRS-HAR-RV-SJ模型展现了最高的预测精度.

The Forecasting Performance of the High-Frequency Volatility Models Based on Jumps,Good-Bad Volatility and Markov Regime-Switching

Based on the perspective of jumps and good-bad volatility,this paper adopts ADS detection method which is more robust than ABD detection to screen the jumps,then this paper proposes extended HAR models.And further considering nonlinear and highly persistent dynamics of realized volatility,this paper introduces the Markov Regime-Switching to construct the corresponding MRS-HAR model,derives its parameter estimation method and is used to discriminate the prediction results of the extended model and the comparison model by using the rolling time windows prediction technique and the newly MCS test and taking a different window period for robustness testing.Taking the Shanghai Futures Exchange’s gold continuous(AU0)futures contract as the research object,the empirical research shows that combined with the Markov Regime-Switching,the volatility of jumps will suppress the future volatility when the market is rising.Combined with the Markov Regime-Switching,good volatility and the bad volatility are relatively balanced when the market is rising,while the market is falling good(bad) volatility suppression(increase) future volatility.MCS test confirms that the MRS-HAR family model combined with the Markov Regime-Switching has better prediction accuracy than the HAR-RV family model,and further considering good volatility,bad volatility and signed jump which are corrected by ADS detection method the volatility model can improve the predictive power of the volatility model.The MRS-HAR-RV-SJ model based on signed jump and the Markov Regime-Switching exhibits the highest prediction accuracy.