已实现波动率估计中不同降噪方法的比较分析及实证

Realized Volatility Estimation for High-Frequency Date

已实现方差作为波动率的估计,在理论上是二次变差的无偏和一致估计。但在现实中,它受到市场微观结构噪声的影响而偏离其理论值。本文考虑如何降低市场微观结构噪声干扰而有效地估计波动率,主要讨论已实现核估计和双频子抽样已实现方差估计这两种纠偏降噪技术。我们进行了比较分析,并以来自中国股票市场的高频交易数据对各种已实现波动率估计进行了全面深入的实证研究,验证了各种统计量的性质。实证结果表明,目前常用的利用时间间隔数据的已实现波动率估计低估了风险,因此纠偏降噪技术对于风险管理具有非常重要的现实意义。

The realized variance （RV） is an unbiased and consistent estimator of the quadratic variance in theory. In the real world, however, the realized variance is distorted by market microstructure noise and the distortion is even severe under high-frequency settings. It is a key how to reduce the impact of noise in the high-frequency financial data. We turn to the techniques of volatility estimation under microstructure noise disturbing. Two categories of RV-type estimators are discussed, namely realized kernel variance and two scales subsampling realized variance. We also show the properties of, and make comparison between, these RV- type estimators by empirical study using data from Chinese stock market. Our empirical results show that the noise reduction technologies are very important in risk management.