SIGNAL EXTRACTION IN LONG MEMORY STOCHASTIC VOLATILITY
针对长记忆随机波动率模型估计复杂的问题，提出一种基于频域半参数维纳-科尔莫戈罗夫滤波的波动率提取新方法，无需设定参数形式，适用于平稳和非平稳信号，并通过蒙特卡洛和道琼斯工业指数日收益率数据验证了有效性。
Long memory in stochastic volatility (LMSV) models are flexible tools for the modeling of persistent dynamic volatility, which is a typical characteristic of financial time series. However, their empirical applicability is limited because of the complications inherent in the estimation of the model and in the extraction of the volatility component. This paper proposes a new technique for volatility extraction, based on a semiparametric version of the optimal Wiener–Kolmogorov filter in the frequency domain. Its main characteristics are its simplicity and generality, because no parametric specification is needed for the volatility component and it remains valid for both stationary and nonstationary signals. The applicability of the proposal is shown in a Monte Carlo and in a daily series of returns from the Dow Jones Industrial index.