近年来，随着高频金融数据可获得性的提高，基于高频数据的数量分析方法日益受到重视。Andersen和Bollerslev等人使用已实现波动率来研究金融资产的波动行为，大大改进了波动率度量的精度。建立在金融高频数据基础上的已实现波动测度是一种新的波动率度量方法，具有无模型、计算方便等优点，并且是波动率的一致估计量，近年来广泛应用于高频金融数据的研究中。一些常用的已实现波动测度包括已实现波动（RV），已实现双幂变差（RBV）和已实现极差（RRV）等。金融波动的长记忆性是金融市场中的一个重要现象。波动的长记忆性是指序列中相距较远的时间间隔具有显著的自相关性，即历史事件的影响会持续影响着未来，而且这种影响在相当长的时间滞后仍然存在。由于金融市场的波动性反映了金融动态风险的大小，因而对金融波动长记忆性的研究，可以更好地分析风险的变化趋势，进一步降低风险的不可知因素，从而达到规避风险、控制风险的目的。在大量的实证研究中，已经发现已实现波动测度存在显著的长记忆性。同时，由于已实现波动测度是建立在高频的数据基础上的，不可避免地受到数据微观结构噪声的影响，特别当抽样频率越高，影响就越显著。因此，在实证当中构建已实现波动测度需要考虑不同抽样频率的影响。本文的目的是研究外汇市场的主要货币及中国上证股票市场的部分股票的已实现波动测度的长记忆特征、长记忆性与抽样频率之间的关系、几种已实现波动测度方法在描述长记忆性特征上的效果，以及长记忆性的产生机制。采用的已实现波动测度包括RV、RBV和RRV，采用的长记忆性研究方法是去势波动分析（DFA）方法。我们的研究表明：外汇市场主要货币及中国上证股票市场股票的三种已实现波动测度都具有长记忆性，随着抽样间隔增大，大部分的货币与股票已实现波动测度的长记忆性减弱，但是部分货币与股票对于抽样频率不敏感，没有表现出相关性。在研究中，我们也发现已实现极差在描述长记忆性方面具有稳定性，是较好的描述长记忆性的估计量，并且通过实证分析证明长记忆性现象是数据产生过程的一个固有的特征。

In recent years, with the improvement of availability of high-frequency financial data, more attention is paid on the data analysis methods based on high-frequency data. The work of Andersen and Bollerslev exploiting realized volatility to study the volatility of financial asset has greatly improved the precision of volatility measurement. Realized volatility measures constructed from high-frequency intraday returns, which are not only model-free, easily computed but also unbiased ex-post estimators of daily return volatility, are widely used in the research of high-frequency financial data in recent years. Realized volatility (RV), realized bi-power variation (RBV) and realized range-based volatility (RRV) are key methods of realized volatility measures.Long-term memory of financial volatility in the financial markets is an important phenomenon. Long-term memory of volatility actually refers to the autocorrelation function of the volatility time series decay exponentially according to a negative power law. It means that historical events will affect the future continually, and the influence lasts for a considerable period. Financial asset price volatility is often manifested in rich, complex features. It is very important to study the long-term memory of financial market volatility in theory and practice of financial risk management.A number of previous empirical studies have documented the realized volatility measures of financial asset returns often exhibit a long-term memory property. Meanwhile, market microstructure noise effect cannot be avoided because of high sampling frequency. In particular, microstructure noise effect will be exacerbated as the sampling frequency of the volatility increases. Hence, the factor of different sampling frequencies must be considered when constructing realized volatility measures in empirical analysis.The aim of this paper is to discuss the long-term memory properties exploiting DFA method, relationship between sampling frequencies and long-term memory of realized volatility measures, the effect of RV, RBV and RRV estimators in both foreign exchange market and China’s stock market, and the mechanism of long-term memory through empirical results. We find that, first, realized volatility measures exhibit significant long-term memory from one-minute to sixty-minute sampling time interval in both markets. Second, with the time interval increases, the long-term memory of realized volatility measures weakened gradually of most currencies and stocks, but some are shown the insensitivity to sampling frequency. Third, we also find that RRV presents relatively stable property on long-term memory, so it can be chosen as a more efficient estimator for market predicting models. Last, we prove that a long-term memory phenomenon is an inherent characteristic of the data generating process.