长记忆随机波动模型的估计与波动率预测——基于中国股市高频数据的研究  

Estimating Long Memory Stochastic Volatility Model and Forecasting Using Volatility Based on High Frequency Data from Chinese Stock Markets

作　　者：王春峰[1] 庄泓刚[1] 房振明[1] 卢涛[1] 

机构地区：[1]天津大学管理学院,天津300072

出　　处：《系统工程》2008年第7期29-34,共6页Systems Engineering

基　　金：国家杰出青年科学基金资助项目(70225002)

摘　　要：为了有效捕捉中国股市波动率的长记忆性,提高远期波动率的预测精度,本文基于中国股市高频数据建立了长记忆随机波动模型,检验高频数据中时变的"日历效应"成分的频率,有效地对"日历效应"进行滤波。使用频域内拟极大似然方法估计LMSV模型参数,为了提高计算效率应用混沌优化算法进行最优搜索。对比了高频数据直接建模和已实现波动率方法建模的预测结果发现,通过高频数据估计的LMSV模型可以很好保留高频数据中所包含的信息量,克服信息丢失问题,预测结果要优于已实现波动率方法建模预测的结果。To catch long memory character of china stock market volatility and improve forecasting precision of long-rang, a long memory stochastic volatility model based on high frequency return of Chinese stock markets is established, a new method of detecting and fitting the calendar effect component is introduced firstly. Secondly, estimating long memory stochastic volatility model using the frequency domain quasi maximum likelihood estimation and chaos optimization algorithm, which can improve the efficiency. At last, we compare the forecasting performance of the model using high frequency return and the model using realized volatility, finding that LMSV model using high frequency return can preserve the information of the data and improve estimation and forecasting precision of long-rang.

关 键 词：LMSV模型 高频数据 ARFIMA模型 日历效应 已实现波动性 

分 类 号：F830[经济管理—金融学]