文献类型：期刊文献

中文题名：基于经验似然贝叶斯计算的稳定分布参数估计

英文题名：Parameter Estimation of Stable Distribution Based on Bayesian Computation With Empirical Likelihood

作者：钱瑾[1];钱夕元[2]

机构：[1]携程旅游网络技术有限公司;[2]华东理工大学理学院,上海200237

年份：2018

卷号：0

期号：7

起止页码：18

中文期刊名：统计与决策

外文期刊名：Statistics & Decision

收录：CSTPCD;;国家哲学社会科学学术期刊数据库;北大核心:【北大核心2017】；CSSCI:【CSSCI2017_2018】；

基金：国家高科技研究发展计划(“863计划”)资助项目(2015AA20107);上海市经信委专项资金资助项目(140304)

语种：中文

中文关键词：高峰厚尾;稳定分布;经验似然;贝叶斯计算

外文关键词：heavy tail and excess kurtosis; stable distribution; empirical likelihood; Bayesian computation

摘要：金融市场数据通常具有“高峰厚尾”的特征,普通正态分布很难拟合这类数据。稳定分布族不存在显式密度函数表达式,常用的一些估计方法无法处理它的参数估计问题。可是稳定分布被证明可以较好地拟合金融市场收益率这类具有“高峰厚尾”特征的数据。文章提出了一种新颖的解决稳定分布参数估计的方法：基于经验似然的贝叶斯计算的稳定分布参数估计方法,将其对比正态分布应用到上证指数收益率数据的分布拟合中去,并用稳定化的PP图检验拟合效果。同时,为了解决基于历史数据的金融数据的预测问题,提出了基于经验似然的贝叶斯预测,并利用该方法结合稳定分布预测上证指数收益率的特征分布情况。
Financial data always has a feature of heavy tail and excess kurtosis, and common normal distribution can not fit such data properly. Stable distribution does not have explicit expression of density function and the traditional computing method can not meet the requirement of dealing with its parameter estimation. Nevertheless, stable distribution has been proved to be able to relatively better fit such data characterized by feature of heavy tail and excess kurtosis as financial market returns. This paper presents a novel parameters estimate method of stable distribution which is based upon Bayesian computation with empirical like- lihood. And then the paper applies the contrast normal distribution to the distribution fitting of the yield rate data of Shanghai Se- curities Composite index, and uses the stable [PP] plot to check the fitting effect. In the meantime, in order to solve the problem of financial data based on the historical data, the paper puts forward Bayesian prediction based on the empirical likelihood, and uses this method and combines stable distribution to forecast the featured log return distribution of Shanghai Securities Composite index.