分位数自回归变点模型的贝叶斯分析及应用  

Bayesian Analysis and Application of Quantile Auto-regression Model

作　　者：郭婧 何幼桦[1] 

机构地区：[1]上海大学理学院,上海200444

出　　处：《统计与决策》2017年第23期14-18,共5页Statistics & Decision

基　　金：国家自然科学基金资助项目(11371242)

摘　　要：文章使用贝叶斯方法对分位数自回归模型系数的变点问题进行分析。基于对分位数回归模型的Gibbs抽样方法的研究,给出了变点模型参数的满条件分布以及MCMC抽样算法。仿真结果表明,所得到变点时刻的MCMC抽样链条有很好的收敛性。使用分位数自回归变点模型对中小板综合指数极差数据进行实证分析,结果表明了数据的波动具有非对称性,在较低分位数上波动的滞后性要弱于高分位数。在不同分位数上得到的变点时刻的估计,与该时间段中小板市场波动的实际表现相一致。This paper employs Bayesian method to analyze the change points of coefficients of quantile auto-regression model. Based on the study of Gibbs sampling method for quantile regression model, the paper gives the full conditional distribution of the parameters of the changing point model and the MCMC sampling algorithm. Simulation result shows that the MCMC sampling chains obtained by MCMC method have a good convergence property. And then the paper uses quantile autoregressive change point model to empirically analyze the comprehensive range data of the Chinese small and medium capital stocks, and the analysis results show that the data fluctuations are asymmetric; the lag phenomenon in the lower quantile is weaker than in the high quantiles; estimates of change-points at different quantiles are consistent with the actual performance of the market during the same time period.

关 键 词：贝叶斯方法 分位数自回归模型 变点问题 MCMC抽样 M-H算法 

分 类 号：O212[理学—概率论与数理统计]