摘要
本文将基于Gibbs抽样的MCMC算法引入GJR-CAViaR模型,实现模型的贝叶斯推断。GJR-CAViaR模型是含有递归形式的分位数回归方程,尚未有文献提出如何对其进行贝叶斯分析和MCMC估计。本文首先利用不对称拉普拉斯分布建立GJR-CAViaR模型的似然函数,并通过引入标准指数分布和标准正态分布的混合分布得到不对称拉普拉斯分布的参数解析的条件分布,然后讨论模型的Gibbs抽样过程以及算法实现。对上证综指日收益率数据建立GJR-CAViaR模型,并得到模型参数的贝叶斯估计值。在马尔科夫链收敛的前提下,发现中国证券市场VaR具有自回归性质,且呈现收益对风险的不对称特征。这一特征不会受到样本容量大小及置信水平的影响。
The paper aims to introduce the MCMC algorithm based on Gibbs sampling into the GJR-CA- ViaR model, and to carry out the Bayesian estimation of the model. The GJR-CAViaR model is quantile re- gression model with recursive function, and there is no related work on how to do the Bayesian estimation with MCMC algorithm. This paper firstly uses the asymmetric Laplace distribution to establish the likelihood func- tion of the GJR-CAViaR model, then gets the parameter conditional distribution through the mixture of stand- ard exponential distribution and normal distribution to represent the asymmetric Laplace distribution. Finally, the Gibbs sampling process and implementation of the model are discussed. Empirical analysis of the Bayesian estimation of Shanghai Composite Index is carried out by using the GJR-CAViaR model. When Markov Chain is converged, we find the VaR in China Security market is strongly autocorrelated and exists the effect of returns to risk. Moreover, the conclusion remains unchanged under different sample size dence level asymmetric and confi-
作者
张颖
傅强
ZHANG Ying FU Qiang
出处
《中央财经大学学报》
CSSCI
北大核心
2017年第7期87-95,共9页
Journal of Central University of Finance & Economics
