针对股指波动所具有的动态结构信息特征,在状态空间建模理论的框架下,将服从Markov过程的潜在波动状态变量引入状态方程,同时在观测方程中考虑极值点的影响,构造出一类非高斯Markov随机波动状态空间模型。针对传统的MCMC方法对该类模型估计时效率低下的缺陷,设计了基于序贯Monte Carlo方法的贝叶斯滤波算法进行仿真分析,并且从算法效率和准确性方面对两种方法进行了比较。通过对沪深300股指波动的实证研究表明:对于一类非线性非高斯状态空间模型,贝叶斯滤波算法在保证估计精度的同时较MCMC方法更加有效率,能够有效刻画股指波动的动态结构特征。
To demystify the regime-switching information hidden in the stock index,a kind of non-Gauss nonlinear state space model is brought forward to allow for fat-tails in the mean equation innovation to capture the changes in volatility caused by economic forces and for Markov switching process in the latent volatility equation.In the sequential Bayesian perspective we provide a Bayesian filtering algorithm for parameter learning and state filtering of the model.In empirical study,the regime-switching information based on the stochastic volatility model is demystified by using this algorithm on CSI300 index spots open price futures.In the applications,we compare the algorithm with MCMC method in both efficiency and accuracy.We find that the Bayesian filtering algorithm outperforms existing MCMC.