基于MCMC的贝叶斯变结构金融时序GARCH模型研究

Bayesian Inference on Structure Change in GARCH Models Based on MCMC

针对变结构GARCH模型没有解析形式的条件后验分布的问题。借助辅助变量把没有具体解析形式的后验分布转化为一系列完全条件分布,实现了变结构GARCH模型参数的贝叶斯估计。中国外汇市场波动性的实证研究,表明了辅助变量-Gibbs抽样有效的解决了贝叶斯变结构GARCH模型中的高维数值计算问题,并发现其波动持续性是由时间序列的状态转移引起的。

In the GARCH model with structural changes, simple Gibbs sampler is not feasible to simulate its posterior densities directly, because the analytical knowledge of conditional posterior densities is not available. After the introduction of auxiliary variables, the full conditionals can substitute for the awkward forms of conditional posterior densities to implement Gibbs iteration, which carried out the estimation of the GARCH model. The empirical analysis of the Chinese foreign exchange market illustrates that auxiliary sampler resolved the difficulties of the high dimension numerical integral in structure changing GARCH model effectively and the serious pseudo-persistence is caused by regime switching of the time series.