基于MCMC方法的两类波动模型的应用比较

MCMC-based comparison of two classes of volatility models

采用中国股市数据,运用基于Gibbs抽样的MCMC方法对ARCH GARCH模型族与SV模型族进行了比较分析.实证结果显示,无论是从收益率序列峰度系数的描述看,还是从平方收益率序列自相关函数的描述来看,SV模型族都优于GARCH模型族;进一步,基于t分布的模型模拟效果总是优于基于正态分布的模型,看出股市收益率序列并不服从正态分布.

Using data from China's stock markets and a Gibbs sampling based MCMC method a comparison of ARCH-GARCH models with SV models are made in this paper. The models-based simulation results suggest that in view of either description of kurtosis of the return rate series, or reproduction of the strong autocorrelations of squared return rate series, SV models are superior to ARCH-GARCH models. Furthermore, in comparison of t distribution-based models with N-distribution-based models, the former is often better than the latter, and it can, thus, be declared that the return series of stock market does not obey the N-distribution.