论文部分内容阅读
为了更全面细致的刻画时间序列变结构性的特征及其相依性,提出了一类马尔可夫变结构分位自回归模型。利用非对称Laplace分布构建了模型的似然函数,证明了当回归系数的先验分布选择为扩散先验分布时,参数的各阶后验矩都是存在的,并给出了能确定变点位置和性质的隐含变量的后验完全条件分布。仿真分析结果发现马尔可夫变结构分位自回归模型可以全面有效地实现对时间序列数据变结构性的刻画。并应用贝叶斯Markov分位自回归方法分析了中国证券市场的变结构性,结果发现中国证券市场在不同阶段尾部表现出不同的相依性。
In order to characterize the time series variable structure characteristics and their dependencies in a more comprehensive and meticulous manner, a novel Markov variable structure quantile autoregressive model is proposed. The likelihood function of the model is constructed by using asymmetric Laplace distribution. It is proved that when the prior distribution of regression coefficients is prior distribution of diffusion, all posterior moments of parameters exist, Posterior Complete Conditional Distribution of Implicit Variables of Positions and Properties. Simulation results show that the Markov variable structure autoregressive model can effectively and completely depict the variable structure of time series data. The Bayesian Markov Regression autoregression method is used to analyze the variable structure of China’s securities market. The result shows that the Chinese securities market shows different dependences at different stages.