具有GARCH-normal-error时序的贝叶斯单位根检验

Bayesian Unit Root Test for Time Series With GARCH-normal-error

针对时间序列呈现尖峰厚尾、条件方差时变特征时,DF检验功效性较低的问题,文章基于贝叶斯理论设计了MCMC抽样算法,检验具有GARCH-normal-error时序的平稳性,并用Monte Carlo模拟仿真验证贝叶斯单位根检验的可行性与有效性,结果表明:贝叶斯单位根检验的稳健性较好,样本容量、ρ的大小及波动参数和单整程度的差异对单位根检验势的影响均不大;当ρ<0.7时,DF检验的势略高于贝叶斯方法,但当ρ> 0.7时,贝叶斯方法的势远高于DF检验;DF k检验与DFτ检验的可靠性受样本容量的影响较大。

In view of the low efficacy of DF test when the time series is characterized by sharp peak and thick tail and time-varying conditional variance, this paper designs MCMC sampling algorithm based on Bayesian theory to test the stability of time series with GARCH-normal-error, and verifies the feasibility and validity of Bayesian unit root test by Monte Carlo simulation. The results go as below: The robustness of the Bayesian unit root test is good, and the differences of sample size, ρ, fluctuation parameters and integration degree have little influence on the potential of unit root test. When ρ <0.7, the potential of DF test is slightly higher than that of Bayesian method, but when ρ >0.7, the potential of Bayesian method is much higher than that of DF test. The reliability of DF k test and DF τ test is greatly affected by sample size.