基于MCMC算法的贝叶斯面板单位根检验

针对面板单位根检验存在检验势不稳定和原假设设置主观选择的问题,提出基于面板数据分位自回归模型,选择非对称Laplace分布的似然函数对模型进行贝叶斯分位回归分析.结合参数的完全条件分布设计MCMC抽样算法,进行贝叶斯分位单位根检验,并利用Monte Carlo模拟实验研究了贝叶斯分位单位根检验的有效性与可行性.研究结果表明,基于面板数据分位自回归模型的贝叶斯单位根检验方法解决了检验势不稳定以及原假设主观设置的问题,能够给出更全面稳健的单位根检验判断.

Bayesian Unit Root Tests in Panel Data by Using MCMC Algorithm

Because the test power of the traditional panel unit root tests is unstable and the choice of the null hypothesis of traditional panel unit root tests is subjective, this paper proposed a Bayesian quantile unit root test for panel data based on asymmetric Laplace distribution. On the basis of quantile autoregression panel data model, the full conditional distributions of parameters were inferred and MCMC algorithm was designed. And then, Bayesian quantile unit root tests were conducted. Numerical results were produced via a combination of Monte Carlo simulation, from which we find that Bayesian quantile unit root tests are noticeably efficient and feasible. As a result, it is shown that Bayesian quantile unit root tests solve unstable power problems and the subjective choice of the null hypothesis. Furthermore, the tests are more robust and can provide more complete information.