基于动态因子结构的贝叶斯分位面板协整研究

针对传统面板协整检验在建模过程中易受异常值影响以及其原假设设置的主观选择问题,本文利用动态公共因子刻画面板数据潜在的截面相关结构,提出基于动态因子的截面相关结构的贝叶斯分位面板协整检验,结合各个主要分位数水平下参数的条件后验分布,设计结合卡尔曼滤波的Gibbs抽样算法,进行贝叶斯分位面板协整检验;并进行Monte Carlo仿真实验验证贝叶斯分位面板协整检验的可行性与有效性。同时,采用中国各省金融发展和经济增长的面板数据进行实证研究,结果发现在各主要分位数水平下中国金融发展和经济增长之间具有协整关系。研究结果表明:贝叶斯分位面板协整检验方法避免了传统面板数据协整方法由于原假设设置不同而发生误判的问题,克服了异常值的影响,能够提供全面准确的模型参数估计和协整检验结果。

Test for Bayesian Quantile Panel Cointegration Based on Dynamic Factor Model

The traditional panel cointegration methods are based on conditional mean models, which is relatively sensitive to outliers. Meanwhile, the cointegration results may differ when the null hypothesis varies, which causes the difficulty of the cointegration tests in the case of different null hypothesis. This paper proposes Bayesian quantile panel cointegration method on the basis of the panel data dynamic factor model, which implements dynamic common factor to characterize the possible cross-section dependence in panel data. To get the parameter estimates of the model and conduct Bayesian quantile panel cointegration test in the major quantiles, the Gibbs sampling algorithms are designed. We also conduct a Monte Carlo study to illustrate the feasibility and validity of Bayesian quantile panel cointegration. Finally, through the empirical application of Chinese financial development and economic growth, we find there are cointegration relationships between Chinese financial development and economic growth in the major quantiles. It shows that Bayesian quantile cointegration method avoids the bias caused by the set of null hypothesis in the traditional panel cointegration and tends to be impervious to outliers. It also shows that Bayesian quantile panel cointegration can provide full and exact information on parameter estimation and cointegration tests.