似无关动态协整模型中的水平扭曲与校正

The Size Distortion and Correction for Seemingly Unrelated Dynamic Cointegrating Regression

本文在解析似无关动态协整模型及其动态最小二乘估计的基础上,从理论上揭示了关于协整参数的假设检验存在严重的水平扭曲,即对协整参数约束的Wald检验统计量的渐近卡方分布存在严重的有限样本扭曲。进一步,本文应用自举抽样技术对水平扭曲进行了有效校正。基于本文的发现,我们建议在对似无关动态协整模型中的参数进行假设检验时,为保证结论的准确性,应使用自举抽样推断技术产生统计量值并由此来形成检验结论。

This paper, based on analyzing the seemingly unrelated dynamic cointegrating regression and its feasible generalized least square estimate, designs a Monte Carlo experiment to test the true null hypotheses for exposing the size distortion. The experiment results reveal that there are severe size distortions in testing the hypotheses of cointegration parameters under finite sample. Furthermore, for correcting the size distortion, we conduct the bootstraps to the null hypotheses. The results show that bootstrapping the asymptotic distribution significantly correct the distortion. Thus, this paper provides us some powerful evidences to employ the bootstrap statistics for testing some restrictions on cointegration parameters in empirical studies.