基于Johansen程序的协整参数自举分析

Bootstrap Analysis of Cointegration Parameters Basing on Johansen Procedure

考虑静态和动态两类数据生成过程,利用蒙特卡罗模拟方法,从估计偏差、实际检验水平和检验功效三个方面对基于Johansen程序的长期参数渐近分析和自举分析进行全面比较。结果表明,与渐近分析相比,自举分析可以减小实际检验水平对名义水平的偏差,但要以检验功效的降低为代价。严格意义上,自举分析是降低了"拒真"错误出现的概率,如果VAR(Vector Autoregression)模型能够很好地拟合数据,自举分析可能导致实际检验水平低于名义水平,此时应该慎用。使用Johansen程序估计协整参数时,容易出现异常估计值,因而不宜通过自举法修正估计偏差。

Considering static and dynamic data generating process, this paper compares asymptotic analysis with bootstrap analysis from three aspects that include estimate bias, empirical size and the power of the test by Monte Carlo method. The results show that bootstrap can be used to reduce the distance between empirical size and its nominal level at the cost of reducing the power of the test. Strictly speaking, bootstrap is used to reduce the probability of refusing the truth. If the data can be fitted by VAR model properly, Bootstrap analysis should be used carefully, because it may induce the empirical size less than the nominal size. Furthermore, if cointegration parameters are estimated by Johansen Procedure, there always is a little outlier estimates, so the biases can＇t be corrected by bootstrap.