关键词: 协整, 自举法, Johansen程序, 蒙特卡罗模拟

Abstract:

Considering static and dynamic data generating process, this paper compares asymptotic analysis with bootstrap analysis from three aspects including 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.

Key words: Cointegration, Bootstrap, Johansen Procedure, Monte Carlo Simulation