An improved density approximation for the Zivot–Andrews test
针对Zivot-Andrews单位根检验，通过大规模模拟用高斯混合近似分布，提供更精确的临界值和p值计算算法，改进原方法在小样本下的表现。
The unit-root test by Zivot and Andrews(1992) for series with possible structural breaks has been the industry standard for over 30 years. All available software reports the critical values presented in the original article, which were computed with the technology available at the time. By a much larger simulation exercise, the relevant distributions are approximated by Gaussian mixtures. This makes the computation of p -values straightforward and handles finite-sample issues very naturally. We show that the discrepancies between the original critical values and ours are relatively minor, but not negligible in some cases. • The Zivot-Andrews unit root test has been widely used for over 30 years. • Critical values and p-vales can be calculated by larger simulations. • We provide a simple but effective algorithm for calculating p-values.