基于奇异值分解去势的非线性趋势序列单位根检验研究

Unit Root Test for Nonlinear Trend Series Based on De-trending with SVD

如何克服ADF与PP单位根检验法对非线性趋势平稳序列的伪检验,提高单位根检验的功效,是非平稳时间序列分析的重要问题。本文基于奇异值分解的思路,构造出检验非平稳时间序列单位根的SVD-RMA检验法,此方法将时间序列的趋势项与干扰项分离,然后用递归均值调整法对干扰项进行检验。仿真实验表明,SVD-RMA法对线性与非线性趋势、甚至结构突变过程的检验功效都非常好;对非线性趋势平稳的检验而言,SVD-RMA检验得到正确结论的可能性要远远好于ADF与PP检验。

To nonlinear trend stationary series, how to overcome the Spurious Tests with ADF and PP unit root test methods is an important problem. This paper proposes the SVD-RMA test approach, which de-trends time series with SVD firstly and then tests unit root for residuals with RMA. Monte Carlo test shows that it has effective power to linear and nonlinear, even structure mutation process. For nonlinear trend process, SVD-RMA test is far better than ADF or PP test.