STAR非线性平稳性检验中误设定的伪检验研究

本文通过理论分析和蒙特卡洛仿真模拟,研究平稳性检验中选用的统计量与数据生成过程不一致时,非线性ESTAR、LSTAR与线性DF检验法能否得出正确的结论.研究表明,二阶LSTAR与ESTAR模型可用相同的检验方法,但前者的非线性特征更强.当数据生成过程为线性AR,或非线性ESTAR、二阶LSTAR模型时,使用DF或ESTAR检验法可得出大致正确的结论,但LSTAR检验法完全失败.数据生成过程的非线性特征越强,ESTAR较DF检验方法的功效增益越高；线性特征越强,DF的功效增益越高.当转移函数F(θ,c,zt)中θ较大导致一阶泰勒近似误差较大或c非0时,标准ESTAR与LSTAR非线性检验法失去应用条件.θ较大或c偏离0较远时,数据生成过程中线性成分增强,用线性DF检验可获得更好的检验结果.

Spurious Test Research of Misspecification in STAR Nonlinear Stationary Test

This paper addresses whether stationary test including nonlinear ESTAR, LSTAR and linear DF methods can draw right conclusions when the chosen test statistics are inconsistent with DGP(data generating process) through theoretical analysis and Monte Carlo simulation. The results show that the second-order LSTAR can use the same test method as ESTAR model but it has stronger nonlinearity. When DGP is nonlinear ESTAR, second-order LSTAR or linear AR, DF or ESTAR method can generally achieve right results, but LSTAR fails completely. Stronger nonlinearity of DGP brings ESTAR higher power than DF, and stronger linearity brings DF higher power than ESTAR. When is big, leading to big approximation error of first-order Talyor Expansion or if is nonzero in transition function , the standard ESTAR and LSTAR will fail to satisfy the applicable conditions. If is big or is far from 0, DGP will have more linear components, therefore linear DF test can achieve better results.