STAR模型中退势单位根检验的小样本性质研究  

The Small Sample Properties for De-trending Unit Root Tests in STAR Frameworks

作　　者：刘雪燕[1,2] 

机构地区：[1]南开大学经济学院数量经济研究所 [2]南开大学滨海学院

出　　处：《统计研究》2009年第3期102-108,共7页Statistical Research

摘　　要：Kapetanios et al.(2003)和刘雪燕(2008)提出了ESTAR和LSTAR模型单位根检验的方法。本文将时间序列退势的OLS和GLS方法与他们提出的单位根检验方法结合,通过蒙特卡洛试验发现,在STAR模型中,对时间序列退势能不同程度的改善单位根检验的功效。若时间序列只存在非零均值,ESTAR模型中OLS退势存在优势;LSTAR模型,样本容量较小时(T<=50),OLS退势的优势较明显,样本容量较大(T>100)时,GLS退势具有了微弱的优势。若序列存在非零的均值和趋势,且样本容量较小时,LSTAR模型中GLS退势的优势较明显,ESTAR模型中OLS退势的优势较明显;样本容量较大时,LSTAR模型中二者功效都很高,ESTAR模型中GLS退势的优势较明显。This paper focuses on the OLS and GLS detrending procedure for unit root tests against alternative hypothesis where the time series data under investigation follow either globally stationary LSTAR or ESTAR processes with deterministic components being present via Monte Carlo simulation. It is found that the proposed testing procedures have considerable power gains against existing nonlinear unit root tests recently proposed by Kapetanios et al. （2003） and Liu Xueyan（2008）. If there is only concept, OLS is better than GLS in ESTAR. In LSTAR, when T 〈 50, OLS is better, but when T 〉 100, GLS is better. If there is concept and trend, when T〈50, GLS is better in LSTAR, but OLS is better in ESTAR. When T〉 100, GLS is better in ESTAR, but both are very good in LSTAR.

关 键 词：非线性 STAR模型 单位根检验 小样本性质 蒙特卡洛模拟 

分 类 号：O212[理学—概率论与数理统计]