ADF检验中滞后长度的选择——基于ARIMA(0,1,q)过程的模拟证据  

The Lag Length Selection in ADF Test:Simulation Evidence from an ARIMA(0,1,q) Process

作　　者：邓露[1] 张晓峒[1] 

机构地区：[1]南开大学经济学院

出　　处：《数量经济技术经济研究》2008年第9期126-138,共13页Journal of Quantitative & Technological Economics

基　　金：国家自然科学基金资助;编号:70571039

摘　　要：在进行ADF检验时如何确定最优的滞后长度一直是研究者们关注的问题。最近的研究表明,不同的滞后长度选择方法对ADF检验的统计推断影响很大。本文在已有研究的基础上,模拟了更为一般的ARIMA(0,1,q)过程,分析了在不同的数据生成过程、检验式以及样本容量下,各种滞后长度选择方法对ADF检验功效和实际检验水平的影响,认为修正的信息准则通常具有较合理的实际检验水平,而从一般到特殊法具有更为稳健的ADF检验性质。The optimal lag length in estimating Augmented Dickey-Fuller statistics have been concentrated on for years. Previous research indicated that different lag length selection models affect a lot on the statistical inference of ADF test. Based on all the researches available, this paper simulates a more general ARIMA （0, 1, q） process and analyzes the influence of lag length selection criterions to the size and power of the ADF test with different data generating processes, ADF regressions, and sample sizes. Finally, it is proved that the Modified In- formation Criteria always shows a more proper size and the General to Special Criteria has more robust properties in ADF test.

关 键 词：ADF检验 滞后长度 信息准则 修正信息准则 一般到特殊 

分 类 号：F224.0[经济管理—国民经济]