面板单位根Hadri检验的有偏性及其修正

The Modification to Bias of Hadri Unit Root Test in Panel Data

Hadri(2000)根据单一时间序列的KPSS检验,提出了以平稳性为原假设的面板数据单位根检验。但我们的仿真试验表明,对短时间序列数据,其基于残差的拉格朗日乘数(LM)统计量是有偏的,使得在此基础之上进行的Hadri检验不再服从标准正态渐进分布。本文通过蒙特卡罗仿真对LM统计量进行了修正,修正之后的Hadri检验统计量的渐进分布为标准正态分布。仿真的结果显示,修正了LM值的Hadri检验具有更好的小样本性质和更高的检验势。

According to the KPSS test of single time series, Hadri （2000） proposed a unit root test in panel data for the null hypothesis that the individual series are stationary. But we find that the residual-based Lagrange Multiplier （LM） test is biased for the short time series, which leads the asymptotic distribution of the Hadri test not to be standard normal. We correct the LM statistics by the Monte Carlo simulation, and the asymptotic distribution of the modified Hadri statistics is standard normal. By simulation we find that the modified Hadri test performs better in the sizes and powers in small samples than the unmodified one.