格兰杰因果检验中不同检验方法的功效比较

The Efficacy Comparison for Different Test Methods in Granger Causality Test

文章运用蒙特卡洛模拟方法,对F、LR、Wald、Mwald四种格兰杰因果检验统计量的检验效果进行了小样本模拟。研究结果表明:(1)极小样本条件下(n￡50),LR统计量检验效果最佳,MWald统计量检验效果较差。(2)随着样本的增大,Wald、Mwald统计量检验效果将会逐步改善。(3)较大样本条件下(n=200,400),F、LR、Wald、Mwald四个统计量检验效果差异变小;在二维系统中,Wald、Mwald统计量检验效果最佳,LR统计量检验效果最差。(4)F统计量检验效果比较稳定,不随样本的增加出现较大改变。因此,在极小样本情况下,适合采用LR统计量检验序列间的格兰杰因果关系;在较大样本情况下,适合采用Wald、Mwald统计量检验序列间的格兰杰因果关系;而F统计量既适用极小样本的检验,也适用较大样本的检验。

This paper utilizes Monte Carlo simulation method to conduct a small sample simulation on＇the granger causality test statistics for F,LR, Wald and Mwald. The study results indicate that （1） under the condition of very small sample （n ≤ 50）, the test result of LR is best, yet Mwald relatively worse; （2） as the sample quantity increases, the test results of LR and Mwald improve gradually; （3） under the condition of large sample （n=200,400）, the Difference in test results of F, LR, Wald and Mwald shrinks, and in two-dimensional system, the test results of Wald and Mwald perform best, yet LR the worst; （4） the test result of F remains relatively stable without much change with sample increasing. Based on these characteristics, the paper draws a conclusion： under the condition of a very small sample, the LR is suitable for the Granger causality for sequences; under the condition of a large sam- ple, the Wald and Mwald are more applicable; however, the F statistics is appropriate for both Minimal sample test and large sam- ple test simultaneously.