统计研究 ›› 2011, Vol. 28 ›› Issue (3): 99-106.

• 论文 • 上一篇    

协整参数的自举推断

针对完全修正最小二乘(full-modified ordinary least square,简称FMOLS)估计方法,给出一种协整参数的自举推断程序,证明零假设下自举统计量与检验统计量具有相同的渐近分布。关于检验功效的研究表明,虽然有约束自举的实际检验水平表现良好,但如果零假设不成立,自举统计量的分布是不确定的,因而其经验分布不能作为检验统计量精确分布的有效估计。实际应用中建议使用无约束自举,因为无论观测数据是否满足零假设,其自举统计量与零假设下检验统计量都具有相同的渐近分布。最后,利用蒙特卡洛模拟对自举推断和渐近推断的有限样本表现进行比较研究。   

  • 出版日期:2011-03-15 发布日期:2011-04-19

Bootstrapping the Inference on Cointegration Parameters

In this paper, the bootstrap inference procedure based on FMOLS method is proposed, and the bootstrap statistics is proven to have the same asymptotic distribution as the test statistics under the null hypothesis. The study on the power of bootstrap test reveals that the distribution of the restricted bootstrap statistics is indeterminate when the sample data doesn’t meet the null hypothesis. So its empirical distribution can’t be used to estimate the exact distribution of test statistics, although it has the good finite sample performances on test size. It is suggested that the unrestricted bootstrap should be applied in the empirical research, because its bootstrap statistics has the same asymptotic distribution as the test statistics whether the sample data meet the null hypothesis or not. At last, the finite sample performances of bootstrap inference and asymptotic inference are compared by Monte Carlo simulation.   

  • Online:2011-03-15 Published:2011-04-19

摘要: 针对完全修正最小二乘(full-modified ordinary least square,简称FMOLS)估计方法,给出一种协整参数的自举推断程序,证明零假设下自举统计量与检验统计量具有相同的渐近分布。关于检验功效的研究表明,虽然有约束自举的实际检验水平表现良好,但如果零假设不成立,自举统计量的分布是不确定的,因而其经验分布不能作为检验统计量精确分布的有效估计。实际应用中建议使用无约束自举,因为无论观测数据是否满足零假设,其自举统计量与零假设下检验统计量都具有相同的渐近分布。最后,利用蒙特卡洛模拟对自举推断和渐近推断的有限样本表现进行比较研究。

关键词: 协整, 自举法, 完全修正最小二乘法

Abstract: In this paper, the bootstrap inference procedure based on FMOLS method is proposed, and the bootstrap statistics is proven to have the same asymptotic distribution as the test statistics under the null hypothesis. The study on the power of bootstrap test reveals that the distribution of the restricted bootstrap statistics is indeterminate when the sample data doesn’t meet the null hypothesis. So its empirical distribution can’t be used to estimate the exact distribution of test statistics, although it has the good finite sample performances on test size. It is suggested that the unrestricted bootstrap should be applied in the empirical research, because its bootstrap statistics has the same asymptotic distribution as the test statistics whether the sample data meet the null hypothesis or not. At last, the finite sample performances of bootstrap inference and asymptotic inference are compared by Monte Carlo simulation.

Key words: Cointegration, Bootstrap, FMOLS