近单整时间序列Scheffe预测模型的内生性及其修正

对于包含近单整时间序列的预测模型,在进行Scheffe检验时由于内生性问题的影响,导致参数统计量的检验水平过于保守,由此也相应降低了检验功效。通过加入解释变量的超前项与滞后项差分项的动态方法进行修正,并对修正前后的统计量有限样本性质进行仿真比较,结果显示这一修正方法可以有效降低内生性问题对Scheffe检验结果的影响。在小样本条件下,经过修正的Scheffe检验不仅提高了统计量的检验功效,同时明显减少了检验水平的扭曲现象。

The Dynamic Scheffe Estimation in Models with Nearly Integrated Regressor

When the regressors employed in cointegration display a great deal of persistence, the commonly used Scheffe test are typically conservative with low power. The reason is that the limiting distribution of Wald statistic that used for the Scheffe test depends on both the local-to-unity parameter and endogenity. In this paper we use the leads and lags of the first differences of the near integrated regressor to eliminate the endogenity. A series of Monte-Carlo experiments show that the statistics of new test demonstrates an increase in power without size distortion in small samples.