面板单位根检验(gap 是不同省份的城乡收入差距):
困惑I:如何选择滞后项?从大到小,怎么知道大的对还是小的对?信息准则,怎么在面板单位根检验时做?
如我选择滞后项到2:
. levinlin gap,lags(2) trend
Levin-Lin-Chu test for gap Deterministics chosen: constant & trend
Pooled ADF test, N,T = (29,29) Obs = 754
Augmented by 2 lags (average) Truncation: 9 lags
coefficient t-value t-star P > t
-0.35505 -12.583 -1.66097 0.0484
如选择项到5:
. levinlin gap,lags(5) trend
Levin-Lin-Chu test for gap Deterministics chosen: constant & trend
Pooled ADF test, N,T = (29,29) Obs = 667
Augmented by 5 lags (average) Truncation: 9 lags
coefficient t-value t-star P > t
-0.43809 -11.916 3.34629 0.9996
. 得出结论全然不同啊。
困惑2:使用fisher时,dfuller和pp差别怎么那么大??
如我选dfuller,则
xtunitroot fisher gap,dfuller lags(5) tr
………………
Inverse chi-squared(58) P 48.6963 0.8029
Inverse normal Z 0.9184 0.8208
Inverse logit t(149) L* 0.8528 0.8024
Modified inv. chi-squared Pm -0.8638 0.8062
如我选择pp,则:
. xtunitroot fisher gap,pp lags(5) tr
Inverse chi-squared(58) P 185.1546 0.0000
Inverse normal Z -6.7263 0.0000
Inverse logit t(149) L* -8.3156 0.0000
Modified inv. chi-squared Pm 11.8060 0.0000
简直是清尘浊水,永不相见的俩东西嘛!!诸位大佬中佬三思后给解答下吧……