是这样的,小子我初学VAR,根据最优滞后阶数原则来看,VAR是4阶(小样本,lag length criteria最高到4阶。),
Lag LogL LR FPE AIC SC HQ
0 -13.32119 NA 0.001545 2.040149 2.185010 2.047567
1 50.77663 96.14673 1.63e-06 -4.847078 -4.267637 -4.817406
2 67.93770 19.30620 6.87e-07 -5.867212 -4.853190 -5.815286
3 78.08733 7.612227 9.42e-07 -6.010917 -4.562313 -5.936736
4 129.4833 19.27348* 1.70e-08* -11.31041* -9.427226* -11.21398*
然后选择1-4滞后,VAR模型就不稳定了
Root Modulus
1.609511 1.609511
-0.203942 - 1.030096i 1.050090
-0.203942 + 1.030096i 1.050090
-0.814984 - 0.644495i 1.039025
-0.814984 + 0.644495i 1.039025
1.033610 1.033610
0.309807 - 0.862965i 0.916891
0.309807 + 0.862965i 0.916891
0.904443 0.904443
0.543086 - 0.655189i 0.851008
0.543086 + 0.655189i 0.851008
-0.646788 0.646788
但是如果lag length criteria 选择3阶,结果就如下,
Lag LogL LR FPE AIC SC HQ
0 -22.31843 NA 0.003948 2.978639 3.125676 2.993254
1 54.90455 118.1057 1.32e-06 -5.047594 -4.459443 -4.989131
2 73.29175 21.63201* 4.99e-07* -6.151971 -5.122708* -6.049660
3 83.05770 8.042547 6.57e-07 -6.242083* -4.771706 -6.095925*
最优阶数就是2阶。然后,也稳定了。
Root Modulus
0.950446 0.950446
-0.257879 - 0.780251i 0.821762
-0.257879 + 0.780251i 0.821762
0.765494 0.765494
0.592193 - 0.389639i 0.708880
0.592193 + 0.389639i 0.708880
PS(都是一阶单整,而且通过了协整检验)