我跑了VARMA-GARCH 和 BEKK-GARCH
但是其結果卻非常的不同??

system(model=var0)
variables DLUS DLTINDEX
lags 1
det constant
end(system)
garch(p=1,q=1,model=var0,mv=cc,variance=varma,pmethod=simplex,piters=10,hmatrices=hh0,rvectors=rr0)


MV-GARCH, CC with VARMA Variances - Estimation by BFGS
Convergence in 38 Iterations. Final criterion was0.0000076 <=0.0000100
Usable Observations 1959
Log Likelihood 14527.9087

Variable Coeff Std Error T-Stat Signif
************************************************************************************
1.DLUS{1} -5.5169e-003 0.0266 -0.207100.83592845
2.DLTINDEX{1} -0.01084.6858e-003 -2.307620.02102052
3.Constant -5.9571e-0056.5685e-005 -0.906920.36444671
4.DLUS{1} -0.2690 0.0831 -3.236830.00120865
5.DLTINDEX{1} 0.0273 0.0250 1.088640.27631239
6.Constant 4.0219e-0042.4330e-004 1.653070.09831601
7.C(1) 7.1112e-0085.9797e-008 1.189220.23435141
8.C(2) 2.6977e-0068.1983e-007 3.290500.00100009
9.A(1,1) 0.1012 0.0153 6.618590.00000000
10. A(1,2) 7.5732e-0033.3414e-003 2.266450.02342373
11. A(2,1) 0.0873 0.0515 1.694760.09012066
12. A(2,2) 0.0737 0.0112 6.581660.00000000
13. B(1,1) 0.8720 0.0191 45.662100.00000000
14. B(1,2) -0.02818.9331e-003 -3.145190.00165981
15. B(2,1) 0.0455 0.1253 0.363450.71626627
16. B(2,2) 0.9235 0.0137 67.345800.00000000
17. R(2,1) -0.3285 0.0127 -25.797200.00000000


system(model=var0)
variables DLUS DLTINDEX
lags 1
det constant
end(system)
garch(p=1,q=1,model=var0,mv=bek,pmethod=simplex,piters=10,hmatrices=hh0,rvectors=rr0)

MV-GARCH, BEKK - Estimation by BFGS
Convergence in 36 Iterations. Final criterion was0.0000038 <=0.0000100
Usable Observations 1959
Log Likelihood 14525.0768

Variable Coeff Std Error T-Stat Signif
************************************************************************************
1.DLUS{1} -0.0121530520.022566589 -0.538540.59020306
2.DLTINDEX{1} -0.0112482250.004240145 -2.652790.00798289
3.Constant -0.0000672110.000058968 -1.139790.25437341
4.DLUS{1} -0.2699092640.082878573 -3.256680.00112722
5.DLTINDEX{1} 0.0323478890.022102495 1.463540.14331964
6.Constant 0.0004673590.000247091 1.891450.05856484
7.C(1,1) 0.0004093390.000057248 7.150260.00000000
8.C(2,1) 0.0003815100.000322301 1.183710.23652918
9.C(2,2) 0.0013715520.000220222 6.228040.00000000
10. A(1,1) 0.2991886160.019854065 15.069390.00000000
11. A(1,2) 0.0117210560.073571687 0.159310.87342089
12. A(2,1) 0.0068530760.004100286 1.671370.09464950
13. A(2,2) 0.2619406230.022351627 11.719090.00000000
14. B(1,1) 0.9469683870.006509017 145.485610.00000000
15. B(1,2) -0.0192618160.024969676 -0.771410.44046496
16. B(2,1) -0.0033268610.001443724 -2.304360.02120242
17. B(2,2) 0.9600270190.006811212 140.948050.00000000

VARMA-GARCH的A(1,2)和B(1,2)的顯著性和BEKK-GARCH的A(1,2)和B(1,2)幾乎是完全相反??
為什麼不同的模型會差這麼多??
我換了四五個變數都是結果相反，是VARMA-GARCH解釋的方向和BEKK-GARCH解釋的方向不同嗎??
已經濟合理性來看VARMA-GARCH的結果比較合理，應該是美元的異常變動會造成股票波動變動
BEKK-GARCH的解釋方向就反過來了，怎麼會??