基于自回归的bekk-garch比较复杂,eviews做不了,winrats比较容易。最近在做这个,把代码放上来吧,其实也很短。我是8阶滞后的var,garch(1,1)。vec-bekk-garch同理。
Code:
system(model=var11)
variables lnstock lnfuture
lags 1 to 8
det constant
end(system)
estimate
*
garch(p=1,q=1,model=var11,mv=bekk,pmethod=simplex,piters=10) / lnstock lnfuture
结果(只写了条件方差-协方差矩阵的系数):
Variable Coeff Std T-Stat Signif
1 C(1,1) 0.001032 0.000120 8.587830 0.000000
2 C(2,1) 0.001549 0.000154 10.071760 0.000000
3 C(2,2) 0.000158 0.000138 1.145280 0.252094
4 A(1,1) 0.157956 0.031076 5.082970 0.000000
5 A(1,2) -0.109753 0.035981 -3.050280 0.002286
6 A(2,1) 0.058407 0.027533 2.121320 0.033895
7 A(2,2) 0.352988 0.031287 11.282160 0.000000
8 B(1,1) 0.999249 0.003176 314.673280 0.000000
9 B(1,2) 0.047299 0.002357 20.066890 0.000000
10 B(2,1) -0.024188 0.003297 -7.336500 0.000000
11 B(2,2) 0.920728 0.003615 254.666180 0.000000
如果不基于VAR模型的bekk结果:
Variable Coeff Std T-Stat Signif
1 Mean(LNSTOCK) 0.00030894 0.000177923 1.73637 0.08249855
2 Mean(LNFUTURE) 0.000277494 0.000188376 1.47309 0.14072808
3 C(1,1) 0.000805 0.000136 5.920140 0.000000
4 C(2,1) 0.001232 0.000151 8.176810 0.000000
5 C(2,2) 0.000000 0.000292 0.000344 0.999725
6 A(1,1) 0.220629 0.020851 10.581490 0.000000
7 A(1,2) 0.038536 0.024150 1.595740 0.110547
8 A(2,1) -0.055154 0.011318 -4.872960 0.000001
9 A(2,2) 0.154234 0.012055 12.793660 0.000000
10 B(1,1) 0.963497 0.003973 242.498370 0.000000
11 B(1,2) -0.001466 0.004606 -0.318250 0.750296
12 B(2,1) 0.014122 0.001460 9.671210 0.000000
13 B(2,2) 0.969268 0.002558 378.923420 0.000000
14 Shape(GED) 1.927284 0.063626 30.290610 0.000000
可以很明显地看到,如果不基于VAR模型,只有2对1有波动溢出,1对2没有。但是在VAR基础上构建bekk-garch以后,波动溢出是双向的。