宝笙宝姑娘 发表于 2011-12-24 20:05
0.1743224
-0.0188469
0.0002171
你列出的LR test statistic公式没有错!
就 nomean 11个系数full_bekk而言,
S-plus 算出的likelihood就是loglikelihood !
S-plus : -7470.418
Matlab : -7.4700e+003
Winrats : -7470.0652
如果都没错,那LR的值怎那么大(likelihood
-7980.331),这是你的疑问
######
full_bekk VS diagonal_bekk
也就是full_bekk
a11 a12 b11 b12
a21 a22 b21 b22
中的系数a12,a21,b12,b21设为零,就是diagonal_bekk
a12,a21,b12,b21设为零之后,diagonal_bekk就没有spillover effects
S-plus中LR test可以根据example 90 去做,
但计算结果似乎有误,因为系数显然与diagonal_bekk不符
这可能就是page 532/1016所说的
BHHH algorithm performs
local optimization in the sense that
the optimal solution it finds may well be just a local optimum
instead of the
global optimum.To make sure that the global optimum
has indeed been reached, start the algorithm using a few different
starting values and see if they all lead to the same optimum.
#################
diagonal_bekk在matlab & e-views的结果接近:
Coefficient Std. Error z-Statistic
OMEGA(1) 0.518868 0.043023 12.06025
BETA(1) 0.949176 0.006293 150.8325
ALPHA(1) 0.210438 0.012218 17.22400
OMEGA(3) 0.406375 0.023036 17.64115
OMEGA(2) 0.321527 0.041833 7.685968
BETA(2) 0.866632 0.014308 60.56796
ALPHA(2) 0.334758 0.008604 38.90611
Log likelihood
-7486.394
Log likelihood -7486.394应该是你想要的
所以建议先做Wald test