自己顶一下吧

额，我用的是s-plus做的，数据是ibm的9190个对数收益率AR(2)-GARCH(1,1)-t5计算VaR
(1)先运用数据拟合AR(2)-GARCH(1,1)模型
s-plus语句：
>spec= garch(ibm~ar(2), ~garch(1,1),cond.dist="t",dist.par=5)
Iteration
0
Step Size =
1.00000
Likelihood = -1.00000e+010
Iteration
0
Step Size = 0.200000
Likelihood =
2.88635
Iteration
1
Step Size =
1.00000
Likelihood =
2.88205
Iteration
1
Step Size = 0.200000
Likelihood =
2.88788
Iteration
2
Step Size =
1.00000
Likelihood =
2.89014
Iteration
2
Step Size =
2.00000
Likelihood = -1.00000e+010
Iteration
3
Step Size =
1.00000
Likelihood =
2.89267
Iteration
3
Step Size =
2.00000
Likelihood =
2.89140
Iteration
4
Step Size =
1.00000
Likelihood =
2.89281
Iteration
4
Step Size =
2.00000
Likelihood =
2.89279

Convergence R-Square = 0.00002040485 isless than tolerance = 0.0001
Convergence reached.
> spec

Call:
garch(formula.mean = ibm ~ ar(2),formula.var =
~ garch(1, 1), cond.dist =

"t",dist.par = 5)

Mean Equation: structure(.Data = ibm ~ar(2)
, class = "formula"
)

Conditional Variance Equation:structure(.Data =
~ garch(1, 1)
, class = "formula"
)

Coefficients:



C
3.604e-004

AR(1)
1.803e-003

AR(2) -3.145e-002

A
2.656e-006

ARCH(1)
4.935e-002
GARCH(1)
9.385e-001
因此，可以求出GARCH模型的各个参数。
（2）运用GARCH模型单步向前预测
首先运用s-plus语句，求出条件方差
> names(spec)
[1] "residuals"   "sigma.t"     "df.residual" "coef"        "model"      
[6] "cond.dist"   "likelihood"  "opt.index"   "cov"         "prediction"
[11] "call"        "asymp.sd"    "series"     
> spec$sigma.t
将所得条件标准差平方得条件方差，然后代入GARCH模型中进行向前一步预测（当然还有第9189天和第9190天的收益率），求出第9191天的方差和收益率。
（3）将方差和收益率代入VaR模型中计算，VaR=r-1.65*标准差

额，大家可以提提意见和建议

eveiws不能直接计算，但系数求出来带入你的VaR模型不就可以了吗？