s*********e
发帖数: 1051 | 1 below was posted on my blog before and hope it helpful.
/************************************************
* ARMA - GARCH ESTIMATION USING PROC MODEL *
************************************************/
data garch;
lu = 0;
lh = 0;
ly = 0;
do i = 1 to 1000;
h = 0.3 + 0.4 * lu ** 2 + 0.5 * lh;
u = sqrt(h) * rannor(1);
y = 1 + 0.6 * (ly - 1) + u - 0.7 * lu;
lu = u;
lh = h;
ly = y;
output;
end;
run;
proc model data = garch;
parms ar1 ma1 mu;
y = mu + ar1 ... 阅读全帖 |
|
j******u
发帖数: 7 | 2 fiml the whole system like to estimate each equation by OLS?
如果没甚麽区别,是不是FIML the whole system 比较好?
一见matix 就头疼的说 |
|
m****i
发帖数: 1 | 3 I guess FIML estimates are the same as running reduced form. It
yields consistent estimates. It's definitely different from OLS.
FIML is better. |
|
g*****r
发帖数: 4 | 4 Just finished my econometric comp. FIML is asymptotically
equivalent to 3SLS without constraint on the variance matrix.
We know that if there exists heteroskedasticity or in the case
of simultaneous equation model, 3SLS is preferred, and thus FIML. |
|
j******u
发帖数: 7 | 5 is there some software could estimate the parameter by fiml?
you mean it is equal to 3sls, so is there some software do 3sls?
i knew eview can, but i think it is too complicated.
其实,我不明白FIML的本质的说.
I knew a little about 2sls, is 3sls like 2sls? |
|
c**p
发帖数: 2 | 6 FIML is asymp effiecient if there is the normality assumption--reach CR bound.
3SLS is an extention of IV--2SLS. ---It should definitely consistent.---The
idea of using 3SLS is the stacked simultaneous equations.
Compared with the OLS equation-by-equation, when stacking for the equations,
more information are used now-- that is why we call it Full Information.
FIML & 3SLS are asymp equivaleint if there is no error covariance
resitriction. |
|
g*****r
发帖数: 4 | 7 If your sample size is big, FIML is equivalent to 3SLS, and
LIML is equivalent to 2SLS. The difference between 3SLS and
2SLS is the M matrix. I wonder if you know Q=1/2*H'MH. In
2SLS, M=inv(1/T*W'W), where W is the instrumental matrix. In
3SLS, M=inv(variance matrix/T*W'W). These are all in
homoskadastic case. If the variance matrix is not diagonal
with 1's, 2SLS is not efficient, ie you have to use 3SLS. If
you are sure there exists heteroskadasticity, you have to
construct the M matrix |
|
s********l
发帖数: 439 | 8 what are the basic difference between 3SLS and FIML?
Thanks |
|
c*****i
发帖数: 51 | 9 FIML use idea of MLE s.t. zero blocks restrictions
3SLS use the idea of GLS, weighted LS
or you can think of 3SLS as GMM, with moment E[X'*(Y-Z*delta)] |
|
