Hello everybody,
first off, thank you all for suggestions.
Riccardo, I know that anything bigger than arma(2,1) is going to be a crappy model. Im
forecating volatility and arma is one of the options, the obviously sloppy one but
nevertheless the benchmark and I can prove nowehere else than on out of sample that its
great fit is just an illusion. That's actually the point...
So, the ARMA(6,5)-GARCH(1,1) model with T-distributed residuals in Eviews gives estimates
which are to be seen at
http://eubie.sweb.cz/gretl_forum/Eviews_estimation.PNG
When I plug in the very same numbers as initial values into my gretl code, I get
convergence, however, I arrive at different estimates, see
http://eubie.sweb.cz/gretl_forum/gretl_forum_correct_starting_values.txt
When I however change those starting values just a little bit (namely ga, arc and dof by
as little as 0.05), convergence is gone, see
http://eubie.sweb.cz/gretl_forum/gretl_forum_slightly_incorrect_starting_...
The most interesting point is that Eviews is telling me that LL = 3978 when MLE is found.
When I start gretl with those "correct" values, the reported LL = 2036. Hence, I
would suspect a difference in LL function. Eviews claims to be using
http://www.scribd.com/doc/44246798/EViews-7-Users-Guide-II page 198 Eq 7.11 which I have
coded as
mle ll = -0.5*log(pi*(dof-2)) - lngamma(dof/2) + lngamma((dof + 1)/2) -0.5*log(h) -
0.5*(dof + 1)*log(1 + (dof - 2)^(-1)*h^(-1)*e^2)
and I cant see the difference. My script is at
http://eubie.sweb.cz/gretl_forum/ARMA65garch11t%20pro%20gretl%20forum.inp
I dont know where to look for a possible error anymore.
Thank you all once again,
Daniel