[Gretl-users] Re: combine figure and multivariate distributions
On Wed, Aug 10, 2022 at 8:17 AM Olasehinde Timmy <timmexdareal(a)gmail.com> wrote:
Let me express my second question clearly.
I am trying to estimate multivariate garch models using the full information maximum
likelihood method. I am quite familiar with the usage of the single equation type, but not
with the system types.
I thought there should be a way of representing the multivariate maximum likelihood in a
matrix form for estimation purposes. My major concern is when there are many variables
involved, and I believe representing the maximum likelihood in matrix form would make it
easy.
In a multivariate system it may perhaps be convenient to express the
loglikelihood in matrix form (say, as a g x T matrix L, where g is the
number of dependent variables and T the number of observations), but
for use with gretl's "mle" you need to provide either a scalar value
for the total loglikelihood (across variables and observations) or,
preferably, a series or column vector containing the per-observation
values. The notional L matrix would have to be "compressed".
However, unless the dependent variables are mutually independent you
presumably want to maximize their joint loglikelihood, which would not
just be a matter of adding up individually calculated l_t's.
Allin