Re: [Gretl-users] Once again: Questions about estiamtors in models for panel data
On Mon, 3 Sep 2007, Mariusz Doszy? wrote:
I've got a question in context of (fixed effects) panel data
models in case were there is a common constant, cross and
time-specific effects (see Green, Econometric Analysis, 2003,
page 291): yit=x[it]'*beta+m+alpha[i]+gamma[t]+e[it]. m is a
common constant, alpha[i] - group specific intercept, gamma[t] -
time effects (0-1 dummy variables), x[it] contains independent
variables. Unfortunately I can't find formulas for estimators of
[standard errors of] parameters of that kind of models.
If you use the panel command in gretl, with fixed effects and with
the --time-dummies flag, you'll get most of what you want, but not
the standard errors (or coefficient estimates) for the alpha[i]s.
(The assumption in gretl is that most of the time people are not
interested in the specific per-unit constants, only in their joint
significance.)
You can get the alpha[i] estimates after using the panel command
using the accessor $ahat, but right now you can't grab the
standard errors in that way. What you can do is estimate the
fixed effects model manually:
# create the cross-sectional dummies, du_*
genr unitdum
# and the time-period dummies, dt_*
genr timedum
# delete the first of each to avoid perfect collinearity
delete du_1 dt_1
# estimate
ols y 0 x du_* dt_*
This will produce the same estimates as the panel command, but
you'll see all the dummy coefficients and their standard errors.
Allin Cottrell