Re: [Gretl-users] Generalized Methods of Moments in Gretl
On Wed, February 15, 2006 12:08, Al Leong wrote:
Hi,
Thank you for developing a great package. I do enjoy
using Gretl.
I have a query and some suggestion:
Query:
Are you planning to develop a GMM module in gretl?
GMM is increasingly popular amongst economist, see
Hayashi(2001) and Alastair R. Hall (2005).
This comes up now and then. Yes, we have some plans, but I'm afraid not for
the near future. *Some* special cases of GMM estimators might be done sooner
(panel data estimators, for example --- see below).
2. Very often, I add new variables by defining the
variables myself. But this is 2-3 layers down the
menu.Possible to Define New variables up one level?
Suggestion: use the console or write a script. Productivity increases by
several orders of magnitude.
3. Heteroskedasticity-correction should apply to panel
models or pooled models as well (may wish to take a
leave from Eviews 5).
4. Aspects of panel data modules could be improved,
making it easier to do fixed effect and random
effect.Panel data is very popular nowadays.
As Allin knows, I am not particularly happy with the current state of panel
data estimators. The thing that IMO we need most badly is support for dynamic
panel data models; I think that at least the Arellano-Bond estimator should be
put in asap.
I have a couple of ideas in mind; Allin and I discussed the issues at some
length when he came to Ancona last year, and IIRC we agreed that our panel
code needs some serious reshaping. It can be done, but it'll take some time.
In the meantime, we got a bit carried away with time-series related work plus
some other stuff which is not ready for public consumption yet. But it's
definitely in our long-term todo list.
5. Running White's tests for heteroskedasticity often
crash the program.
The trouble with White's test is that with a large number of regressors you
get a HUGE number of additional variables in the auxiliary regression used to
compute the actual test statistic. Your problem may be related to that, in
terms of memory issues, of the program simply taking too long to deliver the
result and only seemingly crashing. Could you post a test case?
--
Riccardo "Jack" Lucchetti
Dipartimento di Economia
Facoltà di Economia "G. Fuà"
Ancona