10:30 a.m.
Hi everybody,
I'm wondering whether it would be an idea to also obtain the Hansen
over-id test after the dpanel command (dynamic Arellano/Blundell-Bond
style GMM estimator). Currently gretl only shows the non-robust Sargan
test. This seems to be a little bit inconsistent considering that by
default robust standard errors are reported.
I am aware that for example Stata's built-in xtdpdsys command does _not_
report that result, either. (Other contributed Stata routines do.) Don't
know why the Stata people decided against it.
So two relatively concrete questions:
Is it possible with the content of the $model bundle after dpanel to
construct the Hansen test? I tend to think no, but maybe it is.
Secondly, is it actually possible to switch off the robust standard
errors in dpanel? Just out of curiosity and for comparison.
thanks
sven
5:54 p.m.
Am 25.01.2021 um 11:30 schrieb Sven Schreiber:
So two relatively concrete questions:
Is it possible with the content of the $model bundle after dpanel to
construct the Hansen test? I tend to think no, but maybe it is.
Follow-up on this:
perhaps the stuff that is saved to the bundle with
the option --keep-extra would help here?
Secondly, is it actually possible to switch off the robust standard
errors in dpanel? Just out of curiosity and for comparison.
Another question about the dpanel GUI:
It seems that when running using the dpanel dialog window, the variant
is automatically switched to dpdstyle (judging from the remark in the
printout), even though the option cannot be set. Is this intended? I
thought it's just an alternative that has to be switched on explicitly
(via --dpdstyle, scripting only).
thanks
sven
10:48 a.m.
Am 26.01.2021 um 18:54 schrieb Sven Schreiber:
Am 25.01.2021 um 11:30 schrieb Sven Schreiber:
>
> So two relatively concrete questions:
> Is it possible with the content of the $model bundle after dpanel to
> construct the Hansen test? I tend to think no, but maybe it is.
Follow-up on this: perhaps the stuff that is saved to the bundle with
the option --keep-extra would help here?
>
> Secondly, is it actually possible to switch off the robust standard
> errors in dpanel? Just out of curiosity and for comparison.
OK, to give future mailing list readers some more info here: Actually
there is the section 24.5 in the gretl user guide that says something
about the Sargan-vs.-Hansen issue. Funny that nobody remembered that
part of the documentation ;-) There it is already mentioned that in the
literature there is no clearcut distinction between the labels "Sargan"
and "Hansen". BTW, Stata's own xtdpdsys also talks about Sargan only.
Currently my understanding is that some people call a Hansen over-id
test what in gretl (and elsewhere) is the Sargan over-id test after the
2-step GMM estimation. The 2-step aspect is important here. And you can
easily try and verify that the Sargan tests in gretl will yield
different results after a 1-step vs. a 2-step estimation.
There is perhaps a slight inconsistency in gretl's dpanel 1-step
handling: The reported standard errors (vcov-matrix) are of the robust
variant (if I understand correctly), but the Sargan test after 1-step is
only valid under stronger assumptions. Not necessarily any need for
action, just saying.
Another question about the dpanel GUI:
It seems that when running using the dpanel dialog window, the variant
is automatically switched to dpdstyle (judging from the remark in the
printout), even though the option cannot be set. Is this intended? I
thought it's just an alternative that has to be switched on explicitly
(via --dpdstyle, scripting only).
Intended or not, indeed running dpanel from the GUI always adds the
--dpdstyle option, as can be seen by inspecting the command log. If you
want to switch that off, currently you have to write the dpanel command
line in a script (or the console). But of course dpdstyle doesn't do any
harm.
cheers
sven
7:36 p.m.
On Thu, 28 Jan 2021, Sven Schreiber wrote:
OK, to give future mailing list readers some more info here:
Actually
there is the section 24.5 in the gretl user guide that says something
about the Sargan-vs.-Hansen issue. Funny that nobody remembered that
part of the documentation ;-) There it is already mentioned that in the
literature there is no clearcut distinction between the labels "Sargan"
and "Hansen". BTW, Stata's own xtdpdsys also talks about Sargan only.
Currently my understanding is that some people call a Hansen over-id
test what in gretl (and elsewhere) is the Sargan over-id test after the
2-step GMM estimation.
Yes. The xtabond2 "Hansen test" is identical to the gretl "Sargan
test" after two-step estimation. As Roodman says (p. 98 of his "How
to do xtabond2"), "When the user requests the Sargan test for
'robust' one-step GMM [xtabond2 quietly performs] the second GMM
step to obtain and report a consistent Hansen statistic."
> Another question about the dpanel GUI:
>
> It seems that when running using the dpanel dialog window, the variant
> is automatically switched to dpdstyle (judging from the remark in the
> printout), even though the option cannot be set. Is this intended? I
> thought it's just an alternative that has to be switched on explicitly
> (via --dpdstyle, scripting only).
Intended or not, indeed running dpanel from the GUI always adds the
--dpdstyle option, as can be seen by inspecting the command log. If you
want to switch that off, currently you have to write the dpanel command
line in a script (or the console). But of course dpdstyle doesn't do any
harm.
That's now fixed in git. The GUI behaves in the same way as the
command: "dpdstyle" is a selectable option rather than the default.
Allin
1423
days inactive
1426
days old
3 comments
2 participants
participants (2)
-
Allin Cottrell -
Sven Schreiber