4:02 p.m.
Dear All,
To reproduce:
<hansl>
nulldata 20
set seed 13
y1 = normal()
y2 = normal()
x1 = normal()
x2 = normal()
y = normal()
z1 = normal()
z2 = normal()
z3 = normal()
setinfo y --description="the dependent variable"
setinfo y1 --description="endogenous variable"
setinfo y2 --description="another endogenous variable"
setinfo x1 --description="exogenous variable"
setinfo x2 --description="another exogenous variable"
setinfo z1 --description="instrument I"
setinfo z2 --description="instrument II"
setinfo z3 --description="instrument III"
mt1 <- tsls y 0 y1 y2 x1 x2; x1 x2 z1 z2 z3
mt1a <- tsls y 0 y1 y2 x1 x2; 0 x1 x2 z1 z2 z3
# mt1 and mt1a gives the same
catch tsls y 0 y1 y2 x1 x2; x1 x2 z1 z2
mt2 <- tsls y 0 y1 y2 x1 x2; 0 x1 x2 z1 z2
</hansl>
???
Oleh
4:06 p.m.
Am 10.10.2017 um 18:02 schrieb oleg_komashko(a)ukr.net:
mt1 <- tsls y 0 y1 y2 x1 x2; x1 x2 z1 z2 z3
mt1a <- tsls y 0 y1 y2 x1 x2; 0 x1 x2 z1 z2 z3
# mt1 and mt1a gives the same
catch tsls y 0 y1 y2 x1 x2; x1 x2 z1 z2
mt2 <- tsls y 0 y1 y2 x1 x2; 0 x1 x2 z1 z2
</hansl>
???
Sorry I'm probably too slow -- what did you expect and what did surprise
you (and why)?
thanks,
sven
4:36 p.m.
On Tue, 10 Oct 2017, oleg_komashko(a)ukr.net wrote:
mt1 <- tsls y 0 y1 y2 x1 x2; x1 x2 z1 z2 z3
mt1a <- tsls y 0 y1 y2 x1 x2; 0 x1 x2 z1 z2 z3
# mt1 and mt1a gives the same
If this is what surprises you, there is a very simple explanation: if
there is a constant among the regressors but not among the instruments, it
is added automatically. This makes good sense to me, but in fact someone
(I can't remember whether it was Allin or myself) added the following
comment in the C source (line 386 in tsls.c, if anyone's interested):
/* found const in reglist but not instlist:
needs fixing -- or is this debatable? */
-------------------------------------------------------
Riccardo (Jack) Lucchetti
Dipartimento di Scienze Economiche e Sociali (DiSES)
Università Politecnica delle Marche
(formerly known as Università di Ancona)
r.lucchetti(a)univpm.it
http://www2.econ.univpm.it/servizi/hpp/lucchetti
-------------------------------------------------------
5:14 p.m.
The matter is as follows:
Under over-identification the
constant is added automatically,
but under just identification
it is not added. That's why
I've talked on different
behavior
Oleh
10 жовтня 2017, 19:37:02, від "Riccardo (Jack) Lucchetti"
<r.lucchetti(a)univpm.it>:
On Tue, 10 Oct 2017, > oleg_komashko(a)ukr.net wrote:
> mt1 <- tsls y 0 y1 y2 x1 x2; x1 x2 z1 z2 z3
> mt1a <- tsls y 0 y1 y2 x1 x2; 0 x1 x2 z1 z2 z3
> # mt1 and mt1a gives the same
If this is what surprises you, there is a very simple explanation: if
there is a constant among the regressors but not among the instruments, it
is added automatically. This makes good sense to me, but in fact someone
(I can't remember whether it was Allin or myself) added the following
comment in the C source (line 386 in tsls.c, if anyone's interested):
/* found const in reglist but not instlist:
needs fixing -- or is this debatable? */
-------------------------------------------------------
Riccardo (Jack) Lucchetti
Dipartimento di Scienze Economiche e Sociali (DiSES)
Università Politecnica delle Marche
(formerly known as Università di Ancona)
r.lucchetti@univpm.ithttp://www2.econ.univpm.it/servizi/hpp/lucchetti
-------------------------------------------------------
_______________________________________________
Gretl-devel mailing list
Gretl-devel@lists.wfu.eduhttp://lists.wfu.edu/mailman/listinfo/gretl-devel
8:06 p.m.
On Tue, 10 Oct 2017, oleg_komashko(a)ukr.net wrote:
The matter is as follows:
Under over-identification the
constant is added automatically,
but under just identification
it is not added.
OK, that's now fixed in git. For anyone not tuned into this thread
before, this means that (for example)
open data4-10
tsls ENROLL const INCOME ; COLLEGE
will now work, whereas before you'd have got an error message saying
that the order condition was not satisfied. In one sense the latter
is true (2 regressors but just 1 instrument) but in a substantive
sense it's false since there's no way "const" can be endogenous; we
should treat it as (its own) instrument even if it is not explicitly
listed among the instruments.
In fact we were doing this before, but as Oleh pointed out, only in
the over-identified case (an oversight).
Allin
5:54 p.m.
New subject: different beh... an attempt for clarification
Using hansl there should be
something of this kind inside:
check1 = inlist($model.instlist,const)
if !check1
$model.instlist = $model.instlist const
endif
check2 = nelem($model.instlist - $xlist) - nelem($model.endolist)
if (check2 < 0)
funcerr "order condition for identification failed"
endif
It seems the c code has
the reverse order of c
analogues of computing
check1 and check2
Oleh
10 жовтня 2017, 19:37:02, від "Riccardo (Jack) Lucchetti"
<r.lucchetti(a)univpm.it>:
On Tue, 10 Oct 2017, > oleg_komashko(a)ukr.net wrote:
> mt1 <- tsls y 0 y1 y2 x1 x2; x1 x2 z1 z2 z3
> mt1a <- tsls y 0 y1 y2 x1 x2; 0 x1 x2 z1 z2 z3
> # mt1 and mt1a gives the same
If this is what surprises you, there is a very simple explanation: if
there is a constant among the regressors but not among the instruments, it
is added automatically. This makes good sense to me, but in fact someone
(I can't remember whether it was Allin or myself) added the following
comment in the C source (line 386 in tsls.c, if anyone's interested):
/* found const in reglist but not instlist:
needs fixing -- or is this debatable? */
-------------------------------------------------------
Riccardo (Jack) Lucchetti
Dipartimento di Scienze Economiche e Sociali (DiSES)
Università Politecnica delle Marche
(formerly known as Università di Ancona)
r.lucchetti@univpm.ithttp://www2.econ.univpm.it/servizi/hpp/lucchetti
-------------------------------------------------------
_______________________________________________
Gretl-devel mailing list
Gretl-devel@lists.wfu.eduhttp://lists.wfu.edu/mailman/listinfo/gretl-devel
8:13 p.m.
New subject: different beh... an attempt for clarification
On Tue, 10 Oct 2017, oleg_komashko(a)ukr.net wrote:
Using hansl there should be
something of this kind inside:
check1 = inlist($model.instlist,const)
if !check1
$model.instlist = $model.instlist const
endif
check2 = nelem($model.instlist - $xlist) - nelem($model.endolist)
if (check2 < 0)
funcerr "order condition for identification failed"
endif
It seems the c code has
the reverse order of c
analogues of computing
check1 and check2
Thanks, Oleh; that's a clear statement of the case. We'll take a
look.
Allin
6:51 p.m.
New subject: different beh... attempt 2 to clarify
In tsls.c
MODEL tsls (const int *list, DATASET *dset, gretlopt opt)
have first
tsls.errcode = ivreg_process_lists(list, ®list, &instlist);
and only than
reglist_check_for_const(reglist, dset);
That's why under just identification
tsls gives error and do not have
a possibility to add const
Oleh
10 жовтня 2017, 19:37:02, від "Riccardo (Jack) Lucchetti"
<r.lucchetti(a)univpm.it>:
On Tue, 10 Oct 2017, > oleg_komashko(a)ukr.net wrote:
> mt1 <- tsls y 0 y1 y2 x1 x2; x1 x2 z1 z2 z3
> mt1a <- tsls y 0 y1 y2 x1 x2; 0 x1 x2 z1 z2 z3
> # mt1 and mt1a gives the same
If this is what surprises you, there is a very simple explanation: if
there is a constant among the regressors but not among the instruments, it
is added automatically. This makes good sense to me, but in fact someone
(I can't remember whether it was Allin or myself) added the following
comment in the C source (line 386 in tsls.c, if anyone's interested):
/* found const in reglist but not instlist:
needs fixing -- or is this debatable? */
-------------------------------------------------------
Riccardo (Jack) Lucchetti
Dipartimento di Scienze Economiche e Sociali (DiSES)
Università Politecnica delle Marche
(formerly known as Università di Ancona)
r.lucchetti@univpm.ithttp://www2.econ.univpm.it/servizi/hpp/lucchetti
-------------------------------------------------------
_______________________________________________
Gretl-devel mailing list
Gretl-devel@lists.wfu.eduhttp://lists.wfu.edu/mailman/listinfo/gretl-devel
2625
days inactive
2635
days old
7 comments
4 participants
participants (4)
-
Allin Cottrell -
oleg_komashko＠ukr.net -
Riccardo (Jack) Lucchetti -
Sven Schreiber