4:54 a.m.
Hi,
apparently, the ols command does not accept expressions for rhs variables, e.g.
ols a 0 b+c
Am I missing something and if not, are there any plans to change this?
Many thanks
Fred
5:28 a.m.
On Thu, 23 Jul 2020, fmorkel(a)yahoo.de wrote:
Hi,
apparently, the ols command does not accept expressions for rhs variables, e.g.
ols a 0 b+c
Am I missing something
No, you're not, although _some_ expressions are allowed. Example:
<hansl>
open AWM18.gdt
ols log(PCR) 0 log(YER)
</hansl>
and if not, are there any plans to change this?
Not as of now. But I guess this is something that we may want to revisit
for version 2020d (2020c is going to be released shortly).
-------------------------------------------------------
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
-------------------------------------------------------
5:56 a.m.
Am 23.07.2020 um 11:28 schrieb Riccardo (Jack) Lucchetti:
On Thu, 23 Jul 2020, fmorkel(a)yahoo.de wrote:
> Hi,
>
> apparently, the ols command does not accept expressions for rhs
> variables, e.g.
> ols a 0 b+c
> Am I missing something
No, you're not, although _some_ expressions are allowed. Example:
<hansl>
open AWM18.gdt
ols log(PCR) 0 log(YER)
</hansl>
> and if not, are there any plans to change this?
Not as of now. But I guess this is something that we may want to revisit
for version 2020d (2020c is going to be released shortly).
Well, isn't the problem that when you create such expressions
on-the-fly, then there doesn't exist any name or label for that regressor?
log() is obviously treated as an exception and the prefix "l_" is added
automatically, presumably because it's so common (and easy), but what's
the name that you would assign in general for printout and so forth?
I'm a big fan of conciseness, but even I would say it's not a big
problem to add the line:
series bplusc = b + c
before the ols command.
cheers
sven
7:45 a.m.
With a small set of few regressions, you're right. But if you run several hundred
regressions with several regressors, introducing temporary variables is not very
convenient and increases the data set considerably. The names for the expressions could be
generated by the machine somehow!?
Not found in version 2020d but I I still hope that this feature makes it into the
program.
Thanks
Fred
7:58 a.m.
On Fri, 4 Sep 2020, Fred Morkel wrote:
With a small set of few regressions, you're right. But if you run
several hundred regressions with several regressors, introducing
temporary variables is not very convenient and increases the data set
considerably. The names for the expressions could be generated by the
machine somehow!?
Could you give us a practical example when this feature would be needed?
Maybe someone could come up with an alternative solution.
-------------------------------------------------------
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
-------------------------------------------------------
9:05 a.m.
One example: In our model, we estimate energy consumption by sectors with real production
by sectors and the relative prices by sectors which is calculated as energy prices by
sectors divided by output prices by sectors.
For this example, we then have 20 regressions due to 20 sectors and thus need to introduce
20 temporary variables which depict the aforementioned relative prices by sectors.
As we have many more estimations like this, partly for many more than 20 sectors, you
might get the picture. Most of the time, the temporary variables would not be used
further.
Does this explain the problem better?
Thanks
Fred
9:40 a.m.
On Fri, 4 Sep 2020, Fred Morkel wrote:
One example: In our model, we estimate energy consumption by sectors
with real production by sectors and the relative prices by sectors which is calculated as
energy prices by sectors divided by output prices by sectors.
For this example, we then have 20 regressions due to 20 sectors and thus need to
introduce 20 temporary variables which depict the aforementioned relative prices by
sectors.
As we have many more estimations like this, partly for many more than 20 sectors, you
might get the picture. Most of the time, the temporary variables would not be used
further.
Does this explain the problem better?
Yes, thanks, this is much clearer. But in a case such as yours, I would
argue that a judicious use of loops would be advisable anyway. If
temnporary variables are a problem, deleting them after use within the
loop itself is rather simple. Better still, you could use a function and
the temporary variables would automagically disappear after use. For
example,
<hansl>
function void do_weird_regression(series y, series x)
series tmp1 = log(x)
series tmp2 = cos(x)
series tmp3 = x/exp(x)
ols y const tmp1 tmp2 tmp3
end function
nulldata 100
# generate example dataset
loop foreach i FOO BAR BAZ
series y_$i = normal()
series x_$i = uniform()
endloop
# example
do_weird_regression(y_FOO, x_FOO)
# better
loop foreach i FOO BAR BAZ
do_weird_regression(y_$i, x_$i)
endloop
</hansl>
-------------------------------------------------------
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
-------------------------------------------------------
10:09 a.m.
If I adopt my problem to your example, I would need something like:
ols y_FOO const log(x_FOO)
ols y_BAR x_FOO/x_BAR
ols y_BAZ x_BAZ x_BAR x_FOO
If I'm not mistaken, I can't introduce a loop or use a function here.
BTW: Many thanks for your support!
Fred
10:28 a.m.
On Fri, 4 Sep 2020, Fred Morkel wrote:
If I adopt my problem to your example, I would need something like:
ols y_FOO const log(x_FOO)
ols y_BAR x_FOO/x_BAR
ols y_BAZ x_BAZ x_BAR x_FOO
If I'm not mistaken, I can't introduce a loop or use a function here.
Something like this should do it:
<hansl>
function void do_regressions (list LY, list LX)
loop foreach i LY
if i == 1
series lx = log(LX[1])
ols $i const lx
elif i == 2
series x_x = LX[1] / LX[2]
ols $i x_x
else
ols $i LX
endif
endloop
end function
list LY = y_FOO y_BAR y_BAZ
list LX = x_FOO x_BAR x_BAZ
do_regressions(LY, LX)
</hansl>
Allin
1493
days inactive
1536
days old
9 comments
5 participants
participants (5)
-
Allin Cottrell -
fmorkel＠yahoo.de -
Fred Morkel
-
Riccardo (Jack) Lucchetti -
Sven Schreiber