Re: [Gretl-devel] ols fcast omit problem
On Tue, 27 Sep 2011, Paweł Kufel wrote:
I have noticed a problem, with fcast command, which appears in such
script:
<script>
open data2-3.gdt
smpl 1959 1990
model1 <- ols unemp const cpi infl wggr
fcast 1991 1995
omit --auto
fcast 1991 1995
</script>
Both fcast commands give equal forecast results (corresponding to
model1), during the script without saving model to icon works ok:
<script>
open data2-3.gdt
smpl 1959 1990
ols unemp const cpi infl wggr
fcast 1991 1995
omit --auto
fcast 1991 1995
</scritp>
This script gives different forecast results. First fcast is based on
model given by ols, second one is based on model after omitting
variables.
Since the --auto form of "omit" always results in replacement
of the original model (provided something is in fact omitted),
the correct result is that the second "fcast" differs from the
first.
But I find that this is what happens. I've tried your script
in both gretlcli and the gretl GUI, and even in the GUI with
the first form of the script (where "model1" is saved) I'm
getting different forecasts before and after the "omit --auto"
command. For example, the prediction for unemp in 1991 is 7.3
from the first model and 7.6 for the second, after omission
of infl and wggr.
The language in which gretl is running obviously shouldn't
make any difference, but I tried in Polish too and saw the
same thing.
So I'm puzzled.
Allin