[Gretl-users] Re: Out of sample forecasting
Am 13.03.2021 um 12:30 schrieb Periklis Gogas:
Dear Sven,
It was my fault!
I created the logged first differences after restricting the full
sample, thus when I tried to forecast there was no ld_y out-of-sample
available!
My bad.
Hm, actually I would have expected that lagged endogenous terms are not
a problem and gretl would know what to do automatically.
But indeed the following does not work (error about missing values):
<hansl>
open denmark
series y = log(LRM)
series dy = ldiff(LRM)
ols y const dy(-1)
dataset addobs 4
fcast y --out-of-sample
</hansl>
I guess this isn't a bug, strictly speaking, since section 35.3 in the
user guide says: " ... does not contain any stochastic regressors
other than lags of y", and in this case it's not just a lag, but also a
difference involved.
So in order to get a multistep forecast to work here one would have to
run a restricted ols estimation in (log) levels or perhaps formulate
this as a nls setup. (Haven't tested either.)
cheers
sven