Am 31.01.2023 um 12:10 schrieb Sven Schreiber:
"just" about simulation. It would be interesting to see whether the
dynamic accounting relationships could be specified in gretl's system
block as (dynamic) identities. It's not clear to me whether that is
supported by gretl. (Also noting that the doc in ch. 34 says that
identities only matter for FIML estimation, so another question is
about their role in forecasting.)
OK, here's a partial answer with respect to gretl's handling of dynamic
identities in forecasting from a system: It seems to be honored alright.
A trivial and purely artificial but concrete example:
<hansl>
nulldata 50
setobs 1 1 --time-series # pseudo-annual
smpl 1 40 # hold back the last 10 obs
series capital = normal()
series output = capital + 0.1*normal() # everything artificial
series dcap = diff(capital)
# now some pseudo-exogenous future dcap values
smpl 41 50
dcap = normal()
# back to the estimation sample
smpl 1 40
system method=ols
equation output const capital
identity capital = dcap + capital(-1)
end system
# perform the forecast
fcast output capital --out-of-sample
# save the forecast values in the o-o-s range
smpl 41 50
series outf = $fcast[,1]
series capf = $fcast[,2]
# check whether the identity holds in the o-o-s range
series dev = capf - dcap - capf(-1)
eval max(abs(dev)) # should be and is indeed zero
</hansl>
Again, this is just a partial exploration and a partial answer to the
questions that have been asked.
cheers
sven
_______________________________________________
Gretl-users mailing list -- gretl-users@gretlml.univpm.it
To unsubscribe send an email to gretl-users-le...@gretlml.univpm.it
Website:
https://gretlml.univpm.it/postorius/lists/gretl-users.gretlml.univpm.it/
