[Gretl-users] Re: (ARIMA forecasts, was no subject)
On Sat, 23 Nov 2024, Brian Revell wrote:
Is there any reason why recursive forecasts cannot be generated and
included in the ARIMA option for univariate modelling forecasts (together
with their confidence intervals ) when no exogenous variables have been
included in the specification Clearly post sample data the MA terms would
drop out.
Brian, do you mean "recursive" in the specific sense used in the
doeumentation for gretl's "fcast" command -- with re-estimation of the
model specification on a progressively expanding sample at each time
step?
I ask because with ARIMA (without stochastic exogenous terms) a
"dynamic" forecast can proceed out of sample, and is recursive in the
more general sense of employing the chain rule of forecasting. As in,
for example,
open data9-7.gdt
dataset addobs 10
arima 1 0 1 ; QNC
fcast 1991:1 1993:2 --plot=display
Allin