[Gretl-users] Re: multi-step-ahead-forecasting

Thanks for your reply,maybe, that one is clearer. This is for one-step ahead forecasting. I am trying to generate 5 (and 22) -step-ahead out-of-sample forecasts using the equation (ols djclose const djclose(-1)) starting of the sample period 1+i  – 100+i. However, the equation is estimated recursively every time and forecasted every time using the new equation fitted values.  open djclose.gdtset verbose offseries frcst = NAloop i=1..200 -q         #out-of-sample (10 observations)#    smpl 1+i 100+i        #initial sample (5 observations)#    ols djclose const djclose(-1)          fcast 101+i 101+i    #one-step-ahead-rolling-windows-forecasting#    frcst[101 + i] = $fcastendloop On Thursday, January 7, 2021, 09:35:14 PM GMT, Sven Schreiber <svetosch(a)gmx.net&gt; wrote: Am 07.01.2021 um 22:16 schrieb Burak Korkusuz: Hm, still don't really understand what a rolling initial sample is; I'm asking because gretl's builtin --recursive option might be relevant. But anyway, just define your horizon, e.g.: horizon = 5 and replace your fcast lines with the adapted ones:       fcast 5+i+horizon 5+i+horizon       frcst[5+i+horizon] = $fcast Notice, however, that in a model of your type (ols LRM const LRY), the RHS variable LRY is treated as exogenous. So your "forecast" in this case does nothing else than take the observed value of LRY five steps ahead and plugs it into the estimated equation. But that comes from your model specification, not from the fcast command. hth, sven _______________________________________________ Gretl-users mailing list -- gretl-users(a)gretlml.univpm.it To unsubscribe send an email to gretl-users-leave(a)gretlml.univpm.it Website: https://gretlml.univpm.it/postorius/lists/gretl-users.gretlml.univpm.it/