On Mon, 11 Oct 2021, Sven Schreiber wrote: > the following happened with 2021d: I open denmark.gdt and estimate a toy > model in the GUI. Then I go to Analysis/Forecasts in the model window. > Gretl complains about unavailable out-of-sample obs. I dismiss this > message window and I switch to the main window and add 5 obs. Switching > back to the still open model window I attempt forecasting again. -> But > I get the same complaint, which now is no longer true. What's the model specification? Does it include any variables other than the dependent?

If "addobs" is going to help, you should get a dialog offering to add some observations when you access the forecast menu item.

Am 11.10.2021 um 16:02 schrieb Allin Cottrell: > On Mon, 11 Oct 2021, Sven Schreiber wrote: >> the following happened with 2021d: I open denmark.gdt and estimate a toy >> model in the GUI. Then I go to Analysis/Forecasts in the model window. >> Gretl complains about unavailable out-of-sample obs. I dismiss this >> message window and I switch to the main window and add 5 obs. Switching >> back to the still open model window I attempt forecasting again. -> But >> I get the same complaint, which now is no longer true. > What's the model specification? Does it include any variables other > than the dependent? Without any deeper meaning and just by coincidence, I was regressing LRY from the denmark dataset on LRM(-3) -- not LRM(-1 to -3). I re-checked that when I add just two new obs until 1988:1 or so, gretl still says no obs available for forecasting. LRM_3 should be available until 1988:2 automatically.

Am 14.10.2021 um 16:29 schrieb Allin Cottrell: > On Mon, 11 Oct 2021, Sven Schreiber wrote: > > > Without any deeper meaning and just by coincidence, I was regressing LRY > > from the denmark dataset on LRM(-3) -- not LRM(-1 to -3). > > > > I re-checked that when I add just two new obs until 1988:1 or so, gretl > > still says no obs available for forecasting. LRM_3 should be available > > until 1988:2 automatically. > > This script shows what's happening. LRM_3 is not automatically > available in the added observations. To get that to happen, gretl > would have to recognize, on the basis of its metadata, that LRM_3 is a > series that can be extended after "addobs". That should be possible > but it's not in place at present. Right, thanks, Allin. When I wrote "should be available" I meant that conceptually there is no problem. I would call a bug only the misleading error message, the rest is a missing feature, I think. However, the more I think about it, I believe that potentially quite many models in predictive regression format would be affected. So maybe this is actually a bit more severe.

Am 14.10.2021 um 18:14 schrieb Allin Cottrell: > > That's now upgraded in git: given a time-series dataset, recognized > lag series are extended by "dataset addobs". That's not yet the case > for panel data. > Wow, that was quick! I expected it to be more involved and complex to change that handling... Will test next week at the latest.