Thanks for that. To be clear, what I would like to do is save the random selected observations and estimate a regression equation. This is my training sample. The excluded observations would be my test sample for forecasts - how could I save them as the test set? Can it be done with the GUI or perhaps it needs a script? Thanks.

Again, CvDataSplitter by Artur sounds like it could be what you want.

Apart from that, with gretl's GUI you could do the following, although it's maybe a bit tedious:

(the labels may not be exactly that, I'm too lazy to restart gretl in English, so I'm re-translating on the fly)

1. Draw the random sample (menu Sample/ Random sample)

2. Create a sample range dummy (menu Add / Sample range dummy); say you call that sd1

3. Restore the full sample (menu Sample / restore full range)

4. Apply the sample which complements the previous one: menu Sample / restrict by condition, and in the condition field type in the formula: "missing(sd1)" (without the quotes).

5. Create another sample range dummy for this second sample range: like in step 2 (say, call it non_sd1)

You can now go back and forth between those samples if you wish. If the missing values in the dummies bother you and you prefer to have zeros there, you can use the misszero() function (but not directly in the GUI, AFAIK).

cheers

sven