Thank you Allin for your reply. As I understand, it's not possible to force Gretl to keep the initial range of dates once you remove some dates within this range (?). Anyway, there is clearly a problem of continguity for HAC estimator.

I stll have a problem with the sub-sampling function. When I restrict a variable to positive values and run OLS, it turns out that when I do this OLS estimation a second time, the sample shrinks and the estimates change accordingly. I don't understand why the restriction does not operate once for all ?

Thnaks for your help