The potential issue with large shocks?

[jpfeifer]

That’s not a good idea. See point 1 at:

Shock to standard deviation

Mostly yes. But You are using a log-log-specification, whose moments do not exist. See Andreasen (2010) I would recommend against GIRFs. See https://github.com/JohannesPfeifer/DSGE_mod/blob/master/Basu_Bundick_2017/Basu_Bundick_2017.mod for an alternative.