My understanding is that IRFs are always deterministic, and thus I should get exactly the same values every time I run my .mod file, no matter what seed I choose in set_dynare_seed(). However, I’m getting slightly different results each time.
Is there any way the IRFs could be non-deterministic?
That is only true at first order. At higher order, GIRFs are computed, which are based on simulations.
Great, thanks.
If I set order=1 in the stoch_simul() command, will both the IRFs and simulation be created with first-order approximation?
Yes.The order-option governs all results displayed.
Is it possible to compute 2nd-order IRFs deterministically? (without computing GIRFs)
That is only feasible if you condition on a particular path of past and future shocks.