Different results using higer order approximation

Dear friends,

I have encountered a weird problem that the IRFs under first-order is different from those under third-order approximation.

The response variables and exogenous shocks used are the same.
i.e. employment responses to first moment technology shock. The IRF behaves even in the opposite direction. By the way, I also use pruning at third-order.

Would any friend like to help me with this problem?
Thanks in advance!

Have you tried what happens if you decrease the shock variances? In that case, the IRFs should be close to first order case.

Hi, professor Pfeifer, thanks for your reply.
Yes, I have tried it but still failed.
May be there is some thing wrong with my model.

Did model_diagnostics return anything?

Hi, professor Pfeifer!

Sorry, it returns nothing. I should think it over and check it again.

Thanks for your help and reply!

Please provide me with the file.

Hi, professor!

Sorry! I was going to send you my file. But I may fix the problem after setting shook variances below 0.01. I was in a hurry and did not quitely understand your advice at first. Sorry for that! But I have seen others’ code that set shook variances to 1 and it works normally.

Does it mean higher order approximation make IRFs more unstable than first-order approximation? Thanks a lot!

Dear friends,

I have a problem about how to read or save the simulated moments when using third-order approximation?
Thanks a lot!

Regarding your first point, see

of See “Remark 10 (Scaling With a Factor 100)” in Pfeifer (2013): “A Guide to Specifying Observation Equations for the Estimation of DSGE Models”. When you use 1 in a nonlinear model for the shock variance, you are actually using 100%.
Regarding simulated data at higher order: they should be saved the same way as at order=1 if you specify the



Hi! Prof. Pfeifer, thanks for your help !!