DSGE-VAR with less observables than shocks

Dear all,

I have estimated a DSGE model with 7 shocks and 2 additional news shocks. I would like to estimate a DSGE-VAR to compare the IRFs of these two news shocks with the original DSGE IRFs.
I cannot estimate the DSGE-VAR with fewer observables than shocks but I wouldn’t like to add additional observables to the model. I thought that I could take the smoothed series of the news shocks from the first DSGE estimation and add them as observables for the DSGE-VAR estimation. Does this invalidate the comparison between DSGE-VAR and DSGE IRFs as an exercise of model performance?

Thank you very much in advance!
Best regards

Yes, that would be a problem.

  1. Smoothed objects use the full dataset, while the VAR is backward-looking. As such, the information sets would not match.
  2. Your DSGE-VAR would be conditional on the results of the DSGE model for the news part. That does not strike me as a valid comparison.
1 Like

Thanks a lot for your quick response!

Then, it is just not possible to recover the IRFs for the news shocks using the DSGE-VAR unless I include additional observables, right?

A VAR assumes that there are as many observables as shocks. You cannot have more shocks than observables in this case.

1 Like