I have two questions regarding the DSGEVAR procedure (specified with the option dsge_var in the estimation command).

Where exactly does dynare store the prior (for the BVAR) generated by the DSGE model? The Bayesian IRFs for the DSGEVAR that are generated by dynare rely on identification based on the DSGE model (which of course makes perfect sense). I would like to compare these IRFs to IRFS from an identification scheme based on sign-restrictions, therefore I need the prior to estimate the reduced form BVAR myself and then apply another rotation (one based on sign restrictions).

Does anyone know whether it is possible to use the prior generated let’s say for 6 variables for a 12 variables BVAR, for instance through combining it with a Minnesota prior for the 6 variables that did not show up in the DSGE model? The problem is that the DSGE model I am using does not explicitly model some of the variables I would like to include in my BVAR.

with regard to my first question, does anyone know who implemented the option ‘dsge_var’ for estimation(.) ? Was it Stéphane Adjemian?

I have looked quite a bit, but it seems as if the DSGE-based prior for the BVAR is not explicitly stored anywhere… maybe it can be obtained from one of the .m files? Given that dynare produces IRFs for the DSGEVAR, the DSGE-based prior for the BVAR has to be somewhere, right?.

Dear all, let me specify my question and ask in a different way.

I think one can construct the priors once one extracted certain auto covariance matrices. I have typed up a document based on the Dynare wiki.

What I am looking for in the dynare output are the matrices GAMMA_ZZ, GAMMA_ZY, GAMMA_YZ, GAMMA_YY, the auto-covariance matrices of the DSGE model’s VAR approximation. Does anyone know whether dynare stores these somehwhere, or whether dynare stores parts that can be used to construct these autocovariance matrices?

No, currently we do not save these. The DSGE-VAR is a functionality that requires some work.

We will try to look at this in detail, but it will take some time. All the matrices should be there, it is mostly a matter of saving them (which we currently do not do)

If the DSGE model is silent about the priors for other variables, you can of course combine this prior with a prior of your own for the other variables. The questions then are only whether you get a tractable posterior and whether you can defend your prior.

Yes, the implementation is due to Stéphane Adjemian

I found the function ‘dsge_var_likelihood.m’ which basically shows what is done, it even generates the prior as an output. So far, I have not climbed down that deep into the dynare machinery.

Do you have any references, links, example etc that show how certain parts of the .m files underlying dynare can be executed? Is there any decent documentation on that?