Procedure of Estimating a DSGE-VAR model

Hello Everyone,

I have a question on the procedure of estimating a DSGE VAR model.

My current setup

I have a model with 73 variables.
I estimated the model with 26 observed variables and the estimation ran without any problems.

How I am thinking of estimating a DSGE VAR of this model.

  1. Estimate the model with 26 observed variables
  2. Then simulate the model
  3. Save the Simulated data to a file (it will contain 73 variables)
  4. Create a DSGE-VAR model. which is just the original model with new shocks added to match the shock count to the obs variable count (73 now)
  5. Estimate the DSGE-VAR using the simulated data as the observed data

Is this the correct way to doing this or did I get it wrong?

I wish to thank you in advance for sharing your insights.


Dear Richard,

You do not need all these steps.You do not need to turn your model into a VAR, adding all these shocks, to estimate a DSGE-VAR model. The idea here is to use a DSGE model to formulate a prior on the autoregressive matrices (and covariance matrix of the errors), not to estimate a VAR with all the variables of your DSGE model.

So you can keep the set of observed variables, just add the option doge_var in the estimation command, see the reference manual here. The only constraint is that you must have as many shocks as observed variables.Usually it is much easier to estimate a DSGE-VAR model than a DSGE model, and it is also much faster (because we do not need to run a Kalman filter to evaluate the likelihood).