Estimation with the Kalman filter

Dear all,

I have a code in DYNARE which makes Bayesian estimation of an RBC model. I set the observable variables to be composed by investments, output, hours, utilization and capital.
I add the “smoother” command in the estimation to get the estimates of all endogenous variables of the model. After running the code, I notice that the estimation for capital is quite far from the observed variable (even if this is given as observable!). I suspect that this is due to a high “observation error”. Would it be possible in DYNARE to set this observation error close to zero? And, if yes, how?

Thank you.


Hi Alice,
Dynare only adds estimation error, if you explicitly specified it. If you want to fix this estimation error at a certain (maximum) level, you have three possibilities:

  1. In the estimated_params - block set the upper bound of the prior for the measurement error’s standard deviation to the desired maximum value.
  2. Omitt the measurement error from estimated_params and fix it at the desired value by putting this value in the shocks-block.
  3. If you have more shocks than observables, you can completely drop the measurement error without having stochastic singularity.

However, remember that the problem you describe may be an indication of model misspecification or a not yet converged Bayesian/ML-estimation.