stoch_simul after estimation vs stoch_simul using posterior

[superlily]

I have a mod file where I estimate via Bayesian and then run stoch_simul to get the theoretical variances. But I find that if I write a new mod file using the posterior mean of parameters and then run stoch_simul to get the theoretical variances, the variances are different from those above. Why are they different?

The Dynare version is 4.2.4

[jpfeifer]

That should not be the case. Does the issue persist in Dynare 4.2.5? If yes, could you post the mod-file and the data-file.

[ilobayesian]

I am experiencing a similar issue: if i run stoch_simul after estimation, I get my output, but if I impose the parameter values I just estimated in a separate file and I use stoch_simul, Dynare tells me that B-K conditions are not satisfied. It seems that does not depend on the version of Dynare I am using. How is that possible?

[jpfeifer]

Every parameter vector must satisfy the Blanchard-Kahn-conditions. However, the parameters after Bayesian estimation are not the mean parameter vector, but the vector of parameter means. This might explain the difference. You can try to get the mode by using

Code: Select all

xparam = get_posterior_parameters('mode')


after completing estimation.

[ilobayesian]

Sorry, I do not think I am getting it. The vector of parameter means, i.e. the estimation result, should satisfy the Blanchard-Kahn-conditions. Otherwise how could stoch_simul (runned after the estimation) be able to solve the model at the posterior mean?

Thank you for the clarification!

[jpfeifer]

For every parameter vector the BK conditions are satisfied in estimation. But the estimation result is none of these parameters. It is not the median, but the mean. The mean is a convex combination of the estimated parameters. If the stability region is a convex set of the parameters, this particular convex-combination called mean will be inside the stability region. However, in some cases the stability region is not a convex set and the mean parameter vector is outside of it. In these cases, stoch_simul after estimation will crash because the BK-conditions are violated.