Posterior Predictive Checks

[student84]

Hi everybody,

I've done the Bayesian estimation in dynare using the estimation-command. Everything is running and looks good. Now I want to do some model evaluation using posterior predicitve checks. I know the mathematical background and I already read the post on Bayesian model evaluation in the forum, and I know this is exactly what I need to do. But I don't know how to do it. How can I compute them? Once I have done the estimation in dynare, what are the next steps? I know I need the posterior predictive distribution. How can I get it?
I really would appreciate any help in form of examples and comments.
Thank you very much!

[jpfeifer]

What do you want to predict?

[student84]

I'm interested in checking the absolute fit of the model to the data.
This link describes very clearly what I'm searching for: http://www.bayesian-inference.com/posteriorpredictivechecks. In the end I would like to show scatter plots like in the attached pdf, figure 7 or 9.
Computation is described in the following way:
1. Simulate theta from the posterior distribution, p(theta|y)
2. Simulate y_rep from the predictive distribution, p(y_rep|theta,y)
3. Compare y to the replicated datasets y_rep.
(y is the observed data, theta the vector of parameters)
The first point of computation is clear. It's what I have done by using the estimation command in Dynare. But I don't know what to do at point two in Dynare.
Thank you very much!

[jpfeifer]

I am working on it, but it will take some time as it requires new functionality in Dynare.

[student84]

Thanks a lot! I'm also working on it. So far, I guess I need to set up a external program written in Matlab.

[jpfeifer]

It will still take some time until the functionality is in Dynare. A first draft is at https://github.com/DynareTeam/dynare/pull/871