VAR(2,0) NumPyro vs PyMC-Extras benchmark example

[juanitorduz]

In this PR, I want to collect feedback on some initial benchmark of a VAR(2,0) model. I have more experience with NumPyro for time-series models, but to learn more about the state-space module in PyMC-Extras, I decided to try an analogous implementation. I still have some questions which I will add on the PR.

Original blogpost (for myself: https://juanitorduz.github.io/var_numpyro/

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[review-notebook-app]

View / edit / reply to this conversation on ReviewNB

juanitorduz commented on 2025-12-30T19:49:44Z
----------------------------------------------------------------

I am following the docs: https://www.pymc.io/projects/extras/en/latest/statespace/generated/pymc_extras.statespace.models.BayesianVARMAX.html#pymc_extras.statespace.models.BayesianVARMAX

I tried with stationary_initialization=True,but got an issue with the model initialization.

I would love to get some feedback on the model configuration and priors so that they match the NumPyro model above :)

[review-notebook-app]

View / edit / reply to this conversation on ReviewNB

juanitorduz commented on 2025-12-30T19:49:45Z
----------------------------------------------------------------

It seems the pymc-extras model output the same posterior for the Phi parameters. So the question is about the priors and how to make it comparable with the NumPyro model :)