# Contributions to State-Space Models & Project Ideas for PyMC-Extras

**URL:** <https://discourse.pymc.io/t/contributions-to-state-space-models-project-ideas-for-pymc-extras/17935>\
**Category:** General\
**Tags:** modeling, contributing-to-pymc, gsoc\
**Created:** [September 13, 2026, 7:25am UTC](https://discourse.pymc.io/t/contributions-to-state-space-models-project-ideas-for-pymc-extras/17935 "2026-09-13T07:25:15Z")\
**Posts on this page:** 3\
**Page:** 1

<div class="post-metadata">

**Author:** ![Dhairya](https://yyz2.discourse-cdn.com/flex036/user_avatar/discourse.pymc.io/dhairya/32/9733_2.png) [@Dhairya](https://discourse.pymc.io/u/Dhairya)\
**Post date:** [September 13, 2026, 7:25am UTC](https://discourse.pymc.io/t/contributions-to-state-space-models-project-ideas-for-pymc-extras/17935/1 "2026-09-13T07:25:15Z")

</div>

Hi all,

I have contributed a few fixes within the PyMC projects:

- [pymc#8322] (merged): Propagating dimensions in `LatentKron` Gaussian Processes.
- [pymc-extras#763]: Mode using point estimates for post-estimation in `statespace` module.

Contributing to this would help me understand mathematical concepts behind state-space models better. While I’m still a beginner in statistics and quantitative modeling, I’d be glad to dive deeper into the subject and find some interesting GSoC project ideas concerning `statespace`.

I’d like to know from maintainers (@jessegrabowski and other PyMC maintainers), what is something you are currently looking to develop within the `statespace` models? is there anything math-related that could be an extension of the current features?

Thank you.

---

<div class="post-metadata">

**Author:** ![jessegrabowski](https://yyz2.discourse-cdn.com/flex036/user_avatar/discourse.pymc.io/jessegrabowski/32/5010_2.png) [@jessegrabowski](https://discourse.pymc.io/u/jessegrabowski)\
**Post date:** [September 14, 2026, 4:51am UTC](https://discourse.pymc.io/t/contributions-to-state-space-models-project-ideas-for-pymc-extras/17935/2 "2026-09-14T04:51:52Z")

</div>

I strongly believe you can’t understand a package like pymc by working on the code. You need to find some applied problems you’re interested in, write some models, and communicate your results to critical peers. Things to do in the code base fall out of using the tools.

---

<div class="post-metadata">

**Author:** ![Dhairya](https://yyz2.discourse-cdn.com/flex036/user_avatar/discourse.pymc.io/dhairya/32/9733_2.png) [@Dhairya](https://discourse.pymc.io/u/Dhairya)\
**Post date:** [September 14, 2026, 7:33am UTC](https://discourse.pymc.io/t/contributions-to-state-space-models-project-ideas-for-pymc-extras/17935/3 "2026-09-14T07:33:40Z")

</div>

hi Jesse,

Makes sense, indeed, fixing code internals vs understanding the pain points of model building are two entirely different things, and a power-user is probably the only way to go.

I’m really keen about finance and I love backtesting and implementing algorithmic trading strategies and research papers. Interestingly, I tried to take an applied approach recently while Ethan did his ADVI Trainer for streaming ([PR #8333](https://github.com/pymc-devs/pymc/pull/8333)). For the sake of stress-testing his out-of-core pipeline on non-stationary data, I made a [regime-switching volatility benchmark notebook](https://github.com/user-attachments/files/29807707/quant_stress_test_executed.ipynb) (2-state HMM on 100k equity-style returns). It turned out to be a good exercise that showed the limitations of mean-field ADVI on temporal structures.

Taking your advice, I would like to spend some time being a better user of PyMC and building new notebooks that utilizes `pymc-extras.statespace` in connection with trading concepts, like a Kalman filter-based pairs trading strategy or time-varying beta.

Since my goal is to thoroughly prepare for the upcoming GSoC session (specifically under State Space Models project), are there any specific papers or resources on state-space models applied to econometrics/finance that you’d recommend I read?

I’d love to spend the next few months studying the underlying math and building applied notebooks. I plan to share the models I build here on Discourse to get your critiques and feedback as I learn, so I can eventually contribute more meaningfully.