# Poisson processes for many related univariate time series

**URL:** <https://discourse.julialang.org/t/poisson-processes-for-many-related-univariate-time-series/45211>\
**Category:** Probabilistic Programming\
**Tags:** time-series, turing\
**Created:** [August 19, 2020, 2:34pm UTC](https://discourse.julialang.org/t/poisson-processes-for-many-related-univariate-time-series/45211 "2020-08-19T14:34:01Z")\
**Posts on this page:** 5\
**Page:** 1

<div class="post-metadata">

**Author:** ![mystis](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/mystis/32/9327_2.png) [@mystis](https://discourse.julialang.org/u/mystis)\
**Post date:** [August 19, 2020, 2:34pm UTC](https://discourse.julialang.org/t/poisson-processes-for-many-related-univariate-time-series/45211/1 "2020-08-19T14:34:01Z")

</div>

Hi everyone, I’ve been wanting to do some time series forecasting with Turing, because I’d like forecasting intervals as well.

For practice, I’d like to forecast the [norway car sales dataset](https://www.kaggle.com/dmi3kno/newcarsalesnorway). In short, it has ten years of monthly car sales data, from 2007 to 2017.

- Indices (rows) are a monthly observations of car sales.

- Columns are car brands.

I would like to model this as time series of counts, as well as including the fact that some sub-groups are probably linked together (depending on the economic conjoncture).

I’m not sure at all how I would go about doing this in Turing though, even something simple with one lag for example.

Can you guys guide me ?

Thanks

---

<div class="post-metadata">

**Author:** ![cpfiffer](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/cpfiffer/32/208747_2.png) [@cpfiffer](https://discourse.julialang.org/u/cpfiffer)\
**Post date:** [August 19, 2020, 3:07pm UTC](https://discourse.julialang.org/t/poisson-processes-for-many-related-univariate-time-series/45211/2 "2020-08-19T15:07:34Z")

</div>

Do you have a model in mind? It’s a little tough to help without some kind of model framework.

---

<div class="post-metadata">

**Author:** ![mystis](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/mystis/32/9327_2.png) [@mystis](https://discourse.julialang.org/u/mystis)\
**Post date:** [August 19, 2020, 3:20pm UTC](https://discourse.julialang.org/t/poisson-processes-for-many-related-univariate-time-series/45211/3 "2020-08-19T15:20:59Z")

</div>

For now, something simple like an GLM with a time component as in [this pymc3 example](https://docs.pymc.io/notebooks/GLM-rolling-regression.html). Ideally I’d like to try GLARMA with a poisson distribution as in this [R package](https://cran.r-project.org/web/packages/glarma/vignettes/glarma.pdf), but that’s a bit too much for now.

---

<div class="post-metadata">

**Author:** ![cpfiffer](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/cpfiffer/32/208747_2.png) [@cpfiffer](https://discourse.julialang.org/u/cpfiffer)\
**Post date:** [August 20, 2020, 1:57pm UTC](https://discourse.julialang.org/t/poisson-processes-for-many-related-univariate-time-series/45211/4 "2020-08-20T13:57:45Z")

</div>

Tight, we’ve actually got a tutorial for GLM [here](https://turing.ml/dev/tutorials/5-linearregression/). If you want to see a different link distribution (logistic) you can check out our logistic regression tutorial [here](https://turing.ml/dev/tutorials/2-logisticregression/).

---

<div class="post-metadata">

**Author:** ![mystis](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/mystis/32/9327_2.png) [@mystis](https://discourse.julialang.org/u/mystis)\
**Post date:** [August 20, 2020, 4:19pm UTC](https://discourse.julialang.org/t/poisson-processes-for-many-related-univariate-time-series/45211/5 "2020-08-20T16:19:21Z")

</div>

Thanks i appreciate it