Bayesian Estimation and Forecasting of Time Series in statsmodels, for Scipy 2022 conference
Statsmodels
, a Python library for statistical and econometric analysis,
has traditionally focused on frequentist inference, including in its
models for time series data. This paper and Poster illustrates the powerful
features for Bayesian inference of time series models that exist in
statsmodels
, with applications to model fitting, forecasting, time series
decomposition, data simulation, and impulse response functions.
Suggestions for reading and using this Poster
Our suggestion for using this Poster is as follows.
Poster.ipynb
, and it can be (a)More details about each of these follow:
1. Paper
Included in this repository is a draft of the paper
Bayesian Estimation and Forecasting of Time Series in Statsmodels.pdf
that is
forthcoming in the Proceedings of the 21st Python in Science Conference (SciPy
2022). This paper introduces the time series models included in Statsmodels and
shows how to estimate their parameters using Bayesian methods. It also briefly
describes the relationship to other popular Python libraries for Bayesian
inference, including PyMC, PyStan, and ArviZ.
2. Poster
The paper also provides code samples for several applications that include
parameter estimation, forecasting, and causal inference. To keep the paper
readable, the code included inline is only brief, but the Poster included in
this repostory, Poster.ipynb
, is a Jupyter notebook that contains the complete
code for performing the Bayesian analysis in the paper.
3. Additional resources for Bayesian analysis of time series models in Statsmodels
SARIMAX