This talk will introduce the PyFlux library for time series analysis in Python. I will walk through the modelling and inference options in an accessible, high-level way. I will focus on score-driven (GAS) models, which are a new flexible alternative to traditional time series models. This Python library represents the first comprehensive implementation of a GAS library, a model type that has the potential to be as widespread as ARIMA models. I will demonstrate the usefulness of the models through some fun examples in PyFlux, such as a power rating model for NFL football teams, as well as some finance examples, which I shall attempt to make fun (challenge accepted).

The preliminary outline of the talk will be as follows:

• Brief introduction to the PyFlux library
• Introduction to GAS models for time series
• GAS models as an extension of ARIMA models (with examples)
• GAS models as an extension of state space models (with examples)
• Inference options for GAS models (with examples)
• A simple GAS model for predicting NFL games
• Conclusion

After this talk, you will be able to:

• Understand GAS models, and their benefits versus traditional time series models.
• Estimate and run predictions with GAS models, using PyFlux
• Understand the different types of inference that can be used for time series problems, including variational inference.
• Perform multiple methods of inference on time series problems using PyFlux