12-08, 17:30–18:00 (UTC), Data Track
Unlock robust statistical inference for time series data with tsbootstrap, a new open source Python library implementing specialized bootstrapping techniques.
Proper analysis of time series requires quantifying uncertainty. Bootstrapping enables confidence intervals, distribution-free inference, and model validation. However, classical bootstrapping relies on independent identically distributed (IID) data and fails for time series where observations have dependencies.
This talk will explain the need for specialized time series bootstrapping and introduce tsbootstrap, an open source library implementing essential techniques like block, Markov, and sieve bootstrapping. We'll walk through code examples applying these methods to sample financial data.
The target audience is Python practitioners looking to improve their time series modeling and forecasting workflows. Key takeaways include understanding why naive bootstrapping fails, getting an overview of key techniques, seeing hands-on examples using tsbootstrap, and learning how to incorporate robust resampling into analysis.
The structure will explain the limitations of classical bootstrapping, introduce the tsbootstrap library, demonstrate techniques on sample data, and summarize the core learnings so attendees are equipped to enhance their own time series analyses.
No previous knowledge expected
Sankalp Gilda is an MLE by trade and an astronomer by training, obsessed with uncertainty quantification, and causality, and open source software. When not building production ML pipelines or working on statistical modeling research, you'll find him jumping from planes, boats, or cliffs -- seeking the thrill of the outdoors.