tsbootstrap: Distribution-Free Uncertainty Quantification and Conformal Prediction for Time Series

2026-07-07 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Software Development & Engineering · Depth: Advanced, quick

Summary

tsbootstrap is a new MIT-licensed (v0.6.1) software library designed for distribution-free uncertainty quantification and conformal prediction in time series data. It addresses the common issue where finance, sensing, and demand streams violate the exchangeability assumptions of IID conformal prediction and bootstrap methods. The library offers a single typed API integrating block, residual, sieve, and wild resampling techniques, alongside classical bootstrap confidence intervals and adaptive conformal calibrators like EnbPI, ACI, NexCP, and AgACI. A controlled study showed IID bootstrap undercovers sharply under dependence, while tsbootstrap's methods reduce this deficit, with sieve resampling performing nearest to nominal under short-memory linear dependence. Its compiled backend runs several times faster than arch, and a streaming reduce limits peak extra memory to $O(B)$.

Key takeaway

For data scientists building predictive models or quantifying uncertainty in time series, you should consider adopting tsbootstrap. Its specialized resampling and conformal prediction methods directly address the dependence inherent in finance, sensing, and demand data, preventing the severe undercoverage issues common with IID approaches. Integrating this library can significantly improve the reliability of your confidence intervals and prediction sets, ensuring more accurate and trustworthy forecasts. Explore its API to leverage its performance benefits and advanced calibration techniques.

Key insights

tsbootstrap unifies advanced resampling and conformal prediction to provide robust, distribution-free uncertainty quantification for time series.

Principles

• Time series data frequently violates IID exchangeability assumptions.
• Dependence-aware resampling methods improve prediction coverage accuracy.
• Sieve resampling is effective for short-memory linear dependence.

Method

The library provides a single typed API to select various block, residual, sieve, and wild resampling methods, along with adaptive conformal calibrators (EnbPI, ACI, NexCP, AgACI) via a specification object.

In practice

• Apply tsbootstrap to financial, sensing, or demand stream analysis.
• Utilize dependence-aware methods to correct IID bootstrap undercoverage.
• Benefit from the compiled backend for faster computation.

Topics

• Time Series Analysis
• Conformal Prediction
• Uncertainty Quantification
• Bootstrap Resampling
• Machine Learning Libraries
• Financial Modeling

Best for: AI Engineer, Research Scientist, Machine Learning Engineer, Data Scientist, AI Scientist

Related on AIssential

• Cheap Bootstrap for Fast Uncertainty Quantification of Stochastic Gradient Descent — New "cheap bootstrap" methods enable faster, more efficient uncertainty quantification for Stochastic Gradient Descent solutions.
• Spectral Adaptive Conformal Prediction for Structured Non-Exchangeable Data — Spectral Adaptive Conformal Prediction uses local frequency content and online adaptation to calibrate prediction intervals for non-exchangeable time series.
• On Stability and Decomposition of Sample Quantiles under Heavy-Tailed Distributions — This paper addresses the instability of sample quantiles, particularly Value-at-Risk (VaR), when estimated parameters like projection direction and quantile threshold are derived from heavy-tailed financial returns.
• Audited Conformal Prediction for Classification under Unknown Distribution Shift — Audited Conformal Prediction improves uncertainty quantification under distribution shift using an auxiliary audit model.
• Probabilistic Signature Inversion: Learning Conditional Distributions from Truncated Signatures — Truncated signature inversion is reframed as learning a path's conditional distribution using flow matching, quantifying irreducible uncertainty.
• Data augmented bootstrap: Unifying confidence interval construction by approximate invariance — Data Augmented Bootstrap (DAB) unifies confidence interval construction by leveraging approximate data invariances, extending classical and modern methods.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.