Sankalp Gilda - IID Got You Down? Resample Time Series Like A Pro | PyData Global 2023

Traditional bootstrapping methods fail with time series data due to autocorrelation and temporal dependencies between data points

TS Bootstrap library provides multiple bootstrapping techniques specifically designed for time series:

• Block bootstrapping
• Moving block bootstrapping
• Circular block bootstrapping
• Residual bootstrapping
• Markov bootstrapping
• Sieve bootstrapping

Block bootstrapping maintains chronological integrity by resampling blocks of consecutive observations rather than individual points

Residual bootstrapping fits a model to obtain residuals, resamples them, and adds them back to fitted values - better for handling changing distributions over time

Block length selection is crucial:

• Can be fixed or drawn from distributions
• Affects bias-variance tradeoff
• Different weighting schemes possible within blocks

The library offers:

• Easy pip installation
• Config-based usage pattern
• Integration with common statistical models (ARIMA, SARIMA)
• Enterprise-grade code with testing
• Modular architecture
• Extensive documentation

Library implementation focuses on preserving:

• Time dependencies
• Marginal distributions
• Short-run dynamics
• Autocorrelation structure

Not recommended for data with:

• Strong seasonality
• Significant trends
• No cyclical components
• No significant dependencies

Planned integration with SKTime for improved uncertainty quantification capabilities

Supports both univariate and multivariate time series analysis