Encyclopedia · simulation · statistics
Block bootstrap
Resampling contiguous blocks of returns to preserve autocorrelation and volatility clustering.
The naive bootstrap resamples individual returns independently, destroying the very features that make financial series dangerous: volatility clustering, streaks, serial dependence. The block bootstrap (Künsch, 1989; stationary variant by Politis & Romano, 1994) resamples contiguous blocks instead, so simulated paths inherit realistic runs of good and bad days.
In portfolio simulation the block is drawn across all strategies at once — the same dates move together — which preserves the cross-strategy correlation structure, including its ugly episodes. That is what makes block-bootstrap Monte Carlo a portfolio-level stress test rather than a per-strategy toy.
Related entries