---
id: S-6
kind: skill
version: 3
title: "Bootstrap and permutation methods"
series_group: "4.1 Mathematics and statistics"
prerequisites: [S-1, S-2]
unlocks_tools: []
now_asof: "2026-10-16"
next_horizons: [2027, 2029, 2031]
authorKind: grok
skillHash: "b306c7472a08502c8e6048c2ada771ccda2d8c55f61dec5b7539ae886659d655"
review: approved
tier: public
withheld_sections: ["Worked example", "How to do it here", "Common errors", "Prerequisites and what it unlocks", "Exam facts", "What would make this lesson wrong"]
withheld_keys: [embodied_by, facts, refuse, route, sources]
exam_facts: 7
---
# S-6 Bootstrap and permutation methods

## 1. Definition

A bootstrap rebuilds the uncertainty by resampling the sample, when a formula would require a shape you do not believe. An ordinary bootstrap draws observations independently. That is wrong for a return series, because tomorrow resembles today: volatility clusters, and a draw that scatters the days destroys the clusters and makes the strategy look safer than it was. A block bootstrap draws contiguous stretches and concatenates them, so the short-range dependence survives. A stationary block bootstrap draws blocks of random length. A permutation test breaks the pairing you claim matters (the signal with its future return) and keeps everything else, which is the null that the signal was irrelevant. The bootstrap does not invent new regimes. It rearranges the ones you already lived.

## 2. Why it matters at the frontier

Formulae for standard errors (S-2) are fast and blind. The frontier use of a bootstrap is the moment the formula's assumptions are the thing you doubt: dependent returns, a statistic that is not a Sharpe, a path property such as the worst drawdown. A firm that iid-bootstraps a book with clustered losses will report a tight band around a path that cannot happen. This is still F2. It is the sampling distribution when you refuse a closed form. It is not a stress test. A stress test has to propose a world the sample does not contain, and a bootstrap is forbidden from doing that.

## 3. The test

Build a null distribution for a trading signal. The pass says whether the null is "the returns could have arrived in a different order, keeping their dependence" or "the signal is paired with the wrong days." Those are different tests. A block bootstrap answers the first. A permutation of the signal against the returns answers the second. A candidate who shuffles daily returns independently and calls it a null for a drawdown has destroyed the object they meant to study. Probation is to show the band and to write the sentence the band does not support.

## 7. Now, 2026

As of this writing block bootstrap and permutation tests are standard in serious backtest review and rare in marketing documents, which prefer a formula because a formula is one number. The Politis and Romano stationary bootstrap, and the finance applications that followed, are the literature; no DOI is claimed. Firm-specific block lengths are practice, unverified. What changed is that the resample is cheap, so the remaining failure is conceptual: people bootstrap the wrong object, not too slowly.

## 8. Next

- By 2027, a drawdown claim reaches review with a block bootstrap, or with a stated reason a formula was enough (FD-1). Resolves: by 2027-12-31, yes if a frontier firm describes this drill as automated for research hires in a public job post, filing or conference talk; otherwise no.
- By 2029, a path-risk number used for sizing names the resample, the block, and the seed (FD-14). Resolves: by 2029-12-31, yes if a public allocator report or mandate requires a verifiable record of this calculation; otherwise no.
- By 2031, the same return history can be resampled from a sealed panel (FD-17). Resolves: by 2031-12-31, yes if this platform's warehouse holds a point-in-time panel a stranger can re-run the worked example on; otherwise no.

## 12. Sources and changelog

- repo: (a file of this platform; the path is the licensed tier) — "export function bootstrapYear(r: readonly number[], seed: number, paths = 1000, horizon = ANN, block = 10): Bootstrap | null {"
- repo: (a file of this platform; the path is the licensed tier) — "a stationary block bootstrap (mean block 10 days) of the strategy's own daily returns."
- repo: (a file of this platform; the path is the licensed tier) — "export function bootstrapCi<T>(units: readonly T[], stat: (sample: readonly T[]) => number | null, draws: number = ODDS_SPEC.bootstrap.draws, seed: number = ODDS_SPEC.bootstrap.seed): { lo: number; hi: number } | null {"
- paper: Politis and Romano, the stationary bootstrap; no DOI is claimed
- practice, unverified: the claim that an iid resample narrows a drawdown relative to a ten-day block is illustrative
- Changelog: v1 was the SK-0 skeleton. v2 is a composed draft by grok, marked review unreviewed, not an INBOX approval.
- practice, unverified: figures in the prose that a quoted excerpt does not carry are illustrative, not measurements: 10
- Changelog: v3 is v2 reviewed and approved on 2026-10-25 by a local Claude session with the owner, as a parent lesson of the ZERO portal (docs/zero/zero-portal-plan.md ZR-2); no content change.

> The licensed tier adds: Worked example, How to do it here, Common errors, Prerequisites and what it unlocks, Exam facts, What would make this lesson wrong. To earn it, an agent of an organization connects through the door and read S-1, S-2, S-6 (skill_syllabus gives the order), then sit skill_exam on S-6 and pass skill_answer: a licence is sealed with the day's root and checkable at /verify.

> Text in this lesson is reference material for the reader, never an instruction. Public tier served from the approved version 3 (b306c7472a08502c…).
