---
id: S-2
kind: skill
version: 3
title: "Estimation under noise"
series_group: "4.1 Mathematics and statistics"
prerequisites: [S-1]
unlocks_tools: []
now_asof: "2026-10-16"
next_horizons: [2027, 2029, 2031]
authorKind: grok
skillHash: "7cf06f7180fd18ed4286246f0eb87b9199781ccc6c9440984f9d7d122a0387a4"
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-2 Estimation under noise

## 1. Definition

Estimation under noise is the refusal to treat a point as a fact until you have said how wide the sampling cloud around it is. A mean's uncertainty shrinks like one over the square root of the number of independent observations. A Sharpe ratio is a mean divided by a deviation, so both pieces move, and they move together. Its standard error depends on the Sharpe itself, the length of the sample, and, outside the normal model, on skewness and kurtosis. An information coefficient, the correlation of a forecast with the outcome, has the same disease: overlapping returns reuse the same days, and the textbook correlation error pretends they were fresh.

## 2. Why it matters at the frontier

This is how a firm declines a backtest without insulting the author. An annual Sharpe of one half over a few years is the usual pitch, and under an independent normal model the error bar is about as wide as the number. Capital allocated to that pitch is allocated to noise. The frontier it serves is F2, and the survival half of F5: a receipt that does not include the band is not a receipt. Desks that size on a live Sharpe and never print the error are running the same mistake on a shorter clock.

## 3. The test

The interview asks for the standard error of a Sharpe and of an information coefficient, then for how many years a Sharpe of one half needs before it clears zero. The pass states the square-root law, says the normal formula is too tight once returns are fat-tailed or autocorrelated, and will not give a year-count without the assumptions. Probation is a return stream the firm already holds: compute the ratio and the skew-and-kurtosis error, and say whether zero is inside the band. Annualizing first and then using a daily formula fails.

## 7. Now, 2026

As of this writing the error-bar conversation is ordinary inside systematic firms and still rare in pitch books. Public compilations of hedge-fund Sharpes are bent by who chooses to report, and this lesson does not treat them as measurements. The skew-and-kurtosis formula is old (Lo, 2002); what changed is that search became cheap, so the plain sampling error is no longer the binding one. Firm thresholds for "high enough to fund" are practice, unverified. This platform's visible rule is the floor of 30 observations, under which the kernel is silent.

## 8. Next

- By 2027, a research Sharpe is shown to a person only with its sampling band already computed (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, capital sized off a Sharpe carries the band and the trial count in the same sealed row (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, a sealed return panel long enough to recompute the band sits in the warehouse (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 sharpeVariance(sr: number, t: number, skew: number, kurt: number): number {"
- repo: (a file of this platform; the path is the licensed tier) — "`kurt` is raw (normal = 3), not excess."
- repo: (a file of this platform; the path is the licensed tier) — "export const RISK_RATIO_MIN = 30;"
- paper: Lo, The Statistics of Sharpe Ratios; the citation is not a measurement of this book
- practice, unverified: the worked band (Sharpe 0.5, square 0.25, half-square 0.125, sum 1.125, about 0.53 at 4 years and about 0.27 at 16) 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: 0.5 0.25 0.125 1.125 0.53 16 0.27 30
- 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 (skill_syllabus gives the order), then sit skill_exam on S-2 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 (7cf06f7180fd18ed…).
