---
id: S-4
kind: skill
version: 3
title: "Time-series analysis"
series_group: "4.1 Mathematics and statistics"
prerequisites: [S-1, S-3]
unlocks_tools: []
now_asof: "2026-10-16"
next_horizons: [2027, 2029, 2031]
authorKind: grok
skillHash: "c96ba093e6607eb3dea1fa4bc668e8bb20a56077bc63bd9646848d4c9af08210"
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-4 Time-series analysis

## 1. Definition

A time series is a sample in which order is the point. Stationarity means the distribution of the shocks does not wander off as time passes, so an average computed on the past is about the future. A unit root means it does wander: the level remembers every shock forever, and a regression of one wandering level on another will look brilliant and mean nothing. ARMA is a short-memory model of a stationary series. GARCH is a model of the width of the shocks, which clusters. Cointegration is the exception that saves a pair of wandering series: each level is a unit root, and one linear combination is stationary, so the gap between them comes back. Most findings of that exception are spurious. A finite sample of two random walks produces a combination that looks mean-reverting often enough to fill a notebook.

## 2. Why it matters at the frontier

Pairs trading, rates-curve trades, and any "this spread will close" story are this skill or they are a story. The frontier test is not whether you can run an augmented Dickey-Fuller. It is whether you know why most of the passes are accidents, and what else has to be true (a half-life short enough to pay the costs, a sample that was not chosen after seeing the p-value, a combination that was not the best of a hundred pairs). This is F2 on dependent data. Getting the dependence wrong does not add noise. It invents relationships.

## 3. The test

Design a pairs test and explain why most cointegration findings are spurious. The pass pre-registers the pair, fits the combination only on the past, tests the residual, demands a half-life inside a window that can pay costs, and says what a negative control is (a pair you expect not to cointegrate, or a shuffled clock). A candidate who quotes a p-value from a full-sample fit and stops has failed. The spuriousness argument has to be said in words: two unit roots will correlate in sample.

## 7. Now, 2026

As of this writing Engle-Granger and Johansen are still what desks run, and the better desks pre-register the pair or charge for the search. GARCH remains the workhorse for variance clustering even though its forecasts are often beaten by simpler smoothers; that comparison is practice, unverified at the firm level. What changed is the number of pairs a person can test in an afternoon. The spurious-regression warning is older than the computers that made it urgent. Critical-value tables have not become more generous. The search has.

## 8. Next

- By 2027, a pairs candidate reaches a person only after a past-only fit, a half-life, and a trial count (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 spread used for sizing carries the test, the bucket, and the search size in the 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, the same two series can be retested from a sealed warehouse 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 adfT(e: readonly number[]): number | null {"
- repo: (a file of this platform; the path is the licensed tier) — "never interpolated past what the table supports"
- paper: Engle and Granger on cointegration; MacKinnon on the critical values; no DOI is claimed
- practice, unverified: the two random walks and the "often large" sample correlation are 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: 0.031
- 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-3, S-4 (skill_syllabus gives the order), then sit skill_exam on S-4 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 (c96ba093e6607eb3…).
