---
id: S-17
kind: skill
version: 3
title: "Survival analysis"
series_group: "4.1 Mathematics and statistics"
prerequisites: [S-1, S-4]
unlocks_tools: []
now_asof: "2026-10-16"
next_horizons: [2027, 2029, 2031]
authorKind: grok
skillHash: "44c947b4a28cd97898300aed4c42b3db0f759ce458d1ad38d43c4937e58594d0"
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-17 Survival analysis

## 1. Definition

Survival analysis is how you talk about a waiting time when some of the waits have not finished. A hazard is the chance the event happens now, given it has not happened yet. Defaults, the death of a signal, and a client who has not churned are all waiting times, and the ones still waiting are data, not missing data. Dropping them biases the average life toward the ones that already ended. A half-life is the waiting time until half the gap, or half the cohort, is gone. For a mean-reverting spread it is how fast the residual closes. For an alpha it is how fast the forecast stops being a forecast. The interview asks you to model the half-life of an alpha, and the trap is to compute it on the backtest that survived.

## 2. Why it matters at the frontier

Capacity, signal retirement, and credit all run on lifetimes. A firm that quotes a half-life from the strategies it kept will think its alphas live longer than they do, because the short-lived ones were deleted before the average. That is censoring by the research process, and it is the same shape as a default study that ignores names still solvent. F2 on a clock. The pairs trade in S-4 is the special case where the lifetime is a parameter of a residual, and even there a half-life inside a pretty band is not a profit. Costs are paid per trade, and a fast half-life can still be too expensive.

## 3. The test

Model the half-life of an alpha. The pass defines the event (the forecast correlation crossing a floor, or the residual gap halving), says which forecasts are still alive and must not be dropped, and separates a half-life fit on the past from one fit on the whole history of a survivor. A candidate who averages the life of the strategies in the book has described the survivors. Probation is a cohort that includes the signals you killed.

## 7. Now, 2026

As of this writing hazard models are ordinary in credit and in customer analytics, and half-lives of signals are ordinary language inside quantitative research with very uneven statistics underneath. The careful version keeps the killed signals in the denominator. Firm-specific decay curves are practice, unverified. What changed is the speed at which signals are born and killed, which makes the censoring worse if the ledger of deaths is not kept. This platform keeps the pairs gate. It does not yet keep that ledger as a survival model.

## 8. Next

- By 2027, a signal retirement is recorded as a censored lifetime, not dropped from the average (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 half-life used to size a holding names the cohort, including the deaths, 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, a sealed cohort of signals can be refit by a stranger (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) — "half-life in [2, 30] bars"
- paper: survival methods in credit and in duration data; no DOI is claimed
- practice, unverified: ten signals of which six died inside a quarter is a sketch of censoring, not a cohort in the ledger
- 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: 30 -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); one stale cross-reference fixed (S-38 is written).

> 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-4, S-17 (skill_syllabus gives the order), then sit skill_exam on S-17 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 (44c947b4a28cd978…).
