---
id: S-26
kind: skill
version: 3
title: "Panel-data econometrics"
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: "7c063e4799e7d12794d1f2d28eb1ef9cfef44220515f917660021c9f729458e2"
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-26 Panel-data econometrics

## 1. Definition

A panel is many names observed many times. The rows are not a sample of independent draws. The same name repeats, the same day hits every name, and a t-statistic that pretends otherwise will be enormous. Fixed effects subtract a name's own average, or a day's, so the slope is identified from the changes rather than from the permanent differences. Clustered standard errors let the residuals inside a name, or inside a day, move together, and they widen the error until the sample size is closer to the number of clusters than to the number of rows. Fama-MacBeth runs a cross-sectional regression each period and then averages the slopes, and the standard error of that average is the honest one only if you let the slopes be dependent across time. The interview asks why the t-statistics in a cross-sectional study are overstated without clustering. The answer is that the study counted rows, and the shocks were fewer than the rows.

## 2. Why it matters at the frontier

Every "we have a hundred thousand observations" slide in equity research is this lesson or it is a sales document. The frontier failure is a factor t-statistic computed as if each stock-day were a fresh experiment. It is not. A market shock is one shock, shared. F2 on dependent data, the panel version of S-3. F5 is the clustering unit written next to the t-statistic, so a stranger can see whether you clustered by name, by day, or not at all. Fixed effects are not a causal claim. They remove a level. S-16 is what you need if you want the word cause.

## 3. The test

Why are the t-statistics overstated without clustering? The pass points at the repeated name or the repeated day, says the effective sample is the number of independent shocks, and says what the cluster should be. A candidate who says "the sample is large" fails. A candidate who adds fixed effects and then reads the slope as the effect of a policy fails a different question, S-16's. Probation is a regression of many names on a factor that is the same for every name on a day: the cluster is the day, and a row-level t-statistic is theater.

## 7. Now, 2026

As of this writing clustered errors and fixed effects are the expected hygiene in empirical finance, and slides that ignore them are still common in marketing and in first drafts. Fama-MacBeth remains the language of cross-sectional factor papers. Which production risk system clusters which way is practice, unverified. What changed is the size of the panel a person can regress in a minute, which makes the fake precision cheaper to produce. The shock count did not grow as fast as the row count.

## 8. Next

- By 2027, a panel t-statistic reaches a person with the clustering unit, or it is not shown (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 factor slope used in a genome carries the error estimator 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 peer panel can be refit from filings and prices 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) — "peer fixed effects, Newey-West errors."
- repo: (a file of this platform; the path is the licensed tier) — "the lines are NOT added up."
- paper: Fama and MacBeth; the clustered-error literature; no DOI is claimed
- practice, unverified: five hundred stocks and a few hundred daily shocks against a hundred thousand rows is a counting sketch
- Changelog: v1 was the SK-0 skeleton. v2 is a composed draft by grok, marked review unreviewed, not an INBOX approval.
- 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-26 (skill_syllabus gives the order), then sit skill_exam on S-26 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 (7c063e4799e7d127…).
