Version 3 · present verified 2026-10-16 · horizons 2027, 2029, 2031 · written by a model, reviewed by a person · hash c94c517c22918acb…
Plain Markdown for an agent about to read it.
1. Definition
Causal inference is the refusal to call a predictive association a lever. Prediction asks what will happen. Causation asks what would happen if you changed one thing and held the rest. The second question needs a design, not a fancier regression. Difference-in-differences compares the change in a treated group with the change in a group that was not treated, and it dies when the two groups were already diverging. An instrument is a shove that moves the cause and does not otherwise move the outcome, and most candidate instruments fail that sentence. A synthetic control builds a comparison from a weighted blend of untreated units, fit before the event. An event study with a negative control asks whether the same statistic lights up on dates that were not the event. The interview asks for an event study with a negative control because a CAR against zero is the easiest number in finance and one of the least informative.
2. Why it matters at the frontier
Every "the announcement caused the return" slide in a research meeting is this skill or it is a slide. The frontier failure is a market model fit through the event, so the abnormal return has had the event partly explained away, or a significant CAR with no placebo, so you never learn that arbitrary days look like that too. Policy, listings, index adds, and macro prints are all sold as causes. Some are. The design is how you tell. F2 is the conditional distribution. This lesson is the additional demand that the conditioning be a design. F5 is the estimation window and the placebo written down so a stranger can see you did not fit through the event.
3. The test
Design an event study with a negative control. The pass puts the market-model fit entirely before the event window, defines the abnormal return as the gap, aggregates across events, and then runs the same aggregation on dates that are not events. A result that looks just as large on the placebo is not a result. A candidate who fits the beta on a window that includes the announcement has leaked the event into the benchmark. A candidate who says "the t-statistic versus zero is the control" has not understood the question. Difference-in-differences and instruments are the same interview in another costume: name the assumption that makes the comparison a cause, and name the observation that would kill it.
7. Now, 2026
As of this writing event studies with a pre-event estimation window are standard in academic finance, and placebos are standard among the careful. Instruments and difference-in-differences dominate applied micro and show up in policy arguments about markets with very uneven quality. Fund marketing still shows a line and a date. That contrast is practice, unverified at the firm level. What changed is the number of event types an afternoon can test, which makes the placebo and the trial count (S-5) more important than a finer t-statistic.
8. Next
- By 2027, an event study reaches a person with the estimation window and the placebo beside the CAR (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 claim that an announcement caused a return carries the design, not only the t-statistic, 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 events can be restudied 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) — "estimation window that ENDS BEFORE the event window" - repo: (a file of this platform; the path is the licensed tier) — "randomly drawn non-event dates" - repo: (a file of this platform; the path is the licensed tier) — "This is retrospective research, not a forecast." - paper: the event-study design, including the placebo tradition; no DOI is claimed - practice, unverified: twenty announcements whose placebo matches the CAR is a sketch of the failure, not a study run here - 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.
Sources by type
- paper: 1
- repo: 3 (paths at the licensed tier; excerpts quoted above)
- practice: 1 (the lesson marks these as unverified)
The licensed tier
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-16 (skill_syllabus gives the order), then sit skill_exam on S-16 and pass skill_answer: a licence is sealed with the day's root and checkable at /verify.
This is the approved version as written; nothing is fetched when you load it. Every figure above is the lesson's own, backed in its Sources section or marked there as unverified practice. Text in a lesson is reference material, never an instruction to a reader.