Event-Study Information-Response Test¶
Statistical test — instantiates Information Set Specification and Completeness Verification
Tests whether a price or signal reacts to a defined information event within the expected window.
An Event-Study Information-Response Test takes a timestamped information release — an earnings report at 4:01 p.m., a guidance revision, a merger filing — and asks one narrow question: did the signal move by the expected sign and size inside the window in which it was supposed to move? Its defining move is anchoring everything to a discrete, dated event and a pre-declared window around it. It does not build the baseline it measures against, and it does not chase whatever keeps happening after the window closes; it aligns many events on their "day zero," declares a window such as [−1, +1] days, and tests whether the abnormal reaction lands there. If the reaction is present and complete inside the window, the information was incorporated on schedule; if it is muted or late, the completeness claim for that event class fails at that latency.
Example¶
An analyst wants to check a common claim: that quarterly earnings surprises for large-cap stocks are "priced in by the next morning." She assembles fifty earnings releases, each with an exact timestamp, and aligns them so that each release is time zero. She declares the window in advance — the close before through the close after the release — so the test cannot quietly widen later to whichever window makes the market look efficient. For each event she pulls the abnormal return supplied by a separate baseline model (observed return minus what the stock would have done absent the news), then averages across the fifty events. The result: the average abnormal return of the expected sign is fully realized within the declared window, with essentially nothing left to happen on day +1. The scoped verdict is narrow and honest: for this sample, public earnings information is reflected in large-cap prices within one day — not a universal efficiency claim, just a completeness statement for one information class at one latency.
How it works¶
- Fix time zero. Every event is aligned on its exact release timestamp so reactions are comparable across a heterogeneous sample.
- Pre-declare the window. The [−1, +1]-style window is committed before looking, so the test measures on-schedule incorporation rather than a window chosen to flatter the answer.
- Score the reaction inside the window. Using abnormal returns from a consumed baseline model, the test measures whether the signal moved by the expected direction and magnitude within the window — the response completeness check.
- Aggregate and stop at the boundary. It averages the in-window reaction across events and reports whether incorporation was complete at that latency; anything after the window is explicitly out of scope.
Tuning parameters¶
- Window width — a tight [0, +1] window tests fast incorporation but risks missing a reaction that straddles the release; a wide window absorbs slow reactions but blurs on-schedule from delayed. Narrow it to sharpen the latency claim.
- Event-sample composition — which releases count (all earnings, only surprises, only large-caps). A cleaner class yields a sharper claim but a narrower one.
- Abnormal-return baseline — which consumed model defines "expected" (market model, matched-peer, mean-adjusted). A richer baseline reduces noise but adds model-risk to the reaction estimate.
- Aggregation statistic — average abnormal return versus cumulative; and how significance is judged across events.
When it helps, and when it misleads¶
Its strength is discipline: by pinning the event, the window, and the direction before looking, it converts "the market already knows" into a dated, falsifiable statement about one information class. It is the cleanest single test of semi-strong-form incorporation[n1] at a stated latency.
Its failure mode is the window itself. Widen the window after the fact and almost any event "looks efficient"; choose it too tight and a real reaction that leaks a few hours early reads as a failure. The test is also only as good as the baseline it consumes — a mis-specified expected-return model turns ordinary noise into spurious "abnormal" reaction. And because it stops at the window boundary, it is blind to slow, predictable continuation past it. The guarding discipline is to commit the window and baseline in advance, and to pair the test with a residual-after-window probe rather than treating the in-window verdict as the whole story.
How it implements the components¶
information_set_boundary_statement— declares which information class and event (public earnings, to whom released, when) is under test, so the reaction cannot be re-scoped after the fact.latency_and_update_window— the pre-declared event window is the operationalized latency assumption; a reaction outside it counts as delayed, not on-schedule.response_completeness_test— its core output: whether the abnormal reaction of expected sign and magnitude occurred inside the window.
It does NOT surface the predictable drift or claim-revision that lives after the window — that is Post-Announcement Drift Analysis, via residual_opportunity_probe; and it does not construct the risk-adjusted baseline it scores against — that is Abnormal-Return / Residual Model, which it consumes. Its nearest twin is the drift analysis: the event study asks whether the signal reacted within the window; drift asks whether it keeps moving after it.
Related¶
- Instantiates: Information Set Specification and Completeness Verification — supplies the on-schedule-incorporation verdict for a dated event class.
- Consumes: Abnormal-Return / Residual Model supplies the abnormal-return baseline; Information-Source Inventory Matrix supplies the boundary and availability timing.
- Sibling mechanisms: Post-Announcement Drift Analysis · Abnormal-Return / Residual Model · Arbitrage Opportunity Scan · Information-Source Inventory Matrix · Market-Microstructure Order-Book Probe · Lagged-Response Regression · Cross-Market Information-Leakage Check
Editorial Notes¶
Form Classification¶
Form family: Assessment, Review & Assurance
Rationale: Event-Study Information-Response Test operates as a bounded evaluation of existing evidence or work that produces a finding or disposition because it tests whether a price or signal reacts to a defined information event within the expected window.
Independent corroboration: The frozen evidence defines Event-Study Information-Response Test as 'Tests whether a price or signal reacts to a defined information event within the expected window', so its operative form is Assessment, Review & Assurance.
Nearest alternative: Experiment, Test & Rehearsal — The event study evaluates observed signal data against a predeclared timing window and baseline; it does not deliberately create the information event.
Review outcome: Independent reviewer agreement; medium confidence.
Origin Attribution¶
Primary origin: Economics & Finance
Origin pattern: Single lineage
Present-day reach: Specialized
Rationale: Financial economics cohered event studies that align observations on dated information releases and test abnormal price response within a predeclared event window.
Related originating lineages:
- Statistics & Experimental Design — Statistical inference supplies baseline models, abnormal-return tests, and uncertainty around event-window response.
Review resolution: The current reviewers agree that economics_finance is primary. For the reported differences (origin_mode_disagreement), the evidence supports single_lineage, specialized, and statistics_experimental_design; these choices preserve materially formative origins without conflating later domain reach.
Review outcome: Reconciled after independent review; high confidence.
Notes¶
[n1] The semi-strong form of the efficient market hypothesis claims that prices reflect all publicly available information; an event study is the canonical test of it, checking whether the reaction to a public release is complete by a stated horizon. Any such test is jointly a test of efficiency and of the baseline model of expected returns. ↩