How market reaction is measured
This page documents the measurement behind every readout: what counts as a news event, how each event is classed, and how the market reaction is read. The dataset holds 26M+ news events over 20+ years, recorded point-in-time.
What counts as a news event
A news event is a discrete, timestamped disclosure attributable to one listed company. The taxonomy classes each event at capture; the classes most relevant to IR work:
- Earnings
- Quarterly and annual results. Where consensus exists, the event is the reported figure against it — the surprise, not the level.
- Guidance
- Forward statements issued, raised, reiterated, lowered, or withdrawn. Classed separately from the results they accompany.
- Filings
- SEC filings, including 8-K material events, taken at the filing timestamp.
- Analyst actions
- Rating changes, initiations, and price-target changes, attributed to the issuing firm.
- Contracts and orders
- Awards, renewals, and cancellations disclosed by the company.
- Management changes
- Appointments and departures of named officers and directors.
Further classes — dividends and buybacks, offerings, M&A, regulatory and clinical decisions among them — follow the same rules. Every class is measured the same way.
How each event is classed
Each event is recorded with four fields at capture: the ticker it is attributable to, the source (wire, filing, or publisher), the event class from the taxonomy above, and the publication timestamp. One disclosure can produce more than one event — a results release that also lowers guidance is an earnings event and a guidance event, measured separately.
How the market reaction is measured
- The clock starts at the news timestamp, not the trading day. A release at 8:02 and a release at 15:55 are different events even when the headline is identical.
- Horizons are fixed: the price move is measured from the timestamp at 30 minutes, at the session close, and three trading sessions out.
- Every reaction carries volume context:traded volume as a multiple of the stock’s average daily volume. A +2% move on 0.3× volume and a +2% move on 4× volume are different readouts.
- Each reaction is read against history: 20+ years of the same event class — what this class of event did, the last N times it happened.
- “No measurable reaction” is a recorded outcome, not missing data. A release the market did not price is stored as exactly that.
Point-in-time integrity
Every event is stored as it appeared at publication — headline, timestamp, classification — and is never revised with later information. Corrections, restatements, and subsequent context are new events, not edits. Zero look-ahead: nothing measured at a horizon uses data from after that horizon.
The constraint is structural. A reaction history is only valid if it was recorded live; you cannot backfill 20 years of point-in-time reactions after the fact. The dataset has been recorded this way from the start, which is why it reaches 20+ years.
Peer-reviewed in 2016
The event-reaction methodology was examined in published academic research built on this dataset.
Levi (Tel Aviv), Livnat (NYU Stern), Zhang (Rutgers) & Zhang (UC Berkeley Haas), 2016 — “Are extended hours prices predictive of subsequent stock returns?” — peer-reviewed research built on News Quantified data. The study validates the event-reaction methodology: the measured market reaction to a news event predicts the subsequent drift.
Read the study →Institutional clients since 2012
News Quantified has supplied event-reaction data to institutional clients since 2012. Clients include NASDAQ and Business Wire. The readout for public companies runs on the same dataset — same taxonomy, same horizons, same point-in-time record.