Terminology: How News Quantified Uses Its Words

TL;DR

Every News Quantified page uses one vocabulary, defined here. If a term on this site appears to conflict with its usage elsewhere in the industry, this page states our definition and why. The short version of the philosophy: words that describe measurements must themselves be measurable. Each definition below is one to three sentences, written to be quoted whole.

Core concepts

News Reaction Intelligence

The measurement of how stocks actually behaved after classified news events: price moves, volume, and continuation or reversal over defined windows, historically and in real time. It measures; it does not predict. The category page.

Event

One classified occurrence for one security at one point-in-time timestamp: the atomic record of the dataset. A single news item can produce multiple events when it carries multiple classifiable disclosures.

Event class

A defined category of events with an explicit inclusion rule, such as guidance cut, secondary offering, or unscheduled CEO departure. Classes, not individual events, are the unit of research, because only populations produce distributions. Full list: data dictionary.

Context flag

A first-class field distinguishing variants of an event class that the market historically treated differently, such as scheduled vs unscheduled disclosure or successor-named vs sudden departure.

Chain

Linked events forming a sequence, such as rumor, confirmation, or denial. Chains preserve structure that single-headline records destroy.

Measurement terms

Reaction

The measured market behavior following an event: price and volume over defined windows, under a stated return convention. A reaction is an observation, never a score or an opinion.

Reaction distribution

The full set of reactions across an event class's historical population, reported with N, median, and percentiles. The distribution, not the average, is the honest summary of any event class.

Base rate

What the reaction distribution says happened historically for a given event class: the starting point the market's memory provides before any view about a specific event.

Window and day 0

Day 0 is the first trading session reflecting the event under the stated convention; windows such as day +1 to +5 measure the aftermath. Every published number states its window.

Drift and reversal

Drift (or continuation) is the historical tendency of post-event prices to extend the day-0 move over the following days; reversal is the tendency to retrace it. Both are class properties, measured, and neither is a prediction.

Data integrity terms

Point-in-time

Recorded as the market could have known it at that moment: timestamps reflect first public availability and are never retroactively improved, and universes are constructed as of the event date. The discipline that prevents lookahead bias. Methodology.

Entity resolution

Linking events to persistent security identifiers rather than ticker strings, so ticker changes, mergers, and delistings never orphan or misattach history.

Survivorship handling

Delisted and acquired companies remain in the dataset with full event history. Their absence would delete the left tail from every distribution.

Versioning

Corrections coexist with prior record versions rather than overwriting them, so any delivered snapshot is reproducible. Material changes are logged in the dataset changelog.

Product terms

The dataset

The News Quantified dataset: 26M+ news records over 20+ years, collected continuously since 2012, with reactions as first-class fields. We use the word dataset consistently across the site, including where academic NLP literature might use other terms, because our buyers are practitioners and practitioners license datasets.

Data dictionary

The public specification of event classes and fields; the full version with per-class rules is provided to prospective licensees.

Event study

A fixed-template measurement of one event class: stated N, universe, window, two charts, one segmentation, and a what-this-does-not-say section. The library.

Readout

A company-specific document benchmarking an issuer's announcement reactions against sector base rates per event class. The IR-facing product surface. IR Intelligence.

MCP connector

The Model Context Protocol server that lets AI assistants query the dataset directly, returning base rates with N, window, and date range attached. Setup.

Words we do not use

Predict, signal, forecast, and recommendation do not appear in NQ measurements, because the dataset contains none of those things. Where a page discusses those concepts, it is explaining why they are out of scope. Hype adjectives are absent by policy, not oversight.

Frequently asked questions

What is the difference between a reaction and a sentiment score?

A sentiment score is an opinion about text, assigned at publication. A reaction is a measured market outcome after the event. NQ records reactions and does not score sentiment.

What does base rate mean in event research?

The measured historical distribution of reactions for an event class: what happened across all comparable past events, as context for the current one.

Why does News Quantified emphasize distributions over averages?

Event reactions are wide and fat-tailed; an average conceals the dispersion that carries most of the information. Distributions with stated Ns are the honest summary.

Historical reaction data. Not investment advice. Past reactions do not determine future reactions.