News & social sentiment

Definition

News and social sentiment quantify the tone and volume of text about markets — professional media, wire flow, and social platforms — as tradeable data. The peer-reviewed anchor is Tetlock (2007): high media pessimism predicted short-horizon price pressure followed by REVERSAL — media tone behaved as sentiment (noise that mean-reverts), not information. The social-platform era added a stronger claim: attention flows now MOVE prices directly in retail-heavy names — documented at scale in the SEC’s 2021 report.

How it works / structure

• The measurement stack (engine-executable): tone scoring (dictionary or model-based — the platform pins the scorer version per replay; scorer drift silently rewrites history), VOLUME/attention (mentions, story counts — frequently more informative than tone), novelty (first story vs echo — news decays within minutes in liquid names), and source tiering (wire > outlet > aggregator > anonymous forum).
• The documented effects: tone extremes revert (Tetlock); attention spikes in retail-heavy small caps precede volatility and volume regardless of tone (the 2021 mechanics: coordinated attention → options flow → dealer hedging feedback — indicator-options-flow, sent-short-interest fuel); news momentum (post-news drift) exists in some samples at short horizons.
• The information boundary: trading nonpublic material information is illegal — the platform’s social data is public-forum flow only; rumor-stage M&A chatter (event-mergers-acquisitions) is treated as attention data, never as fact.
• Regime scope: sentiment data’s effect sizes concentrate in small, retail-heavy, hard-to-borrow names; mega-caps absorb attention flows with minimal footprint.

When it applies

Attention-spike screens (volatility-anticipation, not direction); crowding context on shorts (sent-short-interest

• social attention = squeeze-fragility composite); event narrative tracking (qualitative-analysis reaction-vs-news reading needs the news series); tone extremes as reversal context in the Tetlock pattern.

Risk profile & failure modes

• Bot and manipulation contamination: social volume is gameable and gamed (pump groups are documented enforcement targets); volume without account-quality filtering measures the manipulation, not the crowd.
• Scorer nonstationarity: language drifts (irony, ticker slang); a scorer trained on 2015 text misreads 2026 forums; version pinning + revalidation is mandatory.
• Echo inflation: syndication multiplies one story into hundreds of “mentions” — novelty filtering before volume counts.
• Direction overclaim: attention predicts VARIANCE far better than SIGN; directional social-sentiment systems mostly fail replay (the platform’s stance until a replay proves otherwise).

Evidence & limits

Tetlock (2007) is the peer-reviewed tone anchor; the SEC 2021 staff report documents the attention-flow mechanics at their historical extreme. Commercial sentiment feeds carry unpublished construction and survivorship choices — labeled vendor data. Directional claims from sentiment alone are folklore until replayed.

Falsifiable-thesis examples

Illustrations only, not signals:

• “Names entering the top attention decile (novelty-filtered) will realize top-quartile volatility over the following week (variance thesis)” — falsified by the realized-vol ranks.
• “Extreme-pessimism tone scores on X will be followed by positive 2-week returns more often than not this year (Tetlock-reversal thesis)” — falsified by the conditional tally.

Cross-references

• The frame: lens-sentiment; the behavior: bias-herding
• The squeeze composite: sent-short-interest, indicator-options-flow
• The narrative discipline: qualitative-analysis
• The legal boundary and rumor handling: event-mergers-acquisitions