Price tells you what traders paid to express belief. News and social text tell you why they might have paid—sometimes. Sentiment work in prediction markets is not a black-box “bullish meter.” It is structured triage: tier the source, score impact on the contract’s resolution rules, and compare narrative velocity to π and volume.
Text updates your Bayesian beliefs; price validates whether the crowd agreed. Information-flow chapters described headline jumps; strategy chapters split momentum and contrarian responses—sentiment is the text layer those modules assume you already read.
Layers in the signal stack
Market mid and spread show current belief and friction. Volume and open interest show whether capital backed the move. Wires and official releases carry fundamental shocks. Social feeds carry attention, narrative, and manipulation risk. Polls and models supply outside view. Your journal supplies discipline. None alone is sufficient.
News taxonomy
Tier zero is official: agencies, courts, regulators, company filings. Markets usually spike fast; leader venues move first.
Tier one is credible press with named sources and editorial standards.
Tier two is commentary—strategists, op-eds—often small price impact.
Tier three is social rumor—anonymous accounts, screenshots without provenance.
Tier four is meme and irony—dominant on play-money venues, irrelevant to bankroll.
Tag every headline before trading. If you cannot tag it, you should not size it.
Scoring impact, not mood
For each item, score −2 to +2 on whether it moves YES probability under the written rules, not whether it sounds scary. A court denial that keeps a candidate on the ballot is −2 for YES on “disqualified before ballot” even if viral posts scream the opposite. Weight tier: emphasize tier zero, discount tier three. After the third duplicate headline, discount weight again—recycled content adds heat, not information.
A daily sentiment index is optional—a weighted sum you compare to overnight Δπ. Agreement strengthens narrative; divergence forces you to find rule, venue, or liquidity explanations.
Election legal headline example
YES on disqualification trades down eight cents after AP reports a court denied the motion. Campaign statements of “fighting on” are narrative zero. Viral “he’s out” posts are high emotional impact but tier-three weight near zero under true rules. Mild negative sentiment aligns with price—no contrarian buy unless your p′ still beats the ask after fees.
If π fell eight cents but your p′ only moved three, contrarian entry is plausible—if the ask is below p′ net of spread.
Macro whisper trap
Threads promising “huge CPI miss” without sources are tier three. If π rises six cents anyway, price may be leading text—check leader venue, economist row, and volume before betting the whisper. Whisper trades are lotteries against tier-zero releases.
Social volume versus market volume
Social measures engagement; markets measure dollars at risk. Social up, price flat often means attention without belief. Price up, social flat may be informed flow or whales. Both up together raises cascade risk—second-hand citations, not primary documents. Social up, price reverting is a fade candidate only with positive net edge, not because Twitter was loud.
Cascade checklist before you chase or fade
Was the move larger than twice the usual spread? Are people citing each other instead of sources? Did new or surge accounts appear? Do related contracts disagree on the tree? Is expiry inside seventy-two hours where momentum intensifies? Affirmative answers demand slower size, not heroics.
Cascades are the social form of spike overreaction—classification first, click second.
Combining sentiment with consensus π
Morning π from venues. Compute S since prior close. If sign(S) matches sign(Δπ), narrative and capital agree. If they oppose, find rule, venue, or liquidity story before trading. Bayes-update to p′. Trade only if p′ minus ask clears buffer and sizing caps.
Platform echo chambers
Global crypto-native books echo crypto Twitter on token and policy stories. U.S. regulated books echo FinTwit on Fed path. Capped retail echoes political Twitter with stale extremes. Play-money echoes Discord memes. Weight social by where capital actually trades, not where outrage is loudest.
Morning narrative scan (conceptual)
List active dossiers. Pull overnight Δπ. Scan wires tier zero and one first. Score S for movers. Tag leader venue. Confirm volume and open interest. Run cascade checks on spikes. Update p′; compare to asks. Queue trades meeting sizing rules. Evening: log whether S and Δπ agreed—build personal calibration over months.
Ethics and information hygiene
Read primary sources before sizing. Timestamp screenshots for disputes. Avoid engagement bait tier four. Disclose conflicts that feed overconfidence. Sentiment is input to p′, not justification for oversize.
When sentiment and price disagree
If your S is strongly negative but π rose, ask whether the market knows a tier-zero fact you missed, whether another venue led with size, or whether the contract text makes the social story irrelevant. If S is positive but π fell, check for profit-taking on good news or a rule twist in an official document. Divergence is a research assignment, not a signal to double size.
Core concepts to remember
Tier sources before scoring. Impact on rules, not mood. Compare S to Δπ. Cascades need checklists. Social follows capital, not noise alone.
Polls in the stack
Polls are not sentiment—they are structured samples with known error. Use them in the prior step of mispricing audits, not as tier-three social. When polls and π diverge, ask whether the market prices turnout, timing, or question wording differently than the survey.
What comes next in this module
Sentiment is the last input before the mispricing capstone. Run tier discipline before sizing; let p′ face executable prices, not viral certainty. Superforecasting modules later sharpen the outside view you feed into p′ today.
What comes next
Signals, consensus, macro context, and sentiment converge in a single mispricing audit—the capstone of this module.
End each week with three rows: event, sign of S, sign of Δπ—agreement rate is your skill score.
Next: Identifying Mispriced Markets: A Step-by-Step Framework