Foundations compared polls, experts, and markets. Habits chapters installed disciplined workflows. A stubborn pattern remains: famous experts often lose to aggregated crowds and superforecasters on testable questions—while still being invaluable for context, definitions, and process. Philip Tetlock's hedgehog versus fox lens is how serious traders weight punditry without betting the bankroll on aura.
This is not anti-expert. It is pro-calibration: steal drivers, not conclusions.
Hedgehog and fox
The hedgehog knows one big thing—single framework, bold thesis, memorable book title. The tell is "this time is different because…" applied to every cycle.
The fox knows many things—eclectic models, granular probabilities, public revision. The tell is "base rate X, factors A and B move me to about fifty-eight percent," with explicit invalidation.
Outside view is a fox move. Single-driver sermons are hedgehog risk inside your own head.
Why experts fail on forecasting leaderboards
Reputational inertia slows public revision while market price moves. Verbal "likely" does not map to Kelly sizing. Accountability asymmetry lets wrong pundits keep booking; traders lose real money. Domain overreach puts economists on virology contracts. Narrative coherence beats base rate. Entertainment incentives reward certainty, not Brier.
Good Judgment findings: aggregated fox-like amateurs and trained teams often beat average subject-matter hedgehogs on structured questions—not on every task needing classified context or deep process reading.
| Failure mode | Expert panel | Prediction market |
|---|---|---|
| Slow revision | Quote lags price | You pay now |
| Verbal buckets | "Likely" | Needs cents |
| Accountability | Reputation | P&L |
Worked contrast: re-election pundit
Event: incumbent wins re-election YES.
Hedgehog pundit: "structural realignment—lock YES," verbally ~ninety percent. Poll average maybe mid-fifties. Market fifty-seven cents. Your outside view fifty-four percent, inside bump fifty-eight percent.
Buying YES at sixty-eight cents because of pundit certainty pays for hedgehog theater. Fox plan: edge only if belief minus price clears costs after calibration shrink—here roughly pass.
When expert input is gold
Use experts for scenario drivers feeding Bayes, data definitions for resolution disputes, historical analogies for base rates, process knowledge for oracle edge cases. Do not use them for final p without numbers, or for marquee liquid prices already embedding their public views.
Steal the drivers, not the headline call.
Panels, markets, and superforecasters
Expert panels offer depth and slow context. Markets offer continuous price and skin in game—with thin-rule and manipulation risks. Superforecaster teams offer calibration and aggregation—may miss true black swans. Polls offer sampling transparency without penalty for error.
Liquid election marquees: consensus plus your audited belief. Novel regulation: expert drivers, wide uncertainty, often no trade. Macro prints: futures path plus PM. Niche props: expert process; frequently no trade.
Cascade plus lagging quotes
Contract: agency approves drug by Q3.
Monday panel says "likely" (~seventy percent verbal); market fifty-two cents. Tuesday rumor spikes seventy-one cents on volume; panel quotes unchanged Wednesday.
Fox read: what likelihood justifies seventy-one cents without tier-zero FDA document? If your decomposed belief is fifty-eight percent, do not chase; consider fade if economics allow. Hedgehog retail buys because Monday sounded confident.
Sentiment and expert media
Headline sentiment confirms priors—tag tier three. TV certainty anchors p before blind estimate—ban until p₀ logged. Twitter consensus is herding. Market memes are not edge.
One paragraph steel-man NO for every expert YES sermon you almost believed.
Hedgehog traps in prediction-market culture
Ideology tickets always YES on tribe. Model worship on one poll aggregator. Whale worship without audited belief. "Smart money" stories without economics. Expert halo where CV beats calibration.
Inside your own head
The hedgehog you fight is often yourself: one party, one macro story, one whale to copy. Fox discipline is private aggregation before public certainty. When you feel "slam dunk," that is a prompt to run dragonfly and check bins, not to max size.
Contrarian trading versus fox discipline
Betting against consensus is not fox-like by itself. Fox discipline requires a numbered f, a reason price is wrong, and economics that clear costs. Contrarian identity without calibration is a hedgehog in reverse—still one big story, just inverted.
Expert intake filter
Did they state numeric p? What would change their mind? Is it their domain? Did price move before the quote? Do your bins beat them on this class—if not, shadow only.
Watch probabilistic analysts who revise—they are driver mines, not obedience targets.
Narrative fallacy in your feed
The hedgehog inside you wants one cause: "turnout only," "inflation only," "the judge only." Fox discipline keeps parallel partial explanations alive until aggregation. When you close a trade, note whether you were fox or hedgehog in process—not whether you won. Process labels predict next quarter's Brier better than one outcome.
What comes next
Fox discipline is learnable in part. The closing chapter asks what evidence says about training, ceilings, and the handoff to decentralized markets—where honest event probability is necessary but not sufficient for getting paid on-chain.
Key ideas to carry forward
Experts supply drivers; you supply scored p. Hedgehogs sell certainty; foxes sell updates. Cascades plus lagging quotes trap chasers. Steel-man NO.
Next: Can Superforecasting Be Learned? Evidence and Methods