Anchoring taught you to drop sticky first prices. Overconfidence is stickier: you believe your probability is accurate and narrow—edges look huge, sizing formulas scream, and bankroll caps feel insulting. Trader psychology catalogs already flagged overconfidence as an edge killer; here we make it measurable with calibration bins, Brier scores, and honest journals.
Overconfidence is not arrogance alone. It is miscalibration plus oversizing plus under-search disguised as conviction.
Two faces of overconfidence
Calibration overconfidence shows up when eighty-percent forecasts happen sixty percent of the time—reliability diagrams bend wrong. Resolution overconfidence claims tiny Brier spreads that overfit noise. Sizing overconfidence runs Kelly fractions that threaten ruin. Information overconfidence posts fifteen-cent edges on marquee contracts without private data.
Hedgehog forecasters marry one big story; foxes distribute uncertainty. Skill is learnable—overconfidence is what unlearns it fastest when you skip scoring.
Good confidence versus bad confidence
Good confidence is calibrated: when you say seventy percent, seven of ten resolve YES over many trials. Bad confidence is loud: tight language, huge size, thin evidence. The market rewards the first in the long run and taxes the second every fee day.
How biases stack
Hope inflates probability; confirmation defends it; overconfidence shouts “I’m eighty-five percent” while the crowd sits near fifty-five. Anchoring keeps the story glued to an old level. Tag stacks in weekly review—they rarely travel alone.
Why markets punish overconfidence twice
You eat Brier penalties on stated probability and P&L penalties on fees plus adverse selection. Capital traps in “lock” positions while opportunity cost compounds. Public leaderboards reward lucky months that hid bad bins.
You can win P&L while losing calibration—lucky sizing on a biased probability. Fix bins before you scale. Bragging rights from one lucky month are the fuel overconfidence loves—keep a rolling fifty-trade bin table visible when you size.
Words that hide wide uncertainty
“Lock,” “certain,” “free money,” and “can't lose” are linguistic narrow bins. Ban them from journals and group chats. Replace with ranges and invalidation sentences. Language trains the posterior whether you notice or not.
Marquee edge mirage (worked example)
“Party wins state X” YES trades near 46¢ on one liquid venue and 44¢ on another. You claim 72%.
Implied edge near twenty-six cents should trigger pause: two liquid books plus polls near 48% disagree with you. Likely truth: overconfidence plus confirmation. After outside view plus capped inside bumps, 54% might face 46¢—eight cents of edge, half-Kelly under a three-percent cap—not a hero size.
Thinking you are twenty cents smarter than aggregated price without tier-one private information is a classic illusion.
Calibration bin disaster (worked example)
Last twenty trades tagged seventy-to-eighty percent; only 55% resolved YES.
Response: shave eight points on new seventy-to-eighty calls until twenty clean forecasts; halt size increases; paper-trade a week; tag overconfidence in the journal. Expected-value steps should use adjusted probability, not raw bravado.
Kelly wants too much (worked example)
Bankroll $10,000. Probability 60%, price 45¢, fees ignored for teaching.
Edge per share about $0.15; full Kelly might suggest roughly 27% of bankroll; half-Kelly near 13.5%. A three-percent cap exists for a reason. Overconfidence sizes fifteen percent because “edge is huge.” One bad cycle plus correlated political book is a crisis.
Use quarter-Kelly on politicals, correlation heat limits, and caps as non-negotiable.
Dunning–Kruger in tournaments
Under fifty trades, extreme five and ninety-five percent calls are common. Between fifty and two hundred, tails still too wide. Veterans sometimes swing underconfident—that is rare and earned through bins, not vibes.
Top forecasters in long-running competitions are bold and calibrated—not parked at fifty percent forever.
Confidence hygiene habits
Pre-mortem: “I’m wrong because…” required. If edge exceeds twelve cents on liquid marquee, pause for third-party probability. Buddy independent estimate when possible. Shrink map toward market when Brier worsens. Ban words like lock, sure, and free money. Manifold drills at low stakes build bins without bankroll damage.
Structured forecast gates: estimate blind, compare to market—if gap exceeds fifteen cents, mandatory review, cap inside deltas, size through Kelly then cap then minimum, track Brier monthly.
Superforecaster shrink (worked example)
Raw 77% YES versus market 61¢; last quarter you were optimistic in seventy-to-ninety bins.
Shrink toward market might land 69% for expected value—still positive edge near eight cents but size near 2% bankroll, not ten percent at seventy-seven.
Red flags—stop sizing up
Edge above fifteen cents on liquid marquee without review. No losing trades logged this month—luck review time. Ninety-percent language—cap at eighty until bins heal. Skip steel-man—confirmation block. Brier worse than always betting market—shrink toward consensus temporarily.
Weekly overconfidence audit (twenty minutes)
Pull resolved trades with stated probability. Plot hit rate by bin. Compute mean forecast minus outcome for YES bets; if mean above plus five points, global shave three points. List three largest gaps at entry—how many won? Re-read overprecision examples from fallacy chapters. If your largest edges cluster in illiquid names, you are not a genius—you are measuring your own noise.
When the market is your tutor
Consensus price is not truth, but it is a harsh tutor. Asking “why am I fifteen points away?” is healthier than asking “why is everyone stupid?” Sometimes you have private information; usually you have a stacked bias. Default to shrink toward market until Brier proves you earn the wings.
Paper trading without bins
Paper P&L without calibration bins is fantasy sports. Run paper the same way as live: stated probability, locked at commit, scored after resolve. Otherwise paper only trains overconfidence because you never face the Brier sting.
Allocator and expert theater
Briefings sometimes reward confident tone over accurate ranges. If you consume expert media, date-stamp their claims and score them like your own. Experts can be overconfident too; markets are the scoreboard that does not care about credentials.
What comes next
Tight probability and size set up the exit failure mode. Loss aversion makes realizing red feel worse than green feels good—driving hold-losers, sell-winners, and toxic averaging down. The next chapter separates thesis stops from P&L color.
Next: Loss Aversion: Why Losses Hurt Twice as Much as Gains Feel Good