Confirmation bias makes you select evidence to defend a belief. Anchoring is different: the first number you see sticks, and later adjustments stay too close even when fundamentals moved. In prediction markets anchors are everywhere—opening print after a shock, your entry average, a friend’s “fair value,” last week’s poll, a viral screenshot of the current price.
This chapter extends anchoring from psychology catalogs and ties it to implied-probability literacy, cross-venue reading, and journals that separate entry from belief.
Definition and mechanism
Anchoring is over-reliance on an initial value when making judgments, even when that value is random or irrelevant.
Your entry price feels emotionally relevant but should not set probability—only forward expected value at today’s price matters. The first price you saw today is partial public information, not gospel. Round numbers like 50¢ magnetize “coin flip” thinking. Poll headlines measure a different question than your contract. Analyst targets are often inside-view stories.
Classic experiments used arbitrary wheels of fortune; you get the same effect at 2 a.m. watching a thin book rip five cents on rumor.
Anchoring versus neighbors
Anchoring sticks to levels; confirmation sticks to beliefs; loss aversion sticks to pain frames; endowment (later in this module) sticks to ownership. The gambler’s fallacy—“due for a bounce”—misreads streaks; anchoring misreads levels. Outside-view reference classes are the deliberate counter-anchor if you write them first. Write outside view before you open the trading app, not after you have already seen the mid.
Poll headlines versus contract questions
A poll about “generic ballot” is not the same contract as “wins state X.” Anchoring to a national headline while trading a local race is a category error dressed as data. When you cite a poll, quote the exact question text and date; if it does not match resolution, it cannot anchor your probability.
Three anchor classes for traders
Price anchors: you bought YES at 35¢; market is 52¢; you say “still cheap” instead of asking fresh probability at 52¢.
Probability anchors: you started at 50% and never moved despite tier-one news.
Narrative anchors: “it’s a coin flip forever” while structure shifted under you.
Entry price is not fair value (worked example)
Contract: “Bill passes committee by March 1” YES. You bought at 41¢ three weeks ago. Market is 63¢; thesis partially fulfilled but not resolved.
Anchored thinking: “I’m up; hold for a dollar.” Re-anchored thinking: close the book, run a fresh forecast, update outside pass rates, cap inside bumps, land near 58% against a 63¢ ask—sell marginal edge and free attention.
Entry 41¢ is sunk; decision is forward expected value at 63¢, not comfort from green P&L. Holding to “get even” at 41¢ is anchoring plus loss aversion, not edge.
Opening print on a news spike (worked example)
A court ruling hits at 10:00. YES opens 78¢ on a thin book that lifted stray bids from 41¢.
Trap: “It was forty-one; seventy-eight must be wrong” and fade without reading resolution. Process: log probability before and after shock, build outside view for this posture, maybe land near 71% against 74¢ two hours later—update beliefs, do not anchor to the pre-shock level.
Pre-register how event class usually moves markets; do not let the first print after shock be your only outside view.
The fifty-cent magnet
Binary contracts cluster near 50¢ when uncertain. Traders anchor to coin-flip language and under-update.
Before announcement, true base rate might be 12% while price sits at 50¢. After a leak, price still near 50¢ while likelihood tables say otherwise. Late favorites at 94¢ hide tail risk if you anchor to “only six percent upset.”
Calibration discipline: if you always say fifty-fifty, you are hiding inside the anchor instead of scoring bins honestly.
Re-anchoring before material decisions
Hide entry P&L in the journal view. Write outside probability from reference classes. List inside deltas with caps. Open the book; read executable price and depth. Compute edge versus what you can actually trade. Log an anchor note if belief moved more than eight points—if ten points without tier-one URL, delay twenty-four hours or get a second reviewer.
Anchoring across venues
Polymarket at 67¢ and Kalshi at 54¢ is not free money by default—fees and resolution text differ. PredictIt caps can leave stale anchors. Manifold play money at 80% is not a USD anchor.
For trade decisions, executable second venue matters; for belief, your model matters. Poll headlines are common anchors—steel-man what poll would make you drop your level.
Journal fields that help
Record probability estimated before seeing the market price. Record market price at decision. Record entry if any—but do not feed entry into revised probability. Self-score anchor risk one to five. Note whether you ran a full re-anchor pass.
Example row: Fed cut March YES—blind 34%, market 41¢, entry last week 38¢, re-anchored 36%, negative edge, exit because whip count weakened—not “take profit” language. The exit reason field should describe belief change, not P&L color.
Spikes without a model
After shocks, traders anchor to the pre-spike level and fade mechanically. Sometimes fade is correct; often it is nostalgia. Pair every spike reaction with resolution text and an event-class history: how did similar shocks propagate over the next twenty-four hours? If you lack that table, default to smaller size and limits, not conviction shorts.
Cross-venue anchoring
The first app you open becomes the anchor for the day. Rotate which venue you check first on neutral mornings, or build a weighted consensus before any trade. “Polymarket moved so Kalshi must follow” is anchoring plus narrative unless depth and rules align.
Red flags
“I’ll sell at entry” is a price-target anchor. Probability unchanged through major news is stale anchor. Justifying hold with hours already spent is sunk narrative. Round fifty percent for everything is lazy anchor. Fading a spike only because “it was lower” is level anchor without a model.
Take-profit language
Journal exits labeled “take profit” often hide anchor to entry, not belief change. Prefer “probability below price” or “thesis invalidated by X.” Wording trains post-mortems you will trust in Module 14 case studies.
What comes next
Anchoring freed you from sticky first prices. Overconfidence is stickier: you believe probability is accurate and narrow—edges look huge, sizing screams, and caps feel insulting. The next chapter makes miscalibration measurable.
Anchoring is subtle because it feels like memory, not bias. The cure is repetitive: blind estimate, outside view, then market—not the reverse.
Next: Overconfidence Effect: Why You Think You're Better Than You Are