Ownership distorts exits. Recency bias attacks the timing of belief: you overweight the latest poll, print, tweet, or price tick—and underweight base rates, full dossiers, and your own pre-news probability. Trader psychology catalogs previewed recency; here we connect measurement, habits, and explicit links to news jumps, spikes, and revision schedules.
Recency is why superforecasters look boring: they update on evidence calendars, not push notifications.
What you overweight versus underweight
You overweight the last poll, the last CPI print, the last cent move, the viral clip, and your friend’s latest win story. You underweight poll averages and margin of error, twelve-month macro regimes, spread noise, resolution text, and your calibration bins.
Availability overlaps recency—treat “I just saw” as a red flag, not a likelihood.
Why markets are recency machines
Twenty-four-hour books always offer a “latest” price. News tabs invite immediate orders. Social velocity rewards surprise. Political cadence serves weekly overreaction. Play-money leaderboards praise last-week hits.
Crowd wisdom needs integrated information; recency traders price the last tweet. Superforecasters earn reputation by being dull on Twitter: same probability for days, then a crisp move when tier-one data arrives.
The twitch versus the dossier
If your dossier is thin, every twitch feels like information. Building dossiers for recurring event classes—FOMC, jobs, debates, court rulings—is how you buy time against recency. The dossier is the slow memory; the tape is the fast noise.
Recency versus legitimate Bayes
Poll inside margin of error should nudge probability a few points, not eight. Price up three cents with no news is noise—no update. Jobs beat by three tenths of a sigma needs a pre-written likelihood table, not a chase. Opponent gaffe video needs persistence base rates, not a ten-cent narrative bump.
Gate: if you cannot name the likelihood change, the move is recency—not updating. Write the likelihood sentence in the journal before you click; empty fields after a trade are a confession of narrative trading.
Post-poll chase (worked example)
State race YES near 52¢; new poll shows party plus four with margin of error plus or minus three; price jumps to 58¢ in twenty minutes.
Recency path market-buys at the top. Disciplined path writes old probability, models historical poll impact near two to three cents, lands near 56%, sees thin edge, uses limit or passes. Crowd may still be wrong at 58¢—edge needs math, not freshness.
Macro whiplash (worked example)
“CPI YoY below 3% next release” YES held at 44¢; hot CPI sends price to 31¢.
Release-day panic sell at the bottom. Next day only bear blogs. By day three disciplined review might show 38% versus 28¢—small buy if edge survives fees. By day seven scheduled review remembers both prints; recency forgot the hot one.
Macro dossiers exist so one print does not erase twelve months of base rate.
Spikes and sentiment tiers
At T+0 after a spike, buying the wick is classic recency; thirty-minute cooling rooms exist for a reason. At T+24h, treating the move as permanent ignores mean reversion history for the event class. At T+7d, ignoring the spike in the journal wastes learning.
Separate information jumps from noise. Tier-zero and tier-one sources may move probability; tier-three alerts may wake you but not move belief without corroboration.
Cross-venue recency
Polymarket moved while Kalshi flat does not mean Polymarket “knows”—depth and rules differ first. All venues plus five cents might be correlated flow, not truth. Stale venue lag is often fees and withdrawal friction, not a gift.
Build consensus with weights—not whichever app you opened last.
Shadow forecast beats chase (worked example)
Cabinet-pick rumor lifts price 47¢ to 54¢ without official naming rule trigger.
Market buy at 54¢ pays rumor plus resolution risk. Shadow log 51% without trade collects calibration data. Limit at 50¢ is patient. Rumor fails; price 49¢. Chasers paid tuition; shadow upgraded skill without spread tax.
Recency firewall (daily habits)
Morning dossier before apps. Disable price alerts inside margin-of-error bands. For each headline, write whether likelihood changed and why. No market orders first thirty minutes after tier-zero releases. Probability moves above five cents need two independent source tiers. Evening scheduled probability update—compare to market once. Friday list trades tagged “latest”; shrink next week if more than two. Shadow forecasts that never become trades are tuition you can afford—recency chases that become trades are tuition you pay twice after fees.
Notifications as traps
Price alerts inside noise bands train recency. Alert only on thresholds tied to your pre-written scenarios, not on every cent. Mobile-first products intensify the loop; disable pushes for positions you are holding on a thesis measured in weeks.
Event-week calendar (elections, FOMC)
Seven days out: pre-write likelihood table for scenarios. One day out: reduce size; widen limits. T-zero: observe spread; log mids; no hero trades. Plus one hour: first allowed update pass. Plus one day: compare move to historical class. At resolve: score process, not P&L story.
Pairings in Module 13
Confirmation makes latest news fit your side. Anchoring makes latest price the new sticky level. Overconfidence claims you react faster than the crowd. Loss aversion panic-sells the latest dip on your lot. Endowment makes the latest dip on your contract feel unique.
Streak psychology
Three winning weeks make the next headline feel like genius; three losing weeks make the same headline feel like doom. Streaks are recency stacked on P&L color. Scorer day reviews process tags, not streak length.
Mean reversion fantasies
Recency after a spike sometimes pairs with gambler’s fallacy—“it moved five cents, it must snap back.” Sometimes it mean-reverts; sometimes it jumps again on real news. Event-class history beats gut snap-back stories.
Long-dated contracts
Months-long holds amplify recency because the latest headline is the only thing that changed this week. Long-dated discipline means fewer touches: scheduled reviews, not daily narrative reinvention.
What comes next
Recency worships the last signal; hindsight rewrites memory after the fact so outcomes feel inevitable and journals lie. The next chapter keeps Brier scores honest once contracts pay zero or one.
Next: Hindsight Bias: "I Knew It All Along"