The fingerprint
of bad decisions.
Most platforms track what traders do. We track why they do it. These 12 biases are not abstract psychology — they are measurable, repeating signals embedded in real trade data.
You hold onto a failing investment for years, unable to sell because selling makes the loss 'real'. The position is down 40%, but closing it feels like admitting defeat.
Traders systematically hold losing positions 2× longer than winning ones. They exit profits early to 'lock in gains' while losses run indefinitely.
BiasBro computes each trader's asymmetric hold-time coefficient — the ratio of average loss-hold duration to average win-hold duration across all closed positions.
You buy at the top of a crypto rally because every headline and group chat is celebrating gains. You weren't planning to buy — the crowd pulled you in.
Late-entry surges in instrument participation that precede sharp reversals. The classic bull trap: retail enters at peak enthusiasm, smart money exits into that demand.
CrowdPulse flags instruments with abnormal participation velocity — the rate at which new traders open positions. Extreme velocity is a contrarian signal.
You sell the shares that have doubled while holding the ones that have halved, hoping they'll 'come back'. You've locked in small gains and large losses.
Retail traders close profitable positions 47% faster than unprofitable ones on average. This systematic pattern destroys long-term returns compounding.
alex_debias constructs positions counter to the crowd's disposition bias — extracting the premium embedded in their systematic error at scale.
You sell your entire portfolio when markets drop 5%, crystalising losses at the worst possible moment — only to watch everything recover over the following weeks.
Coordinated retail exits trigger stop-loss cascades that overshoot fundamental value by a predictable margin. A measurable, repeating phenomenon.
BiasBro identifies traders entering 'panic state' by detecting sequential close-event clustering — sudden multi-position exits within compressed time windows.
A jacket marked down from £200 to £120 feels like a bargain, even if £120 is still overpriced. The anchor — £200 — rewired your value reference point.
Traders anchor to their entry price. Decisions are made relative to that number rather than current fundamentals — 'I'll sell when I'm back to breakeven.'
Feature engineering captures the gap between trader entry-price anchors and market-justified valuations — a measure of rational detachment from sunk-cost thinking.
After a near-miss car accident, you suddenly drive 10mph under the limit for a month. The accident rate hasn't changed — your perception of it has.
After a market crash, traders slash exposure and miss the subsequent recovery. After a bull run, they over-extend into the next correction.
BiasBro tracks trailing allocation shifts relative to instrument-level volatility regime changes, isolating recency-driven over-reaction from genuine risk management.
You confidently book a restaurant based on one good review, certain the experience will match your expectations. You ignore the base rate of disappointing meals.
Retail traders underestimate risk in familiar instruments and over-concentrate beyond what their actual track record justifies. Self-assessed skill exceeds measured skill by a wide margin.
Calibration scoring compares implied confidence — measured by position size — against historical accuracy at equivalent conviction levels. The gap is the overconfidence premium.
After five reds at roulette, you bet heavily on black — 'it's due.' The roulette wheel carries no record of its past. Each spin is independent.
After five consecutive losing trades, traders increase position size expecting the pattern to reverse. Sequential outcomes are treated as dependent when they are not.
Sequential outcome analysis flags traders exhibiting reversal-expectation behaviour after loss streaks — a characteristic increase in stake size following consecutive losses.
You join a long queue outside a restaurant you've never heard of, assuming the crowd knows something you don't. You wait 45 minutes for a mediocre meal.
Instrument participation spikes sharply before major reversals as the final retail participants pile in. This is the signal, not noise — the herd is the exit indicator.
You choose the hotel with 2,000 reviews over the one with 15, even though the second hotel opened last month and the reviews are more recent and detailed.
Popular eToro copy-traders attract disproportionate copy volume even when recent verified performance has deteriorated significantly. Popularity decouples from alpha.
You accept lifestyle advice from a doctor without questioning it, despite medicine and nutrition being distinct disciplines with contested evidence bases.
Traders copy self-labelled 'professional investors' whose independently verified returns do not support the credential claim. The label does the work; the performance doesn't need to.
You allocate money to a fund manager who has beaten the market three years running, ignoring that hundreds of managers tried and failed — and the three-year sample is too small to distinguish skill from chance.
Traders chase recently outperforming peers during streaks that empirically do not persist. The hottest trader of last quarter is frequently not the hottest trader next quarter.
Sharpe-normalised attribution separates skill from luck across rolling time windows of varying lengths. Genuine edge shows consistency across regimes — hot streaks do not.
From pattern to signal.
Each bias above maps to one or more engineered features in the ML pipeline. Here's the journey from raw trade event to quantified behavioural fingerprint.
Open/close timestamps, entry price, allocated capital, realised return. 81,056 traders. 5+ years of history.
BiasBro reconstructs full capital history — handling deposits, withdrawals, leverage, concurrent positions with state-machine fidelity.
335 behavioural and financial features. Loss-aversion coefficients, calibration scores, sequential outcome analysis, disposition scoring.
Per-trader behavioural profile: 12 bias scores, risk-adjusted performance, regime consistency. The fingerprint that distinguishes skill from luck.
See bias signals in action.
CrowdPulse surfaces real-time crowd behaviour across 15,497 instruments.