Heurística da Representatividade — O Bias de Pattern Matching

A Heurística da Representatividade (Representativeness Heuristic) é a tendência de julgar probabilidade pela similarity a um caso prototypical — “isso parece X, então provavelmente é X”. Em trading, isso significa: “Essa pattern parece um head-and-shoulders, então provavelmente é um head-and-shoulders” — ignoring base rates, sample size e alternative explanations. O resultado: overestimation de pattern reliability, false pattern recognition e systematic errors em probability assessment.

O Experimento de Kahneman e Tversky

O experimento clássico demonstra o bias:

Description de Steve: “Steve é muito shy e introverted, sempre helpful, mas tem pouco interesse em people ou world affairs. É meek e tidy, needs order e structure.”

Pergunta: Steve é mais provavelmente um (A) librarian ou (B) farmer?

Majority answer: Librarian — porque description “represents” stereotype de librarian.

Correct answer: Farmer — porque há 20x mais farmers que librarians (base rate).

Representativeness bias: similarity ao stereotype de librarian overrides base rate de 20:1 farmers:librarians. O “looks like” dominates o “statistically likely”.

Manifestações em Trading Cripto

1. False Pattern Recognition

Traders frequentemente “see” patterns que não são realmente there:

• “Essa formation looks como um triangle — logo vai breakout”
• “Essas 3 candles looks como morning star — reversão incoming”
• Similarity ao prototypical pattern → probability overestimated
• Reality: random price movements podem resemble任何 pattern — base rate de false patterns é high

2. Overconfidence em Technical Patterns

Patterns técnicos são treated como se fossem predictive com alta probabilidade:

• “Head-and-shoulders = reversão confirmada” → probabilidade real: ~55-65%, não 90%
• “Double bottom = reversão guaranteed” → probabilidade real: ~60%
• Representativeness faz pattern look “perfect” → probability perceived como muito higher que real

3. Trend Following Bias

Em uptrend, traders overestimate continuation:

• “BTC subiu 5 dias consecutivos — uptrend confirmed”
• Representativeness: 5 dias green “represents” uptrend → continuation likely
• Base rate: após 5 dias consecutive green, probability de continuation ≈ 50-55%, não 80-90%
• Similarity ao “strong uptrend” → probability overestimated

4. Recency como Representativeness

Eventos recentes são seen como mais “representative”:

• “BTC crashed 30% last week — market está bearish”
• Representativeness: crash recente “represents” bear market → bearish likely
• Base rate: single crash não necessarily indica bear regime → random volatility possible
• Recent events seem mais representative → overweighted em probability assessment

5. Altcoin Classification

Traders classify altcoins por representativeness:

• “Essa coin tem whitepaper como Ethereum — provavelmente será como Ethereum”
• Representativeness: similarity ao ETH prototype → success likely
• Base rate: > 90% de altcoins com “ETH-like” whitepapers fail → success probability very low
• Similarity ao success prototype ≠ probability de success

Base Rate Neglect

O core error de representativeness é neglect de base rates:

Trading Base Rates

• Win rate de random entries: ~50% (coin flip)
• Win rate de breakout patterns: ~55-65%
• Win rate de OB entries: ~60-70%
• Percentage de altcoins que succeed: < 5%
• Percentage de traders que profitable long-term: < 10%

Representativeness bias makes traders:

• Treat pattern como se win rate fosse 80-90% (overestimate)
• Treat altcoin como se success probability fosse 30-50% (overestimate)
• Ignore statistical reality em favor de “looks like”

Sample Size Neglect

Representativeness também ignores sample size:

• “Meus últimos 3 OB trades foram wins — OB é 100% reliable”
• 3 trades = tiny sample → result pode be 100% luck
• Representativeness: 3 wins “represent” perfect system → probability overestimated
• Adequate sample: 100+ trades → win rate meaningful → probability accurate

O Conjunction Fallacy

Representativeness leads ao conjunction fallacy — judging conjunction como more likely que either component alone:

Exemplo: “BTC vai fazer breakout acima de R$300.000 E continuar até R$350.000”

• Probability de breakout: P(A) = 60%
• Probability de continuation se breakout: P(B|A) = 55%
• Probability de both: P(A∩B) = 60% × 55% = 33%

Representativeness faz “breakout + continuation” seem mais likely que “breakout alone” — mathematical impossibility.

Em trading: “OB bullish + BOS + continuation” → conjunction probability é sempre ≤ single component probability.

Countermeasures

1. Base Rate Integration

Antes de qualquer probability assessment:

1. Identify base rate: qual é statistical probability deste outcome em general?
2. Adjust por specifics: como este caso differs do average?
3. Final estimate: base rate × specific adjustment → realistic probability

Exemplo:

• Base rate de breakout success: ~60%
• This breakout: volume confirming, OB nearby → +5%
• Final estimate: 65% (não 90%)

2. Pattern Probability Table

Maintain uma table de pattern probabilities baseada em backtesting data:

• Head-and-shoulders: 55-65% reversão
• Double bottom: 60% reversão
• OB bullish: 60-70% continuation
• Engulfing bullish: 55-65% reversão

Quando “see” pattern → check table → use base rate, não representativeness

3. Sample Size Check

Antes de drawing conclusions:

• Minimo 30 trades para初步 statistics
• Minimo 100 trades para meaningful win rate
• Minimo 300 trades para robust system validation
• Se sample < 30 → don’t generalize → result likely noise

4. Alternative Explanation Search

Quando pattern emerges, busca alternative explanations:

• “Essa formation looks como head-and-shoulders”
• Alternative 1: random noise resembling H&S
• Alternative 2: correction em uptrend (não reversão)
• Alternative 3: manipulation (whale activity)
• Weight alternatives por base rates → mais accurate assessment

5. Conjunction Probability Check

Para任何 multi-step prediction:

1. Estimate probability de each step independently
2. Multiply probabilities → conjunction probability
3. Compare com single-step probability → conjunction always ≤ single
4. Use conjunction probability como realistic estimate

6. Anti-Pattern Recognition

Quando see “perfect pattern”:

• “Is this pattern genuinely present, ou am I forcing data into a template?”
• “What percentage de random data would also ‘look like’ this pattern?”
• “If I remove one candle/bar, does the pattern still exist?”
• Anti-pattern removes representativeness overlay → raw data assessment

7. Journal de Pattern Accuracy

Track no journal:

• Pattern identified → outcome → win ou loss
• Over 50+ trades: calculate actual win rate per pattern type
• Compare actual win rate com perceived probability (representativeness estimate)
• Gap = representativeness bias magnitude → adjust future estimates

Conclusão

A Heurística da Representatividade é o pattern matching bias que faz traders confuse “looks like” com “probably is” — systematically overestimating probability de outcomes que resemble prototypical cases e neglecting base rates, sample size e alternative explanations. Em trading cripto, onde visual patterns são salient e stories são compelling, representativeness é constantemente activated. Countermeasures — base rate integration, pattern probability table, sample size check, alternative search e conjunction check — são analytical tools que override representativeness shortcut. O insight fundamental: probability é mathematical, não visual — “looks like” é feelings, “probably is” é statistics. Trading profitable requires后者, não前者.

Heurística da Representatividade