The Psychology of Prediction Markets: Why Smart People Get Prices Wrong

Prediction markets are fascinating because they expose our biases mathematically. A belief becomes a number with real money attached. The cognitive errors that humans share create exploitable patterns.

Why Prediction Markets Work (When They Do)

The wisdom of crowds is real — under specific conditions:

• Participants must have diverse information
• Participants must have real incentives (money)
• Aggregation must be independent (opinions can't cascade)

When these conditions hold, prediction market prices are extraordinarily accurate. The 2024 US election saw Polymarket prices track outcomes more accurately than every major poll.

The 5 Biases That Create Tradeable Edges

1. Recency Bias

Traders overweight recent events. After a dramatic event (election upset, market crash), they assign too much probability to similar events happening again soon.

The trade: If a 10% probability event just happened, traders often reprice the next similar event at 20-30% even though the base rate is still 10%.

Example: After a surprise Fed rate hike, markets overprice subsequent surprise hikes for the next 1-2 months.

2. Round Number Anchoring

People anchor to round numbers. A market trading at "50%" (i.e., even odds) has psychological gravity — people resist pushing past it even when evidence supports it.

The trade: Asymmetric evidence against the 50-50 anchor creates mispricing. If you have strong evidence for 70%, but market is at 55%, that's edge.

3. Scope Insensitivity

Humans are bad at distinguishing between 1%, 3%, and 5% probabilities. We mentally round them all to "unlikely." This creates systematic mispricing of tail events.

The trade: Rare but calculable events (a company missing earnings by >20%) are often underpriced because traders can't viscerally feel the difference between 2% and 5%.

4. Status Quo Bias

Traders underweight the probability of change. Markets often underprice regime changes, policy reversals, and paradigm shifts — because "things usually stay the same."

The trade: When structural change is clearly underway (regulatory shifts, technological disruption), markets are often slow to price the full probability.

5. Narrative Bias

A compelling story makes an outcome feel more probable. Vivid, specific scenarios feel likely; abstract, vague outcomes don't — even when base rates are identical.

The trade: Outcomes with boring narratives are often underpriced. Outcomes with exciting narratives are often overpriced.

How to Actually Find Edge

Step 1: Find your reference class What's the historical base rate for this type of event? If you're betting on "will Company X beat earnings," the base rate for S&P 500 companies beating earnings is ~75%. Start there.

Step 2: Apply specific adjustments What unique factors push above or below the base rate for this specific case? Be specific and quantified.

Step 3: Compare to market price If your estimate is 70% and the market is 55%, you have potential edge. The larger the gap, the better.

Step 4: Size appropriately Kelly criterion: fraction = (p * (b+1) - 1) / b where p = your probability, b = odds.

def kelly_bet(your_prob: float, market_prob: float, max_bet_pct: float = 0.05) -> float:
    """
    Calculate Kelly optimal bet size.
    market_prob: Current market price (e.g., 0.55 = 55 cents)
    your_prob: Your estimated true probability
    max_bet_pct: Cap at 5% of bankroll (fractional Kelly)
    """
    b = (1 - market_prob) / market_prob  # Decimal odds
    p = your_prob
    q = 1 - p
    
    kelly = (p * b - q) / b
    kelly_half = kelly / 2  # Half-Kelly is more conservative
    
    return min(kelly_half, max_bet_pct)

# Example: Market prices outcome at 40% (you think it's really 60%)
bet = kelly_bet(0.60, 0.40)
print(f"Optimal bet size: {bet:.1%} of bankroll")  # ~16.7%
print(f"Capped at 5% per half-Kelly: {min(bet/2, 0.05):.1%}")

The Information Edge

In prediction markets, the most durable edge comes from information others don't have, not from being "smarter" about processing public information.

Sources of genuine information edge:

• Domain expertise (a doctor predicting health policy)
• Network access (knowing industry insiders)
• Data analysis (processing public data better than market participants)
• Speed (knowing results before they propagate)

Traps disguised as edge:

• Being very confident based on narrative
• Over-indexing on recent data
• Using public polls that the market already incorporated

The best prediction market traders treat it like a second-order game: not "what do I think will happen?" but "where is the market wrong, and why?"