Survivorship Bias

#Definition

Survivorship bias is the logical error of concentrating on the people or things that "survived" a selection process while overlooking those that didn't. In prediction markets, survivorship bias distorts evaluation of trading strategies, forecaster track records, and market performance by focusing only on winners and ignoring the many failures that never became visible.

This bias leads to systematically overestimating success rates and adopting strategies that appear profitable only because failures have been hidden from view.

#Why It Matters in Prediction Markets

Survivorship bias affects almost every aspect of prediction market analysis:

Strategy evaluation: Published trading strategies are disproportionately winners. The many failed strategies never get written about, making successful approaches seem more common than they are.

Forecaster track records: Famous forecasters are famous because they made correct calls. The many forecasters who made identical predictions but happened to be wrong are forgotten, inflating apparent skill.

Platform success stories: Prediction market platforms highlight big winners. The many participants who lost money receive no attention, distorting perception of typical outcomes.

Historical backtesting: Strategies tested only on surviving markets (those that had enough liquidity and interest) ignore markets that failed or were delisted, biasing backtest results.

Self-assessment: Traders remember their wins more readily than losses, creating an inflated sense of personal skill.

#Quantified Impact: Research Findings

Academic research has quantified how severe survivorship bias can be:

During the 2008-2009 financial crisis, hedge fund attrition rose to 31%, with survivorship bias ranging from 1.68% to 6.48% depending on methodology.

The implication for prediction markets: any analysis using only "successful" traders, strategies, or markets significantly overstates expected performance.

#How It Works

#The Basic Mechanism

1. Large initial population: Many traders, strategies, or predictions start
2. Selection process: Some succeed, many fail
3. Visibility filter: Only successes receive attention, publication, or memory
4. Biased conclusion: Observer sees only successes and concludes success is common

#The Classic Example

Imagine 1,000 prediction market traders each making random 50/50 bets:

• After 5 correct predictions in a row: ~31 traders (by chance alone)
• After 10 correct predictions: ~1 trader

That one trader looks like a genius. They get interviewed. They write about their "system." But their success was statistically inevitable given 1,000 starting participants—no skill required.

#Simulation: The Hidden Returns

import random

def simulate_survivorship(n_traders=1000):
    traders = [{'id': i, 'return': 0.0, 'active': True} for i in range(n_traders)]
    
    # Simulate one year of trading
    for _ in range(12): 
        for t in traders:
            if not t['active']: continue
            
            monthly_ret = random.gauss(0.01, 0.05) # Avg 1% gain, 5% volatility
            t['return'] += monthly_ret
            
            # Drop out if return drops below -20% (Ruin)
            if t['return'] < -0.20:
                t['active'] = False

    # Calculate averages
    survivors = [t['return'] for t in traders if t['active']]
    all_traders = [t['return'] for t in traders]
    
    print(f"Avg Return (Survivors only): {sum(survivors)/len(survivors):.2%}")
    print(f"Avg Return (True Population): {sum(all_traders)/len(all_traders):.2%}")
    # Result: Survivors might show +15%, while True Population is +1%

#Numerical Example: Strategy Publication

Scenario: 100 traders independently develop trading strategies

Actual results:

• 15 strategies are profitable (some by skill, some by luck)
• 85 strategies lose money
• Only profitable strategies get published or discussed

What the public sees:

• 15 "successful" strategies
• 100% of visible strategies appear to work

Reality:

• Only 15% of strategies actually worked
• Many "successful" strategies may have been lucky, not skilled

#Survivorship Bias in Track Records

The forecaster selection problem:

Before an event:

• 100 forecasters make predictions
• 50 predict Outcome A, 50 predict Outcome B

After the event (suppose A wins):

• 50 forecasters were "right"
• Media interviews the most confident ones
• These forecasters build reputations as experts
• The 50 who were wrong fade into obscurity

The "experts" appear skilled, but the selection process guaranteed some would appear successful regardless of actual forecasting ability.

#Examples

"Top trader" profiles: A prediction market platform profiles its most successful traders. These traders share strategies that seem insightful. But for every profiled winner, dozens of traders using similar strategies lost money. The platform doesn't profile losers, creating the illusion that the winners' approaches are reliably successful.

Backtesting on active markets: A trader backtests a strategy using historical data from currently active prediction markets. The backtest looks great. But markets that failed, were manipulated, or had resolution disputes aren't in the dataset. The strategy may have performed poorly on those markets—but they're invisible.

Forecaster fame: A political analyst correctly predicted three consecutive election upsets. They become a media fixture. But hundreds of analysts made similar predictions on at least one upset and were wrong on others. The one who happened to get all three right appears skilled; the rest are forgotten.

Strategy forums: Online forums for prediction market strategies overflow with posts about winning approaches. Losing strategies aren't posted (or are quickly forgotten). New traders read the forums and assume most strategies work, when actually most shared strategies represent survivor selection.

Personal memory: A trader recalls their brilliant calls—buying an underdog at $0.15 before they won. They forget the five similar "brilliant" calls where the underdog lost. Their self-assessed skill level far exceeds their actual performance.

#Risks and Common Mistakes

Copying "winning" strategies: Strategies that appear successful may have survived by luck. Copying them doesn't transfer skill that may not exist.

Trusting single forecasters: A forecaster's track record may reflect survivorship selection rather than skill. Even skilled forecasters benefit from some survivorship bias in their visibility.

Underestimating failure rates: Seeing only survivors makes success look achievable. Actual failure rates in prediction market trading are much higher than visible evidence suggests.

Ignoring selection effects in data: Historical analysis using only surviving markets, active traders, or resolved events misses the full picture.

Overconfidence from selective memory: Remembering wins and forgetting losses creates false confidence in one's abilities.

#How to Counter Survivorship Bias

Seek failure data actively: Ask "what happened to the failures?" Look for information about strategies that didn't work, forecasters who were wrong, and traders who lost money.

Consider the base rate: Before being impressed by a success, estimate how many attempts occurred. If 1,000 people tried and 10 succeeded, that's 1% success—not impressive even if those 10 look amazing.

Track your full record: Keep detailed logs of all trades, including losses. Regularly review complete performance, not just highlights.

Apply "pre-registration" thinking: Decide criteria for success before seeing results. This prevents post-hoc selection of successful subsets.

Look for contradicting evidence: Actively search for examples of strategies that failed, forecasters who were wrong, and times when apparent patterns didn't hold.

#Practical Tips for Traders

• Maintain complete trading logs: Record every trade, win or loss. Review full performance regularly, not just memorable successes.

• Be skeptical of published strategies: Assume published strategies are survivors of a selection process. Demand out-of-sample evidence.

• Discount impressive track records: When evaluating forecasters, consider how many forecasters existed making similar predictions. Solo success from a large field is less impressive.

• Test strategies on failed markets too: When backtesting, include markets that had problems—low liquidity, disputed resolutions, or platform failures.

• Ask "who didn't survive?": When seeing success stories, explicitly consider the failures that aren't visible.

• Use prospective tracking: Start tracking a strategy or forecaster from today forward rather than relying on historical claims.

• Apply probabilistic thinking: Even genuine skill produces some failures. A track record with no losses is suspicious—it may reflect selective reporting rather than perfect execution.

#Related Terms

• Hindsight Bias
• Heuristics
• Expected Value (EV)
• Risk Management
• Market Efficiency
• Drawdown
• Kelly Criterion

#FAQ

#What is survivorship bias in simple terms?

Survivorship bias is when you only see the winners and forget about the losers. In prediction markets, you hear about traders who made fortunes, strategies that worked, and forecasters who called upsets correctly. You don't hear about the many traders who lost money, strategies that failed, or forecasters who were wrong. This makes success look more common than it actually is.

#How does survivorship bias affect strategy evaluation?

Strategies that get published, shared, or discussed are disproportionately successful ones. Failed strategies are forgotten or never shared. This means the "strategy universe" you see is heavily biased toward survivors. A new trader reading about "proven" strategies sees a distorted picture—the success rate among visible strategies is much higher than the success rate among all strategies ever attempted.

#Can survivorship bias make someone look skilled when they're not?

Absolutely. If 1,000 people make random predictions, some will be right multiple times by pure chance. The ones who happen to be right become visible—interviewed, followed, published. They look skilled, but their success was statistically inevitable given the large starting population. This doesn't mean no one has skill, but it means apparent skill must exceed what random chance would produce from a large field.

#How do I know if a forecaster's track record reflects skill or survivorship?

Consider: How many forecasters made similar predictions? If many people predicted the same thing and one happened to be consistently right, that's less impressive than if one person was uniquely correct against consensus. Look for predictions that were specific, contrarian, and correct—these are harder to achieve by luck. Also check if the track record is verified or self-reported.

#What's the relationship between survivorship bias and hindsight bias?

They often work together. Survivorship bias selects who gets attention (the winners). Hindsight bias then makes their success seem inevitable in retrospect. Together, they create narratives where the winners' success appears obviously predictable, ignoring both the many losers and the genuine uncertainty that existed before outcomes were known.