Survivorship Bias
Definition
Survivorship bias is the systematic error of evaluating trading systems by looking only at the surviving examples — the EAs still being marketed, the vendors still in business, the backtests that didn't blow up — while ignoring the larger population of failed ones that disappeared from view. The bias makes EA evaluation systematically optimistic.
In-depth: Survivorship Bias
Survivorship bias is one of the most pervasive and underrecognised errors in trading-system evaluation. The mechanism is structural: the EA market continuously sheds failures while keeping successes visible, so any snapshot of the current market is biased toward winners.
Concrete examples of survivorship bias in EA evaluation:
• **"Best EAs of the past 3 years" lists**: these aggregate the EAs that are still being marketed in 2026. EAs that were highly marketed in 2024 but have since been quietly withdrawn don't appear on the list. The remaining survivors' performance overstates the average performance of the EA universe as it existed in 2024. • **Vendor reputation scores**: vendors whose past EAs failed badly enough to damage their reputation often rebrand or exit the market; the vendor reputations visible in 2026 are biased toward those whose products performed well enough to maintain market presence. • **Backtest databases**: open-source EA libraries and marketplace listings preserve EAs whose backtests look attractive; EAs with poor backtests are deleted, not preserved as warnings. The aggregate quality of available backtests overstates the realistic backtest quality of newly-developed systems. • **"Top vendor strategies" promoted by signal-service platforms**: these are filtered to highlight currently-performing vendors; the same vendors had failed strategies in their portfolio that are not shown.
The scale of the bias: rough industry estimates put EA failure rates (defined as either vendor abandonment or 50%+ drawdown leading to account-blow-up among users) at 50-75% within 2 years of release. The buyer evaluating today's marketplace is seeing the 25-50% survivor cohort, not the full population. Average expected outcomes for newly-purchased EAs are materially worse than the average performance of currently-marketed EAs would suggest.
Mitigations for survivorship bias in buyer evaluation:
1. **Demand 12+ month verified live track records** (vendors with shorter track records may be filtering out their early failure modes) 2. **Apply vendor longevity heuristics** — vendors active 24+ months across multiple EAs have at least demonstrated business persistence 3. **Cross-reference vendor portfolio for retired EAs** — vendors with retired EAs you can examine provide useful failure-mode evidence; vendors who present only the current product hide that information 4. **Discount marketed performance metrics by 30-50%** when computing realistic expected outcomes — the realised future performance distribution is materially worse than the publicly-marketed survivor-bias-affected distribution 5. **Refuse "best EAs of all time" framings** — these are pure survivorship bias and have no predictive value for buyer decisions
In the broader trading literature, survivorship bias is the most-studied form of selection bias in performance evaluation. Academic research dating to Brown, Goetzmann, Ibbotson (1992) established that survivorship-corrected mutual-fund performance is materially lower than uncorrected performance. The same principle applies more starkly to the EA market because the failure-and-removal cycle is faster and more opaque than in regulated investment vehicles.