Survivorship bias
Survivorship bias is the error that occurs when you analyze only the instruments that “survived” to the end of your sample (or to today) and exclude instruments that were delisted, became inactive, or otherwise dropped out. In trading research, it typically shows up when your historical universe is built from a current symbol list, so the past is missing markets that were actually tradable at the time. The result is an overly optimistic picture of returns, liquidity, and risk.
Why it matters
Backtests with survivorship bias tend to overstate returns, understate drawdowns, and overestimate liquidity because many of the worst outcomes are removed from the dataset. It also distorts ranking rules like “top N by volume” because the candidate set is incorrectly restricted to survivors. In crypto—where listings and delistings are frequent—the distortion can be large even over a few years.
How it works
Survivorship bias commonly happens when you:
- pull a symbol catalog “as of today” and then request history for that static list
- filter to symbols with long histories, unintentionally dropping symbols that ended early
- rely on datasets that do not preserve delisted or inactive markets
A robust fix is to use a point-in-time universe so membership is determined “as-of” each decision date.
How do you avoid survivorship bias in crypto backtests?
Build the universe point-in-time: for each date (or rebalance), include all symbols that were eligible and had coverage at that time, including those that later disappeared. Do not start from today’s symbol list and “work backward,” because that removes historical dropouts by construction. Finally, store the universe membership per date so the backtest can be reproduced and audited.
What is a survivorship-bias-free universe?
A survivorship-bias-free universe includes all instruments that met your eligibility rules at that time, even if some instruments later delist or go inactive. Practically, that means the universe is time-varying and built with “as-of” membership plus coverage-window checks. It also means your dataset retains historical symbols so that failures and attrition are represented.
Is survivorship bias the same as lookahead bias?
They are related but not identical. Survivorship bias is specifically about excluding “dropouts” (delisted/inactive symbols) from historical samples, while lookahead bias is any use of future information in past decisions. Many backtests suffer from both when they use today’s survivors and also compute rankings with forward-looking windows.
Real-world example
You backtest a 2018–2021 altcoin strategy using only symbols that are active today on major exchanges. Tokens that were listed in 2018 and later delisted (often after severe drawdowns) are missing, so the backtest never experiences the losses and liquidity collapse those markets would have caused. The strategy’s Sharpe ratio and max drawdown look materially better than what a real trader could have achieved.
How CoinAPI Market Data API helps
CoinAPI’s Market Data API provides symbol metadata via GET /v1/symbols, including coverage fields like data_start / data_end and, where available, type-specific windows such as data_trade_start / data_trade_end. Those fields help you keep ended markets in historical research and apply time-bounded eligibility rules so your universe better reflects what existed at each point in time. Combined with a stored “as-of” membership list per rebalance, this supports survivorship-bias-aware backtesting.