Survivorship Bias in Crypto Datasets
Assets delist, venues close, and tickers disappear. Research built only on what survives systematically overstates returns and understates risk.
Survivorship bias occurs when analysis uses only entities that still exist at the end of the sample. In crypto the problem is acute: many tokens that once traded actively have since collapsed or been delisted, and several venues that once held significant market share no longer operate.
How it distorts results
- Cross-sectional strategies appear more profitable because losers that went to zero are missing.
- Drawdown and default risk are understated.
- Venue-level liquidity histories omit platforms that failed, often abruptly.
- Index backtests look smoother than any index that was investable in real time.
Point-in-time data
The remedy is a point-in-time universe: for every historical date, the set of instruments and venues that were actually available on that date, with their data retained after delisting. Symbol maps must also be point-in-time, since tickers are sometimes reused for different assets.
This publication is provided for informational purposes only and does not constitute investment, legal, or tax advice, or an offer or solicitation to buy or sell any asset. Live figures are computed from third-party public market data and may be delayed, incomplete, or inaccurate.
