· 4 min read
Crypto Scanner Backtests: Avoid Survivorship Bias in the Coin List
Check whether a historical crypto scanner test used the assets available then or only today’s survivors. Follow a simple basket example and record universe changes.
By TickerAI
A crypto scanner backtest can look better when it studies only coins that remain available today. Assets that disappeared, changed listing, or no longer satisfy the current filter may be missing from the historical experiment.
The research question matters: testing today's chosen assets over old data is different from testing a rule that would have selected assets using information available at each earlier date. Label the first honestly rather than presenting it as the second.
A small example shows the selection effect
Assume five hypothetical assets each receive $100 at the start, with no rebalancing or costs. Their final returns are +20%, +10%, −5%, −50%, and −100%. The final value of the original $500 basket is $375, a 25% loss.
Now imagine a later dataset omits the last two assets. The remaining three appear to turn $300 into $325, a gain of approximately 8.33%.
| Sample shown | Initial value | Final value | Simple return |
|---|---|---|---|
| All five original assets | $500 | $375 | −25% |
| Only the three retained assets | $300 | $325 | +8.33% |
These are invented outcomes designed to show selection arithmetic. The −100% assumption represents total loss in this example; a real delisting does not automatically imply that exact economic recovery value.
Define the universe as a dated rule
Record which exchange, quote currency, listing status, minimum history, and liquidity condition determine eligibility. Specify when the list is evaluated and when changes take effect.
A rule using later-known market capitalization, eventual survival, or a current popularity list has information that may not have existed earlier. A dynamic rule needs historical inputs, not just today's endpoint response.
Binance's exchange information documentation describes symbol metadata. A current response should not be mistaken for a complete point-in-time archive of historical listings and eligibility.
Missing data needs an explicit treatment
Distinguish a failed download from an instrument not yet listed, a suspended market, a renamed asset, and genuinely missing observations. Silently dropping every failed symbol can change the sample systematically.
Record excluded symbols and reasons alongside successful ones. A minimum-history requirement can be valid if it is specified in advance, but it also changes which newly listed assets can enter the study.
Do not stitch different symbols into one price history without documenting conversions and economic continuity. A name change and a new token contract are not automatically the same research event.
Costs and implementation still need their own model
Correcting the coin list does not make historical prices executable. Include spread, fees, order-size constraints, and any data or venue limitations that affect the proposed trading rule.
A monthly universe update should also specify how existing positions are handled when an asset becomes ineligible. Exiting at the final convenient historical close can introduce another favorable assumption if that price was not realistically available.
Use sensitivity checks for unresolved cases rather than selecting the version that produces the largest return. Preserve the selection rule and rejected variants in the research record.
What TickerAI's scanner does today
TickerAI's crypto-majors preset is a defined list in the repository. It is a current research convenience, not a reconstructed historical membership series or a claim to cover every traded coin.
The scanner reports symbols that fail to load separately from successful results. Keep those failures in the review rather than describing the returned shortlist as a complete market survey.
A scan of current candles is also not a historical backtest. The current product does not expose a dedicated point-in-time universe reconstruction or automated replay engine for that purpose.
Evaluate the claim, not just the chart
The SEC's performance-claims bulletin cautions that backtested performance is hypothetical and highlights the problem of selectively presented results. A survivorship check addresses one specific selection issue, not every research bias.
Ask TickerAI to state the universe used, the scan date, failures, and any historical information it cannot verify. Continue with crypto watchlists and strategy-testing principles. A transparent sample makes a test interpretable; it does not establish future profitability.