All lessons
Backtest crypto data honestly
A dataset can forget failed tokens or miss an exchange outage. Those missing pages can make a backtest look better than reality.
Build a historical simulation that respects what was tradable and knowable at the time.
A backtest applies written rules to historical data. Keep timestamps, venue, pair and price field explicit. Signals based on a candle's close cannot assume entry earlier within that candle; indicators must use only information available at each decision.
Include assets that failed or delisted when evaluating a historical universe. Selecting today's survivors can hide losses. Model fees, spread, slippage, partial fills and outages. Derivative simulations additionally need contract specifications, historical funding and liquidation rules; spot data alone cannot reproduce them.
Separate development data from an untouched later evaluation period. Trying many settings and keeping the best can fit noise. Record every version and compare sensitivity to costs and parameters. A promising historical result remains a hypothesis for forward testing, not proof of future earnings.
Let’s use numbers · hypothetical example
A paper strategy makes USD 30 gross across 20 round trips. At an assumed USD 2 total cost per round trip, costs are USD 40 and the result is −USD 10. Omitting costs reverses the conclusion.
Keep in mind
Even a detailed simulation cannot reproduce every live liquidity constraint or emotional decision.
Your turn
Audit one sample trade for signal time, earliest possible fill, fees and available data; then double the cost assumption.
A quick check before we move on
Why can testing only today's listed tokens mislead?
- Failed or delisted tokens may be excluded
- Today's list contains every historical failure
- Listing automatically proves long-term quality
Check answer
Failed or delisted tokens may be excluded Excluding assets that disappeared creates survivorship bias and can make a historical strategy look better than it was.
Sources and further reading
Content reviewed: 29 October 2026
Educational material with hypothetical examples. Trading can cause substantial losses. A course, pattern, stop order or past result cannot guarantee a profit or limit every loss. Practise in simulation and check current rules and product terms in your jurisdiction.