Why Most Crypto Backtests Lie — and How to Test a Strategy Honestly

A strategy that doubles your money in a backtest and then bleeds in live trading is the most common story in algorithmic crypto trading. Almost always, the backtest was lying.

1. Data from one exchange

Candles from a single venue include its own wicks, outages and thin moments. A stop hit by a one-exchange wick would not have been hit on the market as a whole. Testing on volume aggregated across major exchanges is much closer to the price you would actually trade.

2. Fees, slippage and funding left out

A strategy trading three times a day pays about 0.1% per round trip in taker fees, plus slippage and funding every eight hours on perpetuals. Many edges are smaller than these costs. A backtest must subtract them on every trade.

3. Look-ahead bias

Using the close of a candle to enter on the same candle, or indicators that are recalculated with future data, makes results impossible to repeat. Entries must happen only on information available at that moment.

4. Overfitting

Try enough parameter combinations and one of them will fit history perfectly by chance. The defence is out-of-sample testing: choose parameters on the first part of history, then judge the strategy only on the part it never saw.

A checklist

• Aggregated, multi-exchange data — not one venue's candles.
• Fees, slippage and funding on every trade.
• Signals only from closed candles; no peeking.
• Parameters fitted on one period, judged on another.
• Enough trades (dozens, not five) and a look at drawdown, not only profit.
• Compare with simply buying and holding over the same period.

The AggPro strategy builder tests on data from 15 exchanges with fees and funding included, fits thresholds on the first part of history and shows the rest as an honest check. You can try it right now without signing up.