How to Backtest a Crypto Trading Strategy With Real Fees

Most backtests lie. Not because the math is wrong, but because they leave out the two things that decide whether a strategy survives contact with a real exchange: fees and honest data handling. A strategy showing 30% annual returns at zero fees can lose money at 0.25% per trade — and most free backtesters default to zero.

Why Fees Kill More Strategies Than Bad Signals

Fees scale with trade frequency, and frequency is exactly what automation gives you. Run the numbers:

• A bot that trades 500 round trips a year at 0.25% per side pays roughly 250% of its position size in fees annually.
• To break even, your average trade has to clear 0.5% before it earns anything.
• Scalping strategies with 0.3–0.4% average wins are mathematically dead on arrival at retail fee tiers — no matter how good the win rate looks.

This is why the first honest thing a backtest can do is charge you the fees you'll actually pay. On Kraken, check your real tier (spot taker fees start around 0.40% and fall with 30-day volume) and use that number — not the platform default.

The Five Backtest Sins

1. Zero or fantasy fees

Covered above. If your backtester doesn't let you set maker/taker fees per side, don't trust its output.

2. Lookahead bias

Your strategy sees data it wouldn't have had in real time — e.g., trading on a candle's close price at its open, or using an indicator computed on the full dataset. Symptoms: impossibly smooth equity curves.

3. Overfitting

Tune enough parameters against enough history and any strategy looks perfect — on that history. If performance collapses when you shift the test window by a month, you fit noise, not signal.

4. Ignoring slippage

Backtests fill at the exact signal price. Real markets don't, especially on market orders in thin books. Add a slippage assumption (even a flat 0.05–0.1%) and see if the strategy survives.

5. Cherry-picked windows

Testing a long-only strategy on 2023–2025 crypto data proves nothing except that markets went up. Test across trending, ranging, and crashing regimes, or admit you have a bull-market-only system.

A Sane Backtesting Workflow

1. Write the rules first. Entry, exit, position size — fully specified before you look at results. If you can't write it as code, it isn't a strategy yet.
2. Set real costs. Your actual exchange fee tier, both sides, plus a slippage haircut.
3. Test in-sample, validate out-of-sample. Tune on one period, then run untouched data. Only the out-of-sample number counts.
4. Read per-strategy attribution. If you run multiple strategies, you need to know which one made the money. Blended results hide losers inside winners.
5. Paper trade before live. A backtest that survives fees still has to survive live data feeds, latency, and partial fills.

What a Passing Grade Actually Looks Like

A strategy worth deploying typically shows: positive expectancy after fees and slippage, out-of-sample results in the same ballpark as in-sample, a max drawdown you could psychologically and financially survive, and enough trades (hundreds, not dozens) for the stats to mean anything. If any of those fail, iterate — deployment is the reward for passing, not the next thing to try when results are mediocre.

Frequently Asked Questions

What fee rate should I use when backtesting on Kraken?

Use your actual tier from Kraken's fee schedule. Spot taker fees start around 0.40% for low-volume accounts and decrease with 30-day volume; Kraken Pro tiers are lower. When in doubt, backtest at a pessimistic rate — a strategy that survives high fees only gets better at lower ones.

How much historical data do I need for a meaningful backtest?

Enough to cover multiple market regimes — trending up, trending down, and ranging. For intraday strategies that's typically tens of thousands of candles across at least 12–24 months, producing several hundred trades.

Why does my strategy work in backtests but lose money live?

The usual suspects, in order: fees and slippage weren't modeled, lookahead bias in the backtest, overfitting to the test window, or live data-quality issues (gaps, stale candles) the backtest never faced.

Is backtesting enough before going live?

No. Backtesting filters out losers; it doesn't certify winners. Paper trade or run minimal position sizes first, and use platform-level risk limits so an early failure is survivable.