Crypto Backtesting: Pitfalls, Fees, Slippage, and Honest Expectancy

What Backtesting Can and Cannot Do

Backtesting runs your rules on historical candles or trades.

It shows how the rule would have behaved — not how it must behave next.

• Can: estimate rough win rate, R:R, drawdown shape
• Cannot: guarantee future liquidity or funding regime
• Cannot: capture exchange outage or delisting
• Output is hypothesis — forward test before size

Always Include Fees in the Simulation

Zero-fee backtests inflate win rate on short-hold strategies.

Use maker and taker assumptions that match live execution.

• Apply fee on entry and exit notional each trade
• Scalps: assume taker-taker unless proven otherwise
• Swing: maker entry plus taker exit is common baseline
• VIP tier change — rerun at realistic rate for your volume

Slippage and Spread Modeling

Historical close price is not always your fill price.

Add slippage ticks on entries, exits, and stops.

• Fixed tick slippage per side — simple first pass
• Percent slippage scales with volatility on alts
• Stop slippage often worse than entry — stress test wider
• Low-volume periods in history may understate live slip

Funding on Perpetual Strategies

Multi-hour holds through funding intervals need carry in the backtest.

Historical funding series exist on many data providers.

• Apply funding at each interval for open position
• Long-short hedges net funding — model both legs
• Funding sign changes regime — full history helps
• Ignore funding and long-hold backtests lie

Overfitting and Parameter Mining

Optimizing twenty parameters on one dataset finds noise, not edge.

Simple rules with stable performance across splits beat perfect curves.

• In-sample vs out-of-sample split mandatory
• Walk-forward: train window, test next window, roll
• Fewer parameters — harder to accidentally fit noise
• If equity curve needs one magic setting, suspect overfit

Survivorship and Data Quality Bias

Testing only coins that survived ignores delisted blowups.

Bad tick data and missing candles distort signals.

• Include delisted pairs if strategy traded alts broadly
• Exchange-specific candles differ — match live venue
• Flash wicks trigger stops — use realistic stop fill model
• Clean data first — garbage in, confidence out

Look-Ahead and Execution Assumptions

Using future candle data in signal logic invalidates results.

Same-bar entry on close is optimistic unless proven fillable.

• Signal on close, enter next open — conservative default
• Indicators needing future bars are invalid in live
• Limit fill assumptions — verify depth existed
• Latency on API strategies not visible in candle backtest

From Backtest to Forward Test

Paper trade or minimum size live before scaling.

Compare live slippage and fees to backtest assumptions monthly.

• Forward test minimum 30 trades or one month
• Track live vs backtest expectancy drift
• Update fee assumptions when VIP tier changes
• Kill strategy if live drawdown exceeds backtest max by wide margin

Quick Summary

Crypto backtesting is useful only with fees, slippage, and funding included.

Overfitting and bad data produce pretty equity curves that die live.

Compare assumed fees to real venue tiers before trusting the output.

• Model round-trip fees on every simulated trade
• Add slippage — especially on stops
• Out-of-sample test before allocating real size