How to backtest a crypto trading strategy using technical indicators?

Data Collection and Preparation

1. Historical price data must be sourced from reliable exchanges or aggregators such as Binance, Bybit, or Kaiko, ensuring timestamps are aligned to UTC and include open, high, low, close, and volume fields.

2. Data cleaning involves handling missing candles, correcting exchange-specific anomalies like zero-volume periods, and filtering out pump-and-dump artifacts caused by wash trading.

3. Resampling is applied when indicators require specific timeframes—e.g., converting 1-minute OHLCV into 15-minute bars for MACD calculation without lookahead bias.

4. Asset selection matters: backtesting across BTC/USDT, ETH/USDT, and mid-cap tokens like SOL/USDT reveals strategy robustness beyond single-asset noise.

5. Slippage and fee modeling must reflect real execution conditions—using dynamic taker fees (0.02%–0.075%) and bid-ask spread approximations derived from order book depth snapshots.

Indicator Selection and Parameter Tuning

1. RSI(14) and EMA(50)/EMA(200) crossovers form the core signal engine for trend-following strategies, with thresholds calibrated using walk-forward optimization over rolling 90-day windows.

2. Bollinger Bands width expansion triggers volatility-based entries only when standard deviation exceeds 1.8x its 60-day median, reducing false breakouts during low-volatility consolidation.

3. Volume-weighted average price (VWAP) deviations exceeding ±1.2% serve as mean-reversion confirmation filters, discarding signals that contradict intraday liquidity structure.

4. Stochastic RSI(3,3,14) divergence detection requires strict alignment between price swing highs/lows and oscillator peaks/troughs within ±3 candles—no interpolation allowed.

5. Indicator stacking avoids redundancy: combining ADX(14) > 25 with +DI > -DI ensures directional strength before acting on RSI oversold readings below 30.

Execution Logic and Signal Generation

1. Entry rules mandate simultaneous satisfaction of three conditions: EMA(50) > EMA(200), RSI crossing above 30 from below, and 3-candle bullish engulfing pattern confirmed on closing basis.

2. Position sizing enforces fixed fractional risk—1.5% of equity per trade—with stop-loss placed at the recent swing low minus 0.3% buffer to absorb microstructure noise.

3. Take-profit logic uses trailing stops activated only after price moves 2× ATR(14) in favor, then repositioned every 0.8× ATR(14) increment without look-ahead.

4. Short entries require mirrored logic but enforce additional constraint: funding rate must be negative for ≥12 hours prior to entry to avoid contango-driven losses.

5. Signal suppression occurs during top-5 exchange scheduled maintenance windows and 30 minutes before major economic releases tracked via CryptoPanic API feeds.

Performance Evaluation Metrics

1. Sharpe ratio is computed using daily returns, with risk-free rate set to 0%—a deliberate choice reflecting crypto’s non-correlation with traditional fixed income.

2. Maximum drawdown isolates peak-to-trough equity erosion excluding flash crash events lasting under 90 seconds, identified via 5-second tick volatility spikes >15× 24h average.

3. Win rate aggregates all closed trades where exit price exceeded entry by ≥0.5%, excluding breakeven outcomes rounded to nearest 0.01%.

4. Profit factor compares gross profits to gross losses, with values below 1.3 triggering immediate parameter recalibration across all assets.

5. Expectancy per trade is measured in basis points, normalized against median position size across the test period—not nominal USD amounts.

Frequently Asked Questions

Q: Does backtesting on aggregated multi-exchange data introduce survivorship bias?Yes. Aggregators often exclude delisted pairs or exchanges that ceased operations pre-2020. Using only data from exchanges active throughout the entire backtest horizon mitigates this.

Q: How do you handle candlestick wicks that trigger stops but reverse within the same bar?Stop orders execute at the wick extreme only if volume during that sub-candle interval exceeds 120% of the bar’s average volume—otherwise, execution defaults to the bar’s close price.

Q: Can Ichimoku Cloud signals be reliably backtested on altcoin pairs with low liquidity?No. Tenkan-sen and Kijun-sen calculations require at least 180 consecutive non-zero-volume candles; altcoins failing this threshold are excluded from Ichimoku-based rule sets.

Q: Is it valid to use on-chain metrics like active addresses as inputs in technical indicator backtests?On-chain metrics introduce asynchronous latency—daily active address counts arrive with 18–36 hour delays. Their inclusion violates causal integrity and is prohibited in strict technical backtests.