Build a Custom EA — Step-by-Step Series for MT5 Development

Before writing code, document the strategy in plain English: entry conditions (when to open position), exit conditions (stop loss, take profit, time-based exit), position sizing rule (fixed lots, percent risk, Kelly fraction), instruments traded, timeframes, market regime expectations.

Example: 'Buy EURUSD on H1 timeframe when 20-period EMA crosses above 50-period EMA AND RSI(14) above 50. Risk 1% per trade with stop at recent swing low. Take profit at 2× risk. Skip Friday afternoons and major news events.'

Critical: the strategy should have a hypothesis about WHY it works. Trend-following hypothesis: 'persistent trends in forex driven by macro factors create exploitable directional moves'. Without a hypothesis, strategy is curve-fit pattern matching that won't generalise.

Anti-pattern: starting with backtest optimisation rather than strategy hypothesis. Optimising parameters to maximise backtest return produces over-fit strategies that fail live.

Project structure: separate files for strategy logic, risk management, position management. Avoid monolithic single-file EAs that mix concerns.

Use Standard Library extensively: CTrade for execution, CPositionInfo for position queries, CSymbolInfo for symbol properties. Avoid raw OrderSend() unless you have specific reason.

Magic number convention: use unique magic number per EA so the EA can identify its own positions and avoid interfering with other EAs or manual trades. Different strategies = different magic numbers.

Error handling: every trade.Buy()/trade.Sell() should be followed by trade.ResultRetcode() check. Log failures; don't silently ignore. Common retcodes: TRADE_RETCODE_DONE (success), TRADE_RETCODE_REJECT (broker rejected — investigate), TRADE_RETCODE_REQUOTE (price moved — retry).

Time and market session handling: use TimeCurrent() and TimeToStruct() to filter trading hours. Many strategies are session-specific; running outside intended session degrades performance.

MT5 Strategy Tester runs your EA on historical data with simulated execution. Critical settings: (1) Modeling: 'Every tick based on real ticks' is most accurate but slowest; 'Every tick' for medium accuracy; 'OHLC 1 min' for fast initial tests. (2) Currency: USD. (3) Initial deposit: realistic ($5,000-$10,000). (4) Leverage: match production broker leverage.

Multi-pair and multi-timeframe testing: don't test on one pair/timeframe only. Trend-followers should work on multiple major pairs across H1-D1. If strategy only works on EURUSD H4, it's likely curve-fit.

Realistic spreads and slippage: Strategy Tester uses recorded historical spreads but may not capture slippage realistically. Add buffer (e.g., +0.5 pip to typical spread) for production safety margin.

Avoid look-ahead bias: use only data available at the time of decision. MQL5's iClose(symbol, timeframe, 0) returns current (incomplete) bar; index 1 is the just-closed bar. Most EAs should base decisions on closed bars (index 1+) to avoid look-ahead.

Common backtest pitfalls: over-fitting (too many parameters tuned to historical data), survivorship bias (only testing currently-active pairs), spread variance (using current spreads on historical data when historical was different).

Walk-forward optimisation: optimise parameters on in-sample window (e.g., 2020-2022), test on out-of-sample window (2023). Re-optimise on (2021-2023), test on 2024. Continue rolling. Strategy should perform similarly out-of-sample as in-sample; large degradation = over-fitting.

Demo account testing: deploy on demo broker account for 2-4 weeks. Verify behaviour matches backtest expectations under real broker execution. Common surprises: slippage worse than backtest, missed trades during network issues, broker rejection of certain orders.

Production deployment: use VPS for 24/7 operation (Beeks Financial Cloud, Equinix Cloud, or broker-provided VPS). Set up monitoring (Myfxbook auto-sync, custom alerts on drawdown thresholds). Start with conservative position sizing; scale up gradually as confidence grows with live results.