[Backtest] [Backtest] Great backtest, bad live results: 6 reasons why

"200% a year in the backtest, negative on the live account" — anyone who builds EAs hits this at some point. The cause is almost always one of these six.

1. Overfitting

• Settings tuned perfectly to past data won't fit the future.
• Check: split the period — find settings on the early part, confirm on the later part (walk-forward). Make sure nearby values (20 → 18, 22) also work.

2. Spread set too low

• The tester's default or "current" spread is kinder than reality.
• Gold spreads widen several times at news and rollover. Re-run with a generous spread, above average.

3. Slippage and execution delay

• Backtests fill instantly at the requested price. Live trading doesn't.
• The shorter and more frequent the trades, the bigger the gap. Stops slipping just $1–2 can flip the result.

4. Modeling quality

• "Open prices only" ignores movement inside the candle. If stop and target sit inside the same candle, the result is wrong.
• Do the final check with Every tick based on real ticks.

5. Missing commissions and swaps

• Leave out ECN commissions and overnight swaps and long-hold strategies lose a big chunk of profit.

6. Market regime change

• Gold in 2020–2023 and gold in 2024–2025 had very different volatility. An EA fitted to one regime stalls when it changes.
• Test on at least 2–3 years that include up, down and sideways markets.

Closing the gap, step by step

1. Backtest with real ticks + generous spread + commissions
2. 2–4 weeks on demo (same settings)
3. One month on a small live account → compare with the backtest
4. If the gap is within 50%, scale up

Live results typically come in 20–50% below the backtest. Publish your live record with an EADesk live link and you can analyze the gap together with other members.