[Course] [Build an EA 8] 10 common beginner EA mistakes
Most EA problems posted in Q&A are one of these ten. Read this once before building and once before going live.
- No stop loss
EAs that assume "it'll come back eventually" usually lose the account to one big trend. A stop is mandatory.
- Martingale and averaging-down grids
Doubling size after every loss gives a beautiful curve until it hits zero at once. The backtest period just didn't contain a big trend.
- Deciding on an unclosed candle
Crosses appear and vanish, so the backtest and live results differ completely. Decide once per new candle, as in part 4.
- Ordering on every tick
Orders keep firing while the condition holds, stacking dozens of positions. Always add a guard like "sit out if a position is open."
- No magic number
It touches other EAs' or manual orders and closes the wrong positions.
- Ignoring spread and commission
Backtests with a fixed zero spread mostly look good. Use the real spread.
- Backtests that are too short
Three months tells you nothing. The bar is 2+ years and 100+ trades.
- Overfitting
Numbers fitted perfectly to the past won't fit the future. Find them on an early period and confirm on a later one.
- Ignoring weekend gaps and news
A position held on Friday night can open Monday past its stop. Consider an option to pause around the jobs report (NFP) and FOMC.
- Not keeping records
If you don't note what you changed and what happened, you'll repeat mistakes. Log settings and results for each version.
Final check before going live
- ☐ Every order has a stop and a target
- ☐ One stop-out loses only 1–2% of the account
- ☐ Backtested 2+ years with real ticks and a real spread
- ☐ Ran 2+ weeks on demo without problems
- ☐ A stop rule (max drawdown) is decided
All checked? Publish it with EADesk live linking. A record with nothing to hide is the strongest proof.
