A good EA is a risk profile, not a return machine
Most buyers of an Expert Advisor ask first: "How many percent does it make?" That is the wrong question. Any return can be inflated by taking more risk — right up to total loss. The right question is: "How much risk does this EA carry for each unit of return, and is its edge real?"
This guide gives you a framework to evaluate any EA — bought commercially or self-built — against the same objective criteria. It builds on the core trading metrics; here we apply them to EA selection.
Criterion 1: Out-of-sample and forward results
This is the knockout criterion. A backtest alone proves nothing — it only shows that parameters can be fitted to the past.
- Demand forward results: a serious vendor shows live or demo forward tracking (e.g. via Myfxbook or FXBlue) that started after the EA was developed.
- At least 6–12 months of live history with 100+ trades. Anything less is statistical noise.
- Check the backtest-to-live ratio: if live performance falls well short of the backtest, the EA was probably over-optimised. The Backtest vs live: why EAs fail guide explains how and why this happens.
Criterion 2: Drawdown relative to return
A return without drawdown context is worthless. Two key measures:
- Recovery factor = net profit / maximum drawdown. Values above 3 (across a full market cycle) suggest a healthy ratio.
- Maximum relative drawdown: how far did equity fall, in percent, from its prior peak? An EA with 40% return and 35% max DD is more dangerous than one with 15% return and 8% max DD.
Always ask: would you have held through the deepest drawdown in the history without switching the EA off? If not, the EA is too aggressive for you — regardless of the return.
Criterion 3: Profit factor and trade distribution
Profit factor (gross profit / gross loss) is useful but easy to manipulate.
| Profit factor | Reading | |---|---| | < 1.1 | Edge too thin, eaten by costs | | 1.3 – 1.8 | Realistic and sustainable | | > 2.5 (over many trades) | Rarely honest — check for curve-fitting or grid/martingale |
What matters is the trade distribution: does the profit come from many similarly sized trades, or from a few outliers? If the entire PF hangs on three lucky trades, it is not reproducible.
Criterion 4: Trade frequency and sample size
An EA with 12 trades a year needs decades to become statistically meaningful. A scalper with 2,000 trades delivers a valid sample in months — but is far more sensitive to spread and slippage.
- Too few trades: no statistical significance, high curve-fitting risk.
- Very many trades: execution quality and costs (slippage, spread, commission) become the dominant factor.
Always relate frequency to the broker: a high-frequency EA fails at an expensive broker even if the logic is sound. See the broker comparison for EA/algo trading.
Criterion 5: Robustness
Robustness means the edge does not depend on one exact parameter combination.
- Parameter plateau: does the EA stay profitable when you shift each parameter by ±15%? A sharp peak instead of a broad plateau is a curve-fitting indicator.
- Multi-symbol test: does the logic work with similar settings on related pairs?
- Different market regimes: has the EA lived through trending, ranging and volatile phases — including a genuine crisis?
Criterion 6: Money management
How does the EA scale lot size?
- Good: risk-based sizing (a fixed percentage of capital per trade), a clearly defined stop-loss per position. Details in the position sizing & risk per trade guide.
- Acceptable with caution: fixed lots, provided they fit the capital.
- Warning sign: lot sizes increased after losses — that is a martingale mechanic.
Red flags: how to spot a problematic EA
- A flawlessly smooth equity curve with no visible drawdown — almost always a hidden grid or martingale system.
- No stop-loss on individual positions.
- Backtest-only results, no verified live tracking.
- Marketing built on return promises instead of risk metrics.
- "Set and forget" with no risk rules — no serious EA runs without drawdown protection.
- Lot size grows after losses — the classic precursor to an account blow-up.
Conclusion
A good forex EA is not defined by a high return but by a healthy return-to-drawdown ratio, an edge proven through out-of-sample and forward testing, a sufficiently large trade sample, and transparent, risk-based money management. Anyone who screens by this framework filters out most marketing products before the first real money is at stake. The next step is to combine vetted EAs sensibly — the EA portfolio management guide shows how.