You buy an EA for $299. The seller's backtest curve looks like a straight line going up. You put it on a live account. Two weeks later, the account is down 30%. You contact the seller. "Your parameters aren't set right," they say. Another week passes. Account is blown.
This happens every day. The EA market is full of products where the backtest looks incredible and the live performance is nonexistent. By some estimates, fewer than 5% of EAs sold online can produce consistent profits over 12+ months. The rest are either overfitted to historical data or outright designed to separate you from your money.
There's no shortcut to finding a good EA. But there is a process.
The scams you'll encounter
Three patterns show up over and over.
Fabricated backtests. The developer shows a backtest with 300% annual return and 5% drawdown. What they don't show is that the EA uses a "future function" — it peeks at price data that wouldn't have been available in real time. The backtest looks perfect because the EA literally cheated. Run the same EA live and it's a different story entirely. We've seen dozens of these — the red flags are always the same. More on spotting them in our EA scam guide.
Cherry-picked live results. Developer opens 10 live accounts. Nine blow up, one gets lucky and makes 50% in a month. Guess which one ends up in the marketing? This is survivorship bias in action, and it's everywhere. Verified track records from independent sources (like Myfxbook with verified broker connection, or our signal system) are far more trustworthy than screenshots.
Martingale dressed up as magic. Win rate above 95%. Smooth curve with almost zero drawdown. Looks too good to be true because it is. The strategy doubles position size after every loss — it recovers from most drawdowns, until the one time it can't, and the entire account evaporates. Martingale isn't inherently a scam, but advertising its smooth curve while hiding the tail risk is dishonest.
Rule of thumb: if an EA's marketing looks too good to be true, it is. No exceptions.
Metrics worth checking
Ignore the screenshots. Ignore the testimonials. Look at these numbers.
Profit factor
Total gross profit divided by total gross loss. This is the single most reliable indicator of whether a strategy has edge.
Below 1.0 means it loses money. Between 1.3 and 2.0 is the sweet spot for real strategies. Above 3.0 with less than 300 trades? Suspicious — either too few trades for the number to be meaningful, or the strategy is overfitted.
An EA that maintains a profit factor of 1.5 over two years of live trading is far more impressive than one showing 5.0 over three months of backtesting.
Maximum drawdown
The biggest drop from peak to trough. This number tells you how bad it gets.
Most people overestimate their risk tolerance. You think you can handle a 30% drawdown. Then your $10,000 account drops to $7,000 and you can't sleep. A realistic rule: the drawdown you can actually tolerate is probably half of what you think.
| Max drawdown | What it feels like |
|---|---|
| Under 10% | Manageable for most people |
| 10–20% | Uncomfortable but survivable |
| 20–30% | Tests your conviction. Most people start second-guessing the EA here |
| 30–50% | Painful. High chance you'll manually shut it off |
| Over 50% | One bad day from a blown account |
Sharpe ratio
Return adjusted for risk. How much profit per unit of risk taken.
Below 0.5: not worth the volatility. 0.5 to 1.0: passable. Above 1.0: genuinely good. This metric prevents you from chasing raw returns while ignoring the wild ride needed to get there. According to Investopedia, even most professional fund managers struggle to sustain a Sharpe ratio above 1.0 long-term.
Backtest quality matters more than length
A 10-year backtest on daily candles is less reliable than a 3-year backtest on real tick data with 99%+ modeling quality.
Quality backtesting requirements:
- Tick data with 99%+ modeling quality (MT5's "every tick based on real ticks" mode, or tick data tools on MT4)
- At least 3 years covering different market regimes (trending and ranging)
- Spreads set to realistic levels (not the default fixed 1 pip)
- Out-of-sample validation: optimize on one data period, test on a separate period the EA never saw. If results collapse on out-of-sample data, the strategy is overfitted
We have a complete breakdown of how to read and evaluate backtests in our backtest guide and backtest report guide.
| Metric | Passing | Good | Warning |
|---|---|---|---|
| Profit factor | > 1.2 | 1.5–2.0 | > 3.0 (probably inflated) |
| Max drawdown | < 30% | < 15% | > 50% (too aggressive) |
| Sharpe ratio | > 0.5 | > 1.0 | < 0 (losing strategy) |
| Trade count | > 200 | > 500 | < 50 (statistically useless) |
Why great backtests fail live
This is the question that confuses every beginner. The EA looked perfect in testing — what happened?
Execution costs. Backtests typically use fixed spreads and zero slippage. Live trading has variable spreads that widen during news and low-liquidity periods, plus slippage on every market order. For strategies with thin edges (scalpers especially), the difference between backtest and live costs alone can flip profitability.
Overfitting. The developer adjusts parameters until the backtest curve is perfect — but the parameters are memorizing historical noise, not discovering real patterns. When the future inevitably looks different from the past, the strategy breaks. We've written extensively about this in our overfitting guide.
A quick sanity check: count the adjustable parameters. More than 10 is a yellow flag. More than 20 is almost certainly overfitting. The best strategies we've seen typically have 3–5 core parameters.
Strategy types and their risk profiles
Knowing what type of strategy you're running is more important than any single metric.
Trend following — Opens positions in the direction of the prevailing trend. Low win rate (30–40%), high reward per winning trade. You'll experience long losing streaks. If you can't emotionally handle 12 consecutive losses while knowing the math still works, this isn't for you.
Grid — Places orders at fixed intervals, profiting from price oscillation within a range. Comfortable during ranging markets. Dangerous during strong trends because it keeps adding positions against the move. Without a hard stop loss, one trending market can wipe out months of profits.
Martingale — Doubles position size after each loss to recover. High win rate, smooth curve, catastrophic tail risk. Before using any Martingale EA, you need to know: maximum number of layers, maximum position size at max layers, and whether your account can survive the worst case. Read our full Martingale risk breakdown.
Scalping — Small profits per trade, high frequency. Extremely sensitive to execution quality. The VPS location, broker spread, and network latency matter more than the strategy logic itself. The gap between backtest and live results is typically largest with scalping EAs.
The right way to test an EA
After buying an EA, follow this order. Skipping steps is how accounts get blown.
1. Demo account — at least one month. Same broker and account type as your intended live setup. Run default parameters. The goal isn't to see if it makes money — it's to understand what the EA does. When does it enter? How long does it hold? What triggers its stop loss?
2. Small live account — at least two months. An amount you can lose completely without losing sleep. $200, $500, whatever. Real money introduces slippage and execution realities that demo can't replicate.
3. Compare demo vs live. If live performance is significantly worse than demo, the EA is too sensitive to execution conditions. Scaling it up will only amplify the problem.
4. Scale slowly. Once live results are stable, increase the account size. Each increase should be no more than 50% of the current amount. Leave yourself observation windows between increases.
We see traders skip straight to step 2 with $5,000+ all the time. The ones who survive are the ones who don't.
Setting realistic expectations
The long-term annual return of a genuinely reliable EA is probably 15–30%.
That might sound low for something that costs money and requires a VPS. But consider the benchmarks: the world's top hedge funds average 15–25% annually. Warren Buffett's long-term average is around 20%. An EA producing 25% annually with under 15% drawdown is an excellent result by any standard.
Monthly expectations:
- 2–5% average, with losing months mixed in
- Some months will be negative even with a winning strategy
- The equity curve trends upward but the path is never smooth
If someone promises "20% monthly with no drawdown," close the tab. It doesn't exist. Our realistic income expectations guide goes deeper into the math.
Money management is more important than EA selection
The same EA, running at 0.01 lots vs 1.0 lot on the same account, can produce completely different outcomes. The first survives. The second blows up in a week. Same strategy — the difference is position sizing.
Hard rules:
- Single trade risk: 1–2% of account. On a $10,000 account, max loss per trade is $100–200.
- Never use all your margin. Keep used margin under 20% of account equity at all times.
- Plan for 1.5x historical drawdown. If the EA's max backtest drawdown is 20%, your account needs to survive 30% without margin issues.
- Diversify if possible. Two uncorrelated EAs on separate pairs is safer than one EA with double the lot size. We cover this in detail in our risk management guide.
EAs are tools. Even a great tool in the wrong hands produces bad results. Position sizing is where most of the "skill" in running EAs actually lives.
About the author: The FXTool team builds and tests MetaTrader trading tools daily. We run every EA we sell on live accounts and publish the results. This guide reflects what we've learned from building 50+ EAs and working with thousands of retail traders.
Forex trading involves significant risk and may result in total loss of capital. This article is for educational purposes only and is not investment advice. Understand the risks and consider your financial situation before trading.