When analyzing a backtest, almost out of habit we all look at the exact same four numbers: Win Rate, Total Net Profit, cash Drawdown, and Profit Factor.
Of course, the fact that we do not stop here to explain what Profit Factor or Net Profit is does not mean at all that they are unimportant — on the contrary, they are the essential foundation of any analysis. However, since everyone already checks those figures out of habit, stopping only at them is a very common mistake. Inside the official MetaTrader 5 report there are 5 other fundamental metrics that almost nobody looks at — plus one invisible factor regarding execution latency — that truly allow us to evaluate the complete viability and real safety of an EA.
1. Balance vs. Equity Relative Drawdown (The Grid & Martingale Detector)
In the report you will see two relative drawdown figures side by side: Balance Drawdown Relative and Equity Drawdown Relative. Comparing these two numbers is the most reliable way to uncover what a bot really does when a trade goes wrong.
- Balance Drawdown: Measures the percentage drop counting only trades that have already been closed.
- Equity Drawdown: Measures the real drop your money suffered while trades were still open and floating in loss.
Why does it help you detect runaway Grids or hidden Martingales?
Bots that rely on dangerous Martingales or unbounded Grids refuse to close losing trades: when price moves against them, they open more and more lots and wait weeks hoping price will eventually bounce back. Because of that, their reports show a misleading Balance Drawdown of 1% or 2% (a blue line that looks perfect), while right next to it you will see an Equity Drawdown of 40%, 60%, or 80%. That huge gap reveals that the bot held massive floating losses that came close to blowing the account. Conversely, when both percentages move virtually hand in hand, you know the bot accepts and cuts its losses in a healthy, controlled way without hiding floating risk.
What numbers are considered good?
- Overall value: A relative drawdown below 10% – 15% is considered very healthy, and below 5% is excellent.
- Gap between Balance and Equity: Equity Drawdown should never exceed Balance Drawdown by more than 2% – 5%. The closer both percentages are to each other, the cleaner and safer the system is.
2. Recovery Factor
What is it and why does it matter?
It is calculated by dividing Total Net Profit by Maximal Drawdown. Simply put: it tells you how many times over the bot has been able to multiply its worst losing streak. Keep in mind that this metric is cumulative over time: the more years a backtest spans, the higher the accumulated profit relative to the maximum drawdown. A high Recovery Factor ensures that when a natural market dip occurs, the bot has the strength to climb out of the drawdown quickly and reach new highs without staying stuck for months.
What numbers are considered good?
- Below 1.0: Poor (the bot has not even managed to earn back the equivalent of its worst drop).
- Between 1.5 and 2.5 (on 1-year tests): Good and healthy (meaning that in a single year it has already made around double its worst dip).
- Above 3.0 – 5.0 (on multi-year tests): Very good (professional grade).
- Above 10.0 (on long 5+ year histories): Exceptional.
3. Sharpe Ratio
What is it and why does it matter?
This is the go-to metric for investment funds. It measures how much profit the system generates relative to the ups and downs (volatility) of its equity curve. Two bots can finish the year making the exact same amount of money, but one might do it with huge monthly rollercoaster swings while the other adds steady, calm profits week after week. The higher the Sharpe Ratio, the smoother, more stable, and more predictable the path to making that profit.
What numbers are considered good?
- Below 1.0: Unstable (too much rollercoaster volatility for the profit offered).
- Between 1.0 and 2.0: Good.
- Between 2.0 and 3.0: Very good.
- Above 3.0: Extraordinary (typical of systems with very high week-to-week consistency).
4. LR Standard Error (and LR Correlation)
What is it and why does it matter?
Imagine taking a ruler and drawing a perfect diagonal straight line from the initial deposit to the final balance of the backtest.
- LR Correlation (Linear Regression Correlation): Measures from 0.00 to 1.00 how closely the bot's chart follows that upward straight line.
- LR Standard Error: Measures in cash ($ USD) how much the bot's curve deviates on average (up or down) from that ideal straight line across the period. If it is very high or exceeds the Maximal Drawdown itself, it means the system goes through long dry spells or big spikes and drops; if it is very low, growth stays tightly hugged to its trendline from start to finish.
What numbers are considered good?
- LR Correlation: Above 0.80 – 0.85 is good on 1-year windows, and between 0.90 and 0.99 is excellent on multi-year tests.
- LR Standard Error — Poor: If it exceeds the Maximal Drawdown or represents a constant deviation above 5% – 10% of capital (indicating an erratic curve that wanders too far from the diagonal).
- LR Standard Error — Good / Excellent: When it represents less than 1.5% – 3% of initial capital and stays clearly below the Maximal Drawdown.
5. Total Number of Trades (Total Trades)
What is it and why does it matter?
In statistics, a fundamental law applies: the larger the sample size, the smaller the role of luck. A backtest boasting a 90% win rate or a sky-high Profit Factor over only 40, 60, or 80 trades across several years has no statistical validity: it is very easy for an optimizer to tweak a handful of parameters to catch half a dozen past moves purely by chance (overfitting). In addition, right next to that figure on the same row, always check that there is a healthy balance between Short Trades (sells) and Long Trades (buys), confirming the system knows how to profit in both market directions.
What is the minimum acceptable sample and what numbers are considered good?
- Fewer than 100 trades: Insufficient. The sample is too small, and the risk that the results come from luck or curve-fitting is very high.
- Between 200 and 300 trades: The minimum acceptable threshold in quantitative trading to start granting statistical validity to a strategy (or to evaluate a 1-year test window).
- More than 500 trades in a single year (or 1,000 to 5,000+ across multi-year tests): A solid, statistically reliable sample where luck is diluted and the Win Rate and Profit Factor reflect the algorithm's true capability.
The Invisible Detail Missing from Every Published Backtest: Random Delay (Latency & Slippage)
There is one last decisive factor to keep in mind whenever you look at a published backtest screenshot: a MetaTrader 5 report never displays what execution latency was used when running the test. Two published reports can both show the green 100% real ticks bar while hiding opposite conditions:
- Zero latency (0 ms) execution: By default, many backtests are launched assuming 0 ms latency — an idealized world where orders fill at the exact millisecond requested with zero broker queues or slippage. Many scalpers achieve brilliant curves this way in the Tester, yet on a live account — where tens or hundreds of milliseconds always pass between your VPS and the broker's server — they enter a few ticks late on every trade and their performance changes completely.
- Random delay execution: When enabled in MetaTrader 5, the Tester injects variable, unpredictable pauses before filling each order at the actual price available after that delay, simulating live-market friction. If an EA keeps its curve solid under Random delay, it proves its edge is real and does not depend on sniping impossible micro-ticks.
How can you verify it if a published backtest does not show it?
Because that setting is never printed on a published screenshot, the only real verification is in your own hands: download the Free Demo version of the EA (for example from mql5.com, which lets you test any product inside the Strategy Tester), select Every tick based on real ticks + Delay: Random delay on your own computer, and click Start:
- If the published backtest relied on running with Zero latency (0 ms), you will notice that the growth curve loses stability and no longer resembles the officially shown chart.
- If the EA was honestly engineered and validated to withstand real-world latency, your test will preserve the same stability and consistency (keeping in mind that no two runs will ever be 100% identical, precisely because of that random delay on each execution).


