What Is Overoptimization and Why Is It So Dangerous?
Overoptimization (Overfitting or Curve-Fitting) happens when a developer uses the MetaTrader 5 optimizer to test millions of combinations until finding the exact parameters that dodge historical losses. The algorithm doesn't learn a real market edge; it simply memorizes past noise.
Why is it the most dangerous trap?
Because, unlike Martingales or Grids, an overoptimized EA looks flawless in the backtest:
- It uses a real Stop Loss and hides no floating risk.
- It shows minimal Drawdown and a sky-high Profit Factor.
- It seemingly checks every risk-management box.
However, once attached to a live account, the market prints new prices. The millimeter-tuned filters fail, and the system starts draining the account from week one.
8 Key Checks to Spot an Overoptimized EA
Statistical manipulation leaves visible footprints. Always check these points:
1. Millimeter-Tuned Parameters
A robust strategy uses standard, canonical numbers (Period = 20, RSI = 30 / 70). If you see hyper-specific inputs like RSI Period = 17, Entry Level = 26.4, Stop Loss = 213 points, you are looking at genetic optimization footprints. That artificial precision holds no value in the future.
2. "Swiss Cheese" Schedules and Asymmetries
Restricting trading to a specific session (e.g., Asian session) makes sense. Curve-fitting is exposed by arbitrary patches:
- Trading from 01:00 to 08:00, but specifically blocking 03:00.
- Day vetoes ("Disable Thursdays") with no macroeconomic reason.
- Different rules or hours for Buys than for Sells.
The author simply saw losses during those specific slots in the past and switched them off to polish the report.
3. The Unnaturally Perfect Straight Line
No genuine strategy wins continuously for years without flat periods or natural pullbacks. A ruler-straight diagonal line with a Profit Factor above 3.0 usually hides massive parameter overfitting or hardcoded date blacklists to avoid historical crashes.
4. Constant Updates with "New .set Files"
Check the What's New tab on mql5.com. If the seller posts monthly updates with "New .set files for current conditions", it means the old ones failed live. A robust algorithm keeps its core logic stable for years without monthly parameter surgery.
5. The "Publication Date" Test (Forced OOS)
No one can predict the market after uploading the EA. Download the Demo and run a backtest strictly from the day after the publication date to today:
- If the flawless curve turns flat or negative right after publication, you have mathematical proof of curve-fitting.
- If it maintains consistency, the edge is genuine.
6. Testing on a DIFFERENT Broker with "Random Delay"
Many EAs are tuned to a single broker's exact ticks at 0 ms latency. Download the Demo and test it on your own broker's terminal, enabling Random delay. An overoptimized bot breaks down against natural inter-broker quote differences; a robust one stays stable.
7. The Sensitivity Test ("Plateau vs. Cliff")
If the EA exposes inputs (e.g., indicator periods or Stop Loss), change one core parameter slightly by 5% or 10% and re-run:
- Cliff Effect (Overoptimized): A tiny nudge turns a winning curve into a losing one.
- Plateau Effect (Robust): Profits vary slightly, but the system remains profitable and structurally stable.
8. 1 Minute OHLC vs Real Ticks
Many developers optimize using 1 Minute OHLC to save time (compressing a minute into just 4 prices, ignoring internal noise). Never trust a backtest unless you run it under Every tick based on real ticks. If it shines in OHLC but fails on real ticks, its edge was a simulation artifact.


