Avoid Curve Fitting: 7 Robustness Checks Before You Trust Any EA Backtest
My Trading

Avoid Curve Fitting: 7 Robustness Checks Before You Trust Any EA Backtest

7 January 2026, 16:00
Flora Rosa Seeholzer
Flora Rosa Seeholzer
0
65

If you’ve been around EAs long enough, you’ve seen it:

A backtest that looks like a straight line up.
Tiny drawdown. Massive profit. “Too good to be true.”

And then live performance collapses.

That’s usually not “bad luck”.

That’s curve fitting.

Curve fitting (over-optimization) happens when an EA is tuned to match the past perfectly… but has no real edge that survives new market data.

This post gives you 7 evergreen checks to spot curve fitting before you waste money, time, and confidence.

The uncomfortable truth: most backtests are marketing, not evidence

A backtest is not proof.

A backtest is a hypothesis.

Your job is to pressure-test the hypothesis until only robust systems survive.

If you remember one sentence:

The cleaner the curve, the more suspicious you should be.

Robustness Check #1 — Does it survive spread increases?

Many EAs only work under “ideal spreads”.

But live markets are not ideal, especially during volatility.

Test the backtest with:

  • higher spread assumptions

  • worse execution conditions

If results collapse immediately, the EA doesn’t have a real edge.

This is also why broker choice is part of robustness.

Brokers I recommend for EA execution:

If you run EAs on poor execution, you’ll misjudge everything.

Robustness Check #2 — Does it survive slippage?

Slippage matters most for:

  • breakout systems

  • Gold (XAUUSD)

  • fast continuation moves

If your strategy depends on perfect fills, it’s fragile.

A robust EA should remain acceptable even with realistic slippage.

Robustness Check #3 — Does it work across different market regimes?

A curve-fitted EA often looks great in one regime:

  • strong trend period

  • stable volatility

  • clean ranges

Then it fails when regime changes.

A robust strategy should have:

  • up periods

  • flat periods

  • drawdowns
    …but not complete structural collapse.

Robustness Check #4 — Does it depend on one tiny parameter?

If changing one number slightly breaks performance, it’s not robust.

Signs of fragility:

  • performance flips from great to terrible with small changes

  • “sweet spot” settings that must be exact

  • dozens of parameters that interact in unpredictable ways

This is why simpler EAs often survive better: fewer moving parts to overfit.

Robustness Check #5 — Does it hold on different time windows?

A classic curve-fit trick:

  • optimize on a window that flatters the strategy

  • show only the pretty part

A robust EA should make sense across:

  • multiple years

  • multiple conditions

  • not just one “golden period”

Look for:

  • consistent logic

  • not just consistent profit

Robustness Check #6 — Does it rely on dangerous recovery mechanics?

Many curve-fitted equity curves are secretly supported by:

  • grid

  • martingale

  • “recovery mode”

  • exposure stacking that explodes rarely

It can look perfect for months… until it doesn’t recover.

Robust systems control downside without needing “infinite margin” logic.

Robustness Check #7 — Can you validate it forward with a real workflow?

No matter how good a backtest looks, the final check is always the same:

Demo → small live → stability sample size.

If you don’t follow that workflow, you’ll keep buying “pretty curves” and staying confused.

A clean workflow is here (bookmark it):

  • Choose broker first (execution matters)

  • Demo to verify behavior

  • Small live to confirm real conditions

  • Run long enough to validate stability

Why simple systems often win long-term

There’s a reason many retail EAs fail:

  • too many settings

  • too many degrees of freedom

  • too many ways to over-optimize

Simple, rule-based EAs are easier to:

  • test properly

  • run consistently

  • avoid user misconfiguration

That’s also why my ProTrading EAs are designed to be straightforward and stable, not “1000-parameter puzzles”.

Two examples of simple, robust automation frameworks

USDJPY Trend (H1)

JPY Trend EA ProTrading (74 USD)
MT5: https://www.mql5.com/en/market/product/157484
MT4: https://www.mql5.com/en/market/product/157485

XAUUSD Breakouts (M15)

Gold Trend Breakout EA ProTrading (74 USD)
MT5: https://www.mql5.com/en/market/product/157465
MT4: https://www.mql5.com/en/market/product/157466

They’re built to be:

  • easy to run

  • hard to misconfigure

  • focused on execution + risk control rather than “curve perfection”

If your end goal is scaling, curve fitting is your biggest enemy (Axi Select)

Scaling capital only makes sense if your strategy is robust.

If your system is fragile, scaling just makes the blow-up bigger.

If you want a prop-style capital path that aligns better with consistency than typical challenge models, compare Axi Select:
https://bit.ly/48TlcAc

It’s worth understanding if you’re serious about systematic trading.

Copy/paste checklist: 7 robustness checks

  1. Spread stress test (does it survive wider spreads?)

  2. Slippage stress test (especially breakouts / Gold)

  3. Regime test (trend vs range vs high volatility)

  4. Parameter sensitivity (small changes shouldn’t kill it)

  5. Multiple time windows (not one cherry-picked period)

  6. No dangerous recovery logic (grid/martingale/exposure stacking)

  7. Forward validation (demo → small live → stability)

Execution matters:

Scaling path to compare:

Quick Links (EAs + Brokers + Axi)

JPY Trend EA ProTrading (74 USD)
MT5: https://www.mql5.com/en/market/product/157484
MT4: https://www.mql5.com/en/market/product/157485

Gold Trend Breakout EA ProTrading (74 USD)
MT5: https://www.mql5.com/en/market/product/157465
MT4: https://www.mql5.com/en/market/product/157466

IC Trading: https://bit.ly/3KvI9RO
Pepperstone: https://bit.ly/4ophy72
Axi Select: https://bit.ly/48TlcAc