What Monte Carlo is (in practice)
Monte Carlo is not a tool to improve a strategy.
It is a stress test.
It takes your existing system and runs multiple variations of it under slightly different conditions:
price sequence changes
spread variation
slippage
starting point shifts

The goal is simple:
to see whether the strategy is dependent on perfect conditions
or if it can survive realistic market imperfections

The settings I use
I keep my Monte Carlo settings simple and focused on realism.

Market variations
Randomize history data → ON
The market never moves exactly the same way twice.
Small differences in price sequencing are normal.
This test checks whether the strategy depends on perfect candle structure.
This is one of the most important tests.

Randomize spread → ON
Spread is never constant.
Broker conditions change continuously.
Especially on XAUUSD, spread has a major impact.
Without this, backtests are often too optimistic.

Execution problems

Randomize slippage → ON
Slippage is part of live trading.
Especially during volatility and fast moves.
This tests whether the system can handle execution imperfections.

Randomly skip position entry → OFF
This option simulates missed or skipped entries.

That can happen in specific situations, for example with news filters, spread filters, or execution restrictions.

But in my workflow, I do not use it as a standard Monte Carlo condition.

I use Monte Carlo mainly to test the broader structure of a system under realistic market stress, not to simulate every possible exceptional filter or execution scenario.

For that reason, I leave this option OFF.

Randomly skip position exit → OFF
This option simulates missed or skipped exits.
That is not something I consider a realistic standard condition for normal EA execution, so I leave it OFF.

Randomly close position → OFF
This simulates manual interference.

In my approach:
I do not interfere with robots.
That is a hard rule in my workflow.

The only time I broke that rule was earlier on, when I had a stack of XAUUSD robots all going against me at the same time with wide stop losses. That was an exposure mistake on my side, and I had to partial close twice because the total exposure was too high.

I learned from that. I later made structural adjustments to prevent that situation from happening again, which I explained in more detail in the previous topic:
https://forexsb.com/forum/topic/10058/why-multiple-profitable-eas-can-still-hurt-your-account/

Backtest start

Randomize backtest starting bar → ON
This checks whether the starting point matters.
A robust system should not depend on a perfect start moment.
Together with history randomization, this is one of the strongest robustness checks.

Strategy variations

Randomize indicator parameters → OFF

Important clarification:
This is not re-optimization.
It slightly shifts parameters to test sensitivity.

Sequence:
Reactor → optimization
Monte Carlo → validate stability

Not:
Reactor → Monte Carlo → search for new parameters

This is an important distinction.

How I use the Confidence Table
The Confidence Table is where most of the real information is.

Each row represents a confidence level:
20% → optimistic scenario
50% → median
85–95% → stress zone
100% → worst case

These are not new backtests.
They are variations of the same system under stress.

What I look at
I do not read the table row by row.

I read it vertically:
How fast does performance degrade
How stable is Return / DD
How does drawdown evolve
Does SQN collapse or remain meaningful

My key metric

For me, Return / DD is one of the most important signals.

Example:
If the original Return / DD is around 12, then ideally I want to see at least around half of that still remaining at the 80–85% confidence zone.

So in that example, a value around 6 would be very strong.
That means the system still holds a significant part of its structure.

That is a strong sign of robustness.

That said, I am not overly strict with it.

If the rest of the system still looks good in terms of equity curve, overall metrics, and general behavior, I can still accept a Return / DD around 4–5 in the 80–85% confidence zone.

If I were too strict on Monte Carlo alone, I would end up throwing away too many systems that are still good enough on paper and often still worth testing further.

What I want to see

A good system:
Profit decreases gradually
Drawdown increases in a controlled way
SQN declines but stays relevant
Behavior remains consistent

A bad system:
Profit collapses quickly
Drawdown spikes aggressively
SQN breaks down
Equity structure changes completely

Number of simulations

In my workflow, I use 50 Monte Carlo simulations as standard.

That is my default test.

For me, 50 runs are already enough to judge whether a strategy is structurally stable or not.

Sometimes, after a system passes well on 50 runs, I do an extra test with 100 simulations.

Not because 50 is suddenly not enough.
Not because I want to make the test artificially stricter.
And not because I am looking for perfection.

I do it for one reason only:

to see whether anything changes structurally when the number of simulations is increased.

If the behavior stays broadly the same at 100 runs, that gives me extra confidence that the result at 50 runs was not just noise.

So to be clear:

50 runs = my standard Monte Carlo test
100 runs = optional extra confirmation

The exact number is not the most important factor.

What matters most is whether the system behaves consistently when stress is increased.

Important note

Monte Carlo is not a guarantee.

Passing Monte Carlo does not mean a system is ready for live trading.

It only means:
the system has survived a realistic stress test

It still needs:
Out-of-Sample validation
demo phase
live validation

Final perspective

Monte Carlo is one of the core filters in my workflow.

If a system fails here clearly,
and the rest of the metrics are not convincing,
I discard it immediately.

There is no reason to continue with a weak structure.

Over time, your eye becomes trained.
You start to recognize stability patterns very quickly.

But that intuition always starts with a solid Monte Carlo foundation.

This is not about finding perfect systems.

It is about filtering out fragile ones before they ever reach demo or live trading.

Monte Carlo example (real system)

Below you can see two Monte Carlo tests of one of my systems (EA 628):

50 simulations (my standard test)
100 simulations (extra confirmation)

This is not a random system.

This robot has already been:

backtested (tick data MT4)
tested in EA Studio
validated on demo
and traded live

So what you are seeing here is not theory
this is a system that has already proven itself, now being stress-tested.

What stands out immediately

If you look at the simulation charts, you can see:

All curves follow the same general structure
No chaotic divergence
No random collapse scenarios
The system keeps its shape under stress

That is exactly what you want to see.

A fragile system would show:

wide spread between curves
inconsistent behavior
completely different equity structures

That is not the case here.

Confidence table interpretation

Now the most important part: the confidence table

What matters is not the exact numbers, but the structure.

You can clearly see:

Profit decreases gradually
Drawdown increases in a controlled way
Return / DD stays relatively stable
SQN declines slightly but remains meaningful

This is what I call a stable degradation curve.

50 runs vs 100 runs

This is where it becomes interesting.

When moving from 50 → 100 simulations:

The structure remains the same
No new weaknesses appear
No sudden breakdown in performance
The behavior is consistent

That confirms that the 50-run result was not random noise.

This is exactly why I sometimes run 100 simulations:
not to make the test stricter,
but to confirm that the structure holds.

Key takeaway

Monte Carlo is not about perfect numbers.

It is about structure.

This system shows:

consistency
controlled degradation
stability under stress

That is what robustness looks like in practice.

Also keep in mind:

This is a system that already proved itself in:

backtest
demo
live trading

Monte Carlo is not used here to “find” something.

It is used to confirm that the structure remains intact under variation.