Monte Carlo simulation
A Monte Carlo simulation tests a result against a large number of randomised alternatives instead of taking it at face value. In trading, that means comparing your strategy against thousands of random versions of itself — if random entries do just as well, the edge was luck.
In plain terms
Any single backtest is one path through history. It tells you what happened with this strategy on this data, and nothing about what would have happened by chance. Monte Carlo methods fill in the missing comparison: run the same test hundreds or thousands of times with something randomised — entry timing, trade order, market path — and see where the real result lands in that distribution.
The version that matters for strategy validation is the randomised-entry control. Take the strategy's real trades, and for each one build a substitute trade on the same day and instrument with the entry time drawn at random, the direction decided by a coin flip, and the same stop distance, target and costs. If those random trades perform about as well as the real ones, the entry signal — the thing the whole strategy is built around — is not adding anything.
Why it matters
A profitable backtest with no random benchmark is an unfalsifiable claim. Of course it made money — so might a coin flip in a rising market. Without a comparison set of random alternatives, there is no way to tell a genuine edge from a lucky sequence.
This is the same logic as a placebo arm in a drug trial. The question is never just "did the patient improve" but "did the patient improve more than they would have without the treatment". Monte Carlo simulation is how that second question gets asked in a backtest, and it is routinely skipped in retail trading because the honest answer is often unflattering.
How Atlas Edge handles it
Every rigor check includes a Monte Carlo randomised-entry comparison, run as five seeded draws within each walk-forward window and pooled across all of them — so a verdict built on hundreds of real trades is compared against thousands of random ones. The seeds are fixed, which makes the result deterministic: re-run the same strategy and you get the identical placebo figure, not a fresh random opinion.
The comparison is disclosed inline in every verdict rather than folded into a black-box score, so you can see exactly what was compared and how before deciding whether to trust the result.
Answers to the usual ones
- Is Monte Carlo simulation the same as a permutation test or bootstrap?
- No. Permutation tests shuffle the observed data and bootstraps resample from it. The Monte Carlo control used here generates fresh random trades with matched risk and costs, which answers a more direct question: does the entry signal beat random timing?
- Why use fixed seeds instead of fresh randomness each run?
- Reproducibility. A verdict that changes every time you re-run it is not a verdict. Fixed seeds mean the same strategy always produces the same comparison, and any disagreement about the result can be checked rather than re-rolled.