LEARN · EN · easyquanttrading.com

Monte Carlo simulation for traders: what it tests and what it does not

You ran one backtest and got one equity curve. But the same rules, with the same trades in a different order, produce a different curve — sometimes a dramatically different one. Monte Carlo simulation is the tool for seeing that range instead of a single point, and it is easy to over-read.

The mechanic

Take the list of trade results from your backtest. Shuffle them into a random order. Recompute the equity curve and record the maximum drawdown. Repeat a thousand times. You now have a distribution of drawdowns for a strategy whose trades were exactly as profitable as the backtest, just sequenced differently.

That single change of perspective is why the method is popular. It answers a question a single curve cannot: how much of the drawdown I saw was luck of ordering, and how bad could it plausibly have been?

What a wide distribution tells you

If the thousand simulated drawdowns cluster tightly around the observed one, the strategy's risk profile is not very sensitive to sequence. If they spread widely — with a meaningful tail well beyond anything in the backtest — then your observed drawdown was one draw from a wide range, and the next one could be considerably worse.

This is genuinely useful information and it costs almost nothing to compute. It is also routinely misreported as a confidence interval on future returns, which it is not.

The assumptions it makes

Monte Carlo on trade sequences makes two assumptions that are worth stating out loud, because both are usually violated.

AssumptionWhy it often fails
Trades are independentReal strategies cluster: losing streaks happen in specific conditions
The sample is representativeIf the backtest period missed a crisis, no shuffle will invent one
Trade sizes are fixedCompounding or risk-based sizing changes every curve
No path dependenceStops, margin calls and time exits depend on the path, not just the trades

Four assumptions. The first two matter most, and neither can be fixed by adding more simulations.

What it does not do

It does not make your strategy more likely to work. It does not add information that was not in the trade list. It does not test whether the trades themselves are representative — if the backtest missed a regime, the simulation inherits that omission and cannot reveal it.

In particular, a Monte Carlo run cannot warn you about a strategy whose edge disappears in conditions that were absent from the sample. It will happily produce a tight, reassuring distribution around a result that depended entirely on one favourable period.

How to use it well

  • Report the distribution, not a single number. 'Median simulated drawdown 12%, 95th percentile 26%' is informative; 'expected drawdown 12%' is not.
  • Compare the 95th percentile against what you could tolerate, rather than comparing the median to what you hope for.
  • Run it again after any change to sizing, because sizing changes the whole distribution.
  • Treat it as one check among several, and never as a substitute for out-of-sample testing.

FAQ

How many simulations are enough?
A few thousand is usually plenty for the shape of the distribution to stabilise. Beyond that you are refining a number whose main uncertainty comes from the sample itself, not from simulation noise.
Is Monte Carlo the same as bootstrapping?
They are related. Shuffling without replacement is a permutation test; sampling with replacement is a bootstrap. Both produce distributions of outcomes from a single trade list, and both inherit the same assumption that the trade list is representative.
Does a tight Monte Carlo distribution mean the strategy is safe?
No. It means the result is not very sensitive to the order of the trades you happened to observe. It says nothing about whether those trades resemble the ones you will get next.

More guides

Not investment advice. Historical results do not guarantee future performance. EasyQuant is a research factory — you execute on accounts you control.