Monte Carlo simulation for trading strategies, and what the range doesn't tell you

What is Monte Carlo simulation in backtesting?

Monte Carlo simulation in backtesting reshuffles or resamples a strategy's own trade sequence, then re-runs the equity curve thousands of times to show the range of outcomes the same edge could have produced by chance alone. It measures how much of a result depends on trade order rather than the edge itself.

The inputs are usually the trade list a backtest already produced: the sequence of wins and losses, or the daily return series. Reshuffling that sequence, or drawing from it with replacement, and replotting the equity curve each time turns one line into a distribution of lines, typically summarised as a percentile band around the original.

Is Monte Carlo simulation the same as walk-forward testing?

No. Walk-forward tests whether a rule holds up on data it was not tuned on, moving forward through calendar time. Monte Carlo instead reshuffles trades a strategy already produced to size how much of the equity curve was sequence luck. The two answer different questions and neither substitutes for the other.

A strategy can pass walk-forward cleanly and still owe most of its headline drawdown figure to one unlucky ordering of trades, and a strategy can show a tight Monte Carlo band while still being tuned to noise, because resampling the same 40 in-sample trades a thousand times never tests whether those 40 trades were the right 40 to trust. Seewalk-forward vs CPCVandin-sample vs out-of-samplefor what each structural test actually proves.

Which platforms actually run it

Crossing two assessed fields that have never been crossed on this site before: whether a platform runs Monte Carlo resampling of trades or the equity curve at all, and whether it pairs that resampled distribution with a deflated or probabilistic Sharpe, or a formal overfitting metric, so a reader gets a significance read rather than a chart to eyeball.

* Commercial relationship with this site, scored on the same published rubric as every other platform. Disclosure

What doesn't a Monte Carlo result tell you?

A Monte Carlo result does not tell you whether the strategy was fit to noise in the first place, because it only resamples the trades that survived selection, never tests the trades that were tried and discarded, and says nothing about whether the original rule was the right one to keep out of everything else attempted.

1. It does not correct for the number of variants tried.Reshuffling the winning rule's trades a thousand times says nothing about the rules that were tried and discarded first. Seeprobability of backtest overfittingfor the test that does.
2. It does not detect a leak. If the underlying trade list was produced with look-ahead bias or an unrealistic fill model, resampling that list faithfully reproduces the leak in every path. Seelook-ahead biasandexecution realism.
3. A tight band is not proof of skill. A strategy with very few trades can show a deceptively tight resampled range simply because there is little to reshuffle, which is the opposite of what a tight band is usually read to mean.
4. Reshuffling and block bootstrapping are not interchangeable.Plain reshuffling assumes trades are independent; returns with autocorrelation or regime clustering need a block bootstrap that preserves runs, or the resampled range understates how correlated real drawdowns can get.

Frequently asked questions

Related

• Backtest overfitting: how to detect it
• Walk-forward vs CPCV, and what each one proves
• Probability of backtest overfitting
• Statistical validity of a backtest
• In-sample vs out-of-sample