Monte Carlo Analysis for Trading Systems

Monte Carlo analysistakes results a strategy has already produced — usually the list of closed trades from a backtest — and re-runs them thousands of times with the order randomized or the trades resampled. Each run rebuilds the equity curve from the same underlying statistics, and across all the runs the single historical line turns into a distribution of paths.

The reason this matters is that trade sequencedrives drawdown. The total profit of a fixed list of trades does not depend on their order, but the journey does: six $200 losses scattered across a year are six shallow dips, while the same six losses in a row are a single $1,200 drawdown — same trades, same total, very different experience and very different margin pressure along the way.

A run is mechanical: take the trade results, randomize them, replay them one by one, and record the path. Repeat until the statistics stabilize — typically 1,000 to 10,000 runs — then aggregate.

The randomization method matters. A pure shufflere-orders the exact historical trades, so every run ends at the same final profit and only the path varies — the cleanest way to isolate sequence risk. Resampling with replacement draws trades at random from the list, repeating some and skipping others, so both the path and the ending vary. That asks a broader question: what if the future is statistically like the past without literally repeating it? Percentage-based position sizing also spreads the endings, because the same trades compound differently depending on when they land.

The output of a Monte Carlo pass is not a verdict but three families of numbers, each answering a different question about the same strategy:

The last of these connects directly to risk of ruin: instead of a formula built on assumptions, the simulation simply counts how often the replayed history crossed the level you could not tolerate.

A single number — “max drawdown 9%” — invites false precision. The simulated distribution is read in percentiles: the median is the typical case, and the 90th or 95th percentile is the plausible bad case the same statistics could have produced.

The practical shift is in what you plan around. If only the most favorable orderings keep the account above your pain threshold, the strategy is not robust — it was lucky. You can run these numbers on your own inputs with the free Monte Carlo Trading Simulator.

This is the method’s real job. A backtest report presents one path as if it were the strategy, but that path is one draw from a distribution the report never shows. Monte Carlo asks how fragile the conclusion is: if reshuffling the same trades routinely produces drawdowns twice as deep as the reported one, the smooth reported curve was partly an accident of ordering.

It pairs naturally with the other checks covered in why backtests can be misleading: out-of-sample testing probes whether the statistics are real, while Monte Carlo probes how much pain those statistics can deliver even if they are. A strategy can pass the first test and still fail the second — positive expectancy with a 95th-percentile drawdown no realistic account size survives.

Every simulated path is assembled from trades that already happened. The method rearranges history; it does not imagine anything outside it:

• It assumes the input statistics persist— the same win rate, average win and average loss going forward. A regime change invalidates every percentile.
• It cannot conjure unseen risks: a loss larger than any in the sample, a spread blowout, a weekend gap or a liquidity event leaves no trace in the input and therefore none in the output.
• It inherits the backtest’s flaws. Overfit parameters, optimistic costs or bad data flow straight through into a precise-looking distribution.
• Independent resampling ignores correlation between trades— losing streaks caused by a persistent market condition can be longer than random ordering suggests.