Monte Carlo Simulation for Trading Strategies: What It Tells You That a Backtest Doesn't

A backtest gives you exactly one path through history. Your winning trades happened in a specific order, your losing streaks clustered where they clustered, and your equity curve drew exactly one shape. Monte Carlo simulation's whole job is to ask the question your single backtest can't answer: if the same edge produced the same trades in a different order, how bad could it have gotten? That's it. It's not a crystal ball, it's not a forecast, and it doesn't make a weak strategy strong — but almost every retail trader either skips it entirely or massively overrates what it tells them once they do run it.
What Monte Carlo Simulation Actually Does
Take the list of trade results from your backtest — say, 400 trades with their individual P&L. A Monte Carlo simulation on a trading strategy typically does one of two things with that list: shuffles the order of those same trades thousands of times (sequence/order randomization), or resamples with replacement from that same pool (bootstrap resampling, i.e. random sampling with replacement) to generate thousands of alternate equity curves built from the exact same statistical distribution of outcomes your strategy produced. Run it 5,000 or 10,000 times and you get a distribution: a range of maximum drawdowns, a range of terminal equity values, a range of longest losing streaks — instead of the single number your backtest report gave you. Each of those 10,000 runs is effectively one of the alternative histories your strategy could have lived through with the exact same edge; the real backtest is just the one history that actually happened to occur.
This directly answers the sequence-risk question. A strategy that shows a 18% max drawdown in its one historical backtest might show a 95th-percentile Monte Carlo drawdown of 35-40% once you shuffle trade order a few thousand times. That's not a flaw in the strategy — it's the same edge, same win rate, same average win/loss, just experienced in a less favorable order. If your position sizing was built around the 18% number, you were sizing for a lucky path, not for the actual risk the strategy carries.

The Specific Numbers Monte Carlo Gives You
A properly run simulation on a decent-sized trade sample (200+ trades minimum; fewer than that and you're mostly resampling noise) typically outputs:
Drawdown percentiles — median expected drawdown, 90th percentile, 95th/99th percentile ("worst realistic case" bands). Risk of ruin — the percentage of simulated paths that breach a defined loss threshold (this is where prop firm traders should be looking before picking an account size, not after). Distribution of final equity — not just "did it make money" but how wide the spread of outcomes is around that average, and how skewed that spread is (return asymmetry matters here: a strategy with a few huge winners and frequent small losses produces a very different Monte Carlo distribution shape than one with steady small gains, even at an identical average return). Longest losing streak distribution — useful for psychological prep and for sizing so a bad streak doesn't force you out of the strategy right before it recovers.
None of these numbers existed in your original backtest. The backtest gave you one drawdown figure. Monte Carlo gives you the shape of the distribution that single figure was sampled from.
Where People Oversimplify This (and It Costs Them)
The biggest myth floating around trading forums and prop-firm blogs is some version of "run a Monte Carlo simulation to see if your strategy will pass the evaluation." That's backwards. Monte Carlo simulation doesn't test whether a strategy is good — it tests how much variance a given edge carries. If the underlying edge is weak or the sample is small, running 10,000 shuffles of a bad or barely-profitable strategy just gives you a very precise picture of how a bad strategy loses money in different orders. It will not turn 50 lucky trades into a validated edge, and it will not catch overfitting, lookahead bias, or curve-fitting — those are backtest construction problems, and Monte Carlo simulation operates entirely downstream of them, on the trade list your (possibly broken) backtest already produced.
This is the exact point researchers like Marcos Lopez de Prado and David Bailey have hammered on with tools like the Probability of Backtest Overfitting and the Deflated Sharpe Ratio: if you tested 50 parameter variations and cherry-picked the one with the best Sharpe ratio, no amount of downstream Monte Carlo resampling fixes the selection bias baked into that result. Garbage in, precisely-shuffled garbage out. A firm that proudly tells you "we ran 1,000 simulations" on a model built on flawed historical assumptions hasn't de-risked anything — they've just run the same flawed assumption 1,000 times and called the resulting spread "robustness." The same caution applies to monthly returns tables built from simulation analysis: a smooth-looking distribution of monthly returns across 1,000 runs tells you about variance in the model, not about whether the model's underlying edge is real.
The second common mistake is treating shuffled-order Monte Carlo and bootstrap-resampled Monte Carlo as interchangeable. Pure order shuffling preserves your exact trade list and only randomizes sequence — it answers "how bad could sequence risk alone get." Bootstrap resampling (drawing trades with replacement) also changes trade frequency and can duplicate or drop individual trades — it answers a slightly different question, closer to "what if my sample of outcomes was itself just one draw from a larger population." Mixing these up, or not stating which one a report used, is how two traders can show wildly different "Monte Carlo drawdown" numbers for strategies with identical trade histories.
How to Actually Use It
Run it after your backtest is already validated out-of-sample and walk-forward tested, not as a substitute for that validation. Use a trade sample large enough to mean something — 200+ trades is a reasonable floor, under 100 and your percentile bands are mostly sampling error dressed up as statistics. State explicitly which method you used (shuffle vs. bootstrap) and at what confidence level you're reading drawdown (the 95th percentile number, not the median, is the one that should drive position sizing). Size your account and risk-per-trade around the 90th-95th percentile drawdown figure from the simulation, not the single number your original backtest happened to produce — that's the entire practical payoff of doing this analysis in the first place. And treat risk-of-ruin output as a hard sizing constraint: if 8% of simulated paths breach your account's max drawdown limit before reaching profit target, that's not an edge case to shrug off, that's one in twelve realistic outcomes ending in a blown account.
The Honest Bottom Line
Monte Carlo simulation is a risk-sizing tool, not a strategy-validation tool. It takes the one path your backtest handed you and shows you the family of paths that same statistical edge could plausibly have produced, which is genuinely useful for setting realistic drawdown expectations and avoiding the trap of sizing a live account around your best-case historical run. What it can't do is rescue an overfit strategy, validate a small sample, or substitute for proper out-of-sample testing. If someone's pitching you a strategy and the main evidence is "we ran a Monte Carlo simulation and it looked solid," ask what the underlying trade sample size was and whether the backtest it was built from was walk-forward tested first — the simulation is only as honest as the data you feed it.