Your backtest produced one equity curve. It looks fine. What it doesn't tell you is that the same strategy, with the same trades in a different order, could have handed you a drawdown twice as deep — and that order was pure luck.
Monte Carlo simulation is how you see the other outcomes.
The idea
Take your trade results. Shuffle them into a random order. Build the equity curve. Do it again. Do it ten thousand times. You now have a distribution of what this strategy could plausibly have done, rather than the single history you happened to live through.
What comes out is genuinely uncomfortable. A system with a 12% worst drawdown in backtest routinely shows 25% or worse across the simulations. Not because the strategy changed — because your actual history got a lucky ordering, and you'd been treating that luck as a property of the system.
The drawdown in your backtest isn't your worst case. It's one sample from a distribution you never looked at.
What it answers
How bad can the drawdown get? The one that matters for sizing, and the one nobody plans for honestly (drawdown).
What's the chance of ruin? Run the simulations, count how many hit zero or your quit-point. That's your risk of ruin, measured rather than assumed.
How long can a losing streak run? Knowing the answer in advance is the difference between sitting through the streak and abandoning a working system in the middle of it.
How wide is the range of outcomes? If the simulations spread from +40% to -30%, your "expected" return is a lot less meaningful than it looked.
The honest caveats
Monte Carlo is a stress test, not a crystal ball, and it has real limits.
Shuffling assumes trades are independent. They usually aren't. Losses cluster because market conditions cluster — a bad regime produces a run of losses together, not scattered evenly. Simple reshuffling can understate that clustering, which means it may understate your real drawdown.
It can only resample what you gave it. If your history never contained a 2008 or a 2020, no amount of shuffling will invent one. Monte Carlo explores the outcomes your data implies, not the ones it never saw.
Garbage in, confident garbage out. Run it on an overfitted backtest and you get a beautifully rigorous distribution of a fantasy.
Using it properly
The practical value isn't the headline number. It's the humility. You size for the drawdown the simulations show, not the one your backtest happened to produce. You decide in advance what losing streak you'll sit through, because you've seen how long they run. And you stop mistaking one lucky path for a property of your system.
That's the whole point: to meet your worst case in a spreadsheet, at leisure, instead of live, at speed, with money on the line.
See also: reading your equity curve and variance vs edge.