Take a two-year backtest with 300 trades and a maximum drawdown of 8 percent. Now shuffle those exact 300 results into a different order and plot the curve again. Do it a thousand times. In a decent share of those runs the drawdown will not be 8 percent, it will be 14 or 18, because the losers clustered differently. Nothing about the strategy changed. Only the order did.
That gap is the whole point of Monte Carlo testing. The historical sequence is one draw from a distribution of possible sequences, and it is the draw you happened to fit your expectations to. If your position sizing survives only that one arrangement, it does not survive.
What the method actually does
Start with the trade list from a backtest or, better, from live results: entry, exit, result in R or in account percentage. Monte Carlo then generates many alternative histories from that list. The common variants are worth separating, because they answer different questions.
Reshuffling keeps every trade exactly once and only changes the order. This tests sequence risk: how bad the losing runs could have been. Resampling with replacement, the bootstrap, draws trades at random and allows repeats, which produces series that contain more or fewer winners than history did. That tests both sequence and composition. A third variant randomises the parameters slightly, entry price by a tick, exit by a few pips, to see whether the result depends on execution being perfect.
The output is not a single number. It is a distribution: the 5th percentile drawdown, the median final equity, the share of runs that ended below the starting balance. Read percentiles, not averages. An average outcome tells you nothing about whether you would have survived the bad tail long enough to reach it.
Sequence risk is the real finding
Most traders size positions against the drawdown they saw in the backtest. Monte Carlo consistently shows that number is optimistic, often by a factor of one and a half to two at the higher percentiles. Losses do not arrive politely alternated with wins. They cluster, and the cluster that did not happen in your sample is entirely capable of happening in your first month live.
This matters most for anyone trading against a hard floor. A funded account with a fixed loss limit does not care about your expected value. It cares whether one bad run crosses a line, and once it does the account is gone regardless of what the next hundred trades would have delivered. Anyone working through prop firm drawdown rules should be running the reshuffle before choosing a risk per trade, not after failing a challenge.
Monte Carlo cannot invent risk your sample never contained. If your trade list has no trades from a liquidity shock, no reshuffle of it will produce one. It stress-tests ordering and composition, not regime change.
Where the method quietly lies
The standard reshuffle assumes trades are independent. Many are not. If your system takes correlated positions, several majors long against the dollar on the same signal, then history's "trades" are not separate draws at all, and shuffling them breaks up clusters that in reality arrive together. That makes the simulation more optimistic than the truth, which is the opposite of what you wanted. The fix is to group correlated positions into a single event before sampling, or to sample by day rather than by trade.
The second problem is the source data. A Monte Carlo run on an overfitted backtest produces beautifully rigorous nonsense: you are stress-testing a curve that was fitted to noise. Run it on out-of-sample results or on live trades wherever you can. The technique tests whether your sizing survives a bad run. It says nothing about whether an edge exists.
Third, sample size. A hundred trades is not enough to describe a distribution. The percentile estimates will swing wildly depending on which hundred you had. Treat anything under a few hundred trades as directional at best.
Reading the output like a risk manager
Three numbers earn their place in a review.
The 95th percentile maximum drawdown answers "how deep does this go in a bad but not absurd run". Size so that this number sits comfortably inside your tolerance, and inside any account rule you trade under. The share of runs ending in a loss answers whether the edge is thin enough that ordering alone decides the year. And the longest losing streak across runs answers the psychological question, because the streak that breaks a trader is usually shorter than the one their system can produce. That connects directly to trading psychology, since the plan you abandon is not a plan.
If the 95th percentile drawdown is double your comfort level, the honest response is to cut risk per trade until it is not. Halving position size halves that percentile roughly proportionally, and it changes nothing about whether the edge is real.
Tools and how this fits the rest of testing
You do not need specialist software. A trade list exported to a spreadsheet and a short script that shuffles a column and tracks a running peak will do the job, and writing it yourself forces you to understand what is being randomised. Several journal and analytics packages include a Monte Carlo tab, and MT5's own tester can be pushed in this direction with forward runs, though most traders find the export-and-script route clearer. The wider tool set is covered in backtesting tools.
Order of operations matters. Backtest to see whether the idea has any expectancy at all. Check for overfitting by holding out data. Then run Monte Carlo on what survives, to set position size. Then run it forward on demo or small size, as in forward testing, because the only distribution that finally counts is the one your live execution produces, spread, slippage and hesitation included.
All trading carries a high risk of loss, and no simulation removes that. What Monte Carlo removes is the illusion that the one curve you have seen is the one you will get.
"The backtest shows you the history you got. Monte Carlo shows you the histories you could just as easily have got, and that is the one you size your account against."
— Alex Onta, Executive Director, SINGUARD
Key Takeaways
- A single backtest curve is one sequence out of many, and its drawdown figure is usually the flattering one.
- Reshuffling tests sequence risk, bootstrapping tests composition, and both should be read as percentiles rather than averages.
- Correlated positions break the independence assumption, so group them into one event before sampling.
- Use the 95th percentile drawdown to set position size, especially when trading against a hard account loss limit.
Frequently Asked Questions
How many Monte Carlo runs are enough?
A thousand runs is usually enough to stabilise the middle percentiles, and ten thousand tightens the tails. The bigger constraint is the input, since percentile estimates from a trade list of under a few hundred trades will move a lot depending on which trades you happened to record.
Can Monte Carlo prove a strategy will be profitable?
No. It reorders and resamples results you already have, so it inherits any bias in that sample. If the underlying backtest was overfitted, the simulation stress-tests a fitted curve. Its proper use is sizing and survival analysis, not proof of edge.
Why do reshuffled runs show a worse drawdown than the original backtest?
Because losses cluster differently in different orderings. Your historical sequence happened to space its losers in one particular way. Other orderings put several of them back to back, which deepens the equity trough even though the set of trades is identical.
About the Author
Alex Onta is an Executive Director at SINGUARD. He built eTrader, the terminal, the mobile apps, eTrader Broker, Copytrading, Business and Community, along with the worldwide clustered-server infrastructure it all runs on, with his brother Roman Onta helping on the design, and he leads that division today. Together with Roman he builds the Prop Firm CRM, the Broker CRM, Scalegram and CopySignals, and the two of them carry worldwide compliance, payment processing and international business structuring side by side. He lives and works in Dubai for most of the year. Meet the executive duo leading Singuard's five divisions.