A backtest is a single path through history. Your EA ran across those specific years, in that specific sequence of price movements, and produced those specific results. It is one trajectory through an enormous space of possible trajectories, and there is no particular reason to believe it is representative of what the strategy will encounter in live trading.
Monte Carlo simulation is the method that addresses this directly. Instead of treating the single historical path as the truth about your strategy’s behavior, it generates hundreds or thousands of alternative paths from the same underlying trade distribution and asks: across all these possible realities, what does the risk profile actually look like? What is the realistic worst case, not the worst case that happened to occur in this specific backtest, but the worst case that the strategy’s statistical properties can plausibly produce?
This distinction matters more than most traders realize. A backtest showing a 7% maximum drawdown does not mean the strategy has a 7% maximum drawdown. It means the strategy produced a 7% maximum drawdown in one particular sequence of market conditions. The Monte Carlo distribution of that same strategy might show a P95 drawdown of 19%. Under 95% of plausible outcome sequences, the drawdown stays below 19%, but under the remaining 5%, it goes further. That 5% is not hypothetical risk. It is the concrete risk you carry every time you fund the EA on a live account.
This guide covers the complete methodology: how Monte Carlo simulation works, the two fundamentally different approaches and when to use each, how to interpret results correctly for MT4, MT5, and cTrader backtests, what the numbers actually mean for position sizing, and how to combine MC results with binomial significance testing to build a complete statistical picture of any EA.
What Monte Carlo Simulation Actually Does
The core idea is straightforward. Your backtest produces a list of trade outcomes. Each trade has a profit or loss, an entry and exit time, and a duration. Monte Carlo simulation takes that list and asks: what if these same trades had occurred in a different order?
This is the resampling method, and it is the correct approach for EA validation. You shuffle the sequence of trade outcomes, calculate the equity curve and drawdown statistics for that new sequence, then repeat the process, typically 1,000 to 10,000 times. The result is a distribution of outcomes, not a single outcome. Instead of one maximum drawdown figure, you get a full probability distribution of maximum drawdowns across all simulated sequences.
From that distribution, you extract percentile statistics. The P50 drawdown is the median. Half of all simulated sequences produce a drawdown above this level, half below. The P95 drawdown is the level below which 95% of all sequences fall. The P99 drawdown is the level that only the worst 1% of sequences exceed. These percentile figures give you a statistically grounded range of outcomes rather than the single number your backtest happened to produce.
The mathematical justification is straightforward. If your EA has no look-ahead bias and the trades are independent, then the order in which they occurred historically is essentially arbitrary. The same trades could have happened in any sequence, and the strategy’s edge (or lack of it) would be identical. What changes with the order is the path-dependent risk: specifically the depth and duration of drawdowns. Resampling reveals the full distribution of that path-dependent risk.
Resampling vs Generated Data: Which Method to Use
There are two distinct approaches to Monte Carlo simulation in trading, and confusing them leads to misinterpreted results.
The resampling method takes the actual trade outcomes from your backtest and shuffles their order. It preserves the exact distribution of win sizes, loss sizes, and win rate. The fundamental statistical properties of the strategy remain identical. Only the sequence changes. This is the correct method for answering a key question: “given that my strategy produces these kinds of trades, what is the realistic range of drawdown outcomes?” It is the method used by professional-grade tools including the ErgodicLabs Monte Carlo tool, which implements the percentage-based resampling approach that matches Quant Analyzer’s exact methodology.
The generated data method uses the historical return distribution (mean and standard deviation) to generate entirely synthetic price series, then runs the strategy on those synthetic series. This approach is more common in academic research and derivative pricing but is less appropriate for retail EA validation. The reason: it assumes returns are normally distributed, which they are not. Real EA trade outcomes have skew and kurtosis (fat tails, occasional large losses or large wins) that a normal distribution model systematically underestimates. Applying a normal generator to a strategy with fat-tailed returns will underestimate the true tail risk, and that underestimation is most dangerous at exactly the point where it matters.
For practical EA backtesting validation, use the resampling method. It makes no distributional assumptions. It uses your actual trade outcomes exactly as they occurred, just reordered. The only assumption required is that the trades are approximately independent, which is worth verifying separately but holds for most systematic strategies without grid or martingale behavior.
Percentage-Based vs Fixed-Dollar Resampling
Within the resampling method, there is an important implementation distinction that most articles never address: whether to resample in percentage terms or in fixed dollar terms.
Fixed-dollar resampling takes each trade’s profit or loss in absolute currency terms and shuffles those figures. This is simple but introduces a distortion: it assumes all trades were taken at the same account equity level, which is not true. A $500 loss when the account stands at $10,000 has a different percentage impact than a $500 loss when the account has dropped to $8,000 during a drawdown.
Percentage-based resampling converts each trade to its percentage return on account equity at the time it was taken, then shuffles those percentages and compounds them sequentially. This correctly models how equity curves actually behave, since each trade’s impact depends on the account level at the time it occurs. The resulting drawdown distribution is more accurate, and for strategies with significant drawdown depth, the difference between the two methods is material.
The practical implication: if your Monte Carlo tool reports drawdown figures in absolute terms without specifying its resampling method, treat the results with caution. Percentage-based resampling is the correct approach, and it is what the ErgodicLabs tool implements.
Running Monte Carlo on MT4, MT5, and cTrader Backtests
The input data for Monte Carlo simulation comes from your strategy tester’s trade history. Each platform exports this differently, and the format matters for correct analysis.
MT4 backtests export an HTML report from the Strategy Tester containing a complete trade list. Each row includes ticket number, open time, type, lots, symbol, open price, close time, close price, profit, and balance. The profit column is what Monte Carlo simulation operates on. A key validation step before running simulation: check that no trades show identical open and close times with zero drawdown, which indicates a backtest run with insufficient tick data, which invalidates the results regardless of Monte Carlo analysis.
MT5 backtests export similar HTML reports but include additional detail: the separate “Deals” tab showing individual position entries and exits, and the “Orders” tab showing pending order history. For Monte Carlo purposes, use the Deals data rather than the Orders data, since it shows actual executed fills at real prices. MT5 also exports in XML format, which is more reliable for automated parsing. One MT5-specific issue to watch: the breakeven trade filter. MT5 sometimes logs trades closed at exactly breakeven (zero profit/loss) as separate entries. These should be excluded from Monte Carlo analysis because they contribute no information to the return distribution and inflate the apparent trade count.
cTrader backtests export trade history in CSV format from the History tab in the cTrader platform. The export includes Net Profit (after commission and swap), Gross Profit, Commission, and Swap columns separately. For Monte Carlo analysis, always use Net Profit, not Gross Profit. The difference is significant for high-frequency scalping strategies where commission and swap costs can represent 30 to 50% of gross profit. Using gross profit produces an optimistic Monte Carlo distribution that does not reflect the actual returns a live account would produce.
cTrader also provides a built-in Optimization results export, which is useful for walk-forward analysis but should not be used as input for Monte Carlo simulation, since it contains aggregated results across parameter sets rather than individual trade outcomes. The individual trade history from a single backtest run is the correct input.
How to Read Monte Carlo Results Correctly
The output of a Monte Carlo simulation is a distribution of outcomes, typically visualized as a histogram or percentile table. Reading these results correctly requires understanding what each figure represents, and critically, what it does not represent.
The P50 drawdown is the median outcome. In a 1,000-simulation run, 500 sequences produced a drawdown below this level and 500 above. If the P50 drawdown is 8%, you should expect that in roughly half of all plausible live trading periods of comparable length, the strategy will experience a drawdown of 8% or more. This is not a worst case. It sits at the middle of the distribution.
The P95 drawdown is the figure most relevant to practical risk management. It represents the drawdown level that 95% of all simulated sequences stay below. If the P95 drawdown is 18%, then under 5% of plausible outcome sequences, the strategy experiences a drawdown exceeding 18%. Use this figure when setting account risk limits, funding levels, and drawdown halt thresholds. If you are not prepared to experience an 18% drawdown, this strategy should not be funded regardless of what the historical backtest showed.
The P99 drawdown represents the tail risk: the level exceeded only by the worst 1% of sequences. For strategies you intend to run for multiple years, the P99 figure is the relevant reference for catastrophic risk assessment. A strategy with a P99 drawdown of 35% will, in approximately one in a hundred comparable trading periods, experience a 35% drawdown. That is not an event you can dismiss as impossible. It is an event you should explicitly plan for.
The drawdown expansion ratio (the P95 drawdown divided by the historical maximum drawdown) is one of the most useful single numbers produced by Monte Carlo analysis. A ratio close to 1.0 means the historical drawdown was a reasonably representative worst case. A ratio of 2.0 or above means the historical backtest significantly underestimated the strategy’s actual drawdown risk. Ratios above 2.5 are a serious warning sign, often pointing to a small trade sample where the historical path happened to produce an unusually mild drawdown sequence, or to path-dependent behavior like partial averaging or multi-level grid logic.
The Binomial Layer: Combining MC with Statistical Significance
Monte Carlo simulation answers the question of drawdown risk. It does not answer the question of whether the strategy’s edge is statistically real. These are separate questions, and both need answering before capital is committed.
The binomial test addresses edge significance directly. Given the strategy’s win rate and total trade count, it calculates the probability that the observed win rate is attributable to random chance rather than genuine predictive edge. A p-value below 0.05 means the win rate is unlikely to be random at the 95% confidence level. Below 0.01 means it is unlikely at the 99% level.
Running both tests gives a two-dimensional view of the strategy that neither test provides alone:
A strategy can pass the binomial test with a statistically significant win rate while failing Monte Carlo analysis because the drawdown distribution is too wide or too unpredictable. This pattern typically indicates a strategy with real edge but insufficient sample size. The edge is detectable, but there is not enough data to constrain the tail risk estimates reliably.
Conversely, a strategy can show a tight, well-behaved Monte Carlo distribution while failing the binomial test. This is the more dangerous pattern. The smooth equity curve and controlled drawdown statistics look reassuring, but the win rate itself is not distinguishable from random. This is the signature profile of a martingale or grid EA: precisely controlled drawdown appearance, no statistically defensible edge in the win rate signal.
A strategy that passes both tests (statistically significant win rate at p < 0.01, and a P95-to-historical drawdown expansion ratio below 2.0) is in the minority of all backtests that circulate publicly. Most fail at least one of these tests. Most fail both.
What the Monte Carlo Results Cannot Tell You
Monte Carlo simulation is a powerful tool with specific, well-defined limitations. Understanding them is as important as understanding the method itself.
It assumes trade independence. The resampling method shuffles trade outcomes and compounds them. This is only valid if the trades are approximately independent, meaning the outcome of one trade does not systematically depend on the outcome of preceding trades. For martingale and grid EAs, this assumption is violated by construction because lot sizes scale with prior outcomes. Monte Carlo simulation run on a martingale strategy produces a wildly optimistic distribution because it does not capture the compounding effect of sequential position escalation. Martingale detection must precede Monte Carlo analysis, not follow it.
It does not capture regime change. The historical trade distribution reflects the market conditions that existed during the backtest period. If market conditions change substantially (volatility regime shift, liquidity changes, correlation breakdown), the historical distribution may no longer represent what the strategy will encounter going forward. Monte Carlo simulation generates paths from the historical distribution, so it inherits any regime-specific bias present in the backtest data.
It cannot correct for curve fitting. If the strategy was optimized on the historical data, the trade outcomes it produced are artificially favorable. The parameters were selected precisely because they performed well in that specific period. Monte Carlo simulation run on an overfit backtest produces a distribution of outcomes from an overfit sample, which remains systematically optimistic. Walk-forward analysis and out-of-sample testing are required to address curve fitting. Monte Carlo simulation does not substitute for them.
Minimum trade count applies. Monte Carlo simulation requires enough trades to produce a statistically stable distribution. With fewer than 200 trades, the resampled paths are drawn from a small pool and the resulting distribution reflects sampling variation as much as strategy behavior. The percentile figures become unstable. P95 calculated from 150 trades is a different number every time you run the simulation, because there are only so many ways to arrange 150 items. Treat Monte Carlo results from backtests with fewer than 200 trades as directional only, not as precise risk estimates.
Position Sizing from Monte Carlo Output
The practical purpose of Monte Carlo simulation for most retail traders is not academic. It is to determine the correct position sizing for a live account. The P95 drawdown figure is the key input.
The standard approach: decide the maximum drawdown you are willing to accept in live trading, then size the account such that the P95 drawdown equals that threshold. If your EA’s Monte Carlo P95 drawdown is 18% and your maximum acceptable drawdown is 20%, you can fund the strategy at approximately full size, since the P95 outcome sits within your threshold. If the P95 drawdown is 35% and your threshold is 20%, you need to either reduce position size by roughly 43% (20/35) or not fund the strategy at all.
This gives a concrete, quantitative answer to the question that most EA discussions never resolve: not just whether the EA has acceptable risk, but at what position size its Monte Carlo risk profile fits within your specific account constraints. The answer changes with every strategy and every account size. Deriving it from the Monte Carlo distribution rather than from the historical drawdown alone is the difference between risk management and risk theatre.
A Practical Checklist Before Interpreting Any Monte Carlo Output
Before treating Monte Carlo results as actionable, verify the following:
Trade count is sufficient. Minimum 200 trades for directional guidance, 500+ for reliable percentile estimates. Below 200, the simulation is running on too small a pool for stable distributions.
Martingale has been ruled out. Check the MG Indicator (average win size after a loss divided by average win size after a win). Ratios above 1.5 indicate position scaling behavior that invalidates the independence assumption. Monte Carlo results from such strategies are not meaningful.
The backtest used high-quality tick data. For MT4 and MT5, confirm the modelling quality is 99% with real tick data or a tick data suite. For cTrader, the backtester natively uses high-quality historical data, but verify that the commission and spread settings match the live broker conditions you intend to trade on.
The simulation method is percentage-based. Fixed-dollar resampling understates tail risk for strategies with material drawdown depth. Confirm which method your tool uses before interpreting the percentile figures.
The binomial test has been run separately. Monte Carlo confirms the risk profile. The binomial test confirms the edge is real. Both are required. A positive Monte Carlo result on a strategy with a statistically insignificant win rate is a smooth distribution around a random outcome.
Monte Carlo simulation is the most important single tool in systematic EA validation, precisely because it converts a single historical outcome into a statistically grounded range of outcomes. The number your backtest produced is one point in a distribution. Monte Carlo shows you the shape of that distribution: its median, its tail, and its relationship to the single historical observation you have available. That relationship is what determines whether the strategy is deployable, and at what size.
The free Monte Carlo tool at ErgodicLabs runs this analysis directly from MT4, MT5, and cTrader HTML reports. Upload a backtest, get the full percentile distribution, the expansion ratio, and a plain-language risk interpretation in under thirty seconds. No installation, no account required.