Monte Carlo simulation in trading explained

By Ken Chigbo, Founder, KenMacro. Published 2026-05-13.

What is Monte Carlo simulation?

Monte Carlo simulation is a statistical method that uses repeated random sampling to model the range of outcomes a trading system could produce. Rather than relying on a single historical sequence of trades, the desk runs thousands of randomised permutations of the same trade results, or draws synthetic trades from a fitted distribution, to build a probability distribution of equity curves. The output typically includes percentile drawdowns, median return paths, and the probability of hitting a defined ruin threshold. It is widely used in quantitative risk management to separate genuine edge from path-dependent luck embedded in backtest histories.

How traders use Monte Carlo simulation

Retail systematic traders feed their backtest trade log into Monte Carlo software such as Python scripts, R, or platforms like StrategyQuant and Market System Analyzer. The desk typically runs ten thousand iterations, shuffling trade order while preserving win rate and trade size distribution. Outputs include the 95th percentile maximum drawdown, median terminal equity, and probability of breaching account ruin thresholds. Institutional risk teams extend this by randomising entry slippage, fill probability, and parameter drift to stress test execution assumptions. The practical use is position sizing: if the 95th percentile drawdown is twice the historical backtest drawdown, the trader scales risk down accordingly. It also exposes strategies whose backtest results depended on a lucky trade ordering rather than persistent edge.

Common misconceptions about Monte Carlo simulation

The first misconception is that Monte Carlo predicts future returns. It does not. It only redistributes the outcomes already present in the trade sample, so a biased or curve-fitted backtest produces biased simulations. The second is that reshuffling alone captures all risk. Standard trade-order resampling assumes trades are independent and identically distributed, which ignores regime shifts, volatility clustering, and correlation breakdowns. A third error is treating the median outcome as the expected result while ignoring the tails. The desk uses the lower percentiles, typically the 5th, as the realistic planning case for sizing and ruin analysis.

Open a Vantage raw-spread account

Frequently asked

How many Monte Carlo iterations should a trader run?

Most practitioners run between one thousand and ten thousand iterations. Below one thousand, the tail percentiles become unstable and shift noticeably between runs. Above ten thousand, additional iterations rarely change the headline metrics by a meaningful amount. For production risk reporting, the desk recommends ten thousand as a standard, with twenty thousand or more when the focus is on extreme tail events such as one percent ruin probability, where rare outcomes need adequate sampling.

Does Monte Carlo simulation work for discretionary trading?

Yes, provided the trader keeps a structured journal with consistent risk per trade and a sufficient sample, typically one hundred trades or more. The simulation reshuffles these recorded outcomes to estimate drawdown distributions. The caveat is that discretionary results often reflect changing market regimes and evolving trader skill, so the underlying assumption of identically distributed trades is weaker. Results should be treated as indicative rather than precise probability statements.

What is the difference between Monte Carlo and bootstrap resampling?

Both involve random sampling, but they differ in technique. Pure Monte Carlo often draws trades from a fitted statistical distribution, such as a normal or empirical distribution of returns. Bootstrap resampling draws with replacement from the actual historical trade sample, preserving its real characteristics. In trading practice the two terms are often used interchangeably, with most retail tools using bootstrap methods while still calling the output a Monte Carlo simulation.

Can Monte Carlo simulation detect a curve-fitted strategy?

Indirectly. If a backtest shows a small maximum drawdown but the Monte Carlo 95th percentile drawdown is several times larger, the original sequence was likely fortunate rather than representative. However, Monte Carlo cannot detect curve fitting in the parameter selection itself, since it only resamples outcomes that the fitted strategy already produced. Walk-forward analysis and out-of-sample testing remain the primary tools for identifying overfitting; Monte Carlo complements them by quantifying path risk.

Educational analysis only. Past performance does not guarantee future results. Manage risk against your own portfolio.