Monte Carlo Simulation

Monte Carlo simulation tests trading strategies by running thousands of randomized scenarios using your historical trade data:

Step 1 — Collect trade history. Gather your strategy's historical trades: each trade's return (win or loss percentage), holding period, and any other relevant parameters. You need at least 100 trades for meaningful results.

Step 2 — Random sampling. The simulation randomly selects trades from your history (with replacement) to create synthetic sequences. Each run might pick different trades in different orders — one run might cluster your worst trades together, another might alternate wins and losses evenly.

Step 3 — Build equity curves. Each random sequence generates an equity curve showing how the account would have grown (or shrunk). Run 1,000-10,000 of these simulations to create a distribution of possible outcomes.

Step 4 — Analyze the distribution. From the thousands of equity curves, calculate: median final account value, worst-case drawdown at 95th percentile, probability of reaching your target return, probability of ruin (hitting a catastrophic loss), and the range of possible outcomes.

Why randomize? Your actual trade history happened in one specific order. But it could have happened differently — the same trades in a different sequence produce different drawdowns and different emotional experiences. Monte Carlo reveals what could have happened, not just what did happen. A strategy that showed a 15% max drawdown in backtesting might show a 95th percentile drawdown of 30% in Monte Carlo — meaning there was a 5% chance of a much worse experience than the historical backtest showed.

Drawdown distribution: The most valuable output. Instead of knowing your historical max drawdown was 20%, Monte Carlo tells you: 50th percentile drawdown is 15%, 75th percentile is 22%, 95th percentile is 32%, and 99th percentile is 45%. This reveals the realistic range of pain you should expect. If the 95th percentile drawdown exceeds what you can tolerate, reduce position size.

Risk of ruin probability: What percentage of simulations result in the account hitting a defined ruin level (e.g., 50% loss)? A strategy should show risk of ruin below 1% at your planned position size. If Monte Carlo shows 5% risk of ruin, you are sizing too aggressively.

CAGR distribution: The range of possible compound annual growth rates. A strategy's median CAGR might be 25%, but the 10th percentile might be 5% and the 90th percentile 50%. This tells you how variable your returns could be.

Optimal position sizing: Run Monte Carlo at different position sizes (0.5%, 1%, 2%, 5% risk per trade) and compare the risk/return tradeoffs. Typically, there is a sweet spot where increasing size improves returns rapidly, then a point where further increases barely improve returns but dramatically increase ruin risk.

Strategy comparison: When choosing between two strategies, compare their Monte Carlo distributions rather than their single backtest results. Strategy A with better median returns but worse 95th percentile drawdown may be inferior to Strategy B with lower median returns but tighter drawdown distribution.

Limitation 1 — Assumes stationarity. Monte Carlo assumes your future trades will have similar characteristics to historical ones. If market conditions change (different volatility regime, new regulations, structural market changes), the historical distribution may not predict the future. Always combine Monte Carlo with regime analysis.

Limitation 2 — Ignores trade dependency. Standard Monte Carlo treats each trade as independent. In reality, trades may be correlated — losing trades often cluster during unfavorable market conditions. Bootstrapped Monte Carlo (sampling blocks of consecutive trades rather than individual trades) partially addresses this.

Limitation 3 — Garbage in, garbage out. If your backtest is overfitted (too many optimized parameters, look-ahead bias, survivorship bias), Monte Carlo on those results will produce meaningless optimism. Clean backtest data is essential.

Best practices:

• Minimum 200+ trades for reliable Monte Carlo analysis. Fewer trades produce unstable distributions.
• Run 5,000+ simulations for smooth distributions. 1,000 is a minimum; 10,000 is ideal.
• Focus on the 95th percentile drawdown, not the average or median. Your risk management should handle the bad scenarios, not the average ones.
• Re-run quarterly as new trade data accumulates. Your trade distribution changes over time.
• Test different position sizes to find the optimal balance between growth and ruin risk. The Kelly Criterion gives a theoretical optimum; Monte Carlo validates it empirically.
• Compare forward Monte Carlo vs realized results. If your actual drawdowns exceed the 95th percentile from Monte Carlo, either your strategy has degraded or the simulation assumptions were wrong.