The most expensive mistake in algorithmic trading isn’t a bad strategy. It’s a good strategy with the wrong position size.
A trader finds an EA that returned 80% annually in backtesting with 12% maximum drawdown. They calculate: “I can tolerate 25% drawdown, so I’ll run it at double the backtest lot size. That gives me 24% theoretical drawdown with plenty of margin.” Three months later, the account is down 40% and the EA gets turned off — permanently.
The error wasn’t the strategy. The error was sizing based on a single backtest drawdown instead of the statistically realistic drawdown. Monte Carlo analysis provides the correct number. Here’s how to use it.
Why Backtest Drawdown Is the Wrong Number
Your backtest maximum drawdown is a single observation from a probability distribution. It tells you what happened with one specific ordering of trades over one specific historical period. It does not tell you what drawdown to expect.
Consider a simple analogy. You roll a standard die 50 times and the highest single roll is 5. You conclude: “This die maxes out at 5.” Obviously, this is wrong — you just didn’t happen to roll a 6 in that sample. Given enough rolls, you will.
Backtest drawdown works the same way. Your EA executed 200 trades in a particular order. The worst peak-to-trough decline happened to be 12%. But rearrange those 200 trades into any of the 200! possible orderings, and many of those sequences would produce drawdowns of 18%, 25%, or worse. You only saw one sequence. Your live account will see a different one.
Monte Carlo quantifies this. By reshuffling the trade order thousands of times, you get a distribution of maximum drawdowns. The 95th percentile of that distribution — the value exceeded by only 5% of simulations — is the number you should size against. You can calculate this for free right now with any MT4 or MT5 backtest report.
The Risk Multiplier Formula
The calculation is simple once you have Monte Carlo results:
Risk Multiplier = Original Backtest DD ÷ 95% CI DD
This multiplier tells you how to adjust your lot size relative to the backtest. If the multiplier is 0.6, use 60% of your backtest lot size. If it’s 1.2, you may actually have room to scale up by 20%.
Let’s work through real examples:
Example 1 — Conservative EA: An EA backtested on EURUSD with 0.10 lots shows 4.2% original drawdown. Monte Carlo 95% CI shows 5.8%. Multiplier: 4.2 / 5.8 = 0.72x. Using ~0.07 lots instead of 0.10 would maintain the same drawdown profile. The EA’s risk was slightly understated by the backtest — a common outcome.
Example 2 — Aggressive scalper: A gold scalper backtested with 0.50 lots shows 8.1% drawdown. Monte Carlo 95% CI shows 22.4%. Multiplier: 8.1 / 22.4 = 0.36x. That suggests ~0.18 lots — roughly a third of the backtest size. This EA’s backtest was significantly lucky on drawdown. The original trade sequence avoided the worst clustering of losses, but live trading likely won’t be as fortunate.
Example 3 — Robust trend follower: A trend-following EA with 0.20 lots shows 15.3% drawdown. Monte Carlo 95% CI shows 14.1%. Multiplier: 15.3 / 14.1 = 1.09x. The original drawdown was actually worse than what Monte Carlo predicts at 95% confidence. This typically indicates a robust strategy where the original trade ordering was slightly unlucky — a good sign.
Our free Monte Carlo tool calculates the risk multiplier automatically and shows you the formula with your specific numbers.
Why 95% and Not 99%?
The choice of confidence level involves a tradeoff between safety and capital efficiency.
95% CI means you accept a 1-in-20 chance of exceeding the target drawdown due to trade ordering alone. For most individual EAs, this is the standard professional risk management threshold. It provides meaningful protection without being so conservative that returns become negligible.
99% CI means only 1-in-100 chance of exceeding. This is appropriate when running a portfolio of multiple EAs where a single blowup could cascade, or when managing client capital with strict drawdown limits. The tradeoff is significantly smaller position sizes — typically 30-50% smaller than the 95% sizing.
90% CI is too loose for primary risk management. A 10% chance of exceeding your drawdown target is uncomfortably high for any serious operation.
One important caveat: these confidence intervals only account for trade ordering risk. They don’t protect against overfitting, regime changes, execution quality degradation, or black swan events. In practice, many professional traders apply an additional safety margin on top of the Monte Carlo sizing — especially for live accounts with real capital.
The Capital Reserve Calculation
Monte Carlo drawdown also helps answer a critical question: how much capital do you need to safely run an EA?
The formula:
Minimum Capital = (Lot Size × Contract Value × 95% CI DD) ÷ Target Max DD%
Say you want to run a EURUSD EA at 0.10 standard lots (contract value $100,000), the Monte Carlo 95% CI drawdown is 6.2%, and you want to limit account drawdown to 20%:
Minimum Capital = (0.10 × $100,000 × 6.2%) ÷ 20% = $620 ÷ 0.20 = $3,100
With $3,100, a 6.2% drawdown on the notional ($620) represents exactly 20% of your account. Below this capital level, you’re either taking on more risk than intended or need to reduce lot size.
For gold (XAUUSD) with its higher volatility, the same calculation often yields capital requirements 2-3x higher than forex pairs. This is one reason undercapitalized gold EA accounts are among the most common blowup scenarios in retail trading.
Portfolio-Level Position Sizing
When running multiple EAs simultaneously, position sizing becomes more complex. Individual Monte Carlo analysis tells you the drawdown distribution per EA, but correlated drawdowns across EAs can compound.
A practical approach: size each EA individually using Monte Carlo 95% CI, then apply a portfolio-level reduction factor based on how many EAs share the same instrument or market regime. Two EAs trading EURUSD tend to have correlated drawdowns — their bad days often overlap. Two EAs trading EURUSD and USDJPY have moderate correlation. An EA on gold and an EA on EURAUD typically have lower correlation.
As a rough guideline: if running N correlated EAs, multiply each individual lot size by 1/√N. For two correlated EAs, that’s ~0.71x each. For four, ~0.50x each. This is a simplified version of portfolio variance calculation, but it captures the essential insight that correlated strategies don’t diversify risk as much as traders often assume.
When to Re-Run the Analysis
Position sizing isn’t a set-and-forget decision. Re-run Monte Carlo analysis when:
Every 3 months of live trading: Add real trades to the backtest data and re-simulate. Live performance almost always shows different characteristics than backtesting — spreads are wider, slippage exists, and execution timing varies. Your Monte Carlo distribution will shift, and your sizing should shift with it.
After any EA update: Even minor changes to entry logic, exit logic, or filters change the trade distribution. A modified EA is a different strategy from a statistical perspective.
After a market regime change: If your EA was developed during low-volatility conditions and the market shifts to high volatility (or vice versa), the historical trade distribution may no longer represent future trades. Re-run with recent data weighted more heavily.
When drawdown exceeds the 90% CI: If live trading drawdown surpasses the Monte Carlo 90th percentile, this is an early warning. Don’t wait for the 95% level — consider reducing position size proactively.
Calculate Your Risk Multiplier
Use our free Monte Carlo analyzer to get your EA’s risk multiplier right now. Upload any MT4 or MT5 backtest report and you’ll see the 95% CI drawdown, the risk multiplier, and a lot size adjustment recommendation — all calculated in your browser with no data leaving your machine.
For comprehensive validation including temporal stability, sample size adequacy, and concentration risk analysis, Edge Matrix runs 17 professional-grade tests with a single click.
Size for the drawdown that’s statistically likely, not the drawdown that happened to show up in one backtest.