The Monte Carlo Test: What Your Strategy Looks Like in the Worst Case
A backtest gives you one equity curve: the specific path your strategy would have taken through the specific sequence of historical trades. This single path is the most misleading number in your entire analysis.
The real question is not what happened in that one historical sequence. It is: across all the plausible sequences of the same trades, what does the distribution of outcomes look like?
That is what Monte Carlo simulation answers.
How Monte Carlo Works
Monte Carlo takes your backtest's full trade list — every individual entry, exit, win, and loss — and randomly shuffles the sequence 1000 times. Each shuffle produces a different equity curve. Some orderings result in all the early trades being winners, producing fast initial growth. Others front-load the losses, creating deep early drawdowns.
The result is 1000 equity curves. You can look at the best case (top 5%), the typical case (median), and the worst case (bottom 5%). The 5th percentile equity curve — the one worse than 95% of simulated outcomes — is the scenario you need to be able to survive.
Why Sequence Matters
Consider a simple example: a strategy makes 10 trades. 6 are winners of $5 each. 4 are losers of $4 each. Net profit: $14.
Now consider two possible orderings:
- Best case: WWWWWWLLLL (all winners first). Peak equity grows to $30 before any losses. Final equity: $14.
- Worst case: LLLLWWWWWW (all losers first). Account drops to -$16 before recovering. Final equity: $14.
Same 10 trades. Same final result. But the first sequence was comfortable; the second required surviving a 16-unit drawdown before seeing any profit. If your account only had $20 and the losses wiped it out before the wins arrived, the strategy is over.
This is why the single equity curve is misleading. If history happened to present the favorable sequence, your backtest looks smooth. But the same trades in a different order could have been ruinous.
Reading the Monte Carlo Output
Median equity curve: This is the 50th percentile — half of simulated sequences did better, half did worse. It is a reasonable central estimate of strategy performance, but it is not the number you should plan around.
5th percentile drawdown: The maximum drawdown in the worst 5% of simulated sequences. This is the key risk metric. If this drawdown exceeds your maximum tolerable loss — either psychologically or in terms of account survival — you need to reduce position size or modify the strategy.
Band width (25th to 75th percentile): The distance between these two curves measures variance. A narrow band means outcomes are consistent regardless of trade sequence. A wide band means results are highly sequence-dependent — the strategy works great in some market conditions and terribly in others. Wide bands require more conservative sizing.
Ruin probability: What percentage of simulated sequences resulted in a loss exceeding your defined risk tolerance? If more than 10% of simulations breach your maximum drawdown threshold, the strategy requires modification before live deployment.
Using Monte Carlo to Set Position Size
The correct process is:
- Run the backtest.
- Run Monte Carlo.
- Look at the 5th percentile drawdown in percentage terms.
- Choose a position size such that 5th percentile drawdown in dollar terms is an amount you can genuinely sustain.
If the 5th percentile drawdown is 35% of the backtest capital and your tolerance is a 15% account drawdown, you should allocate no more than 43% of your account to this strategy (15% / 35% = 0.43).
This calculation feels conservative. It is. The point is that if the bad sequence occurs — and eventually it will — your account survives and you have the capital and the psychological resolve to continue running the strategy through recovery.
The Honesty of Monte Carlo
Standard backtests reward lucky sequences. Monte Carlo exposes them. A strategy that looks excellent in a standard backtest because history happened to present a favorable trade sequence will show a wide Monte Carlo band and a brutal 5th percentile.
A strategy that looks only decent in a standard backtest but shows tight Monte Carlo bands and a mild 5th percentile is genuinely robust. The edge is consistent regardless of sequence. That is the strategy worth deploying.
The purpose of Monte Carlo is not to make you pessimistic about your strategy. It is to give you an honest assessment of variance so you can size correctly. Correctly sized strategies survive drawdowns. Strategies without a survivable drawdown plan end in forced liquidation.