Monte Carlo Simulation: Why Your Backtest Is Lying to You
You ran a backtest. Your strategy made money. The equity curve climbed, there were some drawdowns, and the final balance was satisfying. Three months into live trading the account is down 30% and the strategy "isn't working." What happened is almost never what traders think. The strategy didn't break. What happened is that the backtest showed you one possible sequence of results, and you traded a different one.
Each faint trail is one simulated future for this strategy. The bold line is the median — half of runs finished above it, half below. The shaded band spans the lucky 10th to unlucky 90th percentile, so an outcome inside the band was reasonable to expect; outside, less so.
- ✓ A backtest produces one sequence of results — the actual historical sequence
- ✓ Different orderings of the same wins and losses produce radically different equity curves
- ✓ Monte Carlo simulation runs hundreds of randomised sequences to show the full probability range
- ✓ The difference between best and worst simulated paths reveals your real risk exposure
- ✓ Use the Strategy Reality Check to run 400 Monte Carlo simulations on your strategy
50% of simulations beat. Your realistic expected result.
Probability band: 10th–90th percentile outcomes. Width reveals sequence risk.
Probability of ruin: the % of simulations that hit your defined loss threshold.
Worst-case path: the sequence you need to survive psychologically and financially.
Why the backtest result is always optimistic
The historical backtest sequence is not a random sample from the distribution of possible sequences. You are more likely to backtest a strategy over a period when it performed well than when it struggled — survivorship bias within your own test. Monte Carlo simulates sequences that never happened and sequences that are unlikely but possible. The gap between your backtest equity curve and the Monte Carlo median is the cost of that historical optimism.
A practical example
Strategy: 55% win rate, 1.5:1 R:R, 2 trades/day, 20 trading days/month, 1% risk, $10,000 starting capital.
| Metric | Backtest (historical) | Monte Carlo (400 sims) |
|---|---|---|
| 3-year final balance | $38,400 | Median: $34,668 |
| 10th percentile (bad year) | — | $18,200 |
| 90th percentile (good year) | — | $67,400 |
| Max drawdown | 14% | 41% (worst sim) |
| Probability of 50% ruin | 0% | 2.3% |
The backtest showed $38,400. The Monte Carlo median is $34,668 — the historical sequence was slightly above median. More importantly, the 10th percentile of $18,200 is what a trader should plan for as a realistic adverse scenario, not the backtest's optimistic figure.
What sequence risk means for position sizing
The width of the Monte Carlo band directly informs position sizing. A wide band means sequence risk is high — the same mathematical edge can produce very different lived experiences. Size so that the 10th percentile outcome is survivable both financially and psychologically. If your realistic bad-case 1-year outcome would cause you to abandon the strategy, your position size is too large regardless of what the median shows.
FAQ
Is Monte Carlo simulation the same as backtesting?
No. Backtesting replays one specific historical sequence. Monte Carlo generates many randomised sequences using the same statistical parameters. Backtesting tells you what happened; Monte Carlo tells you the range of things that could happen.
How many simulations are enough?
400–1,000 simulations is typically sufficient for stable percentile estimates. Below 100, percentile bands are noisy. Above 10,000, computation time increases without meaningful accuracy improvement for retail trading parameters.
My live results are within the Monte Carlo band but feel terrible. What's happening?
You may be on a 10th–20th percentile path — statistically normal, psychologically brutal. This is exactly why Monte Carlo matters before you trade live: knowing your realistic bad scenario in advance makes it possible to tolerate rather than interpret it as strategy failure.
Should I only trade strategies with tight Monte Carlo bands?
Not necessarily — tight bands can mean limited upside. The goal is that the 10th percentile outcome is survivable at your chosen position size. Wide-band strategies can still be traded at smaller sizes.