Monte Carlo simulation in trading usually means generating many possible equity paths by randomizing historical returns or trade sequences, then studying the distribution of outcomes (final wealth, drawdown, ruin).
It is a stress tool, not a crystal ball. Done well, it clarifies uncertainty. Done poorly, it implies precision where none exists.
What people hope to learn
- How deep drawdowns might get if return order changes.
- Whether results depend on one lucky streak of trades.
- Rough confidence bands around ending equity or CAGR.
Common techniques (plain language)
Shuffle daily returns - Keep the same set of historical daily returns but reorder them randomly many times. Preserves return magnitudes, destroys autocorrelation structure.
Bootstrap returns - Resample returns with replacement from history to build synthetic paths.
Shuffle trade P&L - Randomize the order of closed trades while keeping trade sizes fixed.
Each method answers a slightly different question. Pick one that matches your risk worry.
What Monte Carlo preserves and destroys
| Preserved (often) | Destroyed (often) |
|---|---|
| Histogram of daily moves | Serial correlation, momentum, mean reversion |
| Average return level | Exact crisis sequencing |
If your edge lives in regime structure, naive shuffles understate tail risk. Combine simulation with scenario analysis (2008, 2020, 2022-style windows) and walk-forward tests.
Reading results
Look at percentiles, not only the mean path:
- 5th percentile max drawdown (bad luck case)
- Median ending wealth
- Probability of exceeding a loss threshold
Compare to deterministic max drawdown from the historical path. Large gaps mean order risk matters.
Pitfalls
- Too few simulations - Noise in the noise.
- Independent draws when returns cluster - Volatility clustering matters.
- Ignoring costs - Turnover strategies need slippage and commissions in the draw distribution.
- Replacing judgment - A pretty cone chart does not validate a broken signal.
When it helps most
- Assessing sequence risk after a solid backtest.
- Communicating uncertainty to stakeholders who ask "worst case?"
- Checking whether Calmar or Sharpe is driven by one short sample.
Monte Carlo belongs in the toolkit with honest limits. Use it to widen your view of paths, not to bless overfit parameters.
