Common simulation designs
Reordering historical trades
Existing trade outcomes are randomly reordered. This preserves the observed outcomes but assumes order is interchangeable and future trades resemble the sample.Bootstrap resampling
Trades are sampled with replacement. This allows some outcomes to appear more than once and others not at all. Block bootstrap methods can preserve short-term dependence.Parametric simulation
Outcomes are drawn from a fitted probability distribution. This can create values not present in the sample but adds model risk, especially in the tails.Strategy and market simulation
Prices, signals, fills, and portfolio rules are modeled directly. This can represent more detail but requires many assumptions and careful validation.A practical workflow
1
Prepare net trade data
Include spread, commission, slippage, financing, missed trades, and execution differences.
2
Choose the dependence model
Decide whether independent resampling is defensible or whether regimes, blocks, or correlated positions must be preserved.
3
Apply the real sizing policy
Recalculate position size, aggregate exposure, and account value at each step.
4
Implement account rules
Model daily resets, open P&L, trailing thresholds, and the stop condition exactly.
5
Run many paths
Use enough simulations for the tail probabilities being estimated.
6
Report distributions
Show percentiles and breach frequencies rather than one average curve.
Useful outputs
- median and percentile terminal return;
- maximum drawdown distribution;
- longest losing streak distribution;
- time under water;
- probability of finishing below starting capital;
- probability and timing of daily or maximum-loss breach;
- sensitivity to worse costs, lower expectancy, and higher correlation.
Limitations
A small or biased backtest produces a biased simulation. Random reordering can understate risk when losses cluster by regime. Historical data may omit an event more severe than anything observed. Report assumptions and rerun the model under less favorable inputs.Losing streaks
Understand why sequence modeling matters.
Risk of ruin
Define the failure event to simulate.
Random-entry benchmark
Compare a strategy with a neutral baseline.
Prop risk management
Apply the simulation to account rules.