← All case studies
Quantitative Simulation

Multi-Model Monte Carlo Testing

A multi-model Monte Carlo engine that stress-tests strategies across thousands of scenarios — so decisions are judged by their distribution of outcomes, not one lucky backtest.

PythonNumPyMonte CarloSimulation
Multi-Model Monte Carlo Testing — illustrative photography

The challenge

Strategy decisions were being made off a single historical backtest — one path through history, treated as if it were the future. That gives a comforting number and no sense of the range of things that could actually happen, or how badly the worst cases could hurt.

The process

End to end — how data moves from source to decision.

Model Ensemble
GBM, jump-diff, bootstrap
Scenario Engine
Thousands of paths
Aggregate
Percentiles & tails
Distribution
P5 · P50 · P95

Our approach

How we structured the work.

01
Built an ensemble rather than a single model — geometric Brownian motion, jump-diffusion, historical bootstrap and a regime-switching variant — so no one set of assumptions dominates.
02
Generated thousands of scenario paths per run in vectorised NumPy, keeping each run fast enough to explore assumptions interactively.
03
Aggregated results into full outcome distributions with percentile and tail statistics, not just an average.
04
Framed the output around downside — P5, median and P95 — so the conversation includes how bad the bad cases are.

What the reporting looks like

An illustrative view of the dashboards this engagement produces. All figures are sample data for illustration only — not client results.

Monte Carlo — Outcome Distribution
Sample data · illustrative
10,000
Simulation runs
P5 -12%
Downside
P50 +9%
Median
P95 +34%
Upside

Simulated 12-month return distribution

P5P50P95

Model ensemble weight

GBMJump-diffHist-bootRegime

Outcome

What changed
Strategies are now judged on the whole distribution of outcomes and their tail risk, across a family of models — replacing a single, over-trusted backtest with an honest picture of what could happen.

Related work

Other engagements with a similar shape.

Process × Intelligence

Have a similar problem?

Start with a conversation. We'll map the process, pressure-test the goal, and come back with a plan.

Get a Quote →