Monte Carlo Simulation

A backtest hands you one path through history: one trade order, one equity curve, one max drawdown. But the order of those trades was luck. Monte Carlo simulation asks what the distribution looks like across thousands of alternate orderings of the same trades.

The procedure

Take the backtest's individual trade returns. Shuffle their order, or resample with replacement. Rebuild the equity curve and record its drawdown, final return, and streaks. Repeat 1,000 to 10,000 times and read the distribution.

What it tells you

The historical max drawdown was one draw from a distribution. If the reshuffles put the 95th-percentile drawdown at 38%, that is the number to size leverage and nerves against. The fraction of paths that breach your uncle point before compounding away from it is a risk-of-ruin estimate. And sensitivity checks fall out for free: if removing the five best trades flips the system negative, the "edge" was a handful of outliers.

The caveat

Reshuffling assumes trades are independent, which understates risk when losses cluster, and they do cluster in regime shifts. Treat Monte Carlo bounds as optimistic floors. Its job, like all validation, is to make overconfidence expensive before the market does.

On AlphaProve

The engine ships the resampling described above: it can shuffle trade order, skip trades, or perturb individual returns, then rebuild the equity curve thousands of times to produce the distribution rather than the single path. A risk-of-ruin figure falls out of the fraction of paths that breach a chosen floor, and prop-firm-style challenge presets let you ask whether a strategy would have passed a fixed drawdown-and-target rule across those resampled runs.