Monte Carlo and Robustness

After a backtest, open the Monte Carlo tab of the results, choose a method and the number of paths, and run it. It costs one backtest credit and uses the same strategy, data and costs as the run above it. Through the API, call POST /backtest/engine/monte-carlo.

Trade methods run inside DuckDB as one query over every path (up to 100,000 paths), with seeded draws: the same seed gives the same result. Market-path methods re-run your strategy on every path, so they take longer and are capped at 500 paths. They work for every language, including builder and QuaScript strategies.

• Probability of loss: the share of paths that end below the starting capital. Above 30% the edge is weak relative to its noise.
• Probability of ruin: the share of paths whose equity touches the ruin level (50% below the start by default) at any point. Size positions so this stays near zero.
• Drawdown percentiles: plan for the 5th percentile (one path in twenty is worse), not for the drawdown your single backtest happened to show.
• Path bundle: up to 100 sampled equity paths with the best, worst and median path and the real trade order. Switch between all paths, the top 10 and the 5th–95th envelope.
• Risk corridor: the 5th–95th and 25th–75th percentile of equity after each trade, with the ruin barrier and a drawdown pane. A wide corridor means the outcome depends on luck in trade order.
• Terminal wealth: the histogram of final equity across paths against a normal fit, and the risk-reward frontier: each path's max drawdown against its total return.
• Losing streaks: the share of paths by their longest run of consecutive losing trades. Size positions so the 95th percentile streak is survivable.
• Actual Sharpe percentile (market paths): where the real run ranks among the re-runs. Above 90, the real history was unusually kind to the strategy.