Monte Carlo in a backtest: what it measures and what it does not

An equity curve shows one historical path. Deepwick's Monte Carlo panel asks what other paths might look like when constructed from the closed trades in that backtest. The distinction matters: the original sample still supplies all the information.

Open Monte Carlo from a saved report

Go to Backtests, open a completed report and choose quant report. The Monte Carlo bootstrap panel shows the simulation count, final-equity percentiles and a drawdown measure. You need a session and at least one closed trade to obtain results; a small sample can produce numbers with very little information behind them.

Record the symbol, provider, timeframe, historical window and backtest costs before interpreting the panel. If you do not have a report yet, start with your first backtest.

What changes in each simulation

Deepwick samples closed-trade P&L with replacement: one trade may appear several times while another never appears. Each path contains as many draws as the original sample contained trades. This is more than rearranging a fixed list.

The NIST statistical handbook describes this bootstrap principle: resampling observations to explore uncertainty in a statistic.

The current report uses 1,000 simulations and a reference starting capital of 10,000. It adds monetary P&L amounts; it does not rerun the script or recalculate position sizing. If your backtest starts with different capital, do not read these percentages as a direct reproduction of your account. Recalculation can change the figures because sampling is random.

Read the percentiles in context

5th %, Median and 95th % describe positions within the simulated final-equity distribution. They do not promise that future performance will fall between those endpoints.

The 95th-percentile drawdown describes an adverse part of the generated simulations. Worse paths can exist, including among those simulations. It cannot account for losses absent from the original sample.

Ruin prob (final < 50%) counts simulations ending below half the reference starting capital. It does not necessarily count every path that crossed that threshold along the way, or estimate a real account's liquidation probability.

Record an investigation, not a verdict

Keep the trade count, percentiles, reference capital and analysis date in your notes. Check whether a few large winners dominate the sample and what changes when you study a different period.

This resampling does not preserve historical trade clusters or introduce new regimes, execution failures or different costs on its own. Use it to develop questions about the sample, alongside walk-forward with separate training and test windows. The partition panel in the Quant report does not replace that process of reoptimizing and evaluating on subsequent data.