Shuffling trades to see how bad the drawdown could have been
Posted: Sat Oct 31, 2026 6:57 pm
A backtest produces one sequence of trades, and the maximum drawdown reflects that particular order. The same trades in a different order would produce a different drawdown. A Monte Carlo resampling approach reveals the range.
Take the list of trade results, shuffle it randomly many times, and compute the maximum drawdown for each shuffle. The distribution shows what is plausible, not just what happened.
Typically, the historical drawdown sits somewhere in the middle of the distribution, and the worst cases in the shuffles are much larger. The 95th percentile drawdown is a more conservative planning figure than the single historical value.
This matters for scalpers because of the large number of trades. Even with a positive edge, adverse sequences can occur, and the account must be able to survive them.
The method assumes trades are independent, which is not always true, since results cluster by market conditions. A block bootstrap, which shuffles chunks of consecutive trades, preserves some of that structure.
Use the results to set position size so that the plausible worst case is tolerable.
What does the 95th percentile drawdown look like compared with the one in the backtest?
Take the list of trade results, shuffle it randomly many times, and compute the maximum drawdown for each shuffle. The distribution shows what is plausible, not just what happened.
Typically, the historical drawdown sits somewhere in the middle of the distribution, and the worst cases in the shuffles are much larger. The 95th percentile drawdown is a more conservative planning figure than the single historical value.
This matters for scalpers because of the large number of trades. Even with a positive edge, adverse sequences can occur, and the account must be able to survive them.
The method assumes trades are independent, which is not always true, since results cluster by market conditions. A block bootstrap, which shuffles chunks of consecutive trades, preserves some of that structure.
Use the results to set position size so that the plausible worst case is tolerable.
What does the 95th percentile drawdown look like compared with the one in the backtest?