Worst-Case Path Construction

A normal backtest asks "how did this strategy do over history?" Worst-case path construction asks a harder question: "what is the nastiest sequence of events, still consistent with the market's known behavior, that this strategy could run into?" Instead of drawing paths at random and hoping the bad ones show up, the analyst deliberately assembles a path designed to hurt — clustering the strategy's worst historical days back to back, or reordering returns so all the losses land early before the equity curve has any cushion. The goal isn't to predict the future; it's to find the edge of what the strategy can survive.

A useful analogy is crash-testing a car by choosing the worst plausible impact angle rather than driving it randomly into things and hoping one crash is bad enough. The engineer picks the scenario, not chance.

In practice this means taking the actual pool of historical daily (or trade-level) returns and re-ordering them, rather than inventing new numbers: for example, sorting a year of trade P&L from worst to best and asking what the drawdown looks like if the five worst trades all happen in the first month, when capital and confidence are lowest. This produces a maximum drawdown and time-to-recovery figure that is usually far worse than the one seen in the actual historical sequence, because ordinary history rarely delivers its worst events consecutively.

The technique is closely related to block bootstrapping and Monte Carlo path resampling, but it deliberately abandons randomness in favor of an adversarial, worst-plausible ordering. It's most useful as a sanity check before sizing a strategy: if a worst-case reordering of the same historical returns wipes out the account, the position size or leverage was too aggressive even though the historical backtest looked fine.