Drawdown

Drawdown is the distance below the high-water mark. It is a path statistic, not a return statistic: two strategies with identical return distributions can produce very different drawdowns depending on how their wins and losses happen to be ordered. That property is what makes drawdown informative — it measures the experience of holding the strategy, not just its average — and also what makes it treacherous to compare and easy to game.

Two structural facts are routinely ignored. First, maximum drawdown grows with track length. Even for a strategy whose behaviour never changes, the deepest excursion observed over ten years will, in expectation, exceed the deepest observed over two — more history means more opportunities for a bad sequence. Comparing maximum drawdowns across records of different lengths without adjustment is comparing apples to a longer list of apples. Second, backtested drawdown is biased shallow. A configuration selected from many trials is selected partly for having dodged the historical data’s worst sequences — a form of fitting to noise that backtest overfitting produces automatically. The live drawdown distribution is systematically worse than the one in the pitch document, before anything about the market changes at all.

Depth is also only half of the statistic. Drawdown duration — time under water, from peak to recovery — is frequently the operative number, because capital, patience, and mandates expire on the clock, not on the percentage.

What converts drawdown from discomfort into termination is a barrier. Prop-firm loss limits, margin requirements, investor redemption triggers, and a trader’s own capitulation point all place a hard floor under the equity curve; a statistical excursion that crosses the floor is not an excursion but an ending. That interaction — path statistics meeting hard barriers — is the domain of risk of ruin, and its sharpest practical form is the trailing drawdown rule, where the floor itself moves.