Drawdown Duration and Time Under Water

Two funds both fall 20% from their peak. Fund A recovers to a new high in six weeks. Fund B takes four years, and an investor who redeemed partway through never sees the recovery at all. Both funds report an identical "20% max drawdown," and that statistic alone cannot tell them apart. What separates a survivable slump from a career-ending one is usually not the depth, it is the duration: how long you spend below the old high-water mark, "underwater," before the account is whole again.

The analogy

Picture two swimmers who both get pulled a little way out by a rip current. One is back on the beach in two minutes; the other treads water for two hours before the current lets go. The distance from shore at the worst moment might have been similar for both, but the two-hour swimmer is the one at real risk of drowning, and the one whose family panics and calls it off. For an investor, "time under water" is exactly that stopwatch: the clock starts the moment the account first dips below a prior peak and does not stop until a new peak is set.

Building the idea

Recall a drawdown reading $D_t$ measures how far below the running peak the account sits at time $t$ (zero exactly at a new high, positive otherwise, as in Max Drawdown). Define an underwater episode as any maximal stretch of consecutive periods where $D_t > 0$. Its duration is simply the length of that stretch:

$\text{Duration} = t_{\text{recover}} - t_{\text{peak}},$

the number of periods from the old peak to the moment a new peak is finally set. In words: count every day (or month) the account spends strictly below its old high, from the instant it first slips to the instant it fully recovers; that count is the duration of one underwater episode. A track record typically has many episodes of different lengths, and two summary statistics matter most: the longest time under water observed, and the average time under water across all episodes.

Generate several paths and imagine a horizontal dashed line at each path's running peak so far. Duration is the horizontal width of the gap between where the path first dips below that line and where it next crosses back above it, not the vertical depth of the dip. A shallow but wide dip can have a longer, more painful duration than a deep but narrow one.

Worked example 1: reading duration off an equity curve

An account's monthly running-peak-relative values (percent below peak) are: Jan 0, Feb 3, Mar 7, Apr 5, May 2, Jun 0, Jul 0, Aug 4, Sep 9, Oct 6, Nov 1, Dec 0.

Episode 1 starts in February (first month below peak after a high in January) and ends in June, the first month back at a new peak: a duration of 5 months (Feb–Jun). July is at a peak too (0), so no episode. Episode 2 runs August to December: 5 months again. Longest and average time under water: 5 months either way. Notice the deepest reading in Episode 2 (9% in September) was worse than Episode 1's deepest (7% in March), but both episodes took the same time to resolve, depth and duration are genuinely separate axes of pain.

Worked example 2: same max drawdown, very different duration

Fund X drops 15% in one month, then rallies back to a new high the next: duration of 1 month. Fund Y also drops 15%, matching Fund X's max drawdown exactly, but grinds sideways for 30 months before clearing its old peak: duration of 30 months. An investor evaluating both funds purely on max drawdown would rate them identically. But a retiree or endowment needing the money back within two years would find Fund Y essentially unusable despite the "tied" statistic, because for 30 of those months the capital simply was not there to spend.

What this means in practice

Allocators set explicit duration limits alongside depth limits ("no more than 12 months underwater at the 95th percentile") because redemption behavior tracks duration more than depth. Duration also interacts with fee structures: high-water-mark performance fees are not earned again until the fund clears its old peak, so a long duration directly delays the manager's own compensation too.