Drawdown Distributions — Finance Lessons

A drawdown is the gap between where your account was at its best and where it is now — the pain of having been richer. Most people quote one number, “max drawdown 30%,” as if it were a fixed property of a strategy. It isn’t. Run the same system again and the worst drawdown comes out different, because drawdown is a random variable — it has a depth distribution, a duration distribution, and a brutal cousin called time under water. This lesson teaches you to think in those distributions, because the single backtest number you trust is just one draw from them — and rarely the worst one the future holds.

Before you read — take a guess

Your backtest reports a maximum drawdown of 25%. What does that number actually tell you about the drawdown you'll experience trading it live?

It guarantees the average drawdown will be 25%Relatively little on its own — it is one sample from a distribution; live trading can easily produce a deeper drawdown the backtest never happened to hitIt is the exact drawdown you will experience every yearIt is a hard ceiling — live drawdown can never exceed the backtested 25%

Drawdown depth: how far down from the peak

Analogy. Picture hiking and tracking your highest altitude so far. A drawdown is how far below that high-water mark you currently are. You can be at a perfectly fine altitude and still be in a deep drawdown if you’d previously summited higher. The pain is relative to your best, not your start.

Definition. At any time $t$ , the drawdown is the percentage fall from the running peak: $\text{DD}_t = \frac{\text{Peak}_t - \text{Equity}_t}{\text{Peak}_t},$ where $\text{Peak}_t$ is the highest equity reached up to time $t$ . It’s 0 when you’re at a new high and grows as you fall below it. The maximum drawdown (MDD) is the worst value $\text{DD}_t$ takes over the whole path — the deepest peak-to-trough plunge.

Worked example. An equity curve runs 100 → 134 (new peak) → 76 (trough) → 101. The deepest fall is from the 134 peak to the 76 trough: $\text{MDD} = \frac{134 - 76}{134} = \frac{58}{134} \approx 0.433 = 43.3\%.$ Note the recovery to 101 doesn’t change the max drawdown — MDD records the worst pain along the way, regardless of where you end up. Note also the asymmetry: recovering from that 43% drawdown back to the 134 peak requires a gain of $58/76 \approx 76\%$ , far more than the 43% you lost. Deep drawdowns are arithmetically expensive to climb out of.

Maximum drawdown: the worst peak-to-trough fall

The recovery asymmetry

Losses and the gains needed to undo them are not symmetric. A 10% loss needs an 11% gain; a 25% loss needs 33%; a 50% loss needs 100%; a 75% loss needs 300%. This convex ‘recovery wall’ is why deep drawdowns are so dangerous — and why they shade into ruin. Below some depth, the gain required to recover is so large it’s effectively unreachable.

An equity curve goes 200 (peak) → 120 (trough) → 180. What is the maximum drawdown, and what gain from the trough was needed to fully recover the peak?

40% drawdown; a 40% gain recovers it, since losses and gains are symmetric40% drawdown; a 66.7% gain was needed to recover (120 → 200 is +80/120)33% drawdown; measured from the trough up to the peak80% drawdown; the account fell by 80 points

Maximum drawdown is a distribution, not a number

Here’s the conceptual leap. The maximum drawdown of a strategy isn’t a fixed attribute — it’s the outcome of a particular sequence of wins and losses. Shuffle the same trades into a different order, or generate a fresh statistically-identical run, and you get a different MDD. So MDD has a distribution.

The shape. The distribution of maximum drawdown is right-skewed: most runs cluster around a “typical” worst drawdown (the median), but there’s a long tail of unlucky runs with much deeper drawdowns. The single backtest you have is one random draw — and there’s no reason it landed on the bad tail. The expected max drawdown over many runs is usually deeper than the median, dragged up by that tail.

Why it matters. If you size your strategy so that your backtested 25% drawdown is just barely survivable, you’re betting that the future won’t draw from the tail — and over a long enough life, it will. Robust sizing plans for a drawdown well beyond the historical worst: a common rule of thumb is to assume the real max drawdown could be 1.5–2× the backtested one.

The same system, many runs — max drawdown is a distributionMedian max drawdown: 27%

Risk per trade

Each run is the same statistics in a different random order, so each has a different worst drawdown. The result is a right-skewed distribution: most runs cluster near the median, but a long tail of unlucky orderings reaches far deeper. Your one backtest is a single sample from this — and rarely the tail. Raise risk-per-trade and the whole distribution slides deeper.

Maximum drawdown as a random variable.

Pick the right option for each blank, then check.

Maximum drawdown is . Its distribution is , so the drawdown you actually experience could be . Sizing should therefore assume a drawdown .

Duration and time under water

Depth is only half the suffering. The other half is how long you’re stuck below your high-water mark.

Definitions.

Drawdown duration is the time from a peak until equity recovers back to that peak (the full underwater episode, start to finish).
Time under water (TUW) is the total or longest stretch spent below the previous high-water mark — how long you go without setting a new equity high.

These are also random variables with their own distributions, and they’re often more psychologically and commercially damaging than depth. A 20% drawdown that recovers in a month is forgettable; a 20% drawdown that grinds on for three years destroys investor patience, triggers redemptions, and breaks the trader’s nerve — even though the depth is identical.

Worked example. A fund peaks in January 2021, drops 18% by October 2021, then claws back to its old high only in March 2024. The drawdown depth is 18%, but the duration is about 38 months — over three years under water. For an investor who entered at the peak, that’s three years of no new highs, watching every quarterly statement show red against their entry. Many redeem long before the recovery, converting a survivable depth into a realized loss and a closed fund — a drawdown that became ruin through duration, not depth.

Depth gets the headlines; duration ends careers

A deep, fast drawdown is terrifying but brief. A shallow, endless one is what actually breaks people — investors redeem, allocators cut you, and you abandon the system right before it recovers. When you evaluate a strategy, look at the time-under-water distribution as hard as the depth distribution. The longest underwater stretch is often the real survival test.

Match each drawdown concept to its precise meaning.

Pick a term, then click its definition.

Two strategies both have an 18% maximum drawdown. Strategy A recovers in 2 months; Strategy B stays underwater for 3 years. Why might B be far more dangerous despite equal depth?

B is not more dangerous — only depth matters for riskLong time under water drives investor redemptions and trader abandonment, so the survivable depth can turn into realized ruin before recoveryA is more dangerous because fast drawdowns are unpredictableB must have a higher max drawdown that the backtest hid

Using drawdown distributions

What to do with them

Size against the tail, not the median. Assume a realized max drawdown meaningfully deeper than your backtest’s — a 1.5–2× cushion is a sane default — so a tail draw is a bad month, not the end.
Set a drawdown budget that is also a survival limit. Decide the depth at which you’d stop (your ruin barrier from lesson 1), and size so that hitting it is genuinely unlikely under the distribution, not just absent from one backtest.
Watch time under water for capital-stability risk. If your money can leave (investors, your own nerve), the duration distribution may bind harder than the depth distribution.

Pitfalls

Over-trusting a single backtest’s MDD — it’s one sample, and survivorship bias means the strategies you’re even looking at are the ones that got lucky on drawdown.
Ignoring the order dependence — the same trades reshuffled give a different MDD, which is the whole reason Monte Carlo (next lesson but one) resamples them to map the distribution.
Confusing depth and duration — a strategy can be excellent on one and lethal on the other.

Is there a rough formula linking drawdown to a strategy’s return and volatility?

Yes, in approximation. For a strategy modeled as a random walk with drift (per-period mean return $\mu$ and volatility $\sigma$ ), the expected maximum drawdown over a horizon grows roughly with $\sigma^2/\mu$ — meaning drawdown is dominated by the ratio of variance to drift. Double the volatility and expected max drawdown roughly quadruples (it scales with $\sigma^2$ ); double the drift and it roughly halves. This is why low-Sharpe strategies (small $\mu$ relative to $\sigma$ ) suffer punishing drawdowns even when they’re profitable on average: the drift that pulls you back to new highs is weak relative to the noise that drags you down. It also explains the time dimension — expected time under water scales with $\sigma^2/\mu^2$ , so a strategy with half the Sharpe spends roughly four times as long underwater. The practical upshot: a strategy’s drawdown profile is largely a function of its Sharpe ratio, and chasing return without controlling volatility buys you deep, long drawdowns. These are approximations (real returns have fat tails and autocorrelation that worsen the tails), but the scaling intuition — drawdown grows with variance and shrinks with drift — is robust and worth carrying around.

Why is sizing a strategy so that its BACKTESTED maximum drawdown is 'just survivable' a dangerous practice?

The backtested MDD is one draw from a right-skewed distribution; the live drawdown can easily land deeper in the tail and breach your limitBacktested drawdowns are always larger than live ones, so it is overly cautiousIt is fine — the backtest is a guaranteed worst caseDrawdowns never recur once they have appeared in a backtest

Putting it together

A drawdown is the fall from your running peak, and maximum drawdown is the deepest such fall over a path — recorded regardless of recovery, and arithmetically expensive to climb out of (the recovery asymmetry: a 50% loss needs a 100% gain). The key mental shift is that max drawdown is a random variable, not a fixed number: its distribution is right-skewed, your one backtest is a single draw rarely on the bad tail, and live drawdowns can run well past the historical worst. Duration and time under water are separate distributions that often matter more than depth — a shallow, endless drawdown ends careers and triggers redemptions, turning survivable pain into realized ruin. Size against the tail of these distributions (a 1.5–2× cushion over backtest), budget your drawdown as a survival limit, and remember that a strategy’s whole drawdown profile is largely set by its Sharpe ratio — variance deepens and lengthens the pain, drift shortens it.

Big picture

Drawdown distributions — the whole picture

Drawdown distributions
- Depth
  - DD = (peak − equity) / peak
  - Max drawdown = deepest peak-to-trough
  - Recovery asymmetry: 50% loss needs 100% gain
- A distribution, not a number
  - Right-skewed with a long deep tail
  - Backtest MDD is one random draw
  - Live can exceed historical worst
- Duration & time under water
  - Duration = peak to recovery
  - TUW = longest below the high
  - Duration ends careers, triggers redemptions
- Using them
  - Size against the tail (1.5–2× cushion)
  - Budget drawdown as a survival limit
  - Profile is set by the Sharpe ratio

Drawdown is a random variable: depth, duration, and time under water all have skewed distributions, so the backtest number is one draw — size against the tail.

Recap: drawdown distributions

Question 1 of 40 correct

An account peaks at 500, falls to 350, then recovers to 480. What is the maximum drawdown?

70% — the account fell 150 points30% — (500 − 350)/500 = 150/500, measured from the peak to the trough regardless of recovery4% — (500 − 480)/500, measured to the final value43% — (500 − 350)/350, measured from the trough

Check your answer to continue.

Next up — sequencing risk — the most counterintuitive result in the topic: with no cashflows the order of your returns is irrelevant, but add withdrawals or contributions and order becomes destiny. The retiree who meets a crash in year one can run dry while an identical retiree who meets it in year fifteen sails through.