Quant Analytics: from scratch to pro · Portfolio, sizing and risk metrics · Intermediate
Drawdown statistics: maximum drawdown, duration and the Calmar ratio
Drawdown measures the pain between a peak and the next low. Maximum drawdown, time under water and the Calmar ratio describe how bad it got, how long it lasted and what the return paid for it. Even a strategy with no edge produces deep drawdowns over long samples.
In this lesson
- Compute the drawdown series and maximum drawdown from a list of values
- Measure drawdown duration and recovery time
- Compute a Calmar ratio and say what it leaves out
- Explain why the maximum drawdown of a sample grows with its length
Read first: Drawdowns and the maths of recovery, Volatility explained
A portfolio that gains 10% a year with a 50% hole in the middle is a different thing from one that gains 10% smoothly. Volatility treats both as fluctuations around an average. Drawdown asks the question investors actually feel: how far below the previous high have I been, and for how long?
Definitions
The running peak at time t is the highest value up to t. The drawdown is the percentage distance below that peak: DD(t) = V(t) / peak(t) - 1. It is zero at a new high and negative otherwise. The maximum drawdown is the most negative value the drawdown ever reaches: the worst peak-to-trough loss in the sample. See maximum drawdown and the recovery arithmetic in drawdowns and recovery maths.
Two more statistics complete the picture. Drawdown duration is the time from a peak until a new high; time under water is the share of the sample spent below the running peak. The deepest drawdown and the longest one are often different episodes.
Worked example
A portfolio's year-end values over six years are 100, 120, 90, 105, 130, 104, 140.
- The peak of 120 is followed by a low of 90: a drawdown of 25%. The portfolio is under water until year 4 (130), three years.
- The peak of 130 is followed by 104: a drawdown of 20%, recovered in year 6.
- The maximum drawdown is 25%. To recover it needs a gain of
120 / 90 - 1, which is 33.3%.
The annualised return over six years is (140 / 100)^(1/6) - 1, which is 5.77%.
The Calmar ratio
The Calmar ratio divides the annualised return by the maximum drawdown: Calmar = annualised return / |maximum drawdown|. Here it is 5.77% / 25%, which is 0.23. It plays the role of the Sharpe ratio with a different risk measure, the worst loss instead of the typical swing. Conventions differ on the window (often the last three years) so state it. See Calmar ratio.
Drawdown depends on sample length
The maximum is the worst of everything seen so far, so the longer the sample, the deeper it tends to be. For a Brownian motion with zero drift, Magdon-Ismail and co-authors (2004) give an exact expected value. It reduces to E[MDD] = sqrt(pi/2) x sigma x sqrt(T), about 1.25 x sigma x sqrt(T), with sigma the annual volatility and T the years. This is an approximation when applied to percentage drawdowns of a compounding portfolio.
With volatility of 15%, the expected maximum drawdown is 18.8% over one year and 37.6% over four, as the chart shows. A process with no edge at all produces both. A positive drift reduces these figures, but the message holds: a short, shallow drawdown is weak evidence of low risk, and a record of ten years deserves a deeper expected number than a record of one.
Real episodes are deeper than a normal model suggests, because volatility clusters and losses come in runs. See the 2008 financial crisis and the 2020 Covid crash.
Common mistakes
- Reading the maximum drawdown of a short record as the risk of the strategy.
- Comparing drawdowns across different frequencies. Intraday values give deeper numbers than month-end values.
- Quoting a Calmar ratio without its window.
- Ignoring duration. A shallow drawdown that lasts five years is hard to live through.
- Using the maximum drawdown as an optimisation target. It is one episode and a noisy statistic.
- Computing drawdown on returns that are not compounded (adding percent changes instead of chaining them).
In practice
Report maximum drawdown with its date range, the longest time under water and the expected-versus-observed comparison for the sample length. A longer record tends to show a deeper worst drawdown than a shorter one. Analytics on FinUties applies measures like these to the markets it covers, so they are at hand when you read a chart.
Expected maximum drawdown against sample length, no edge
- volatility 10%
- volatility 15%
Worked numbers
Each result is recomputed from the formula shown.
| What | Formula | Result |
|---|---|---|
| first drawdown, 120 to 90 | 90 / 120 - 1 | 25% |
| second drawdown, 130 to 104 | 104 / 130 - 1 | 20% |
| gain to recover 25% | 120 / 90 - 1 | 33.3% |
| annualised return, 100 to 140 over 6 years | (140 / 100)**(1 / 6) - 1 | 5.77% |
| Calmar ratio | cagr / 0.25 | 0.23 |
| expected max drawdown over 1 year, volatility 15% | sqrt(pi / 2) * 0.15 * sqrt(1) | 18.8% |
| expected max drawdown over 4 years | sqrt(pi / 2) * 0.15 * sqrt(4) | 37.6% |
Check yourself
A portfolio falls from 200 to 150. What drawdown is that, and what gain restores the peak?
150 / 200 - 1 = -25%. To get back to 200 it must gain 200 / 150 - 1 = 33.3%.
Annual return 12%, maximum drawdown 30%. What is the Calmar ratio?
12% / 30% = 0.40.
A track record shows a maximum drawdown of 8% over one year. Does that describe the risk of the strategy?
Only weakly. The maximum drawdown is the worst of one sample. A longer record, or the same process with a different order of returns, typically shows a deeper one. For a driftless process the expected maximum drawdown over T years is about 1.25 x volatility x sqrt(T).
Why is the maximum drawdown a poor target to optimise?
It depends on a single path and a single worst episode, so it is a noisy statistic. Optimising to it fits that one episode.
Key terms
Sources
- On the maximum drawdown of a Brownian motion. Journal of Applied Probability 41(1), 147-161 (Malik Magdon-Ismail, Amir F. Atiya, Amrit Pratap, Yaser S. Abu-Mostafa), 2004. Accessed 9 Nov 2026
- The Sharpe Ratio. The Journal of Portfolio Management 21(1), 49-58 (William F. Sharpe), 1994. Accessed 9 Nov 2026