QTSurfer beta
Learn

Drawdown

Learn how drawdown measures the fall from a previous equity peak, how to calculate maximum drawdown and its duration, and why it changes how a backtest should be read.

A drawdown is the decline in account value from a previous peak. It starts the moment equity drops below its running maximum and ends only when a new maximum is set. Maximum drawdown is the deepest such decline over the whole period measured.

Total return says where a strategy finished. Drawdown says what it cost to get there: how much of the account was lost from a high point, how long the loss lasted, and whether the trader would have still been running the strategy when it recovered.

Calculation

Let E(t) be the equity at time t and P(t) the running peak, P(t) = max E(s) for all s ≤ t. The drawdown at t is:

DD(t)  = E(t) − P(t)             absolute, in quote currency
DD%(t) = E(t) / P(t) − 1         relative to the peak

Both are zero or negative. Maximum drawdown is the minimum of either series over the period:

MaxDD  = min DD(t)
MaxDD% = min DD%(t)

Two duration measures usually travel with the depth:

  • Drawdown duration — time from the peak to the trough.
  • Recovery time — time from the trough back to a new peak. A drawdown still open at the end of the period has no recovery time, and that absence is itself information.

Worked example

An account starts at 100, rises to 110.5, then falls to 90.25. The running peak at the third point is 110.5, so:

DD  = 90.25 − 110.5        = −20.25
DD% = 90.25 / 110.5 − 1    = −0.183   →  −18.3 %

Note that the relative drawdown is measured from the peak, not from the starting capital. Against the initial 100 the account is down 9.75 %; against its own peak it is down 18.3 %. The second number is the one a trader would have experienced.

What drawdown adds to a result

  • Path risk. Two strategies can end at the same equity while one spent months 30 % below its peak and the other never fell more than 5 %. The final return is identical; the experience, the margin requirements, and the probability of abandoning the strategy are not.
  • Position sizing. A strategy’s historical maximum drawdown is the minimum loss to plan for, not the maximum. Future drawdowns tend to exceed the largest one seen in a finite sample.
  • Return per unit of pain. Ratios such as CAGR / |MaxDD%| (often called the Calmar ratio) compare growth with the deepest loss required to obtain it.
  • Consistency. A single deep drawdown surrounded by smooth gains suggests dependence on one market episode; many shallow drawdowns suggest a strategy whose risk is spread across time.

Common mistakes

  • Comparing absolute drawdowns across different capital. −20.25 means something different on an account of 100 and an account of 10,000. Compare percentages, or normalise first.
  • Reading depth without duration. A 10 % drawdown recovered in a week and a 10 % drawdown that lasted a year are not the same risk.
  • Treating the historical maximum as a ceiling. The sample contains one path. Longer samples and live trading almost always find a deeper one.
  • Optimising for the smallest drawdown. A strategy that rarely trades has a small drawdown and little else. Drawdown is a constraint to respect, not an objective to minimise on its own.
  • Measuring from a downsampled curve. A curve reduced to a few hundred points can miss the exact trough or the exact peak that defined the maximum drawdown. Compute it from the full series or use the metric the engine reports.

Drawdown in QTSurfer

A completed backtest reports both forms in its results: maxDrawdown in quote currency and maxDrawdownPercent relative to the peak, alongside the Sharpe ratio, Sortino ratio, CAGR, and trade count. The equity curve behind them is available in the same response, so the depth can be located in time.

In a parameter sweep, every leaderboard row carries maxDdPct, and maxdd is one of the four objectives a sweep can rank by. Rank by drawdown only together with the trade count: rows below the trade floor are flagged belowTradeFloor precisely because a low drawdown built on very few trades is not evidence of control.

When an equity curve is resampled for display, the transform preserves the first and last points and the global extrema, so the highest and lowest equity values survive. The specific peak-to-trough pair that defines the maximum drawdown is not guaranteed to, which is why the reported metric, not a value read off a compact chart, is the figure to quote.