Drawdown and Sharpe ratio explained

What drawdown actually measures

A drawdown is the decline in account equity from a previous peak to a subsequent trough, expressed as a percentage of the peak. At any moment, your current drawdown is how far below your all-time-high equity you sit. Maximum drawdown is the worst such decline over the whole test or trading period — the single deepest valley on your equity curve. If equity climbs to $12,000 and then sags to $9,000 before making a new high, that episode is a 25% drawdown. Max drawdown is the largest one of these you ever endured.

It matters because returns are abstract until you have to live through the path that produced them. A strategy that "made 30% a year" but spent six months down 40% along the way is a completely different experience from one that ground out 30% with a worst dip of 8%. Drawdown captures that lived experience — the part that makes people panic and abandon a perfectly good system at exactly the wrong moment.

Why drawdown decides survival

The first job of any trader or bot is not to make money — it is to not die. A deep enough drawdown ends the game permanently, because below a certain point recovery becomes mathematically implausible and psychologically impossible. There are three ways drawdown kills an account:

• Mathematically — the deeper you fall, the larger the gain required to climb out (see the recovery math below).
• Psychologically — most people switch off the bot or override it during the worst of a drawdown, locking in the loss right before the rebound.
• Mechanically — leverage and margin calls can force liquidation at the bottom, turning a temporary paper loss into a permanent realised one.

This is why experienced practitioners obsess over drawdown more than over return. A strategy you can actually hold through its worst stretch is worth more than a higher-returning one you will abandon. You can study a strategy's depth and recovery directly on the equity curve in the backtester, and pair it with sane risk management and position sizing.

The brutal recovery math

Here is the asymmetry that catches beginners off guard. A loss and the gain needed to undo it are not equal, because the recovery gain is calculated on a smaller base. The formula is simple:

recovery mathrequired_gain = 1 / (1 - drawdown) - 1

10% loss  -> need +11.1% to recover
20% loss  -> need +25.0%
33% loss  -> need +50.0%
50% loss  -> need +100.0%   (must DOUBLE)
75% loss  -> need +300.0%
90% loss  -> need +900.0%

Lose half your account and you do not need another 50% to get back to even — you need to double what is left. Lose 90% and you need a tenfold gain just to break even, which essentially never happens. This is the entire argument for keeping drawdowns shallow: small losses are recoverable, large ones compound into a trap. It is also why uncapped leverage is so dangerous — it magnifies the very drawdowns that the recovery math punishes most severely.

The Sharpe ratio: return per unit of risk

The Sharpe ratio, introduced by William F. Sharpe, answers a question raw return cannot: how much extra return did you earn for each unit of volatility you accepted? It rewards smooth, consistent gains and penalises wild swings. Two strategies that both return 12% a year are not equal if one did it calmly and the other lurched between +40% and -25% months. The formula divides excess return by the standard deviation of returns:

sharpe ratioSharpe = (Rp - Rf) / sigma_p

  Rp      = portfolio (strategy) return
  Rf      = risk-free rate (e.g. T-bill yield)
  sigma_p = standard deviation of the strategy's returns

# annualise from periodic returns:
Sharpe_annual = mean(r) / std(r) * sqrt(periods_per_year)
# periods_per_year = 252 for daily, 12 for monthly

The numerator is your reward above a do-nothing risk-free return; the denominator is the volatility — the risk — you stomached to get it. A higher Sharpe means more reward per unit of risk. Crucially, Sharpe says nothing about direction on its own; you annualise it consistently with the data frequency, and you must compare Sharpe numbers only when they were computed the same way.

What counts as a "good" Sharpe

There is no universal threshold, but a working rule of thumb for a real, live, after-cost strategy looks like this:

1. Below 1.0 — weak. The returns barely compensate for the volatility; often not worth the effort versus simply holding an index.
2. 1.0 to 2.0 — solid. This is where most genuinely good, sustainable strategies live over long periods.
3. 2.0 to 3.0 — strong, and rare to sustain after real-world costs.
4. Above 3.0 — in a short retail backtest, treat this as a red flag for overfitting rather than a trophy. Sustained Sharpes that high usually require infrastructure and edges retail does not have.

Context changes everything: the risk-free rate used, the measurement period, and the return frequency all move the number. The right way to read a Sharpe is alongside the equity curve and the drawdown, never in isolation.

Why a dazzling backtest Sharpe usually lies

The Sharpe ratio is trivially easy to inflate on historical data, which is exactly why a sky-high backtest Sharpe should make you suspicious rather than excited. The common ways a fake-good Sharpe appears:

• Curve-fitting — tuning parameters until the past looks perfect bakes in noise that will not repeat. This is classic overfitting.
• Ignoring costs — leaving out fees, spread and slippage flatters both return and consistency; real trading erodes both.
• A short, lucky window — a strategy can post a wonderful Sharpe over one favourable market regime and fall apart in another.
• Look-ahead bias — accidentally using future information makes the equity curve unnaturally smooth, spiking Sharpe artificially. This is one of the main reasons most trading bots fail.

The antidote is honest testing: hold out data the strategy never saw, then run forward testing and paper trading before risking a cent. A modest Sharpe of 1.3 that survives out-of-sample is worth far more than a 4.0 that only exists in-sample. The numbers only mean something once you understand what a trading strategy is and whether it has a real edge.

Sortino and Calmar: the close cousins

Two related metrics fix specific blind spots in Sharpe. The Sortino ratio replaces total volatility with downside deviation only — it stops punishing a strategy for big upside swings, which most traders would happily accept. The Calmar ratio ties return directly to the pain measure from earlier: it divides annualised return by the maximum drawdown, so it directly rewards strategies that achieve returns without deep holes.

related ratiosSortino = (Rp - Rf) / sigma_downside   # penalises only downside vol
Calmar  = annual_return / abs(max_drawdown)  # return per unit of worst loss

No single number is sufficient. The disciplined way to evaluate a strategy is to read drawdown, Sharpe, Sortino and Calmar together, then sanity-check them against the raw equity curve — and never trust any of them until they have held up out-of-sample. For more on judging a system as a whole, see how to evaluate a trading strategy.

Frequently asked questions