The Sharpe ratio: what it measures, and the four ways it misleads you

The definition

Subtract the risk-free rate from the average return, then divide by the standard deviation of returns. That is the whole formula. The result answers a precise question: how much return did you get per unit of how much the returns bounced around?

The precision is worth holding onto, because most misuse comes from asking the ratio to answer a question it was never built for. It says nothing about direction of the volatility, nothing about the order in which returns arrived, and nothing about the size of positions.

Failure mode one: it punishes upside

Standard deviation treats a large gain and a large loss as the same kind of event. A strategy that occasionally produces an unusually good month is penalised for it. This is not a quirk; it is a direct consequence of using variance as the risk measure.

For strategies with genuinely asymmetric return distributions — trend following being the classic case — this systematically understates quality. The Sortino ratio exists to address exactly this, by dividing by downside deviation only.

Failure mode two: it ignores sequence

Standard deviation does not care about order. A strategy that returns +10, -10, +10, -10 has the same Sharpe as one that returns -10, -10, +10, +10. In the second case you spent two periods underwater before recovering, and you may well have quit.

Maximum drawdown is the measure that captures this, and it is why a Sharpe quoted without a drawdown is an incomplete sentence. Two strategies with a Sharpe of 1.0 can have maximum drawdowns of 8% and 35%.

Failure mode three: it assumes a shape that returns do not have

The statistical interpretation of Sharpe assumes returns are roughly normally distributed. Real trading returns are not: they have fat tails, and they cluster. That means the confidence intervals people attach to Sharpe ratios are usually too narrow, especially over short samples.

The practical consequence: a Sharpe of 2.0 measured over three months is not evidence of a Sharpe of 2.0. It is evidence of a Sharpe somewhere in a wide range that includes zero.

Failure mode four: it scales with leverage and with frequency

This is the one that causes the most confusion. If you double your position size, both your average return and your volatility double, so the Sharpe ratio stays roughly the same while your risk of ruin changes completely.

A related effect works across timeframes. Higher-frequency returns have lower volatility per period, and annualising multiplies by the square root of the number of periods, which flatters high-frequency strategies on paper. Comparing Sharpe ratios computed at different frequencies without adjusting is a common and quiet error.

How to use it anyway

• Quote it with the sample length, always. A Sharpe without a period is not a number.
• Pair it with maximum drawdown, which captures the sequence risk Sharpe ignores.
• Check the distribution: if the strategy relies on rare large wins, look at Sortino as well.
• Be sceptical of very high values. Sharpe above about 3 on a liquid market over a long period usually means an assumption problem, not genius.
• Remember that a search inflates it. Test many variants and the best Sharpe is biased upward — statistical corrections exist for exactly this reason.

The honest summary

Sharpe is a good screening tool and a poor decision rule. It is useful for throwing out strategies that take a lot of volatility for a little return. It is not useful for choosing between two survivors, because it is blind to precisely the thing you have to live with: how the losses arrive.