Skip to content
Financial Context Wiki

Sortino Ratio: Return Above a Target per Unit of Downside Deviation

For educational purposes only; not investment advice.

Direct answer

Section titled “Direct answer”

The Sortino ratio compares a portfolio’s average return above a specified target with the variability of returns below that target:

Sortino ratio = [mean(R) - T] / downside deviation relative to T

T is the minimum acceptable return, or MAR, for the same period as each observation. Unlike the Sharpe ratio’s standard-deviation denominator, downside deviation does not penalize returns merely for being above the target. The result depends heavily on the target and exact downside-deviation convention, so a reported number without those definitions is incomplete.

Target and denominator conventions

Section titled “Target and denominator conventions”

For returns Ri and periodic target T, one common full-sample definition is:

downside deviation = √[Σ min(0, Ri - T)² / N]

Every observation is represented in N; periods at or above the target contribute zero. Another implementation divides by only the number of returns below the target, sometimes with sample-degree adjustments. These are not algebraically equivalent. A data provider may also use a risk-free rate, zero, an actuarial hurdle, or a custom required return as T.

The target must use the same frequency and compounding basis as returns. An annual target cannot be inserted directly into monthly deviations. State whether returns include distributions, are gross or net of fees, and are arithmetic period returns or another series.

Annualization requires an explicit method. Under restrictive independent, identically distributed assumptions, a periodic ratio is sometimes multiplied by √N. Serial correlation, smoothed valuations, changing target rates, volatility clustering, and asymmetric returns can invalidate that shortcut. Calculating downside deviation from annual returns is not necessarily equivalent to scaling a monthly result.

Sortino answers a preference-specific question: how much average return exceeded a threshold per unit of below-threshold dispersion. It does not measure maximum drawdown, recovery time, loss probability, expected shortfall, liquidity, or the size of a single ruinous tail event.

Worked example

Section titled “Worked example”

Suppose six monthly returns are 4%, -2%, 3%, -1%, 2%, 0%, and the monthly target is T = 0%. Their arithmetic mean is 1%. Only -2% and -1% contribute squared shortfalls.

Using all six observations in the denominator:

downside deviation = √[(2² + 1²) / 6] = 0.91%

Sortino ratio = (1% - 0%) / 0.91% = 1.10

Using only the two below-target observations instead:

downside deviation = √[(2² + 1²) / 2] = 1.58%

Sortino ratio = 1% / 1.58% = 0.63

Neither value is a transcription error; they answer the same broad idea using different denominator conventions. A comparison is valid only after fixing the convention, target, sample dates, frequency, and fee treatment.

Now change the monthly target to 1%. The 0% month also becomes a shortfall, the numerator becomes zero, and the ratio becomes zero under either denominator convention. The return series did not change; the investor’s hurdle did. This target sensitivity is part of the metric, not a defect to conceal.

Interpretation checklist

Section titled “Interpretation checklist”
  • Publish the formula, target source, return frequency, denominator convention, annualization method, sample period, observation count, and fee basis.
  • Count below-target observations. A very high ratio based on one small downside observation is statistically fragile.
  • Inspect the worst returns and their sequence. Squaring shortfalls gives large losses more weight, but a short sample can entirely miss rare tail events.
  • Check stale or model-based prices. Smoothed marks can suppress measured downside deviation without reducing economic risk.
  • Use a target appropriate to the decision. Zero, cash return, inflation, liability growth, and a required portfolio return represent different objectives.
  • Compare portfolios over identical dates and currencies. Different market regimes or targets invalidate a simple ranking.
  • Pair the ratio with cumulative return, maximum drawdown, recovery duration, expected shortfall, skewness, leverage, liquidity, and stress scenarios.
  • Avoid optimizing a strategy solely for historical Sortino. Multiple testing and asymmetric payoff strategies can generate attractive backtests with hidden crash exposure.

If no observations fall below the target, downside deviation is zero and the ratio is undefined or infinite depending on software. That is a warning about insufficient observed downside, not proof of a risk-free strategy.

Common misconceptions

Section titled “Common misconceptions”
  • “Sortino ignores losses.” It ignores deviations above the target; losses and sub-target gains contribute to downside deviation.
  • “The target must be zero.” MAR is a chosen hurdle and must be disclosed.
  • “All platforms calculate the same number.” Denominator, target, frequency, annualization, and fee conventions vary.
  • “A higher Sortino always means a safer investment.” It describes one historical return distribution relative to a target and omits many risks.
  • “Sortino is always superior to Sharpe.” They answer different questions; choosing downside risk embeds a preference rather than removing assumptions.
  • “No downside observations means perfect performance.” It may mean the sample is short, prices are smoothed, or a rare loss has not occurred.
Section titled “Related topics”

Authoritative sources

Section titled “Authoritative sources”