Skip to content
CyclepediaThe trading encyclopedia
ITENESPT-BRFRDE

Sortino ratio and downside deviation

Mean return above a target per unit of downside deviation, with the target, frequency, denominator and annualisation convention made explicit.

On this page

Who this is for — Readers distinguishing total dispersion from returns that fall below a required objective, with the target recognised as part of the metric.

The Sortino ratio compares mean return above a minimum acceptable return with downside deviation calculated relative to that same target. Unlike the Sharpe ratio, its denominator does not penalise every deviation from the mean; it includes only observations below the target under a stated convention.

The target is often called MAR (minimum acceptable return). It may be zero, a risk-free rate, a periodic objective, inflation or a liability-related required return. Choosing zero by habit changes the question: a positive return below the true objective would be counted as success.

Sortino ratio and downside deviation Risk is measured relative to a stated target return Sortino ratio and downside deviation Risk is measured relative to a stated target return Sortino = mean(rₚ − T) / √mean[min(0, rₚ − T)²] Target T It may be zero, a hurdle oranother target, but it must bestated. Downside deviation Only shortfalls below thetarget enter the squaredsemideviation. Sample Frequency, window andobservations above targetaffect the ratio. Comparison It differs from Sharpe throughthe risk definition, not merelythe name. Cyclepedia · educational diagram: state conventions, period and data
The target divides the distribution. Changing it changes both differential return and the observations that enter downside risk.

A reproducible definition

For periodic returns Rₜ and periodic targets MARₜ, one common convention is:

DD = √[(1/N) × ∑ₜ min(0, Rₜ − MARₜ)²]
Sortino = [mean(Rₜ) − mean(MARₜ)] / DD

For a constant target, the numerator is mean(R)−MAR. The displayed denominator divides by all N observations. Dividing only by the number below MAR produces a higher downside deviation and a different ratio. Under this convention it is an error, not an equivalent form. Lower-partial- moment formulations with other orders and normalisations also exist, so the name alone does not guarantee identical provider calculations.

The target's role

MAR should share the returns' frequency, period, currency and FX treatment, nominal or real basis, gross or net status, calendar and horizon. An annual target is not always divided by twelve; a coherent compound conversion for equivalent months is (1+MAR_annual)^(1/12)−1. A time-varying target requires preserving the full MARₜ series.

A dollar risk-free rate is not automatically a euro portfolio's target, and headline inflation may not match the relevant liability. Economic purpose precedes calculation.

Sortino and Sharpe ask different questions

Aspect Sharpe Sortino
reference benchmark for differential return minimum acceptable return
risk standard deviation of the full differential series deviations below target
positive deviations increase dispersion do not enter downside denominator
main sensitivity mean, volatility, autocorrelation mean, MAR, shortfall frequency
degenerate case standard deviation zero downside deviation zero

A higher Sortino than Sharpe does not prove a better strategy; the denominators are different. Sortino can become very large when few shortfalls occur, which is also when downside-risk uncertainty may be greatest.

Example with two conventions

Four hypothetical percentage returns are 2, 1, −1, 2, with MAR zero and mean 1%. Using all observations:

DD = √[(0² + 0² + (−1)² + 0²) / 4] = 0.5%

Periodic Sortino is 1/0.5 = 2. Dividing only by the single negative observation gives downside deviation 1% and ratio 1. Formula and denominator must therefore accompany the result. Neither number creates a universal quality band.

Annualisation and limits

Numerator and downside deviation may be annualised under coherent assumptions, but multiplying the ratio by √m is not a universal law. Autocorrelation, a changing target, conditional volatility, overlapping returns and smoothing change scaling. A cautious report may calculate directly at the desired frequency and also show an unannualised result.

Sortino depends strongly on MAR; compresses shortfall frequency and severity into one quadratic measure; ignores return order and underwater duration; can be unstable with few failures; does not describe unobserved tails; and does not correct selection bias, cost or illiquidity. It is undefined when downside deviation is zero and does not prove skill or causality.

A payoff with frequent small returns above MAR and one rare severe loss can look attractive until the tail event enters the sample. Drawdown, scenarios, the full distribution and valuation quality remain necessary.

Reporting checklist

Publish the formula and denominator; target and economic rationale; frequency, window and currency; simple or log returns; price or total return; costs and fees; annualisation and serial-dependence treatment; number and depth of shortfalls; drawdowns and scenarios; and ex-ante, backtest or live status.

Common error — Dividing downside deviation only by negative periods and comparing it with a provider that divides by all observations. The same label then hides different ratios.

Sources

Suggest an edit My proposals