Downside deviation
Also called downside risk, target downside deviation
A volatility measure that counts only returns below a chosen target, such as zero or a minimum acceptable return. Unlike standard deviation, it ignores returns above the target, so upside swings do not count as risk.

On this page· 5 sections
Standard deviation treats a surprise gain exactly like a surprise loss. It assumes a symmetrical return distribution, so upside and downside deviations are penalized equally[1]. Downside deviation drops the gains from the count and measures only how far returns fall short of a goal.
Harry Markowitz made the same distinction in 1959 with a measure called semivariance, but the calculation was too heavy without computers, and standard deviation became the standard[1].
How it works
First pick a target return. Kidd calls it the minimum acceptable return (MAR). It may be an absolute return, an index return, the risk-free rate or zero[1]. Semideviation, the older cousin, measures shortfalls from the mean. Downside deviation measures them from the target instead[1].
For each period, the shortfall is how far the return fell below the target. A return at or above the target counts as zero.
Downside deviation = √(sum of squared shortfalls ÷ number of periods)
The divisor is the total number of returns, not only the ones below the target. Dividing by the count of the losers is a common mistake[1].
Several versions exist. The simple “discrete” method looks only at what happened. Sortino and Forsey (1996) argue it can understate risk, especially when most returns are positive, and prefer a method that fits a continuous curve to the returns. Kidd’s example is the Japanese equity market of the 1980s, where 45 monthly observations and a monthly MAR of 0.08% gave 2.74% by the discrete method and 3.20% by the continuous one[1]. The same article warns against annualizing a discrete downside deviation the way a standard deviation is annualized[1].
Example
Say a strategy’s six monthly returns are +3%, −2%, +1%, −4%, +5% and 0%, with a target of 0%.
- Shortfalls: 0, 2, 0, 4, 0 and 0 points. Squares: 4 and 16, adding up to 20.
- Divide by all six months: 20 ÷ 6 = 3.33.
- Square root: √3.33 ≈ 1.83%.
Dividing by the two losing months instead gives √10 ≈ 3.16%, which overstates it. Now change the +5% month to +10%. Downside deviation stays at 1.83%, since nothing below the target changed. The sample standard deviation climbs from 3.27% to 4.89%.
Watch out
The target changes the answer. In the R package that implements the Sortino ratio, the MAR defaults to zero. Its documentation notes that a MAR set too low will not capture the risks that concern the investor, and one set too high will portray a sound investment unfavorably[2].
Few down periods make a thin sample. With mostly positive returns, the discrete method can understate the risk by a wide margin[1].
The input is history. Investors use these statistics to judge future volatility, but they are computed from past returns[1].
Sources
- Professional bodyCFA InstituteThe Sortino Ratio: Is Downside Risk the Only Risk that Matters? (Deborah Kidd, CFA, Investment Performance Measurement, 2012)Checked November 9, 2026
- Platform docsCRAN, The R ProjectSortinoRatio: calculate Sortino Ratio of performance over downside risk (PerformanceAnalytics)Checked November 9, 2026
