What it means
Standard deviation, the usual measure of investment volatility, treats an unexpectedly good month exactly like an unexpectedly bad one. Most investors do not think that way, because a surprise gain is not a problem to be managed.
Downside deviation fixes this by counting only the periods where the return fell below a threshold the investor cares about. That threshold is called the minimum acceptable return, or MAR.
It is often set at zero, meaning any losing period counts, but it can equally be the risk-free rate, an inflation target or the return a pension scheme needs to meet its obligations. Choosing the MAR is a judgement about what the money is actually for, not a technical detail.
The calculation takes each period's shortfall below the MAR, squares it, adds the squares, divides by the total number of periods and takes the square root. Periods above the MAR contribute zero rather than being dropped from the count, which keeps consistently steady funds from being penalised.
The result is expressed in the same units as the returns themselves, usually a monthly or annual percentage. Downside deviation is most often seen as the denominator of the Sortino ratio, which divides excess return by downside deviation to show how much reward an investor earned per unit of bad volatility.
Two funds with identical standard deviations can have very different downside deviations if one of them achieves its volatility mostly through upside surprises. That difference is exactly what a cautious investor wants to see.
One nuance trips people up frequently. Some providers divide by the number of periods that fell below the MAR rather than the total number of periods, which produces a larger figure and makes comparison across sources unreliable.
Always check which convention is in use before comparing two published numbers.
In practice
Real-world examples
Example 1
A charity's investment committee compares two balanced funds with identical 9% annual standard deviations. Fund A has a downside deviation of 5.5% and Fund B has 8.2%, so the committee selects Fund A because its volatility comes mostly from strong months rather than losing ones.
Example 2
A family office sets its minimum acceptable return at 4%, the rate it needs to cover charitable distributions. Measured against that threshold rather than zero, a bond portfolio that looked calm on a standard deviation basis shows a meaningful downside deviation, because several years returned 2% or 3%.
Example 3
A hedge fund manager reports a Sortino ratio alongside the Sharpe ratio in her quarterly letter. Because her strategy produces occasional large gains, the downside deviation is far lower than the standard deviation, and the Sortino ratio of 1.9 tells a more flattering and more accurate story than the Sharpe ratio of 0.9.
Think of it
Imagine driving a car with a speed limit. Standard deviation measures every change in your speed, whether you speed up or slow down. Downside deviation only cares when you drop below the minimum speed limit and risk stalling.
Formula
Calculation
Downside deviation = square root of [ sum of (the smaller of zero and (Return - MAR)) squared, divided by n ]
Where n is the total number of periods observed.
A fund reports eight monthly returns: 3%, 5%, -4%, 2%, 6%, -4%, 1% and 4%. The investor sets the MAR at 0%, so only the two losing months contribute.
Shortfalls: -4% and -4%
Squared shortfalls: 16 and 16, giving a total of 32
Divide by the total number of periods: 32 / 8 = 4
Downside deviation = square root of 4 = 2.00% per month
To annualise, multiply by the square root of 12: 2.00% x 3.464 = 6.93%.
The Sortino ratio follows directly. The average monthly return is (3 + 5 - 4 + 2 + 6 - 4 + 1 + 4) / 8 = 13 / 8 = 1.625%, so the ratio against a 0% MAR is 1.625% / 2.00% = 0.81. The fund earned 0.81% of average monthly return for every 1% of monthly downside deviation.Case study
Seen in the real world
Meridian Trust Partners is a fictional multi-family office used here for illustrative purposes. Its clients were unsettled by a year in which the reported volatility of their portfolios rose sharply, even though almost none of them had lost money over the period.
The investment team recalculated the portfolios using downside deviation against a MAR of 3%, the rate most clients needed to preserve purchasing power. The exercise showed that roughly two thirds of the reported volatility came from months that beat the threshold, often by a wide margin. Downside deviation for the flagship portfolio was 4.1% a year against a standard deviation of 11.2%.
Meridian rebuilt its client reporting around the new measure, showing standard deviation, downside deviation and the Sortino ratio side by side with a plain-English note explaining the difference. Redemption enquiries fell over the following two quarters, and the team found that conversations about risk became far more productive once clients could see that not all volatility was working against them.
Watch out
Common mistakes
- Assuming downside deviation is simply half of standard deviation, when the relationship depends entirely on how the returns are distributed.
- Dividing by the number of losing periods instead of the total number of periods, which inflates the figure and makes the fund look worse than it is.
- Comparing downside deviations calculated against different minimum acceptable returns, which is like comparing two exam scores marked out of different totals.
Questions
People also ask
What minimum acceptable return should I use?
Use the return the money genuinely needs to earn, which is often zero for a general portfolio, the risk-free rate for a trading strategy, or a funding target for a pension or endowment.
Does a low downside deviation mean the investment is safe?
No, it means losing periods have been small or rare in the data observed, and a strategy that sells insurance-like exposure can show a very low figure right up until a single large loss appears.
How many periods do I need for the figure to mean anything?
Thirty-six monthly observations is a common working minimum, because short samples can easily contain no losing months at all and produce a downside deviation of zero.