Volatility & bands

Downside Deviation

Volatility measured only from returns that fall below a target — the risk that actually hurts.

Works in most conditionsEngine-computed on a fixed sample series
14512096Rising = expanding, falling = fading
Downside Deviation 8.43How to read Downside Deviation on the chart — the callouts mark what to look for.

The formula

For each return, take its shortfall below the target (the return minus the target when that is negative, otherwise zero), square it, average across all N observations, and take the square root. Upside returns contribute nothing.

Downside Deviation = √( Σ (shortfall below target) squared ÷ N )
Worked example
ReturnShortfall below 0%Squared
+3%00
−2%−2%4
+1%00
−4%−4%16
+2%00
Sum20

√(20 ÷ 5) = √4 = 2% downside deviation (target = 0%)

What it is

Downside Deviation is a risk measure that counts only the volatility that actually hurts — the returns that fall below a target — while ignoring upside swings entirely. Standard deviation treats every deviation from the mean as risk, penalising a strategy just as much for large gains as for large losses, which many investors find unfair, since nobody complains about upside surprises. Downside Deviation fixes that by measuring only the dispersion of returns beneath a chosen minimum acceptable level, often zero or the risk-free rate. For a beginner, it answers a more honest question than plain volatility: how badly and how often does this strategy fall short of what I need, ignoring how pleasantly it sometimes overshoots? It is the risk lens behind the Sortino ratio and is especially suited to judging strategies whose returns are lopsided rather than symmetric. It isolates harmful volatility from benign gains.

How it is calculated

You begin by choosing a minimum acceptable return, sometimes called the target or MAR, which is frequently zero or the risk-free rate. For each period, you compute the shortfall below that target: if the return is below the target, the shortfall is the amount by which it fell short, and if the return is at or above the target, the shortfall counts as zero. You then square each of those shortfalls, average them, and take the square root — much like a standard deviation, but with only the below-target shortfalls contributing and the above-target periods entering as zeros. A subtle but standard convention is that the average is taken over the total number of periods, including the zero-contribution winning periods, not just over the losing ones. The result is a single number in the same units as the returns, expressing the typical magnitude of downside relative to the target.

Reading it, step by step

A higher Downside Deviation means larger or more frequent shortfalls below your target, and a lower one means the downside is well contained. Because it ignores upside, two assets with identical standard deviations can have very different Downside Deviations — the one that falls more violently or more often below the target will score higher, correctly flagging it as the riskier of the two for someone who cares about losses. Read it always in relation to the target you chose, since the same return stream produces different values against a zero target versus a risk-free-rate target. It is most illuminating when comparing strategies: the one with the lower Downside Deviation for a given return delivered its results with less painful volatility. Interpreted alone it is just a number, but as a comparative and as the denominator of the Sortino ratio it becomes a genuine judge of downside-adjusted quality.

Best timeframes

  • ScalpingRarely usedtoo few losses
  • Day tradingDaily returns, 1 – 3 mo
  • SwingDaily/weekly, 6 – 12 mo
  • PositionMonthly returns, 3+ yrmost robust

This is a portfolio-level risk statistic computed over a return series, not a chart timeframe — it needs enough downside observations to be stable.

Downside deviation vs other risk measures

Downside Dev.Std DeviationMax Drawdown
Counts only lossesYesNoYes
Needs a target returnYesNoNo
Captures worst-case tailNoNoYes
Feeds ratioSortinoSharpeCalmar

Best timeframes and settings

The two settings that matter are the return frequency and the target. Downside Deviation is computed from a series of periodic returns — daily, monthly, or annual — and the choice should match the horizon you care about, with monthly returns common for evaluating strategies and funds. The target return is the pivotal input: zero treats any loss as downside, while the risk-free rate treats underperformance versus cash as downside, and each produces a different value. It is not a chart overlay tuned for scalping or swing trading but a portfolio-level statistic, so the relevant decision is how many observations you feed it and which threshold you set. More observations make the estimate more stable, which matters because the measure depends on having enough below-target periods to be reliable. Always report the target alongside the number, since the value is meaningless without it.

When and where to use it

Downside Deviation is used to evaluate and compare investment strategies, funds, and return streams rather than to time individual trades, and it fits any asset class where you have a return history. It is especially valuable for asymmetric or option-like return profiles, where standard deviation misleads by penalising the very upside that makes such strategies attractive. It shines when comparing two strategies with similar total volatility but different loss behaviour, revealing which one achieves its returns with less harmful downside. Avoid relying on it when you have too few loss observations to be statistically stable, and never treat it as a complete risk picture, because it describes typical downside, not the worst-case tail. Pair it with drawdown and tail-risk analysis for the full story. Its natural home is performance evaluation and strategy selection, particularly through the Sortino ratio.

Strategies that use it

The first application is Sortino-ratio strategy selection: divide a strategy's excess return over the target by its Downside Deviation, and prefer strategies with higher Sortino ratios, since they deliver return per unit of harmful volatility rather than per unit of total volatility. The second is asymmetric-strategy evaluation: for option-selling, trend-following, or other lopsided return streams, use Downside Deviation instead of standard deviation so the strategy is not unfairly penalised for its upside variance. The third is comparative risk screening across a portfolio of strategies or funds: rank candidates by Downside Deviation against a common target to surface those whose losses are best contained. In each case the target return must be chosen deliberately and held consistent across the comparison, because changing it changes every number, and the whole point is a fair, downside-focused comparison rather than an absolute verdict.

Combining it with other indicators

Downside Deviation is one piece of a risk mosaic and is most useful alongside complementary measures. It combines directly with return to form the Sortino ratio, which is its primary and most important pairing. It should sit next to drawdown analysis, because Downside Deviation describes typical below-target volatility while maximum drawdown describes the worst peak-to-trough loss, and the two answer different questions. Standard deviation shown beside it reveals how asymmetric a return stream is: a large gap between the two suggests upside variance is inflating the standard deviation. Volatility estimators such as Parkinson or Garman-Klass and range measures like ATR round out the picture for the underlying instrument. The unifying idea is that no single risk number is sufficient, and Downside Deviation specifically covers harmful dispersion, which must be paired with a tail measure to be complete.

Where it fails

The measure is highly sensitive to the chosen target return, so a careless or inconsistent threshold can make one strategy look better or worse than another for no real reason. It also needs enough below-target observations to be stable, and with short histories or rare losses the estimate can be noisy and unreliable. Its biggest blind spot is the tail: Downside Deviation summarises typical downside but says nothing about the worst-case catastrophe, so a strategy with modest routine downside can still harbour a devastating rare loss that this measure will not reveal. Beginners sometimes treat it as a full risk verdict or compare values computed against different targets. The fixes are to fix the target explicitly and keep it consistent across comparisons, to require a sufficient sample of losing periods, and always to pair it with drawdown and tail analysis so the rare disaster is not overlooked.

A worked example

Suppose a strategy produces six monthly returns of plus three, minus two, plus one, minus four, plus two, and minus one percent, and you set the target at zero. Only the negative months count as shortfalls, so the below-target values are minus two, minus four, and minus one, while the three positive months contribute zero. Squaring the shortfalls gives four, sixteen, and one, which sum to twenty-one, and dividing by the total of six periods — including the winning months as zeros, per the standard convention — gives three point five. The square root of three point five is about one point eight seven percent, so the monthly Downside Deviation is roughly 1.87 percent. If this strategy earned an average monthly excess return of one percent over the target, its Sortino ratio would be one divided by 1.87, about 0.53. Compared against a rival strategy with the same average return but a Downside Deviation of only one percent, and thus a Sortino of one, the rival is clearly delivering its returns with less harmful volatility — the exact judgement Downside Deviation is designed to make.

Common mistakes

  • Ignoring how sensitive the result is to the chosen target (minimum acceptable return).
  • Computing it from too few loss observations, so the figure is unstable.
  • Comparing two assets' downside deviation measured against different targets.
  • Reading it as a worst-case measure — it says nothing about the tail; pair it with drawdown.
  • Failing to annualize consistently when comparing across return frequencies.