What Is Downside Risk?
Michael Whitman
·
Trading since 2017, full-time since 2020
·
YouTube creator, 137K subscribers
·
Published 2026-09-14
·
Updated
Downside risk measures variation only in returns that fall below a chosen threshold, ignoring movement above it entirely. It corrects the standard assumption that upside and downside variation are equally undesirable, at the cost of estimating the figure from roughly half the observations available.
Risk, as most people use the word, means losing money. Standard deviation does not measure that — it measures variation in both directions, and downside risk is the correction.
How it works
Choose a threshold — zero, a cash rate, or a required return — and measure deviation only for observations falling below it.
Returns above the threshold contribute nothing. A period that gained 12% adds zero to the measure, which matches how it was experienced.
Standard deviation cannot do this. It is symmetric by construction and treats a 12% gain and a 12% loss as identical contributions to risk.
Where the correction actually matters
For symmetric returns it changes almost nothing. If gains and losses are distributed alike, downside deviation is roughly the standard deviation divided by a constant, and the ranking is unchanged.
For skewed returns it matters enormously. Trend following produces occasional large gains; option selling produces occasional large losses, and standard deviation cannot tell them apart.
So it is a correction that is unnecessary in the easy cases and essential in exactly the cases where conventional measures mislead most.
A worked example
Take the S&P 500 since 1950. It closed below a prior record on 93% of trading days, its worst fall was 56.8%, from 2007 to 2009, and its longest stretch below a record ran 7.5 years, from 1973 to 1980.
A standard deviation measured on that history counts every day. A downside measure counts only those below the chosen threshold — a substantially smaller set.
Set the threshold at zero and the measure asks about losing periods. Set it at a required return of 6% and every period below 6% counts, including modestly positive ones.
Those two produce meaningfully different numbers from identical data, which is why a downside risk figure without its threshold stated is not interpretable.
The related measures
Semi-deviation is the classical version, using the mean as the threshold, and is the oldest form of this idea.
Value at risk states a loss level that will not be exceeded on a given percentage of periods, which is a threshold rather than a deviation and says nothing about how bad the exceedances are.
Expected shortfall averages the losses beyond that point, answering the question value at risk skips and requiring more data to estimate.
And maximum drawdown is the simplest of all — the worst peak-to-trough decline — which requires no distributional assumption and describes exactly one historical path.
The original data
The S&P 500 record shows how much material the downside view has to work with. From 1950 to 2026, 46.9% of trading days closed lower, and the index went through 68 separate stretches closing 5% or more below a record, about 0.89 a year.
The 93%-below-record figure is the one worth sitting with. An asset that rises over time spends almost all of its life below a previous high, which means downside observations are the normal case rather than the exception.
And acting on the measure pays a spread and commission on every round trip. A risk measure that prompts frequent adjustment pays that repeatedly, which is a cost no risk calculation includes.
Why standard deviation survives anyway
It uses every observation. Halving a sample doubles the estimation noise, and for short records that cost can exceed the benefit of measuring the right thing.
It has clean mathematics. Portfolio variance combines from the variances and covariances of the holdings; downside measures do not aggregate as neatly, which makes optimization harder.
And it is the convention. Comparisons across managers, funds and studies are made on it, so a different measure is not comparable to the published universe.
Which leaves an honest conclusion rather than a recommendation. Downside measures describe risk better and estimate it worse, and the right choice depends on the length of the record and how asymmetric the returns are.
Why the threshold choice matters so much
Zero is the intuitive default. It asks how much of the variation came from losing periods, which is the plain-language meaning of risk.
A cash rate is more defensible. Falling short of what a deposit would have paid is a real shortfall, and treating a 1% gain as costless when cash paid 4% understates the problem.
A required return targets the actual objective. A pension scheme needing 6% treats anything below 6% as a shortfall, because for them it is one.
Each produces a different number from identical data, and the ranking of two investments can reverse between them — which is why the threshold belongs beside the figure whenever one is quoted.
How this connects
The Sharpe ratio is the measure it most often replaces. It divides return above cash by standard deviation, so a strategy with lumpy gains is penalized for volatility investors would welcome.
The Calmar ratio takes the path-based route instead, dividing return by the worst drawdown, and the ulcer index weights both how deep and how long the declines ran.
Selling a covered call reshapes the payoff deliberately. It caps the upside and leaves the downside intact, which is the kind of asymmetry symmetric measures blur.
Option markets price this asymmetry directly. The volatility smile shows protective puts typically costing more than equivalent calls, which is the market’s estimate of downside risk rather than a historical one.
For a trader, risk of ruin is the extreme version of the same question, and expectancy is what has to stay positive while the downside is being survived.
Much of it cannot be diversified away. Systematic risk hits every holding at once, which is why a hedge rather than more positions is the usual answer to it.
A backtest will understate it whenever the sample lacks a severe decline.
And personal risk tolerance is really a statement about downside, since nobody abandons a plan because it went up too much.
When it fails
The characteristic failure is a strategy whose downside has not yet occurred. An option-selling approach runs three years with no losing month, the downside deviation is near zero, and every downside-based measure reports exceptional risk-adjusted performance. The measure is arithmetically correct and it is describing an empty set — there are no adverse observations in the sample, which is a statement about the period rather than about the strategy. The figure is at its most flattering immediately before the event it is structurally unable to see.
A second failure is comparing figures with different thresholds, which are not the same measurement.
A third is using it on a short record, where a handful of observations determine the result.
A fourth is treating a low figure as low risk rather than as few observed losses.
The question: No losing months. Is the downside zero?
And a fifth is expecting it to capture the tail. It describes the losses in the sample, and the ones that matter most are the ones not in it.
Related
Sortino ratio covers the best-known measure built on this. Tail risk covers the extreme losses beyond what this typically captures. And volatility risk covers the symmetric measure this corrects.
Standard deviation says a portfolio that jumped 15% was risky. Nobody who held it felt that way. Downside risk measures what people actually mean by the word, which is an improvement, and it is measured from fewer observations, which is the price.
— Michael Whitman
This page is educational, not financial advice. Test every idea on your own charts before risking money.