Semivariance measures the average squared deviation of observations that fall below a selected threshold. In finance, it is used to measure unfavorable return dispersion without penalizing returns above the threshold.
The threshold may be the sample mean, zero, a benchmark, or a minimum acceptable return. Because definitions and denominators vary, “semivariance” is incomplete unless the calculation convention is stated.
Key Takeaways
- Variance uses squared deviations on both sides of the mean; semivariance uses only deviations below a threshold.
- Below-mean semivariance and target semivariance answer different questions.
- Dividing by all observations versus only below-target observations changes the result.
- Downside deviation is generally the square root of target semivariance.
- Semivariance is historical unless it is calculated from a forecast or simulated distribution.
- A lower semivariance does not establish higher return, greater liquidity, or lower tail loss.
For returns \(R_i\), target \(T\), and \(n\) observations, a common target-semivariance definition is:
$$
\operatorname{Semivariance}(T)
=
\frac{1}{n}
\sum_{i=1}^{n}
\left[\min(R_i-T,0)\right]^2
$$
The equivalent shortfall form is:
$$
\operatorname{Semivariance}(T)
=
\frac{1}{n}
\sum_{i=1}^{n}
\left[\max(T-R_i,0)\right]^2
$$
Downside deviation is:
$$
\operatorname{DownsideDeviation}(T)
=
\sqrt{\operatorname{Semivariance}(T)}
$$
If \(T\) is the sample mean, the result is below-mean semivariance. If \(T\) is a required return, it is target semivariance or a second-order lower partial moment.
Worked Example
Assume monthly returns of:
4%, 2%, -3%, 1%, and -5%
Using a target of 0%, only -3% and -5% create shortfalls. With all five observations in the denominator:
$$
\operatorname{Semivariance}(0)
=
\frac{(-3\%)^2+(-5\%)^2}{5}
=
6.8\ \text{percentage points squared}
$$
The corresponding downside deviation is approximately 2.61%.
If the denominator includes only the two below-target observations:
$$
\frac{(-3\%)^2+(-5\%)^2}{2}
=
17
$$
The square root is approximately 4.12%. The example shows why the denominator must be disclosed.
Below-Mean vs. Target Semivariance
| Version | Threshold | Main interpretation |
|---|
| Below-mean semivariance | Sample or expected mean | Dispersion of outcomes below average |
| Zero-target semivariance | 0% return | Dispersion of nominal losses |
| Required-return semivariance | Hurdle or minimum acceptable return | Dispersion of failures to meet the objective |
| Benchmark semivariance | Benchmark return | Dispersion of underperformance |
| Liability-relative semivariance | Required funding return | Dispersion of funding shortfalls |
Using the mean as the threshold makes the threshold move with the sample. A strategy with a low average return can show limited below-mean semivariance even if it repeatedly misses an investor’s required return.
Semivariance vs. Variance
Standard Deviation is the square root of variance and measures total dispersion around the mean. Semivariance focuses on one side.
Suppose two return series have similar variance:
- Series A alternates between moderate gains and moderate losses.
- Series B has many large gains and only small losses.
Their total volatility may be similar, while Series B has lower downside semivariance under a zero target. Whether that makes Series B preferable depends on valuation, liquidity, tail exposure, and the reliability of the history.
Relation to the Sortino Ratio
The Sortino ratio commonly divides return above a target by downside deviation:
$$
\operatorname{Sortino}
=
\frac{R_p-T}{\operatorname{DownsideDeviation}(T)}
$$
Different downside-deviation conventions therefore produce different Sortino ratios. Before comparing funds or strategies, align:
- return frequency
- annualization
- target return
- denominator convention
- fees and cash flows
- lookback period
How to Calculate Semivariance
- Select the return series and observation frequency.
- Decide whether returns are arithmetic, logarithmic, gross, or net.
- Define the threshold.
- Replace above-threshold deviations with zero.
- Square the remaining deviations.
- Choose and disclose the denominator.
- Average the squared shortfalls.
- Take the square root only if downside deviation is needed.
For portfolio analysis, use consistent total-return data and address distributions, fees, leverage, currency, and external cash flows.
Annualization
Downside deviation is sometimes annualized by multiplying a periodic estimate by the square root of periods per year. For monthly data:
$$
\operatorname{AnnualizedDownsideDeviation}
\approx
\operatorname{MonthlyDownsideDeviation}\sqrt{12}
$$
This shortcut assumes stable and sufficiently independent returns and a consistent target. Volatility clustering, serial correlation, illiquidity, and path-dependent strategies can make the scaling misleading.
Semivariance itself is in squared units. Multiplying the square-root measure and multiplying semivariance require different scaling.
What Semivariance Captures
Semivariance can help when:
- losses matter more than gains
- a portfolio has asymmetric returns
- the objective includes a minimum acceptable return
- performance is compared with a liability or benchmark
- upside volatility should not increase the stated risk measure
It does not directly capture:
- the worst loss
- recovery time
- peak-to-trough drawdown
- loss probability beyond a selected quantile
- liquidity or market impact
- leverage and margin calls
Use Downside Risk for the broader measurement framework.
Risks and Limitations
- Threshold dependence: rankings can change when the target changes.
- Denominator ambiguity: published values may use incompatible conventions.
- Sample dependence: a calm or short history can understate downside.
- Squared weighting: extreme observations dominate the result.
- No path information: the same semivariance can arise from different drawdown sequences.
- Survivorship bias: failed products may be absent from the dataset.
- Annualization: square-root scaling may not fit the return process.
- No scenario outside history: unprecedented events remain unmeasured.
Common Mistakes
- Calling semivariance the variance of negative returns without defining the threshold.
- Switching between the mean and zero without disclosure.
- Dividing by below-target observations in one series and all observations in another.
- Taking the square root and still labeling the result semivariance.
- Comparing annualized and periodic figures.
- Assuming lower semivariance means lower maximum loss.
- Ignoring return smoothing and stale asset prices.
- Using semivariance as a substitute for stress testing.
Authoritative Context
These sources provide statistical and financial-risk context. They do not prescribe one semivariance target or denominator for all investment reports.
- Downside Risk: The broader framework for evaluating outcomes below a defined threshold.
- Standard Deviation: Measures total dispersion around the mean rather than only below-target variation.
- Sortino Ratio: Uses downside deviation to relate excess return to unfavorable variability.
- Expected Shortfall: Estimates average loss beyond a selected quantile instead of averaging all squared shortfalls.
- Ulcer Index: Measures the depth and persistence of drawdowns from prior peaks.
FAQs
What is semivariance in simple terms?
It is the average squared shortfall below a chosen threshold, such as zero return, the mean, or a required return.
What is the difference between semivariance and downside deviation?
Downside deviation is generally the square root of semivariance calculated against the same target and convention.
Can two providers report different semivariance for the same returns?
Yes. They may use different thresholds, denominators, return definitions, frequencies, or annualization methods.
Educational Use
This article provides general financial education. It is not personalized investment, portfolio-construction, statistical, trading, tax, legal, or risk-management advice.