Downside Deviation: Measure Returns Below a Stated Target

Define the target and the calculation

Choose a target return T for each observation interval. For each return r, keep min(r − T, 0), square it, average the squared shortfalls using the stated denominator, then take the square root. Returns above the target contribute zero under this definition. They are not replaced by negative values or subtracted from losing observations.

PerformanceAnalytics documents full-series and subset denominator options. Its target must have the same periodicity as the returns. This page uses the full number of observations unless explicitly comparing the subset calculation.

Work through a small sample

Take hypothetical monthly returns of −4%, −1%, +2% and +3%, with a monthly target of 0%. The shortfalls are −4, −1, 0 and 0 percentage points. Their squared sum is 16 + 1 = 17. Dividing by all four observations gives 4.25; the square root is approximately 2.06155 percentage points per month.

If you divide by only the two below-target observations, the result is the square root of 8.5, approximately 2.91548. Neither difference is a rounding issue. The subset version conditions the denominator on shortfalls; the full-series version includes how frequently they occur. Label the convention before comparing tools or reports.

Changing the target changes the question

Now use a +1% monthly target for the same four returns. Shortfalls become −5, −2, 0 and 0 percentage points. The full-series result is the square root of (25 + 4) ÷ 4, approximately 2.69258. The data did not change; the benchmark did.

An annual target cannot simply be inserted as a monthly threshold. For example, a hypothetical 12% effective annual target corresponds to about 0.948879% per month under geometric conversion, not 12% per month. This arithmetic example explains unit alignment and does not propose an appropriate target for a trading account.

Read the limitations before using a ratio

Downside deviation is often used in a target-relative performance ratio. If every observed return meets the target, the denominator can be zero. Display that as a boundary condition rather than an infinitely reliable score. A short sample, missing valuation dates or smoothed marks can hide downside events.

The statistic does not preserve the sequence of losses, maximum drawdown or the size of unobserved tail events. Annualizing it by a square-root rule also introduces assumptions about time aggregation that should not be treated as universal. Keep the raw return series, target, period count and denominator setting in the report.

When comparing copied accounts, use aligned dates and cash-flow-adjusted returns in a common currency basis. Otherwise a different deposit schedule or valuation time may explain the observed shortfalls. This reference defines a descriptive calculation; it does not identify an acceptable risk level or establish future performance.