Transparency is a feature. This page explains what volatility, downside deviation, and semivariance measure and why we compute them — in plain terms. The precise formulas and numeric conventions are documented in our internal methodology; the summary here is deliberate, not an omission.
These three measures all describe how much a return series moves around — its risk — starting from a history of simple period returns expressed as decimal fractions.
Volatility is the standard deviation of returns — how widely each period's return scatters around its own average — rescaled from the native frequency of the data (daily, weekly, monthly) into a comparable annual figure. It answers "how bumpy is this?": a higher number means larger swings in either direction. We use the sample (unbiased) form of standard deviation, and we annualize with the standard square-root-of-time rule, so the caller declares whether the returns are daily, weekly, or monthly rather than us assuming it. Volatility is always zero or positive; a perfectly flat series has no dispersion, so zero is a valid answer, not an error. It needs at least two observations to have any spread to measure, and it is a historical (realized) measure, not a forecast.
Downside deviation is a one-sided volatility that counts only the periods that fell short — it measures the size of the returns below a chosen threshold (the minimum acceptable return, which defaults to zero, so by default only outright losses count). Upside moves do not add to it. Crucially, it averages those shortfalls over every period, not just the losing ones, so a series that rarely disappoints scores lower than one that disappoints often. It is reported per period (not annualized here), is always zero or positive, and equals zero when nothing fell below the threshold. It is the natural risk input when you care about losses specifically rather than volatility in both directions.
Semivariance is simply downside deviation before the final square root — the average squared shortfall below the threshold, using the same all-periods averaging. It carries exactly the same information as downside deviation but is reported in squared-return units rather than as a return fraction, which is convenient when it feeds directly into other calculations. Like the others it is always zero or positive and defaults to treating only losses as downside.
