Description
Summarise how bad the bad side of an outcome distribution is by computing variance and skewness only below a chosen benchmark, giving a downside deviation and downside-adjusted rankings.
Standard deviation treats a surprisingly fast task the same as a surprisingly slow one, although only one hurts users. Lower partial moments measure dispersion only on the unwanted side of a benchmark (zero, a target, a competitor or the current release): the downside variance (semivariance) averages squared shortfalls below it, its square root is the downside deviation, and the third lower moment captures how long the bad tail is. Vinod and Reagle combine spread and skewness into a single risk index, RSK = SD minus c times skewness, where c reflects how much the decision maker dislikes long bad tails, and replace the mean-over-SD ratio with mean over downside deviation. These sit alongside quantile measures such as value-at-risk and expected shortfall, which answer a different question: how bad is a stated small percentage of outcomes.
Attributes
- Discipline
- Statistics & Modelling
- Loss type
- Unmigrated
- Stage
- Evaluation, Post-launch
- Data
- Quantitative
- Effort
- Hours
Steps
- Choose the direction that counts as bad (longer times, lower scores) and a benchmark: zero, a service-level target, a competitor's median or the current release.
- For each option, keep only observations on the bad side of the benchmark and compute the mean squared shortfall (semivariance) and its square root (downside deviation).
- Compute the third lower partial moment or downside skewness to describe how far the bad tail stretches.
- Where one index is needed, combine as SD minus c times skewness, setting c from stakeholders' tolerance for long bad tails, or compute the mean divided by downside deviation.
- Compare options on downside measures as well as means and full SDs, and report a quantile measure (VaR or expected shortfall) alongside for the extreme tail.
- Bootstrap the measures (resampling in blocks or with a maximum entropy bootstrap for time series) to give intervals.
Sources
- Preparing for the Worst: Incorporating Downside Risk in Stock Market Investments (Vinod & Reagle, 2005)
Also known as
- Lower partial moments
- Semivariance
- Downside standard deviation
- Downside deviation
- Downside Sharpe ratio
- Downside risk measurement
Categories
- Analytics & Behaviour
- Statistics & Modelling