Skewness

Skewness is built by taking each observation's deviation from the mean, cubing it, and averaging the result. Cubing preserves sign, so large negative deviations contribute large negative values while large positive deviations contribute large positive ones; in a symmetric distribution the two sides cancel and the sum collapses toward zero. Because cubed deviations carry the units of the variable raised to the third power, the figure is divided by the standard deviation cubed, producing a pure number comparable across assets and horizons.

For a sample, the conventional estimator multiplies the average cubed standardized deviation by n divided by the product of (n minus one) and (n minus two), a small-sample correction that fades as the observation count grows. The cubing operation makes the statistic highly sensitive to outliers: a single extreme month can dominate the estimate, which is why short samples produce unstable skewness readings.

Sign identifies the long tail, not where the bulk of observations lie. Under right skew the mass of outcomes clusters below the mean and a thin string of large positive results drags the average upward, so the mean exceeds the median. Left skew reverses this: frequent modest outcomes sit above a mean pulled down by rare severe losses. The distance between mean and median is therefore a fast diagnostic that requires no cubing at all, and a histogram or box plot usually confirms it visually.

Magnitude matters as well as direction. As a working convention, absolute skewness beyond roughly 0.5 is treated as economically meaningful for typical financial samples, while values near zero justify a symmetric approximation. Note that a distribution can be perfectly symmetric and still be dangerous through fat tails, which skewness cannot detect; that is the province of kurtosis.

Asymmetry is structural in finance rather than incidental. Equity prices modelled as lognormal are positively skewed because prices are bounded below at zero but unbounded above, while the returns on broad equity portfolios are typically negatively skewed as correlations converge in stress. Option positions manufacture skewness directly: a long call or long straddle buys a small, frequent premium loss in exchange for a rare large gain, giving positive skew, whereas writing options, covered-call overlays and carry trades collect steady premiums against occasional severe losses, giving negative skew.

Credit exposures share the same profile, since a bond's upside is capped at contractual cash flows while default truncates the downside sharply. Strategies that appear attractive on mean and volatility alone, including many arbitrage and yield-enhancement programmes, frequently owe their smooth track records to concealed negative skew.

A candidate should be able to move fluently between the three representations of skewness: the sign of the third standardized moment, the ordering of mean, median and mode, and the shape of a plotted distribution. Given summary statistics, state which tail is longer and which side the mean lies on relative to the median. Given a short data series, apply the standardized cubed-deviation formula with the small-sample factor and interpret the result against the significance convention.

The analytical payoff is critique. Mean-variance optimisation, the Sharpe ratio and value-at-risk figures derived from a normal assumption all treat upside and downside dispersion as equivalent and therefore understate loss potential for negatively skewed series. Recognising that investors generally pay a premium for positive skew, and demand compensation for bearing negative skew, connects the statistic to observed risk premia.