Sample Standard Deviation

Builds onArithmetic Mean Return — if this page feels steep, start there.

s = \sqrt{\frac{\sum_{i=1}^{n}(R_i - \bar{R})^2}{n-1}}

$s$: the sample standard deviation — the 'typical distance' of a return from the average
$R_i - \bar{R}$: each observation minus the mean: how far that period strayed from average
$(\cdots)^2$: squaring makes every deviation positive (misses count both ways) and punishes big misses extra
$n-1$: divide by one less than the count — a correction because the mean was estimated from the same data
$\sqrt{\cdots}$: the square root undoes the earlier squaring, returning the answer to ordinary return units

Why it must be true

Risk, in statistics, is distance from the average — squared so that ups and downs both count, then averaged, then square-rooted back into return units. That's all a standard deviation is: the typical size of a surprise.

Why divide by $n-1$ and not $n$ ? Because the sample mean was computed from the same data — the observations are automatically a little closer to it than to the true mean, understating dispersion. One degree of freedom was spent estimating $\bar{R}$ ; dividing by $n-1$ pays for it.

The derivation

Measure each observation's deviation from the sample mean, square (sign-blind, penalizes big misses), and average over the $n-1$ independent deviations — the last deviation is determined by the others, since deviations from the sample mean always sum to zero:

s^2 = \frac{\sum_{i=1}^{n}(R_i - \bar{R})^2}{n-1}

Take the square root to return to the original units:

s = \sqrt{s^2}

When to reach for it

Dispersion of a sample of returns — the standard risk number behind volatility, tracking error and risk limits.

Listen for

“sample of returns”“volatility / dispersion”“standard deviation of the last n years”

Back-of-the-envelope

Estimate it in your head first — then the calculator only confirms.

≈
Range/4 rule: for a handful of observations, s ≈ (max − min) ÷ 4 is a startlingly good first guess. Returns from −2% to 14% → s ≈ 4%.
≈
Bounds: s is positive and can't exceed the full range. Any choice bigger than max − min is noise.
≈
The variance is s² — in decimal terms a tiny number (0.0016 for s = 4%). If a candidate answer looks absurdly small, it's the variance distractor.

Traps in applying it

✗Dividing by n instead of n − 1 for a sample.
✗Reporting the variance (squared units) where the question wants σ.
✗Annualizing by multiplying σ by t instead of √t.

Limits & criticisms

Standard deviation is symmetric — it punishes a windfall exactly like a crash, which is why downside measures (semideviation, VaR) exist. It assumes the distribution is stable while real volatility clusters and regime-shifts, and it badly understates the risk of fat-tailed returns — the core post-2008 criticism of σ-based risk management.

Where it came from

Karl Pearson coined "standard deviation" in 1893; the $n-1$ correction is older, inherited from Friedrich Bessel's astronomy. The decisive moment for finance came in 1952, when Harry Markowitz declared that risk is the standard deviation of returns — turning a statistician's dispersion measure into the industry's definition of risk (and earning a Nobel Prize).

Today volatility is quoted like a price: fund factsheets, VaR models, options implied vols, and risk limits are all denominated in standard deviations.

Where it leads

Master this and the following come almost for free:

Coefficient of Variation (CV)Sharpe Ratio Correlation from Covariance Beta

Drill this formula Back to the library