Sample variance (Bessel's correction)
The average squared distance from the sample mean, but divided by n − 1 instead of n. Dividing by n would underestimate the true variance on average.
01The formula
- the sample variance
- the i-th observation
- the sample mean
- the number of observations
- the true variance
02See it move
03Where it is used
- Risk managerEstimates the volatility of a position from recent returns.
- Quant researcherComputes the noise term in t-statistics and Sharpe ratios.
- Options deskMeasures realised volatility to compare with implied volatility.
04Why traders use it
Volatility is the core input to risk and option pricing. With short windows the choice between n and n − 1 changes the answer noticeably.
05Worked example
Over 400 simulated samples of 5 observations, the ÷(n − 1) estimates average 103.8% of the true variance, while the ÷n estimates average 83.1%. Theory says ÷n should give 80.0%.
06How it is built
- Split each deviation: .
- Squaring and summing, the cross term vanishes, so .
- Take expectations: .
- So dividing by gives an unbiased estimate; dividing by is too small by the factor .
07The common mistake
Mixing up the defaults. numpy's np.var divides by n unless you pass ddof=1, while pandas' .var() divides by n − 1, so the same data give two different answers.
08In Python and LaTeX
import numpy as np
rng = np.random.default_rng(7)
n, M = 5, 400
X = rng.standard_normal((M, n)) # true variance is 1
unbiased = X.var(axis=1, ddof=1)
biased = X.var(axis=1, ddof=0)
print('theory for /n:', (n - 1) / n)s^2 = \frac{1}{n - 1}\sum_{i=1}^{n}(x_i - \bar{x})^2, \qquad E\!\left[s^2\right] = \sigma^2, \qquad E\!\left[\tfrac{n-1}{n}s^2\right] = \tfrac{n-1}{n}\sigma^209Where it fits
Rather not code the maths yourself? Describe your trading idea and let AI build the bot.
Build a bot with AgenKitThe worked example and the model run the same code, so every number matches. Models simplify markets on purpose. This is education, not financial advice.