Variance
Variance is a statistical measure of the dispersion of a set of returns around their mean, calculated as the average of the squared deviations from the mean. In finance, variance is the fundamental building block of portfolio risk, underpinning the mean-variance optimization framework, value at risk calculations, and virtually all quantitative risk models.
Key takeaways
- Variance equals the average squared deviation from the mean: σ² = Σ(Rᵢ - R̄)² / (N-1) for a sample.
- Standard deviation — the square root of variance — is the more intuitive risk measure because it is expressed in the same units as returns.
- Portfolio variance depends on both individual asset variances and the covariances (or correlations) between all pairs of assets.
- Variance treats upside and downside deviations symmetrically; semi-variance and downside deviation address this limitation for non-normal return distributions.
- Variance minimization is the objective in Markowitz mean-variance optimization, which forms the theoretical foundation of modern portfolio theory.
Explanation
Variance is the second central moment of a probability distribution. For a discrete set of N return observations R₁, R₂, ..., Rₙ with mean R̄, variance is computed as the sum of squared deviations from the mean divided by N (for population variance) or N-1 (for sample variance, applying Bessel's correction to produce an unbiased estimator). The result is always non-negative, and its square root — the standard deviation — is the more commonly cited risk measure because it shares the same units as the underlying returns (e.g., percent per year).
In portfolio theory, variance takes on a structural role far beyond describing a single return series. Harry Markowitz's 1952 mean-variance framework demonstrated that the variance of a portfolio is not simply a weighted average of constituent variances — it also incorporates the pairwise covariances between all assets. For a two-asset portfolio, variance = w₁²σ₁² + w₂²σ₂² + 2w₁w₂σ₁₂, where σ₁₂ is the covariance between assets 1 and 2. This equation embeds the fundamental principle of diversification: when assets are less than perfectly correlated (σ₁₂ < σ₁σ₂), portfolio variance is lower than the weighted average of individual variances. For large portfolios with N assets, variance is dominated by the N(N-1)/2 covariance terms, which outnumber the N variance terms as N grows — emphasizing that diversification primarily reduces covariance-driven risk.
Variance is the central input to Value at Risk (VaR) and volatility models. Under the assumption of normally distributed returns, portfolio VaR at confidence level α is simply a function of portfolio variance (equivalently, standard deviation): VaR = μ - z_α × σ, where z_α is the appropriate normal quantile. However, empirical return distributions exhibit excess kurtosis (fat tails) and negative skewness, meaning that variance alone understates extreme loss probability. This limitation motivates the use of conditional VaR (CVaR), extreme value theory, and stress testing as complements to variance-based risk measures.
In practice, variance is estimated from historical returns (historical variance), from options-implied volatility (which provides a market consensus forward-looking variance estimate), or from factor-based risk models (which decompose portfolio variance into factor and idiosyncratic components). Factor-based variance estimation is particularly valuable for large portfolios because it dramatically reduces the number of parameters to estimate — a portfolio of 500 stocks requires estimation of 500×499/2 ≈ 125,000 covariances directly, versus a much smaller set of factor exposures and factor covariances in a factor model.
Formula
σ² = Σ(Rᵢ - R̄)² / (N-1) [sample]; Portfolio variance = wᵀΣw, where w is the weight vector and Σ is the covariance matrix
Example
A risk analyst is evaluating a long/short equity fund with five years of monthly returns. The monthly returns have a mean (R̄) of 0.8% and the squared deviations from this mean sum to 0.0432 (in decimal form) across 60 months. Sample variance = 0.0432 / 59 = 0.000732, or 0.0732% per month in decimal units. Annualizing: monthly variance × 12 = 0.000732 × 12 = 0.008789. The annualized standard deviation = √0.008789 ≈ 9.37%. By contrast, the S&P 500 had an annualized standard deviation of 14.2% over the same period. The fund's lower variance reflects its long/short structure, which reduces systematic market exposure. The analyst notes the fund's correlation with the S&P 500 is 0.45, meaning meaningful diversification benefit exists when combining the fund with a long-only equity portfolio.
Related terms
Correlation Covariance Cross Hedge Diversification Equity Factor Model Fat Tails Forced Liquidation Implied Volatility Kurtosis Long Hedge Mean Variance Optimization