Understanding Portfolio Variance: Key Concepts and Calculation Formula

Search Fintrev

Search our guides, reviews and glossary.

Portfolio variance is a statistical measure of how much a portfolio’s total return spreads around its average. It depends on three inputs: each holding’s weight, each holding’s own variance, and the covariance between every pair of holdings. Its square root is the portfolio standard deviation, the figure usually reported as portfolio volatility. Because covariances enter the calculation, a portfolio’s variance is generally lower than a simple weighted average of its holdings’ variances, and that gap is the arithmetic behind diversification.

How portfolio variance works

Variance for a single asset measures the average squared distance between its returns and its mean return. A portfolio adds a second layer: assets do not move independently, so the way they move together matters as much as how much each one moves alone.

Covariance captures that joint movement. It can also be written as correlation multiplied by the two standard deviations:

Cov(1,2) = ρ(1,2) × σ1 × σ2

Correlation (ρ) runs from −1 to +1. At +1 the two assets move in lockstep and offer no diversification. At 0 their movements are unrelated. At −1 they move in exact opposition, so a suitable mix can cancel out variance entirely. Modern Portfolio Theory (MPT), introduced by Harry Markowitz in 1952, rests on this point: a volatile asset can still lower total portfolio risk if it tends to rise when the other holdings fall. What matters is its contribution to portfolio risk, not its standalone risk.

Portfolio variance formula

Two-asset portfolio

For two assets, the portfolio variance formula is:

σp² = w1²σ1² + w2²σ2² + 2·w1·w2·Cov(1,2)

Where:

• w1, w2 = portfolio weights of each asset (summing to 1)
• σ1, σ2 = standard deviations of each asset’s returns
• Cov(1,2) = covariance of the two assets’ returns, equal to ρ(1,2)·σ1·σ2

The first two terms are each asset’s own risk, scaled by the square of its weight. The third term is the interaction effect. It is positive when the assets move together, negative when they move in opposite directions, and zero when they are uncorrelated.

Three-asset portfolio

Adding a third asset adds one variance term and two more covariance terms:

σp² = w1²σ1² + w2²σ2² + w3²σ3² + 2w1w2Cov(1,2) + 2w1w3Cov(1,3) + 2w2w3Cov(2,3)

Any number of assets

In general, a portfolio of n assets has n variance terms and n(n−1)/2 distinct covariance terms. Three assets give 6 terms in total, five assets give 15, and 30 assets give 30 variances plus 435 covariances. The growth is quadratic, which is why the calculation is usually written in matrix form:

σp² = wᵀ Σ w

Here w is the column of weights and Σ is the variance-covariance matrix, with each asset’s variance on the diagonal and pairwise covariances off it. In a spreadsheet this is a matrix multiplication, for example with MMULT and TRANSPOSE.

Example of portfolio variance

Hypothetical example for illustration only. A portfolio holds 60% in Asset A, with an annual standard deviation of 15%, and 40% in Asset B, with an annual standard deviation of 5%. The correlation between them is 0.2. All figures are in decimal form.

1. Covariance: 0.2 × 0.15 × 0.05 = 0.0015
2. Weighted variance of A: 0.6² × 0.15² = 0.36 × 0.0225 = 0.0081
3. Weighted variance of B: 0.4² × 0.05² = 0.16 × 0.0025 = 0.0004
4. Covariance term: 2 × 0.6 × 0.4 × 0.0015 = 0.00072
5. Portfolio variance: 0.0081 + 0.0004 + 0.00072 = 0.00922
6. Portfolio standard deviation: √0.00922 ≈ 0.096, or about 9.6%

The weighted average of the two standard deviations is 0.6 × 15% + 0.4 × 5% = 11%. The portfolio’s actual volatility is about 9.6% because the correlation is well below +1. Changing only the correlation shows its effect:

• At ρ = +1, the portfolio standard deviation equals the weighted average: 11%.
• At ρ = 0, the covariance term drops out: √0.0085 ≈ 9.2%.
• At ρ = −1, it falls to |0.09 − 0.02| = 7%.

Note the units. The variance of 0.00922 is 92.2 “percent squared,” not 0.922%. Standard deviation puts risk back into the same units as returns, which is why analysts rarely interpret variance directly.

How variance is used in building portfolios

Diversification and the efficient frontier

Under MPT, investors seek the highest expected return for a given level of risk. Plotting every possible mix of assets by expected return against standard deviation produces the efficient frontier, the set of portfolios offering the best return at their level of variance. Holdings with low or negative correlation, classically stocks and bonds, push the frontier up and to the left.

Minimum variance portfolio

The leftmost point on the frontier is the minimum variance portfolio, the mix with the lowest possible variance. For two assets, the weight in Asset 1 is:

w1 = (σ2² − Cov(1,2)) / (σ1² + σ2² − 2·Cov(1,2))

Using the example inputs: (0.0025 − 0.0015) / (0.0225 + 0.0025 − 0.003) = 0.001 / 0.022 ≈ 4.5% in Asset A and 95.5% in Asset B. That mix has a standard deviation of about 4.95%, slightly below Asset B’s 5% on its own: adding a small amount of the riskier asset lowered total risk. With 30 or more assets there is no simple closed form, so optimization software works from the full covariance matrix.

Equally weighted portfolios and the limit of diversification

For an equally weighted portfolio of n assets, the formula simplifies to:

σp² = (1/n) × average variance + (1 − 1/n) × average covariance

As n grows, the first term shrinks toward zero and portfolio variance approaches the average covariance. Asset-specific (unsystematic) risk can be diversified away; the shared, market-wide component, often called systematic variance, cannot. This is why a broad index fund or ETF still fluctuates with the market.

Contribution to portfolio variance

Each holding’s contribution to variance equals its weight multiplied by its covariance with the whole portfolio, and the contributions add up to total portfolio variance. A holding that is small by weight can account for a large share of risk if it is volatile and highly correlated with everything else. Risk budgeting and rebalancing often start from this breakdown.

Risk-adjusted performance

Portfolio standard deviation feeds directly into risk-adjusted measures. The Sharpe ratio, for example, divides a portfolio’s excess return by its standard deviation, so two portfolios with the same return can rank very differently once variance is considered.

Limitations

• Estimates, not facts. Variances and correlations come from historical returns and change with the time window and return frequency used. A large matrix holds hundreds of estimated covariances, each with estimation error.
• Unstable correlations. Relationships between assets shift over time and can rise sharply in a crisis, reducing diversification exactly when it is needed.
• Upside counts as risk. Variance penalizes large gains as heavily as large losses.
• Tail risk. Variance describes typical dispersion and does not fully capture rare, extreme losses, particularly when returns are skewed or fat-tailed.
• Time scaling assumptions. Converting monthly variance to annual by multiplying by 12 (and standard deviation by √12) assumes returns are independent from period to period.

Portfolio variance vs. related terms

• Standard deviation (volatility): the square root of variance, in the same units as returns.
• Covariance and correlation: the pairwise inputs to portfolio variance. Correlation is covariance standardized to a −1 to +1 scale.
• Weighted average variance: ignores the covariance terms and overstates risk whenever correlations are below +1.
• Beta: sensitivity to a benchmark, which captures systematic risk only. Portfolio variance measures total risk.
• Expected return: a weighted average of the holdings’ expected returns, with no interaction terms.

FAQs

The Bottom Line

Portfolio variance matters to anyone weighing how holdings combine, because correlations decide whether adding an asset raises or lowers total risk. A volatile asset can reduce portfolio risk if it tends to rise when others fall, while market-wide risk cannot be diversified away. The inputs are estimates from historical data, and correlations can rise sharply in a crisis, so the result works best as a guide to typical swings.