Portfolio Variance: How Weights, Volatility, and Covariance Combine

What is Portfolio Variance?

Portfolio variance measures the dispersion of a portfolio's returns around its average return.

It is the square of portfolio standard deviation:

1Portfolio Variance = (Portfolio Standard Deviation)²

The important part is not the squaring. It is how portfolio variance is built.

Unlike expected portfolio return, which is a weighted average of component expected returns, portfolio risk depends on how the holdings move together. Asset weights and standalone variances matter, but so do Covariance and Correlation.

The two-asset formula

For assets A and B:

1σp²
2= wA² σA²
3+ wB² σB²
4+ 2 wA wB Cov(A,B)

An equivalent correlation form is:

1σp²
2= wA² σA²
3+ wB² σB²
4+ 2 wA wB σA σB ρAB

where:

• wA and wB are portfolio weights;
• σA and σB are asset standard deviations;
• Cov(A,B) is covariance; and
• ρAB is correlation.

The cross term is what makes portfolio risk different from a simple average of standalone risk.

A worked example

Suppose a portfolio is 50% Asset A and 50% Asset B.

Assume:

1Asset A volatility = 20%
2Asset B volatility = 10%
3Correlation = 0.20

Then:

1Portfolio Variance
2= (0.50² × 0.20²)
3+ (0.50² × 0.10²)
4+ (2 × 0.50 × 0.50 × 0.20 × 0.10 × 0.20)
5
6= 0.0100 + 0.0025 + 0.0020
7= 0.0145

Portfolio volatility is the square root:

1sqrt(0.0145) ≈ 12.0%

That is lower than the 15% weighted average of the two standalone volatilities.

The difference comes from imperfect correlation.

Portfolio variance is not a weighted average of standalone variances

This is a common mistake.

If an investor calculates:

150% × Variance A + 50% × Variance B

the result ignores how returns interact.

Portfolio variance is not a weighted average of standalone variances.

The weights are squared in the individual variance terms, and the covariance terms matter.

The same holdings can produce very different portfolio variance depending on their co-movement.

Correlation can dominate the result

Suppose the two assets in the example both have 15% volatility and equal weights.

If correlation is +1, the portfolio remains at 15% volatility.

If correlation is 0, portfolio volatility falls because the returns do not move together perfectly.

If correlation is negative, the reduction can be larger.

This is the quantitative foundation of Diversification.

Adding a volatile asset can even reduce total portfolio volatility when its relationship with existing holdings is sufficiently diversifying.

That can sound counterintuitive until the covariance terms are made explicit.

The matrix form scales to larger portfolios

For many assets, portfolio variance is commonly written as:

1Portfolio Variance = w' Σ w

where:

• w is the vector of portfolio weights; and
• Σ is the covariance matrix.

The covariance matrix contains individual variances on the diagonal and pairwise covariances elsewhere.

This compact formula powers much of mean-variance portfolio construction, including the Efficient Frontier and Minimum Variance Portfolio.

Historical variance is only one estimate

A portfolio-variance calculation can be mathematically exact given its inputs while the inputs themselves are uncertain.

Historical estimates depend on:

• return frequency;
• lookback window;
• price synchronization;
• changing volatility;
• changing correlations;
• corporate actions and distributions;
• currency treatment; and
• whether the current portfolio resembles the historical one.

A portfolio built today may contain weights that never existed during the historical sample.

The resulting variance is therefore a model-based combination of estimated relationships, not an observed future fact.

Forecast covariance matrices add model risk

Professional portfolio systems often forecast future covariance instead of using a raw sample matrix.

Methods can include:

• factor models;
• exponentially weighted estimates;
• shrinkage estimators;
• regime-dependent assumptions; and
• implied or forward-looking inputs.

These methods can improve stability, but none removes uncertainty.

A precise output such as 11.43% expected volatility should not be mistaken for precision about the future simply because the calculation contains several decimal places.

Portfolio variance can hide concentration

Two portfolios can have the same estimated variance while carrying very different economic risks.

One may be diversified across many independent exposures. Another may have a concentrated position hedged by an asset whose historical relationship is fragile.

If the hedge correlation changes, the second portfolio can behave very differently from the model.

Variance should therefore be paired with exposure analysis, position concentration, liquidity, leverage, and stress scenarios.

Portfolio variance is not drawdown or tail loss

Portfolio variance is not maximum drawdown or tail loss.

Variance treats return deviations around the mean symmetrically. Large positive returns increase variance just as large negative returns do.

It also does not preserve the order of returns, unlike Maximum Drawdown.

And it does not directly answer questions such as:

• what loss threshold corresponds to a 95% or 99% confidence level;
• how severe losses are beyond that threshold; or
• what the worst possible loss might be.

Value at Risk and Expected Shortfall address different tail questions, each with its own assumptions and limitations.

Leverage changes variance quickly

If a portfolio's positions are scaled proportionally by a leverage factor L, its variance scales approximately with L² under the same return relationships:

1Levered Variance ≈ L² × Unlevered Variance

and standard deviation scales roughly with |L|.

But real leveraged portfolios can face changing financing costs, margin requirements, dynamic rebalancing, and forced position reductions. Those effects may not appear in a static variance calculation.

Optimization can exploit tiny input differences

A mean-variance optimizer searches across possible weight combinations.

If the covariance matrix says one asset pair is slightly more diversifying than another, the optimizer can assign a large weight to exploit that difference.

When the estimate is noisy, the optimized portfolio can be unstable.

Useful checks include:

• perturbing correlations and volatilities;
• comparing several lookback windows;
• imposing sensible weight constraints;
• testing turnover and transaction costs; and
• comparing the optimized result with simpler allocations.

Robustness matters more than a visually smooth frontier.

How investors should use portfolio variance

A disciplined review asks:

1. Are the weights current and correctly normalized?
2. What variance and covariance estimates are being used?
3. What sample period and return frequency produced them?
4. Is the covariance matrix historical or forecast?
5. How sensitive is the result to higher correlations?
6. Are leverage and derivatives represented correctly?
7. What concentrations are hidden beneath the aggregate statistic?
8. What do drawdown and tail-risk measures show separately?
9. How stable is the estimate across reasonable alternative inputs?

Grizzly Bulls' Models can be evaluated with these portfolio-risk distinctions without turning this page into a live optimizer. The Macroeconomic Conditions Index provides separate market-regime context, but it does not determine the covariance matrix or portfolio weights used in a variance calculation.

Sources and further reading

• CFA Institute, 2026, Portfolio Risk and Return: Part I: https://www.cfainstitute.org/insights/professional-learning/refresher-readings/2026/portfolio-risk-return-part-1
• CFA Institute, 2026, Portfolio Mathematics: https://www.cfainstitute.org/insights/professional-learning/refresher-readings/2026/portfolio-mathematics
• CFA Institute, 2026, Using Multifactor Models: https://www.cfainstitute.org/insights/professional-learning/refresher-readings/2026/using-multifactor-models

Explore more topics in the Financial Research Encyclopedia.