Portfolio Volatility: How Correlation Drives Diversification

Direct Answer

Portfolio volatility is the standard deviation of a portfolio's combined returns, and it is almost always lower than a simple weighted average of the individual holdings' volatilities. The gap between the two numbers is driven entirely by correlation: the less two assets move together, the more their combined swings partially cancel out, which is the actual mathematics behind the idea that diversification is "the only free lunch in investing."

This page works through the two-asset volatility formula step by step, shows a fully worked numeric example proving the diversification effect, contrasts it against a high-correlation scenario where that effect nearly disappears, and covers how to annualize volatility correctly and where standard deviation falls short as a risk measure.

Key Takeaways

• Portfolio volatility is not a weighted average of individual asset volatilities except in the special case where correlation equals exactly 1.
• The two-asset formula is σp = sqrt(w1²σ1² + w2²σ2² + 2·w1·w2·ρ12·σ1·σ2).
• Lower correlation between holdings produces a larger reduction in portfolio volatility relative to the naive weighted average.
• Annualize volatility by multiplying by the square root of the number of periods in a year, not by the number of periods itself.
• Standard deviation penalizes upside and downside moves equally and assumes volatility is stable, which it usually is not.
• Correlations tend to rise toward 1 during market crashes, eroding the diversification benefit exactly when it matters most.

The Two-Asset Volatility Formula

Combine each asset's weighted variance with a correlation-scaled cross term, then take the square root.

For a portfolio holding two assets, portfolio volatility (the standard deviation of the portfolio's returns) is:

σp = sqrt(w1²σ1² + w2²σ2² + 2·w1·w2·ρ12·σ1·σ2)

Where w1 and w2 are the portfolio weights of each asset (they sum to 1, or 100%), σ1 and σ2 are each asset's individual volatility measured as a standard deviation of returns, and ρ12 is the correlation coefficient between the two assets' returns, a number between −1 (perfectly opposite) and 1 (perfectly aligned).

The first two terms under the square root are unremarkable: they are simply each asset's own variance (volatility squared), scaled down by how much of the portfolio it represents. If those were the only two terms, doubling either weight would double that asset's contribution to risk in a straightforward, additive way. The third term is where the interesting behavior lives. It scales directly with ρ12, so its size depends entirely on how the two assets move relative to each other, not on how volatile either one is in isolation.

Why the formula has a plus sign, not something more complicated

It helps to see the formula as a special case of a more general rule: variance of a sum equals the sum of variances plus twice the covariance between the parts. A two-asset portfolio's return is just a weighted sum of two asset returns, so its variance follows exactly that rule. Covariance between two assets is itself defined as ρ12·σ1·σ2 — correlation multiplied by both standard deviations — which is why that specific combination appears in the cross term. When correlation is positive, the cross term adds to portfolio variance; when correlation is negative, it subtracts from it, which is how two individually volatile assets can combine into a portfolio calmer than either one alone.

The edge cases worth knowing by heart

• ρ12 = 1 (perfectly correlated): The formula collapses algebraically to σp = w1σ1 + w2σ2 — the simple weighted average. This is the only correlation value at which the naive weighted-average shortcut is exactly correct.
• ρ12 = 0 (uncorrelated): The cross term disappears entirely, and portfolio volatility is simply sqrt(w1²σ1² + w2²σ2²), always less than the weighted average whenever both weights are positive.
• ρ12 = −1 (perfectly opposite): The formula collapses to σp = |w1σ1 − w2σ2|. If the two weighted volatilities happen to be equal, portfolio volatility falls all the way to zero — a textbook perfect hedge, though one that essentially never occurs exactly with real assets.

Common mistake: assuming that a portfolio split 60/40 between a 15%-volatility asset and a 55%-volatility asset must itself sit somewhere between those two figures. It can, and typically does, sit meaningfully below the lower of the two numbers once the third-term reduction from imperfect correlation is large enough — though not below zero, and the exact outcome depends on the correlation value.

Worked Example: The Diversification Effect at Low Correlation

A 60/40 stocks-and-crypto blend at 0.3 correlation lands well below the naive weighted average.

Consider a portfolio holding 60% in a stock index with 15% annualized volatility and 40% in a crypto asset with 55% annualized volatility, and assume the correlation between the two return streams is 0.3 — a modest, positive but far from lockstep relationship.

Step 1 — the naive weighted average (what many traders assume the answer is):

Naive average = (w1 × σ1) + (w2 × σ2)
Naive average = (0.60 × 15%) + (0.40 × 55%)
Naive average = 9% + 22% = 31%

If you simply blended the two volatility figures by weight, you would expect a portfolio volatility around 31%. This is the number a spreadsheet produces if someone mistakenly averages volatilities the same way they would average expected returns.

Step 2 — the actual formula, term by term:

w1²σ1² = 0.60² × 0.15² = 0.36 × 0.0225 = 0.0081
w2²σ2² = 0.40² × 0.55² = 0.16 × 0.3025 = 0.0484
2·w1·w2·ρ12·σ1·σ2 = 2 × 0.60 × 0.40 × 0.30 × 0.15 × 0.55 = 0.01188

Sum = 0.0081 + 0.0484 + 0.01188 = 0.06838
σp = sqrt(0.06838) ≈ 0.2615 = 26.1%

The key insight: actual portfolio volatility comes out to roughly 26.1%, nearly 5 percentage points below the 31% naive weighted average — a relative reduction of about 16%. Every dollar of exposure is still fully invested in the same two assets at the same weights; nothing about the position sizes changed. The entire reduction comes from the fact that stocks and crypto do not move in perfect unison, so on days when one drags the portfolio down, the other partially offsets it, and vice versa. This is the literal mathematics behind the claim that diversification is "the only free lunch in investing" — it is one of the only ways to reduce risk without giving up expected return, and it costs nothing beyond holding assets that do not all move for the same reason.

Contrast: What Happens When Correlation Is High

The same weights and volatilities produce almost no diversification benefit once correlation rises to 0.9.

Keep every input identical — the same 60/40 split, the same 15% and 55% individual volatilities — and change only the correlation, from 0.3 up to 0.9, representing two assets that tend to move together most of the time.

2·w1·w2·ρ12·σ1·σ2 = 2 × 0.60 × 0.40 × 0.90 × 0.15 × 0.55 = 0.03564

Sum = 0.0081 + 0.0484 + 0.03564 = 0.09214
σp = sqrt(0.09214) ≈ 0.3035 = 30.4%

At 0.9 correlation, portfolio volatility comes out to about 30.4% — barely six-tenths of a percentage point below the 31% naive average, a relative reduction of roughly 2% instead of 16%. The diversification benefit has almost entirely evaporated, even though the portfolio still technically holds two different asset classes at the same weights as before.

This is the point worth internalizing: correlation, not the number of assets held, is what drives the diversification benefit. A portfolio of ten highly correlated assets can carry nearly as much combined risk as a single position of the same size, while a portfolio of just two assets with genuinely low or negative correlation can meaningfully cut risk relative to holding either one alone. Counting positions is not the same as measuring diversification.

Side-by-side comparison

Common mistake: concluding that because a portfolio holds assets from "different categories" — stocks and crypto, for instance — it must be meaningfully diversified. Category labels are a weak proxy for correlation. Two assets from different categories can still carry a high correlation, particularly during periods when both are driven by the same macro liquidity or risk-appetite conditions, in which case the diversification benefit shown in the first example simply does not materialize.

Beyond Two Assets: The Covariance Matrix in Plain English

With more than two holdings, portfolio volatility generalizes to σp = sqrt(w′Σw) — every pair of assets contributes a term.

The two-asset formula is a special case of a more general result that applies to a portfolio of any number of assets: σp = sqrt(w′Σw), where w is the list of portfolio weights and Σ (capital sigma) is the covariance matrix. This notation looks like it requires linear algebra to use, but the underlying idea does not.

Picture a grid with one row and one column for every asset in the portfolio. Each cell in that grid holds the covariance between the asset in its row and the asset in its column — a number that combines both assets' volatilities with their correlation, exactly the ρ·σ1·σ2 term from the two-asset formula. The cells running down the diagonal, where a row and column refer to the same asset, hold that asset's own variance (its volatility squared), since an asset is perfectly correlated with itself. Every other cell is an off-diagonal covariance between two different assets.

Portfolio variance is calculated by weighting every single cell in that grid by the weights of the two assets it connects, and adding all of the weighted cells together. The diagonal cells reproduce the "own variance" terms from the two-asset formula. The off-diagonal cells reproduce the cross terms — and because the grid is symmetric (the covariance between asset A and asset B is the same as between B and A), each pair of assets effectively contributes its cross term twice, which is exactly why the two-asset formula has that factor of 2 in front of its single cross term.

The practical consequence of this structure is that the number of pairwise relationships grows much faster than the number of assets. A three-asset portfolio has three pairs of covariance terms; a ten-asset portfolio has forty-five. As a portfolio grows, its risk profile is determined less by any single asset's own volatility and increasingly by the average correlation between all of the pairs — which is precisely why professional diversification analysis focuses on correlation structure across the whole book, not on trimming the single most volatile individual holding.

What this means without doing the matrix algebra

• Every asset contributes its own variance to the total, scaled by how large a position it is.
• Every pair of assets contributes an additional term based on how strongly and in what direction they move together.
• A large, well-diversified portfolio's volatility is driven mostly by the average correlation across all of its holdings, not by any individual asset's volatility alone.
• Adding an asset that is highly correlated with the existing average barely changes this picture; adding one with low or negative correlation to the existing average can meaningfully change it.

Annualizing Volatility From Daily or Monthly Data

Multiply by the square root of the number of periods in a year — never by the number of periods itself.

Volatility is almost always calculated first from a shorter data interval — daily or monthly returns — and then scaled up to an annual figure for comparison against other assets or benchmarks. The scaling rule is:

Annual volatility = Daily volatility × sqrt(252)
Annual volatility = Monthly volatility × sqrt(12)

252 is the approximate number of trading days in a calendar year; 12 is the number of months. The square root, rather than the raw multiplier, is the detail most often gotten wrong.

Why square root and not a straight multiple

Under the standard assumption that returns from one period to the next are independent and drawn from the same distribution, variance is additive across time: the variance of a full year's return is the sum of the variances of each individual day's return, so annual variance equals daily variance multiplied by 252. Standard deviation, however, is the square root of variance, not variance itself. Taking the square root of both sides of "annual variance = 252 × daily variance" gives "annual standard deviation = sqrt(252) × daily standard deviation" — the square root has to be applied to the scaling factor as well as to the variance. Multiplying volatility directly by 252 instead of by sqrt(252) — roughly 15.87 rather than 252 — overstates annualized volatility by more than a factor of 15.

The same logic applies at any interval: weekly volatility annualizes by sqrt(52), quarterly by sqrt(4), and so on. The general form is that standard deviation scales with the square root of elapsed time, sometimes called the "square root of time" rule, while variance scales linearly with elapsed time.

Where the assumption breaks down

The square-root-of-time rule depends on returns being uncorrelated from one period to the next and on volatility itself being constant over the measurement window. Neither assumption holds exactly in real markets. Momentum and mean-reversion effects introduce period-to-period autocorrelation that the simple scaling rule ignores, and — as covered next — volatility itself is not constant, so a daily figure measured during a calm stretch will understate the annualized risk that shows up once a turbulent stretch begins.

Common mistake: annualizing a volatility figure calculated from a short, unusually calm or unusually turbulent window and treating the result as a stable, forward-looking estimate, when the underlying daily or monthly figure it was scaled from may not be representative of a full market cycle.

Limitations of Standard Deviation as a Risk Measure

Standard deviation treats big gains and big losses identically, and assumes volatility is stable when it usually is not.

Standard deviation is calculated from squared deviations from the mean, and squaring a number erases its sign. A return that surprises 10% to the upside and one that surprises 10% to the downside contribute exactly the same amount to the volatility calculation. This means an asset that only ever moves in large positive jumps — rare but not impossible for a fast-growing crypto token, for example — can register a high standard deviation and appear statistically "risky," even though every one of its large moves has been a gain rather than a loss. Most traders and investors intuitively think of risk as the chance of losing money, not the chance of an outsized move in either direction, and standard deviation does not distinguish between the two.

This gap between the statistical definition and the intuitive one is exactly what downside-focused measures like the Sortino ratio and downside deviation are built to address: they calculate a version of "volatility" using only returns that fall below a minimum acceptable return or target, ignoring upside surprises entirely. A full treatment of downside deviation is a separate topic, but it is worth knowing that standard deviation is not the only lens available, and that a genuinely asymmetric return profile can look misleadingly risky — or misleadingly safe — through the standard-deviation lens alone.

Volatility clustering and regime shifts

A second limitation is more subtle: the formulas on this page treat σ1, σ2, and ρ12 as fixed numbers, but in real markets none of them are constant. Volatility tends to cluster — a large move is statistically more likely to be followed by another large move than by a return to calm conditions — and markets shift between distinct volatility regimes, from extended calm periods to sudden turbulent stretches, sometimes with little warning. A standard deviation calculated over a trailing window is a backward-looking average across whatever regime that window happened to capture; it is not a forecast of the regime that comes next. A portfolio volatility figure calculated during a calm quarter can understate the risk that materializes the moment conditions shift, and the correlation input carries the same instability, as covered in the next section.

Common Mistakes When Estimating Volatility and Correlation

Practical checklist

• Avoid estimating volatility or correlation from a short, unrepresentative lookback window that happens to capture only calm or only turbulent conditions.
• Recalculate correlation periodically rather than treating a single historical estimate as permanent.
• Do not assume category diversity (different asset classes, sectors, or tickers) is a substitute for measured low correlation.
• Remember that correlations tend to rise toward 1 during broad market selloffs, precisely when diversification is needed most.
• Distinguish volatility (a measure of dispersion in both directions) from downside risk (a measure of loss potential specifically) when the return distribution is asymmetric.
• Re-annualize using the square root of the correct number of periods for the data interval actually used, not a generic multiplier copied from a different context.

The lookback-window problem and the correlation-during-crashes problem are related and worth treating as a pair. A correlation estimate built from twelve months of unusually calm markets can show two assets as nearly independent, supporting a diversification decision that looks well-justified on paper. The same two assets, observed instead during a broad selloff, can behave as though correlation jumped to 0.8 or higher — common macro shocks such as a liquidity crunch or a market-wide flight to cash tend to move many assets in the same direction at once, and forced selling by leveraged holders often liquidates whatever is easiest to sell rather than whatever caused the original loss. The diversification benefit calculated from calm-market data is real during calm markets and can shrink sharply during the exact conditions a diversified portfolio is meant to protect against. Treating a single historical correlation figure as a permanent property of two assets, rather than as a conditional estimate that can shift with market regime, is one of the more consequential mistakes in applying this math.

Common mistake: using a lookback window short enough to have missed the last major drawdown, then relying on the resulting low-correlation estimate to size a portfolio for the next one.

Misconceptions Versus Reality

Risks, Limitations, and Exceptions

• Standard deviation treats upside and downside deviations identically, which can misrepresent risk for assets with asymmetric return distributions.
• Volatility and correlation estimates depend on the lookback window used and are not fixed properties of an asset or asset pair.
• Correlations tend to rise toward 1 during broad market stress, reducing the diversification benefit precisely when it is most needed.
• Volatility clusters and shifts between regimes rather than remaining constant, so a trailing calculation can lag current conditions.
• The square-root-of-time annualizing rule assumes returns are uncorrelated period to period, an assumption that autocorrelated (trending or mean-reverting) return series can violate.
• The two-asset and n-asset formulas describe historical or assumed statistical relationships; they do not predict future returns or guarantee any specific level of future portfolio risk.
• A near-zero or negative correlation observed historically between two assets is not guaranteed to persist, particularly if the underlying economic relationship driving that correlation changes.

Practical Implementation Checklist

1. List each portfolio holding's weight and its own individual volatility, measured over a representative lookback window.
2. Estimate the correlation between every relevant pair of holdings, not just the largest positions.
3. Apply the two-asset formula, or the general covariance-matrix version for more than two holdings, rather than defaulting to a weighted average.
4. Compare the resulting portfolio volatility against the naive weighted average to see how much diversification benefit the current mix is actually providing.
5. Annualize using the square root of the correct period count if the underlying data is daily or monthly.
6. Re-estimate correlation periodically, and specifically check how correlated the holdings became during the most recent period of market stress.
7. Pair the volatility figure with a downside-focused measure if the portfolio's return distribution is meaningfully asymmetric.
8. Treat a low measured correlation as a current estimate rather than a permanent guarantee when sizing positions.

Frequently Asked Questions

Sources and Methodology

The formulas and framing on this page draw on the foundational mean-variance portfolio literature and standard professional investment curricula. Key sources include:

• Harry Markowitz, "Portfolio Selection," The Journal of Finance, Vol. 7, No. 1 (March 1952), pp. 77–91: The foundational paper introducing mean-variance portfolio theory, including the mathematics of combining individual asset volatilities and their correlations into a single portfolio-level risk figure.
• CFA Institute curriculum, "Portfolio Risk and Return" readings: Standard professional reference for the two-asset and multi-asset (covariance-matrix) portfolio variance formulas used throughout institutional investment analysis.
• Investopedia, "Modern Portfolio Theory (MPT)": An accessible general-audience explanation of the diversification principle and the mean-variance framework underlying the calculations on this page.

All worked numeric examples on this page use illustrative, hypothetical inputs chosen to demonstrate the mathematics; they are not derived from any specific fund, index, or real trading account, and are not a forecast of any asset's future volatility or correlation.

This content was reviewed by the Swoopr Markets Education Team in August 2026.

Conclusion

Portfolio volatility is not a weighted average of individual asset volatilities — it is a function of both individual volatilities and how those assets move together.

The gap between the naive weighted-average figure and the true portfolio volatility is the measurable size of the diversification benefit, and that gap is driven by correlation, not by the number of assets held. Use this page alongside the correlation and diversification-ratio guide for a deeper look at measuring that relationship directly, and alongside the value-at-risk guide to see how portfolio volatility feeds into a loss-probability estimate.

Related Reading

• Portfolio performance metrics hub — the parent guide this page belongs to.
• Correlation and the diversification ratio — a direct extension of the correlation concept covered on this page.
• Alpha and beta — how a portfolio's volatility relates to its sensitivity to a benchmark.
• Value at risk — a loss-probability estimate that uses portfolio volatility as a direct input.
• Portfolio risk: correlation, concentration, drawdowns, and portfolio heat — the broader portfolio risk-management framework this metric supports.