Is Your Portfolio Really Diversified? Calculate Correlation and Risk Contributions

Ten funds need not provide ten independent sources of return. If they hold the same large technology companies or depend on cheap funding and one currency, a long account statement can hide a single dominant exposure. The number of positions is a poor test of diversification.

A useful audit moves through three layers: economic exposures, covariance-aware risk accounting, and adverse scenarios. They answer different questions—where the money sits, what drives ordinary fluctuations, and what could happen when cash is needed. The assets below are hypothetical; no fixed stock–bond relationship is assumed.

What correlation measures—and what it does not

Correlation measures linear co-movement in returns, from −1 to +1. Positive values mean returns tend to deviate from their respective averages in the same direction; negative values indicate offsetting deviations. It is not the probability that both prices rise, and it does not establish causation. Zero correlation is not independence or protection against simultaneous tail losses.

Use comparable returns, not trending price levels. Adjust consistently for distributions and splits, translate into the account currency and align timestamps. A market that has already incorporated news should not be casually paired with a closing price recorded before that news. Non-synchronous trading can make same-day correlation appear artificially low.

Short windows respond faster but generate noisier estimates; long windows can blend regimes that no longer apply. Compare defensible rolling windows without selecting whichever looks most reassuring after the fact. The familiar square-root-of-time annualization also needs assumptions about serial covariance. Matching daily frequency alone does not settle those issues.

A 60/40 mix is not a weighted average of volatilities

Assume annual volatility of 20% for asset A and 10% for B, with capital weights of 60% and 40%. Use the same currency and return frequency, with no leverage, option nonlinearity or costs. Portfolio variance is wA²σA² + wB²σB² + 2wAwBσAσBρ. Take its square root for volatility.

In decimal units, the calculation becomes 0.0144 + 0.0016 + 0.0096ρ. Holding everything else fixed gives the following outcomes. These are conditional calculations, not forecasts.

Simply adding 60%×20% and 40%×10% produces 16%, the perfectly positive-correlated case here. At the opposite extreme, even correlation of −1 does not remove all volatility: A's weighted volatility is 12%, while B's is only 4%. The offset is not equal in size. A negative coefficient cannot substitute for position sizing.

Sixty percent of the money can generate ninety percent of the risk

At zero correlation, total variance is 0.016. A supplies 0.0144 and B 0.0016, or 90% and 10%. A capital-allocation pie chart therefore understates how strongly A drives this portfolio's modelled fluctuations.

More generally, an asset's component contribution to volatility is its weight multiplied by its covariance with portfolio returns, divided by portfolio volatility. Components sum to portfolio volatility; dividing each by that total gives percentage contributions. In the zero-correlation example, A contributes about 11.38 percentage points and B 1.26, adding to 12.65%. These are not loss probabilities, maximum-loss allocations or recommended trading weights.

At correlation −0.50, B's percentage contribution is approximately −7.14% and A's 107.14%. The negative component means B offsets some volatility under the current weights and covariance assumptions; B can still lose money. Removing a holding and renormalizing the others creates a different portfolio. Its risk cannot be found by mechanically subtracting a local component from the old total.

Equal risk is not minimum risk or maximum return

With the same 20% and 10% volatilities and zero correlation, one-third in A and two-thirds in B equalize weighted volatilities. Each supplies half the volatility risk and portfolio volatility is about 9.43%. A lower-volatility asset requires more capital to supply the same risk in this particular setup.

But the minimum-variance two-asset mix under these assumptions is 20% A and 80% B, with volatility of about 8.94%. Equal risk and minimum variance solve different problems. Neither establishes the best portfolio without expected returns, objectives, liabilities and constraints. For futures, margin cash is not the economic exposure weight; for options, a linear covariance model omits nonlinear effects.

Price moves change both capital weights and risk contributions. The arithmetic of portfolio rebalancing, allocation drift and trading costs explains how to restore a target mix. Returning to 60/40 does not necessarily restore the old risk, however: volatility and correlation may have changed. Refresh those assumptions before trading merely to repair yesterday's pie chart.

Stress volatility, dependence and liquidity together

Raising correlation while leaving calm-market volatilities unchanged can understate an adverse scenario. If A's volatility rises to 30%, B's to 15%, and correlation to 0.80, the unchanged 60/40 mix has modelled volatility of about 23.08%, versus 12.65% in the zero-correlation baseline. This tests a joint assumption; it does not claim all crises produce the same correlation.

Standard deviation is not a loss ceiling. In a separate one-period scenario, A falls 30% and B 10%. With starting weights of 60/40 and no interim trades or flows, the portfolio loses 22%. That scenario loss is not the same statistic as 23.08% annual volatility. If assets must be sold, add wider spreads, limited depth, margin demands and currency moves to the analysis.

Look through fund holdings to issuers, sectors, interest-rate exposure and currencies. Several funds tracking the same index do not make the underlying companies independent. A rarely traded asset can also show deceptively stable prices while being difficult to liquidate. Low recorded volatility alone does not make it a reliable crisis buffer.

A diversification audit that can be updated

1. Map actual economic exposures and underlying holdings. Combine overlapping issuers, sectors, currencies and common factors; show leverage separately.
2. Standardize return definitions, distributions, currency, timestamps and frequency. Document missing observations and non-synchronous trading.
3. Record current weights, volatilities and the full covariance matrix. Reconcile component contributions to the portfolio total.
4. Compare reasonable estimation windows and retain their dates. Recalculate when holdings or data change, not only when results look uncomfortable.
5. Test joint selloffs, higher volatility, changing dependence and weaker liquidity. Keep standard deviation distinct from cash losses.
6. Choose adjustments only after objectives, constraints, costs and withdrawal needs are explicit. Identify which changed assumption invalidates the old diversification claim.

Diversification is not the number of lines on a statement. It is the variety of economic responses available under the same shock. Correlation measures one part of that relationship, risk contributions identify the dominant exposures, and stress scenarios test whether the apparent protection survives when it matters.