Correlation and why it fails when needed

Diversification depends on assets not moving together, and the extent to which they do not is measured by correlation — which is unstable in a specific and inconvenient way.

What correlation measures

The basics.

A statistic between minus one and plus one describing how two assets have moved relative to each other over a period.

Zero indicates no linear relationship; positive indicates they move together; negative indicates they move oppositely.

Combining assets with correlations below one reduces portfolio volatility relative to the weighted average of the individual volatilities, which is the mathematical basis of diversification.

The instability problem

Which is the point.

Correlations are not constant: they are estimated over a period and change.

Critically, correlations between risky assets tend to rise during periods of market stress.

Which means diversification provides least benefit at exactly the moment it is most wanted.

This has been observed repeatedly across crises, and it is a structural feature rather than an anomaly — during a liquidity event, investors sell whatever they can, which pushes prices down together regardless of fundamentals.

The bond and equity relationship

The most consequential example.

High-quality government bonds have historically tended to rise or hold value during equity declines, which is the basis of the balanced portfolio.

The relationship depends on the driver of the decline: a growth shock produces falling equities and falling yields, which means rising bond prices.

An inflation shock produces falling equities and rising yields, which means falling bond prices — both fall together.

Recent periods demonstrated this, with balanced portfolios experiencing simultaneous losses in both components.

Which is not a failure of the concept — it is a demonstration that the relationship is conditional on the environment.

Where diversification still works

For balance.

Across individual companies, where diversification removes specific risk reliably.

Across sectors and geographies, where correlations rise during stress and remain below one.

Between cash and everything else, since cash is the only asset that does not fall in nominal terms.

Over long periods, where the averaging effect of holding many assets persists even if short-term correlations spike.

Which means diversification is genuinely valuable and is less protective in the short term during crises than the statistics suggest.

The measurement problems

Which flatter some assets.

Illiquid assets valued by appraisal rather than by market transactions show smoothed returns, which artificially lowers measured volatility and correlation.

Which makes direct property, private equity and some private credit look better diversifying than they are.

Correlations estimated over short periods are noisy.

And correlations estimated over periods that did not include a crisis do not tell you what happens in one.

Which means correlation figures in marketing material should be treated with substantial scepticism.

Tail dependence

A more useful concept.

What matters is not average correlation but behaviour in the extreme cases.

Two assets can have low average correlation and high tail dependence, meaning they fall together in the worst scenarios while behaving independently the rest of the time.

Which is precisely the failure mode that undermines diversification when it matters.

Examining how a proposed holding behaved during previous crises is more informative than its correlation statistic.

What to do about it

Practical responses.

Hold cash, which is the only genuinely uncorrelated asset in nominal terms and which funds spending without forced selling.

Diversify across genuinely different sources of risk rather than across many variations of the same exposure.

Do not rely on historical correlation statistics for products marketed as uncorrelated.

Size positions so that a simultaneous decline across the portfolio is survivable.

And plan for the scenario in which everything falls together, since it has happened and will happen again.

The realistic expectation

Worth setting.

A diversified portfolio will fall during major market stress, sometimes substantially, and diversification reduces rather than eliminates the fall.

Anyone expecting a diversified portfolio to be protected during a crisis will be disappointed and may act on the disappointment.

Which is why understanding the limits in advance is a behavioural protection as well as an analytical one.

The alternative to correlation

A simpler framing.

Rather than optimising correlations, consider what each holding is for and what would have to happen for the whole portfolio to fail.

Which produces a more robust structure than statistical optimisation on historical data, since optimisation on unstable inputs produces confident and fragile portfolios.

And which is why simple broadly diversified portfolios have generally outperformed complex optimised ones out of sample.

General information only, not investment advice. Investments can fall in value and diversification does not eliminate risk. Consult a regulated financial adviser.