Correlations rise in the weeks you need them low

The points below about correlation between holdings are ordered by how much difference they make, not by how often they get repeated.

What correlation describes

Correlation measures the tendency of two holdings to move in the same direction, and it is always calculated over a chosen period. That choice is not neutral, because correlation between the same two assets varies enormously depending on what the period happened to contain. A number calculated across a decade of calm markets describes a relationship that may not survive contact with stress.

This is not a flaw in the statistic; it is a limit on what any backward-looking average can tell you. Portfolio construction that leans on stable correlations is leaning on the least stable input in the entire calculation.

Why correlations rise in a crisis

During severe declines, selling is frequently driven by a need for cash rather than by any view about the individual assets. Investors sell what they are able to sell rather than what they would most like to sell, spreading pressure across unrelated holdings. Leverage amplifies this, because forced liquidation is entirely indiscriminate about which assets it happens to touch.

On an ordinary week, the result is that things which normally diverge move downwards together for a period of weeks or months. The effect is well documented across past episodes, although its precise size has differed in every one of them.

What still tends to help

High-quality government bonds have historically behaved differently from equities during equity declines, though not in every single episode. The relationship is not a law and has broken down in periods where inflation rather than growth was the driving concern. Cash is the only holding that reliably does not fall in nominal terms, at the cost of lagging badly over long periods.

Diversification across many equities within one market helps far less in a crisis than diversification across genuinely different asset types. Anybody promising an asset that always rises when equities fall is describing a product rather than reporting a finding.

The diversification that is mostly cosmetic

Holding several equity funds tracking overlapping indices produces a portfolio with one dominant risk and several separate statements. Sector funds within a single market share the same underlying economy and the same market-wide pressures when those arrive.

Adding a fund because it has performed differently lately is chasing a correlation measured over exactly the wrong period. The question worth asking about any addition is what it would do in the conditions that would damage everything else you hold. That question is answerable from the nature of the asset rather than from any correlation table.

What to do with the knowledge

Assume that in a severe decline most growth assets will fall together, and size the defensive part of the portfolio accordingly. Base that defensive allocation on what you would need to draw or hold through such a period rather than on a correlation estimate.

Rebalancing works precisely because it forces action across a boundary that a crisis has just moved. A plan that survives correlated falls does not need correlations to behave themselves, which makes it a more robust design. This is an argument for simplicity again, because a simpler portfolio makes its real risk very much easier to see.

Adjust the size of it until it is something you would actually do tired.

Reading correlation figures honestly

When a document quotes a correlation, find the period it was measured over before reading anything at all into the number. Ask whether that period contained a severe decline, because a figure drawn from a calm decade describes calm decades.

Put simply, treat any correlation as a description of the past rather than as a property belonging to the asset itself. Where a product is marketed on low correlation, that is the claim most worth checking against a genuinely stressed period. The useful version of the concept is qualitative: what does this hold, and what conditions would make it fall.

Everything above, in order of what to do first

1. What correlation describes. Correlation measures the tendency of two holdings to move in the same direction, and it is always calculated over a chosen period.
2. Why correlations rise in a crisis. During severe declines, selling is frequently driven by a need for cash rather than by any view about the individual assets.
3. What still tends to help. High-quality government bonds have historically behaved differently from equities during equity declines, though not in every single episode.
4. The diversification that is mostly cosmetic. Holding several equity funds tracking overlapping indices produces a portfolio with one dominant risk and several separate statements.
5. What to do with the knowledge. Assume that in a severe decline most growth assets will fall together, and size the defensive part of the portfolio accordingly.
6. Reading correlation figures honestly. When a document quotes a correlation, find the period it was measured over before reading anything at all into the number.

The takeaway

Build the portfolio so it survives everything falling at once. Then correlations do not have to cooperate.

Pick the one that costs you least, and let the rest wait.