Diversification is generally treated as a property of a portfolio - something a holding either provides or does not. It is a property of a correlation, and correlation is a variable that moves with the market regime.

The entire mechanism rests on assets not moving together. What follows concerns how much of that benefit survives a crisis, using the standard two-asset variance identity and the published record on how correlations behave in the tails.

Correlation Rises in Bear Markets and Not in Bull Ones

The asymmetry is the finding. Longin and Solnik’s study of extreme correlation in international equity markets, published in the Journal of Finance in 2001, modelled the tails of the joint distribution directly using extreme value theory rather than inferring them from full-sample correlations. Correlation rose in bear markets. It did not rise in bull markets.

Their second result matters as much as the first: correlation was related to the market trend, not to volatility as such. That rules out the comfortable reading in which correlation simply rises with turbulence and falls back symmetrically. The move is directional, and it runs against the portfolio.

The observed magnitudes are consistent across episodes. Pairwise equity correlations that sat near 0.40 in normal conditions ran roughly 0.70 to 0.80 through the 2008 financial crisis, and spiked to about 0.75 within a few months in the 2020 drawdown.

Two-Thirds of the Benefit Sits in That Exact Range

Portfolio volatility as a percentage of holding a single asset outright, as correlation between two equally-weighted, equally volatile assets rises from 0 to 1.

At a correlation of 0.4, a 50/50 portfolio’s volatility runs at about 84% of a single asset’s - a 16% reduction from diversification alone. At 0.8, that reduction shrinks to roughly 5%.

The decline is not gradual. Around two-thirds of the benefit disappears across precisely the range that correlations traverse in a real crisis, which means the loss of diversification is concentrated in the same weeks the portfolio most needs it. Longin and Solnik’s own conclusion was that the benefits of international diversification are vastly reduced in bear markets, when they are most needed, and that allocation should adapt to that fact.

The Assets Did Not Change; the Regime Did

It is worth being precise about what fails here. The underlying holdings do not stop being different businesses in a crisis, and diversification does not fail as a strategy. What changes is the input the benefit is computed from.

Fear and forced selling push prices to move together irrespective of what the individual businesses are doing. A portfolio measured as well diversified through calm markets was measured against a correlation that no longer holds when it matters - and a backtest run on full-sample correlations will not reveal this, because the full sample is dominated by the calm periods.

What This Means for Allocators

Stress-test the portfolio at crisis correlations, not historical average ones. Re-running the variance calculation at 0.8 rather than 0.4 is a two-minute exercise and it is the number that will apply when the result matters.

Look for assets whose correlation is structurally rather than empirically low - a different cash-flow driver, a different currency, a different liquidity profile. Low measured correlation across a calm sample is the property most likely to disappear under stress.

And treat the diversification figure as conditional. It is real, it is worth having, and it is smallest exactly when the portfolio is being tested.

Back to Writing