The Stability Assumption That Fails When You Need It Most
Portfolio construction using correlation relies on an assumption that is rarely stated explicitly: that the correlations measured from historical data will persist into the future — or at least will be roughly stable during the periods when the portfolio is tested. This assumption is reasonable in normal market conditions. It breaks down catastrophically in crises.
When financial markets are under severe stress, investors facing losses in one asset class tend to sell other assets to raise cash, meet margin calls, or reduce overall risk. This simultaneous selling creates a correlation spike: assets that were largely unrelated under normal conditions begin to fall together, not because their underlying fundamentals have suddenly aligned, but because they are all owned by the same panicking investors.
The result is that precisely when diversification is most needed — during a market crisis — it provides less protection than historical correlation data would suggest.
The 2008-09 Crisis as a Case Study
The global financial crisis of 2008-09 is the most striking modern illustration of correlation breakdown. In the years preceding the crisis, a wide range of asset classes — developed market equities, emerging market equities, high-yield corporate bonds, commodities, real estate investment trusts, and even some hedge fund strategies — showed moderate positive correlations with one another but appeared to offer useful diversification in combination.
When Lehman Brothers collapsed in September 2008, these correlations converged sharply toward +1. Developed market equities fell in parallel with emerging market equities. Corporate credit spreads exploded. Commodities sold off alongside equities. Real estate declined globally. The only assets that reliably held value or appreciated were government bonds of the highest-rated sovereigns (primarily US Treasuries and German Bunds) and, to a lesser extent, gold.
Investors who believed they had built a diversified portfolio across many asset classes discovered that their diversification had been largely illusory — it existed only in calm conditions.
Correlation Asymmetry
Researchers have documented what is called correlation asymmetry: the empirical finding that correlations between risky assets are systematically higher during down markets than during up markets. This means the diversification benefit measured in normal conditions overstates the benefit available in precisely the scenarios where investors most need protection.
This asymmetry has a clear implication: portfolio models that use a single average correlation coefficient — whether calculated over a full market cycle or over recent history — will underestimate portfolio risk in tail scenarios. A model built on 10-year average correlations may look well-diversified on paper while concealing dangerous levels of crisis-period concentration.
Practical Responses
The insight that correlations spike in crises does not invalidate diversification — it refines how to implement it. Several practical responses follow:
First, stress-test portfolios using crisis-period correlation matrices, not normal-period ones. When examining how a portfolio would behave in a 2008-style scenario, use the correlations observed during that crisis rather than the long-run averages.
Second, allocate meaningfully to assets that have historically maintained low or negative correlations even in crises — high-quality government bonds, gold, and cash. These tend to hold their properties better when other correlations spike, because their safe-haven demand increases (rather than decreases) during panics.
Third, size positions assuming correlation can spike. A position that looks well-hedged at a 0.2 correlation may be dangerously large if that correlation moves to 0.8 during the period it matters most.
The broader lesson is that correlations are not constants. They are regime-dependent. A robust portfolio is one that has been designed with an awareness of what happens to its correlations when the environment changes most violently.