(Image: A Satire of Tulip Mania, Jan Brueghel the Younger, 1640)
“Your investment approach must account for the truth that there are going to be large drops in every asset class, and that ‘diversification’ alone is not enough (correlation goes to 1 at the worst times, like it did in 2008).”
– Daniel Taylor, October 2018
There is a version of the diversification story that goes like this: spread your capital across uncorrelated asset classes, and the mathematics of modern portfolio theory will protect you. Bonds will rally when equities fall. International markets will diverge from domestic ones. Alternatives will zig when everything else zags. It is a tidy theory — Nobel Prize–winning, in fact — and for long stretches it appears to work. The problem is that it tends to stop working precisely when it matters most.
Much like we saw yet again last month (March 2026), correlation is not a structural property of asset classes. It is a regime-dependent phenomenon — and in the three great dislocations of the past two decades, the correlations investors had spent years building portfolios around simply ceased to exist. Last month was decidedly a unprecedented paradigm for markets, and as we’ll see, the failure of diversification at such times should not have been unexpected.
Understanding why diversification fails — and what actually offers protection when it does — is not a theoretical exercise. It’s the central question of risk management in our era of persistent macro volatility.
THREE TIMES DIVERSIFICATION FAILED
Let’s start with 2022, because it’s the most recent (so most easily remembered) and it’s unambiguous. The 60/40 portfolio — the canonical expression of diversification wisdom — delivered one of its worst performances in modern history. The S&P 500 fell 18.1%. Simultaneously., the Bloomberg U.S. Aggregate Bond Index suffered its worst drawdown in benchmark history. The stock-bond correlation, which had been persistently negative for over a decade following the Global Financial Crisis, turned sharply positive as inflation became the dominant macro variable — averaging 0.41 across the post-2022 period against a prior-decade average of -0.37. A portfolio constructed to hold up when equities fell instead amplified the loss. The hedge became the risk.
March 2020 tells a different version of the same story. The dominant narrative of that month was the COVID-19 crash: the DJIA fell roughly 26% in four trading days, and the S&P 500 dropped more than 30% peak-to-trough. Treasuries initially provided shelter. But as the crisis deepened, a global move to cash overwhelmed even the safest markets: bid-ask spreads on 10-year Treasury bonds widened to crisis levels, investment-grade corporate bonds sold off alongside equities, and even gold — the classic hedge — temporarily fell as leveraged investors liquidated everything to meet margin calls. The mechanism through which diversification was supposed to work broke down during precisely the most violent two weeks of the crisis.
And 2008 during the GFC is the archetype. Pairwise equity correlations across global markets spiked from approximately 40% pre-crisis to nearly 70% during the most acute phase, and remained elevated for more than five years. A study published in the Financial Analysts Journal found that a portfolio diversified across U.S. stocks, bonds, international equities, emerging markets, and REITs saw its effective equity beta rise from 0.65 to 0.95 — everything moved like stocks. Diversification failed across styles, sizes, geographies, and alternative assets. Private equity fell. Hedge funds fell. REITs fell. High-yield bonds fell alongside equities, for precisely the structural reason Merton’s model(1) predicts: as default probability rises, debt becomes equity.
WHAT THESE MOMENTS HAVE IN COMMON
Three episodes, three different failures – but one underlying cause. In each case, a macro regime shift altered the dominant factor driving asset returns, which in turn eliminated the low-correlation properties that had made diversification appear structural.
In 2022, the operative regime was inflation. When CPI is the dominant risk variable, both equities and bonds are repriced simultaneously: equities because higher discount rates reduce valuations, bonds because rising yields destroy principal. The stock-bond correlation, which had been reliably negative in a low-inflation, growth-driven regime since the late 1990s, simply inverts in a sustained inflationary environment. The negative correlation investors came to regard as a natural law was, in historical terms, an anomaly — a feature of temporary low-inflation equilibrium, not a structural property of the asset classes themselves.
In March 2020, the operative shock was liquidity. When fear is acute enough, investors stop asking what something is worth and start asking whether they can sell it at all. In that environment, the usual distinctions between risky and safe assets become temporarily meaningless: everything liquid gets sold by someone who needs cash urgently. This is why even U.S. Treasuries — the deepest, most liquid market on earth — experienced deteriorating bid-ask spreads and forced the Federal Reserve into extraordinary interventions within days.
In 2008, the regime was credit contagion and leverage. Assets that appeared uncorrelated in a world of functioning credit markets turned out to be exposed to the same underlying factor: the continued functioning of the financial system itself. When that factor becomes the dominant variable — when the question is not “which asset will outperform” but “will the system survive” — correlations converge not because asset classes are similar, but because the single systemic risk overwhelms all idiosyncratic asset class considerations.
The common thread across all three episodes is that diversification is only as good as the stability of the regime that produced the historical correlation from which the data was derived. Correlations drawn from 10-year rolling periods will always be wrong at the moment of maximum stress, because the macro regime that is breaking down is the same regime that generated the data. You are being protected by the past precisely when the past no longer applies.
One other note on these downturns: while 2022 and 2020 were technically bears, in that they fell more than 20% from their highs, I would argue they were not true bear markets. Indeed, we have Managing Directors at all the major banks and asset managers who have never seen a real bear market in their professional careers. The true bear markets of 1972-74 (-43%), 1987-89 (-30%), 2000-2002 (-45%), and the GFC (-51%) saw an average drawdown in the S&P 34% and took an average of 4 years to fully recover. So, again in the context of history, our recent v-bottom downturns and quick recoveries to new highs look like the anomaly. Markets don’t always come right back.
WHAT GENUINELY UNCORRELATED RETURN SOURCES LOOK LIKE
If traditional asset class diversification is regime-dependent, the question then becomes what real structural non-correlation actually looks like — and the answer requires a completely different framework.
The clearest example is systematic trend-following, specifically managed futures. The BTOP 50 index of managed futures strategies returned +17.7% in 2008 while the S&P 500 fell more than 38%. In 2022, the SG Trend Index gained substantially as both equities and bonds collapsed. The reason for this robustness is structural: trend-following derives its returns from the persistence of price movements, which tends to occur precisely when macro regime shifts are occurring — the same environment in which static diversification collapses. As Alex Greyserman and Kathryn Kaminski document in Trend Following with Managed Futures, an 800-year dataset confirms low correlation to traditional asset classes, positive (right tail) skewness, and strong crisis performance — not because trend following is immune to all environments, but because crisis periods are precisely when trends are most sustained.(2)
Global macro strategies — discretionary or systematic — offer a second source of genuine uncorrelatedness, to the extent that returns are driven by some mechanism meant to identify macro regime changes rather than broad exposure to risk premia. A manager who is structurally short Treasuries going into a rate-hiking cycle, or long commodity futures during an inflationary supply shock, generates returns from the regime transition itself. The return stream is not “uncorrelated to stocks” because it holds a different asset class — it is uncorrelated because it is designed to exploit the macro variable that is undergoing a change.
Volatility strategies represent a third category. Long volatility benefits from the feature of market crises that destroys everything else: the rapid repricing of uncertainty. The cost is the carry drag. This is not a bug but a diagnostic truth about what genuine diversification requires: true uncorrelatedness is not free. It extracts its price in the years when it is not needed, precisely because it pays most in the years when nothing else does.
What does not belong in this category, despite common assumptions, is private equity, private credit, or real assets valued on an appraisal basis. The smoothed returns that make these appear uncorrelated in a portfolio model are really just the product of infrequent valuation — the underlying economic exposures are deeply correlated to the same regime factors that move public markets. The 2008 data are instructive: private equity indices posted catastrophic losses once mark-to-model valuations gave way to market reality.
THE LAPSED INSURANCE POLICY OF DIVERSIFICATION
The standard professional response to 2022 was to note that the 60/40 portfolio “recovered” in 2023. While this is technically true, it is also precisely the wrong lesson to draw. The question is not whether diversification eventually works over a 10-year rolling period — it generally does, which is why the institutional consensus remains anchored to it. The question is whether the protection it promises will materialize at the specific moment it is most needed.
The lesson of 2008, March 2020, and 2022 — read together — is that traditional diversification, de facto, functions as insurance that lapses under the conditions most likely to trigger a claim. The macro regime that generates a comfortable correlation matrix is, by definition, not the macro regime that is in the process of breaking down.
This does not argue against diversification, but rather for a more precise understanding of what correlation is, and when it holds. Investors who build portfolios on static historical correlations are implicitly assuming future regime stability — and that assumption is most dangerous when macro uncertainty is highest, inflation is volatile, or financial system stress is acute.
Genuinely uncorrelated return streams do not offer free protection. They require patience, some fee tolerance, and the willingness to hold strategies that may underperform in rocket-ship, extended bull markets (i.e – not succumbing to FOMO when everyone else is piling in). This is the price of genuine diversification.
So, the question that remains, and that no historical dataset can fully resolve: if you construct a portfolio for the regime that could change (and eventually will, often dramatically), are you managing risk — or simply postponing it?
_______________
Footnotes:
- Robert Merton’s 1974 structural credit model treats a firm’s equity as a call option on its assets and its debt as equivalent to holding those assets while short the same call. Under this framework, as a firm’s asset value approaches the face value of its debt (i.e – default probability rises) the statistical properties of the debt converge toward those of the equity. The practical consequence in a systemic crisis is that high-yield and even investment-grade corporate bonds lose their fixed-income character and begin trading as distressed equity proxies, moving in lockstep with stock markets rather than offsetting them. This is precisely what occurred in late 2008, when credit spreads and equity volatility became nearly perfectly correlated across the capital structure. Merton’s original paper, “On the Pricing of Corporate Debt: The Risk Structure of Interest Rates,” was published in the Journal of Finance in 1974.
- Managed futures strategies are not homogeneous. Short-term trend followers may underperform during sharp initial dislocations and outperform during extended trends; multi-strategy approaches show different crisis profiles. The “crisis alpha” argument is strongest for intermediate-to-long holding period trend systems across diversified futures universes. Investors should evaluate specific program characteristics rather than relying on category-level generalizations.