Diversification: Why It Fails When You Need It Most

Ask AI · Why Does Asset Correlation Tend Toward 1 in a Crisis?

Diversification can eliminate unsystematic risk, but it can’t eliminate systematic risk. In reality, retail investors lose huge amounts of money almost entirely because of systematic risk.

Diversification can eliminate unsystematic risk, but it can’t eliminate systematic risk. In reality, retail investors lose huge amounts of money almost entirely because of systematic risk. Even more deadly, the mathematical foundation of diversification—correlation staying stable across assets—collapses in a crisis. It provides the least protection when you need it most.

This is not a denial of diversification. Diversification is indeed effective in normal market conditions. But if a retail investor thinks, “Buying 10 stocks means I’m managing risk,” then their understanding of risk management is still at kindergarten level.

#01

Two Types of Risk: One Can Be Diversified, One Cannot

Modern Portfolio Theory (MPT) divides risk into two categories:

Unsystematic risk (idiosyncratic risk): Financial fraud at a company, policy crackdowns on an industry, sudden bad news for a particular stock. This type of risk affects only specific assets and has nothing to do with other assets.

Systematic risk (systematic risk): The entire market declines together. Economic recession, financial crises, liquidity dry-ups, abrupt changes in central bank policy. This type of risk affects all assets.

Diversification eliminates the first type. Holding 30 stocks across different industries means the impact of a black swan in any single stock on the portfolio can be diluted to an ignorable level. This is mathematically provable—and it’s exactly where diversification truly has value.

But for the second type of risk, diversification can do nothing.

A basic conclusion of the CAPM model: in a sufficiently diversified portfolio, unsystematic risk approaches zero, and what remains is systematic risk (beta). In other words, after diversification is taken to the extreme, the portfolio’s risk becomes equal to the market’s risk. If the market drops by how much, the portfolio drops by the same amount.

This is a fact that many retail investors don’t understand: diversification’s endpoint isn’t “no risk,” but “only systematic risk left.” And systematic risk is precisely the biggest—and most deadly—one.

#02

Correlation Tends Toward 1 in a Crisis: The Mathematical Root of Diversification Failure

Markowitz’s mean-variance model is the mathematical foundation of diversification. The core logic of this model is: if the correlation between two types of assets is less than 1, then when you combine them into a portfolio, the portfolio’s risk will be lower than the risk of holding either asset type by itself. The lower the correlation, the better the risk reduction. If correlation is negative, the effect is best.

This logic is mathematically flawless. The problem lies in one premise assumption: that correlation between assets is stable.

In reality, correlation is not stable. And the way it becomes unstable is the most vicious kind—correlation is lower in normal markets, but surges in crises, tending toward 1.

The 2008 Financial Crisis: The S&P 500 fell from 1565 points in October 2007 to 666 points in March 2009, a decline of 56.8%. In the same period, various asset classes also fell sharply (the following drawdowns are estimated based on major benchmarks): the MSCI Emerging Markets Index fell by about 54%. The S&P Global REITs Index fell by about 68%. The high-yield corporate bond index fell by about 26%. The commodities index (CRB) fell by about 57%.

In normal years, the correlation between US stocks and emerging markets is roughly between 0.5 and 0.7; between US stocks and REITs around 0.5; and between US stocks and commodities around 0.3. But in 2008, all these figures jumped to above 0.85.

A “diversified portfolio” holding US stocks, emerging markets, REITs, and commodities in 2008 had an actual decline nearly indistinguishable from holding only US stocks. The protection diversification provided was close to zero.

The COVID Crash in March 2020: The S&P 500 fell 34% between February 19 and March 23. In the same period, investment-grade corporate bonds fell by about 15%, gold fell by about 12% between March 9 and 19, and WTI crude oil fell to negative $37.63 on April 20.

Gold is a traditional safe-haven asset. Over the past several decades, the correlation between gold and the stock market has been close to zero or even negative. Many asset allocation plans use gold as a “hedging tool when the stock market falls.” But in the two weeks of extreme panic in March 2020, gold and stocks fell together. Because when liquidity dried up, everyone was selling everything they could to exchange for cash. Correlation became 1 at that moment.

In 2022: The S&P 500 fell by about 19%, and the Bloomberg US Aggregate Bond Index fell by about 13%. The 60/40 portfolio suffered its worst annual performance since 1937.

The 60/40 portfolio is the most classic asset allocation strategy on Wall Street over the past 40 years. Its effectiveness is built on an assumption: stocks and bonds are negatively correlated. When stocks fall, bonds rise; the return on bonds hedges the losses on stocks.

That assumption roughly held during the low-inflation, rate-cutting cycle from 1980 to 2021. But in 2022, the Fed aggressively hiked rates, inflation surged, and stocks and bonds fell together. The 40-year negative-correlation assumption collapsed within a single year.

These three cases point to the same conclusion: correlation tends toward 1 in a crisis. The mathematical foundation of diversification—correlation between assets stays stable and relatively low—collapses precisely when you need protection the most.

#03

Two Lethal Assumptions of the Markowitz Model

The reason the mean-variance model fails in a crisis is that it relies on two premise assumptions:

Assumption 1: Asset returns follow a normal distribution.

A normal distribution implies that the probability of extreme events (such as a single-day drop of more than 10%) is very low. Based on a normal distribution, the probability of the S&P 500 dropping more than 7% in a single day is about once every few thousand years. But in reality, in March 2020 there were four circuit breakers, including a 12% single-day drop on March 16.

The true distribution of financial returns is “fat-tailed”—extreme events occur much more frequently than predicted by the normal distribution. Taleb discussed this systematically in Black Swan: the normal distribution underestimates tail risk, and it is precisely tail risk that causes the biggest losses.

If the return distribution is fat-tailed, then the “optimal diversified portfolio” computed based on the normal distribution assumption will perform far worse in extreme events than the model predicts.

Assumption 2: Correlation between assets is stable.

We’ve already used data from 2008, 2020, and 2022 to demonstrate how fragile this assumption is. Correlation is not constant—it’s a variable that changes with market conditions. And the direction of its change is the most unfavorable: in bull markets correlation is lower (diversification appears effective), while in bear markets correlation spikes (diversification fails when you need it most).

When these two assumptions collapse at the same time, it means the Markowitz model’s guidance value in a crisis is close to zero. A “best portfolio” built with this model performs well in normal years, but in crisis years it’s almost no different from not diversifying at all.

#04

Empirical Evidence from A-Share Stocks: In Front of Thousands of Stocks Hitting Limit Down, Diversification Equals Zero

From June 12, 2015 to July 8, the SSE Composite Index fell from 5178 points to 3507 points, a decline of 32%.

On July 8, more than 1800 A-share stocks hit the daily limit down. Hitting the limit down means you can’t sell.

A retail investor portfolio holding 30 A-share stocks across different industries had no essential difference on that day from holding just 1 stock—everything hit limit down, and everything was unable to trade. The protection diversification provides is zero.

The deeper problem is that the industry correlation within A-share markets is far higher than in US stocks. In A-share markets, price increases and decreases are driven more by liquidity conditions and policy than by differences in the fundamentals of each industry. When the central bank tightens liquidity or when regulatory policies abruptly change, all industries decline in sync. Under this market structure, the effectiveness of diversification by industry is naturally weaker than in mature markets.

The stock crash in 2015 was not an exception. In 2018, the SSE Composite Index fell 24.6% for the full year, and almost all industries had negative returns. In January 2016, there was a circuit breaker on January 4: within less than half an hour after the opening, two circuit breakers were triggered, and trading across the entire market was halted. In moments like these, how many stocks you hold doesn’t matter.

#05

An Institution’s Real Risk Control: Not Diversification—It’s Position Sizing and Hedging

Inside quantitative institutions, diversification is only the outermost layer of the risk-control system. It’s treated as “basic hygiene habits,” not a “core defensive tool.” What truly matters are two things.

The first is position management.

The maximum position size for an institution’s single strategy is typically no more than 20%-30% of total capital. This means that even if a strategy suffers extreme losses, the impact on total capital is kept within a tolerable range. The remaining 60%-80% is cash or short-term government bonds. In a crisis, it not only won’t lose money, but also provides ammunition for buying at lower levels.

A common approach for retail investors is going all-in. Going all-in means no buffer, no ammunition, and no exit. When the market drops 30%, a retail investor who is all-in loses 30%, while an institution with a 30% position loses only 9%. The difference doesn’t come from how many types of assets they hold—it comes from how much of their capital is allocated to positions.

The second is tail hedging.

Institutions buy downside protection for the stock positions they hold using put options. In normal markets, these options are pure cost—every month you pay a premium, and most of the time they expire worthless. But during a crash, the value of these options can surge tens of times or even hundreds of times, offsetting most of the losses from the stock positions.

Universa Investments is the tail-hedging fund where Taleb served as an advisor. During the COVID crash in March 2020, Universa’s flagship fund returned over 4000% that month. In the same period, the S&P 500 fell by 34%. That return came from their long-term holding of deep out-of-the-money put options—losing a little premium each month in normal times, but when a crash hits, it earns back years of costs in one event and generates massive profit.

During the 2008 financial crisis, Bridgewater’s All Weather fund fell far less than the S&P 500, which dropped 38.5% over the same period. The core of the All Weather fund is not “holding more types of assets,” but “doing volatility equalization and risk budgeting for each asset class”—in essence, a more precise approach to position management rather than simple diversification.

Retail investors don’t have these tools. Retail investors don’t have options accounts (in A-share markets, the threshold for individual investors to open options trading is 500k in capital plus six months of trading experience plus passing an exam), they don’t have futures accounts, they don’t understand volatility equalization, and they don’t understand tail hedging. The “risk management” retail investors can do is just buying 10 different stocks.

This is the most brutal layer of information asymmetry: retail investors think they’re managing risk, but in reality they’re only doing the most superficial and most powerless layer of risk management. The truly defensive fortifications—position control, tail hedging, and liquidity management—are invisible and out of reach for retail investors.

#06

The Real Value and Boundaries of Diversification

After saying all this about the limitations of diversification, there’s one point that needs to be made clear: diversification isn’t useless—it’s just been overestimated.

The real value of diversification lies in eliminating unsystematic risk. If a retail investor holds only one stock, any negative event involving that stock will cause a devastating blow. Holding 10-30 stocks across different industries can effectively dilute the black swan risk of any single stock. This value is real and shouldn’t be denied.

But the boundaries of diversification are also very clear:

Boundary 1: It can’t eliminate systematic risk. When the market as a whole falls, the drawdown difference between a diversified portfolio and a concentrated holding is small.

Boundary 2: Its effectiveness depends on correlation staying stable. Correlation spikes in crises, causing the diversification effect to shrink drastically right when you need it most.

Boundary 3: It’s not all of risk management—it’s only the most basic layer. Without position management and hedging tools, the protection diversification provides is extremely limited.

Boundary 4: Over-diversification dilutes returns. A portfolio holding 50 stocks will have a return rate that approaches the index. If the goal is to outperform the index, over-diversification becomes an obstacle.

#07

For Retail Investors, Three Things Matter More Than Diversification

First, position control. Never go fully invested. Keep at least 30% cash as buffer and ammunition during crises. This matters ten times more than how many stocks you hold.

Second, recognize that systematic risk exists. Don’t think that buying 10 stocks makes you safe. When the entire market enters a downtrend cycle, the difference between 10 stocks and 1 stock becomes almost negligible. Identifying the market’s overall condition—whether it’s in an accumulation phase or a distribution phase, whether it’s in an uptrend or a downtrend—is more important than choosing which stocks to hold.

Third, accept the fact that “some risks can’t be diversified away.” Systematic risk can’t be eliminated through diversification; it can only be dealt with by reducing position size or hedging. If you don’t have hedging tools, the only option is to lower your position when risk increases. This isn’t running away—it’s survival.

Diversification is a good habit. But if a retail investor treats it as all of risk management, then their understanding of risk still stays at the surface of the saying, “Don’t put all your eggs in one basket.”

When an earthquake comes, baskets of any number are useless. What matters is whether, before the earthquake, you put part of the eggs into a safe deposit box.