Why does diversification thin out exactly when it is needed?
The most important limit on everything in this course is not theoretical. It is a repeated, observed pattern: correlations rise in crises. The number the whole machine depends on moves against you at the worst possible moment.
What the record shows
- October 2008. In the acute phase of the financial crisis, the dispersion between sectors, countries and styles collapsed. Assets that had spent years behaving differently sold off together; the S&P 500 finished 2008 down about 37% including dividends, with very little inside it spared.
- March 2020. During the fastest major drawdown in modern market history, even traditional shelters wobbled — gold fell for several sessions in the middle of the selloff as investors sold whatever could be sold to raise cash.
- 2022. The classic pairing failed in the open. Rising rates hit stocks and bonds simultaneously: the S&P 500 fell roughly 18% and the broad US investment-grade bond index roughly 13% in the same calendar year — a combination that had not occurred in decades.
Measures of average pairwise correlation among large-cap stocks that sit around 0.2 to 0.3 in calm markets have repeatedly spiked toward 0.6 and beyond during panics. Put that into Unit 2's effective-bets formula and watch a portfolio's number of independent bets collapse in real time.
Why it happens
Three mechanisms, none of them mysterious.
Liquidity, not fundamentals, sets prices in a panic. Forced sellers — margin calls, fund redemptions, risk limits — sell what they can, not what they would prefer to. That indiscriminate selling imposes a common factor on assets with nothing else in common.
One risk becomes the only risk. In normal times thousands of separate stories drive thousands of separate prices. In a systemic event, one story drives every price at once. Diversification across stories does not help when there is only one story.
Shared holders, shared leverage. Assets held by the same leveraged investors get sold by the same leveraged investors. The linkage sits in the ownership structure, not in the underlying economics.
What this does and doesn't mean
It does not mean diversification is useless. The 2008 investor holding one failing bank instead of the index had a categorically worse decade, and the free lunch was served on every ordinary day for years before and after. It means the benefit is state-dependent: largest in normal conditions, smallest in tail events. A portfolio built on the assumption of stable correlations is built on the one number least likely to hold still.
The professional response is honesty in the modelling. Stress-test with elevated correlations rather than historical averages. Treat any calm-period correlation as an upper bound on the protection available. And recognise that what survives a crisis is exactly the systematic risk the last two lessons named — arriving all at once.
Try it now
- The dates are the whole exercise. Navigate both charts below to February–April 2020 and Measure each across those weeks. Two holdings with historically different behaviour: did they diverge, or fall together?
- Now move both to a calm year — 2017 — and measure again. Compare the two pictures, and note which of them a correlation estimated from ordinary history would have described.
- Phrase the limit precisely: correlations measured in calm periods overstate the protection available in a crisis. Say it as an observation about data, which is what it is.