When Genius Failed Ch. 2: The Hidden Assumption — Correlation Is Not a Constant

Investment Background

Chapter 1 left a question. If the models were not wrong and the historical data was not wrong, what was?

The answer is an assumption. It was never written into the model's equations, because it was too obvious — so obvious that nobody thought it needed stating.

The assumption was that the different bets were independent of one another.

LTCM did not put all its money into one trade. Quite the opposite: it was extremely diversified. It ran convergence arbitrage in U.S. Treasuries, in Italian government bonds, in Danish mortgage bonds, in Japanese swap spreads, in U.K. gilts. It also ran equity pairs trades, merger arbitrage, and short equity volatility.

That diversification was not decoration. It was the foundation of the entire risk model.

The Wall Street Translation

Why does diversification permit 25x leverage? Let us do the arithmetic, because it is the key to the whole disaster.

Suppose you hold one hundred bets, each with 10% volatility on its own.

If those hundred bets are perfectly independent, the portfolio's volatility is not 10% but 10% divided by the square root of 100 — that is, 1%.

This is one of the most reliable results in statistics. Diversification really does destroy risk, and does it with remarkable efficiency.

Volatility falling to 1% means you can safely run far more leverage than usual. A 1%-volatility portfolio levered 25 times has an effective volatility of 25% — high, but not absurd for a hedge fund.

LTCM's entire risk architecture rested on that square root.

Now change the assumption.

If those hundred bets are perfectly correlated — meaning they are actually one bet written down a hundred times — the portfolio volatility is not 1% but 10%.

Ten times larger.

And the leverage is 25x. Ten percent volatility multiplied by 25x leverage is an effective volatility of 250%.

In that state, wiping out your capital does not require a crisis. It requires a few trading days.

The Crux: Correlation Is Not Fixed

This is the heart of the chapter, and the single most useful sentence in this book for you.

Correlation is not an intrinsic property of an asset. It is a function of the market's state.

LTCM's models estimated correlations between positions from historical data. In calm periods those numbers were low — and genuinely so. The Italian government bond spread and the Danish mortgage bond spread really do have little to do with one another. One is driven by Italian fiscal policy, the other by Danish mortgage prepayment behavior. Their fundamentals are nearly unrelated.

In normal times, the estimate was accurate.

But under stress, the thing driving prices changes.

When panic arrives, investors stop asking "what is Italy's fiscal position?" They ask exactly one question:

"Is this liquid? Can I sell it right now?"

And along that dimension, every one of LTCM's positions was the same position.

Every convergence trade was structurally identical: buy the illiquid thing, short the liquid thing. That is precisely why the spread existed in the first place — it is a liquidity premium.

So when the entire world began paying up for liquidity at the same moment:

• Everything they owned was falling.
• Everything they were short was rising.
• On the same day, in every country, across every asset class.

They believed they held one hundred independent bets. They actually held one bet — "the liquidity premium will narrow" — written down one hundred times.

The diversification was an illusion. Not because they failed to diversify, but because the dimension they diversified across was not the dimension that mattered in a crisis.

August 17, 1998

The trigger came from a direction almost nobody had priced.

On August 17, 1998, the Russian government did two things: it devalued the ruble, and it defaulted on its own local-currency debt.

The second is the one that mattered.

A sovereign defaulting on debt denominated in its own currency is close to theoretically impossible — a government can always print money to repay obligations in the currency it issues. That is the very basis of the concept "risk-free rate."

Russia defaulted. It chose default over printing.

The direct loss to LTCM from Russia was not by itself fatal. Their Russian exposure was limited.

What was fatal was the effect on the psychology of every investor on earth.

If a sovereign can default on local-currency debt, then every cell in every risk model marked "impossible" needs rewriting. And when the world's risk managers rewrote those cells in the same week, they all reached the same conclusion:

Reduce risk. Sell what is hard to sell. Buy what is easy to sell.

This is called the flight to liquidity.

LTCM's entire portfolio was the perfect inverse of that behavior.

The Numbers

From August into September, LTCM's capital moved as follows:

Date	Capital
January 1998	roughly $4.7 billion
August 21 (single day)	single-day loss of roughly $550 million
End of August	roughly $2.3 billion
End of September	roughly $400 million

Four months. A loss of more than 90%.

Note the August 21 figure. $550 million in one day. In their risk models, a single-day loss of that magnitude was roughly a once-in-ten-thousand-years event.

It happened on a Friday.

Then several more things of similar magnitude happened over the following weeks.

When your model says something occurs once in ten thousand years and it occurs three times in a month, what you have learned is not that you were unlucky. You have learned the model is wrong.

Why Leverage Made It Irreversible

The second effect of leverage has to be stated plainly, because it is far more serious than "it magnifies losses."

An unlevered investor can choose to do nothing while prices fall.

If you hold an index fund and the market drops 40%, your account shrinks — but nobody can force you to sell. You can wait. Historically, waiting has almost always worked.

A levered investor does not have that option.

As LTCM's positions fell, the banks lending to them issued margin calls. Meeting a margin call requires selling assets. And when they sold — given that their positions were enormous relative to those markets — prices fell further.

Which triggered more margin calls.

This is a self-reinforcing loop, and it has no natural stopping point.

Worse, the market knew they were struggling. Wall Street traders knew what LTCM held and knew they had to sell. So those traders sold ahead of them, pushing prices down before LTCM could get out.

At that stage, LTCM's counterparty was no longer an abstract market. It was a group of people who knew you had to sell, and acted on it.

Executable Trading Rules

1. In any portfolio, first ask whether these positions become the same position under stress. Do not look at asset class labels; look at drivers. Three different tech stocks are not diversification. A U.S. fund plus a European fund plus an emerging markets fund will correlate far more tightly in a crisis than calm-period historical data suggests.

2. Plan against crisis correlation, not average correlation. A practical rough method: assume that in a severe crisis all of your equity holdings correlate at about 0.9. If your portfolio is still acceptable under that assumption, it is robust.

3. Distinguish clearly between money that will fluctuate and money that can be forcibly sold. This is the most important line in the chapter. Without leverage, a decline is a paper fluctuation and time is on your side. With leverage, a decline triggers forced selling and time is against you. These are not differences of degree; they are differences of kind.

4. Check your portfolio for hidden leverage. A margin account is explicit. But selling naked options, holding 2x or 3x leveraged ETFs, and certain structured products all embed leverage. If you cannot state your actual leverage multiple immediately, assume it is higher than you think.

5. Understand that genuine diversification has exactly one source: difference in drivers. Cash and short-term Treasuries are truly independent of stocks in a crisis, not because they are "another asset class," but because in a flight to liquidity, they are the thing being fled to.

Relevance to a Retirement Portfolio

This chapter maps directly onto the barbell in Antifragile — but from the opposite direction.

Taleb tells you why to hold extremely safe assets plus a small convex position. This chapter shows you what happens when you believe you are diversified and are not.

For a retirement portfolio the conclusion is concrete:

Your core holding should be a low-cost total-market index fund plus high-quality short-term bonds or cash. That combination is robust not because it is clever, but because in a crisis it triggers nobody's demand that you sell.

LTCM had better models, better mathematics, and better people. The only thing they lacked was the right to do nothing on the worst day.

That right is available to you for free. Its price is giving up leverage.

Chapter 3 takes on a subtler problem: why a group that genuinely understood risk levered 25 to 1 anyway. The answer is not in the mathematics. It is in the incentives.