The Week the Quants Broke: Anatomy of the August 2007 Meltdown

In the second week of August 2007, something strange happened. Quantitative equity funds, the market-neutral, factor-driven, statistically-arbitraging kind, suffered losses that their own risk models said should essentially never occur. Multiple standard deviations, on consecutive days. Some funds lost a third of their value in days.

And then, by the end of the week, much of it came back.

This was not the subprime crisis, at least not directly. These funds did not hold mortgages. They were long and short baskets of ordinary US stocks, hedged to be neutral to the market. They should have been immune to whatever was going on in credit. They were not.

Amir Khandani and Andrew Lo reconstructed the week, and their answer is one of the most important pieces of risk-management writing in modern finance.

The problem: the losses made no sense

Start with what makes this so puzzling.

A market-neutral quant fund is designed not to care about the market. It is long the stocks its models like and short the ones it dislikes, sized so that the market exposure cancels out. If the S&P falls ten percent, the fund is roughly flat. That is the entire point.

In August 2007, the market was not collapsing. The S&P did not crater during the critical days. Yet these funds, which held no mortgage exposure and no market exposure, were hemorrhaging money.

Worse, they were all losing money at the same time, on the same positions, in the same direction. Funds that had never spoken to each other, run by different people with different proprietary models, were suffering nearly identical losses. That is not what independent bets look like.

And then the reversal. The losses substantially reversed within days. A permanent repricing does not reverse. A change in fundamentals does not reverse. Something temporary had happened.

The key idea via analogy: everyone rushing for the same exit

Here is the picture Khandani and Lo assembled.

Imagine a crowded cinema. Everyone in it has, independently, chosen the same seats, because the seats are objectively the best ones. Nobody coordinated. They just all did the same sensible analysis.

Now one person, for reasons of their own that have nothing to do with the film, has to leave in a hurry. They shove their way out. Their shoving disturbs the people around them. Some of those people, seeing the commotion and worried about being trampled, also head for the exit. Now the aisle is jammed, and the shoving intensifies, and more people leave. Nothing about the film has changed. The stampede is entirely self-generated.

That is what the authors argue happened. The chain runs:

1. Somewhere, a large fund needed cash. Most likely a multi-strategy fund or a bank proprietary trading desk that was bleeding badly from its credit book (the subprime crisis was well underway) and got a margin call, or simply decided to cut risk. Its credit positions were illiquid and could not be sold. But its quant equity book was liquid. So that is what it sold.
2. It liquidated fast. Unwinding a large quant equity portfolio means buying back your shorts and selling your longs, in a hurry. That pushes the shorts up and the longs down, which is precisely a loss on the strategy.
3. Everyone else held the same book. Because quant funds all use similar signals (value, momentum, reversal, earnings quality) and all read the same academic literature, their portfolios overlapped heavily. So one fund's forced unwind hit every other fund's positions directly.
4. The losses forced more selling. Funds hitting risk limits, or facing their own margin calls, had to deleverage too. Each unwind pushed prices further, causing more losses, forcing more unwinds. A liquidation spiral.
5. Then the market makers stepped back. Using transactions data, the authors simulated a simple market-making strategy and found something telling: it was profitable before and after the crisis week, but sharply unprofitable during it. That is evidence that the people who normally absorb this kind of selling pressure had withdrawn their capital, exactly when they were most needed, which is why prices moved so violently on the flow.
6. When the forced selling stopped, prices snapped back. Because nothing fundamental had changed. The dislocation was pure supply and demand, so it reversed once the seller was done.

Khandani and Lo also make a broader point about why the system had become so fragile. Quant equity strategies had been extremely profitable in the 1990s. As competition arrived, the raw returns thinned out. Funds responded not by accepting lower returns but by applying more leverage to the same thinning edge. A strategy that earns two percent unlevered and is run at eight times leverage looks like a sixteen percent strategy, right up until the moment it needs to be unwound in a hurry. Leverage does not just amplify returns. It converts a mild dislocation into a forced sale, and forced sales are what turn a bad day into a spiral.

Why it mattered

• It named the risk nobody had on their dashboard. Every risk model measured market exposure, sector exposure, factor exposure. None of them measured how many other people hold exactly what I hold, with how much leverage, and who might be forced to sell it for reasons unrelated to my thesis. Crowding risk went from an unspoken worry to a first-class concept because of this episode.
• It showed contagion travels through balance sheets, not just fundamentals. The channel from subprime mortgages to market-neutral equity funds was not economic. It was a shared owner with a margin call. That is a genuinely different mechanism, and it means diversifying across strategies does not protect you if the strategies share investors.
• It is the definitive case study in liquidity risk. The positions were fine. The models were fine. The exit was the problem. "Your risk is not what you hold, it is what happens when you try to sell what you hold" is the lesson, and it was learned expensively.
• It was a rehearsal for 2008. The same dynamics, forced deleveraging, liquidity evaporation, market makers withdrawing, correlations snapping to one, played out on a vastly larger scale a year later. August 2007 was the tremor before the earthquake, and this paper reads it as such.

The honest limitations

• The trigger is inferred, not observed. Nobody has produced the smoking gun showing exactly which fund began the unwind. The authors are careful to present this as the most plausible reconstruction consistent with the evidence, not a confession. Position-level data for hedge funds is simply not public.
• The strategies are simulated, not actual. Because real fund positions are secret, the analysis relies on reconstructing generic quant strategies and hedge fund index data. Those are good proxies, but they are proxies.
• Crowding is easy to name and hard to measure. The paper convinced everyone that crowding matters. It did not give the industry a reliable way to measure it in advance, and honestly, nobody has since. You generally discover how crowded a trade is on the day it unwinds.
• The reassuring conclusion is dangerous. "It reversed within days" can be misread as "these dislocations always reverse, so just hold on." That is only true if you survive. The funds that were forced to liquidate at the bottom did not get to enjoy the rebound, and leverage is precisely what decides whether you have the option to wait.

The one-line takeaway

Khandani and Lo showed that the August 2007 quant meltdown was not a failure of the models but a stampede for the exit: heavily leveraged funds all held the same positions, one of them was forced to sell for unrelated reasons, and the resulting liquidation spiral proved that in a crowded, levered trade, your real risk is not your portfolio but everyone else's.