Quant Analytics: from scratch to pro · Professional practice and failure case studies · Pro
Case study: the quant quake of August 2007
In August 2007 many quantitative equity funds lost money at once, in a market that was not obviously in crisis. The leading explanation is a forced unwind of similar, leveraged portfolios. The case shows what crowding, leverage and shared factors do when they meet.
In this lesson
- Describe the unwind hypothesis for August 2007 and say how much of it is established
- Explain how portfolios built on shared factors become exposed to each other
- Compute how leverage turns a small adverse move in a market-neutral book into a large loss of capital
- Judge how unlikely a loss is under a normal model, and why that probability misleads in a crowded market
Read first: Leverage and margin, Correlation and its limits, Portfolio maths: variance, correlation and the limits of diversification
In the week of 6 August 2007 a number of quantitative long/short equity hedge funds suffered unusually large losses. Broad stock indices were not in obvious distress. The episode matters to a quant because the funds were built to be diversified and market-neutral, and still lost together.
What happened
Long/short equity funds hold some stocks long and others short, with the aim that market direction roughly cancels. Quantitative funds pick the stocks with rules based on factors such as valuation or recent returns. Khandani and Lo (2007) simulated a simple contrarian strategy of this type and proposed the unwind hypothesis: a rapid liquidation of one or more quantitative portfolios caused temporary price impact, with most of the unwind on 7 to 9 August. Other funds then cut risk, which added to the losses. In their simulation, returns recovered sharply once the liquidations ended around 10 August.
A later paper (Khandani and Lo, 2011) uses factor and transactions data and dates the stress earlier: the quant meltdown began in July 2007 and continued to the end of the year, with unwinds on 1 and 6 August and a withdrawal of market-making capital from 8 August. The authors call their hypotheses speculative and based only on indirect evidence. This lesson keeps to that level; it does not state loss sizes, since the only ones the papers give are from their own simulations.
The mechanism
Shared factors. Funds using similar signals hold similar positions. Each looks diversified alone. Together they are one large, crowded trade.
Leverage. Market-neutral does not mean low risk. A book with long and short positions of several times capital loses several times any adverse move in its spread.
Forced selling and price impact. If one holder must sell, prices move against the others. Their risk limits then force them to sell, which moves prices further. That loop is deleveraging.
Liquidity withdrawal. The 2011 paper points to market makers pulling back from 8 August. When the usual buyers step away, a given sale moves prices more.
Reversal. A price move caused by forced selling and not by information can reverse once the selling ends. Funds that sold at the low locked in the loss. Those who could hold on recovered much of it, in the simulation.
Worked example: what the numbers say
The figures here are illustrative arithmetic, not reported losses.
A market-neutral book has gross positions of 6 times capital. The positions lose 2% in a day. The loss of capital is 6 x 2%, which is 12%. Neutrality removed market direction and left the leverage.
Correlation shows why diversification fails. Two assets of 10% volatility each, equally weighted, give a portfolio volatility of sqrt(0.5 x (1 + rho)) x 10%. At a correlation of 0.2 that is 7.75%. At 0.9 it is 9.75%, nearly the single-asset figure. When many funds hold the same positions, the correlation between their assets rises in a forced sale.
A normal model also under-reports this kind of loss. Suppose the daily standard deviation of a book is 0.3% and a day loses 1.5%. That is a 5-sigma event. Under a normal distribution it has odds of about 1 in 3.49 million days, or 13,843 years of 252 trading days. When such days occur in the data, the model is wrong about the tails. The cause is the crowd, which a model built on one portfolio's history does not contain.
Lessons for a quant
- Crowding: your risk includes the positions of others who use the same signals. See crowded trade.
- Leverage: size gross exposure so that a forced reduction by others, not just your own, can be survived.
- Liquidity: the cost of exit depends on who else is exiting. Test it at the size of the crowd.
- Correlation going to one: stress correlations with a shocked matrix, not the calm one. See correlation and its limits.
- Model risk: a backtest with no other participants overstates how safely a book can be reduced. A live-versus-expected band flags losses quickly. See model risk.
- Operational response: decide before a shock whether to reduce, hold or add. Selling into forced selling locks losses in.
What changed afterwards
The episode is now a standard reference for crowding and forced deleveraging among systematic funds. Its main change is in how risk is framed: the overlap between your positions and everyone else's is a risk in itself, besides each portfolio's own volatility. No specific rule change is attributed to this episode here.
Common mistakes
- Equating market-neutral with low risk.
- Believing independence of managers means independence of positions.
- Quoting a normal-model probability for a loss caused by a crowd.
- Treating the first explanation of a market event as settled. The authors revised theirs.
- Selling at the low of a liquidity-driven move without knowing the cause.
- Using a model's history as if no one else shared its signals.
In practice
A stress table for a factor-based book lists the loss for a given adverse spread move, the loss when correlations are shocked, and the size of an exit relative to typical volume. Research on FinUties is where studies built on methods like these are published, with their method and limits stated.
Volatility of an equal-weight two-asset portfolio, by correlation
Worked numbers
Each result is recomputed from the formula shown.
| What | Formula | Result |
|---|---|---|
| gross exposure as a multiple of capital | 6 | 6 times |
| adverse move across the positions | 0.02 | 2% |
| loss of capital | gross * adverse | 12% |
| volatility of each asset | 0.10 | 10% |
| portfolio volatility at correlation 0.2 | sqrt(0.5 * (1 + 0.2)) * vol_each | 7.75% |
| portfolio volatility at correlation 0.9 | sqrt(0.5 * (1 + 0.9)) * vol_each | 9.75% |
| illustrative daily loss | 0.015 | 1.5% |
| illustrative daily standard deviation | 0.003 | 0.3% |
| size of the loss in standard deviations | day_loss / day_sd | 5-sigma |
| one in how many million days, under a normal model | 1 / phi(-z5) / 1000000 | 3.49 |
| same in years of 252 trading days | 1 / phi(-z5) / 252 | 13,843 |
Check yourself
A market-neutral book is run at 6 times capital in gross positions. Its positions lose 2% in a day. What is the loss of capital?
6 x 2% = 12% of capital. Market neutrality removes market direction, not leverage.
Why can funds with different managers lose together?
If they rank stocks with similar factors, they hold similar longs and shorts. A sale by one pushes prices against the others, who then may cut risk too.
Two assets have a volatility of 10% each, equally weighted. How does the portfolio volatility change as the correlation rises from 0.2 to 0.9?
sqrt(0.5 x (1 + rho)) x 10%: 7.75% at 0.2 and 9.75% at 0.9. Most of the diversification benefit goes.
A 1.5% daily loss is quoted as a 5-sigma event given a 0.3% daily standard deviation. What does a normal model say?
About 1 in 3.49 million days, or roughly 13,800 years of trading days. A model that gives such odds is wrong about the tails, not the world unlucky.
What should a reader do with Khandani and Lo's explanation?
Treat it as the leading, tentative account. The authors themselves call the hypotheses speculative and the evidence indirect, and the revised view dates the stress to July.
Key terms
Checked facts
Every figure below carries how it was checked: Primary checked (read in the cited primary, official, academic or data source), Two sources agree, or Approximate (an order of magnitude only).
- Primary checked
In the week of 6 August 2007 a number of quantitative long/short equity hedge funds suffered unusually large losses.
- Primary checked
Khandani and Lo (2007) proposed that a rapid liquidation of one or more quantitative equity portfolios caused temporary price impact, with most of the unwind on 7 to 9 August.
- Primary checked
In their simulation, returns of a contrarian strategy recovered sharply after the liquidations ended around 10 August.
- Primary checked
Khandani and Lo (2011) find that the quant meltdown began in July 2007 and continued to the end of 2007, with unwinds on 1 and 6 August and a withdrawal of market-making capital from 8 August.
- Primary checked
The 2011 paper describes its hypotheses as speculative and based only on indirect evidence.
Sources
- What Happened To The Quants In August 2007?. Journal of Investment Management 5(4), 5-54 (Amir E. Khandani, Andrew W. Lo), 2007. Accessed 9 Nov 2026
- What happened to the quants in August 2007? Evidence from factors and transactions data. Journal of Financial Markets 14(1), 1-46 (Amir E. Khandani, Andrew W. Lo), 2011. Accessed 9 Nov 2026
- SR 11-7: Supervisory Guidance on Model Risk Management. Board of Governors of the Federal Reserve System, with the OCC, 2011. Accessed 9 Nov 2026