Capacity, Crowding and Decay

In the second week of August 2007 some of the most sophisticated trading firms in the United States lost money in a way their models said was nearly impossible. The stock market barely moved. What moved were the spreads between the stocks that quantitative equity funds owned and the stocks they were short. On 10 August the Wall Street Journal reported that a key Renaissance Technologies fund was down 8.7% so far in the month, that Highbridge's Statistical Opportunities Fund was down 18% as of 8 August and that Tykhe Capital's largest fund was down about 20%; on 14 August it reported Goldman Sachs's Global Equity Opportunities fund down "more than 30%" in a week (as quoted by Khandani and Lo; reported). Renaissance's fund was its Institutional Equities Fund, RIEF, about $29 billion at the time and down 7.4% for the year at that point (Bloomberg, reported). AQR Capital Management's flagship Absolute Return fund lost about 13% from peak to trough within the month before finishing August down 3.4% (Bloomberg Markets, reported).

Jim Simons wrote to RIEF's investors that the plan was to stay the course (as described by the press; the exact wording has not been seen). Those who could do that were largely rewarded within days. Amir Khandani and Andrew Lo, who reconstructed the week with a simple contrarian strategy on S&P 1500 stocks, found that the unwinding of common factor portfolios had begun in July, that market makers withdrew risk capital from 8 August, and that a book-to-market portfolio levered eight to one on the morning of 1 August would have lost about 24% by the close of 7 August. Contrarian positions opened on 8 and 9 August and held for five days then earned 8.20% and 8.96% unlevered (Khandani and Lo, Journal of Investment Management, 2007, and Journal of Financial Markets, 2011). Their explanation, still the leading one, was that a large fund in trouble elsewhere, probably in credit, sold its liquid equity book to raise cash. Because many funds held similar books built from similar signals, its selling moved prices against all of them, their losses forced more selling, and the selling stopped only when the forced sellers had finished.

The episode is called the quant quake, and it is the right place to begin this part of the book because nothing in it was a bad idea. The signals worked before August and kept working after it. What failed was the assumption that a fund's positions were its own. This chapter is about the three ways an edge that has been found, tested and traded is lost without ever being disproved: it is traded in more size than it can carry (capacity), it is held by too many others at once (crowding), and it is competed away over the years (decay). All three sit in the execution and discipline factors of the edge equation, and all three are invisible in a backtest of one strategy run alone.