The fundamental problem with computer models for trading securities is that they don’t effectively take the outside world into account. They don’t have human thoughts, so psychology can never figure into their calculations. Unlike Gary Cohn in the commodities pit, they can’t look into the whites of people’s eyes and see their fear. And as Gary discovered so successfully, a large part of trading is based on understanding other traders’ emotions. Are they scared? Are they panicked?
In the summer of 2007, fear had started to creep into the markets, and the computer models simply couldn’t pick it up. My colleagues and I began to worry about Global Alpha and AQR when we saw something curious going on with the funds’ vital signs. We used to track how closely these funds’ performance correlated to a benchmark such as the S&P 500 Index (a collection of five hundred stocks that acts as a kind of blood pressure gauge for the stock market). Normally, the quant funds traded within 10 to 50 basis points of the S&P. (A basis point –“bip” for short- is a unit equal to a hundredth of a percentage point: 100 bips equals 1 percent.) In the summer of 2007, AQR and Global Alpha were showing a variance of greater than 250 bips from the S&P 500. Highly abnormal.
We needed to figure our what was going on, and to understand what’s really going on in a quant meltdown, you need to talk to a quant. I was fortunate to have had a great one on our team. Like Cliff Asness, Helga had an economics PhD from Chicago. She spoke with her fellow geniuses at other banks and hedge funds and deduced that the quant funds seemed to be falling victim to their own success: there were just too many of them using the exact same model.
It wasn’t just AQR and Global Alpha that used the model. There were other big funds run by PhDs working with variations on Cliff’s special sauce: there was James Simons’s Renaissance Technologies, and there was D.E. Shaw, among many smaller imitators. As a result of all these companies working off similar models, investment opportunities in heavily capitalized mainstream companies were becoming crowded, so the computers were increasingly seeking out more illiquid and less widely held investments. The more out of the way the security, the fewer buyers and sellers for it, so it can be hard to unwind one of these investments. Although quants do think about the dangers of illiquidity a lot, the mistake they made this time was to fail to imagine that everyone would want to get out at the same time. They were so hypnotized by all the relentless success that they just kept doing most of what the computer model was telling them to do.
If the computer spat out, “Buy 10,000 shares of Lukoil,” the fund’s traders went out and bought the Russian oil company. If the computer said, “Sell May wheat futures,” the traders started selling. The programs kept looking for freakier securities that displayed the anomalies the model was looking for-and the fund managers kept trading. Not enough questions were being asked.
Then, suddenly, everybody’s model was saying, “Sell.” Ironically, this fear in the market was actually being driven by something completely different: emerging jitters in the subprime mortgage market. Nothing to do with math; everything to do with emotion.
Fuente: Why I left Goldman Sachs. Greg Smith. Grand Central Publishing. New York. 2012.