Running such computationally intensive simulations has become a lot easier in the last decade. Gregg Berman, a former experimental astrophysicist who left the academy for the world of finance in 1993, is one of what he calls “a plethora of PhDs” at RiskMetrics, a firm that provides models, tools, and data to the majority of important banks, brokerages, and hedge funds. (Among other things, the company tries to predict how a derivative will behave in a variety of market conditions–how it might respond, for instance, to weakening exchange rates or increased interest rates.) When Berman started in the business, he says, “full-blown simulations [of the Monte Carlo type] were rare.” Now that computers can be so easily linked, however, Berman might put as many as 1,000 processors to work at once to run “simulations within simulations,” which might measure risk on a product like a mortgage-backed security.
The net result of this improved ability to assign values to increasingly complex derivatives was an explosion in their variety. That meant there was a derivative to suit every investor’s appetite for risk. In consequence, investors were increasingly willing to put more money into derivatives.
Recently, one of the most popular of these new instruments has been collateralized debt obligations, or CDOs. Crucially for our story, CDOs are also the product most closely associated with the summer’s subprime mess. The CDO has been called a “derivative of a derivative,” and to further confuse things, there are CDOs of CDOs, and even CDOs of CDOs of CDOs. A CDO combines both high- and low-risk securities that might derive their cash flow from mortgages, car loans, or more esoteric sources like movie revenues or airplane leases. Investors in a CDO can buy the rights to different levels of income and associated risk, called “tranches.” Generally, the most risky tranche of a CDO pays the most income. Created by quants and priced by quants, CDOs have become a popular way for hedge funds, pension funds, insurance companies, and other investors to buy pieces of high-risk but high-profit sectors like subprime loans. According to the Securities Industry and Financial Markets Association, annual issues of CDOs worldwide nearly doubled between 2005 and 2006, going from $249.3 billion to $488.6 billion.
The quants who devise such derivatives work more or less in public view. They’re obscured mainly by the complexity of their work. But our knowledge of the quants who design trading strategies is additionally occluded by the secrecy of the big fund operators like Renaissance Technologies. I did manage to speak with some current traders, who gave me a general idea of their approach, and with some ex-traders, who were slightly more specific.
One common method that quants use to identify market opportunities is pairs trading. Pairs trading involves trying to find securities that rise in tandem, or that tend to go in opposite directions. If that relationship falters–if, say, the values of two stocks that travel together suddenly diverge–it’s likely to indicate that one stock is undervalued or overvalued. Which stock is which is irrelevant: a trader who simultaneously bets that one will go up and the other one down will probably make money. It’s a strategy that lends itself to the use of computers, which can sort through huge numbers of price correlations over many years of stored data–although the final decision to speculate on the relative pricing of paired stocks generally rests with a fund’s managers.
Quants have also been pursuing a strategy known as “capital structure arbitrage,” which seeks to exploit inefficient pricing of a company’s bonds versus its stocks. Again, computers do the searching, looking for instances where, for one reason or another, the securities are slightly misaligned.
In a similar technique, Max Kogler, a principal at the newly launched MM Capital in New York, uses computers to look for inconsistencies in value between the option on an index fund and the options on the stocks that compose that index. Kogler has a master’s from the University of Cambridge in pure mathematics with a focus on statistics. He says his algorithms look for “baskets of options that are not doing what they’re supposed to be doing.” When his computers find such a basket, he and his partners discuss whether or not to buy.
Kogler runs his algorithms on “one Linux box.” “Part of the allure of our algorithm,” he said in an e-mail, “is that it cuts down computational requirements dramatically. Nonetheless, you’ll want to have a speedy machine with pretty decent clock speed and a couple of parallel CPUs.”