Crunching for Dollars
A.I. takes aim at Wall Street
In June, a computer its creators call the most powerful ever built for commercial use (and the fifth most powerful in the world) will go online in Los Angeles. The machine, as yet unnamed, will be dedicated to one goal: beating Wall Street.
Whether by hubris or genius, its creatorsa crew of engineers, researchers and investors led by Cayman Islands-based JJX Capitalbelieve their machine will help them predict, with unprecedented speed and accuracy, the future price movements of every stock, bond and commodity traded in the United States. “What we’re talking about here is doing investment strategy optimizations that would take a matter of hours or days on a PC, being able to have them done in a matter of seconds, and do that not just for one stock but the whole market,” says Steve Ward, CEO of Frederick, MD-based Ward Systems, which is providing artificial intelligence software for the project.
The machine itself will consist of 2,048 parallel 1-gigahertz processors, linked to a massive shared memory warehouse. With a throughput of 2 trillion floating point operations per second (teraflops), the machine, its creators say, will race through calculations almost three times faster than the next-fastest commercially dedicated supercomputer, owned by San Francisco-based brokerage Charles Schwab.
This story is only available to subscribers.
Don’t settle for half the story.
Get paywall-free access to technology news for the here and now.
Subscribe now
Already a subscriber?
Sign in
You’ve read all your free stories.
MIT Technology Review provides an
intelligent and independent filter for the
flood of information about technology.
Subscribe now
Already a subscriber?
Sign in
The project is the brainchild of John Fitzpatrick, the 38-year old president of JJX Capital, a startup hedge-fund company. Fitzpatrick believes his company will be able to reduce risk and increase returns for its clients by accelerating both the accuracy and the sheer number of transactions performed on a daily basis. “By increasing the velocity of decision making through high-powered computing, you can increase the velocity of money, and thus increase the working power of a capital base,” says Fitzpatrick.
The basic challenge of quantitative stock, bond or commodity analysis is the sheer volume of datahundreds of indicators per stockthat must be accounted for. And the same indicators change their meaning depending on a situation, so analytical systems must learn from trends, adjust strategies and the relative weight of indicators, and test hypotheses.
Enter artificial intelligence. Many software firms, including Ward Systems, offer PC-based AI software for individual and institutional investors. The software incorporates a combination of artificial intelligence technologies, including fuzzy logic, neural networks and genetic algorithm optimization to help predict the performance of an investment.
“The more complex the models become in financial analysis applications, the longer they take to run on conventional computer systems,” says Fitzpatrick. “To regain that advantage and have very refined risk models, high performance computing is a necessary direction to making that happen…We want to reduce the unknowns to the absolute smallest number we can.”
Fitzpatrick boasts that with the parallel processing capabilities of his supercomputer, he’ll be able to simultaneously run analysis using more than 700 indicators and predictive formulas on every investment vehicle that trades in the U.S. today.
But even that may not be enough, says Dr. David J. Leinweber, a visiting professor at the California Institute of Technology and former partner and managing director of the quantitative investment firm First Quadrant, based in Pasadena, CA. “If you keep looking at the past history of the market, you’ll always find something,” says Leinweber. “Whether what you find is actually useful in predicting the future is another matter.”
To prove his point, in 1998 Leinweber demonstrated that the variation of the annual closing price of the S&P 500 index for the ten years from 1983 to 1993 could be predicted to near-perfect accuracy by looking at three variables: butter production in Bangladesh; total U.S. and Bangladeshi dairy production; and world sheep population.
“If I had just taken a bunch of other variablesinterest rates, oil prices, unemployment, money supply, things that didn’t sound as stupidI could find something that looks at least as good,” says Leinweber.
While he believes that artificial intelligence can play a role in investing, Leinweber cautions that reliably predicting the movements of even a single stock is beyond the capabilities of any computer or software today. “On a daily basis, things like artificial intelligence and supercomputers have some incremental value. But the market is the reflection of not only underlying economic laws but also of human behavior; and until someone can explain all aspects of human behavior, you can’t accurately predict the market.”
