This content is from: Home

It's in the genes; Naming names

Small armies of haggard computer programmers often work around the clock for weeks or months at a time to perfect a complicated econometric model or financial market simulation.

Now machines are pulling the all-nighters.

Running so-called genetic programs, which are just beginning to emerge from academic laboratories, ten Apple Power Mac computers in Cambridge, Massachusetts, are turning out foreign exchange trading models at an unheard-of pace. The machines are whirring at Financial Labs, a joint venture that Ridgewood, New Jersey­based brokerage FX Solutions launched in July with Aaron Sokasian, a 27-year-old genetic programming whiz who holds a Ph.D. in astrophysics from Harvard University.

In genetic programming, evolutionary principles rule. The Power Macs test models against "large, noisy data sets" that represent market events, says Sokasian. "Out of a population of many models, you breed and propagate the most successful ones and discard the others."

Darwin-FX, the first product of Sokasian and his team of four Harvard rocket scientists, is "a force multiplier," says FX Solutions co-founder Thomas Plaut. "We'd need hundreds of thousands of programmers to duplicate what it does."

How good is it? In a test run on Group of Seven currencies between 1995 and 2003, Darwin-FX returned an average 11 percent annually -- two to three times conventional trading results.

"Its trading experience is limited, but it's got a great pedigree," says Anthony Czapla, head of Chicago-based brokerage NT Financial Group, which recently began introducing Darwin-FX to clients. "An evolutionary process that's existed in nature for hundreds of millions of years is hard to argue with."

Naming names

Until last year Language Analysis Systems was an obscure government contractor that only a customs agent or an intelligence analyst might love. Then the private sector woke up to the implications of the USA Patriot Act. Now LAS's software, NameHunter, has begun to make its way into the financial industry's regulatory compliance arsenal.

Passed in the wake of September 11, the Patriot Act tightened rules requiring financial companies to monitor transactions for possible money laundering or terrorist connections. Transaction analysis tools were readily available from companies like Fairfax, Virginia­based Mantas and London's Searchspace, but institutions now have to check customer names against burgeoning databases of suspected wrongdoers. That's where LAS comes in.

Herndon, Virginia­based LAS invented modern name-matching technology by enhancing basic keyword searching with artificial intelligence. Founded in 1984, the company cut its teeth helping the State Department screen visa applicants. Later LAS worked for two Treasury agencies -- the Financial Crimes Enforcement Network and the Office of Foreign Assets Control -- good preparation for the Patriot Act.

Now financial institutions can compare their files against nearly a billion names in LAS's cross-cultural database, which detects matches and potential falsifications in, say, Arabic, Russian or Thai names that don't translate easily or uniformly into English.

"We've come up with 200 variations of 'Mohammed,'" notes LAS chairman and CEO John Hermansen, a 57-year-old Ph.D. in computational linguistics from Georgetown University. "One of our partners came up with only 38 -- which is why they're now partnering with us."

That partner, Las Vegas­based Systems Research and Development, has embedded NameHunter in its Non-Obvious Relationship Awareness data mining and fraud detection system. Similarly, Colorado Springs, Colorado­based Factor 3 Technologies has integrated NameHunter into its compliance system for brokerages."LAS fills big gaps, correcting for input errors, mistranslations and spelling variations," says Factor 3 CEO Peter Gadkowski. "That takes our system to a whole different level." -- J.K.