From the July 01, 2005 issue of Futures Magazine • Subscribe!

QIM: Quantifying profits

After much contemplation, Jaffray Woodriff settled on quantitative behavioral finance to describe his trading methodology. It is a mouthful, as is behavioral finance academics, artificial intelligence and evolutionary programming — all apt descriptions of what he does. It also is understandably complex considering he has designed software that has created and tested hundreds of billions of trading models. The best models are employed in Quantitative Investment Management’s (QIM) Global program.

Woodriff, system designer and one of three principals at the Charlottesville, Va., based QIM, created his first trading system in 1987 as a college student. And while his trading was successful, that program and the subsequent commodity trading advisor (CTA) he formed never gained traction. In 1997 he moved to New York to trade his unique models for a bank. He ended up at Societe Generale were he traded commodities, currencies and individual stocks.

In New York he worked with Michael Geismar, another QIM principal, who was an old college buddy from the University of Virginia. Greyson Williams, pictured, is also a principal.

In December 2001 they went off on their own to trade Woodriff’s methodology with proprietary money. In October 2003, QIM started trading customer funds. The program produced returns of 35.69% in 2003, 22.96% in 2004 and is one of the few CTAs enjoying a positive 2005.

Woodriff’s software, which is based on mathematics, utilizes artificial intelligence techniques that take a group of relatively simple inputs and create a vast number of trading models. Those models are tested throughout 25 years of market data in 40 different markets and the best 1,500 models are used in the program. The models are scored on performance and correlation to each other. They create an overall score, between -500 and 500, for each market. A score above 100 or below -100 will generate a buy or sell signal. It is a compilation of the all the models that recognize market patterns; some are trend following and some are countertrend, while others are neither.

“The models are generated by computer code, they’re not selected because I went out to try and find a countertrend model that would do great at long-term market turns. The models are screened algorithmically based on how well they predict the movement of markets at any time,” Woodriff says. He ends up with many long- and short-term models. The composite signals tends to be the short- to medium-term type signals, holding trades an average of eight to 10 days.

comments powered by Disqus