CAMBRIDGE, Mass. — Human beings have evolved over millions of years from simple life forms. Trading models, on the other hand, have not.
But a group of scientists has taken lessons from evolution and, using high-powered computers, applied them to creating an expansive trading model that employs Darwinian theory to uncover inefficiencies in foreign exchange markets.
The group of five physicists and astrophysicists, led by Harvard-trained astrophysicist Aaron Sokasian and operating under the name Financial Labs LLC, started managing about $6 million in the model earlier this month and plans to launch a hedge fund, all using the firm's new trading model.
Mr. Sokasian's program creates not one but hundreds of models and, using real market data, identifies the strongest ones and builds on those to create what could be called a "supermodel." When a market inefficiency is identified, the program sends out a buy or sell signal.
"The only way a (trading) model can be successful in the long run is to exploit inefficiencies, and the key is that these inefficiencies are dynamic and change in time in ways you might not anticipate," Mr. Sokasian said. "We found empirically that it's difficult to come up with one model that would work and be robust enough to indicate how these inefficiencies might pop up."
That's where the genetic programming comes in.
Financial Labs scientists created a computer program that devises hundreds of trading models, or a "population" of models. Using data from about a decade's worth of Group of Seven currency trading, the models are tested and the ones that perform best are identified.
The best pieces of the best performing models are then combined to create even better models. The process continues, giving off buy or sell signals, Mr. Sokasian said.
"We apply survival-of-the-fittest criteria" to determine the best performing models, Mr. Sokasian said. "That only allows the most successful ones to propagate to the next iteration. And through a number of iterations, you have a very specific population that is good at exploiting inefficiencies."
But because the nature of market inefficiencies changes, the Financial Labs program also uses different criteria to create the models so each one is based on a different set of criteria, which can include volatility moving averages, market indicators and others.
Mr. Sokasian said the model has been tested using fresh data for every test, which he said was an important distinction from other trading models. "Often people who build trading models will develop them on one data set and continue to optimize it on that data set," he said. "Hence the performance is optimized on that data set, but you would not expect (market) behavior to continue on the same data set."
Mr. Sokasian and his colleagues built their trading model, dubbed Darwin-FX, at the request of Thomas Plaut, a co-founder and chief executive officer of FX Solutions, an online currency-trading platform based in Ridgewood, N.J.
Mr. Plaut, who also co-founded Financial Labs with Mr. Sokasian, said the goal was to have a foreign-exchange trading model that could return 8% to 12% per year while avoiding the pitfalls of other systematic trading programs that overoptimize returns.
Mr. Plaut said his target market for the fund is institutional investors "looking for an absolute-return product who want to manage risk in a systematic way." Thus far, clients who have committed funds to the managed account include a global bank, a broker-dealer and some high-net- worth individuals. Mr. Plaut hoped to have about $20 million in the managed account by the end of November.
Tony Czapla, managing director of NT Financial Group LLC, a Chicago-based broker-dealer that specializes in electronic brokerage, said he is ready to give Financial Labs a try.
"The concept of it makes sense by virtue of the fact it's been tested for hundreds of millions of years," Mr. Czapla said. "It's nature. That's what makes it so appealing."