A group of Harvard-trained physicists and astrophysicists are trying their hand at running a hedge fund in the US. Rather aptly they've called their firm Financial Labs and launched their fund at the end of last month.
The five Harvard graduates set up their firm in July last year. Since then they have been developing a technological infrastructure that they feel differentiates them from other funds. Aaron Sokasian, co-founder of Financial Labs, says: "Our approach is extremely objective and numerically based. We don't really use any kind of discretion."
The firm has been running a managed foreign exchange account since January this year. "The forex market interests us for obvious reasons," says Sokasian. "It's continuous, the liquidity is very deep and it's something we can trade, I think, at an advantage to most other people just because we have the technological infrastructure to allow us to connect up to multiple banks."
This ability to connect to multiple banks means Financial Labs can get extremely tight spreads, explains Sokasian. "Instead of going to a single bank where spreads might be three to four pips we can potentially connect up to several banks and have that reduced down to one or two or potentially zero pip spreads," he says. Sokasian believes this gives his firm an advantage because most players in the forex markets will not implement strategies that are only going to gain one to two pips per trade. "The hope of course is that historically these inefficiencies might continue to exist in the future and hence lead to profit," says Sokasian."Our whole point is that with our infrastructure and our large set of strategies we're going to be able to suck up all the nickels in the world."
The firm's hedge fund takes bets, not only in the foreign exchange markets, but across a number of asset classes and nations. It also employs several different strategies. "They can be categorized as directional strategies, statistical arbitrage strategies and pure arbitrage strategies," Sokasian says.
The hedge fund and the managed account models employ what is referred to as genetic programming. "Basically, the idea is that instead of us sitting down saying ?what is a good directional model to implement?' we give that test to the computer right off the bat," says Sokasian.
The team first gives the computer a set of components that might be useful in developing trading models. But the computer determines what combination of these functions it is going to incorporate into the model.
"At first it does this randomly because it has no clue as to what a good model is or what a bad model is," says Sokasian. "So, at the click of a button the computer puts together a population of 200 or so models." He adds: "The trick, and where ?genetic' comes from, is that this group of models is first trained on a historical set of market data from a particular instrument. And, through a pattern of algorithms in genetic programming, it cuts off the weakest ones in terms of performance and only allows the best ones to propagate forward into the gene pool."
The idea is that, during this historical training period, only the models that seem to have a good combination of the components will be the ones the team employs.
Only time will tell and Sokasian believes "a fair representation of our performance will come six months from now."