From the December 01, 2011 issue of Futures Magazine • Subscribe!

GDP and the science of trading

Trader Profile

David Oros already had built the infrastructure of his alternative investment program used in the commodity trading advisor (CTA) he founded, Global Domain Partners (GDP), when he decided to bring in Jacob and Joshua Vogelstein to work on more advanced algorithms.

Jacob Vogelstein came into the futures and investment business cold. "I came in from the perspective of ‘Hey, this is a smart guy. Can we do something cool?" Jacob says.

He and his brother were working on their PhDs when they got the call. "Dave [Oros] contacted my brother and me at the [business’s] inception. We had personal connections with Dave and people who started the fund," he says.

The algorithms the brothers apply to financial markets for the Baltimore-based CTA were drawn from their research in neuroscience and biomedical engineering, research that the brothers continue to work on at Johns Hopkins University, where both received PhDs. Their reseach is making headway in creating advanced prosthetics for injured servicemen.

"Originally the query was, ‘What do you know from your research that would be useful in the context of financial markets?’" Jacob says. "We could have explored a whole number of different approaches, but the one that was most intuitive to us is the one that would give us the best competitive advantage. That was to draw inspiration from our research in neuroscience."

Basically, Jacob is working on translating the communication from the brain and encoding it. "My specific research is on brain computer interfaces, it is a device that allows you to extract information from the brain and use it to control a computing system."

He points out that some of the strongest algorithms and research happening in science are in this area.

"We have come to understand that the brain is probably the best pattern recognition computing system that ever has been created. To the extent that our strategies are finding patterns in historical price data, we thought what better place to look but the brain for how one can efficiently and reliably track visible patterns from past events and use that to predict the future, which is what our brain does every day," he says.

GDP Director of Trading Jason Frost already had conventional signals before they launched their medium-term pattern recognition Vector program in 2006 but basically used those to test their portfolio management and position sizing models. They wanted to do something unique. "Jacob knew about the project and what we were trying to do and made the connection that the algorithm he was working on was something we could plug in," Frost says. "We needed some unique kinds of forecasting. We needed the special sauce and that is where Jacob and Joshua came in."

The brothers immediately improved on the placeholder models. "We originally were using a more conventional regression approach and [Jacob] came and said, ‘You are skipping the magnitude information from your forecast, so you are just doing a classification. Why don’t we use one of these advanced classification algos from our research?’" Frost says. Jacob immediately knew what tools to use. "I have this algo that does something different in our advanced prosthetics research. There is no reason this shouldn’t work better than what you are doing," he said.

The system produces a signal in 17 markets including calendar energy spreads in the reversal program. Each signal is an aggregation of 40 unique models, which are chosen from a larger group of 1,500. Each day the program creates four new models and retires four old ones from the group. The uniqueness of the models was apparent this summer when they captured a sharp reversal in gold at the end of August along with the subsequent rally. "I don’t know of any other model that caught that move," Frost says.

The program is in every market it trades all of the time. Each day a signal is produced that predicts the highs and lows for each market over the next five to 15 days and reacts accordingly. It is based on patterns over a sliding window of six years.

The program targets very low volatility, producing a compound annual return of 4.70% with a worst drawdown of 8.71%. In April 2009 they launched their Sigma series with twice the leverage, which has produced a compound annual return of 7.61% with a worst drawdown of 11.2%. Both programs are up slightly in 2011 through October after a couple of down months.

On his research Jacob says, "You think thoughts faster than you type them, faster than you speak them, you know what you are going to do in advance of actually physically executing the movement. To that extent your body is slowing you down."

Nothing is slowing down the Vogelsteins’ research, whether they are applying it to trading or things even more important.

Comments
comments powered by Disqus