Aistis Raudys was introduced to the trading world as a data scientist working in the asset management department at Deutsche Bank in London. He holds a PhD in Artificial Intelligence, which he earned from Vilnius University where he now lectures. He also holds an M.S. in Software Engineering.
Raudys would go on to take a position with Societe Generale Asset Management in London. He worked within the group’s quantitative/systematic commodity trading advisor (CTA) . “I was developing trading strategies; optimizing and calibrating them,” Raudys says. “At some point I was not agreeing with how we approached the process.”
Raudys is a data scientist and says the scientific approach calls for you to have proof that some trading systems are working on unseen data.
“They weren’t doing it in the scientific way,” Raudys says of his work with Soc Gen. “I wanted to approach trading in a more scientific way. Build strategies and test them on a completely unseen environment to see if they still made money.”
So Raudys went out on his own and began building trading systems, thousands of them.
At Soc Gen, he was working on building longer-term technical systems that mainly built trend following strategies similar to what a lot of the big CTAs have done. On his own, Raudys looked at much shorter time frames and focused more on mean reversion strategies.
He began trading his system for friends and family in Lithuania around 2008, and in 2014 launched Chicago-based CTA Automaton along with Thierry Rabut. The Diversified Program has earned 16.3% since its inception and is up 8.65% year-to-date through August.
Automaton has six core strategies that include a trend following, countertrend and mean reversion approach. But he has literally created thousands of sub-strategies he runs all of the time on 48 of the most liquid futures markets.
“Each strategy can trade 48 instruments, Raudys says. “We try different combinations to see which ones work the best and then select the best ones to build a portfolio [of live strategies].”
He constantly tests the performance of these sub-strategies and selects 400 from the thousands he has created based on testing results and their correlation to each other. “Some strategies started working after the 2008 credit crisis, so you run the simulation and see they were not making money before that and the market changed and some new market inefficiencies appeared which can be used to make money,” he says. “I have some strategies that had been working well until 2011 and then we had the European debt crisis and they don’t work anymore. The market is changing and you can see that with the performance of different strategies.”
“You have a lot of different combinations of the logic, of the markets and of the parameters—thousands of them but I can’t trade all of them,” Raudys says. “I go into this big pot of strategies and select [specific strategies] based on performance and the correlation.”
The program, which holds positions from one to three days is extremely active with 3,000 trades a month, or roughly 9,500 trades a year per million. This makes execution vital to its success, and Raudys, a data scientist at heart, has created an algorithm for that.
Raudys researched executions, looking at whether it would be better to execute actively or passively? What order types would provide the best execution? Should it be aggressive and use market orders? Should it be very passive and use limit orders?
“We did loads of computer simulations and realized that if we can be as passive as possible, that limits slippage,” Raudys says. So he built an execution layer that takes all the trading signals and automatically sends the order to the exchange matching engines in the most advantageous way.
“At the very bottom, the strategies send the order to the execution layer and the execution layer; if it needs to, matches the trades if they are offsetting each other ; if not, it tries to execute them slowly, gradually and as passively as possible,” he says. “It tries to use limit orders. In all our compilations we calculated that it is more profitable this way than jumping straight into the market.”
Raudys says that his scientific background allows him to optimize his strategies in one day as opposed to a month when he was working at large institutions.
“Because I implement everything fast and efficiently, it allows me to do the research cycles quickly and test different ideas faster than the other guys,” he says.