Yahoo Finance Ex-DeepMind Trio Bring Algos, AI Poker Prowess to Tower Research
(Bloomberg) -- A systematic-trading startup founded by three alumni from Google DeepMind is bringing its algorithms to Tower Research Capital — including a little AI poker-playing prowess.
Most Read from Bloomberg
Wall Street Billionaires Are Rushing to Back Trump, Verdict Be Damned
Donald Trump Becomes First Former US President Guilty of Crimes
World’s Largest Nuclear Plant Sits Idle While Energy Needs Soar
EquiLibre Technologies agreed a deal late last year to contribute exclusively to the New York proprietary trading firm, Tower’s Chief Investment Officer John Cogman said in an interview.
That makes the Prague-based firm one of numerous quants working with Tower, a high-frequency market maker that sources trading signals from different independent teams. But EquiLibre stands out thanks to its founders’ pedigree.
All three have doctorates, worked at IBM and collaborated on a system at Alphabet Inc.’s AI subsidiary DeepMind designed to play a wide range of games. Two of the trio co-authored one of the first algos to beat humans in poker, and none had ever worked in finance before starting EquiLibre.
“This was a very exciting opportunity to apply these deep reinforcement learning techniques to financial markets,” Cogman said. “Bringing in people that have looked at similar problems but in different domains and then try and apply them to finance — we see that as very attractive.”
Wall Street quants have long dabbled in a branch of AI called machine learning that can comb through large data sets for complex patterns. The EquiLibre team specializes in reinforcement learning, a technique typically used in games where agents learn to maximize rewards through trial and error. The most Cogman would say about EquiLibre’s algos is that they’re broad enough to be applied to a variety of assets.
Yet applying Silicon Valley’s most cutting-edge techniques in finance has often proved challenging. Financial data are noisy and limited, and any patterns you exploit are prone to change as traders compete and economic regimes shift.
“There’s so much more noise that you cannot just simply take off-the-shelf learning algorithms that people apply in say vision or games or speech recognition and naively apply them,” said Martin Schmid, one of the EquiLibre co-founders.
Launched in in 2022, the startup took more than a year to assemble a prototype, then sought out a partnership to avoid all the duller parts of finance, including regulatory compliance and data access. Under the arrangement with Tower, EquiLibre will continue as an independent company but will not trade on its own, Schmid said.
Only about 10% of EquiLibre staff of around 25 have worked in finance, he said, while the rest come from tech or academia.
“What we have seen with other companies out there is they often see that there’s an AI revolution happening, but all they really do is they hire a few people and give them a few months to try,” said Schmid, who has a PhD in algorithmic game theory from Charles University in Prague. “Then they get to know it is actually really, really hard to get it to work.”
Most Read from Bloomberg Businessweek
Disney Is Banking On Sequels to Help Get Pixar Back on Track
The Secret Ozempic Recipe Behind Novo’s Race to Boost Supplies
US Malls Avoid Death Spiral With Help of Japanese Video Arcades
Why Dave & Buster’s Is Transforming Its Arcades Into Casinos
©2024 Bloomberg L.P.
