Fund manager who traded for Cohen trains algos to copy his brain
(Bloomberg) -- Since he left his job as hedge-fund manager last year, Muhammed Yesilhark has spent his days teaching computers to pick stocks like he did in his 12-year career.
By the end of the year, the former head of European equities at Carmignac Gestion SA, who previously traded for billionaire Steven Cohen and hedge fund York Capital Management, says the algorithms will be ready to help run money for clients of his own investment firm, Q2Q Capital. Yesilhark, who will run the fund from Dubai and London, said he has raised $100 million and aims to gather $800 million in total from 40 investors.
The computer models he designed are meant to accelerate idea generation for Yesilhark’s small investment team and learn from mistakes. More than 40 of them currently trawl through troves of data, scan real-time trading patterns and read company releases using natural language processing techniques to spot the perfect trading opportunity. It’s an area that requires a heavy investment of time and money and is dominated by top firms, including Cohen’s Point72 Asset Management.
“He has certainly got a nice pedigree and the idea is a great one, but the difficulty is in the implementation of it,” said Jacob Schmidt, chief investment analyst at NLP Financial Management, which invests in hedge funds including quant strategies. “It’s a question of money and manpower. All the big guys are doing it. They have huge teams.”
Cohen’s Point72 Asset Management, which invests his $11 billion fortune, is testing models that mimic trades of its portfolio managers. He is also experimenting with automating the work of its execution traders, who place buy and sell orders with brokers on behalf of money managers. Ray Dalio’s Bridgewater Associates is developing algorithms based on employee data to help automate decision-making at the world’s biggest hedge fund.
“More than 50 percent of what we do is repetitive," says Yesilhark, a married father of two. “The best way to solve the problem is technology."
Using computers to automate some of the investing process is a broader trend among hedge funds, who are looking for a high-tech answer to years of underperformance and dwindling fees. Some firms are even turning to computers to identify trading signals themselves and improve on their own, using artificial intelligence.
Such funds, known by the acronym AI, have yet to prove themselves in the long run and on a broader scale. The Eurekahedge AI Hedge Fund Index, which tracks 12 of these money pools, has outperformed hedge fund peers since 2013 but trailed the S&P 500 Index.
The trend toward artificial intelligence is “real but it depends on the resources that one has to spend on technology and hire the best people to retain an edge," said Michele Gesualdi, who oversees $3 billion as the chief investment officer at Kairos Investment Management, which invests in hedge funds. “It’s difficult for a young firm to do this and compete."
Yesilhark says his models, while less ambitious, have shown encouraging performance in the short span he’s used them to find trades. Since September, when the money manager started testing them with his own capital, they returned 38 percent, helped by winning bets on Syngenta AG, seeds giant Monsanto Co. and battery maker Blue Solutions SA. He declined to provide details of his models.
“They never sleep, do not go to toilet, have no girlfriends or boyfriends,” says the 37-old, who was born and raised in Germany. “I think artificial intelligence and investing is a big opportunity."
Learning From Cohen
Yesilhark joined Carmignac in January 2014 and left in March 2016. The Carmignac Portfolio Grande Europe fund that he co-managed gained 10 percent in 2014 and lost 1.4 percent in 2015, according to Carmignac’s website. He also co-managed the Carmignac Euro-Entrepreneurs fund, which gained 10 percent in 2014 and almost 12 percent in 2015, according to their website
The idea to clone his mind was sparked by his trading days at SAC Global Investors, Cohen’s previous investment firm, where Yesilhark was a partner before moving to Carmignac Gestion. Once his models, which can automate as much as 90 percent of the decision-making, detect a potential trade, Yesilhark makes the final call on whether to buy, sell or reject it. His new analysis is codified again and the machine learns not to repeat the same mistake.
The fund will not charge a management fee but a fixed fee for expenses, as well as take a 25 percent cut of the returns generated. The performance fee will support the bonus pool and a charitable foundation that Yesilhark has set up to back not-for-profit organisations.
The money manager said he will not be part of the firm’s bonus pool, using it as an incentive for his staff to work for the firm and help him hire the best talent in the industry. He expects to make money solely from returns on his own investment.
“This industry is ripe for disruption in a major way," he said.