Hedge funds excited on machine learning but may not understand

The number of hedge funds that say they use machine-learning or artificial intelligence has increased significantly over the past couple of years, with a survey from mid-2018 reporting that over half of respondents do. With increased interest in the field, investors have to be weary of funds jumping on the trend. A number of execs at big hedge-fund allocators said they had difficulty conducting due diligence on these funds. Pockets of the industry see the increased interest as an inevitable move to the next wave of investment innovation, building on top of the foundation created by massive quant funds like Renaissance Technologies, D.E. Shaw, and TwoSigma.One of the hedge fund industry's machine-learning pioneers remembers the days — just “three years ago” — when explaining his fund had to be done in the simplest of terms.Now, with more than half of hedge funds using some form of machine learning or artificial intelligence in their investment processes, according to a BarclayHedge report, the “pendulum has almost swung the other way,” says Michael Kharitonov, co-founder of $4 billion Voleon Capital.“We didn't see people from the machine learning community at hedge funds or in finance” a decade ago, he said in an interview with Business Insider. “Now machine learning has become very popular in finance and a lot of people want to work on it.”See more: A new machine-learning tool used by hedge funds to rank their brokers hopes to put an end to the 'old boys network'Voleon was one of the first hedge funds to completely rely on machine-learning systems to trade its portfolio. The fund was founded in 2007 by Kharitonov and Jon McAuliffe, two D.E. Shaw alums with five different college degrees between them.The concept of using machine-learning in any capacity was foreign in finance at that time as machine-learning was still a nascent field, Kharitonov said. The past dozen years have changed that.More mainstreamAs the traditional quantitative space has gotten crowded, much like the fundamental equity space, investors are searching for funds with new techniques to squeeze alpha out of the market. According to BarclayHedge, the average machine-learning fund lost 2% last year, slightly better than the average hedge fund. Through the first quarter of 2019, these funds have posted an average return of roughly 1.6%, below the market and the average fund. Machine-learning and artificial intelligence have made their way into the systems and operations aspect of the business. Citco, which is the fund administrator for more than $1 trillion in hedge fund assets, is using machine-learning to track trades and transactions more efficiently, while Brevan Howard has spun off a company called Aim2 that integrates artificial intelligence into trading. Nomura is reportedly using the Aim2 platform for its fixed-income business.$108 billion investment group Man Group has begun using machine-learning in its trades to determine which clearing platform is most efficient, Keith Haydon, the manager's CIO of Man Solutions, told attendees at the SALT conference this week in Las Vegas. Several in the industry see machine-learning funds as the logical next step from quantitative funds. Specialty investors like MOV37, founded by the CEO of hedge fund investor Protege Partners, focus exclusively on what it calls the “third wave” of investing, while the world's largest hedge fund investor, Blackstone, is watching the space to see what impact these funds have on quant.See more: The explosive growth of quant investing is paving the way for 'super managers' in the hedge-fund industry“I think it raises the possibility of disruption for some [quant] strategies,” said John McCormick, CEO of Blackstone's hedge fund solutions group, at a panel at the Milken Conference in Beverly Hills. “What impact artificial intelligence and machine learning will have on” quant is the next big question for the industry, he added.Automating quantThe biggest difference between traditional quant funds and machine-learning products is that a quant fund still relies on humans to build models, putting a bias into the process, based on how the model-creator views the market. A machine-learning fund further removes the human from the investing process by building systems that then build the models used to trade. The software then updates the models itself, constantly taking in information and data.“In some ways, we are automating quant,” Kharitonov said. “Many of the things quants have humans doing we create systems to do.”Now machine-learning has become a buzzword, investors are putting their money into funds and businesses claiming machine-learning expertise with no way to confirm it.“A little bit of frothiness out there,” he said of the fundraising environment. “I know tech start-ups that have rebranded themselves as machine-learning and now the venture capital is pouring in.”Because these models are not “based on human intuition on how markets are supposed to work,” said McCormick, additional work is required on the investor's part. See more: Silicon Valley has made top data-science talent too expensive for many hedge funds, so they're getting creative to compete“For allocators, it raises a number of difficult questions about how you'll do due diligence on the models that the creators or the humans running the firm have difficulty explaining,” he said. Ray Nolte, the chief investment officer for fund-of-funds SkyBridge Capital, said his firm has not embraced machine-learning funds in part because of the difficulty in evaluating them.“Are they really repeatable over time？” Nolte said. “It's very hard to understand and get comfortable with them.” Another tool for the biggest hedge funds？ To suss out the funds that have limited machine-learning expertise, “you have to look at the people,” Kharitonov said. “You're not going to become a machine-learning expert by taking a three-month online course.”Still, he expects the smartest, and biggest, funds to continue to invest in this type of technology, further pushing up the price of the business in an industry that has become so expensive that launches are at a record low. Larger fund platforms, which have the resources to fill systems with expensive alternative data streams on top of readily available market data, might be able to push out some of the original machine-learning specialists with their scale. Only 15% on the funds surveyed by BarclayHedge that said they used machine-learning or artificial intelligence are over $100 million in assets. While typically smaller funds on average outperform multi-billion platforms, the tide may be turning, thanks in part to the growing cost of doing business. In Blackstone's hedge fund portfolio, the largest funds have been the best performers, McCormick said.It's a trend “we think will continue,” he said.