Recent technological advances – notably in AI – have put hedge funds on the cusp of next-level investment strategies. Is 2023 the year the industry achieves ‘true’ AI? Or has that already happened?
This article first appeared in the April 2023 Technology Insights Report
Recent technological advances – notably in AI – have put hedge funds on the cusp of next-level investment strategies. Is 2023 the year the industry achieves ‘true’ AI? Or has that already happened?
Earlier this year, the new CEO of Bridgewater Associates used a telling phrase to describe the increasingly important role technology, and artificial intelligence (AI) in particular, was playing at the world’s largest hedge fund manager. “Evolve or die,” said Nir Bar Dea. “That’s what’s happening here.” AI, he suggested during the interview with Bloomberg earlier this year, was going to help the $140 billion firm improve returns, increase profitability, and keep pace with its peers. Because “just doing what we’ve been doing, isn’t good enough.”
In some ways, even Bridgewater is playing catch up. Renaissance Technologies, the world’s second largest investment firm by hedge fund AUM, has been generating returns from machine learning techniques for many years; Citadel has a series of AI initiatives to its name and is currently seeking a companywide license for ChatGPT, the open-source AI ‘chat bot’ launched in November; while Man Group, Europe’s largest investment manager by hedge fund AUM, has developed proprietary database technology – ArcticDB – to drive investment decisions across its entire front office, including product innovation.
“At Man Group, we’re often dealing with asset classes that encompass an enormous universe of individual instruments – hundreds of thousands, in some cases – and the ability to represent such a large universe, and its evolution over time, in a single prime dataset and make decisions around that data quickly is invaluable,” Gary Collier, CTO of Man Alpha Technology and head of the team behind ArcticDB, tells Hedgeweek.
Collier and his team’s work on the technology that would become ArcticDB began in 2013. Ten years later, the data needs of the wider industry have evolved in such a way that Man has launched ArcticDB as a product.
Engine, fuel
For the hedge fund industry at large, a story that was first about data has been steadily developing into one in which AI plays a fellow protagonist.
“The two are inextricably connected – the technology is the engine, and the data is the fuel,” says Neil Chapman, CEO of Exabel, an Oslo-based front office technology firm that specialises in AI. “On the one hand, however powerful your engine, a Ferrari with no fuel is going nowhere. But on the other, you may have the highest-grade fuel and not have an engine that can burn it and turn it into something useful.”
Of course, the use of AI by hedge funds is not new. Eurkeahedge’s AI Hedge Fund Index dates back to January 2010, and outperforms the data provider’s main hedge fund index over that time (see fig. 1.1). A recent survey by market intelligence specialist Market Makers, meanwhile, suggests most of the top 50 hedge fund managers will use AI to achieve portfolio returns this year.
But there is a growing sense that advancements in the field over the past 12 months, headlined by ChatGPT, represent a turning point for the way AI is applied.
Speaking on a recent episode of the Goldman Sachs Exchanges podcast, Kash Rangan, head of US software coverage at the investment bank’s research division, said “this new chapter of AI” was exciting because it’s “not just about prediction, but it’s about generating”. Notably, today’s AI can generate code.
Hedgeweek research from Q1 2023 shows that hedge fund managers are utilising disruptive technology – including AI/machine learning, big data, and blockchain – in their quest to generate performance alpha much more than they are within their operational functions.
Some 40% of all hedge funds surveyed now employ AI/machine learning to improve their investment performance (see fig. 1.2). About half employ big data and one-third use blockchain. Only 16% said they were not using any of the three technologies to improve their investment strategy.
By manager AUM, our survey sample suggests most established hedge fund firms – those with at least $1billion in AUM – use AI/machine learning for performance alpha, as do nearly half of firms with AUM <$250m (see fig.1.3). Intriguingly, none of the mid-sized firms surveyed do (AUM $250-999m), potentially belying a group that possesses neither the resources of established firms nor the nimbleness and tech-savviness of start-ups.
For many hedge fund managers, the lack of uptake is a matter of inaccessibility to the technology rather than disinterest in, or distrust of, the concept of AI.
Thomas Schmiedel, CTO and co-founder of AI technologists Rubinstein & Schmiedel (R&S), reports a “far greater openness” towards AI among the hedge funds he speaks with than traditional asset managers. “Hedge fund professionals are more agile in their approach and understand the ideas much earlier than others,” he says. “From what we have seen, most asset managers are not up to speed – Excel is their tool of choice. They are not yet at a level where they could consider themselves digital or AI enabled.”
‘True’ AI
Unsurprisingly, the adoption of AI and machine learning techniques by hedge funds differs by managers’ investment strategy style and technological sophistication.
According to Hedgeweek’s manager survey, 69% of systematic hedge fund firms have adopted AI/machine learning processes to generate alpha, compared to 13% of traditional discretionary hedge funds (see fig. 1.4). More than one-third of tech-savvy discretionary hedge funds employ machine learning techniques in investing, as do 42% of hybrid strategies.
Combined with these groups’ tentative steps into blockchain, this reflects a broad patchwork nature of AI adoption in the industry – one that has generated ambiguity and an increasing amount of debate.
Adrian de Valois-Franklin, CEO at Castle Ridge Asset Management, a Toronto- and New York-based firm which runs a fully machine-based, self-evolving AI market neutral hedge fund, believes there are essentially very few “pure play” AI/machine learning hedge funds, designed by AI from the ground-up, globally.
“While more-and-more hedge funds have started to claim the use of AI for marketing purposes, it is often applied as an afterthought to strategies that were designed by human portfolio managers,” de Valois-Franklin says. “Adding further confusion, many firms now quote a ‘quantamental’ approach – this often translates to a glorified AI screening tool, with humans overriding the AI decisions due to a lack of confidence or transparency in the system.”
Technologists have sought to assist by defining clear ‘levels’ of AI sophistication.
R&S subscribes to the idea of three levels of AI in investment management – with the top level reserved for AI strategies using “deep reinforcement learning” (see boxout). The Switzerland- and Germany-based firm achieved what it calls Level 3 in 2019 through its ‘Marvin’-family of AI trading models, which recently moved from digital assets to equities, with commodities next, “on the path of becoming a fully AI-managed multi asset fund,” says Schmiedel.
Some believe this type of product would be a first for the investment management industry. Others disagree, saying ‘Level 3’ products of this ilk utilising ‘deep reinforcement learning’ have already been brought to market. Of course, it is difficult to measure the claims of any firm working with such sophisticated – and secretive – techniques. It is notable that some managers suggesting the existence of ‘Level 3’ products use hierarchies of AI extending to four levels or more.
Time will tell if the debate over the term ‘AI hedge fund’ has hampered progress, especially with allocators.
Either way, Schmiedel suggests hedge fund firms are well placed to take AI forward, possessing the type of “contrarian thinking” required to adopt unfamiliar technologies. In the recent past, it has seen them embrace algorithmic trading. In the not-too-distant future, it should see them embrace ‘true’ AI – investment strategies without the safety net of meaningful human involvement in the decision-making process.
