Humans are emotional, and tend to unconsciously absorb a lot of fallacies and received wisdom, to their detriment as efficient investors. AI, on the other hand, is coldly logical, basing its decisions solely on data.
Plus machines are much better than people at marshalling and assessing huge quantities of information: it’s impossible for a human to consider every possible investment out there, but entirely possible for a machine to do so.
As a result, a number of hedge funds have sprung up that let the machines make the decisions, such as Cerebellum Capital, Rebellion Research, Hong Kong-based Aidyia, and Sentient Technologies, in which Li Ka-shing’s VC fund is a major investor.
How AI works
Broadly, the AI is given access to a wide range of both public and proprietary data, and then works out investment strategies based on which methods of using that data work best. Crucially, the AI can endlessly refine and improve its strategy. It isn’t the same thing as automated or algorithmic trading, where the advantage is speed; here the advantage is computational sophistication.
And it seems to work: AI hedge funds generated impressive average returns during the six years to 2016, according to research company Eurekahedge's AI/Machine Learning Hedge Fund Index, which tracks 23 such funds.
Training the AI
Among the best known is Cerebellum Capital, which was set up eight years ago, prompted by advances in cloud computing and machine learning, and the increasing amounts of data available. The company recently started its first truly automated machine-learning fund. At the moment it’s mainly investing money from the founders and their associates, but the company is negotiating with institutional investors.
The company’s CEO David Andre says the biggest problem in developing the software has been setting too much store by particular pieces of data, an issue known as overfitting. “Most of the development has come down to one core problem: how can you tell when a model has been overfit or not?” he says. “You find things that do terrifically in the past but not so well in the future.”
Everyone’s at it
For decades, so-called quant hedge funds have used data scientists’ complex mathematical models and the processing power of computers to guide investment decisions.
A lot of them, naturally, are moving into this space: the likes of Quantitative Investment Management, Renaissance Technologies and Two Sigma. Andre says their approach is slightly different from his firm’s. “They’re hiring lots of PhDs and starting with intuition from the quant. We want the machine to find the strategy.”
Mainstream financial institutions have also started to dip their toes in the water, including names as diverse as Bridgewater Associates, BlackRock, Point72 Asset Management, UBS and Morgan Stanley. Most, though, are currently using AI as a way of generating ideas rather than guiding investment strategy: 26 per cent of financial companies use it in their decision-making, according to PricewaterhouseCoopers.
In other words, we’re still far from AI being used to manage individual investors’ portfolios. “It’s a surprisingly slow-moving space,” says Andre. “But a lot of people prefer to call up their investment manager and have a discussion. There are trillions of dollars in that, so change isn’t going to happen overnight.”
The future: AI everywhere
Arguably the logic behind AI investment management suggests that the bots should eventually take over entirely. And if it becomes ubiquitous, it could cease to be a differentiator—but possibly, some algorithms are just better than others.
“A lot of the easy machine learning will go away because so many firms are doing it,” says Andre. “And then you’re going to need clever computers or clever people. Our bet is on the clever computers.”
See also: Millennial Money: The Future of Finance