Walk into an investment firm today, and you'll likely find it crowded with analysts poring over documents and investment managers weighing which stocks to add to their portfolios. Chances are, their days are numbered. Soon, computers enhanced with artificial intelligence (AI) will do much of the stock picking, while fund managers merely ensure the machines run the right algorithms.
Numerous fund companies are already turning to AI in the hope it can deliver better returns than human stock pickers. The systems identify patterns in pricing data, yield curves, how markets execute trades and much more, and then make predictions based on those patterns. Bridgewater Associates, the world's largest hedge fund, said last year that it will replace many of its managers with machines. Research firm Optimas estimates that by 2025, AI use will cause a 10 per cent reduction in the financial services workforce, with 40 per cent of those layoffs in money management operations.
Andrew Dassori, founder and chief investment officer at New York's Wavelength Capital Management, has been using AI in his Wavelength Interest Rate Neutral Fund since 2013. He mostly applies it to the fixed-income side, but is experimenting with stock-picking algorithms on the fund's equity portion too. "The computer can analyze more data than any individual can," he says. "It can look at areas where there's new and rapidly expanding data, like around market sentiment, and measures of buying and selling pressure."
AI also surpasses humans in its powers of prediction. It can determine if one stock or bond is likely to perform better than another based on factors ranging from past returns to weather patterns to who uses a company's products where and when. Dassori created a model seeking out securities that have a specific combination of "carry" (a security's dividend or yield), momentum and value styles. After the 2016 U.S. election, the algorithm suggested he buy a five-year treasury bond instead of a 10– or 30-year bond. It was right, he says.
While Dassori's fund uses "supervised" technology–meaning he monitors and sometimes adjusts what the computer picks–he's also testing an unsupervised version in which he inputs his parameters and lets the machine do the selection. If that takes off, he knows his function as a fund manager may change. "Individual security analysis is done better by a machine," he concedes, though he adds, "People need to test and validate the process."
At London-based Man AHL, an investment industry leader in AI adoption, algorithms look at two-month trends in the market to help determine the best move. AI might decide, for example, to buy into a dip when others might sell, or vice versa. "AI might say, I've seen this pattern before, so I'd like to sell on the small downturn and buy the dips as the market starts to rise again," explains Larry Kissko, a client portfolio manager. "It uses what it's learned from the data to do things slightly differently than human intuition [might suggest]." While the system currently analyzes mostly market-related data, Man AHL may incorporate language from analyst reports, demographic trends and other analytics.
Whether AI will perform better than humans remains to be seen. Both Kissko and Dassori believe the technology is improving their returns but won't say by how much. And dependence on machines comes with risks. Since algorithms rely mostly on historical data, something that's never happened before could trip them up. "What if you train these things and suddenly the model breaks?" asks Davyde Wachell, founder of Responsive, a Vancouver-based company that has created an AI-enabled wealth management program.
The fact that people program AI to analyze specific information is also a factor. What if they tell it to focus on the wrong data sets? And what happens if everyone's AI is looking at the same things? "The market would become perfectly efficient," says Sebastien Betermier, a finance professor at McGill University. "There would be nothing to predict anymore."
However AI ends up affecting the markets, the traditional fund manager's role is set to change dramatically. Many of those who keep their jobs will be charged with creating investing models and ensuring algorithms function properly. "They'll be overseeing and validating what the machine is doing," says Dassori.