AI isn’t ready to take fund manager jobs … yet
How good is artificial intelligence at managing money? To judge by the recent performance of some AI-driven strategies, it doesn’t look like the robots are going to take over from the humans anytime soon.
In August 2018, a quantitative team at Aberdeen Standard Investments started a $10 million Artificial Intelligence Global Equity Fund, betting that an algorithm can be more effective at figuring out the complex world of factor investing than a human portfolio manager. A year later, the fund had underperformed the broader stock market’s powerful rally, and its assets had grown only 8%. Institutional investors say they’ll hold off committing money until they see a longer track record.
AI has penetrated almost every area of our lives, from online customer support to facial recognition to self-driving cars. But investing is proving to be one of the toughest challenges for machine learning.
The main problem is financial market data, according to Bryan Kelly, head of machine learning at $194 billion AQR Capital Management. Market data — unlike photos or road traffic information or chess games — is finite, and the algorithms can learn only from past performance. “This isn’t like a self-driving car where you can drive the car and generate enormous amounts of additional data,” Kelly says. “The dual limitation of very noisy data and not a lot of it in financial markets means that it’s a big ask to want the machine to identify on its own what a good portfolio should look like without the benefit from human insight.”
People who try to predict the stock market or interest rates using AI might end up with flawed analysis that can lead to financial losses, warns Seth Weingram, director of client advisory at $97 billion Acadian Asset Management. “You see market-naive folks who are trying to apply these techniques get into trouble,” he says. “There’s a risk that you don’t actually have enough data to meaningfully train your algorithm.”
What’s being touted as a revolution has been used by quantitative whizzes for years. Almost all quant funds use machine learning to sweep through social media, news articles and earnings reports.
PanAgora Asset Management, a $45 billion quant fund based in Boston, has been creative in using natural language processing to analyze Chinese equities. Its machine-learning tool spiders through online forum posts by retail Chinese traders and identifies cyber slang words they use to avoid government censors, who might crack down on negative language, such as discussions of poor earnings results. Canny Chinese bloggers, for example, replace the word “rubbish” with a phonetically similar expression, “spicy chicken.” PanAgora’s model identifies such similar-sounding words and the context in which they appear to gauge sentiment about Chinese companies.
Naureen Hassan says tools in development will make advisers "faster and smarter" in serving clients.July 13
PanAgora is also looking at using AI to execute trades and spot accounting abnormalities that a simple analysis wouldn’t find. “We have tons of data [on the execution of trades], and now instead of making all these individual decisions using anecdotal evidence from the trading desk, we can make a much more quantitative decision given past results,” says George Mussalli, equities chief investment officer at PanAgora.
One reason Aberdeen Standard and others are turning to robots for help is the recent market environment. Investors are fretting over the end of the bull market as trade tensions and an inverted yield curve flash warning signs for global growth. But they’re afraid to exit too early and miss out on late-cycle returns.
Yet swings in investor sentiment are hard for machines to navigate, too. “If the market becomes unpredictable, it’s always more challenging for AI,” says Anand Rao, global artificial intelligence lead at consulting firm PwC. “This time around, there are different forces acting. But [the collapse of the credit market bubble in] 2007 was also very different, and so was [the end of the dot-com bubble in] 2000. With more data and more history, AI funds will get better.”
So far, machines seem befuddled by these markets. After outperforming the Hedge Fund Research HFRX Equity Hedge Index in four of the last five years, Société Générale’s long-short U.S. stock index based on a machine-learning model has been lagging this year, with a return of less than half that of HFRX. The Eurekahedge Artificial Intelligence Hedge Fund Index, which tracks hedge funds that use machine learning, has also underperformed in 2019: Its gain of 2.3% through Aug. 31 trailed the 6.9% return for the broader HFRX Index.
“This year has been challenging for AI funds. It’s probably the first time in the history of the U.S. that you have a president tweeting these kinds of things,” says Nicky Indradi, an analyst at Eurekahedge. “If you give AI funds more time to better understand the technology they’re using, I’m optimistic that they’ll be able to perform well.”
One of the Socgen Index’s creators, Andrew Lapthorne, says the robot’s strength is in feature recognition: picking biotech stocks that have a better chance of outperforming, for example. But he also warns that the strategy needs time to develop before it can be offered to a larger number of clients —about $39 million of assets currently track SocGen’s machine-learning strategies. In addition, he says double-digit returns are unrealistic.
Nobel Prize-winning economist Robert Shiller is also tempered on AI’s prospects. “I think we still need human oversight,” Shiller says in a London interview. “AI can really mess up when reacting to text because a word may have a new meaning or there could be a typo — we would recognize it as a typo, but the machine might not.”
Still, Boyan Filev, co-head of quantitative equity at Aberdeen Standard, says the advantage of utilizing machine learning to manage a portfolio is that it adapts to the market and improves over time. The fund’s underperformance, he contends, is mainly the result of challenging markets and changing behavior of so-called equity factors, which have led to losses at many quant funds in 2019.
“Our fund was positioned more defensively at the start of the year in line with the bear market of the end of 2018. However, the sharp reversal of equity markets this year hasn’t been particularly helpful,” Filev says. “A more stable and slower-evolving environment is more beneficial to our product. Very sharp reversals in market directions are very hard to position against in the short term.”
Filev expects the fund to adapt to conditions, he says. It just hasn’t done so yet.