AI and ETFs: the machines are coming (but not always winning)

The machines are coming. Artificial intelligence can write songs, converse, and code — but can it supplant investment managers? While a growing number of AI-driven ETFs are trying to prove that it can, many are still struggling to outperform funds whose shareholdings are determined by humans or by companies’ market values.

Several AI-powered ETFs have been launched with much fanfare over the past few years. Like mainstream ETFs, they are funds listed on a stock exchange that give investors exposure to a diversified portfolio of securities. But, unlike their non-machine counterparts, they use machine learning, sentiment analysis and natural language processing to identify patterns and trends that can help them select assets.

Notable examples include South Korean Qraft Technologies’ three offerings — US Large Cap ETF (QRFT), US Large Cap Momentum ETF (AMOM), and the US Next Value ETF (NVQ), as well as the VanEck Social Sentiment ETF (BUZZ), EquBot’s AI Powered Equity ETF (AIEQ), Merlyn. AI’s Bull-Rider Bear-Fighter Index (WIZ) and its SectorSurfer Momentum ETF (DUDE), plus WisdomTree International’s AI Enhanced Value Fund (AIVI).

So far, however, the performance of AI-driven alternatives has been inconsistent.

Over a three-year period to April 19 2023, the conventional index-tracking SPDR S&P 500 ETF (SPY) has delivered returns of 14.8 per cent. QRFT is close behind, with returns of 14.5 per cent. Its other offering, AMOM, has delivered 12 per cent, while AIEQ managed 4.4 per cent, WIZ 6.4 per cent and AIVI 11.9 per cent.

Over a one-year period to the same date the picture is more mixed, with AMOM and AIVI both beating the SPY and the rest falling behind.

“These ETFs are using artificial intelligence to help pick stocks in hopes of outperforming the broader market but, as decades of research has shown, it is hard for active management of any kind to beat a low-cost index based approach,” observes Todd Rosenbluth, head of research at data provider VettaFi.

Experts say that, when picking stocks, AI can struggle to track trend which may not show up in past data, company reports and news media that it analyses using natural language processing.

“We live in a very complex world, and human intuition does add value,” says Joseph Byrum, chief technology officer at Consilience AI. “I can automate the reading of documents for every company in the universe. But what you can’t train a model for is looking at the way central banks are behaving — there’s a level of idiosyncratic risk I don’t think you’ll ever be able to model.” 

[AI] can cover a broad spectrum of stocks really quickly . . . we don’t have HR, it works weekends, it doesn’t take holidays

As a result, AI-based systems still rely on a degree of human oversight and intervention. Wisdom Tree International’s AI ETF offering allows its portfolio managers to review and veto trades, although the company says this “has not happened a lot”. EquBot has operational controls to help it deal with fake or incorrect financial information and decide how often and what type of securities to trade.

Developers say the AI models are, however, already cutting out inefficiencies in human decision-making.

“They can cover a broad spectrum of stocks really quickly,” notes Christina Bargeron, client portfolio manager at Voya, which developed the proprietary model behind Wisdom Tree’s AIVI. “We don’t have HR, it works weekends, it doesn’t take holidays. It can do this so much more quickly than a human could and it can go extremely deep.”

AI ETFs also cut out human bias and ego from the decision-making process. According to Francis Oh, head of AI ETFs at Qraft, the AMOM fund did not include financials in its portfolio at the start of the year, which helped it to avoid the rout that followed the collapse of Silicon Valley Bank and the turmoil at Credit Suisse.

“As a portfolio manager, I have to decide whether to keep strategies or change them,” says Oh. “Should I wait a few more months or change it right now? That decision can cause distress to the fund managers and investors.”

Human fund managers continue to find it difficult to prove their worth against indices and the conventional ETFs that track them. According to data from Morningstar, in 2022, only 48.7 per cent of US equity funds beat their indices, and only 43.2 per cent of global equity funds did so. The picture is much worse over a longer timeline. From mid-2012 to mid-2022, only 12 per cent of US equity funds and roughly 20 per cent of global equity funds offered higher returns.

And the quality of the competing AI technology is only likely to improve. Experts say AI that can be used for stockpicking is still in its relative infancy but, as companies such as OpenAI and Google pour billions into development, its ability to make sound decisions will improve. Adding more sources of data will also give the AI models better foundations on which to base their stockpicking decisions.

“The data continues to explode,” says Chris Natividad, chief investment officer of EquBot. “Our partners at IBM say 90 per cent of the data has been created in the past few years, and we believe we’re going to be seeing that exact same statement every two years from now.”

This story has been amended to correct Joseph Byrum’s job title.