In an attempt to see how close technology really is to replacing Wall Street’s army of analysts, pundits and money brokers, Boomberg challenged ChatGPT, the artificial intelligence tool taking the internet by storm, to create a winning portfolio for the US stock market
26 January 2023 20.28
What happens when you ask the world’s most popular artificial intelligence tool to design an ETF (exchange-traded fund, and define a mutual fund whose primary characteristic is that it is traded on secondary stock markets) that can beat the US stock market? He tells you the same thing every frustrated inventory manager does.
The result: a classic watchdog exercise, with the tool explaining that the market is too unpredictable to design such a fund, while advising on the need to choose investments that align with our goals and appetite for risk.
This was the full response when we instructed ChatGPT to “design an ETF to beat the US stock market and tell us what stocks are in it.”
Mark one for humans. It seems that despite all the hype, AI isn’t ready to take the world of stock picking by storm just yet.
On the other hand, maybe ChatGPT knows the secret to beating the market, but is smart enough not to reveal it? There are already AI-guided investments all over Wall Street, including in the ETF arena, and some are winning the market right now.
A current standout is the AI Powered Equity ETF (ticker AIEQ), a $102 million vehicle that has returned about 9.9% in 2023 as of Wednesday, compared with 4.7% for the S&P Total Return Index. 500.
AIEQ uses a quantitative model that runs 24/7 on IBM Corp.’s Watson platform to assess more than 6,000 publicly traded US companies each day.
It collects regulatory documents, news, management profiles, sentiment indicators, financial models, valuations and more.
The product, developed by EquBot LLC and overseen by ETF Managers Group LLC, can rapidly change positions and exposure levels, making it a sentiment barometer for observers.
He entered 2023 with a mixed assignment. Major holdings currently include home furnishings firm RH, Las Vegas Sands Corp., sustainable energy company Constellation Energy Corp. and JPMorgan Chase & Co.
Performance analysis shows that the ETF’s consumer discretionary holdings, including shares in companies like Caesars Entertainment Inc., Kohl’s Corp. and meme-stock darling GameStop Corp., have been the biggest driver of performance this year.
However, extend the time horizon and AIEQ’s knack for beating the market is undone. Since its inception in 2017, the ETF has delivered around 41% to investors, according to data compiled by Bloomberg. The S&P 500 Total Return Index has returned more than 72% in the same period.
It works best when you can catch boost names in the grow space, said Jessica Rabe, co-founder of DataTrek Research. “He struggled to find momentum names in a highly volatile stock market last year, and when he’s had the best track record, it’s been during bull markets when he favors tech names.”
So maybe ChatGPTChatGPT was smart to refuse to try to beat the market. To give it another go, we asked the tool, like other tests of ChatGPT’s capabilities with hypotheses, to help with a different, never-ending quest for money management: an investment that offers clear diversification from the broader market.
This is what we got when we told ChatGPT to “design an ETF to deliver a return uncorrelated with the US stock market.”
A multi-asset approach, mixing a few alternatives. Not a bad result, according to Eric Balchunas, a senior ETF analyst at Bloomberg Intelligence, even if history shows that human investors tend to like to separate their asset classes.
This is straight out of the institutional playbook, Balchunas said. These are strong recommendations for asset classes that provide uncorrelated returns. This is what most institutional investors invest in. It is clear that one reads in books.
In this response, ChatGPT points out that it is difficult to create an uncorrelated ETF, as there is often some level of co-movement. He urges that any portfolio be chosen through market analysis and, in good news for the financial community, he once again encourages us to speak to an advisor.
Acknowledging the limitations of ChatGPT is an important caveat for our informal experiment. The tool is language based and optimized for dialog; it was not designed to predict markets. OpenAI, the company behind ChatGPT, is transparent about its limitations, such as its “limited knowledge” of anything after 2021.
Since the tool won’t provide us with a new machine-made portfolio in detail, we tested the next best option and asked it to name “best AI-powered ETF.” But for some reason, ChatGPT struggles to identify any, even though AIEQ has the words “AI Powered” in its name. And while it says “there are some ETFs that use artificial intelligence (AI) as part of their investment process,” the tool doesn’t mention any.
If I were naming names, I’d probably mention the $419 million WisdomTree US AI Enhanced Value Fund (AIVL), one of the largest. Along with its sister fund, the $82 million WisdomTree International AI Enhanced Value Fund (AIVI), underwent changes a year ago to incorporate AI and machine learning into its strategy and name.
AIVL has returned around 0.8% last year, while AIVI has lost 2.6%, compared with a 6.1% loss for the S&P 500 Total Return Index. Mixed performance against the benchmark in 2023, with gains of 3.7% and 7.9%, respectively.
Matt Bartolini, head of SPDR Americas Research at State Street, says the use of AI means the fund can analyze a much larger pool of potential investments than humans alone can manage.
AI looks at thousands of pages in mere seconds, Bartolini said. You can increase productivity at the coverage level than with a more humane approach.
To end our experiment, we decided to be direct. We bluntly told ChatGPT to “explain whether artificial intelligence can choose actions better than a human.” This is what we got, in full:
So it seems there is still hope for the humans on Wall Street.
—Sam Potter and Katie Greifeld, cwith the help of Matthew Miller, for Bloomberg