An initial study suggests that most funds managed with artificial intelligence are lagging far behind standard benchmarks. And they will continue to do so until they understand the language – and reality itself.
Since the advent of Chat GTP nearly two years ago, generative AI tools have been hailed as a technology that is “more profound” than fire, with impacts “comparable in magnitude” to the Industrial Revolution, electricity, and the wheel.
These are big shoes to fill at any point in human history. And while we are in exciting times, it seems that once the initial excitement and hype wears off, most of us will have to settle for little more than sophisticated chatbots and virtual assistants of the same ilk.
Not good with money
As Steve Jobs always said, there is “one more thing,” namely that AI is bad, even terrible, when it comes to investing money. This is according to an opinion and analysis piece published last Friday in Scientific American, which Sam Wyatt And Gary Smith from Pomona College in California.
This should come as a relief to all those wealth managers out there who, just a few months ago, were faced with the challenge of offering more to both ultra-high-net-worth and less-wealthy clients, when a flood of viral posts went viral on social media claiming astronomical – and instant – returns from trading stocks using some vague AI-based tool or algorithm.
Towards a century of underperformance
However, the authors went much further than finews.ch and most others in the media-hype-market complex, by claiming that the technical proponents of artificial intelligence have been “over-promising and under-delivering” since at least 1960, that is, for nearly 70 years.
“It is now increasingly clear that GPT and other LLMs are not intelligent in any meaningful way and cannot be relied upon to make important decisions such as hiring decisions, prison sentences, loan approvals, insurance premiums – and investments,” they wrote.
Honest metrics
Wyatt is a student at Pomona, while Smith Fletcher Jones is a professor at the college and the author of over 100 peer-reviewed research articles and 17 books, the most recent of which is entitled: «The Power of Modern Value Investing: Beyond Indexing, Algos and Alpha» which he together with Margaret Smith.
Both were relatively open and matter-of-fact about the matter, saying that AI-powered investing was “particularly interesting” because it was a way to assess the overarching capabilities of the technology itself.
No errors or bias
They started by looking at AIEQ, the first AI-powered ETF, launched in 2017 and touted at the time as a groundbreaking application because of its ability to replicate a veritable army of oppressed equity analysts 24 hours a day, 365 days a year – while eliminating human error and bias.
Two weeks later, another global AI equity fund called MIND was launched, claiming it could recognize patterns and make decisions in the same way as the human brain, but at lightning speed and without bias, overconfidence, cognitive dissonance or emotion.
The volcano question
For all the Trekkies out there, this is an indirect answer to the long-burning question of whether or not Spock would be a good investor, as this is a burning issue in certain hidden corners of the internet (collected Google search).
The answer is that, just like AI, he probably wouldn’t have become a billionaire, a hundred-millionaire, or even a multimillionaire, and would ultimately have settled for the broadest and cheapest ETF on the market, like the rest of us mortals who invest money.
Far behind
The reason for this is that the authors say that both funds have significantly lagged the S&P 500 since their inception. The AIEQ’s total return is 63 percent, while the S&P’s is 108 percent.
MIND fared even worse. The stock closed 2022, but not before its cumulative return was a sobering minus 12 percent, while the S&P was up 65 percent.
No annoying people
But they didn’t stop there: Both conducted an analysis of newer AI-driven funds, but pointed out that these had not yet been peer-reviewed.
The pair found 11 fully AI-driven funds, such as AIEQ and MIND, where decisions were made without the help of pesky or increasingly less pesky humans. They also found 43 other AI funds where some human involvement was actually possible.
Still poor
An example of the latter was Qraft’s AI-powered US Large Cap Momentum ETF (AMOM), which uses AI to select stocks, although human advisors retain “full discretion in making investment decisions.”
Yet even in this limited population, only 10 of the 43 partial AI funds outperformed the S&P 500, with average annual returns a good five percentage points worse. The all-in AI funds, however, somehow managed to top even that.
“Every single one underperformed the S&P 500. Six of 11 funds even lost money,” they wrote.
In addition, six of the eleven all-ins and 25 of the 43 partial AIs have now been closed, which represents a clear, unambiguous and more or less absolute end to the investment competence of generative bots.
No correlation
The reason for this, in their opinion, is that AI tools are unparalleled virtuosos in detecting statistical patterns – but are unable to assess them correctly in order to maintain their functionality.
For example, they pointed to a correlation between daily stock prices and low temperatures in Antelope, Montana, and suggested that AI could (really, unbelievably) use this connection to make investment decisions, since it doesn’t know the temperatures, the stock prices, or whether there is a connection between them.
Unreliable at the moment
“Until AI algorithms understand what words mean and how they relate to the real world, they will continue to be unreliable in making important decisions, including investments,” they claim.
This last fact may also cause Spock to look somewhat quizzically, just as he once raised his eyebrows at Kirk or McCoy. But for private bankers, wealth managers and investment advisors, it should be a clear and present comfort – at least for the moment.
