An initial study suggests most funds managed with artificial intelligence massively underperform standard benchmarks and will continue to do so until they understand language - and reality itself.
Since the advent of Chat GTP almost two years ago, generative AI tools have been lauded as tech «more profound» than fire, with their impact being «comparable in scale» with the industrial revolution, electricity, and the wheel.
Those are big shoes to fill at any point in human history. And although these are heady times, after the first flush of excitement and hype subsides, it seems most of us will have to content ourselves with little more than supercharged chatbots and virtual assistants of the same ilk.
Not Good With Money
As Steve Jobs famously used to say, there is «one more thing» and that is that AI appears to be bad, even terrible, at investing money, as an opinion and analysis piece in the «Scientific American» published last Friday by Sam Wyatt and Gary Smith of Pomona College in California indicates.
That is going to be a relief to all those wealth managers out there facing demands to do better by the ultra-high net worth and not-so-high net worth a few months ago when a spate of viral social media posts did the rounds claiming sky-high - and instant - equity trading returns with some vague AI-based tool or algorithm.
Towards a Century of Underperformance
The authors, however, went a good deal further than finews.asia and most others in the media-hype-market complex by maintaining that the tech proponents of AI have «overpromised and underdelivered» since at least 1960, or almost 70 years.
«It is now increasingly clear that GPT and other LLMs are not intelligent in any meaningful sense and cannot be relied on for important decisions, such as hiring choices, prison sentencing, loan approval, insurance rates—and investing,» they wrote.
Honest Metric
Wyatt is a student at Pomona while Smith is Fletcher Jones Professor at the college and author of more than 100 peer-reviewed research articles and 17 books, the most recent of which was titled «The Power of Modern Value Investing: Beyond Indexing, Algos and Alpha» that he co-authored with Margaret Smith.
Both delved into the matter in a relatively open and clearheaded fashion by saying that AI-powered investing was «particularly interesting» in that it was a way to assess the overarching abilities of the technology itself.
No Errors or Bias
They started with a look at AIEQ, the first AI-driven ETF launched in 2017, then touted as a groundbreaking application with its ability to replicate a veritable army of downtrodden equity analysts around the clock 365 days a year – all the while removing human error and bias.
Another AI global equity fund called MIND launched two weeks later that claimed to be able to recognize patterns and make decisions in the same fashion that the human brain did, but at hyper-fast speeds, and without biases, overconfidence, cognitive dissonance, or emotion.
The Vulcan Question
For all the Trekkies out there, all this indirectly answers that long-burning question as to whether Spock would be a good investor or not, inasmuch as it is a burning issue in certain hidden corners of the web (collated Google search).
The answer is that, like AI, he probably wouldn’t have made billionaire, centi-millionaire, or even multi-millionaire and would have ended up contenting himself, like the rest of us investing mortals, with the broadest-based, cheapest ETF out there.
Trailing Badly
That is because the authors say that both funds have trailed the S&P 500 badly since they were launched. AIEQ’s total return is 63 percent while the S&P’s is at 108 percent.
MIND fared even worse. It was shut down in 2022 but not before its cumulative return was a sobering minus 12 percent against the S&P’s positive 65 percent.
No Pesky Humans
But they didn't stop there, and both performed an analysis of more recent AI-driven funds while cautioning that it has not been peer-reviewed yet.
The two found 11 fully AI-driven, such as AIEQ and MIND, where decisions were made without the help of pesky, or increasingly unpesky, humanity. They also found a further 43 AI funds that did allow for some human involvement.
Still Poor
An example of the latter was the Qraft AI-enhanced US Large Cap Momentum ETF (AMOM) that uses AI to select stocks although human advisers retained «full discretion over investment decisions».
Still, even in that compromised population, only 10 of the 43 part-AI funds did better than the S&P 500, with average annual returns being a good five percentage points worse. The all-in AI funds, however, somehow managed to top even that.
«Every single one did worse than the S&P 500. Six of 11 funds actually lost money,» they wrote.
Moreover, six of the 11 all-ins and 25 of the 43 part-AIs have since been shut down, drawing a clear, plain, and more or less absolute line on the investment acumen of generative bots.
No Correlation
According to them, the reason for this is that AI tools are unparalleled virtuosos at finding statistical patterns - but that they can’t judge them to save their serviceable lives.
For example, they pointed out a correlation between daily stock prices and low temperatures in Antelope, Montana, saying that AI might use that connection (really, unbelievably) to make investment decisions given they don’t know what temperatures or stock prices are or whether they are related.
Unreliable for Now
«Until AI algorithms understand what words mean and how they relate to the real world, they will continue to be unreliable for important decisions, including but not limited to investing,» they maintained.
That last fact is something that would also be likely to draw a somewhat quizzical look from Spock in the same fashion that he used to raise his eyebrow at Kirk or McCoy, although it should be seen as a clear and present source of comfort for private bankers, wealth managers, and investment advisors – at least for now.
