Artificial intelligence poses a major threat to the worst stock pickers on Wall Street. Even the mediocre ones might want to start looking for another line of work.
But the jobs of the best stock pickers are secure — for now.
That’s the conclusion I reached upon analyzing the real-world performance of an AI stock-selection program over the past year. The investment firm that employs the AI emails me, at the beginning of each month, a list of 10 stocks that the AI program predicts will outperform the market over the subsequent 60 days. For each recommended stock I compared its total return to that of the S&P 500
The results are encouraging: 63% of the past year’s stocks beat the S&P 500 over the 60-day period following recommendation. That’s significantly better than the 50% you’d expect if the AI program was no better than random.
To put this in context, I conducted a similar analysis for each of the more than five-dozen investment newsletter model portfolios for which my firm audits performance. For each stock over the past year, I compared its return to that of the S&P 500 over the 60 days subsequent to the date on which the newsletter editor recommended it.
About one-in-five monitored newsletters did at least as well as the AI program, and several did even better. In the case of the best performer among my firm’s sample, 75% of its stocks recommended last year did better than the S&P 500 over the 60 days following recommendation. (I am not naming the AI-focused firm that each month sends me the list of 10 stocks or the newsletters that did just as well or better.)
You shouldn’t be surprised by what I found. The result is similar to what I’ve heard about AI’s potential in other industries. A programmer I know at a leading tech firm tells me AI poses a threat to the job security of the least-qualified developers — but not to those at the top.
Why doesn’t AI do better in the stock market?
This discussion isn’t so much a criticism of AI but a caution against the inflated expectations that many have about what AI can do in the stock market. As I described in a mid-February column, we know that expectations are inflated because stocks perform better when they are owned by an ETF with “AI” in its name. This is reminiscent of the irrational exuberance at the top of the internet bubble, when a stock’s price would jump when it changed its name to include “dot com.”
In that mid-February column I hinted at why AI doesn’t do better at picking stocks: The market is an efficient arena in which the signal-to-noise ratio is extremely low. That in turn makes it close to impossible to predict the future.
Some readers emailed me to ask what this low ratio looks like in practice, so let me offer the following argument made last summer by Stanford University historian Niall Ferguson: “Consider for a moment what we are implicitly asking when we pose the question: Has inflation peaked? We are not only asking about the supply of and demand for 94,000 different commodities, manufactures and services. We are also asking about the future path of interest rates set by the Fed, which — despite the much-vaunted policy of ‘forward guidance’ — is far from certain. We are asking about how long the strength of the dollar will be sustained, as it is currently holding down the price of U.S. imports.
“But there’s more. We are at the same time implicitly asking how long the war in Ukraine will last… We should probably also ask ourselves what the impact on Western labor markets will be of the [Covid pandemic]… Good luck adding all those variables to your model. It is in fact just as impossible to be sure about the future path of inflation as it is to be sure about the future path of the war in Ukraine and the future path of the Covid pandemic.”
I read Ferguson’s argument in a provocative essay entitled “The Illusion of Knowledge,” by Howard Marks, the respected co-founder and co-chairman of Oaktree Capital Management. He believes that, if anything, Ferguson was too optimistic about what’s possible to predict: Marks wrote that “accurately predicting inflation is ‘more impossible’ (if there is such a thing) than predicting the outcomes [of the war in Ukraine and the Covid pandemic]…, since doing so requires being right about both of those outcomes and a thousand other things. How can anyone possibly get all these things right?”
The bottom line? Asking AI to be more than just moderately better than average at predicting stock-market winners is asking the impossible. As always, investors need to keep their expectations in check.
Mark Hulbert is a regular contributor to MarketWatch. His Hulbert Ratings tracks investment newsletters that pay a flat fee to be audited. He can be reached at firstname.lastname@example.org
More: How AI can help us become better investors
Plus: ChatGPT may be good at your job, but AI is a terrible stock picker