John Pearce, a quant and equity portfolio manager at The Illinois Municipal Retirement Fund, sat down with editor Christine Giordano and enjoyed a philosophical and pragmatic discussion on how AI might impact the future of investments, and considerations to be made.
How do you think AI might influence markets 5 years into
the future?
There is a lot of excitement about the impact that
Artificial Intelligence (AI) will have on the economy and markets generally.
Whenever there is this much hype about something, it can be helpful to take a
step back and try to build a framework to aid in separating signals from noise.
Pastor and Veronesi published a paper in 2009 in the
American Economic Review[1] that could be useful for this
purpose. Although it focused on the historic examples of railroads and the
internet, I think the model they developed for how technological innovations
are incorporated into stock prices can be applied to today’s questions about
AI. Their model would suggest that fundamental analysts are now working
diligently to incorporate the expected impact of AI on future cash flows of
companies. The so called “cash flow effect” drives stock prices, and
valuations, higher in the short-run for innovative companies. But there
is still great uncertainty about the magnitude of productivity gains that will
be realized by any given company, let alone the economy overall.
Technological revolutions rarely follow a linear path, nor
do they necessarily deliver lofty growth expectations. Even if “this time is
different” with AI, there remains perhaps even greater ambiguity of how these
advances will affect eventual valuations as AI integration moves from an
idiosyncratic risk component for early adopters to a systematic one as
technology implementation goes more widespread. This change increases discount
rates and brings down multiples. In short, if this model serves as a guide, we
are likely still in the relatively early, and highly uncertain, stage of the AI
technology “revolution.” Using this framework as a guide, I expect that the
sky-high valuations we see today are more likely to revert than to remain
permanently elevated.
How might AI advances impact the portfolios of asset
owners?
Allocators will have to make a frank assessment of how
comfortable they, and their stakeholders, are with making allocations to
strategies that are driven by AI, especially AI strategies that we mere
humans do not (and maybe even cannot) understand. Gennaioli, Shleifer,
and Vishny coined the term “Money Doctors”[2] when describing the importance of
trust in investment management.
Trust in a manager is believed to reduce the perceived risk
of a given investment and allows them to charge a fee that is stickier than
would otherwise be expected. Will asset owners trust the bots as much as we do
humans, especially when the bots point to patterns that defy economic
intuition? Further, we as allocators need to be honest with ourselves about how we
are likely to react when AI driven strategies hit a slump. Referring again to
Shleifer and Vishny, will this prove to be another situation where we realize a
“limit of arbitrage?”[3]
Are investors more likely to throw in the towel on an
otherwise good strategy at the point of greatest pain if we do not understand
it or have a warm human hand to help us see the light? Famously, it was darkest
just before the dawn in the episode that was later called When Genius
Failed[4]. As more investment firms incorporate AI
into their processes, these questions will become even more important than they
already are and are likely to change relationships between asset owners and
managers.
How will the relationship between asset owners and asset
managers change in the age of AI?
Asset owners will need to learn to tell apart the innovators
and the imitators. Ideally, the asset management firms we partner with will
embrace new technologies in a thoughtful and measured fashion. Whether
assessing a manager’s adoption of AI for its own investment processes or how it
will do diligence on potential portfolio companies that are reliant on AI for
growth, asset owners should start thinking about how to identify imposters.
One way to tell may relate to the amount of investment that
firms have made over time in preparing their data and analysis teams to
implement these new tools. There is the old quip that if Abraham Lincoln were
given eight hours to chop down a tree, he would spend the first six sharpening
his axe.
Investment firms with strong quantitative chops likely
already have truly proprietary data sets, decision ready analytics
architectures, and teams of experts who understand the proper use and,
importantly, potential for misuse of models. They are now primed to take
advantage of the increasing power of AI and ML tools.
On the other hand, those who think they can simply flip a
switch with the latest GPT engine are likely to be disappointed or, worse, will
start to learn hard lessons about input integrity and the dangers of data
mining. In short, it is time to start asking whether you believe that an asset
manager has been spending years sharpening or if they just recently started
swinging.
Another way to detect pretenders is to ask whether the team
has the technical expertise in AI to effectively screen out ideas that are not
technologically feasible. Just because it sounds like AI ought to be able to do
something does not mean that it actually can.
On the other side of that coin, even if an idea is
technically feasible it must be commercially viable. I have heard some proposed
use cases that strike me as applying an AI solution to something that is not
really a problem. “Product-market fit” still matters in an AI world.
Finally, in my experience, experts tend to talk about the
promise of new technology in their fields in a rather guarded fashion. Maybe
they are just sandbagging, but I think you are more likely to find
disappointment when you are promised the moon and the stars by overly eager
optimists.
Those who set reasonable expectations and show their work on
what they have done to prepare themselves to be able to leverage new
technologies are the ones I would bet on to be the most successful.
What are some of the limitations of AI?
We have all heard about generative AI tools occasionally
“hallucinating” and producing nonsense results to inquiries. But smaller errors
pose at least as big of a risk. To grossly oversimplify things, LLMs are
programmed to search for the next most likely thing to say given a prompt. The
output of a model may look correct enough that it might convince a non-expert
of its accuracy, but it can be wrong in small and important ways to an expert.
Perhaps you can feed the model better inputs or provide some
supervised training to improve results, but both of those remedies require
contributions from specialists with organic intelligence. Alternatively,
expertise is knowing when the next most likely thing is exactly the wrong
thing. Wisdom to know the difference is extremely valuable as the bots
encourage convergence.
The way that experts are developed will also change with the
continued advancement of AI tools. As a student, my calculus teacher would not
let us use calculators. In hindsight, I am grateful for her insistence that we
learn to do things by hand because it taught us fundamentally different ways to
approach problems. When you learn to do things with and without the aid of
different tool sets you come to see similar problems from a variety of angles.
Likewise for young financial professionals, basic modeling
is the exact sort of practice they need lots of time doing to build expertise.
If we take away those practice repetitions, we deny opportunities to hone their
craft. So, while we need to learn to work with AI, we also need to think
carefully about how using AI will change the way we learn.
What are your thoughts on the “Magnificent Seven?”
The ex-post assignment of clever monikers to cohorts of
outperforming stocks is nothing new. We have a human desire to create
narratives after the fact that make the world seem more predictable than it is
ex-ante. Even though it might not feel like it, there is a lot of rotation in
which stocks lead the market over time.
One way to demonstrate this churn is to take a snapshot of
the ten biggest stocks by market capitalization every ten years. When I did
that for the US market over a recent history, I observed that, on average, six
of the top ten from the decade prior drop out by the next snapshot window.
While this measurement is far from scientific, it is consistent with the
findings of others who have run similar studies.
So, the historic record suggests there are worse than even
odds that a given “Magnificent Seven” stock will still be in the top ten a
decade from now. This is not to say that today’s market leaders are guaranteed
to be losers going forward. But a reasonable base rate expectation is that we
will not still be calling these same seven companies “magnificent” in 2034.
[1] Technological Revolutions and Stock
Prices | NBER
[2] https://scholar.harvard.edu/files/shleifer/files/moneydoctors_journaloffinance.pdf
[3] https://scholar.harvard.edu/files/shleifer/files/limitsofarbitrage.pdf
[4] https://www.amazon.com/When-Genius-Failed-Long-Term-Management/dp/0375758259
John Pearce, an equity portfolio manager at The Illinois Municipal Retirement Fund since 2019, enjoys staying informed on the latest research regarding Artificial Intelligence. He comes from a quant background, having spent eight years as a director of research and analyst at Northpointe Capital.
