A growing corpus of industry publications and media articles concede that AI can solve complex problems in other industries — but it can’t be used to crack the market.
One such publication is AllianceBernstein’s latest Black Book. Bernstein — a firm that offers in-depth, independent, long-term, industrial-strength sell-side research — has been publishing its research annually since 1974, typically in a series of “Black Books.”
This year’s edition, “A Painful Epiphany: Investing in a Post-Pandemic, Post-Global World,” caught my attention because of its title, focus, and thesis: The pandemic presents asset owners with a new investment reality that “demands a change in the approach to asset allocation and the outlook for capital markets — one that will bring about an evolution in the investment industry.”
The invocation of a demand for change, combined with the urgency implied by the choice of the word “epiphany,” signifies that the Black Book’s authors — Inigo Fraser Jenkins, Alla Harmsworth, Robertas Stancikas, Harjaspreet Mand, and Maureen Hughes — believe the investment industry is at an inflection point.
Yet after reading the 202-page tome, I was sorely disappointed. What’s offered is little more than a recapitulation of well-worn investment tropes — for example, “the choice for asset owners is some combination of illiquid assets, factor exposure, leverage, or active management.”
Bernstein’s sole nod to modernity is a recitation of the consensus view that the industry will spend much more time exploring the potential for digital assets in the coming years.
In fact, the firm spends much of the text promoting factor investing — an approach that finds its roots in William Sharpe’s pricing model and the Fama-French three-factor model. Since then, there is a long history of academics and practitioners using linear regression to deconstruct return streams into factors and a residual (alpha).
Asset owners like the British Petroleum (née Amoco) and Weyerhaeuser pension funds were using such techniques to build portfolios as early as the 1980s.
I struggle to find anything original or transformational in these ideas.
Central to the authors’ defense of a curated factor approach is the claim that in the new post-pandemic investment regime, such deconstruction of exposures and the subsequent revealing of “idiosyncratic alpha” will allow for the revival of active management.
Putting aside the fact that “idiosyncratic alpha” is a neologism that adds nothing to the investment lexicon — alpha is, by definition, idiosyncratic — the argument is flawed because the revival of active management is not contingent on a new macroeconomic environment. Its revival rests entirely on manager skill — a manager’s ability to make the investment decisions that generate alpha.
New regime or not, it is disingenuous to assume that active managers using the same well-worn investment methods (factor-based or otherwise) will now be able to generate alpha when the evidence shows these methods, used in various market regimes, have for at least the past 20 years consistently failed to beat their benchmarks net of fees, let alone to generate alpha.
Yet Bernstein’s demand for change (regardless of the catalyst) is certainly timely. For active managers, 2022 was an annus horribilis. As Stefan Hoops, chief executive of DWS, recently told the Financial Times, “The golden decade for asset management is over.”
We need look no farther than a Morningstar report for evidence signaling the end of an era. The report states, “Actively managed funds bled an incredible $926 billion in 2022, roughly triple their second-worst calendar-year outflow in 2018. . . . Passive funds collected $556 billion in 2022.”
By demanding change while offering no true path forward, Bernstein misses a grand opportunity to use its bully pulpit — precisely at the right time — to present to asset owners a truly transformational future of investing: one based on artificial intelligence.
The authors make several naive references to AI in a chapter curiously titled “Metamorphosis: An Investment Industry in Transformation.” They write about a post-Excel world where machine learning could augment existing human investment processes but could not be used to make “actual investment decisions” (are there any other kind?): “In time, this role could extend to actual modeling processes. . . . The adoption of machine learning for manipulating and extracting data seems set to grow dramatically.”
This view of AI is woefully out of step with that of the Black Book’s target readers — asset owners. According to the 2022 CFA Institute Investor Trust Study, “In the last two years, interest in artificial intelligence has grown significantly among institutional investors, with 84 percent now eager to invest in a fund that uses such technology, up from 71 percent.”
This view also ignores, or at least severely discounts, the broad consensus among data scientists, business leaders, and even the general public that AI has reached an inflection point. One observer admitted that changes are coming so quickly that “I can’t keep up with AI advancement.”
Perhaps unbeknownst to the authors, the future is already here. Not only are learning-based AI systems like ChatGPT and AlphaFold solving what were previously thought to be unsolvable scientific problems, they are also transforming our daily lives.
Investment management is not exempt from this shift. What is holding back the wholesale adoption of advanced AI is the industry’s desire to protect the status quo. This defense is supported by structural forces (e.g., educational curricula) and the overriding yet unsubstantiated belief that although AI can achieve superhuman results in other industries, investing is simply different (the underlying markets are noisy, adversarial, volatile, etc.) — and these differences eliminate the possibility of using advanced AI to make autonomous investment decisions.
It is critical to note that nonbelievers present these differences as facts, ex cathedra, without offering empirical evidence (e.g., the track record of an investment strategy based entirely on deep learning or reinforcement learning that has failed) that advanced AI cannot make investment decisions without a human in the loop.
My intention is not to single out Bernstein’s work. It’s clear the authors put a tremendous amount of thought into the research and its presentation.
Rather, I cite it as a recent example of the growing body of publications arguing that AI is merely a handmaiden to human intelligence, capable of enhancing an existing human-based investment process but unable to make accurate autonomous investment decisions.
These arguments present autonomous AI as a future, conditional possibility. And they share a blinkered perspective, one constrained by the traditional investment canon founded on the belief that investing is essentially a human activity.
It is dubious to claim that the current situation “demands a change” and to use terms like “metamorphosis” and “transformation” only to offer ideas that support the status quo.
Such posturing misleads the very investors the Black Book authors seek to educate.
Opinion pieces represent the views of their authors and do not necessarily reflect the views of Institutional Investor.
Angelo Calvello, Ph.D., is co-founder of Rosetta Analytics, an investment manager that uses deep reinforcement learning to build and manage investment strategies for institutional investors.
