ML has begun to make inroads as asset managers realize that the ability to extract value from big data is going to be a key differentiator — and that traditional industry practices will struggle to stay afloat in this mounting flood of real-time data. Analytics using ML can be more robust than traditional financial modelling, as it can tap into the streams of “unstructured” (text-based) data that global digitalization and social media are creating, in addition to the millions of corporate press releases, conference-call transcripts and regulatory filings that are produced every year.
With the industry in a state of flux due to the rise in passive investing and the move away from commissions to level fees, many asset managers are investing heavily in technology to reduce operating costs and to comply with regulators’ ever-increasing demands for transparency.
Leading asset managers are realizing that this also presents a solid opportunity to invest in advanced data analytics and ML capabilities, as well. With the fixed-income bull market coming to an end, fund managers need to begin to implement new investment strategies, reorganizing their operations around next-generation investment systems.
Asset managers are using predictive analytics to generate investment ideas or as an early warning system for assets at risk. At a minimum, A.I.-enhanced data analytics can complement traditional financial analysis by offering unique insights.
Merrill Lynch is experimenting with an A.I. stock-picking tool to help it identify value in small-cap stocks that conventional analysts might have missed. Because quant investment ideas are starting to have shorter expiration dates as trading signals get arbitraged away, BlackRock is steering its quant research toward ML and exploiting social media and web search information. After all, those quant ideas that do work can be turned into smart beta products, or even passive strategies that give exposure to specific return factors.
Still, apart from a few cases of smaller, leading-edge hedge funds, asset managers exhibit an understandable reluctance to place their portfolios and funds fully in the supervision of an ML-based robot. Recent work by researchers at the Wharton School of the University of Pennsylvania and the University of Chicago offer an explanation for this behavior, demonstrating a phenomenon they term algorithm aversion.
It was found that people will often refrain from employing algorithmic (computer-based) decisions even where such approaches demonstrate consistently better results than relying on human “gut” or “seat of the pants” estimations. Since many of the new ML algorithms yield answers that appear opaque, without an understandable explanation for their decisions, these systems feed into this distrust.
Research has shown, however, that if the system users feel they have a degree of influence on the decision process, their acceptance of this machine help is magnified.
Nobel Prize-winning psychologist Daniel Kahneman’s work demonstrated that humans exhibit two types of decision-making reasoning: “thinking slow,” which employs deliberative effort, and the much more natural (to us) “thinking fast,” where we typically lean on mental rules of thumb that yield explainable but sometimes logically dubious judgments.
So while we excel at thinking fast, ML-based computer systems increasingly excel at Kahneman’s more thorough thinking slow — although, given their ability to perform billions of calculations per second, machines can perform this in a small fraction of the human time requirement.
It would seem that there is a natural partnership opportunity in having humans work with ML-based AI assistants, each complementing the other’s ability. Indeed, such a pairing, often termed “cognitive collaboration” or “cognitive augmentation,” is emerging in several areas, most notably freestyle chess.
In freestyle chess, teams of capable human chess players working with one or more computer systems have consistently defeated both world-class computer programs and human grand masters — demonstrating the superiority of human strategic guidance working with a computer’s tactical acuity.
In their search for alpha, under the constraints of governmental regulation, accelerated machine-based trading time frames and magnified competition, asset managers are faced with solving perhaps the most complex class of problem: multi-objective optimization. How can one juggle profitability, earnings, risk, investment objectives and so forth, described by hundreds of attributes changing in real time?
The asset manager needs help — the help of an AI associate that can navigate and present multi-dimensional Pareto (optimal return) mathematical frontiers, producing complex predictive analytics models. In addition, new AI-based learning capabilities will enable this AI associate to accept and learn from the manager’s directive feedback and thus adapt to changing conditions and goals.
The latest AI technologies enable the development of a new virtuous cycle, where the AI assistant is there to provide alternatives and present tradeoffs while the asset manager ultimately decides the course of action, and this, in turn, teaches the AI assistant to better focus its future modeling.
While self-directed, conscious and sentient AI is still the stuff of science fiction, it’s clear that ML has enormous potential and that it’s going to come into its own, as billions of connected devices drive a real-time economy.
If we’ve hit a tipping point, and ML and automated trading technology becomes prevalent, we’ll likely see some industry-changing waves in the asset-management industry as digitization transforms it into a technology-driven one.
Savvy investment managers have cause to think long and hard about the future as the industry is on the cusp of a technological arms race where an intelligent program is poised to join your team as a machine co-pilot.
— By Gary Brackenridge, global head of asset management at Linedata