By Lauren Bailey
In a global marketplace rife with
uncertainty, artificial intelligence tools and their ability to more efficiently mine
data, may just be the lifeboat
institutional investors need to navigate these uncharted waters.
A recent survey by Mercer found nine out of
10 investment managers are currently using or planned to use AI in their
investment processes. It found use of AI across investment strategies and
research has expanded far beyond the traditional ‘quant’ cohort, with nearly
all (91%) managers noting they’re currently using (54%), or planning to (37%)
use, AI within their investment strategy or asset class research.
The
question isn’t whether they should be leveraging these tools, it’s where they
can extract the most value in their adoption. Indeed, these tools can help
fundamental managers identify assets to invest in, said Russ Goyenko, associate professor of finance with McGill University’s Desautels Faculty of Management and president and co-founder of Finaix, an AI-driven asset management company.
One of the biggest benefits to these tools
is that, in times of uncertainty, when markets are reacting to shifting reality
on the ground, they’re built to provide guidance on the best possible outcomes
in any economic environment, he added. “These models are very powerful; in
a sense, they’re built to be more precise and with less [tendency to] overreact as a human
might.”
For
example, Finaix’s AI tool does 100% of the stock selection. In early
February, the outlook on the ground seemed bleak, given the tariff uncertainty
and whispers of a looming recession. But the AI model signaled the team should
hold their clients’ positions, seeing through the volatility to recognize
potential for growth over the medium to long-term. Goyenko said it was a good
decision because the market rallied mid-March after sustaining weeks of losses. A couple weeks later, he said the tool adjusted its analysis, pointing to higher volatility and significant slowdown in growth across the U.S. stock market, suggesting a shift from growth to quality large-cap stocks.
“It’s a very good, effective tool, to
identify what you face [during] huge uncertainty. What these models do
is basically analyze [historical] levels of volatility and new shocks from
different sources and . . . how this risk metric structure shifted
recently. It compares it to the past. It doesn’t look at sentiment, it just
looks at the pure quantitative data.”
Please Click Here to Register Now!
Risks
However, Mercer’s study also found the most-cited barriers for managers in unlocking the technology’s full potential were data quality and availability, followed by
concerns around integration and compatibility, as well as ethical and legal
considerations. Risk associated with divergent regulation was another significant
concern for nearly half of respondents.
During a panel session in Davos, Switzerland, in early January, Carlos Rangel, chief investment officer of the W.K. Kellogg Foundation, urged investment managers to consider the voices present at the table helping inform some of their design choices as they build AI solutions and develop their capacity.
Bias can be introduced in the development of these tools, said Mohammad Rasouli, a researcher at Stanford University and founder and chief executive officer of AIx2, noting their purpose is to turn unstructured data into clean data that helps institutional investors find the alpha and run due diligence.
“Machine learning is looking at the representation in the history . . . to find the best action to take now and in the future. It’s very easy, even for a non-technical person, to see that you are relying on that historical distribution representation. What if the future is not similar to the past distribution? What if your data that you collected for distribution was biased?”
He pointed out that, depending on the historical data analyzed, AI models could introduce bias in their own price positions and that of other assets. As well, he said some niche events could have been excluded from the historical data and may need to be synthetically added to the distribution of the data for unbiased representation of the future. This is just one reason why human oversight of these tools is necessary and will be for some time.
Goyenko agreed with this sentiment, noting, for now, human
oversight through controlled applications is important.
training the model, but once deployed live, get out of the way. Right now, it
uses more information than humans can process. Let it do what you train it to
do.”
In 2024, the Canada Pension Plan Investment
Board’s Insights Institute and the World Economic Forum jointly published a Responsible AI Playbook, which noted that ensuring AI applications are deployed in
an honest, helpful and harmless manner, is “an
important step for enhancing risk-adjusted returns and positioning businesses
for success.”
It pointed to education as a barrier to
responsible use, noting stakeholders must invest in continuous learning to keep
pace with AI advancements, including executive education, forums for investor
dialogue, and public awareness initiatives.
