Data going back decades is also needed and investors are focused globally on emerging markets, corporate sovereigns and investment-grade bonds, he added.
“And then it took a while for us all (as an industry) to build up the necessary infrastructure to collect … even mundane information, such as analytics for all the constituents of all these different benchmarks and have that readily available,” Wuebben said.
Many fixed-income investors focus efforts on trying to determine macro factors including yields, spreads and Federal Reserve policy. Machine learning can help create better factors, Wuebben said.
Many datasets that investors, like hedge funds, use have been geared toward stock selection, said Andrew Chin, the recently named first-ever chief artificial intelligence officer at AllianceBernstein.
Chin sees an opportunity for fixed income “to look at datasets that may not be as easy accessible, stitching together a variety of scrapped data to put together information about fixed-income securities; munis (municipal bonds), I think, is a perfect example,” he said.
Quantitative methods coming to the fore in fixed income is an ongoing process and machine learning and AI will play “ever-increasing” roles, Wuebben predicted. AB manages $294 billion in fixed income as of June 30.
Chin noted that using AI to summarize documents and calls and extract key metrics from corporate filings allows AB's analysts to be a lot more efficient and effective.
And machine learning factors that work in the equity domain can also carry over to fixed income, Wuebben said, pointing to building factors from sentiment analysis.
“Factors that allow us to make statements about … the relative attractiveness of issuers. So that's a factor you find in equities, but you would also find that in, say, corporate bond investing,” he said.
Chin said jokingly that many AI tools, such as classifying humans vs. cats, are dealing with easier, high probability problems. Asset managers must figure out how tools can work for them.
“Our industry is very different … so these tools have to be adapted to see how they work with low signal-to-noise ratio problems, and I think we're just at the start of that, and so I think that will take time,” he said.
'Equity market has come a long way'
“Quants have been using forms of AI for decades,” said Stacie L. Mintz, managing director and head of quantitative equity for PGIM Quantitative Solutions, pointing to definitions that encompass regression models.
Natural language processing — machine learning technology that gives computers the ability to interpret, manipulate and comprehend human language — is an area of AI that PGIM has “strongly” embraced and used for many years now, she said. Mintz said it’s a “very valuable tool and directly has impacted and improved our models and the ability to get beyond just basic financial data of companies.”
Over the last decade, tools and computing power advanced, closing the information gap. Mintz pointed to information that was previously restricted to fundamental analysts talking directly to companies or physically reading items that quantitative models could not read. And non-quants have realized the value of these tools.
“I feel like everybody is a little bit of a quant now because it’s kind of hard not to be with all the information out there you need to digest. So, I think that the equity market has come a long way,” she said.
PGIM recently hired a natural language machine learning expert allowing the team to go even deeper, Mintz added.
“I think quants …are a little bit ahead of the game with a lot of this, mainly because we have the staff, we have the talent who can evaluate new techniques, new sources of data. This is what we’ve done always,” she said.
Stock selection is a “really complicated” question with so much information out there and a huge signal-to-noise ratio, Mintz said, adding that it’s important that “we have signals in our stock selection model that are based in fundamentals and that we understand.”
“If AI finds some interesting signal and it worked in the past, if you don't understand what it is or what the foundation of it is, we wouldn't be comfortable using that in the future because, you know, past performance does not guarantee future results,” she said. “So, I think there is a limit there. I think humans are going to really be the ones that need to determine whether whatever has worked in the past can really be relied on to work in the future.”
Human assessments are still critical, Mintz said, offering the example of the COVID-19 pandemic as a moment in time where investors realized it was unlike something they had ever seen before.
And Mintz sees applications that will support the investment business. She pointed to exploring using AI for automating to some degree responses to requests for proposals, portfolio compliance monitoring, client communications and summarizing information.
“It’s about improving efficiency and increasing productivity of the team we have by really automating some of these more mundane tasks — important but mundane,” she said.
