This article was first published in Pensions & Investments on 25th September 2023 here.
Daily advances in artificial intelligence are leading some to herald it as the next frontier in asset management. AI can generate opportunity with new investment ideas and sources of alpha, as well as improve the risk management of existing quantitative models while creating new ones.
But which developments are real and actionable, and which are simply noise?
“What we’re seeing is, in a way, remarkable. The level of awareness of the technology has increased massively,” said Benjamin Roy, chief technology officer at CFM, a global quantitative and systematic asset management firm. “AI used to be known by some niche players and experts, but now it’s headline news everywhere,” he said, particularly with the release of Open AI’s widely popular ChatGPT, a generative AI tool.
Institutional investors need to look beyond the hype and understand both AI’s place in the asset-management ecosystem today and the ability of an asset manager to benefit client portfolios by implementing AI scientifically and collaboratively. Just because someone — anyone — can scour hundreds of earnings reports and scrape thousands of websites and news articles in a matter of minutes doesn’t mean that the information they extract is useful.
“AI is not a magic wand,” Roy said. “It’s another statistical tool.”
Three key areas for AI
In fact, for decades, the financial services industry has been using AI and large language models, which use algorithms to process data such as human text. AI is already being applied across compliance, investment research, electronic trading, trade matching and liquidity sourcing. It has streamlined investment processes, improved risk management and produced innovation.
Today, CFM sees three key areas where AI can deliver efficiencies in asset management: alpha generation, portfolio construction and execution.
“In all those layers, AI can be beneficial,” Roy said. “For us, it’s not just one area, it’s all three. We are also looking at other areas that are less involved in the investment process, where AI’s recent advancements can also be beneficial, like legal, HR, even research and development. But if we come back to the investment process and how we manage client money, it’s across the three layers.”
Multilayered benefits
Certainly, one of the biggest benefits of AI for asset managers is its ability to filter and analyze bigger and bigger sets of data — a key component of their investment process. AI algorithms can scrape, clean and analyze terabytes of data relatively quickly, including regulatory filings, social media posts, weather reports, securities trading statistics, web traffic statistics and government economic reports, to name just a few. The key challenge for asset managers is to find value in all that data that helps asset owners meet their investment goals.
“We’ve found a lot of new ideas by leveraging deep-learning, natural-language processing and language models,” Roy said, noting that CFM has long used these technologies. “In our investment process, we analyze and crunch hundreds of data sets every day. We have both structured data as well as more and more unstructured data. AI tools and the evolution of those tools have helped us process this data more efficiently and extract more value out of it.”
Testing, validating models
For asset managers, that value-extraction process is a key element in pursuing AI system capabilities, whether new or existing.
“There’s a fast-growing number of open-source models that we can use,” Roy said. “And in a way, that’s where we find some new ideas that can lead us to evolve our approach.” A core part of the process is testing the models, which takes time and expertise.
“What’s important is to spend time testing those models. And to do that, to validate that there is value in them, you need to have a scientific culture and process to make sure that the models provide value-added results versus just noise,” Roy said. “We keep testing and evaluating those models because it’s evolving so, so quickly.”
That process requires “massive” computing capacity, Roy explained. Once tested and validated, models also need to be able to filter more and more data. It’s all about scale, he said. “If you want to be able to leverage and use those new [large language models] efficiently in an investment process, you need scale,” he said. “That’s one big difference” with today’s AI technology over earlier iterations.
That’s where cloud computing comes in.
Leveraging the cloud
Cloud service providers, such as Microsoft Azure, Amazon Web Services and Google Cloud Platform, that have the ability to house and process huge amounts of data, can typically deliver the technology resources for an asset manager to develop its own models. “Working with cloud providers is a must if you want to be able to manage bigger and bigger data sets, use more and more computing power and leverage more and bigger models,” Roy said. “And they keep up with innovation in this space.”
One potential risk is that because new AI tools are readily available, anyone can harness their powers. “Because of OpenAI and the cloud providers, it seems very easy to play with [AI] and find some value,” but that’s not the case, said Roy. “It takes a lot more expertise to make it work at scale, on a daily basis, in a production environment.”
The expertise needed to turn AI into a useful tool goes beyond technology. If the tech experts who work for an asset manager are in a silo, their output may not lead to anything that can be implemented and leveraged to drive investment results. Collaboration between the technology and research teams is critical. At CFM, the two functions have almost blended into one.
“Our experience is that in order to really exploit those models, to find new sources of data, our engineers need to work very closely with our alpha researchers,” Roy said. “The engineers need to have a research mindset. More and more, the lines are being blurred between technology and research.”
Execution efficiency & alpha
AI implementation today comes down to two key components: crunching larger and larger data sets and leveraging that data to source alpha. That can have broader impacts across the investment firm. For example, Roy said, the ability of AI to filter through or screen terabytes of data in minutes rather than hours can increase an asset manager’s efficiency exponentially and potentially open up resources for other, higher-value tasks.
Or consider trading. For a multi-asset manager, trading across asset classes and regions can become expensive and cut into alpha potential. AI can not only route orders more efficiently than humans, but it can also track trades and identify areas for improvement. “We invest in thousands of assets across more than a hundred trading venues,” Roy said.
“We have a huge amount of data on how our execution has been working and how market structures are behaving,” he added. “There are many subtle differences between all those markets, all those assets. Machine learning has been an efficient tool for us to speed up the understanding of all those venues and subtleties of the market. It has delivered value for us and our clients.”
Rigorous approach
If AI has become table stakes for asset managers, how can institutional investors determine who can really deliver the goods? Does their tried-and-true manager selection process need to change? Roy doesn’t think so.
“If you’re looking for a manager that is not quantitative but who’s using AI, you should ask the same type of questions you would ask a quant manager in order to ensure they have a strong scientific process that is reproducible and rigorous,” he said. “What is your approach? Have you tested your process? Do you have traceability and audit capabilities? Can you explain what happened with your simulation?”
Even as new AI models are being released every day, it’s become harder to leverage those models to generate implementable results. To be successful in leveraging the power of AI, asset managers need to have the infrastructure in place, well-tested, repeatable processes for parsing data and a collaborative cross-team approach that can deliver alpha opportunities from existing as well as new data sets.
DISCLAIMER
Any description or information involving investment process or allocations is provided for illustration purposes only. There can be no assurance that these statements are or will prove to be accurate or complete in any way. This article does not constitute an offer or solicitation to subscribe for any security or interest.
