Senior PM & Engineer highlights the distinction between machine learning, large language models, explains how both serve a new set of products
AI has become a shorthand for generative AI, the large language models (LLMs) whose launch in 2022 sparked the so-called ‘AI revolution.’ Adam Cilio, however, has been using AI in his processes for almost a decade. The Senior Portfolio Manager & Engineer at Guardian Capital runs that firm’s i3 investment portfolios, which integrate artificial intelligence, human intelligence, and innovation. The process at i3 rests on a core understanding that there are many forms of AI, and that certain AI models are better suited to portfolio management tasks than others.
Cilio explained that the i3 team has used a machine learning AI tool since late 2017 in their process. He noted that this more ‘traditional’ AI model can be better suited to the numerical tasks that LLMs are often less equipped for. At the same time, he spoke to how the firm now uses LLMs to digest key qualitative inputs and glean necessary insights. All of this combines, he says, into a formula for alpha generation with controlled costs and the capacity for real differentiation.
“I think what sets us apart is not just the AI technology itself. Algorithms can be implemented anywhere. But it’s the combination of applying AI to a stock universe that’s analyzed through a multiple model, multiple feature approach, and then combining that with the experience of the portfolio managers who actually do the portfolio construction and the risk management component of it at every stage,” Cilio says. “We use AI to help us get a focus on companies that have quality earnings growth and then sustainability of cash flow, payouts and dividends over time. We don’t just focus on the size of the payout, but also the sustainability and growth of that payout.”
How AI can inform a dividend strategy
That process is now at work in two new funds from Guardian Capital, the Guardian i3 Global Dividend Growth Fund and the Guardian i3 Canadian Dividend Growth Fund. Cilio explained that the AI tools used in managing these funds are designed to evaluate stocks against global peers. In the case of the Canadian fund, that helps correct against the tendency for Canadian equity strategies to be dominated by a handful of companies.
In both the Canadian and global equity strategies Guardian just launched, Cilio explained that the AI tools allow his team to predict forward dividend growth of a company over time, while they measure the sustainability of those dividends. This allows them to steer the portfolio towards companies that may pay out slightly less, but who grow their dividends while avoiding companies who may be unable to sustain high dividend growth over time. Those outcomes, he says, tend to correlate well to long-term equity performance, especially the protection of capital in the long-term.
The broader application of AI feeds into what Cilio calls the firm’s GPS investment philosophy. GPS stands for growth, payout, and sustainability of cash flows. The AI tools, he explains, allow his management team to focus on those three core metrics in their analysis. The result, he says, is that portfolios can be high conviction, with lower turnover, and a more concentrated set of names. All of that is informed by the application of the right AI for the job.
The right AI tool for the right AI task
The core AI tool that Cilio and his team have been using since 2017 is not an LLM. It’s a numerate machine learning model algorithm that goes beyond traditional linear statistical models. In the case of these products, Cilio and his team use that AI model to looks for dividend growth, dividend sustainability, and earnings growth. That more traditional AI tool tends to be more robust in the face of outliers, spurious data, and the risks of hallucination which can derail the newer LLMs. They can also control for biases in input data and see interrelationships between these data that linear models can’t.
There is also space, Cilio says, for the application of generative AI tools, namely in the digestion and analysis of unstructured text data. LLMs, he says, can replace the three analysts who would need days to pore over 500 documents. The LLMs can speed qualitative analysis while machine learning AI can power quantitative analysis. The whole process still relies on a human at the core.
“AI puts the framework in place where you still have someone who is overseeing and monitoring things and it has to still be at the scale where a human can do that,” Cilio says. “I think for our industry that’s very applicable. You’re not just going to rely on an LLM to give you something and then take it as at face value. You actually have to have specific software development and control in place that’s run by people that can give concrete outputs of not just what an LLM is either saying or summarizing, but what is its level of confidence. You can build all this, but it needs to be done by a person.”
How an AI process benefits investors
Cilio believes that these use of these AI tools can help with cost control and alpha generation for an investment strategy. He notes that his own team is relatively lean, able to operate efficiently with four investment managers, one client portfolio manager, two data scientists, one data engineer, and one software developer.
AI can also inform a higher conviction strategy, Cilio says, by focusing specifically on the metrics by which a fund will be judged, like sustainability of dividend growth. That can result in lower turnover and ensure that portfolio managers aren’t engaging in “closet indexing,” where a small number of names act as a surrogate for a specific index. Key to a positive outcome, Cilio says, is the understanding that certain AI tools are better suited to certain tasks, and that an experienced human at the helm can turn AI into an efficient strategy that works.
“Clients just really understand that in the end the product is still managed by an experienced team of portfolio managers who are now better able to utilize their own domain knowledge combined with AI technology to make better decisions,” Cilio says. “I’d go so far as to call it an active management renaissance. I think it allows us to have a better process which would then translate to a greater trust in active managers.”
