George Patterson, the managing director and the chief investment officer for $103 billion PGIM Quantitative Solutions, has been using artificial intelligence in his quant investing strategies for years. But he doesn’t think the new wave of generative AI of products like ChatGPT are that useful for investment managers like him and his team.
Large language models are “less useful” for quants, explained Patterson, a former rocket scientist turned quant in the early 1990s. “ChatGPT is like is a hammer, and then everyone’s got a nail.” PGIM Quantitative Solutions was founded 50 years ago, making it one of the oldest quant managers. Parent PGIM is the $1.33 trillion global asset management business of Prudential Financial.
Until the Chinese start-up DeepSeek upended markets this week with the launch of new AI models, the promise of the technology had been powering the stock markets and a big chunk of venture capital investments.
Although it’s too early to assess the impact of DeepSeek, Patterson has long brushed aside the usefulness of tools like ChatGPT. “We’re not interested in textual output. I don’t need a paragraph,” he said.
In fact, Patterson has found older large language models useful for his fundamental quant analysis. He uses one slightly earlier technology, called BERT, to conduct sentiment analysis of conference call transcripts to detect any changes in “tone” from management.
“Are they signaling that something is more of a risk than it used to be — or less of a useful risk than it used to be?”
Even though managers have been using AI for years to scour public information, Patterson still finds the technology helpful to detect sentiment changes that are hidden in documents like 10Ks. New products are an example. “What’s the potential? Are investors getting excited about this?” Is management “signaling that this product is growing faster than they expected or not as fast as they expected?” he asked.
PGIM Quantitative Solutions (the name was changed from QMA in 2021) has a team of engineers that build out, or calibrate the AI models, often leveraging academic models or delivering new techniques, he said. The cost of building a model could cost millions of dollars in compute time, he said. “So a lot of times, we want to take one that’s publicly available, and then adjust it or customize it for our particular needs. And that’s increasingly what we see people doing.” (That thinking may change, given DeepSeek’s claim that its model was developed at a fraction of the cost of others.)
Even so, such AI is making it easier for quants to do more fundamental analysis of companies than in the past, according to Patterson, who began his career at NASA’s Jet Propulsion Laboratory after earning a B.S. in physics from the Massachusetts Institute of Technology and a PhD in physics from Boston University.
His first financial job was with Wells Fargo Nikko Investment Advisors, which later became Barclays Global Investors and was then sold to BlackRock. In 2007, Patterson helped started a hedge fund, Menta Capital — just before the quant meltdown that August. “It was probably the worst possible time,” he said. After a move to the East Coast for family reasons, Patterson joined the Bank of Montreal and then PGIM Quantitative Solutions, where he has worked for the past seven years.
“AI will help you deal with large amounts of data,” he said. “But if you ask me, the value is in the data. You can learn how to build these models. But if you don’t have the data, the models are useless.”
He argues that data is going to become more and more valuable. In the past, data wasn’t saved. “We didn’t realize that now you can go and mine this data for new customers, for new products.”
And while the computing power available to analyze the massive troves of data is vastly superior to that in the past, he suggested, “what’s going to be interesting to see is how it evolves over the next decade.”