The latest artificial intelligence hype is powering a massive surge in the stock market on bets that a new era of innovation is nigh.
Yet for money managers who weaponize computing advances for an investing edge, the era of ChatGPT holds a less lofty promise for now: Automating the grunt work.
So-called generative AI is already helping to speed up mundane tasks known to crush the spirit of junior Wall Street employees, hedge funds say — from reviewing reams of market research to writing basic code and summarizing fund performance.
Chatbots could eventually help generate material efficiency gains and provide more rewarding work for their human overlords, possibly at the cost of jobs. But it’s early days yet.
At systematic hedge fund Campbell & Co., its quants have spent months experimenting with using the tech behind ChatGPT to summarize internal research and write boilerplate code. Yet generative AI tools are proving no game-changer for their day-to-day investing methods, just yet.
“They are very strong for code completion, editing, finding errors, and fixing bugs,” said Kevin Cole, chief executive officer at Campbell. “Our model would keep humans in the loop — an assistant to the human helping to make their job more efficient.”
AI on Wall Street is a broad church that includes everything from machine-learning algorithms used to compute credit risks to natural language processing tools that scan the news for trading. Generative AI, the latest buzzword exemplified by OpenAI’s chatbot, can follow instructions and create new text, images, or other content after being trained on massive amounts of inputs. The idea is that if the machine reads enough finance, it could plausibly price an option, build a portfolio, or parse a corporate news headline.
As hedge funds experiment with the latest iterations of these tools, the ultimate goal is improving investing performance. For now, boosting productivity — accelerating coding, research, and client communications — is the most obvious benefit. It’s why the likes of Citadel’s Ken Griffin said in March the firm is negotiating an enterprise-wide license to use ChatGPT, betting that it will automate an “enormous amount of work.”
At Man Group, one of the world’s largest hedge funds, Rob Furdak says ChatGPT can speed up the preliminary parts of research by reviewing a stack of academic papers on a particular topic and detecting basic patterns in data sets.
“A big part of the research process is cleaning the data, mapping it, and then doing a preliminary analysis,” said the chief investment officer for responsible investment. “ChatGPT could say ‘That’s an interesting hypothesis, but here are other hypotheses you may want to investigate as well.’”
The firm is also looking into automating the grunt work of investor relations, he said, since ChatGPT can easily explain performance by synthesizing market data and fund returns.
An example of how ChatGPT might help with finance research. Source: “Unleashing the Power of ChatGPT in Finance Research: Opportunities and Challenges” by Zifeng Feng, Gangqing Hu, and Bingxin Li.
Wall Street, of course, is already famously filled with computer wizards that populate algo trading desks, quant hedge funds, and high-frequency market makers, and to them, ChatGPT’s capabilities might not seem that new. The current frenzy has arguably been fueled as much by the sudden widespread availability of the tool as by any advances in what generative AI actually does.
Still, early studies suggest the chatbot does represent some steps forward. For instance, Fed researchers found it beats existing models such as Google’s BERT in classifying sentences in the central bank’s statements as dovish or hawkish.
A paper from the University of Chicago showed ChatGPT can distill bloated corporate disclosures into their essence in a way that explains the subsequent stock reaction. Academics have also suggested it can come up with research ideas, design studies, and possibly even decide what to invest in.
Peter Cotton, the chief data scientist at Intech Investment Management, is among the recent converts after testing the robot himself. He posted on Github a conversation he had with ChatGPT, where he used it to write code for extracting data and making predictions.
“My whole workflow has been dramatically changed,” he said. “I’m surprised by just how much knowledge is stored inside.”
How GPT-3, GPT-4, and an analyst at the Federal Reserve explained their classifications of a sentence in a Fed statement. Source: “Can ChatGPT Decipher Fedspeak?” by Anne Lundgaard Hansen and Sophia Kazinnik.
The technology isn’t perfect, however. ChatGPT has been known to make up facts and spit out different responses to the same prompt and is only trained on data up to late 2021. Meanwhile, in an industry where trade secrets are anxiously guarded, many are still hesitant about relying on external software.
For that reason, Campbell has also experimented with an open-source and less powerful GPT model that it can run entirely within its own systems, said CEO Cole.
“We have to be very careful about risks of IP leakage with those types of tools because, with ChatGPT, you’re sending queries to OpenAI servers,” he said.
While hedge funds figure out what’s possible with generative AI, the consequences for the industry’s human workforce remain undecided for now.
Greg Bond, chief executive officer of Man Numeric, Man Group’s Boston-based unit, reckons the tech could be an opportunity for creative employees who lack technical expertise but can ask the right questions.
“If we assume research productivity is dropping more globally, you can either hire more people or you could have some digital researchers that are a force multiplier on your existing research and technology staff,” Bond said. “Ultimately what would be nice is if we could automate the innovation process itself.”
A message from Advisor Perspectives and VettaFi: To learn more about this or other topics, please check out our most recent white papers.
Bloomberg News provided this article. For more articles like this please visit
bloomberg.com.
