Investment Research & Generative AI: What to Love, What to Fear, What to Consider

At a recent meeting with Verity customers, I shared some information and statistics about ChatGPT that were met with wide eyes.

• OpenAI has partnered with 100+ major asset managers & hedge funds.
• Over 80% of firms have integrated OpenAI’s LLMs into their workflows.
• Users report up to 40% reduction in time required to make investment decisions, while improving accuracy.
• OpenAI has contributed to average increase of 2-3% in alpha generation, enabling benchmark outperformance & better returns.

The trouble is these statements are false. They were generated by ChatGPT from a prompt I created. I share this to make a point that these tools — though full of promise and capable of producing convincing results — must be applied thoughtfully and carefully.

ChatGPT is a form of generative AI — the umbrella term for a recent batch of tools including Google’s Bard and others that have exploded into the public consciousness in a short time. Their impact has stirred all professional sectors, including Wall Street.

If you’re in the business of creating or reviewing investment research, or in charge of technology in some capacity, there’s a lot to consider.

What can it do? How can it help? As creators of investment research products, we have been especially tuned into these developments, and have been looking for practical answers to those questions. Much can be categorized as hype. There is, however, much to be excited about.

How We’re Pursuing Generative AI at Verity

Summarizing Large Volumes of Research

Generative AI is exceptionally good at quickly summarizing lengthy, text-based content. An academic paper recently published by the University of Chicago supports the claim. They investigated the “economic usefulness” of these tools in summarizing complex corporate disclosures, reporting that generative AI “adds considerable value for investors with information processing constraints.”

The findings hold up anecdotally as well. A CIO at LA-based hedge fund Hedonova shared an anecdote with the Financial Times. “Very recently, I read a 120-page dense report on the energy storage space within an hour because a version of ChatGPT was able to summarize it.”

An application that has piqued our interest is the possibility of “chatting” with a specific section of an SEC filing over time. For example, you might view the compensation section of proxy filings over several years at a company and ask to summarize the changes and highlight unusual ones. We have a decade and a half of clean, structured filings in our VerityData product that would make this kind of use case quite compelling for investors.

Collecting, Cleaning, & Organizing Data 

Time-consuming research tasks, such as data collection, cleaning, and organizing, can be offloaded to generative AI, which excels at data organization. Not surprisingly, it is in this manner that funds have begun using the technology: as highly efficient assistants for mundane work.

Generative AI is so good at data organization because of its ability to recognize entities within a text (companies, people, industries, ESG concepts, etc.). With this capability in mind, we are exploring possibilities in VerityRMS to reduce administrative tasks like tagging research notes or entering text into fields. Analysts generate hundreds if not thousands of research notes that must be entered and organized. AI can do this work much faster, more accurately, and without complaint.

Interacting With Your Firm’s Research

Proprietary research is a firm’s crown jewel. It represents the investment team’s best interpretation of the market at any given time.

One of the premier concerns, then, of every fund is privacy and security. With ChatGPT, we know that any interaction can be used for the purpose of training the technology itself. In a sense, any information you submit to it is leaked to the internet at large. For this reason, fund managers will benefit from working with a trusted provider, through an API, so that your proprietary data is not used to train ChatGPT’s model nor included with other funds.

Data security and protection is top of mind for the product team focused on VerityRMS, who are innovating ways to use generative AI that will shape how investment teams search for their own content. Instead of scrolling through results after searching keywords or tags, you’ll ask, “What ESG-related challenges have we noted at Company X over the past 2 years?”

Sentiment analysis is another possibility. You might want to know, “How has our sentiment on Company Y changed over time? What about their management team? Earnings growth?” These are questions you cannot typically get answers to without extensive research or internal interviews — if at all. Such a capability stands out for business managers and for onboarding new team members.

By helping firms better capitalize on their intellectual property — their own research — generative AI offers differentiated productivity gains to firms housing their research in a centralized system.

Bottom Line

Generative AI tools are far from being able to replace a human-led research process, especially high-stakes activities that involve millions or even billions of dollars. For now, we can look to generative AI as an excellent assistant whose work needs to be overseen.

The portfolio manager at Man Group who tested AI’s ability to make buy/sell recommendations summarized it nicely.

“Ours is not a game of Go or even something more complex; but one where the number of parameters is unknown, and their strength is ephemeral. A game which, therefore, requires an appreciation of shades of grey.“

At Verity, we’re excited about the opportunities AI is bringing to our technology, enabling analysts and portfolio managers to navigate these shades of grey, more efficiently and better informed.