Investing knowledge automation is getting

closer thanks to generative AI.

When trained on huge datasets, large language models (LLMs) like ChatGPT and Google Gemini are highly adept at producing insightful responses to questions. Long Chen, an associate professor and area chair of accounting at the Donald G. Costello College of Business at George Mason University, and Yi Cao, an assistant professor of accounting at the same institution, are actively investigating how individual investors can use LLMs to extract market insights from the bewildering array of company data that is readily available.

Co-authored by Jennifer Wu Tucker of the University of Florida and Chi Wan of the University of Massachusetts Boston, their new working paper, which will be published in the SSRN Electronic Journal, looks at AI’s capacity to recognize “peer firms,” or rivals in a given industry’s product market.

Cao uses the real estate market as an example to illustrate the importance of peer selection. Similar to the real estate market, the value of a corporation in the capital market is partly based on the value of its peers. In the real estate market, the price of a house is determined by comparing it to similar homes in the vicinity, or what are known as ‘comps.’ Our goal in this research is to use LLMs’ ability to find comparables in order to assess business value.”

This work is just as important as it is challenging. Select peers require a great deal of time, talent, and effort to collect, organize, and manage data. But the researchers reasoned that LLMs might do most of the labor-intensive data collecting and analysis on behalf of the individual investors, and generate a peer list that is as legitimate as if it were compiled by human specialists.

“The advantage is in the capability to utilize all the information potentially out there so that it is at least performing as well as other traditional methods that can help us investors and researchers,” Cao explains.

Because “Bard has a greater ability to utilize its pre-training data, which is arguably larger than ChatGPT’s and with more parameters,” according to Cao, Chen and Cao chose Bard from Google, now called as “Gemini,” as their preferred LLM for the study.

The researchers defined “product market competition” and created a prompt for Bard. They then told Bard to restrict its knowledge pool to a particular year between 1981 and 2023 to prevent “look-ahead bias,” which is the effect of future information distorting the outcomes.
Since there is less information available for smaller or private firms, they restricted the focal firms to big, publicly traded corporations. The data-set included more than 300,000 focal firm-years in total.

For a focal firm, the LLM might produce, on average, seven peer firms; this number is comparable to the SEC’s recommendations on the disclosure of a firm’s sectors.

After that, the researchers contrasted the LLM’s output with the lists of 40 top computer software businesses that were created by three human specialists. Surprisingly, the average overlap was a little over 40%.

The federal government’s Standard Industrial Classification (SIC) codes and Text-based Network Industry Classification (TNIC), which compares companies based on linguistic similarities in their 10-K filings, were the other two alternative systems for peer identification that they compared to the AI-identified peer lists.

The result of the LLM and the TNIC substantially overlapped. Furthermore, because their monthly stock returns were more closely aligned with the focus firm, the peers found by the LLM were often a better fit than those from SIC and TNIC.

However, TNIC performed better than the LLM in locating peers for mid-sized businesses in the sample, suggesting that the LLM is not always superior.

“We need to understand that LLMs are actually a very powerful, new tool, unmatched in their efficiency, ability to process vast amounts of information at a low cost, and accessibility to the general public,” Cao writes.
“It’s especially beneficial for individual investors as all the cost concerns that we’re talking about are especially relevant for them,” Chen continues.

Chen says the following about the future of LLM: “Using generative AI always has risks and rewards. It’s unclear when the current systems will become outdated.

In response to a question on the SEC implementing an AI tool for investors, Chen stresses that users must weigh the benefits and drawbacks of utilizing AI in order to make an informed decision “because AI cannot be held responsible for the information it provides or for how it is utilized.”

Chen says in closing, “We have to accept this new technology, but we also have to acknowledge that it is not yet flawless. There is intense competition to advance technologies. Our results may simply indicate the technology’s lower bound of efficacy.”