ChatGPT can level the playing field for retail investors

Across industries, generative AI has unlocked powerful new capabilities to wider groups of users.

Now there are signs that it allows retail traders in the financial markets to be more competitive with better-resourced pros.

A study coauthored by Xiumin Martin, Olin professor of accounting, found that since ChatGPT’s widespread release in November 2022, the trading patterns of retail traders more closely resemble those of sophisticated institutional traders.

This phenomenon is evident around the time of companies’ quarterly earnings calls, when traders must analyze a large amount of corporate information quickly. Institutional investors have relied on AI and machine learning tools to digest the transcripts of these calls since well before the introduction of ChatGPT.

“This advanced technology was very expensive, very complicated, and difficult to use,” Martin said. “Whereas generative AI, particularly the open-source, open-usage tools like ChatGPT, allow almost anybody to use it for free, at least in the current form.”

She said the use of ChatGPT appears to be shrinking the information gap between sophisticated institutional investors and some retail investors.

We believe it might have implications about equality in capital markets.

—Xiumin Martin

Martin and her coauthors have been invited to present their study, “AI (ChatGPT) Democratization, Return Predictability, and Trading Inequality,” at the Ray Ball Journal of Accounting Research (JAR) Conference at the University of Chicago May 2−3.

Taming corporate information

Company earnings calls are important conference calls in which high-ranking corporate personnel — CEOs, chief financial officers, chief operating officers — talk to key investor groups.

“They give overall remarks about the current quarter and how they did,” Martin said. “They present their financial results, and open the floor to analysts, who ask about their future strategy, and the company’s guidance for their (upcoming) earnings.”  

Because of the importance of this information, the U.S. Securities and Exchange Commission requires that companies make it available to the general investing public in real time. But it can be difficult to wade through; calls frequently run 40 minutes to an hour, Martin said. A transcript averages about 7,000 words per call.

Institutional investors use AI to analyze these transcripts looking for company trends and expectations. Before the introduction of ChatGPT, this gave them a significant advantage over retail investors, Martin said.

She and her coauthors were interested in seeing how access to ChatGPT has changed that dynamic. To investigate the question, the team analyzed 19 years of earnings calls with ChatGPT, building a picture of the “AI sentiment” the calls revealed.

Then, they used data provided by regulators to pull out two sets of trading data from the same period — trades made by retail investors and those made by professional traders who use short-selling (borrowing a security to sell it and repurchasing it later at a lower price for a profit).

“Short sellers are, as a group, considered to be very sophisticated institutional investors,” Martin said.

The team compared how closely each group’s trading followed the AI sentiment they had previously uncovered. Short sellers followed AI sentiment before ChatGPT’s release, indicating that they had access to AI tools years before that point. “Our sample period starts in 2010,” she said. “It is likely that short sellers’ use of machine learning technology can go far back to 2010.”

Democratizing trading

On the other hand, retail traders’ trading patterns showed the influence of ChatGPT, aligning with AI sentiment almost immediately after its release. During ChatGPT outages (which are officially reported by its developer Open AI), that alignment fell away, providing further evidence of its influence, Martin said.

“People might argue this was a coincidental result — that there are other things happening along with generative AI arrival,” she said. “If indeed other things are driving the results, then we should expect the results to hold during outages of ChatGPT. What our evidence shows is that during outages, it almost went away.”

In fact, because it has more bandwidth to analyze longer documents, ChatGPT does a better job of predicting future stock returns than the professional-grade AI tool used by many institutions, FinBERT, the team found.

Martin said it’s not clear whether this new democratization caused by ChatGPT will hold up, as institutional traders develop more sophisticated tools and ChatGPT also becomes more robust. It’s something that she wants to continue to study.

“I would like to follow the path and see whether the information gap will increase or decrease,” she said. “I think it can go both ways. The technologies are evolving and will also interact with regulations that are going to develop to regulate AI in the capital markets. All of that could increase or decrease inequality in the marketplace.”

Martin’s coauthors on the study are Anne Chang and Xi Dong of Baruch College and Changyun Zhou of Southwestern University of Finance and Economics.