Generative AI and the stock market: How ChatGPT downtime moves shares

By Vince Chong

SMU Office of Research Governance & Administration – The growing influence of generative artificial intelligence, or GenAI, in daily life is evident everywhere. With its ability to learn and produce text, images, audio, and code, GenAI is now widely used for tasks ranging from simple research to producing professional reports and analyses across diverse industries. 

Unsurprisingly, this includes financial markets, where many investors rely on GenAI to guide their decisions. In one of the first studies of its kind globally, SMU Assistant Professor of Accounting Lin Pengkai examined how GenAI usage influences trading activity across investor groups. 

His paper Does generative AI facilitate investor trading? Early evidence from ChatGPT outages was published in the prestigious Journal of Accounting and Economics and cited by top regulators, including the Financial Stability Board (FSB) and the US Public Company Accounting Oversight Board (PCAOB).

Co-authored with SMU colleagues Professor Cheng Qiang and SMU doctoral candidate Zhao Yue, the study track trading changes in the US stock exchanges during ChatGPT outages. ChatGPT is widely regarded as the leading GenAI engine since its launch in November 2022.

The team found eight major outages that happened during trading hours between February and August 2023, with each period ranging from about 18 minutes to four hours, averaging roughly two hours. As the outages occurred on different days and times, changes in trading were unlikely to be explained by regular market patterns. The researchers also checked news sources to make sure that no other services failed at the same times.

The results were striking: data from the NYSE Trade and Quote database showed “a significant decline in trading volume during the outage periods.” To put the effect into perspective, Professor Lin cited previous research showing that turnover in stock trading fell by three to seven percent on the first full day of an electrical outage affecting 0.5 percent of US electrical customers. By comparison, the ChatGPT outages equated to roughly a 10 percent decline relative to the average level of trading volume.

The impact of his team’s findings is such that the FSB cited their study “as an example of academic methods for capturing the impact of AI adoption on asset price volatility,” Professor Lin told the Office of Research Governance and Administration (ORGA). Similarly, the acting chair of the PCAOB referenced the paper as “growing evidence that investors and analysts are using AI tools to interpret financial information as they make investment decisions.”

“Broadly, I believe our paper can inform regulators about the prevalence of GenAI uptake in the financial sector and its initial effects on price efficiency.” 

Usage details

One key finding was that ChatGPT is unlikely to be used for direct trading execution, as there was no immediate decline in trading volume when outages began. Instead, the study found a marked decline only about 30 minutes after the outages started.

The decline was sharper for companies releasing corporate news on that same day as the outages, but before these glitches ended. This “suggests that ChatGPT outages have a more pronounced impact on investors who rely on it for timely information processing and trading decisions.”

Another implication of the study challenges the view that GenAI helps to level the playing field for retail investors. While these users can access sophisticated investing analyses once reserved for those with outsized resources, GenAI instead “enhances the information advantage of institutional investors,” the research noted. As it observed, larger declines in trading volume were registered among such professionals, compared to retail traders.

Whether GenAI can ultimately level the playing field remains under debate, Professor Lin said. The evidence so far is mixed. 

“Conceptually, whether AI can level the playing field depends on the relative benefits the two types of investors can reap from the tool, subject to their respective cost, knowledge, and skill constraints,” the researcher said.

While GenAI does make it easier for retail investors to collect and process information, its potential can be further unlocked by “more advanced functions, such as automatic, real-time information processing using APIs (application programming interfaces).” 

These may “particularly benefit institutional investors that are resourceful in both talent and organisational capacity,” he said. 

Potentially more pros than cons

As the study qualified, its observations do not reveal the specific ways in which investors use ChatGPT. Future studies could employ surveys or field data to better understand how investors use ChatGPT – whether for “non-trading tasks, information processing tasks, and/or direct trading decisions, such as generating trading recommendations,” the report noted. Another avenue, Professor Lin said, could explore how investor habits shift if other GenAI engines like Gemini and Claude gain popularity, reducing reliance on a single tool.

Still, the preliminary evidence observed in Does generative AI, using several price efficiency measures, implies that “the benefits of using ChatGPT in trading may outweigh its potential risks,” despite concerns about its tendency to provide false information. 

While more research is required, such benefits have been partly backed by concurrent studies, Professor Lin said, including one that found that "hedge funds adopting [GenAI] earn two to four percent higher annualised abnormal returns than non-adopters.”

“What is clear,” he concluded, “is that such technology is revolutionary and here to stay. I am optimistic that we will continue to find more ways to benefit from it.”

Back to Research@SMU February 2026 Issue