When ChatGPT goes dark, Wall Street slows down. During eight major ChatGPT outages in 2023, trading in US stocks fell noticeably, a sign that many investors were already using the chatbot in ways that shape their trades.
The study is by Qiang Cheng, Pengkai Lin and Yue Zhao of Singapore Management University. It was published online on August 13, 2025, in the Journal of Accounting and Economics.
Outages as a natural experiment
Because nobody can see how investors use ChatGPT, the authors used its unexpected failures as a test. They identified eight major outages between February and August 2023 that overlapped with US trading hours, using OpenAI’s incident reports and user complaints logged on DownDetector.
The outages lasted from 18 minutes to four hours, 119 minutes on average, and fell on different days and at different times. The authors checked news reports and found no other major service failures at the same moments.
They then compared trading volume in five-minute intervals during each outage with the same intervals on the previous five trading days. The final sample held 2,615,434 observations covering 2,553 stocks.
A delayed, measurable dip
Trading volume fell during outages by about 5.55% of its standard deviation, or roughly 9.88% of its average level. For comparison, the authors cite an earlier study that found a 1.87% drop in volume when distracted driving laws cut phone use during commutes.
The decline did not start the instant ChatGPT failed. It appeared about 30 minutes in, which the authors read as a sign that investors used the tool to process information and handle routine work rather than to place trades directly.
The drop was bigger for companies that had released news that day before the outage ended, when there was more information to digest. Both retail and non-retail trading fell, with a larger effect for non-retail volume and for stocks with more ownership by transient institutional investors.
Outages of OpenAI’s programming interface also coincided with a significant fall in algorithmic trading volume, suggesting some trading algorithms had the service built in.
Who gains from the chatbot
Market quality shifted too. Short-term price impact and return variance fell during outages, consistent with less informed trading, and bid-ask spreads narrowed, consistent with less information asymmetry among investors.
Taken together, the authors suggest ChatGPT may have widened the edge of institutional investors over retail ones rather than leveling the field. They describe these market-quality effects as modest in size.
Looking at the longer run, stocks whose trading dropped most during outages saw a larger improvement in price informativeness after ChatGPT launched. The authors caution that this measure is noisy and open to selection bias, so it does not prove cause and effect.
They also cannot observe what investors actually ask the chatbot. A field experiment that randomly gives investors access to such tools would be the cleaner test, they write.
Study Details:
- Title: Does generative AI facilitate investor Trading? Early evidence from ChatGPT outages
- Authors: Qiang Cheng, Pengkai Lin, Yue Zhao
- Journal: Journal of Accounting and Economics, Volume 80
- Publication Date: August 13, 2025
- DOI: 10.1016/j.jacceco.2025.101821