The Release of GPT-4 and Trading Activity in U.S. Equity Markets

Abstract

This paper analyzes changes in U.S. stock market trading activity around the release of GPT-4 in March 2023. We use a segmented interrupted time-series design with level-discontinuity and regression-kink components. The analysis uses monthly market-wide trading data from March 2020 to May 2026 and supplementary FINRA measures of ATS, non-ATS and total OTC trading. The preferred specification with a 24-month window before and after the event estimates a 22.6% increase in market-wide dollar volume at the event date. The estimate remains positive under multiple robustness checks. A growth-rate ARDL model also finds a significant event-month impulse, but not a sustained increase in subsequent growth. The estimate loses statistical significance once March 2023 is excluded; May and June 2023 are also plausible structural break dates, and the FINRA measures show no clear immediate discontinuity. The main result is therefore a positive but short-lived increase in trading activity that coincides with the release of GPT-4. The estimates do not establish that GPT-4 caused the increase. The modeled increase in volume reflects repeated turnover of identical securities, not new wealth, investment returns or real output.