Research

AI (ChatGPT) Democratization and Trading Inequality  (with Xi Dong, Xiumin Martin, and Changyun Zhou )

Revise and Resubmit, Journal of Accounting Research

Presented at: JAR Conference 2025, CICF 2025 (scheduled), AI in Finance Conference 2025 (scheduled), AAA Annual Meeting 2025 (scheduled), MFA 2025, FARS 2025, 16th Annual Hedge Fund Research Conference 2025, CFEA 2024, 1st Workshop on LLMs and Generative AI for Finance, “AI Era in Finance” Symposium​ 2024, Australian National University, Baruch College, McGill University, Monash University, Washington University in Saint Louis

Abstract: We present the first analysis of the influence of democratized AI’s (ChatGPT) on investors’ trading activities. We develop an AI-sentiment measure using earnings conference calls. We find that before the introduction of ChatGPT, short-selling activities exhibited alignment with AI-sentiment, whereas retail trading did not. However, following the wide deployment of ChatGPT, we observed a significant increase in the AI alignment of retail traders, accompanied by a decrease in the alignment of short sellers, implying that AI contributes to reducing the trading inequality between retail investors and short sellers. We further find that the primary mechanism driving the AI democratization effect on retail trading is its ability to lower information processing costs for retail investors. Lastly, AI-sentiment positively predicts returns both before and after the democratization of AI in a similar way, with no subsequent return reversal observed in the long run. This evidence suggests that AI-sentiment effectively captures fundamental information, and the increased alignment of retail investors' trading with AI-sentiment does not appear to have a measurable impact on price efficiency. To strengthen our causal inferences, we examined the impact of exogenous ChatGPT outages. These outages significantly reduced the alignment between retail and AI-sentiment, thus reinforcing our conclusions.