9IÖÆ×÷³§Ãâ·Ñ

Xiumin Martin

Professor of Accounting
Olin Business School, Washington University in Saint louis

AI (ChatGPT) Democratization, Return Predictability, and Trading Inequality

Date: Friday, March 21, 2025
Time: 10:30 AM – 12:00 PM
Location: Bronfman building, Room 046

Abstract

We present the first analysis of democratized AI’s (ChatGPT) role in investors’ trading activities, leveraging a 19-year dataset of earnings call transcripts, featuring extensive textual information. We have five key results. First, we develop a long-text (average 7,000 words) AI-sentiment measure that preserves full transcript context. AI-sentiment predicts returns of 1% per month, persisting 12 months, while human-dictionary - based (HD) sentiment shows little or negative predictive power. Second, over the prolonged period preceding ChatGPT’s wide-deployment, short sellers had already been trading in alignment with AI-sentiment following earnings calls, while retail traders had not. Post-deployment, retail traders’ alignment increases up to 23-fold, while short sellers’ alignment can diminish. Third, stocks with higher retail-AI alignment witness significant bid-ask spread reductions. Fourth, exogenous ChatGPT outages notably reduce retail-AI alignment and reverse bid-ask spread improvements. Fifth, AI-sentiment provides investors with long-term return insights by leveraging long-text context and discerning between genuinely and excessively positive HD-sentiment. These findings suggest that in an era of democratizing information access and proliferating big data, democratizing AI—the tools for analyzing information—is crucial to leveling the playing field between privileged and ordinary investors.