Talking to Digital Twins: Selective Disclosure and Belief Measurement in Financial Social Media

Financial influencers (finfluencers) on social media platforms like X significantly impact market dynamics. However, relying on public posts to measure their influence creates a selection bias: researchers only observe what influencers choose to disclose. This leaves the "silent region"—the vast majority of stocks an influencer may have an opinion on but chooses not to post about—unobserved. This paper asks whether it is possible to systematically measure these hidden beliefs and whether they contain predictive information about future stock returns.

To overcome the measurement problem of selective disclosure, the authors construct "digital twins" of financial influencers. These are AI-based agents built from the influencers' public account history and recent content. The researchers implement a fixed, daily interview protocol where these digital twins are queried about their buy, hold, or sell recommendations for a consistent set of large-cap stocks. Because these interviews are conducted and archived in real-time, the study avoids the look-ahead bias common in retrospective LLM analyses. The researchers then test whether the resulting "Net Buy Share" (the difference between buy and sell recommendations) predicts stock returns over the subsequent ten trading days.

Validation tests show that digital twins accurately mirror their human counterparts, with recommendations matching 91.5% of the time. The study finds that the interview-based signals are highly predictive of future returns, particularly in the "silent region" where the human influencers made no public posts. Specifically, a ten-percentage-point increase in the buy-minus-sell interview measure predicts a 50-basis-point increase in excess returns over a ten-day horizon. This suggests that the digital-twin method successfully recovers economically meaningful belief proxies that are otherwise invisible to traditional sentiment analysis.