Research

I fine-tune a large language model to develop a novel measure of social identification with a stock. My model categorizes investors as (i) tribalists, who strongly identify with a stock community, (ii) skeptics, who express disdain toward it, and (iii) social-neutrals, who lack strong identification. Using data from a major investment-focused social media platform, I find that tribalists have more followers than do social-neutrals and experience faster follower growth. Using transaction data from a Bitcoin focused social media site, I find that tribalists tend to wait less before buying after posting about Bitcoin. Tribalists transact faster than social-neutrals and skeptics after selling on a loss. Skeptics wait the least time before their next transaction after selling on a gain. Applying statistical social network models, I find that investors are more likely to follow or be followed by others who share the same social identity. I observe that while positive sentiment investors are more likely to form in-group connections (e.g., tribalists connect with tribalists), investors holding opposing sentiment (bullish vs. bearish) still form social ties with each other if they share the same social identity. Analysis of posts reply dynamics during a Bitcoin bubble reveals that skeptics are consistently negative, while tribalists are consistently positive, indicating echo chamber effects. Out-group interactions become more polarized during the bubble’s peak, with skeptics increasing negative replies and tribalists responding positively, particularly to skeptics.

We investigate the timing of SPAC mergers and subsequent deSPAC performance. Inspired by a simple target acquisition model, we employ duration analysis and find that SPACs often merge late, which is suggestive of high agency costs near the SPAC liquidation deadline. Contingency clauses can mitigate these costs by aligning manager and shareholder interests. Late merging SPACs without such clauses underperform, with 7-9% lower merger announcement returns and 30% lower deSPAC ROA. Notably, SPAC contingency clauses are only performative when agency costs are high, underscoring their importance in improving the performance of late-merging deSPACs.

I model the intertemporal behavior of a risky asset's price when the traders in the economy have limited attention and thus react differently to information signals with different salience levels. My model provides a clear connection between different information environments and the risky asset's corresponding price path. Both the prominence of the news, good or bad, and how much investors react or value such news directly determine the intertemporal behavior of risky asset prices. Furthermore, my model describes the necessary information environment to produce a risky asset price bubble. My model demonstrates that only two conditions are needed to create a price bubble: (i) a series of information shocks that include a positive error term and (ii) the economy is populated by investors who react strongly to non-fundamental news.