Identifying Short Selling Regimes in the Post-Meme-Stock Era

Abstract

Short interest has long served as a robust predictor of negative abnormal returns. The short-selling business fundamentally changed, however, after the meme-stock episode of 2021, raising questions about the continued informativeness of the short-interest signal. This thesis examines whether the predictive power of short interest has weakened in the post-meme era, and whether observable market conditions and retail attention metrics drive transitions in signal informativeness. Using a panel of U.S. equities from January 2010 through December 2024, we construct portfolios sorted on lagged short interest and estimate Carhart four-factor models across pre- and post-meme subsamples. Equal-weighted portfolios of heavily shorted stocks, which generated significant negative alphas in the pre-meme period, produce insignificant positive alphas after January 2021. We then estimate a Time-Varying Transition Probability Hidden Markov Model (TVTP-HMM) that endogenously identifies the two regimes: an Informed regime, in which short interest predicts negative abnormal returns, and a Disrupted regime in which the signal weakens or reverses. We observe similar, significant, positive alpha differentials between Informed and Disrupted regimes as the subsample analysis and a substantial increase in return variance and momentum and small-cap factor coefficients. Transition dynamics reveal that VIX and the equity put/call ratio drive entry into disruption, while retail attention concentration shocks and ticker breadth govern persistence. The Disrupted regime accounts for roughly 4-13% of pre-meme months but 46-65% of post-meme months, suggesting that the impairment of short-interest informativeness has become a semi-permanent feature of the post-meme market environment.