Fat and Fatter: Crash Risk and Retail Trading

Abstract

I estimate ex-ante crash probabilities for individual stocks via novel machine learning methodologies. In particular, I introduce imbalanced learning techniques to facilitate rare events prediction. I show that stocks with high crash probabilities tend to have lower returns. Further results indicate that at least a subset of retail investors, as proxied by Robinhood traders, tend to chase high crash risk stocks, which may bid up their prices and result in lower returns subsequently. Using Robinhood's introduction of commission-free option trading at the end of 2017 as a quasi-natural experiment, together with textual information from Reddit, I document causal evidence that retail participation significantly increases stock crash risk. This effect is stronger for small firms.

Keywords: Crash Risk, Cross-Section of Stock Returns, Imbalanced Learning, Machine Learning, Robinhood, Tail Risk, Wallstreetbets.

JEL Classification: G11, G12, G14