The Folly of Crowds: Social Networks, Asymptotic Learning, and Asset Bubbles

Abstract

In January 2021, an explosion in the stock price of GameStop (GME) triggered worldwide attention as a group of retail traders on the WallStreetBets (WSB) subreddit succeeded in sending the stock’s price to previously unthinkable heights. The episode demonstrated the power of social interactions on a single forum to move markets. Investors, and the world, took notice. That same year, the stock prices of AMC Entertainment Holdings (AMC), BlackBerry (BB), Bed Bath & Beyond (BBBY), and many other “meme stocks” surged as a WSB users hyped the stocks on the forum and coordinated their trades. In 2022, the meme stock frenzy has waned, but the role of social interactions in moving asset prices has never been clearer.

The goal of this thesis is not to tell the stories of individual Reddit-driven asset bubbles: plenty of research into meme stocks has already been done. Rather, this thesis aims to understand how the structure of user interactions on forums like WSB allows large asset bubbles like GME to occur. To accomplish this goal, we represent user interactions using two different models of social learning: naive DeGroot learning and sequential Bayesian learning. For each model, we determine conditions that either guarantee or preclude asymptotic learning.

We then examine a set of WSB submissions spanning the first 8 years of the forum’s history to determine whether user interactions on financial social networks lead to or prevent asymptotic learning. We find that several features of networks that prevent asymptotic learning — specifically those related to the distribution of user influence — are present on WSB.