Estimating a model of herding behavior on social networks

Abstract

In this paper, we estimate an agent-based model (ABM) to investigate herding behaviors in the formation of investor sentiment. We formalize a simple opinion dynamics model in a social network framework and rely on a numerical method to estimate its parameters. We derive a sentiment proxy from the weekly aggregation of online messages concerning 15 US stocks and 5 cryptocurrencies. Our empirical results suggest a strong impact of herding behavior on the formation of sentiment toward highly volatile assets. For such assets, we simultaneously find limited impacts of financial returns and investor attention on the opinion formation process, suggesting that investor sentiment is explained by social interactions. On the other hand, we find a limited influence of social interactions on sentiment regarding less volatile assets, whose formation process is instead explained by the strong influence of financial returns and investor attention. In particular, we find that herding behavior was significantly higher and played a major role in the sentiment formation process regarding cryptocurrencies when the bubble occurred.