A Time Series Analysis of Herd Investor Behavior Using Online and Social Media Data

Abstract

We examine the relationship between market performance of leading cryptocurrencies (Bitcoin and Ethereum), meme-stocks (AMC, GameStop), and subjects of corporate boycotts (Bud Light) using weekly market price and volume data along with social media data of weekly mentions (which total 337 million in this dataset) and net sentiment. Using vector autoregression (VAR) time series analysis along with Granger causality testing and structural breaks, we successfully predict trade volume of these various assets using social media data and price data. We also find that closing price data and trade volume are reliable predictors of net sentiment about crypto in online and social media. However, we struggle to predict the closing price for the group of assets studied. We also employ impulse response functions, finding evidence of a dynamic relationship occurring between online and social media net sentiment and online media volume with closing price and trade volume. These functions show that investor sentiment operates with a short memory lasting around 3 weeks, additionally these functions show that price generates a shock on trade volume but that crypto and meme-stock markets experience this differently. Our findings reinforce the notion that meme-stock traders and herd investors do not trade on market fundamentals but are instead sensitive to herding (or sentiment) movements. Our findings also suggest that compared to these meme-stock investors, crypto markets have more traditional motivations of loss aversion.