Why Don’t We Agree? Evidence from a Social Network of Investors (Summary)

The authors analyze disagreement in investment approaches to understand how much information sets influence investor disagreement as opposed to how much differential interpretations of information affect disagreement. Empirical research is typically unavailable to quantify these two sources of financial market disagreement because it is generally not feasible to predict an investor’s investing model. As a result, the authors analyze investors’ interactions on an investing social network where they self-report on their investing philosophy and designate messages as reflecting bearish or bullish sentiment. The data, which are grouped by investing philosophy, allow the authors to separate overall investor disagreement into within-group and cross-group categories, thus providing novel insights into the differential implications of investing model disagreement and information set disagreement.

The authors’ dataset is sourced from StockTwits, an investor’s social networking platform built for perspective-sharing about stocks. StockTwits provides data on all messages posted between 1 January 2010 and 30 September 2014, which total over 18.3 million message posts by over 107,000 unique users and mentioning 9,755 unique tickers. Posted messages were limited to 140 characters. For each post, the authors observe user identifier, message content, indicators for sentiment, and messages linked to specific stocks and tickers. The authors restrict their sample to message posts between January 2013 and September 2014 because the highest-quality data and best coverage are derived from more recent years. After imposing certain sampling restrictions, the final sample contains 1,442,051 posts by 12,029 discrete users.

To measure sentiment and disagreement, the authors use a maximum entropy-based method in order to incorporate unclassified messages and create a fully classified dataset that has 989,793 bullish messages and 452,258 bearish messages. A sentiment measure is constructed from bullish and bearish message data. The overall disagreement measure is calculated as the standard deviation of expressed sentiment across messages. Three key conclusions from their disagreement measures are presented: the link between the various disagreement measures and trading volume, conclusions on sophisticated versus unsophisticated investors (providing evidence of gradual information diffusion), and insight into how disagreement drives increased volume around earnings announcements. Finally, the authors provide two sets of robustness tests, suggesting that cross-group disagreement may be similar in the larger market despite the specific variations in investment approaches on StockTwits.

The authors find that within-group disagreement leads to 2.5 to 4 times the amount of trading volume as that of cross-group disagreement. The research should enjoy broad application because the authors show how disagreement influences nearly one-third of trading volume increases around earnings announcements. Still, disagreement does not account for most of the increased volume around earnings announcements. But this research does provide a novel perspective on the impact of minimizing information asymmetries by emphasizing the importance of within-group disagreement.