The attention economy of retail trading: Evidence from Robinhood and social media

Is retail participation driven by what investors notice or how they feel? This study develops a hierarchical signal-processing framework and tests it using daily changes in Robinhood holders for 427 U.S. stocks from May 2018 to August 2020. Attention and sentiment are measured using cross-platform principal components from StockTwits, Twitter, and Seeking Alpha. In baseline and stratified regressions, attention is the more consistent conditional predictor: a one-standard-deviation increase corresponds to an expected holdings change equal to 68% of the median absolute daily change, after conditioning on the full control set. This association holds across most firm- and market-state stratifications but weakens under information-competition regimes, notably elevated presidential-tweet volume, consistent with limited-attention crowding out. Sentiment is insignificant on average but informative in validating contexts: positive tone near 52-week highs reinforces momentum narratives, while negative tone matters when valuation or short-interest signals make skepticism salient. Panel vector autoregression (PVAR) evidence qualifies these patterns: in the unconditional system, holdings feed back into attention, and short-horizon directionality is not identified from the data alone; in the state-conditional system, attention’s predictive content becomes significantly negative under high macroeconomic-news competition. Together, the results cast visibility as a consideration-set filter and tone as a context-dependent validation signal.