We propose an empirically-motivated theory of business cycles, driven by fluctuations in sentiment towards a small number of firms. We measure firm-level sentiment with computational linguistics and analyst forecast errors. We find that 50 firms can account for over 70% of the unconditional variation in U.S. sentiment and output over the period 2006-2021. The “Granular Sentiment Index”, measuring sentiment towards the 50 firms, is dominated by firms that are closer to the final consumer, i.e. are downstream. To rationalize our findings, we embed endogenous information choice into a general equilibrium model with heterogeneous upstream and downstream firms. We show that attention centers on downstream firms because they act as natural “information agglomerators”. When calibrated to match select moments of U.S. data, the model shows that orthogonal shocks to sentiment of the 20% most downstream firms explain more than 90% of sentiment-driven (and 20% of total) aggregate fluctuations.
