Forecast Dispersion and Information Selection

I show that dispersion among equity analysts' forecasts increases following earnings announcements. I provide evidence that this is due to information selection: analysts rely on different subsets of public information revealed in earnings announcements when revising their forecasts in a high-dimensional setting. Using text data from analysts' reports, I show that analysts cite different information when they revise their forecasts following earnings announcements. Results from elastic net regressions corroborate these findings but further imply the existence of a common benchmark to which analysts can compare their forecasts. The common benchmark explains the decrease in forecast dispersion in the periods between earnings announcements. To formalize these findings, I develop a simple model of forecast-making processes under a common benchmark to explain the dynamics of forecast dispersion. My results highlight the importance of information selection as a driver of disagreement, especially in high-dimensional environments.