Modeling Earnings Expectations Based on Clusters of Analyst Forecasts

Abstract

This article introduces an approach to modeling market expectations that captures the benefits of both timelines and aggregation. The intuition behind the earnings expectation measure, referred to as the cluster mean, is that analysts' forecasts arrive at the market in a sequence of clearly distinguishable forecast clusters. At any time, the cluster mean expectation is simply the mean of all the forecasts issued in the most recent cluster. The tests show the cluster mean forecast has greater forecast accuracy, higher correlation with security returns, and lower serial correlation in forecast revisions than the consensus forecast, both for the entire tests sample and for a number of different subsamples.