ABSTRACT
Users of I/B/E/S data generally act as if I/B/E/S reported actual earnings represent the earnings analysts were forecasting when they issued their earnings estimates. For example, when assessing analyst forecast accuracy, users of I/B/E/S data compare analysts' forecasts of EPS with I/B/E/S reported actual EPS. I/B/E/S states that it calculates actuals using a “majority rule,” indicating that its actuals often do not represent the earnings that all individual analysts were forecasting. We introduce a method for measuring analyst inferred actuals, and we assess how often I/B/E/S actuals do not represent analyst inferred actuals. We find that I/B/E/S reported Q1 actual EPS differs from analyst inferred actual Q1 EPS by at least one penny 39 percent of the time during our sample period, 36.5 percent of the time when only one analyst follows the firm (hence, this consensus forecast is based on the “majority rule”), and 50 percent of the time during the last three years of our sample period. We document two adverse consequences of this phenomenon. First, studies failing to recognize that I/B/E/S EPS actuals often differ from analyst inferred actuals are likely to obtain less accurate analyst earnings forecasts, smaller analyst earnings forecast revisions conditional on earnings surprises, greater analyst forecast dispersion, and smaller market reaction to earnings surprises than do studies adjusting for these differences. Second, studies failing to recognize that I/B/E/S EPS actuals often differ from analyst inferred actuals may make erroneous inferences.
Returns to buying upward revision and selling downward revision stocks
