Patterns in Analysts' Long-Term Earnings Forecasts

Abstract

This article examines the accuracy of analysts' long-term estimates (made before the beginning of the year being forecast) of annual operating earnings during the 1986–2002 period, as provided in the Thomson First Call I/B/E/S database. The article asks whether an investor can use the pattern of a particular analyst's earnings estimates as a way of discerning analyst accuracy; that is, whether there are biases or signals in those patterns. Examining the change between consecutive earnings estimates, we find larger changes (negative or positive) are associated with lower forecast accuracy. Examining the difference between an analyst's estimate of earnings and the previous year's actual earnings in joint tests, we find that estimates predicting larger increases in earnings are decreasingly accurate, but those predicting larger decreases in earnings are increasingly accurate. Examining the number of estimates produced by an analyst as a signal of quality, we find no relationship in univariate tests, although one has been reported previously in the literature; in joint tests, a negative relationship is found between forecast frequency and accuracy.