Methodological Choices In Detecting Divergent Earnings

Abstract

This paper empirically tests for methodological superiority in detecting divergent earnings (the difference between actual and expected earnings). Divergent earnings are generated using Value Line forecasted and reported earnings data. Two hundred random samples of 100 cases each are drawn. One hundred independent two sample tests are performed with 0%, 1%, 3%, 5%, 7%, and 10 % positive earnings introduced. The two sample tests are performed using both parametric (t test), and nonparametric (Mann Whitney test) statistics. They are performed on the “divergent earnings” data deflated by: 1) forecasted earnings , and 2) the market price of the stock. The results indicate that the superior alternative is nonparametric statistical methods based upon ranks, and the deflator choice under these nonparametric methods is of little consequence.