Methodological Choices In Detecting Divergent Earnings: An Extension

Abstract

Divergent earnings are those that differ from expected.  Doran (2000) provides evidence that nonparametric tests based upon rank values are superior to parametric alternatives in detecting divergent earnings.  He also finds that deflator choice (i.e., forecasted earnings or market price of the stock) is of little importance when superior nonparametric methods are used.  This study extends the efforts of Doran (2000) by testing for benefit derived from the common research method of deflating earnings data.  The data used here is the same as used by Doran (2000), where Value Line is the source of all earnings data.  One hundred independent two sample tests are performed between a positive earnings group and a matched control group.  The tests are performed with various levels (1%, 3%, 5%, 7%, and 10 %) of positive actual earnings introduced.  Failure to reject the null hypothesis of no positive earnings divergence indicates the existence of Type II error (determined using the nonparametric Mann Whitney test).  The Mann Whitney test was performed on the undeflated data, and the same data deflated by: 1) forecasted earnings, and 2) market price of the stock.  Difference in frequency of type II error is determined using the Chi-square test.  The results generally indicate no significant difference in the ability to identify abnormal divergent earnings when utilizing deflated data.  Statistical tests are found here to be at least as powerful when undeflated earnings data are used.  There is weak evidence supporting the notion that deflating earnings data inhibits the ability to detect abnormal earnings.  These findings indicate that the common practice of deflating earnings data is unnecessary, and may actually weaken the power of statistical tests.