Additional Evidence on the Time Series Properties of Reported Earnings Per Share: Comment

BALL AND WATTS ACKNOWLEDGE that the findings of their classic study of the time series properties of measured accounting income are somewhat limited because of an ex post sample selection bias against decreases in income [1, pp. 667-668]. They note, however, that the importance of the bias cannot be determined within their own selected sample of firms [1, p. 667]. The first purpose of this paper is to provide some evidence on the importance of their ex post sample selection bias against income decreases by applying some of their procedures to samples of firms which do not possess the same bias. Ball and Watts state that they were not interested in the behavior of the income time series of individual (outlier) firms [1, p. 680]. However, their stated conclusion that a particular single process-a submartingale process-characterizes the measured income series quite well [1, p. 680] implies (perhaps inadvertently) a similarity in the characteristics of the earnings processes of different firms. The second purpose of this paper is to display data which suggests that there may be more diversity than similarity in some of the characteristics of the earnings processes of different firms.