Following prior research (Venter et al., 2014), subsamples are selected from the main sample before trimming (2,755 observations in Table 1) and are individually trimmed thereafter. The result is that the sum of observations for the subsamples does not always add up exactly to the number of observations in the main sample.

Results are from estimating model (2) with ADJUST omitted. The dependent variable is MV, the cum dividend market value of equity, three months after the reporting date; BV is the book value of equity; GAAP is GAAP earnings; HEAD is headline earnings; DISC is discretionary earnings; COMB is discretionary earnings if reported and headline earnings if not; NEG is an indicator variable set to one if GAAP is negative and zero otherwise. All variables, other than indicator variables, are scaled by number of shares outstanding. Two-tailed p-values based on robust standard errors clustered by firm and year (Petersen, 2009; Cameron et al., 2011) are reported in brackets. An adjustment is made for individual variables where the variance-covariance matrix is not positive-semidefinite.

***,

** and

* denote significance at the 1%, 5% and 10% levels, respectively. The Vuong test (Vuong, 1989) is directional so that a negative test statistic indicates that the second model is superior to the first