Numbers Games? A Natural Experiment Investigating Three (Ir)Regularities in Reported Earnings Per Share

We confirm the existence of three irregularities noted in prior research related to reported earnings per share (EPS). First, the unusual pattern in the second digit of reported EPS noted by Thomas (1989) that the second digit of EPS is more likely to be zero and five and less likely to be nine for profit firms while the pattern does not appear for loss firms. Second, the rounding pattern in (unreported) third decimals of EPS noted by Das and Zhang (2003) that the one tenth of a cent is more likely between five and nine for profit firms. Third, the threshold irregularity in EPS changes attributed to Degeorge et al. (1999) and related to Burgstahler and Dichev (1997), that zero changes and one-cent increases are over-represented and one-cent decreases are underrepresented, relative to expectations. Our study confirms that these patterns exist and enhances confidence in each reported irregularity's validity by identifying matched pairs of firm-years immediately prior to the adoption of SFAS 128. We further extend prior research on EPS irregularities by conducting analysis to discern whether the second digit pattern appears to dominate the threshold irregularity or vice-versa