Several sensitivity tests are also conducted in the returns-earnings analyses. Fist of
all, we re-estimate the return-earnings model by partitioning the samples into positive
and negative earnings. Hayn (1995) suggests that the earnings-response coefficient is
low and not stable when earnings are negative. While we find that all the results are
qualitatively the same when the earnings are positive, there are inconsistent findings
when the earnings are negative. In addition, we also consider alternative measures of
return (12-month window ending at the fiscal year-end) and earnings, including both
the level and the change of earnings. The sensitivity checks do not affect the main
results, and the variable for change of earnings is generally not significant across all
model specifications. Finally, prior studies (Subramanyam and Wild, 1996) indicate
that the persistence and variability of earnings may explain earnings
informativeness[13]. Earnings persistence, earnings variability and their respective
interactive terms with earnings are then included in the returns and earnings model.
The models are re-estimated using both the pooled sample and the panel sample.
Neither interactive term is statistically significant in explaining returns, while the main
results of governance on return-earnings association remain unchanged.