We follow studies such as Francis et al. (2005), Verdi (2006), Doyle et al.
(2007), and Bharath et al. (2008), which used cross-sectional models of accruals
quality. Specifically, as pointed out by Bharath et al. (2008), the cross-sectional
estimates overcome the severe restrictions that apply to time series models,
which introduce survivorship bias in the sample because they need to use
firm-specific time series data. Furthermore, cross-sectional estimation controls
for changes in accruals due to business cycle effects. Moreover, since studies
such as Subramanyan (1996) and DeFond and Subramanyan (1998) gave evidence
that cross-sectional models of abnormal accruals are better specified than their
time-series version, it is more usual in the earnings management literature to
use cross-sectional models, and given that we use different measures of abnormal
accruals in order to test the robustness of our results, we opt to use cross-sectional
models.