04.09.2013 Views

Modelling the accruals process and assessing unexpected accruals*

Modelling the accruals process and assessing unexpected accruals*

Modelling the accruals process and assessing unexpected accruals*

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Hence, <strong>the</strong> correlation between abnormal <strong>and</strong> normal <strong>accruals</strong> could be a plausible explanation as<br />

to why even well-specified <strong>accruals</strong> models are powerful in detecting abnormal <strong>accruals</strong> only in<br />

very restrictive circumstances. When abnormal <strong>and</strong> normal <strong>accruals</strong> are correlated, <strong>the</strong> expected<br />

component will unavoidably embed some abnormal <strong>accruals</strong>. As a result, <strong>unexpected</strong> <strong>accruals</strong><br />

extracted from <strong>the</strong>se models will not capture abnormal <strong>accruals</strong> in full. Specifically, when <strong>the</strong>re<br />

is no heterogeneity between firms, <strong>the</strong> r-squared of <strong>the</strong>se models will be close to one <strong>and</strong> <strong>the</strong><br />

model fully explains <strong>the</strong> variation in <strong>accruals</strong> which consist of both abnormal <strong>and</strong> normal<br />

<strong>accruals</strong>. The resulting <strong>unexpected</strong> <strong>accruals</strong> will be close to zero even for serious earnings<br />

managers. Since <strong>the</strong> high r-squared of a model can be due to its ability in correctly attributing<br />

normal <strong>accruals</strong> as <strong>the</strong> expected component or incorrectly attributing abnormal <strong>accruals</strong> as <strong>the</strong><br />

expected component, this raises doubt over whe<strong>the</strong>r one should judge a model‟s ability to<br />

correctly extract abnormal <strong>accruals</strong> by its r-squared.<br />

In short, <strong>the</strong> correlation between abnormal <strong>and</strong> normal <strong>accruals</strong> inevitably leads well-specified<br />

<strong>accruals</strong> models to incorrectly attribute abnormal <strong>accruals</strong> as <strong>the</strong> expected component <strong>and</strong> less<br />

well-specified <strong>accruals</strong> models to incorrectly attribute normal <strong>accruals</strong> as <strong>the</strong> <strong>unexpected</strong><br />

component. The zero conditional assumption imposed by ordinary least squares requires <strong>the</strong><br />

disturbance term to be uncorrelated with <strong>the</strong> explanatory variables. The correlation between<br />

abnormal <strong>and</strong> normal <strong>accruals</strong> exactly violates this assumption. As a result, <strong>the</strong> approach to<br />

identify earnings management via a least square disturbance term of <strong>the</strong> regression is likely to be<br />

problematic. This echoes <strong>the</strong> points made in McNicols (2000).<br />

40

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!