Inferring Conservatism from the Asymmetric Reporting of Accruals : A Conditional Heteroscedastic Modeling Approach

I develop an ex-post conservativism measure (that identifies conservatism based on whether firms understate or overstate earnings) rather than an ex-ante conservatism measure (that suffices as a governance mechanism [such as conservative accounting rules] to compel managers to be conservative). Prior conditional conservatism models infer asymmetric timeliness from public information (Basu, 1997; Khan and Watts, 2009; Collins et al, 2014) and operating cash flows (Ball and Shivakumar, 2005, 2006, 2008) rather than directly from discretionary accruals (DCA) – the tool available to and which managers intentionally use to understate or overstate earnings (Dechow et al, 1995). I thus propose a model (i.e. the augmented univariate conditional heteroscedastic [UCH] model) that mechanically removes linear dependence in accruals to isolate uncorrelated DCA, which is then used to model the asymmetric reporting of negative versus positive DCA. Intuitively, a firm using substantially more negative than positive DCA is characteristically conservative. I show that this measure positively covaries with extant measures, aligns with economic explanations (e.g. contracting, taxation, litigation, regulation, inter alia) driving conservatism, and explains other firm variables (e.g. crash risk, earnings management)