582 DETZEL ET AL.
cause of the accruals anomaly and determining whether it arises from risk or mispricing is fundamental
in understanding whether markets efficiently price accounting information.
Accruals are less persistent than the cash component of earnings; they negatively predict future
earnings, controlling for current earnings. The seminal study of Sloan (1996) argues that investors
fail to account for this lower persistence and subsequently over-value firms with high accruals. Con-
sistent with this ‘earnings fixation’ hypothesis, many studies find evidence of mispricing of accruals,
particularly for the least persistent and least reliable components (e.g., Green, Hand, & Soliman, 2011;
Hirshleifer, Hou, & Teoh, 2012; Momente, Reggiani, & Richardson, 2015; Richardson, Sloan, Soli-
man, & Tuna, 2005; Xie, 2001). However, other studies find that exposure to a risk factor explains a
large part of the accruals anomaly and interpret this as evidence of rational pricing (e.g., Chichernea,
Holder, & Petkevich, 2015; Guo & Maio, 2016; Khan, 2008).
Ball, Gerakos, Linnainmaa, and Nikolaev (2016) present evidence of a rational profitability-related
explanation for the accruals anomaly. They find that controlling for cash-based operating profitability
(COP), accruals no longer predict the cross-section of returns. Alternatively, others argue that accruals
predict returns because they measure real investment and firms should rationally respond to low dis-
count rates with high investment, all else equal (e.g., Fairfield, Whisenant, & Yohn, 2003; Wu, Zhang,
& Zhang, 2010; Zhang, 2007). Consistent with this ‘investment theory’, Wu et al. (2010) find that
exposure to an investment factor explains a large portion of the average returns on accruals-sorted
Overall, the literature documents mixed evidence for both mispricing and investment-versus-
profitability explanations of accruals. One explanation is that prior studies use varying definitions
of accruals. For example, the seminal study of Sloan (1996) uses ‘operating accruals’ (OA), defined
as working-capital accruals minus depreciation and deferred taxes. However, since OA omits impor-
tant long-term accruals -- such as capitalized intangibles and property, plant, and equipment (PP&E)
-- Richardson et al. (2005) propose the use of change in net operating assets (ΔNOA) as a measure of
total accruals. Wu et al. (2010) and Xie (2001) consider total accruals and/or groups of accruals based
on their reliability or level of reporting discretion. In particular, with the exception of Lewellen &
Resutek (2016), previous work generally does not decompose accruals into components that measure
new investment and those that measure transitory adjustments to earnings. To resolve the mixed evi-
dence in prior accruals studies, we hypothesize that the (total) accruals anomaly is in fact two separate
phenomena: a risk-based investment-accruals premium, and a mispricing of non-investment accruals
that is unrelated to risk.
Following Lewellen and Resutek (2016), we decompose total accruals into three components:
working-capital accruals (ΔWC), long-term investment accruals (IA), and long-term non-investment
or ‘nontransaction’ accruals (NTA). The IA component includes items such as new PP&E that repre-
sent expenditures in real investment. The ΔWC and NTA include items such as accounts payable and
receivable as well as depreciation that do not represent new long-term investment expenditures, but
only transitory accounting adjustments to cash flows. Hence we refer to ΔWC and NTA collectively
as ‘non-investment accruals’.
To test our hypothesis, we first investigate whether exposure to a risk factor explains the abnor-
mal returns on accruals-sorted portfolios. Risk-factor betas are highly correlated with accruals them-
selves and prior studies typically do not test whether these betas explain returns better than accruals.
In one exception, Hirshleifer et al. (2012) perform ‘characteristics-versus-covariances’ tests follow-
ing Daniel and Titman (1997) in which the former sort stocks on OA and then on OA-factor betas
within OA portfolios. This double sorting produces variation in OA-factor risk uncorrelated with OA.
They find that OA factor betas do not earn a return premium after controlling for OA, consistent with