The influence of the accruals generating process on earnings persistence

Abstract

In this paper we categorise accruals on the basis of how they are generated, and empirically evaluate whether this categorisation provides additional insights into future earnings and is relevant to the estimation of firm value. Specifically, we categorise accruals on the basis of whether the underlying cash flows lead or lag earnings recognition, and whether the accruals are initiating or reversing (i.e. a four-way categorisation). We demonstrate that these accrual categories are not homogeneous, have differing implications for earnings persistence and are relevant for firm valuation. Significantly, where cash flows lag earnings recognition (e.g. sales made on credit) they have greater implications for future earnings than where cash flows lead earnings (e.g. unearned revenues) and depreciation. Similarly, initiating accruals have greater implications for the persistence of earnings than reversing accruals. Paradoxically, while depreciation exhibits high persistence, it has less of an implication for the persistence of earnings than either lag or initiating earnings categories. These findings enhance our understanding of the properties of accounting income and how it is impounded into share prices.

Keywords

Accrual anomaly accruals categories accruals components

1. Introduction

Accrual accounting dictates that earnings are recognised in the financial reports in the period in which they are generated, and this is independent of when underlying cash flows occur. Specifically, accruals may arise from cash flows that either lead or lag earnings recognition, and they may be initiating or reversing. Recognising this diversity in accruals, an issue requiring address is whether recognition of how accruals have been generated provides additional insights into differences in the persistence of accruals and earnings. This suggests two important research questions. First, do differences in the process by which accruals are generated have implications for earnings persistence which is relevant to the evaluation of the accrual anomaly identified by Sloan (1996)? Second, does information about how accruals are generated help in predicting earnings which is relevant to the estimation of firm value?

The accrual anomaly was identified by Sloan (1996) who provides evidence of the cash and accruals components of earnings having different implications for the persistence of earnings, and these differences not being fully reflected in stock prices. Investors appear to overestimate the implications of accruals for the persistence of earnings and this leads to the systematic mispricing of accruals. A substantial literature has since developed evaluating if the anomaly is an artefact of research design. Hence a specific objective of this study is to extend this literature, focusing on the assumption that accruals are homogeneous and have equal implications for earnings persistence. This assumption is potentially problematic as it fails to recognise the diversity that exists within accruals and the implications this may have for earnings persistence. Accordingly, we evaluate the robustness of the accrual anomaly, having regard to the process by which accruals are generated, and recognise that accruals can arise from anticipated cash flows that may lead (e.g. unearned revenue) or lag (e.g. credit sales) earnings recognition. We also recognise that accruals may be the first recording of a transaction and hence be an initiating accrual (e.g. credit sales) or the completion of the transaction that is a reversing accrual (e.g. receipt of accounts receivable). This is a four way categorisation of accruals.

A more general objective of this study is to evaluate whether the accruals generating process provides additional information that is relevant to predicting earnings and estimating firm value. While there is evidence that earnings and cash flows each have incremental information content for investors (Bowen et al., 1987), there is a stronger association of stock returns with earnings than cash flows (Dechow, 1994). This suggests that accruals, or the processes of transforming cash flows into earnings, convey relevant information to investors about expected firm performance and firm value. Some support for this view is provided by Barth et al. (2001) who find that the disaggregation of accruals into major categories aids in the prediction of future cash flows. Barth and Clinch (2013) develop this proposition further and consider analytically how the nature of accruals may provide ‘other information’ (Ohlson, 1995) which is relevant to firm valuation. We evaluate this empirically by considering whether the accruals generating process provides insights into future earnings and is relevant to the estimation of firm value.

The common theme in addressing these research questions is that they involve evaluation of the relation between accruals, cash flows and future period earnings. A central tenet of the accrual anomaly is that the accruals and cash flows categories of earnings have different persistence, and different implications for earnings persistence. In contrast, accruals are assumed to be homogeneous, and have equal implications for earnings persistence. A possible explanation for the accrual anomaly is that it is simply a consequence of an error in modelling the relation between accruals categories and future earnings. More generally, if different categories of accruals have different implications for earnings persistence, the composition of accruals is potentially a source of ‘other information’ which is relevant to the prediction of future earnings and estimating firm value (Barth and Clinch, 2013).¹

The potential for accruals categories to have different implications for earnings persistence is recognised in a substantial literature that has evaluated the accrual anomaly in which various definitions and partitions of accruals have been considered. For instance, attention has been focused on operating earnings and working capital accruals. Such accruals will likely reverse within one year and hence are more likely to exhibit similar persistence (e.g. Dechow and Dichev, 2002). Accruals have been disaggregated into current and non-current categories, with non-current accruals being found to be highly persistent but then having relatively few implications for earnings persistence (Richardson et al., 2005). Accruals have also been disaggregated into discretionary accruals and non-discretionary accruals, with the former expected to be less persistent (e.g. Chen and Cheng, 2002; Kothari, 2001; Richardson et al., 2006).² However, this disaggregation is problematic because it relies on the estimation of discretionary accruals, making it disposed to error (e.g. Bernard and Skinner, 1996; Shan et al., 2010). A finer disaggregation of accruals has been undertaken recognising the balance sheet accounts to which they relate (e.g. Chan et al., 2006; Thomas and Zhang, 2002). Critically, while providing some insights into the nature of accounting information, this literature has not provided an adequate explanation of the accrual anomaly.³ Furthermore, whether the categories of accruals provide ‘other information’ which is relevant to the determination of future earnings and firm value is relatively unexplored.

An alternative disaggregation of accruals is pursued in this paper which focuses on the accruals generating process. This recognises that accruals may result from cash flows in periods before earnings recognition (i.e. cash flows lead earnings recognition), or as a consequence of estimated cash flows that occur after earnings recognition (i.e. cash flows lag earnings recognition). Furthermore, accruals will initiate in one period, and reverse in subsequent periods. Potentially each accruals category has different implications for earnings persistence and firm valuation. Thus, separate analysis of the categories of accruals potentially provides ‘other information’ about earnings persistence than aggregate accruals. If this is recognised by investors it may at least in part explain the accrual anomaly. More importantly, it may also provide insights into the properties of accounting information and how this conveys information to investors which is impounded in share prices.

The research approach adopted in this paper follows that of Sloan (1996), with the sample period extending to 2006, enabling consideration of the anomaly subsequent to its initial identification.⁴ Consistent with Sloan’s finding of the market overpricing accruals, further evidence is provided of investors overestimating the contribution of accruals to the persistence of earnings; this extends to periods subsequent to the publication of Sloan. However, perhaps more importantly we demonstrate that the categories of accruals are not homogeneous as assumed by Sloan, and they exhibit differing implications for the persistence of earnings. Significantly, where cash flows lag earnings recognition (e.g. sales made on credit) they have greater implications for the persistence of earnings than when cash flows lead earnings (e.g. unearned revenues) and depreciation. Similarly, initiating accruals have greater implications for the persistence of earnings than reversing accruals which is consistent with findings regarding growth (Allen et al., 2010; Fairfield et al., 2003; Zhang, 2007). While depreciation itself is highly persistent, it has fewer implications for the persistence of earnings than other accruals, which is again consistent with the prior literature (e.g. Francis and Smith, 2005; Richardson et al., 2005). Finally, in addition to evidence of accruals continuing to be overpriced, evidence also exists for investors underpricing cash flows (Clinch et al., 2012; Coakley et al., 2008; Houge and Loughran, 2000). These findings provide a more detailed insight into the nature of earnings categories (both cash flows and accruals categories) and their implications for earnings persistence.

Critically, we provide evidence that earnings categories are potentially diverse, both in terms of their nature as well as their implications for earnings persistence. This suggests further evaluating how these earnings categories vary across firms, and confirms the categories of accruals as potentially a source of ‘other information’ that is relevant to the estimation of firm value, as suggested by Barth and Clinch (2013). This is also relevant for a significant body of financial accounting research which considers the information properties of accounting income and assumes a consistent relation between earnings and price. At a minimum it suggests re-evaluating prior studies, where it is likely that earnings categories are diverse in nature, and considering how this impacts the findings.

The remainder of the paper is organised as follows. Section 2 describes the categorisation of accruals undertaken in this paper. The research design employed is described in Section 3 while the sample data and descriptive statistics are provided in Section 4. The results are presented in Section 5, as are the various sensitivity tests. Finally, Section 6 discusses outcomes and conclusions.

2. The disaggregation of earnings

The central concern of this paper is whether the various cash flow and accrual categories of earnings have different implications for earnings persistence – information which is used by investors in firm valuation. This paper builds on significant literature that has developed since Sloan (1996) that evaluates the accrual anomaly.⁵ In this literature much attention has been focused on the research design employed, and in particular the assumption that accrual categories have equal implications for the persistence of earnings. This paper considers whether this assumption is appropriate and if the accrual anomaly is a consequence of an error in modelling the relation between accrual categories and future earnings. A wider concern is whether differences in accrual categories provide other information that is relevant to estimating future firm performance and firm value. Accordingly, we extend the literature which considers the implications of cash flows and accruals for future earnings by disaggregating accruals on the basis of the accruals generating process. This is outlined below.

Underpinning the literature evaluating the implications of accruals for earnings persistence is the disaggregation of earnings into accruals and cash flows categories, shown in the equation.

{Earnings}_{it} = {Accruals}_{it} + {CF}_{it}

(1)

Where:

Earnings_it	:	Earnings for firm i in period t;
Accruals_it	:	Accruals for firm i in period t;
CF_it	:	Cash flows for firm i recognised in the same period earnings (t).

The implications of the earnings categories for earnings persistence is then evaluated, together with whether these implications are reflected in stock prices. A critical feature of this approach is that it considers the accruals to be homogeneous, and fails to recognise differences that may exist within accruals which may be relevant to determining the persistence of earnings and firm value.

Notwithstanding, differences between accruals categories are recognised in the literature considering the relation between accruals and cash flows over successive periods (e.g. Dechow and Dichev, 2002; Francis and Smith, 2005; Pae, 2005). Here, it is common for accruals to be differentiated on the basis of when the underlying cash flows occur. This is shown in the following equation in which earnings, and in particular accruals, are categorised on the basis of when the cash flows occurred.

{Earnings}_{it} = {CF}_{it - 1} + {CF}_{it} + {CF}_{it + 1}

(2)

Where:

_CFit	_:	Cash flows for firm i that are recognised in the same period earnings (t) (as previously labelled);
_CFit+1	_:	_{Cash flows for firm i that are received in period t+1 and are accelerated into current period earnings (t);}
_CFit-1	_:	_{Cash flows for firm i that were received in period t-1 and are deferred into current period earnings (t).}

In essence, equation (2) recognises that accruals consist of cash flows that either lead or lag earnings recognition. They reflect different types of transactions, and it can be reasonably expected that they will have different implications for the persistence of earnings. This is recognised in Barth and Clinch (2013) who suggest that knowledge of these different categories of accruals represents ‘other information’ that is likely to be relevant to the estimation of firm value.

Time period	t-2	t-1	t	t+1	t+2
Earnings_it-1 =	${CF}_{it - 1}^{lead}$	${CF}_{it}^{current}$	${CF}_{it + 1}^{lag}$
Earnings_it =		${CF}_{it - 1}^{lead}$	${CF}_{it}^{current}$	${CF}_{it + 1}^{lag}$
Earnings_it+1 =			${CF}_{it - 1}^{lead}$	${CF}_{it}^{current}$	${CF}_{it + 1}^{lag}$

This shows that the category of earnings labelled in equation (2) as ${CF}_{it + 1}$ might be more completely described as accruals where the cash flows lead earnings recognition and the accruals are reversing in the current period (i.e. ${CF}_{it - 1}^{lead, rev}$ ). Similarly, with respect to ${CF}_{it + 1}$ these might be described as accruals where the cash flows lag earnings recognition and the accruals initiate in the current period (i.e. ${CF}_{it}^{lag, rev}$ ). It also suggests that current period cash flows previously labelled ${CF}_{it}$ will not only include cash flows recognised in current period earnings (i.e. ${CF}_{it}^{current}$ ), but also cash flows relating to items recognised in ‘prior’ period earnings which might be described as cash flows lagging recognition in earnings and reversing (i.e. ${CF}_{it}^{lag, rev}$ ), as well as cash flows relating to items recognised in ‘future’ period earnings which might be described as cash flows leading recognition in earnings and initiating (i.e. ${CF}_{it}^{lead, init}$ ).⁶ This yields the following equations which show the disaggregation of earnings with cash flows categories relabelled (equation 2a), and the disaggregation of current period cash flow (equation 2b).

{Earnings}_{it} = {CF}_{it - 1}^{lead, rev} + {CF}_{it} + {CF}_{it + 1}^{lag, init}

(2a)

{CF}_{it} = {CF}_{it}^{lead, init} + {CF}_{it}^{current} + {CF}_{it}^{lag, rev}

(2b)

Where:

${CF}_{it - 1}^{lead, rev}$	_:	_{Cash flows for firm i that are received in period t-1 and deferred into current period earnings (t) (previously labelled CFit-1);}
${CF}_{it}^{lead, init}$	_:	_{Cash flows for firm i that are received in period t and deferred to future earnings;}
${CF}_{it}^{current}$	_:	_{Cash flows for firm i that are received in period t and recognised in earnings in the same period;}
${CF}_{it}^{lag, rev}$	_:	_{Cash flows for firm i that are received in period t and were accelerated in prior period earnings;}
${CF}_{it + 1}^{lag, init}$	_:	_{Cash flows for firm i that are received in period t+1 and are accelerated into current earnings (t) (previously labelled CFit+1).}

Combining equations (2a) and (2b) yields the following overall disaggregation of earnings on the basis of when cash flows are received and recognised in earnings:

{Earnings}_{it} = {CF}_{it - 1}^{lead, rev} + {CF}_{it}^{lead, init} + {CF}_{it}^{current} + {CF}_{it}^{lag, rev} + {CF}_{it + 1}^{lag, init}

(3)

Subsequently equation (4) below recognises that accruals arise from cash flows that either lead or lag earnings recognition, and are initiating or reversing. They reflect different categories of transactions, and it can be reasonably expected that they will have different implications for the persistence of earnings. For example, cash flows recognised in the current period, that represent a reversal of an accrual initiated in a prior period (i.e. ${CF}_{it}^{lag, init}$ ), are unlikely to have the same implications for earnings persistence as current period cash flows that are initiating accruals and will be recognised in subsequent period earnings (i.e. ${CF}_{it}^{lead, rev}$ ).

It is an empirical question whether these separate categories of earnings differ across firms, and have different implications for the persistence of earnings. To provide insights into the nature of these differences, we also evaluate the implications of accruals for earnings persistence having regard to accruals categorised more simply on the basis of whether the cash flows lead or lag earnings.

{Earnings}_{it} = {({CF}_{it - 1}^{rev} - {CF}_{it}^{init})}^{lead} + {({CF}_{it + 1}^{init} - {CF}_{it}^{rev})}^{lag} + {CF}_{it}^{current}

(4)

And whether accruals are initiating or reversing.

{Earnings}_{it} = {({CF}_{it + 1}^{lag} - {CF}_{it}^{l e a d})}^{init} + {({CF}_{it - 1}^{l e a d} - {CF}_{it}^{l a g})}^{rev} + {CF}_{it}^{current}

(5)

If these categories of earnings have differing implications for earnings persistence the concern is whether this is reflected in stock prices. This disaggregation of earnings also recognises that current period cash flows include items that are not recognised in earnings in the current period, and this might provide insights into the ‘cash flow anomaly’ that has been identified in some studies (e.g. Anderson et al., 2009; Clinch et al., 2012; Coakley et al., 2008; Koerniadi and Tourani-Rad, 2007).

3. Research design

The research design in this study follows the approach adopted in Sloan (1996) to evaluate the persistence of earnings, and the implications of the categories of earnings for firm value. Emphasis will be given in this section to those aspects of research design where innovation occurs such as the extended disaggregation of earnings.

3.1. Disaggregation of earnings

Attention is first focused on the determination of accounting variables. Consistent with Sloan (1996) we focus on operating earnings and the disaggregation of earnings. Accruals are calculated from balance sheet information as follows.⁷

Accruals = (Δ CurrentAssets - Δ Cash) - (\begin{array}{l} Δ CurrentLiabilities - \\ Δ CurrentDebt - Δ Tax \end{array}) - Dep

(6)

Where:

		Compustat item #
ΔCurrent assets	Change in current assets	4
ΔCash	Change in cash and cash equivalents	1
ΔCurrent liabilities	Change in current liabilities	5
ΔCurrent debt	Change in debt included in current liabilities	34
ΔTax	Change in income tax payable	71
Dep	Depreciation and amortisation expense	14

Accruals estimated in this manner are then deducted from earnings to determine cash flows as follows.

{CF}_{it} = {Earnings}_{it} - {Accruals}_{it}

(7)

This approach to the estimation of cash flows and accruals is undertaken for a number of reasons. First, the focus of this study is on operating earnings – this is considered appropriate as operating activities are responsible for generating earnings and tend to resolve within one period, reducing potential error. Second, the estimation of accruals from balance sheet information is necessary due to the unavailability of cash flow data during the early part of the sample period. However, it introduces bias in the determination of accruals and cash flows (Collins and Hribar, 2002) and to mitigate this, sample firms are restricted by excluding those firms where it is likely to be most pronounced (e.g. firms’ mergers and acquisitions). Finally, although depreciation is based on non-current assets, its inclusion is necessary as depreciation is included in operating earnings.⁸

The further classification of accruals is problematic and assumptions about the level of disaggregation must be made. Some categorisations are relatively straightforward. For example, with some accruals, such as accrued expenses, the cash flows would be expected to lag income recognition. While for prepaid expenses, the cash flows would be expected to lead income recognition. However, others are more complex. For example, inventory may be purchased on credit and sold before payment is made, resulting in recognition as income prior to payment. Conversely, inventory may be purchased and paid for before it is sold, resulting in payment before recognition in income. Reflecting these possibilities, if inventory minus accounts payable is positive, this is identified as cash flows leading earnings (i.e. CF^lead), and if it is negative, this is identified as cash flows that lag earnings (i.e. CF^lag).

Consistent with the accruals generating process outlined above, accruals are disaggregated as follows.

{Accruals}_{it} = {CF}_{it + 1}^{lag, init} - {CF}_{it}^{lag, rev} + {CF}_{it - 1}^{lead, rev} - {CF}_{it}^{lead, init} - {Dep}_{it}

(8)

This is based on the following financial statement disclosures:

			Compustat item#
Panel A:	Balance sheet accounts computed individually
Assets	Receivables		2
	Current assets other		195
	Prepayments		2024
	Prepaid expenses		160
	Deferred charges		356
	Accrued expenses		153
Liabilities	Current liabilities		207–356
Other	Depreciation		14
	Total assets		6
	Earnings		178
Panel B:	Balance sheet accounts computed in combination
	Inventory (I)		3
	Accounts payable (AP)		70
Where:	(I–AP) ≤ 0 adds to: CF^lag
	(I–AP) > 0 adds to: CF^lead
Panel C:	Balance sheet items computed into CF
_CFlead	(#2024) − (#153 + #356)	if (#3 – #70) ≤ 0
_CFlag	(#2 + #195) − (#207 – #356)	if (#3 – #70) > 0

Note: For both lead and lag, the respective initiate and reverse is denoted by time: i.e. period (t+1),(t) or (t–1).

3.2. Evaluation of persistence

The central concern of this paper is whether the categories of earnings have differing implications for the persistence of earnings. This is evaluated on a basis consistent with Sloan (1996), albeit extended to reflect the additional categories. Accordingly, attention is focused first on differences in the implications for earnings persistence of accruals that are leading and lagging.

{Earnings}_{it + 1} = λ_{0} + (\begin{array}{l} λ_{1} {({CF}_{it - 1}^{rev} - {CF}_{it}^{init})}^{lead} + \\ λ_{2} {({CF}_{it + 1}^{init} - {CF}_{it}^{rev})}^{lag} + \\ λ_{4} {Dep}_{it} \end{array}) + λ_{4} {CFO}_{it} + ε_{it}

(9)

Attention is then shifted to accruals that are initiating and reversing.

\begin{array}{l} {Earnings}_{it + 1} = μ_{0} + (μ_{1} {({CF}_{it + 1}^{lag} - {CF}_{it}^{lead})}^{init} + μ_{2} {({CF}_{it - 1}^{lead} - {CF}_{it}^{lag})}^{rev} + \\ μ_{3} {Dep}_{it} + ε_{it} \end{array}

(10)

Finally, consideration is given to the separate categories of earnings (i.e. leading, lagging, initiating and reversing).

\begin{array}{l} {Earnings}_{it + 1} = ω_{0} + ω_{1} {CF}_{it + 1}^{lag, init} - ω_{2} {CF}_{it}^{lag, rev} + \\ ω_{3} {CF}_{it - 1}^{lead, rev} - ω_{4} {CF}_{it}^{lead, init} - \\ ω_{5} {Dep}_{it} + ω_{6} {CFO}_{it} + ε_{it} \end{array}

(11)

The objective of these tests is to establish whether the earnings categories have differing implications for the persistence of earnings.

3.3. Market evaluation of earnings persistence

If earnings categories have differing implications for the persistence of earnings, the efficient market hypothesis predicts that this should be reflected in stock prices. This is evaluated in Sloan (1996) for the cash flows and accruals categories of earnings, and extended here for the categories of earnings that are disaggregated on the basis of the accruals generating process.

Market perceptions of the implications of earnings and earnings components for earnings persistence are inferred (statistically) from stock prices (i.e. Mishkin, 1983). This is extended to the earnings categories described above and is reflected in the following equations for lag and lead categories.

{Earnings}_{it + 1} = λ_{0} + (\begin{array}{l} λ_{1} {({CF}_{it - 1}^{rev} - {CF}_{it}^{init})}^{lead} + \\ λ_{2} {({CF}_{it + 1}^{init} - {CF}_{it}^{rev})}^{lag} + λ_{4} {Dep}_{it} \end{array}) + λ_{4} {CFO}_{it} + ε_{it}

(12)

{Abnormal return}_{it + 1} + β ({Earnings}_{it + 1} - λ_{0} - (λ_{1}^{*} {({CF}_{it - 1}^{rev} - {CF}_{it}^{init})}^{lead} + λ_{2}^{*} {({CF}_{it + 1}^{init} - {CF}_{it}^{rev})}^{lag} + λ_{3}^{*} {Dep}_{it}) - λ_{4}^{*} {CFO}_{it}) + ε_{it + 1}

(13)

Attention is then shifted to cash flow categories that initiate and reverse.

\begin{array}{l} {Earnings}_{it + 1} = μ_{0} + (μ_{1} {({CF}_{it + 1}^{lag} - {CF}_{it}^{lead})}^{init} + \\ μ_{2} {({CF}_{it - 1}^{lead} - {CF}_{it}^{lag})}^{rev} + \\ μ_{3} {Dep}_{it} + μ_{4} {Dep}_{it} + ε_{it} \end{array}

(14)

{Abnormal return}_{it + 1} = β (\begin{array}{l} {Earnings}_{it + 1} - \\ μ_{0} - (μ_{1}^{*} {({CF}_{it + 1}^{lag} - {CF}_{it}^{lead})}^{init} + \\ μ_{2}^{*} {({CF}_{it - 1}^{lead} - {CF}_{it}^{lag})}^{rev} + \\ μ_{3}^{*} {Dep}_{it} \end{array}) - μ_{4}^{*} {CFO}_{it}) + ε_{it + 1}

(15)

Finally, consideration is given to the separate categories of earnings (i.e. leading, lagging, initiating and reversing), which yields the following set of equations.

\begin{array}{l} {Earnings}_{it + 1} = ω_{0} + ω_{1} {CF}_{it + 1}^{lag, init} - ω_{2} {CF}_{it}^{lag, rev} + \\ ω_{3} {CF}_{it - 1}^{lead, rev} - ω_{4} {CF}_{it}^{lead, init} - \\ ω_{5} {Dep}_{it} + ω_{6} {CFO}_{it} + ε_{it} \end{array}

(16)

{Abnormal return}_{it + 1} = β (\begin{array}{l} {Earnings}_{it + 1} - ω_{0} - (ω_{1}^{*} {CF}_{it + 1}^{lag, init} - \\ ω_{2}^{*} {CF}_{it}^{lag, rev} + ω_{3}^{*} {CF}_{it - 1}^{lead, rev} - \\ ω_{4}^{*} {CF}_{it}^{lead, init} - ω_{5}^{*} {Dep}_{it}) + ω_{6}^{*} {CFO}_{it} \end{array}) + ε_{it + 1}

(17)

In these equations, regressions are estimated using simultaneous non-linear least squares regression. The (un-starred) coefficients in the first order auto-regression of earnings are estimated directly against future earnings, while the starred coefficients of the second order auto-regression represent market estimation(s) of the persistence implied in (abnormal) market returns. In an efficient market, investors would correctly identify the persistence of earnings and its categories, resulting in equality of coefficients across the equations.⁹ In this paper abnormal returns are calculated by measuring the size-adjusted annual buy-hold return and deducting the annual buy-hold return on a value-weighted portfolio of stocks having similar market values, by the Centre for Research and Security Prices (CRSP) monthly returns file. Portfolio membership is determined using the market capitalisation at the beginning of the year in which the return accumulation period begins.

4. Data and descriptive statistics

Sample firms for the study are chosen from the period 1962 to 2007, with financial report data available on Compustat (2008) and stock return data available on CRSP.¹⁰ Firms excluded from the study are those from the banking and finance industry (GICS 40000000 to 44999999) for which accruals cannot reliably be determined, and from the information technology and telecom industry (GICS 45000000 to 59999999) where the occurrence of the technology boom may impact the results. This identifies a potential sample of 122,984 firm year observations. Prior research (e.g. Collins and Hribar, 2002) suggests potential bias in estimating accruals when a balance sheet approach is used. To ameliorate this problem firms with so-called ‘articulation events’, such as mergers, acquisitions and divestitures, are removed from the sample. This together with missing data items reduced the sample to 90,999 usable firm year observations. Tests involving the determination of earnings persistence require an extra year of data and this reduces the final sample to 83,272 firm year observations.

Descriptive statistics for variables used in this study are presented in Table 1 and these are scaled by average total assets to facilitate cross sectional comparison of firms. Panel A provides descriptive statistics for earnings, cash flow and accrual categories. This shows a mean (median) value of earnings of 0.051 (0.089), and consistent with expectation the mean (median) values of cash flows and accruals are 0.076 (0.115) and −0.025 (−0.031) respectively. Insights into the impact of alternative disaggregation of earnings are provided in Panels B and C. First, mean (median) accruals where cash flows lag earnings of 0.019 (0.012) are greater than mean (median) accruals where cash flows lead earnings of 0.004 (0.000). Second, mean (median) accruals that are initiating and reversing are −0.080 (−0.075) and 0.103 (0.093) respectively. These are much larger and suggest that there is material offsetting within lag and lead cash flow categories. Panel D presents the results of fully disaggregated accruals (i.e. lag, lead, initiate and reverse), and this shows that there is significant offsetting within these categories. For example, while mean (median) ${({CF}_{it + 1}^{init} - {CF}_{it}^{rev})}^{lag}$ is 0.019 (0.012), the mean (median) values of the separate categories are 0.134 (0.130) for ${CF}_{it - 1}^{lead, rev}$ and 0.154 (0.150) for ${CF}_{it}^{lag, rev}$ .

Table 1.

Descriptive statistics.

Descriptive statistics of earnings and categories of earnings of a sample of US firms from 1962 to 2006.
	Observations	Mean	Median	Standard error	Minimum	Maximum
Panel A:	Earnings and earnings components
Earnings	83,272	0.051	0.089	0.252	−5.892	6.025
Cash flows	83,272	0.076	0.115	0.260	−6.554	6.563
Accruals	83,272	−0.025	−0.031	0.120	−3.620	6.216
Panel B:	Accrual categories: Cash flows leading and lagging earnings recognition
${({CF}_{it - 1}^{rev} - {CF}_{it}^{init})}^{lead}$	83,272	0.004	0.000	0.064	−3.065	1.592
${({CF}_{it + 1}^{init} - {CF}_{it}^{rev})}^{lag}$	83,272	0.019	0.012	0.091	−2.594	6.507
Panel C:	Accrual categories: Initiating and reversing
${({CF}_{it + 1}^{lag} - {CF}_{it}^{lead})}^{init}$	83,272	−0.080	−0.075	0.187	−3.289	6.385
${({CF}_{it - 1}^{lead} - {CF}_{it}^{lag})}^{rev}$	83,272	0.103	0.093	0.188	−2.481	1.637
Panel D:	Accrual categories: Cash flows leading or lagging earnings recognition, initiating or reversing
${+ CF}_{it}^{lead, init}$	83,272	0.055	0.010	0.144	−3.430	1.592
${+ CF}_{it}^{lag, rev}$	83,272	0.154	0.150	0.154	−2.297	1.676
${- CF}_{it - 1}^{lead, rev}$	83,272	0.051	0.009	0.128	−1.580	0.966
${- CF}_{it + 1}^{lag, init}$	83,272	0.134	0.130	0.140	−6.435	1.151
${- Dep}_{it}$	83,272	0.048	0.041	0.038	−0.004	1.482

Note: This table is based on the following financial statement variables used in this investigation, where # represents the Compustat item number:

Earnings = Operating income after depreciation(#178)/ ((TotalAssetst–1(#6)+ TotalAssetst(#6))/2)

Cash Flows = (Earnings(#178) – (Accruals)/ ((TotalAssetst–1(#6)+ TotalAssetst(#6))/2)

CFlead, init = (ΔPre-Payments(#2024+ IAP>0)/ ((TotalAssetst–1(#6)+ TotalAssetst(#6))/2)

CFlag,init = (ΔReceivables(#2)+ΔCurrentAssetsOther(#195) + IAP<0)/((TotalAssetst–1(#6)+ TotalAssetst(#6))/2)

CFlead,rev = (ΔPrepaidExpenses(#160)+ΔDeferredCharges(#356))/((TotalAssetst–1(#6)+ TotalAssetst(#6))/2)

CFlag,rev = (ΔCurrentLiabilitiesOth(#207)–ΔDeferredCharges(#356))/((TotalAssetst–1(#6)+ TotalAssetst(#6))/2)

Depreciation = Depreciation(#14)/ ((TotalAssetst–1(#6)+ TotalAssetst(#6))/2)

IAP> 0 = ΔInventory(#3)–ΔAccountsPayable(#70) ;where >0

IAP< 0 = ΔInventory(#3)–ΔAccountsPayable(#70) ;where <0

Accruals = ${({CF}_{it - 1}^{rev} - {CF}_{it}^{init})}^{lead}$ + ${({CF}_{it + 1}^{init} - {CF}_{it}^{rev})}^{lag} - D e p$ reciation

Accrualslead = ${({CF}_{it - 1}^{rev} - {CF}_{it}^{init})}^{lead}$

Accrualslag = ${({CF}_{it + 1}^{init} - {CF}_{it}^{rev})}^{lag}$

Accrualsinit = ${({CF}_{it + 1}^{lag} - {CF}_{it}^{lead})}^{init}$

Accrualsrev = ${({CF}_{it - 1}^{lead} - {CF}_{it}^{lag})}^{rev}$

Importantly, this confirms that there are major differences in the categories of earnings across firms, and therefore provides potentially significant insights into the persistence of earnings and firm value.

5. Results

5.1. Earnings persistence

Consideration is given in the first instance to the evaluation of the persistence of earnings, with the results provided in Table 2, Panel A. This shows a high level of earnings persistence ( $α_{1}$ = 0.840, t-statistic = 421.638, p-value = 0.000) and an adjusted R² of 68.10%, a result consistent with that reported by Sloan (1996) ( $α_{1}$ = 0.841). Attention is then focused on the categories of earnings. This commences in Table 2, Panel B where evidence of the implications of the earnings categories considered by Sloan (i.e. cash flows and accruals) for earnings persistence are presented. Consistent with expectations, cash flows (γ₂ =0.846, t-statistic = 419.106, p-value = 0.000) have greater implications for the persistence of earnings than accruals categories (γ₁= 0.775, t-statistic = 177.192, p-value = 0.000). Furthermore, equality of the coefficients on accruals and cash flows (i.e. $γ_{1} = γ_{2}$ ) can be rejected at the 1% level (F-statistic = 278.329). Notwithstanding the sample differences, the coefficients are similar to those reported by Sloan (γ₂ = 0.855 and γ₁ = 0.765).

Table 2.

Earnings persistence.

Results of OLS regression of future earnings on current earnings and earnings categories of a sample of US firms from 1962 to 2006.
									Obs.	83272	Adj R2	0.681
Panel A:	${Earnings}_{it + 1} = α_{0} + α_{1} {Earnings}_{it} + ε_{it}$
	α₀	$α_{1}$
Coefficient	0.001	0.840
t-statistic	2.762	421.638
p-value	0.006	0.000
Panel B:	${Earnings}_{it + 1} = γ_{0} + γ_{1} {Accruals}_{it} + γ_{2} Cash {Flows}_{it} + ε_{it}$
	γ₀	$γ_{1}$	$γ_{2}$	γ₁, γ₂
Coefficient	−0.001	0.775	0.846	F-Test
t-statistic	−1.180	177.192	419.106	278.329
p-value	0.238	0.000	0.000	0.000
Panel C:	${Earnings}_{it + 1} = λ_{0} + (λ_{1} {({CF}_{it - 1}^{rev} - {CF}_{it}^{init})}^{lead} + λ_{2} {({CF}_{it + 1}^{init} - {CF}_{it}^{rev})}^{lag} + λ_{4} {Dep}_{it}) + λ_{4} {CFO}_{it} + ε_{it}$
	λ₀	$λ_{1}$ .	$λ_{2}$	λ₂	λ₂			$λ_{2}$	$λ_{4}$	$λ_{2}$	λ₁, λ₂, λ₄, λ₁, λ₂, λ₄	λ₁, λ₂, λ₃, λ₄, λ₁, λ₂, λ₃, λ₄
Coefficient	−0.007	0.692	0.842	−0.661	0.847			F-Test	F-Test	F-Test	F-Test	F-Test
t-statistic	−8.545	87.581	148.671	−49.742	419.895			0.964	394.091	252.111	10644.241	431.561
p-value	0.000	0.000	0.000	0.000	0.000			0.326	0.000	0.000	0.000	0.000
Panel D:	${Earnings}_{it + 1} = μ_{0} + (μ_{1} {({CF}_{it + 1}^{lag} - {CF}_{it}^{lead})}^{init} + μ_{2} {({CF}_{it - 1}^{lead} - {CF}_{it}^{lag})}^{rev} + μ_{3} {Dep}_{it} + μ_{4} {Dep}_{it} + ε_{it}$
	μ₀	μ₁	μ₂	μ₃	μ₄			μ₂, μ₄, λ₂, λ₄	μ₁, μ₄, λ₁, λ₄	μ₁, μ₂, λ₁, λ₂	μ₁, μ₂, μ₄, λ₁, λ₂, λ₄	μ₁, μ₂, μ₃, μ₄, λ₁, λ₂, λ₃, λ₄
Coefficient	−0.007	0.790	0.794	−0.656	0.847			F-Test	F-Test	F-Test	F-Test	F-Test
t-statistic	−8.185	158.667	161.704	−49.299	418.437			121.859	137.624	1.492	63777.687	203.938
p-value	0.000	0.000	0.000	0.000	0.000			0.000	0.000	0.222	0.000	0.000
nel E:	${Earnings}_{it + 1} = ω_{0} + ω_{1} {CF}_{it + 1}^{lag, init} - ω_{2} {CF}_{it}^{lag, rev} + ω_{3} {CF}_{it - 1}^{lead, rev} - ω_{4} {CF}_{it}^{lead, init} - ω_{5} {Dep}_{it} + ω_{6} {CFO}_{it} + ε_{it}$
	ω₀	ω₁	ω₂	ω₃	ω₄	ω₅	ω₆	ω₄, ω₆, λ₂, λ₄	ω₃, ω₆, w₃, w₆	ω₂, ω₆, w₂, w₆	ω₁, ω₆, w₁, w₆	ω₁, ω₂, ω₃, ω₄, ω₅, ω₆, w₁, w₂, w₃, w₄, w₅, w₆
Coefficient	−0.016	0.688	0.847	−0.638	−0.815	−0.603	0.841	F-Test	F-Test	F-Test	F-Test	F-Test
t-statistic	−15.320	87.126	148.458	−71.778	−127.382	−43.751	409.390	385.584	24198.204	0.966	51716.946	2344.123
p-value	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.326	0.000	0.000

Note: This table is based on the financial statement variables used in this investigation that are detailed in Table 1.

In Table 2, Panel C the analysis is extended by distinguishing accruals on the basis of whether the underlying cash flows either lead or lag earnings. There is evidence that where accruals reflect cash flows that lag earnings (λ₂ = 0.842, t-statistic =148.671, p-value = 0.000), they have greater implications for the persistence of earnings than accruals where cash flows lead earnings (λ₁ = 0.692, t-statistic = 87.581, p-value = 0.000). Equality of the coefficients on the separate categories of accruals (i.e. λ₁ = λ₂) can be rejected at the 1% level (F-statistic = 252.111). Where accruals arise from cash flows that lag earnings (λ₂ = 0.842, t-statistic = 148.671, p-value = 0.000), they have almost the same implications for the persistence of earnings as current period cash flows recognised in current period earnings (λ₄ = 0.847, t-statistic = 419.895, p-value = 0.000); the difference in these coefficients is not statistically significant (i.e. λ₂ = λ₄, F-statistic = 0.964, p-value = 0.326). It is also interesting to note that depreciation (λ₃ = −0.661, t-statistic = −49.742, p-value = 0.000), while itself highly persistent, has relatively fewer implications for earnings persistence.

The results with accruals categorised on the basis of whether they are initiating or reversing are presented in Table 2, Panel D. Compared to the results in Panel C, there is relatively little difference in the implications for earnings persistence of accruals that are initiating (μ₁ = 0.790, t-statistic = 158.667, p-value = 0.000) or reversing (μ₂ = 0.794, t-statistic = 161.704, p-value = 0.000). Furthermore, the difference in these coefficients is not statistically significant at the 10% level (μ₁ = μ₂, F-statistic = 1.492, p-value = 0.222). As expected, these coefficients are both different from those of cash flows recognised in current period earnings (μ₄ = 0.847, t-statistic = 418.437, p-value = 0.000). The coefficient on depreciation (μ₃= −0.656, t-statistic = −49.299, p-value = 0.000) shows that this accrual continues to have fewer implications for earnings persistence.

Finally, the implications of the separate categories of accruals for earnings persistence are considered with the results reported in Table 2, Panel E. Consistent with the results above, accruals where cash flows lag earnings (both initiating and reversing) have high implications for earnings persistence (ω₁= 0.688, t-statistic = 87.126, p-value = 0.000; and ω₂ = 0.847, t-statistic = 148.458, p-value = 0.000 respectively) relative to accruals where cash flows lead earnings (both initiating and reversing) (ω₃ = −0.638, t-statistic = −71.778, p-value = 0.000; and ω₄ = −0.815, t-statistic = 127.382, p-value = 0.000 respectively). With the exception of the coefficient on accruals, where cash flows lag earnings recognition and are reversing (ω₂ = ω₆, F-statistic = 0.966), equality of the coefficients on accruals categories with the coefficient on cash flows recognised in current period earnings can be rejected (i.e. ω₁ = ω₆, F-statistic = 51716.946, p-value = 0.000; ω₃ = ω₆, F-statistic = 24198.204, p-value = 0.326; ω₄ = ω₆, F-statistic = 385.584, p-value = 0.000).

Importantly, this analysis identifies different implications for the persistence of earnings for each of these categories of earnings. This supports the proposition that accruals categories of earnings are not homogeneous, and they may provide insights into the nature of the accrual anomaly and information about future earnings (Barth and Clinch, 2013).

5.2. Mishkin test – For rational pricing by the market

Having established that earnings categories are not homogeneous and have different implications for the persistence of earnings, consideration is directed to the issue of whether this is reflected in stock prices. This is addressed statistically through Mishkin tests, with the results reported in Table 3. In Table 3, Panel A market efficiency is considered and it is notable that the coefficients on earnings across the two models (actual and expected) are α₁ = 0.840 and $α_{1}^{*}$ = 0.813 respectively. The difference in these coefficients is statistically significant at the 1% level using a chi-square likelihood ratio statistic (LRS = 1887.486 and marginal significance level; MSL = 0.000). This is problematic as it suggests market inefficiency with respect to earnings and contrasts to the results in Sloan (1996) where the coefficients were α₁ = 0.841 and $α_{1}^{*}$ = 0.840 and the difference was not significant.¹¹

Table 3.

Mishkin test.

Results from non-linear generalised least squares estimation of the stock price reaction to information in current earnings and categories of current earnings (accruals and cash flows) about future earnings using Sloan style abnormal returns of a sample of US firms from 1962 to 2006.
Obs. 83272	Actual coefficient	t-statistic	Expected coefficient	t-statistic	Error from expected
Panel A:	${Earnings}_{it + 1} = α_{0} + α_{1} {Earnings}_{it} + ε_{it}$
	${Abnormal return}_{it + 1} + β ({Earnings}_{it + 1} - α_{0} - α_{1}^{*} E a r n i n g s_{i t}) + ε_{it + 1}$
α₁	0.840	421.644	0.813	70.519	0.026
α₁	0.895	49.880	LRS	1887.486
Panel B:	${Earnings}_{it + 1} = γ_{0} + γ_{1} {Accruals}_{it} + γ_{2} Cash {Flows}_{it} + ε_{it}$
	${Abnormal return}_{it + 1} + β ({Earnings}_{it + 1} - γ_{0} - γ_{1}^{} A c c r u a l s_{i t} - γ_{3}^{} {CFO}_{it}) + ε_{it + 1}$
γ₁	0.775	177.195	0.997	38.408	−0.222
γ₂	0.846	419.114	0.797	67.344	0.048
β	0.886	49.351	LRS	1622.704
Panel C:	${Earnings}_{it + 1} = λ_{0} + (λ_{1} {({CF}_{it - 1}^{rev} - {CF}_{it}^{init})}^{lead} + λ_{2} {({CF}_{it + 1}^{init} - {CF}_{it}^{rev})}^{lag} + λ_{4} {Dep}_{it}) + λ_{4} {CFO}_{it} + ε_{it}$
	${Abnormal return}_{it + 1} + β ({Earnings}_{it + 1} - λ_{0} - (λ_{1}^{} {({CF}_{it - 1}^{rev} - {CF}_{it}^{init})}^{lead} + λ_{2}^{} {({CF}_{it + 1}^{init} - {CF}_{it}^{rev})}^{lag} + λ_{3}^{} {Dep}_{it}) - λ_{4}^{} {CFO}_{it}) + ε_{it + 1}$
λ₁	0.692	87.583	1.002	21.343	−0.310
λ₂	0.842	148.675	0.915	27.421	−0.073
λ₃	0.661	49.744	1.370	17.215	−0.709
λ₄	0.847	419.907	0.793	66.481	0.054
β	0.882	49.005	LRS	1834.543
Panel D:	${Earnings}_{it + 1} = μ_{0} + (μ_{1} {({CF}_{it + 1}^{lag} - {CF}_{it}^{lead})}^{init} + μ_{2} {({CF}_{it - 1}^{lead} - {CF}_{it}^{lag})}^{rev} + μ_{3} {Dep}_{it} + μ_{4} {Dep}_{it} + ε_{it}$
	${Abnormal return}_{it + 1} = β ({Earnings}_{it + 1} - μ_{0} - (μ_{1}^{} {({CF}_{it + 1}^{lag} - {CF}_{it}^{lead})}^{init} + μ_{2}^{} {({CF}_{it - 1}^{lead} - {CF}_{it}^{lag})}^{rev} + μ_{3}^{} {Dep}_{it}) - μ_{4}^{} {CFO}_{it}) + ε_{it + 1}$
μ₁	0.790	158.672	1.011	25.229	−0.221
μ₂	0.794	161.708	0.985	25.003	−0.191
μ₃	0.656	49.301	1.622	14.870	−0.966
μ₄	0.847	418.450	0.776	47.859	0.071
β	0.653	36.329	LRS	1175.428
Panel E:	${Earnings}_{it + 1} = ω_{0} + ω_{1} {CF}_{it + 1}^{lag, init} - ω_{2} {CF}_{it}^{lag, rev} + ω_{3} {CF}_{it - 1}^{lead, rev} - ω_{4} {CF}_{it}^{lead, init} - ω_{5} {Dep}_{it} + ω_{6} {CFO}_{it} + ε_{it}$
	${Abnormal return}_{it + 1} = β ({Earnings}_{it + 1} - ω_{0} - (ω_{1}^{} {CF}_{it + 1}^{lag, init} - ω_{2}^{} {CF}_{it}^{lag, rev} + ω_{3}^{} {CF}_{it - 1}^{lead, rev} - ω_{4}^{} {CF}_{it}^{lead, init} - ω_{5}^{} {Dep}_{it}) + ω_{6}^{} {CFO}_{it}) + ε_{it + 1}$
ω₁	0.688	87.130	0.998	21.296	-0.310
ω₂	0.847	148.464	0.918	27.358	−0.071
ω₃	0.638	71.781	0.940	17.868	−0.302
ω₄	0.815	127.388	0.894	23.745	−0.079
ω₅	0.603	43.753	1.309	15.910	−0.706
ω₆	0.841	409.407	0.788	64.945	0.054
β	0.884	49.084	LRS	1812.607

Note: This table is based on the financial statement variables used in this investigation that are detailed in Table 1. Abnormal returns are calculated by measuring the size-adjusted annual buy-hold return and deducting the annual buy-hold return on a value-weighted portfolio of stocks having similar market values, by CRSP. Portfolio membership is determined using the market capitalisation at the beginning of the year in which the return accumulation period begins. All coefficients are significant at less than 1% level.

Table 3, Panel B reports the results for tests of the markets’ interpretation of the implications of cash flows and accruals for earnings persistence. The actual implications of accruals for the persistence of earnings (γ₁ = 0.775, t-statistic = 177.195, p-value = 0.000) is lower than that inferred from stock prices (γ_^*1 = 0.997, t-statistic = 38.408, p-value = 0.000). This suggests that the market overestimates the implications of accruals for the persistence of earnings. In contrast, actual implications of cash flows for the persistence of earnings (γ₂ = 0.846, t-statistic = 419.114, p-value = 0.000) is greater than that inferred from stock prices (γ_^*2 = 0.797, t-statistic = 67.344, p-value = 0.000). This suggests that the market underestimates the implications of cash flows for the persistence of earnings. The differences in these coefficients is statistically significant at the 1% level (LRS = 1622.704; MSL = 0.000). This result is generally consistent with Sloan (1996) in that it identifies the market’s overpricing accruals. However, unlike Sloan, the market is also underpricing cash flows (i.e. there is evidence of a cash flow anomaly which has also been found in a number of studies including Anderson et al., 2009; Clinch et al., 2012; Coakley et al., 2008; Koerniadi and Tourani-Rad, 2007).

The results where accruals are categorised on the basis of whether cash flows are leading or lagging earnings are presented in Table 3, Panel C. These suggest the market overestimates the implications for earnings persistence of accruals arising from cash flows that lead earnings (λ^*₁ = 1.002, t-statistic = 21.343, p-value = 0.000 and λ₁ = 0.692, t-statistic = 87.586, p-value = 0.000) and accruals arising from cash flows that lag earnings (λ^*₂= 0.915, t-statistic = 27.421, p-value = 0.000 and λ₂ = 0.842, t-statistic = 148.675, p-value = 0.000). The actual implications of cash flows recognised in current period earnings (λ₄ = 0.847, t-statistic = 419.907, p-value = 0.000) for the persistence of earnings is underestimated by the market (λ^*₄ = 0.793, t-statistic = 66.481, p-value = 0.00). Additionally, the implications of depreciation (λ³ = 0.661, t-statistic = 49.744, p-value = 0.000) for the persistence of earnings is overestimated by the market (λ^*₃ = 1.370, t-statistic = 17.215, p-value = 0.000). The difference in these coefficients is statistically significant at the 1% level (LRS = 1834.543; MSL = 0.000).

Table 3, Panel D reports the results for accruals categorised on the basis of whether they are initiating or reversing. There is a strong overestimation by the market of the implication for earnings persistence of both accruals that are initiating (μ^*₁= 1.011, t-statistic = 25.229, p-value = 0.000 compared to μ₁= 0.790, t-statistic = 158.672, p-value = 0.000) as well accruals that are reversing (μ^*₂ = 0.985, t-statistic = 25.003, p-value = 0.000 compared to μ₂= 0.794, t-statistic = 161.708, p-value = 0.000). Furthermore, this overestimation of the implications of accruals for the persistence of earnings extends to depreciation (μ^*₃ = 1.622, t-statistic = 14.870, p-value = 0.000 compared to μ₃ = 0.656, t-statistic = 49.301, p-value = 0.000). As expected, the implication for earnings persistence of cash flows recognised in current period earnings is underestimated. The differences in these coefficients is statistically significant at the 1% level (LRS = 1175.428; MSL = 0.000).

Finally, in Table 3, Panel E the results are reported for the evaluation of the market’s perceptions of the implications of the separate categories of accruals for earnings persistence. It is noteworthy that the implications of accruals for earnings persistence are subject to the greatest overestimation where accruals arise from cash flows that lag earnings recognition and are initiating (ω₁ = 0.688, t-statistic = 87.310, p-value = 0.000; ω_^*1 = 0.998, t-statistic = 21.296, p-value = 0.000), and depreciation (ω₅ = 0.603, t-statistic = 43.753, p-value = 0.000; ω^*₅ = 1.309, t-statistic = 15.910, p-value = 0.000). The overestimation of the implications of accruals for earnings persistence is not, however, limited to these accruals, and this is evidenced across all accrual categories (e.g. for ${CF}_{it - 1}^{lead, rev}$ ω₃ = 0.638, t-statistic = 71.781, p-value = 0.000; ω^*₃ = 0.940, t-statistic = 17.868, p-value = 0.000). The actual implications for earnings persistence of current period cash flows recognised in current period earnings (ω₆ = 0.841, t-statistic = 409.407, p-value = 0.000) remain underestimated by the market (ω^*₆ = 0.788, t-statistic = 64.945, p-value = 0.000). The differences in these coefficients is statistically significant at the 1% level (LRS = 1812.607; MSL = 0.000).

In summary, while there is evidence that the categories of earnings have different implications for the persistence of earnings, this is not reflected in stock prices and the accrual anomaly persists. However, it is notable that there are differences across the categories of accruals, and while not resolving the cause of the accrual anomaly, it does suggest that additional information about future earnings and firm value is likely provided by categorisations of accruals.

5.3. Sensitivity tests

During the course of this study a number of issues were identified as requiring further investigation. Some issues relate to the sample firms included and their possible impact on results, while others relate to research design. The sensitivity of the results to these issues is discussed in this section.¹²

Consideration was given to the possibility that results obtained from the investigation might be affected by extreme observations. Accordingly, tests are re-run with selected data sets where observations outside both the 1st and 99th percentile, as well as 5th and 95th percentile, are first winsorised and second, excluded. Results obtained from these sensitivities remain consistent with those reported elsewhere in this paper.

The literature provides evidence that the relation between earnings and stock returns is likely to be different between profit and loss firms (e.g. Ball and Shivakumar, 2006; Basu, 1997; Govendir and Wells, 2013), and this may be a reflection of lower earnings persistence for loss firms (e.g. Klein and Marquardt, 2006). To address the issue of whether this influences the accrual anomaly, separate consideration is given to profit and loss firms. First, it is notable that profit firms (69,677 usable observations) exceed loss firms (21,362 usable observations) in the population. Second, while there is variation in the persistence of earnings and earnings categories across firms, results for both profit and loss sub-samples remain consistent with results obtained for the total population.

Firm size may be associated with particular types of accruals and the firm’s ability to generate earnings. While this is likely addressed by the scaling of all items by average total assets, the sensitivity of the results to firm size is considered by splitting the sample on the basis of total assets. Results for both sub-samples remain generally consistent with results obtained for the total population.

Consistent with Sloan (1996) and the extant literature, firms in the banking and finance industry (GICS 40000000 to 44999999) for which accruals cannot be reliably determined are removed. However, consideration has also been given to firms in the information technology and telecom industry (GICS 45000000 to 59999999) where the occurrence of the technology boom may have impacted the results. Accordingly, these firms have also been removed from the sample. However, a sample including these firms has also been examined. The results for this sample demonstrate statistically weaker results for the persistence of earnings categories. Notably, despite the inclusion of these firms in this sensitivity sample, the results remain generally consistent with results obtained for the total population.

A possible explanation for Sloan’s findings was that investors were unable to easily differentiate categories of earnings until 1988 when the statement of cash flow was required to be included in the financial statements (SFAS 95). This provides a plausible explanation as to why Sloan came to the conclusion that investors did not differentiate between earnings categories. However, the subsequent literature continues to find the accrual anomaly even where the sample is taken only after the introduction of SFAS 95. Accordingly, tests were re-run with the sample split into the periods from 1962 to 1991 and from 1992 to 2006. Results obtained from these tests remain consistent with those reported elsewhere in this paper.

Finally, in addition to the evaluation of statistical differences between heterogeneous accruals categories of earnings, hedge portfolios are used to evaluate whether the accrual anomaly can be used to generate economic returns. Firms are grouped into portfolios based on the magnitude of accruals to determine whether investing based on particular portfolios generates economic (i.e. share) returns. These are determined on the size of abnormal returns based on categories of accruals. The returns to portfolios constructed are not found to be economically significant.

6. Conclusions

In this paper we empirically evaluate whether the disaggregation of accruals, having regard to the ‘accruals generating process’, provides insights into future earnings and to the estimation of firm value. Accordingly, we disaggregate accruals having regard to whether the underlying cash flows lead or lag earnings recognition and whether the accruals are initiating or reversing. For these different accruals we then evaluate whether they have different implications for earnings persistence.

Evidence is provided of differences in these categories of earnings across firms, and when separately recognised these categories of earnings have different implications for the persistence of earnings. Accruals where cash flows lag earnings recognition (e.g. sales made on credit) have greater implications for the persistence of earnings than accruals where cash flows lead earnings (e.g. unearned revenues) and depreciation. Similarly, initiating accruals have greater implications for the persistence of earnings than reversing accruals. Paradoxically, while depreciation exhibits high persistence, it has fewer implications for the persistence of earnings than either the lag or initiating earnings categories. This suggests that ‘not all earnings are created equal’ and there are likely to be differences in how earnings information is impounded in stock prices, depending on its categories.

This paper makes a number of contributions to the literature. First, evidence is provided on the categories of earnings having different persistence. This is inconsistent with the assumptions implied in the research design for tests of the accrual anomaly. Second, and perhaps most importantly, evidence is provided that earnings categories are potentially diverse in both their nature and their implications for future earnings. Accordingly, information about how accruals are generated potentially provides ‘other information’ that is relevant in predicting earnings and estimating firm value (Barth and Clinch, 2013; Ohlson, 1995). This has relevance for a significant body of financial accounting research which considers the information properties of accounting income as well as for firm valuation. At a minimum it suggests re-evaluating prior studies where it is likely that earnings categories are diverse in both their nature and their implications for future earnings. This would include studies in which profit and loss firms are pooled, and this is increasingly necessary as the proportion of loss firms increases in samples. It might also provide insights for the literature considering pro-forma income, where consideration needs to be given to the implications of items included (excluded) from income for future earnings (e.g. Bhattacharya et al., 2003; Bradshaw and Sloan, 2002; Brown and Sivakumar, 2003).

Footnotes

Appendix 1

A principle objective of financial reporting is to provide economic information to external stakeholders (e.g. debt and equity) that may be used in their assessment of whether to maintain or to increase or decrease their allocation of resources to the firm (i.e. firm valuation). A starting point involves a simple and accurate valuation method such as cash received less cash paid over the lifetime of the firm.

Accordingly, beginning with a simple set of transactions that might occur in a valuation model, where all cash flows occur in the same valuation period (e.g. period t), may be derived from the following transactions:

{CF}_{i t}^{current}

Where all cash received is from sales:

or, where all cash paid is for expenses (e.g. insurance):

However, because external stakeholders require regular valuation we create periodic (annual) reporting periods. This introduces numerous complications because to suggest that earnings of each period only occur in that period would be an incomplete representation as it suggests that some cash flows do not overlap across periods, and we know that in practice they do. Hence, we initiate accruals where there are cash flows belonging to other periods, for example, the collection of outstanding amounts from debtors that were earned in a prior period and collected in t+1, would yield cash flows that lag earnings recognition:

C F_{it + 1}^{lag, init}

Where revenue is earned in the current period and not collected until a future period:

or, where an expense is recognised in a period prior to when it is paid:

Additionally, categorisation of earnings by its cash flows introduces further complications as not all cash may be received in period t is CF^current because some of it will be for earnings across other periods (i.e. t+1, or t–1). Subsequently, cash flows that may be prepaid in period t–1 and belong to a future period t may be described as:

- C F_{it}^{lead, init}

Where cash is received that represents future earnings this might be derived from:

or, where cash is paid in a period before the expenses have been recognised in earnings:

Equally, where the cash is collected in period t that is an accrual reversal it would represent:

C F_{it}^{lag, rev}

These cash flow categories might be derived where the collection of cash occurs in a period following earnings:

or, where a firm pays for expenses incurred in a prior period:

Dr - Accounts payableCr - Cash

Equally, where cash flow occurs prior to earnings and the accrual is reversed this would suggest:

C F_{it - 1}^{lead, rev}

Where sales are brought to account from cash collected in a prior period:

or, where expenses brought to account were paid in a prior period:

This relabelling of earnings into its cash flow categories can more simply be expressed graphically as:

Accordingly, the following diagram uses the above examples to show how the categories of earnings, categorised by the timing of their cash flows into earnings, might track into a firm’s financial statements.

Acknowledgements

We would like to gratefully acknowledge the helpful comments of Greg Clinch, Dan Dhaliwal, Jere Francis, Joseph Weber, workshop participants at the European Accounting Association annual conference 2011 and UTS Business School research seminars.

Final transcript accepted 4 October 2013 by Peter Clarkson (AE Accounting).

Funding

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Notes

References

Allen

Larson

Sloan

(2010) Accrual reversals, earnings and stock returns. Journal of Accounting and Economics 66: 113–129.

Anderson

Woodhouse

Ramsay

. (2009) Incorporating additional variables into the testing of the accrual anomaly in Australia: A piecewise approach. Pacific Accounting Review 21: 5–25.

Ball

Shivakumar

(2006) The role of accruals in asymmetrically timely gain and loss recognition. Journal of Accounting Research 44: 207–242.

Barth

Clinch

(2013) The role of accruals in valuation and cash flow prediction. Working paper.

Barth

Cram

Nelson

(2001) Accruals and the prediction of future cash flows. Accounting Review 76: 27–58.

Basu

(1997) The conservatism principle and the asymmetric timeliness of earnings. Journal of Accounting & Economics 24: 3–37.

Beneish

Vargus

(2002) Insider trading, earnings quality, and accruals mispricing. Accounting Review 77: 755–791.

Bernard

Skinner

(1996) What motivates manager’s choice of discretionary accruals? Journal of Accounting & Economics 22: 313–325.

Bhattacharya

Black

Christensen

. (2003) Assessing the relative informativeness and permanence of pro forma earnings and GAAP operating earnings. Journal of Accounting & Economics 36: 285–319.

10.

Bowen

Burgstahler

Daley

(1987) The incremental information content of accruals versus cash flows. Accounting Review 62: 723–747.

11.

Bradshaw

Sloan

(2002) GAAP versus the street: An empirical assessment of two alternative definitions of earnings. Journal of Accounting Research 40: 41–67.

12.

Brown

Sivakumar

(2003) Comparing the value relevance of two operating income measures. Review of Accounting Studies 8: 561–572.

13.

Chan

LKC

Jegadeesh

. (2006) Earnings quality and stock returns. Journal of Business 79: 1041–1082.

14.

Chen

Cheng

(2002) Abnormal accrual based anomaly and managers motivation to record abnormal accruals. Working paper, University of Wisconsin-Madison, WI.

15.

Clinch

Fuller

Govendir

. (2012) The accrual anomaly: Australian evidence. Accounting and Finance 52: 377–394.

16.

Coakley

Manson

Petrovic

(2008) UK Evidence on limited attention and the accrual anomaly. Working paper, University of Essex, Essex.

17.

Collins

Hribar

(2002) Errors in estimating accruals: Implications for empirical research. Journal of Accounting Research 40: 105–134.

18.

Cotton

(2009) The accrual anomaly: Tests of the naive investor hypothesis using discretionary accruals. In Proceedings of the Academy of Accounting and Financial Studies, Allied Academies International Conference New Orleans, Vol 14: 1.

19.

Dechow

(1994) Accounting earnings and cash flows as measures of firm performance: The role of accounting accruals. Journal of Accounting & Economics 18: 3–42.

20.

Dechow

Dichev

(2002) The quality of accruals and earnings: The role of accrual estimation errors. Accounting Review 77: 35–59.

21.

Defond

Park

(2001) The reversal of abnormal accruals and the market valuation of earnings surprises. Accounting Review 76: 375–404.

22.

Fairfield

Whisenant

Yohn

. (2003) Accrued earnings and growth: Implications for future profitability and market mispricing. Accounting Review 78: 353–371.

23.

Francis

Smith

(2005) A re-examination of the persistence of accruals and cash flows. Journal of Accounting Research 43: 413–451.

24.

Govendir

Wells

(2013) An evaluation of regulated IFRS and non-IFRS firm performance measures. Journal of the Securities Institute of Australia 1: 13–23.

25.

Green

Hand

JRM

Soliman

(2011) Going, going, gone? The demise of the accruals anomaly. Management Science 57: 797–816.

26.

Houge

Loughran

(2000) Cash flow is king? Cognitive errors by investors. Journal of Psychology and Financial Markets 1: 161–175.

27.

Hribar

(2001) The market pricing of components of accruals. Working paper, University of Iowa.

28.

Klein

Marquardt

(2006) Fundamentals of accounting losses. Accounting Review 81: 179–206.

29.

Koerniadi

Tourani-Rad

(2007) Accrual or cash flow anomaly? Evidence from New Zealand. Accounting Research Journal 20: 21–36.

30.

Kothari

(2001) Capital markets research in accounting. Journal of Accounting & Economics 31: 105–231.

31.

Kraft

Leone

Wasley

(2007) Regression-based tests of the market pricing of accounting numbers: The Mishkin test and ordinary least squares. Journal of Accounting Research 45: 1081–1113.

32.

Lev

Nissim

(2006) The persistence of the accrual anomaly. Contemporary Accounting Research 23: 193–226.

33.

Mishkin

(1983) A rational expectations approach to macroeconomics testing policy effectiveness and efficient markets models. The University of Chicago Press for the National Bureau of Economic Research, Chicago, USA.

34.

Ohlson

(1995) Earnings, book values, and dividends in equity valuation. Contemporary Accounting Research 11: 661–687.

35.

Pae

(2005) Expected accrual models: The impact of operating cash flows and reversals of accruals. Review of Quantitative Finance and Accounting 24: 5–22.

36.

Pincus

Rajgopal

Venkatachalam

(2007) The accrual anomaly: International evidence. Accounting Review 82: 169–203.

37.

Richardson

Sloan

Soliman

. (2005) Accrual reliability, earnings persistence and stock prices. Journal of Accounting & Economics 39: 437–485.

38.

Richardson

Sloan

Soliman

. (2006) The implication of accounting distortions and growth for accruals and profitability. Accounting Review 81: 713–743.

39.

Shan

Taylor

Walter

(2010) Errors in estimating unexpected accruals in the presence of large changes in net external financing. Working paper. Accounting and Finance Association of Australia and New Zealand Conference, Adelaide, Australia. pp. 1–54.

40.

Sloan

(1996) Do stock prices fully reflect information in accruals and cash flows about future earnings? Accounting Review 71: 289–315.

41.

Thomas

Zhang

(2002) Inventory changes and future returns. Review of Accounting Studies 7: 163–187.

42.

Xie

(2001) The mispricing of abnormal accruals. Accounting Review 76: 357–373.

43.

Zach

(2003) Inside the accrual anomaly. Working paper, Washington University, St Louis, MO.

44.

Zhang

(2007) Accruals, investment, and the accrual anomaly. Accounting Review 82: 1333–1363.