is the timing role of accrual accounting disappearing? · pdf fileis the timing role of...

43
Is the Timing Role of Accrual Accounting Disappearing? Robert M. Bushman University of North Carolina at Chapel Hill [email protected] Alina Lerman Yale University [email protected] X. Frank Zhang Yale University [email protected] October 2013 Preliminary and incomplete, comments welcome

Upload: hoangbao

Post on 06-Mar-2018

219 views

Category:

Documents


1 download

TRANSCRIPT

Is the Timing Role of Accrual Accounting Disappearing?

Robert M. Bushman

University of North Carolina at Chapel Hill [email protected]

Alina Lerman Yale University

[email protected]

X. Frank Zhang Yale University

[email protected]

October 2013

Preliminary and incomplete, comments welcome

Is the Timing Role of Accrual Accounting Disappearing?

ABSTRACT

This study examines the timing role of accrual accounting, a smoothing of temporary timing fluctuations in operating cash flows via accruals. We show that the timing role has dramatically declined over the past fifty years and has largely disappeared in more recent years. This is demonstrated by a temporal decrease in the adjusted R2 and a temporal increase in the coefficient on contemporaneous cash flows in both the Dechow (1994) and the Dechow and Dichev (2002) accrual models. We explore several potential reasons for the observed attenuation and find that an increase in cash flow volatility partially explains the decline. On the other hand, a temporal change in the matching between revenues and expenses, an increase of the accrual accounting role in the asymmetrically timely recognition of gains and losses, and an increase in one-time items do not significantly contribute to the documented attenuation. Thus, the startling finding that the central role of accrual accounting has significantly declined over time, and for the most part disappeared in recent years, remains largely unexplained.

1

1. Introduction

A central role of accrual accounting is to smooth out temporary fluctuations in cash flows

(e.g., Dechow, 1994; Dechow, Kothari and Watts, 1998), as accrual accounting systems

recognize economic events in firms’ financial statements independently of the timing of cash

flows associated with these events. We refer to this role as the timing role of accruals. By adding

accruals to operating cash flows, accrual accounting systems produce an earnings number that is

a less noisy measure of operating performance than operating cash flows. As Dechow (1994)

points out, a central implication of the timing role of accrual accounting is that contemporaneous

accruals and cash flows are negatively correlated. In this paper, we show that the timing role of

accrual accounting has dramatically diminished over the past half a century and has largely

disappeared in more recent years, as evidenced by a temporal decrease in the adjusted R2 and a

temporal increase in the coefficient on contemporaneous cash flows in accrual models based on

Dechow (1994) and Dechow and Dichev (2002). The negative contemporaneous association

between cash flows and accruals, an inherent implication of the timing role of accrual

accounting, is often taken as given in prior literature. A loss of this property has profound

implications on accounting research and teaching.

We adopt two models to examine the change in the timing role of accrual accounting.

The first one is based on Dechow (1994) and regresses total accruals on contemporaneous

operating cash flows. We run the model both in the levels and in the changes specifications for

each year from 1964 to 2012 (inclusive) and examine the temporal change in the goodness of fit

measure and in the coefficient on contemporaneous cash flows. We find that the adjusted R2

drops from about 70% (90%) in the 1960s to near zero (20%) in more recent years for the levels

(changes) specifications. At the same time, the coefficient on contemporaneous cash flows

2

experiences a drastic increase and converges towards zero. The results suggest that the timing

role of accrual accounting has significantly diminished over the past fifty years in a persistent

and smooth manner.

The second model we use is the Dechow and Dichev (2002) regression of total accruals

on past, current, and future operating cash flows. Again, we find a dramatic decline in the

adjusted R2 of the model and a smooth increase in the coefficient on contemporaneous cash

flows over the fifty year period. The adjusted R2 has dropped from about 70% in the 1960s to

below 20% in the latest years, whereas the coefficient on contemporaneous cash flows has

increased from about -0.8 to -0.2 over the same time period. In contrast, the coefficients on past

and future cash flows show little change in magnitude over time.

Having documented the pronounced and continuous decline in the timing role of accrual

accounting, we explore the potential reasons for this attenuation. First, we consider whether a

change in cash flow volatility, a proxy for the uncertainty of the operating environment, has

contributed to the observed temporal change. We find that cash flow volatility is indeed

negatively associated with the degree of the timing role of accrual accounting. However, the

increase in cash flow volatility over the years only partially explains the gradual decline in the

timing role. Second, we consider the effect of the temporal change in the matching between

revenues and expenses as discussed in Dichev and Tang (2008). We find that the decline in the

matching between revenues and expenses is less drastic than the decline in the timing role of

accrual accounting. Furthermore, the effect of the mismatch on the attenuation of the timing role

of accrual accounting is subsumed by the effect of the changes in cash flow volatility. Next, we

consider whether the observed decline in the timing role of accruals is a result of changes in the

relative timeliness of gain versus loss recognition. We utilize the Ball and Shivakumar (2006)

3

framework to examine the two roles of accrual accounting and find that the timely loss

recognition role is largely unchanged over a significant portion of the sample period. When

considering the two roles together we see a pronounced decline over the years in their joint

ability to explain firm-specific levels of accruals. Thus, the decline in the timing role of accrual

accounting is not driven by an offsetting increase in the timely gain and loss recognition role.

Finally, we consider the effect of a temporal increase in the frequency and magnitude of reported

one-time items and find that they do not significantly contribute to the decline of the timing role

of accrual accounting.

A battery of additional tests extends the main results. First, we expand the smoothing

window to non-adjacent fiscal periods to consider the possibility that accruals map to cash flows

up to three periods preceding and following the current period. Next, we explore the potential

effects of the change in the sample composition over time and redo our analysis on a sample of

the largest 1,000 firms in each year. We also examine the potential effect of mergers and

acquisitions and exclude from the sample the firm-years with significant M&A activities as

measured by either sales or earnings contributions. Finally, we consider whether changes in

operating cash cycles, changes in absolute total accruals, or industry effects contribute to the

observed attenuation. We find that our results on the decline of the timing role of accrual

accounting are robust to all alternative samples and specifications.

Overall, we document a profound decline in the timing role of accrual accounting. We

show evidence that an increase in cash flow volatility partially explains the observed attenuation.

Having considered a multitude of alternative potential causes we conclude that the bulk of the

decrease remains unexplained.

4

Our findings that the central timing role of accrual accounting has significantly declined

(or even disappeared) over the years have broad implications for academics, practitioners, and

regulators. On a conceptual note, our results suggest that we may need to rethink what accrual

accounting means today and what we can teach our students. Given the common position held in

the literature that accruals give accounting earnings its primary role in valuation, contracting, and

performance measurement, our findings suggest that we may need to revisit these issues. Our

evidence may also help to answer such questions as why earning have become less value

relevant over time, why the accrual anomaly has declined, and why earnings-based measures

have become less prominent in debt covenants and compensation contracts. On a more pragmatic

note, we observe that empirical accounting studies, which either examine accrual accounting or

utilize measures of accrual quality such as the Dechow and Dichev (2002) mapping, typically

pool historical data in examining their research questions. Our results suggest that this approach

may not be appropriate, given the temporal change in the information content of accruals.

Similarly, some papers use the historical average of the negative correlation between

contemporaneous accruals and cash flows as a base assumption in modeling other relationships

(e.g. Richardson et al., 2005) and may need to take note of the changing association.

We organize the rest of the paper as follows. Section 2 reviews relevant literature.

Sections 3 describe our sample. Section 4 presents the main empirical results and explores

potential explanations. Section 5 conducts robustness checks, and Section 6 concludes.

2. Prior Research and Background

Accrual accounting recognizes economic events in firms’ financial statements

independently of the timing of cash flows associated with these events. The contrast between

5

cash basis reporting and earnings reported under the accrual system is highlighted in Financial

Accounting Standards Board [FASB] Concept 1:

“[Accrual accounting] recognizes that the buying, producing, selling, and other

operations of an enterprise during a period, as well as other events that affect

enterprise performance, often do not coincide with the cash receipts and payments

of the period.” (paragraph 44)

A central role of accrual accounting, what we refer to as the timing role, is to smooth out random

timing fluctuations in operating cash flows. For example, consider a firm in a steady state with

constant scale of operations over time. An increase in accounts receivable due to a customer

unexpectedly delaying payments would simultaneously reduce cash flows and increase accruals

by the same amount. Similarly, a temporary increase in inventory is associated with growth in

the working capital account “inventory” and a contemporaneous reduction in operating cash

flows. Accrual accounting prevents such transitory fluctuations from affecting the reported

earnings of the firm by keeping the net revenues of the period unchanged in the first case and by

shifting the inventory increase to future period’s cost of goods sold in the second. This

smoothing property of the reporting system can be viewed as a channel by which accruals

increase the informativeness of reported earnings. By adding accruals to operating cash flow,

accrual accounting systems produce an earnings number that is less noisy than operating cash

flow as accruals mitigate the noise in cash flow that arises from exogenous or manipulative

variation in working capital items. Accruals record real economic transactions in a timely

fashion, thus distinguishing our system of accounting from the mere counting of cash. As

Dechow (1994) points out, a central prediction of the timing role of accrual accounting is, thus,

that accruals and cash flows from operations are negatively correlated.

6

The negative contemporaneous association between operating cash flows and total

accruals is observed going back to some of the earlier studies on accrual accounting (Rayburn,

1986; McNichols and Wilson, 1988). Rayburn (1986) records firm-specific Pearson correlation

of -0.81 between the levels of cash flows from operations and total accruals in the 1962-1982

period. McNichols and Wilson (1988) observe Spearman correlation of -0.69 (-0.78) between the

levels (changes) of the two variables in the 1967-1985 period. Later research continues to

explore the association in a more systematic fashion. Dechow (1994), Sloan (1996), and Dechow

et al. (1998) all predict, document, and exploit a negative contemporaneous correlation between

levels or changes of aggregate accruals and operating cash flows. Dechow (1994), in particular,

specifically posits that the negative association is inherent in the system where accruals are being

used to smooth the noisy cash flow metrics. This relation stems from the temporary nature of

cash flow fluctuations and is smaller when measured over longer intervals. Dechow and Dichev

(2002) expand on this role of accruals and introduce a measure, which they term accrual quality,

capturing the mapping of current accruals into last period, current period, and next period cash

flows. In line with the timing role of accrual accounting, their analysis indicates that the

association between working capital accruals and contemporaneous operating cash flows is

strongly negative while that of the accruals and past/future cash flows is positive (albeit of a

much smaller magnitude). Subsequent literature has relied heavily on the Dechow and Dichev

(2002) mapping measure to explore questions pertaining to accruals quality (i.e. Francis et al.,

2004, 2005, Dechow et al., 2010).1

1 There is a debate in the literature on whether the smoothing property of accruals improves or impedes earnings informativeness. In contrast to works noted above, some have adopted the view of smoothing as an earning management mechanism (i.e. Beatty et al., 2002, Leuz et al., 2003). For example, Myers et al. (2007) document a stronger negative correlation between changes in quarterly cash flows and accruals for firms with strings of consecutive EPS increases (although for both control and ‘suspect’ groups the correlation is below -0.9 in the 1963-

7

In summary, the negative association between contemporaneous accruals and cash flows

is well established in the literature. However, there is some sporadic evidence in studies using a

more recent sample period suggesting that the association between accruals and cash flows

becomes less pronounced in recent years. For example, Barone and Magilke (2009) find a

Pearson (Spearman) correlation of -0.04 (-0.33) between levels of operating cash flows and total

accruals on the pooled 1988-2004 time period. Givoly and Hayn (2000) find that the covariance

between accruals and cash flows increased from about -0.01 in the 1960s-1980s to -0.005 in the

1990s. In this paper, we examine whether the negative association between accruals and cash

flows weakens over time and, if so, explore the potential reasons for such attenuation.

Our paper is closely related to Dichev and Tang (2008) but differs in its objective and

empirical analysis. Dichev and Tang (2008) examine the effect of poor matching between

revenues and expenses on the properties of accounting earnings over the 1967-2003 period.

Conceptually, one may envision an accounting reporting evolution where the matching of

revenues and expenses is increasingly disrupted over time, however, the timing role of accruals,

whether intended or nefarious, remains unchanged and vice versa. For example, the imposed

requirement to expense employee stock options should increase the matching of expenses to the

appropriate revenue, but it should not change the smoothing property of accruals. Empirically,

we find that the disappearing timing role of accrual accounting is not explained by the decline in

the matching between revenues and expenses. In terms of magnitude, the decline is much more

dramatic for the timing role of accrual accounting (a drop from about 70% to 10%) than for

Dichev and Tang’s matching between revenues and expenses (a drop from 99% to 94%) in our

1964-2012 sample period.

2004 period). We do not address the question of whether the negative contemporaneous association reflects earnings quality or earnings management and whether those characteristics have changed over time.

8

3. Sample and Definition of Variables

We obtain our sample data from Compustat and limit the sample to firm-years with non-

missing accruals, cash flows, and average total assets variables. Our sample consists of 228,847

firm-year observations from 1964 to 2012 (inclusive). We use the balance sheet approach to

estimate total accruals.2 Specifically, total accruals (TACC) are defined as changes in non-cash

current assets less changes in non-debt current liabilities minus depreciation expense, scaled by

average total assets. Cash flows (CFO) are cash flows from operations measured as earnings

minus total accruals, where earnings (E) are earnings before extraordinary items scaled by

average total assets.

Table 1 presents the descriptive statistics and the correlation matrix of the variables of

interest. The descriptive statistics are generally in line with existing research (such as Table 2 of

Dechow and Dichev, 2002). The mean working capital accruals and cash flows are 0.013 and

0.034, respectively. The mean total accruals are negative because of the depreciation expense.

The Pearson (Spearman) correlation between total accruals and contemporaneous cash flows

from operations on the pooled basis is expectedly negative at -0.29 (-0.47). In line with the

timing role of accruals, the Pearson correlations between both total and working capital accruals

and past and future cash flows from operations, as well as the Spearman correlations between

working capital accruals and past and future cash flows from operations are positive and

statistically significant. As in Dechow and Dichev (2002), the positive Pearson correlations

2 Although Hribar and Collins (2002) document that the balance-sheet-based accruals suffer from measurement errors, especially for firms with merger and acquisition activity or discontinued operations, we are compelled to use this approach because of the long time series period examined. For robustness, we also estimate the total accruals from the statement of cash flows from 1988, the year of implementation of SFAS No. 95. We find qualitatively same results (untabulated) using this measure of accruals on the shortened 1988-2012 time series.

9

between accruals and past/future cash flows are much smaller in magnitude than the negative

correlation between accruals and contemporaneous cash flows. Interestingly, the Spearman

correlation between total accruals and past and future cash flows from operations is small and

negative.

4. Results

4.1 Main results

4.1.1 Dechow (1994) model

We begin our analysis with the exploration between contemporaneous accruals and cash

flows over time. Dechow (1994) shows that accruals and cash flows are negatively correlated

because accruals tend to mitigate timing and matching problems in cash flows when reflecting

firm performance. We capture this relation by regressing total accruals on cash flows from

operations, as shown in equation (1a) below. We run equation (1a) each year and examine β1, the

coefficient on CFO, and the adjusted R2, a measure of the model’s goodness of fit, over time.

𝑇𝐴𝐶𝐶𝑡 = 𝛽0 + 𝛽1𝐶𝐹𝑂𝑡 + 𝑒𝑡 (1a)

where TACC and CFO are total accruals and cash flows from operations, respectively.

The results of the annual regressions are found in Panel A of Table 2. The adjusted R2

from equation (1a) has dropped from about 70% in 1960s to near zero in more recent years,

suggesting that the timing role of accrual accounting has largely disappeared. Note that the R2

cannot drop below zero, which limits the downside for the adjusted R2. Similarly, the coefficient

β1 has increased from about -0.7 to -0.1 over the past fifty years.

10

For completeness, we note that prior research and theory has frequently focused on the

relationship between the changes, rather than the levels, in accruals and cash flows (McNichols

and Wilson, 1988; Leuz et al., 2003). To address this alternative specification we run equation

(1b) in a similar fashion.

∆𝑇𝐴𝐶𝐶𝑡 = 𝛽0 + 𝛽1∆𝐶𝐹𝑂𝑡 + 𝑒𝑡 (1b)

where ∆TACC and ∆CFO are annual changes in total accruals and cash flows from operations,

respectively.

The results of the annual changes regressions are found in Panel B of Table 2 and are

qualitatively similar to the levels results. The adjusted R2 from equation (1b) has dropped from

about 90% in 1960s to 20% in recent years, and the coefficient β1 has increased from about -0.9

to -0.3.

In Panels C and D of Table 2, we examine the changes in the adjusted R2 and the

coefficient β1 from models (1a) and (1b) in a more systematic fashion by regressing each on a

time trend. We observe that the coefficient on the time trend is statistically significantly negative

(positive) for adjusted R2 (β1) and the goodness of fit of the model is over 90% for both levels

and changes specifications. The fitted values for the beginning and ending year of the sample

confirm the drastic decline in the smoothing relationship. Figure 1 presents the results from

Table 2 in graphical form. It highlights the continuity and smoothness of the decline (increase) of

the R2 (β1) over time, suggesting that the pattern is not driven by a regime shift.

4.1.2 Dechow and Dichev (2002) model

11

Next, we consider the Dechow and Dichev (2002) model that regresses total accruals on

past, current, and future cash flows, as shown in equation (2) below.

𝑇𝐴𝐶𝐶𝑡 = 𝛽0 + 𝛽1𝐶𝐹𝑂𝑡−1 + 𝛽2𝐶𝐹𝑂𝑡 + 𝛽3𝐶𝐹𝑂𝑡+1 + 𝑒𝑡 (2)

The results of the annual regressions are found in Panel A of Table 3. The adjusted R2

from equation (2) has dropped from about 70% in the 1960s to below 20% in the latest years,

and the coefficient β2 has increased from about -0.8 to -0.2 over the same time period. In Panel B

of Table 3, we regress the adjusted R2 and cash flow coefficients from model (2) on a time trend.

The coefficient on the time trend is negative (positive) and statistically significant for adjusted R2

(β2) and the goodness of fit of the model is above 90% for both. The fitted values at the

beginning and ending sample period years exhibit an even more pronounced change than that

observed in the results from the annual regressions.

We do not offer a directional prediction regarding the change in the association between

accruals and past and future cash flows. One possibility is that the loss of the negative

association between contemporaneous accruals and cash flows is offset with an increase between

adjacent accruals and cash flows. An alternative possibility is that the attenuation of the

contemporaneous association is partnered with a decline in the positive association with adjacent

cash flows. Turning to the observed coefficients on past and future cash flows in Panel A of

Table 3, we find that the time-series changes are relatively small in magnitude. The coefficient

on past cash flows, denoted as β1 in model (2), has declined from an average of 0.15 in the first

five to an average of 0.08 in the last five years of the sample. A time trend regression in Panel B

shows that the decline is statistically significant but of a much smaller magnitude than an

increase for the contemporaneous cash flow coefficient. The coefficient on future cash flows,

12

denoted as β3 in model (2), appears largely unchanged in Panel A and even shows evidence of a

slight increase in the time trend regression in Panel B. If we carry out the analysis over distinct

time periods, we see that the statistically significant increase in β3 is driven by observations in

the earliest years, while the later coefficients remain largely unchanged. In the time trend

regression in Panel B excluding the first ten years and rerunning the model on the period 1975-

2011 (untabulated), we find the coefficient on the time trend equal to 0.0005 and statistically

insignificant. These relatively small temporal changes in the coefficients on adjacent cash flows

reinforce our conclusion that the dramatic decline in the adjusted R2, a measure of greatest

interest to us, is driven mainly by the loss of the association between contemporaneous accruals

and cash flows.

Figure 2 presents the results from Table 3 in graphical form. Panel A represents the

evolution of the adjusted R2, a summary measure of the mapping between accruals and cash

flows. Again, we observe a relatively smooth and persistent decline of the R2 over time.3 Panel B

shows the temporal variation in the coefficients on the past, current, and future cash flows in the

Dechow and Dichev (2002) model. As discussed above, we note the striking attenuation of the

negative coefficient on contemporaneous cash flows and the small changes of the coefficients on

past and future cash flows.

Overall, the results in Section 4.1 indicate that the timing role of accrual accounting, as

reflected in the negative association between contemporaneous accruals and cash flows, has

dramatically shrunk over the past fifty years and largely disappeared in more recent years. This is

3 In untabulated analysis we consider whether we can extrapolate the loss of the timing role of accrual accounting to a loss of the general predictive ability of accruals as in the Sloan (1996) framework. We regress one period ahead earnings on both total accruals and cash flows from operations. We observe that the overall predictive ability of accruals, both in absolute terms and relative to the predictive ability of cash flows, has not changed dramatically over the past fifty years.

13

evidenced by a pronounced temporal decrease in the adjusted R2 and the temporal increase in the

coefficient on contemporaneous cash flows in the accruals models based on both Dechow (1994)

and Dechow and Dichev (2002). Furthermore, the attenuation of the timing role of accrual

accounting over the years occurred in a smooth and gradual fashion.

4.2 Possible explanations

Having documented the drastic decline in the timing role of accrual accounting, we now

explore the potential reasons for the observed attenuation.

4.2.1 Operating uncertainty and poor matching

First, we consider a possibility that the underlying cash flow volatility, a proxy for the

uncertainty in the operating environment, may have changed over time. Dechow and Dichev

(2002) note that the ability of accruals to map into cash flows is, in theory, related to cash flow

volatility. An increase in cash flow volatility over the sample period could lead to a disruption in

the expected stable relationship between cash flows and accruals stipulated by the timing role of

accrual accounting. Mathematically, the R2 and the cash flow coefficient in equation (1a) can be

written as 𝐶𝑂𝑉(𝐴𝐶𝐶,𝐶𝐹𝑂)𝑉𝐴𝑅(𝐴𝐶𝐶)∗𝑉𝐴𝑅(𝐶𝐹𝑂)

and 𝐶𝑂𝑉(𝐴𝐶𝐶,𝐶𝐹𝑂)𝑉𝐴𝑅(𝐶𝐹𝑂)

, respectively. Thus, cash flow volatility

(VAR(CFO)) directly affects both the R2 measure and the cash flow coefficient.

Second, we consider the effect of the temporal change in the matching between revenues

and expenses (Dichev and Tang, 2008). FASB’s slow push of balance-sheet or mark-to-market

accounting towards greater prominence may have changed the role of accruals over time. Among

the standards which denote the ascent of fair value accounting are those on the determination and

treatment of goodwill, reporting for financial assets, and impairments of fixed assets. Dichev and

14

Tang (2008) document the aggregate effects of the regulatory evolution in the context of the loss

of matching between recognized revenues and expenses. They find a continuous and pronounced

decline in the contemporaneous correlation between revenues and expenses and corresponding

changes in earnings properties such as increased volatility, decreased persistence, and greater

negative autocorrelation.

To examine whether a temporal increase in operating cash flow volatility or a temporal

decrease in the matching of revenues and expenses documented by Dichev and Tang (2008) is

responsible in part or in full for the attenuation of the timing role of accrual accounting, we run

equation (3) on the sample period.

𝐴𝑑𝑗.𝑅2(𝐷𝐷)𝑡 = 𝛽0 + 𝛽1𝑇𝑖𝑚𝑒 + 𝛽2𝑆𝑡𝑑(𝐶𝐹𝑂)𝑡 + 𝛽3𝐴𝑑𝑗.𝑅2(𝐷𝑇)𝑡 + 𝑒𝑡 (3)

where the dependant variable is adjusted R2 from the Dechow and Dichev (2002) regression as

represented in model (2). This variable captures the goodness of fit of the model where accruals

are determined solely by the past, present, and future cash flows and, thus, is a good proxy for

the degree of the timing role of accrual accounting. Time is a time trend represented as the

number of years from 1964. Std(CFO) is the standard deviation of cash flows from operations

calculated annually. Adj.R2(DT) is the adjusted R2 from the Dichev and Tang (2008) model

represented as (4) below, which serves as a proxy for the matching between revenues and

expenses.

𝑆𝐴𝐿𝐸𝑡 = 𝛽0 + 𝛽1𝐸𝑋𝑃𝐸𝑁𝑆𝐸𝑡−1 + 𝛽2𝐸𝑋𝑃𝐸𝑁𝑆𝐸𝑡 + 𝛽3𝐸𝑋𝑃𝐸𝑁𝑆𝐸𝑡+1 + 𝑒𝑡 (4)

where SALE are the net sales scaled by average total assets and Expense are expenses measured

as sales minus earnings before extraordinary items scaled by average total assets.

15

The results of the regression model (3) are reported in Table 4. Column 2 shows that the

coefficient on Std(CFO) is negative and statistically significant , in line with the intuition that

cash flow volatility is negatively correlated with the degree of the timing role of accrual

accounting. However, cash flow volatility does not subsume the statistically significant

coefficient on the time trend and only modestly increases the adjusted R2 from 0.915 in column 1

to 0.929 in column 2. The decline in the magnitude of the coefficient on the time trend variable

from column 1 to column 2 reveals that about 31% [=(0.16-0.11)/0.16 ] of the decline in the

timing role of accrual accounting in our setting is attributable to an increase in cash flow

volatility.

Column 3 shows a positive and statistically significant coefficient on Adj.R2(DT) ,

consistent with the idea that the decline in the matching between revenues and expenses over

time observed by Dichev and Tang (2008) contributes to the loss of the timing role of accrual

accounting. However, the coefficient on the time trend variable remains negative and statistically

significant. The decline in the magnitude of the coefficient on the time trend variable from

column 1 to column 3 reveals that only about 19% [=(0.16-0.13)/0.16 ] of the timing role decline

is related to Dichev and Tang’s (2008) documented mismatch between revenues and expenses.

Finally, we find that the Adj.R2(DT) coefficient becomes statistically insignificant, whereas the

coefficient on cash flow volatility remain highly significant, in the full specification in column 4.

In Figure 3 we plot the variables of interest over time. Panel A graphs the time series

pattern of the standard deviations in earnings and components of earnings: total accruals and

cash flows from operations (the latter alternatively calculated from the balance sheet for the full

sample and from the statement of cash flows for the period 1988 forward). We observe that the

volatility of cash flows does indeed increase over the sample period, particularly from 1990

16

onwards. At the same time, the volatility of accruals remains largely constant, highlighting the

growing disconnect between the behavior of the two components of earnings. We also note that

the volatility of cash flows exhibits specific pronounced changes around 2000, 2003, and 2008

which are not mirrored in the relatively smooth patterns evidenced in Figures 1 and 2. Panel B of

Figure 3 graphically illustrates the loss of the goodness of fit in the models capturing the timing

role of accrual accounting and the model capturing the matching of revenues and expenses. We

observe that the decline in the timing role of accrual accounting in our setting is much more

dramatic than the decline in matching between revenues and expenses in Dichev and Tang

(2008).

Overall, we find that a temporal increase in cash flow volatility partially explains the loss

of the accrual accounting timing role. The decline in the matching between revenues and

expenses in Dichev and Tang (2008) is also related to the observed attenuation of the timing role,

but that effect is subsumed by the changes in cash flow volatility.

4.2.2 Asymmetrically timely recognition of gains and losses

As Ball and Shivakumar (2006) point out, another inherent property of accrual

accounting is asymmetrically timely recognition of gains and losses. Because revisions in the

current period cash flows from a durable asset are likely to be positively correlated with

revisions in its expected future cash flows, the timely gain and loss recognition role of accruals

suggests a positive correlation between accruals and contemporaneous cash flows. This positive

correlation will tend to offset the negative correlation from the timing role of accrual accounting

in a linear specification, such as equation (1). Because losses are generally recognized in a more

timely fashion than gains, the positive correlation will not be symmetric. It is possible that the

17

effect of the asymmetrically timely gain and loss recognition gets stronger over the observed

period and, thus, attenuates the negative coefficient on cash flows from equation (1) in the recent

years.

To incorporate the asymmetrically timely recognition of economic gains and losses via

accruals into our analysis, we follow Ball and Shivakumar (2006) extension of the cash flow

model and run equation (5) over the sample period.

𝑇𝐴𝐶𝐶𝑡 = 𝛽0 + 𝛽1𝐶𝐹𝑂𝑡 + 𝛽2𝐷 + 𝛽3𝐷 ∗ 𝐶𝐹𝑂𝑡 + 𝑒𝑡 (5)

where D is equal to one if CFOt is negative and zero otherwise. We also consider an alternative

specification of D as discussed in Ball and Shivakumar (2006), where it is equal to one if annual

ΔCFOt is negative and zero otherwise. If the observed decline in the timing role of accrual

accounting in Table 2 is due to the change in the asymmetrically timely gain and loss

recognition, then the adjusted R2 from equation (5) should not change significantly over time.

The results of the annual regressions are reported in Panel A of Table 5. The key finding

is that the adjusted R2 from equation (5) has dropped from about 70% in the 1960s to under 15%

in the more recent years, a decline only slightly less pronounced than that from 70% to under

10% in the unadjusted Dechow (1994) model reported in Panel A of Table 2. In Panel B of Table

5 we examine the change in the adjusted R2 and the coefficients β1 and β3 from model (5) in a

more systematic manner by regressing each on a time trend. We observe that that the coefficient

on the time trend is statistically significantly negative (positive) for the R2 (both βs) and the

goodness of fit of the model is very (relatively) high. The fitted values of the beginning and

ending year of the sample for adjusted R2 are 73% and 3%, again very comparable to the decline

in the fitted values from 75% to 9% in the unadjusted Dechow (1994) model reported in Panel C

of Table 2.

18

Figure 4 presents the results from Table 5 in graphical form. As noted above, for

completeness, we follow Ball and Shivakumar (2006) and consider both the specification where

an expectation of loss is represented by a negative CFOt and a negative ΔCFOt. Panels A and B

show a continuous and smooth decline in the adjusted R2 of the model. Panels C and D show the

evolution in coefficients β1 and β3 from model (5). We find that the coefficient on

contemporaneous cash flows from operations becomes less negative over time, particularly for

the changes specification of the model. Interestingly, we also observe that while the coefficient

on the interaction term D*CFOt is indeed increasing over the early part of the sample period, in

line with Ball and Shivakumar’s (2006) conjecture that the conservative recognition of expected

losses has increased over time, it is actually stable or even declining on the second half of the

sample (roughly 1990 to 2012). This seems to suggest that both the timing role and the timely

loss recognition role of accruals have been decreasing over the past twenty years.

Overall, we conclude that the attenuation in the timing role of accruals is not driven by an

offsetting increase in the accruals’ role of timely gain and loss recognition.4 The adjusted R2

from Ball and Shivakumar (2006) non-linear specification declines over time in a similar manner

to the one from Dechow (1994) linear specification.

4.2.3 One-time items

In this subsection, we examine whether one-time items explain our results. Prior literature

shows that non-recurring items have drastically increased in both frequency and magnitude over

time (Bradshaw and Sloan, 2002). Because one-time items are by definition (or at least should

4 We note that our patterns of coefficients appear different from those reported in Figure 1 of Ball and Shivakumar (2006). While our measure of accruals and correspondingly cash flows is based on the balance sheet data for all periods, their measures are taken from cash flow statements when available. We replicate our analysis using cash flow statement data when available and still observe patterns very similar to those that we report in Figure 4. We conjecture that the difference stems from the sample composition.

19

be) transient and do not play into the accrual accounting smoothing of earnings, they may be

contributing to the observed attenuation. Although our main analysis already excludes

extraordinary items from the measure of earnings, we further replicate our work excluding

special items, which have been shown to have drastically increased over the past fifty years in

both frequency and magnitude. Specifically, we define earnings excluding one-time items as

earnings before extraordinary items minus (1-t)*special items, where t is the top statutory tax

rate.

In untabulated analysis, we observe that the declines in the adjusted R2 from both the

Dechow (1994) and the Dechow and Dichev (2002) models are very similar to those reported in

Tables 2 and 3. We acknowledge that some one-time items are buried in recurring items and, as a

result, we cannot completely tease out one-time items from our accruals and cash flow measures.

While we find little evidence that one-time items contribute to the attenuation of the timing role

of accrual accounting, we leave a full investigation of this issue to future research.

5. Robustness Tests

There could be a number of other operations-based and regulatory reasons why the

timing role of accrual accounting has diminished over time. We carry out a battery of additional

tests to explore alternative explanations to the observed decline in the negative association

between accruals and cash flows. Table 6 contains the results of analyses which consider the

impact of expanding the conceptual smoothing window, a change in the sample composition, and

the effect of firms engaging in mergers and acquisitions activity. We briefly address each of

these analyses below.

20

First, we consider the possibility that while the smoothing of cash flows via accruals has

decreased over the adjacent periods, the window of smoothing has expanded – i.e. accruals today

are expected to correspond to cash flows in non-adjacent fiscal periods. Thus, we expand the

Dechow and Dichev (2002) model of mapping the past, present, and future cash flows into

accruals to include years -2 to +2 (Model 1 of Table 6) and years -3 to +3 (Model 2 of Table 6).

We observe that the negative loading of the time trend variable for adjusted R2 is largely

unchanged from that reported in Panel B of Table 3.

Next, we consider a change in the sample composition. Because we are examining a very

long time-series it is feasible that the sample composition had changed significantly over time,

both in terms of specific firms and in terms of distinct industries gaining and losing prominence.

We follow the logic of Dichev and Tang (2008) and redo our analysis on the sample of largest

1,000 firms in each year as measured by total assets. Model 3 of Table 6 shows that the

coefficient on the time trend and the goodness of fit of the model are only slightly decreased (by

about 25% and 20% respectively) as compared to those reported in the main analysis.

Finally, we examine the effect of mergers and acquisitions. As noted previously, because

we want to examine a time-series variation in the role of accruals extending prior to 1988 we rely

on the balance sheet approach of accruals estimation. Because this methodology has been shown

to suffer from measurement errors in the presence of firms with large acquisitions, we exclude

from the sample firm-years with significant M&A activities as measured by either sales

contributions (Models 4 and 5 of Table 6) or earnings contributions (Models 6 and 7 of Table 6).

We find the results regarding the attenuation of the timing role of accrual accounting unchanged.

21

We carry out several analyses which are untabulated for purposes of brevity. First, we

consider whether the length of the operating cycle changes over time. Dechow and Dichev

(2002) suggest that the magnitude of the estimation errors in the accrual generating process is

related to the length of the operating cycle. We examine the average annual operating cycle for

our sample and find no pattern of a systematic change over the past fifty years. Second, we

examine the pattern of absolute total accruals over the sample period, in line with the Dechow

and Dichev (2002) observation that a larger magnitude of the accruals will impede the mapping

of accruals into cash flows. We do not find a significant increase in absolute accruals over time.

Finally, we repeat the analysis by each 2-digit SIC industry and calculate the descriptive

statistics of the adjusted R2 from the Dechow (1994) and the Dechow and Dichev (2002) models

across industries. The mean, 1st quartile, median, and 3rd quartile of the adjusted R2 all decline

substantially over time, indicating that the findings are not driven by a specific industry or a

significant change in the composition of firms across industries. Overall, we find that our results

on the loss of the timing role of accrual accounting are robust to alternative samples, research

designs, and specifications.

6. Conclusions

The negative association between contemporaneous accruals and cash flows is inherent in

accrual accounting. Both accounting research and teaching largely take this negative association

as given. In this paper, we show that it is important to revisit whether this supposition is still

valid today. Using the models based on Dechow (1994) and Dechow and Dichev (2002) to

examine the change in the timing role of accrual accounting over the past fifty years, we find

22

evidence of a pronounced and continuous decline in this property of accrual accounting. In fact,

the negative association between accruals and cash flows has nearly disappeared in the recent

years. The continuity and smoothness of the attenuation between the two components of earnings

suggests that the decline is not due to a specific regulatory or environmental regime shift.

We explore a variety of potential reasons for the decline in the timing role of accrual

accounting including a change in cash flow volatility, the loss of matching between revenues and

expenses, a change in the asymmetrically timely recognition of gains and losses, the effect of

one-time items, and industry effects. We find some evidence that an increase in cash flow

volatility partially explains the decline in the timing role of accrual accounting, but the bulk of

the attenuation remains unexplained.

While we document strong evidence that the timing role of accrual accounting has largely

disappeared, many questions remain unanswered and, thus, are open to future research. For

example, although we find limited evidence on cash flow volatility and are able to rule out a

number of other potential explanations for the decline in the timing role of accrual accounting,

we do not offer a definitive answer as to what is indeed the main factor or factors responsible for

the attenuation. It may be that accruals are increasingly representative of estimation errors or

corrections of prior estimation errors which do not map into cash flows. It is also possible that

FASB’s push towards fair value accounting has made accruals less correlated with cash flows

overall. Lastly, if the smoothing role of accruals is at least partly due to managerial earnings

manipulation, then a temporal change of managerial behavior, due to regulatory or enforcement

developments or capital market shifts, may contribute to the observed attenuation. Even if

accrual accounting still serves its timing role, the increase in magnitude of other elements of

accruals (e.g., noise, fair value adjustments, and earnings management) swamps the timing role

23

of accrual accounting in our models, resulting in a decline in the extent of the negative

correlation between accruals and cash flows. We leave these potential explanations to be

explored in future research. A related question which may be of interest to researchers and

practitioners is whether the overall usefulness of accruals, as utilized in valuation and contracting

settings, has changed over time.

24

References:

Ball, R. and L. Shivakumar. 2006. The role of accruals in asymmetrically timely gain and loss

recognition. Journal of Accounting Research 44(2), 207-242.

Barone, G. and M. Magilke. 2009. An examination of the effects of investor sophistication on the

pricing of accruals and cash flows. Journal of Accounting, Auditing & Finance 24(3),

385-414.

Beatty, A., B. Ke, and K. Petroni. 2002. Earnings management to avoid earnings declines across

publicly and privately held banks. The Accounting Review 77(3), 547-570.

Bradshaw, M. and R. Sloan. 2002. GAAP versus the street: an empirical assessment of two

alternative definitions of earnings. Journal of Accounting Research 40(1), 41-66.

Dechow, P. 1994. Accounting earnings and cash flows as measures of firm performance: The

role of accounting accruals. Journal of Accounting and Economics 18(1), 3–42.

Dechow, P. and I. Dichev. 2002. The quality of accruals and earnings: The role of accrual

estimation errors. The Accounting Review 77(Supplement), 35–59.

Dechow, P., S. P. Kothari, and R. Watts. 1998. The relation between earnings and cash flows.

Journal of Accounting and Economics 25(2), 133–168.

Dechow, P., W. Ge, and C. Schrand. 2010. Understanding earnings quality: A review of the

proxies, their determinants and their consequences. Journal of Accounting and

Economics 50(2-3), 344-401.

Dichev, I. and V. Tang. 2008. Matching and the changing properties of accounting earnings over

the last 40 years. The Accounting Review 83(6), 1425-1460.

Francis, J., R. LaFond, P. Olsson, and K. Schipper. 2004. Costs of equity and earnings attributes.

The Accounting Review 79(4), 967-1010.

Francis, J., R. LaFond, P. Olsson, and K. Schipper. 2005. The market pricing of accruals quality.

Journal of Accounting and Economics 39(2), 295-327.

25

Givoly, D. and C. Hayn. 2000. The changing time-series properties of earnings, cash flows and

accruals: Has financial reporting become more conservative. Journal of Accounting and

Economics 29(3), 287-320.

Hribar, P. and D. Collins. 2002. Errors in estimating accruals: Implications for empirical

research. Journal of Accounting Research 40(1), 105-134.

Leuz, C., D. Nanda, and P. Wysocki. 2003. Earnings management and investor protection: An

international comparison. Journal of Financial Economics 69(3), 505-527.

McNichols, M. and G. P. Wilson. 1988. Evidence of earnings management from the provision

for bad debts. Journal of Accounting Research 26(Supplement), 1-31.

Myers, J., L. Myers, and D. Skinner. 2007. Earnings momentum and earnings management.

Journal of Accounting, Auditing & Finance 22(2), 249-284.

Rayburn, J. 1986. The association of operating cash flow and accruals with security returns.

Journal of Accounting Research 24(Supplement), 112-133.

Richardson, S., R. Sloan, M. Soliman, and I. Tuna. 2005. Accrual reliability, earnings persistence

and stock prices. Journal of Accounting and Economics 39(3), 437-485.

Sloan, R. 1996. Do stock prices fully reflect information in accruals and cash flows about future

earnings? The Accounting Review 71(3), 289-315.

26

Figure 1

The relation between accruals and cash flows over time: Dechow (1994) Panel A: Adjusted R2 – Levels Model: 𝑇𝐴𝐶𝐶𝑡 = 𝛽0 + 𝛽1𝐶𝐹𝑂𝑡 + 𝑒𝑡

Panel B: Adjusted R2 – Changes Model: ∆𝑇𝐴𝐶𝐶𝑡 = 𝛽0 + 𝛽1∆𝐶𝐹𝑂𝑡 + 𝑒𝑡

0.00

0.10

0.20

0.30

0.40

0.50

0.60

0.70

0.80

0.00

0.10

0.20

0.30

0.40

0.50

0.60

0.70

0.80

0.90

1.00

27

Panel C: Coefficient on CFOt (β1) – Levels Model: 𝑇𝐴𝐶𝐶𝑡 = 𝛽0 + 𝛽1𝐶𝐹𝑂𝑡 + 𝑒𝑡

Panel D: Coefficient on CFOt (β1) – Changes Model: ∆𝑇𝐴𝐶𝐶𝑡 = 𝛽0 + 𝛽1∆𝐶𝐹𝑂𝑡 + 𝑒𝑡

TACCt is total accruals measured as changes in non-cash current assets minus changes in non-debt current liabilities minus depreciation expense scaled by average total assets. CFOt is cash flows from operations measured as earnings minus total accruals, where earnings are earnings before extraordinary items scaled by average total assets. The sample includes 228,847 firm-year observations with non-missing TACCt and CFOt from 1964 to 2012. Each year, all variables are Winsorized at 1 percent and 99 percent.

-0.900

-0.800

-0.700

-0.600

-0.500

-0.400

-0.300

-0.200

-0.100

0.00019

64

1966

1968

1970

1972

1974

1976

1978

1980

1982

1984

1986

1988

1990

1992

1994

1996

1998

2000

2002

2004

2006

2008

2010

2012

-1.200

-1.000

-0.800

-0.600

-0.400

-0.200

0.000

1964

1966

1968

1970

1972

1974

1976

1978

1980

1982

1984

1986

1988

1990

1992

1994

1996

1998

2000

2002

2004

2006

2008

2010

2012

28

Figure 2 The relation between accruals and past, current, and future cash flows over time:

Dechow and Dichev (2002)

Panel A: Adjusted R2: 𝑇𝐴𝐶𝐶𝑡 = 𝛽0 + 𝛽1𝐶𝐹𝑂𝑡−1 + 𝛽2𝐶𝐹𝑂𝑡 + 𝛽3𝐶𝐹𝑂𝑡+1 + 𝑒𝑡

Panel B: Coefficients β1 β2 β3: 𝑇𝐴𝐶𝐶𝑡 = 𝛽0 + 𝛽1𝐶𝐹𝑂𝑡−1 + 𝛽2𝐶𝐹𝑂𝑡 + 𝛽3𝐶𝐹𝑂𝑡+1 + 𝑒𝑡

TACCt is total accruals measured as changes in non-cash current assets minus changes in non-debt current liabilities minus depreciation expense scaled by average total assets. CFOt is cash flows from operations measured as earnings minus total accruals, where earnings are earnings before extraordinary items scaled by average total assets. The sample includes 206,199 firm-year observations with non-missing TACCt, CFOt-1, CFOt, and CFOt+1 from 1965 to 2011. Each year, all variables are Winsorized at 1 percent and 99 percent.

0.00

0.10

0.20

0.30

0.40

0.50

0.60

0.70

0.80

0.90

-1.000

-0.800

-0.600

-0.400

-0.200

0.000

0.200

0.400

1965

1967

1969

1971

1973

1975

1977

1979

1981

1983

1985

1987

1989

1991

1993

1995

1997

1999

2001

2003

2005

2007

2009

2011

CFOt-1

CFOt

CFOt+1

29

Figure 3 The time series pattern of cash flow volatility and matching between revenues and expenses

Panel A: The time-series pattern of accruals and cash flow volatility.

Panel B: Adjusted R2 from Dechow (1994), Dechow and Dichev (2002), and Dichev and Tang (2008) models.

Et is earnings before extraordinary items scaled by average total assets. TACCt is total accruals measured as working capital accruals minus depreciation expense scaled by average total assets. CFOt is cash flows from operations measured as earnings minus total accruals. CFO_SCFt is cash flows from operations as reported on the Statement of Cash Flows. Dechow is the adjusted R2 from the model 𝑇𝐴𝐶𝐶𝑡 = 𝛽0 + 𝛽1𝐶𝐹𝑂𝑡 + 𝑒𝑡. Dechow-Dichev is the adjusted R2 from the model 𝑇𝐴𝐶𝐶𝑡 = 𝛽0 + 𝛽1𝐶𝐹𝑂𝑡−1 + 𝛽2𝐶𝐹𝑂𝑡 + 𝛽3𝐶𝐹𝑂𝑡+1 + 𝑒𝑡. Dichev-Tang is the adjusted R2 from the model 𝑆𝐴𝐿𝐸𝑡 = 𝛽0 + 𝛽1𝐸𝑋𝑃𝐸𝑁𝑆𝐸𝑡−1 + 𝛽2𝐸𝑋𝑃𝐸𝑁𝑆𝐸𝑡 + 𝛽3𝐸𝑋𝑃𝐸𝑁𝑆𝐸𝑡+1 + 𝑒𝑡.

0.00

0.05

0.10

0.15

0.20

0.25

0.30

0.35

1964

1966

1968

1970

1972

1974

1976

1978

1980

1982

1984

1986

1988

1990

1992

1994

1996

1998

2000

2002

2004

2006

2008

2010

2012

std(E)

std(TACC)

std(CFO)

std(CFO_SCF)

0

0.2

0.4

0.6

0.8

1

1.2

1964

1966

1968

1970

1972

1974

1976

1978

1980

1982

1984

1986

1988

1990

1992

1994

1996

1998

2000

2002

2004

2006

2008

2010

2012

Dechow

Dechow-Dichev

Dichev-Tang

30

Figure 4 The timely loss recognition role of accruals over time – Ball and Shivakumar (2006)

Panel A: Adjusted R2: 𝑇𝐴𝐶𝐶𝑡 = 𝛽0 + 𝛽1𝐶𝐹𝑂𝑡 + 𝛽2𝐷 + 𝛽3𝐷 ∗ 𝐶𝐹𝑂𝑡 + 𝑒𝑡 (where D is equal to one if CFOt is negative)

Panel B: Adjusted R2: 𝑇𝐴𝐶𝐶𝑡 = 𝛽0 + 𝛽1𝐶𝐹𝑂𝑡 + 𝛽2𝐷 + 𝛽3𝐷 ∗ 𝐶𝐹𝑂𝑡 + 𝑒𝑡 (where D is equal to one if year over year change in CFOt is negative)

0.00

0.10

0.20

0.30

0.40

0.50

0.60

0.70

0.80

0.90

0.00

0.10

0.20

0.30

0.40

0.50

0.60

0.70

0.80

31

Panel C: Coefficients β1 β3: 𝑇𝐴𝐶𝐶𝑡 = 𝛽0 + 𝛽1𝐶𝐹𝑂𝑡 + 𝛽2𝐷 + 𝛽3𝐷 ∗ 𝐶𝐹𝑂𝑡 + 𝑒𝑡 (where D is equal to one if CFOt is negative)

Panel D: Coefficients β1 β3: 𝑇𝐴𝐶𝐶𝑡 = 𝛽0 + 𝛽1𝐶𝐹𝑂𝑡 + 𝛽2𝐷 + 𝛽3𝐷 ∗ 𝐶𝐹𝑂𝑡 + 𝑒𝑡 (where D is equal to one if year over year change in CFOt is negative)

TACCt is total accruals measured as changes in non-cash current assets minus changes in non-debt current liabilities minus depreciation expense scaled by average total assets. CFOt is cash flows from operations measured as earnings minus total accruals, where earnings are earnings before extraordinary items scaled by average total assets. D is a dummy variable with the value of 1 if CFOt is negative in Panels A and C and if ΔCFOt is negative in Panels B and D. The sample includes 228,847 firm-year observations with non-missing TACCt and CFOt from 1964 to 2012. Each year, all variables are Winsorized at 1 percent and 99 percent.

-0.800

-0.600

-0.400

-0.200

0.000

0.200

0.400

0.600

0.80019

6419

6619

6819

7019

7219

7419

7619

7819

8019

8219

8419

8619

8819

9019

9219

9419

9619

9820

0020

0220

0420

0620

0820

1020

12

CFOt

D*CFOt

-0.800

-0.600

-0.400

-0.200

0.000

0.200

0.400

1965

1967

1969

1971

1973

1975

1977

1979

1981

1983

1985

1987

1989

1991

1993

1995

1997

1999

2001

2003

2005

2007

2009

2011

CFO

DD_CFO

32

Table 1 Descriptive statistics

Panel A: Descriptive statistics Variable N Mean Stdev Min Q1 Median Q3 Max

Et 228847 0.000 0.178 -1.472 -0.009 0.040 0.078 0.408

WACCt 228847 0.013 0.090 -0.392 -0.023 0.007 0.047 0.411

TACCt 228847 -0.034 0.100 -0.490 -0.079 -0.035 0.008 0.378

CFOt-1 206199 0.041 0.173 -1.461 0.000 0.069 0.125 0.518

CFOt 228847 0.034 0.185 -1.461 -0.005 0.067 0.124 0.518

CFOt+1 206199 0.040 0.173 -1.461 0.002 0.069 0.124 0.518

MVt 200990 1808 11088 0.5 28 115 565 1819782

BMt 200110 0.73 1.29 -32.15 0.32 0.59 1.02 11.18 Panel B: Correlation matrix for key variables. Pearson (Spearman) correlations are shown above (below) the main diagonal. Et WACCt TACCt CFOt-1 CFOt CFOt+1 Et 1 0.23** 0.26** 0.61** 0.83** 0.60** WACCt 0.25** 1 0.94** 0.09** -0.29** 0.07** TACCt 0.24** 0.89** 1 0.05** -0.29** 0.02** CFOt-1 0.50** 0.09** -0.02** 1 0.46** 0.49** CFOt 0.63** -0.40** -0.47** 0.57** 1 0.57** CFOt+1 0.43** 0.03** -0.08** 0.41** 0.46** 1 ** Significant at the 1 percent level. Et is earnings before extraordinary items scaled by average total assets. WACCt is working capital accruals measured as changes in non-cash current assets minus changes in non-debt current liabilities scaled by average total assets. TACCt is total accruals measured as working capital accruals minus depreciation expense scaled by average total assets. CFOt is cash flows from operations measured as earnings minus total accruals. MVt is the market value of equity at a firm’s fiscal year end. BMt is the book-to-market ratio, calculated as the book value of equity scaled by the market value of equity at fiscal year-end. The sample includes 228,847 firm-year observations with non-missing TACCt and CFOt from 1964 to 2012. Each year, all variables except for MVt are Winsorized at 1 percent and 99 percent.

33

Table 2 The relation between accruals and cash flows over time: Dechow (1994)

Panel A: Regression model on levels: 𝑇𝐴𝐶𝐶𝑡 = 𝛽0 + 𝛽1𝐶𝐹𝑂𝑡 + 𝑒𝑡

Year 𝛽0 𝛽1(CFOt) Adj. R2

year 𝛽0 𝛽1(CFOt) Adj. R2 1964 0.043 -0.695 0.67

1989 -0.011 -0.449 0.32

1965 0.053 -0.718 0.67

1990 -0.030 -0.325 0.18 1966 0.055 -0.708 0.65

1991 -0.036 -0.304 0.17

1967 0.054 -0.785 0.69

1992 -0.027 -0.276 0.17 1968 0.055 -0.810 0.75

1993 -0.026 -0.254 0.15

1969 0.051 -0.789 0.74

1994 -0.016 -0.262 0.16 1970 0.031 -0.748 0.62

1995 -0.020 -0.233 0.13

1971 0.030 -0.759 0.64

1996 -0.028 -0.196 0.11 1972 0.043 -0.800 0.71

1997 -0.031 -0.157 0.08

1973 0.049 -0.753 0.68

1998 -0.043 -0.114 0.05 1974 0.040 -0.706 0.60

1999 -0.051 -0.069 0.02

1975 0.020 -0.689 0.56

2000 -0.054 -0.009 0.00 1976 0.032 -0.690 0.58

2001 -0.079 -0.021 0.00

1977 0.036 -0.693 0.58

2002 -0.062 -0.070 0.03 1978 0.039 -0.688 0.58

2003 -0.048 -0.114 0.07

1979 0.039 -0.670 0.52

2004 -0.035 -0.108 0.07 1980 0.031 -0.689 0.55

2005 -0.039 -0.091 0.05

1981 0.026 -0.642 0.52

2006 -0.034 -0.101 0.06 1982 0.003 -0.591 0.44

2007 -0.035 -0.107 0.07

1983 0.008 -0.594 0.44

2008 -0.044 -0.062 0.04 1984 0.012 -0.536 0.38

2009 -0.051 -0.127 0.09

1985 -0.009 -0.475 0.30

2010 -0.036 -0.110 0.07 1986 -0.010 -0.475 0.31

2011 -0.033 -0.091 0.06

1987 0.002 -0.565 0.41

2012 -0.040 -0.075 0.05 1988 -0.004 -0.475 0.34

34

Panel B: Regression model on changes: ∆𝑇𝐴𝐶𝐶𝑡 = 𝛽0 + 𝛽1∆𝐶𝐹𝑂𝑡 + 𝑒𝑡 Year 𝛽0 𝛽1(ΔCFOt) Adj. R2

year 𝛽0 𝛽1(ΔCFOt) Adj. R2

1964 0.007 -0.937 0.90

1989 -0.012 -0.712 0.61 1965 0.006 -0.910 0.91

1990 -0.017 -0.664 0.55

1966 0.001 -0.928 0.91

1991 -0.013 -0.601 0.47 1967 -0.009 -0.903 0.90

1992 0.000 -0.571 0.43

1968 -0.003 -0.945 0.92

1993 -0.006 -0.542 0.38 1969 -0.006 -0.937 0.92

1994 0.003 -0.502 0.34

1970 -0.016 -0.963 0.89

1995 -0.008 -0.522 0.38 1971 -0.001 -0.909 0.85

1996 -0.014 -0.467 0.34

1972 0.006 -0.935 0.88

1997 -0.009 -0.459 0.32 1973 0.005 -0.907 0.86

1998 -0.017 -0.384 0.25

1974 -0.007 -0.858 0.79

1999 -0.003 -0.389 0.27 1975 -0.010 -0.873 0.84

2000 -0.009 -0.243 0.14

1976 0.006 -0.903 0.82

2001 -0.031 -0.223 0.13 1977 0.001 -0.887 0.81

2002 0.013 -0.225 0.13

1978 0.002 -0.865 0.81

2003 0.012 -0.276 0.21 1979 0.000 -0.855 0.78

2004 0.011 -0.312 0.22

1980 -0.010 -0.869 0.78

2005 -0.004 -0.338 0.25 1981 -0.006 -0.849 0.80

2006 0.003 -0.334 0.26

1982 -0.021 -0.814 0.70

2007 -0.004 -0.409 0.34 1983 -0.001 -0.804 0.67

2008 -0.019 -0.237 0.17

1984 -0.002 -0.771 0.65

2009 -0.008 -0.223 0.15 1985 -0.022 -0.768 0.64

2010 0.018 -0.323 0.25

1986 -0.012 -0.688 0.55

2011 0.002 -0.285 0.23 1987 0.001 -0.744 0.59

2012 -0.009 -0.279 0.23

1988 -0.005 -0.744 0.63

35

Panel C: Regression results for time trends in 𝛽1(CFOt) and the adj. R2 for the levels model 𝛽1(𝐶𝐹𝑂𝑡) = 𝑏0 + 𝑏1𝑇𝑖𝑚𝑒 + 𝜀 𝐴𝑑𝑗.𝑅2 = 𝑏0 + 𝑏1𝑇𝑖𝑚𝑒 + 𝜀

Regression b0

(t-stat) b1

(t-stat) R2 Fitted value year 1964

Fitted value year 2012

𝛽1(𝐶𝐹𝑂𝑡) -0.87

(-36.53) 0.0187 (21.96) 0.911 -0.870 0.028

𝐴𝑑𝑗.𝑅2 0.745

(34.42) -0.0173 (-22.27) 0.913 0.745 -0.085

Panel D: Regression results for time trends in 𝛽1(ΔCFOt) and the adj. R2 for the changes model 𝛽1(∆𝐶𝐹𝑂𝑡) = 𝑏0 + 𝑏1𝑇𝑖𝑚𝑒 + 𝜀 𝐴𝑑𝑗.𝑅2 = 𝑏0 + 𝑏1𝑇𝑖𝑚𝑒 + 𝜀

Regression b0

(t-stat) b1

(t-stat) R2 Fitted value year 1964

Fitted value year 2012

𝛽1(∆𝐶𝐹𝑂𝑡) -1.055

(-48.17) 0.0175 (22.28) 0.911 -1.055 -0.215

𝐴𝑑𝑗.𝑅2 0.997

(45.42) -0.0187 (-23.73) 0.921 0.997 0.099

TACCt is total accruals measured as changes in non-cash current assets minus changes in non-debt current liabilities minus depreciation expense scaled by average total assets. CFOt is cash flows from operations measured as earnings minus total accruals, where earnings are earnings before extraordinary items scaled by average total assets. Time is the number of years since 1964. In Panel C 𝛽1(𝐶𝐹𝑂𝑡) and 𝐴𝑑𝑗.𝑅2 are the coefficient estimate and the adjusted R2 respectively from the levels Dechow (1994) model 𝑇𝐴𝐶𝐶𝑡 = 𝛽0 + 𝛽1𝐶𝐹𝑂𝑡 + 𝑒𝑡 estimated annually. In Panel D 𝛽1(∆𝐶𝐹𝑂𝑡) and 𝐴𝑑𝑗.𝑅2 are the coefficient estimate and the adjusted R2 respectively from the changes Dechow (1994) model ∆𝑇𝐴𝐶𝐶𝑡 = 𝛽0 + 𝛽1∆𝐶𝐹𝑂𝑡 + 𝑒𝑡 estimated annually. The sample includes 228,847 firm-year observations with non-missing TACCt and CFOt from 1964 to 2012. Each year, all variables except for Time are Winsorized at 1 percent and 99 percent.

36

Table 3 The relation between accruals and past, current, and future cash flows over time

Panel A: Regression model: 𝑇𝐴𝐶𝐶𝑡 = 𝛽0 + 𝛽1𝐶𝐹𝑂𝑡−1 + 𝛽2𝐶𝐹𝑂𝑡 + 𝛽3𝐶𝐹𝑂𝑡+1 + 𝑒𝑡

year Intercept CFOt-1 CFOt CFOt+1 Adj. R2 1965 0.040 0.170 -0.817 0.074 0.71 1966 0.045 0.170 -0.785 0.024 0.68 1967 0.043 0.166 -0.857 0.034 0.72 1968 0.044 0.133 -0.867 0.056 0.76 1969 0.039 0.127 -0.847 0.087 0.76 1970 0.023 0.158 -0.813 0.048 0.65 1971 0.020 0.172 -0.819 0.049 0.68 1972 0.031 0.164 -0.847 0.028 0.72 1973 0.042 0.149 -0.803 -0.007 0.71 1974 0.028 0.185 -0.776 0.066 0.65 1975 0.011 0.183 -0.734 0.067 0.65 1976 0.011 0.222 -0.750 0.044 0.65 1977 0.024 0.198 -0.774 0.036 0.64 1978 0.031 0.178 -0.769 0.028 0.63 1979 0.026 0.150 -0.748 0.105 0.58 1980 0.017 0.149 -0.746 0.093 0.61 1981 0.010 0.156 -0.707 0.084 0.58 1982 -0.013 0.155 -0.666 0.127 0.48 1983 -0.006 0.185 -0.690 0.096 0.51 1984 -0.002 0.179 -0.648 0.135 0.48 1985 -0.017 0.180 -0.600 0.137 0.41 1986 -0.021 0.147 -0.550 0.109 0.36 1987 -0.007 0.177 -0.664 0.112 0.48 1988 -0.008 0.132 -0.586 0.126 0.44 1989 -0.018 0.150 -0.560 0.119 0.42 1990 -0.038 0.174 -0.477 0.155 0.33 1991 -0.043 0.127 -0.441 0.139 0.27 1992 -0.036 0.177 -0.456 0.125 0.29 1993 -0.035 0.145 -0.386 0.114 0.22 1994 -0.023 0.140 -0.398 0.112 0.22 1995 -0.026 0.140 -0.389 0.131 0.24 1996 -0.031 0.117 -0.385 0.169 0.24 1997 -0.031 0.107 -0.315 0.126 0.18 1998 -0.042 0.105 -0.285 0.141 0.16 1999 -0.040 0.127 -0.291 0.118 0.19 2000 -0.048 0.073 -0.149 0.129 0.10 2001 -0.075 0.085 -0.126 0.080 0.06 2002 -0.060 0.104 -0.197 0.085 0.12

37

2003 -0.045 0.100 -0.244 0.085 0.16 2004 -0.035 0.109 -0.274 0.117 0.17 2005 -0.037 0.108 -0.275 0.124 0.19 2006 -0.032 0.095 -0.292 0.168 0.22 2007 -0.029 0.120 -0.276 0.090 0.19 2008 -0.047 0.069 -0.143 0.070 0.09 2009 -0.052 0.062 -0.231 0.098 0.16 2010 -0.034 0.108 -0.242 0.079 0.18 2011 -0.031 0.067 -0.219 0.107 0.14

Panel B: Regression results for time trends in 𝛽(CFO) and adj. R2 𝛽1(𝐶𝐹𝑂𝑡−1) = 𝑏0 + 𝑏1𝑇𝑖𝑚𝑒 + 𝜀 𝛽2(𝐶𝐹𝑂𝑡) = 𝑏0 + 𝑏1𝑇𝑖𝑚𝑒 + 𝜀 𝛽3(𝐶𝐹𝑂𝑡+1) = 𝑏0 + 𝑏1𝑇𝑖𝑚𝑒 + 𝜀 𝐴𝑑𝑗.𝑅2 = 𝑏0 + 𝑏1𝑇𝑖𝑚𝑒 + 𝜀

Regression b0

(t-stat) b1

(t-stat) R2 Fitted value year 1965

Fitted value year 2011

𝛽1(𝐶𝐹𝑂𝑡−1) 0.190

(25.82) -0.0021 (-7.81) 0.575 0.188 0.091

𝛽2(𝐶𝐹𝑂𝑡) -0.939

(-47.02) 0.017

(23.52) 0.925 -0.922 -0.140

𝛽3(𝐶𝐹𝑂𝑡+1) 0.054 (5.41)

0.0017 (4.65) 0.324 0.056 0.134

𝐴𝑑𝑗.𝑅2 0.787

(39.36) -0.016

(-21.90) 0.914 0.771 0.035 TACCt is total accruals measured as changes in non-cash current assets minus changes in non-debt current liabilities minus depreciation expense scaled by average total assets. CFOt is cash flows from operations measured as earnings minus total accruals, where earnings are earnings before extraordinary items scaled by average total assets. Time is the number of years since 1964. 𝛽1(𝐶𝐹𝑂𝑡−1), 𝛽2(𝐶𝐹𝑂𝑡), 𝛽3(𝐶𝐹𝑂𝑡+1) and 𝐴𝑑𝑗.𝑅2 are the coefficient estimates and the adjusted R2 respectively from the Dechow and Dichev (2002) model 𝑇𝐴𝐶𝐶𝑡 = 𝛽0 + 𝛽1𝐶𝐹𝑂𝑡−1 + 𝛽2𝐶𝐹𝑂𝑡 +𝛽3𝐶𝐹𝑂𝑡+1 + 𝑒𝑡 estimated annually. The sample includes 206,199 firm-year observations with non-missing TACCt, CFOt-1 , CFOt and CFOt+1 from 1965 to 2011. Each year, all variables except for Time are Winsorized at 1 percent and 99 percent.

38

Table 4 The impact of a temporal change in cash flow volatility and

matching of revenues and expenses 𝑀𝑜𝑑𝑒𝑙: 𝐴𝑑𝑗.𝑅2(𝐷𝑒𝑐ℎ𝑜𝑤_𝐷𝑖𝑐ℎ𝑒𝑣) = 𝑏0 + 𝑏1𝑇𝑖𝑚𝑒 + 𝑏2𝑆𝑡𝑑(𝐶𝐹𝑂) + 𝑏3𝐴𝑑𝑗.𝑅2(𝐷𝑖𝑐ℎ𝑒𝑣_𝑇𝑎𝑛𝑔) + 𝜀

Regression 1 2 3 4

Intercept 0.787

(39.36) 0.843

(32.51) -0.856 (-1.09)

2.699 (1.58)

Time -0.016

(-21.90) -0.011 (-7.48)

-0.013 (-9.04)

-0.011 (-7.28)

Std(CFO) -1.067 (-3.12)

-1.838 (-2.34)

Adj.R2(Dichev_Tang) 1.669 (2.23)

-1.788 (-1.09)

Adj. R2 0.915 0.929 0.922 0.929 Adj.R2(Dechow_Dichev) is the adjusted R2 from the Dechow-Dichev (2002) model, a proxy for the degree of the timing role of accrual accounting. Time is the number of years since 1964. Std(CFO) is the standard deviation of cash flows. Adj.R2(Dichev_Tang) is the adjusted R2 from the Dichev-Tang (2008) model, a proxy for the match between revenue and expenses. Dechow-Dichev model: 𝑇𝐴𝐶𝐶𝑡 = 𝛽0 + 𝛽1𝐶𝐹𝑂𝑡−1 + 𝛽2𝐶𝐹𝑂𝑡 + 𝛽3𝐶𝐹𝑂𝑡+1 + 𝑒𝑡 Dichev-Tang model: 𝑆𝐴𝐿𝐸𝑡 = 𝛽0 + 𝛽1𝐸𝑋𝑃𝐸𝑁𝑆𝐸𝑡−1 + 𝛽2𝐸𝑋𝑃𝐸𝑁𝑆𝐸𝑡 + 𝛽3𝐸𝑋𝑃𝐸𝑁𝑆𝐸𝑡+1 + 𝑒𝑡 where TACCt is total accruals measured as changes in non-cash current assets minus changes in non-debt current liabilities minus depreciation expense scaled by average total assets. CFOt is cash flows from operations measured as earnings minus total accruals scaled by average total assets. SALEt is net sales scaled by average total assets. EXPENSEt is expenses measured as sales minus earnings before extraordinary items scaled by average total assets. The sample includes 206,199 firm-year observations with non-missing TACCt, CFOt-1 , CFOt and CFOt+1 from 1965 to 2011. Each year, all variables except for Time are Winsorized at 1 percent and 99 percent. T-statistics are in parenthesis.

39

Table 5 The timely loss recognition role of accruals over time – Ball and Shivakumar (2006)

Panel A: Regression model: 𝑇𝐴𝐶𝐶𝑡 = 𝛽0 + 𝛽1𝐶𝐹𝑂𝑡 + 𝛽2𝐷 + 𝛽3𝐷 ∗ 𝐶𝐹𝑂𝑡 + 𝑒𝑡

year Intercept CFOt D D*CFOt Adj. R2 1964 0.030 -0.593 0.011 -0.366 0.69 1965 0.036 -0.574 0.007 -0.514 0.70 1966 0.040 -0.575 0.013 -0.357 0.67 1967 0.034 -0.598 0.014 -0.400 0.72 1968 0.029 -0.579 0.018 -0.486 0.78 1969 0.029 -0.579 0.016 -0.377 0.76 1970 0.020 -0.634 0.009 -0.273 0.63 1971 0.022 -0.689 0.003 -0.214 0.65 1972 0.031 -0.688 0.003 -0.301 0.72 1973 0.037 -0.646 0.015 -0.174 0.69 1974 0.029 -0.613 0.008 -0.241 0.60 1975 0.017 -0.673 0.005 -0.038 0.56 1976 0.026 -0.643 0.004 -0.142 0.58 1977 0.030 -0.645 0.010 -0.075 0.58 1978 0.032 -0.627 0.011 -0.115 0.59 1979 0.030 -0.604 0.021 -0.028 0.53 1980 0.029 -0.667 0.007 -0.013 0.55 1981 0.017 -0.584 0.022 -0.006 0.53 1982 0.007 -0.627 -0.002 0.094 0.44 1983 0.017 -0.661 -0.009 0.105 0.44 1984 0.024 -0.657 0.006 0.273 0.39 1985 0.011 -0.657 0.002 0.403 0.33 1986 0.006 -0.628 0.005 0.315 0.33 1987 0.011 -0.644 -0.001 0.168 0.41 1988 0.011 -0.610 0.008 0.313 0.36 1989 0.002 -0.560 -0.001 0.239 0.34 1990 0.001 -0.586 0.006 0.589 0.25 1991 -0.012 -0.522 0.005 0.483 0.22 1992 -0.005 -0.511 0.018 0.465 0.23 1993 -0.002 -0.518 0.030 0.530 0.23 1994 0.008 -0.526 0.031 0.528 0.23 1995 0.004 -0.492 0.032 0.518 0.22 1996 0.007 -0.540 0.028 0.598 0.22 1997 0.005 -0.526 0.016 0.558 0.18 1998 0.000 -0.554 0.008 0.615 0.17 1999 -0.007 -0.493 -0.003 0.542 0.13 2000 -0.014 -0.422 0.002 0.510 0.10 2001 -0.024 -0.527 -0.017 0.619 0.14

40

2002 -0.027 -0.383 -0.008 0.412 0.11 2003 -0.020 -0.368 0.000 0.377 0.15 2004 -0.007 -0.381 0.011 0.415 0.16 2005 -0.007 -0.385 0.004 0.419 0.15 2006 -0.005 -0.384 0.014 0.422 0.17 2007 -0.008 -0.366 0.002 0.350 0.13 2008 -0.010 -0.383 -0.010 0.391 0.12 2009 -0.033 -0.309 0.006 0.268 0.13 2010 -0.018 -0.298 0.018 0.305 0.13 2011 -0.010 -0.319 0.009 0.337 0.13 2012 -0.018 -0.305 -0.001 0.300 0.11

Panel B: Regression results for time trends in coefficient estimates and adjusted R2 𝛽1(𝐶𝐹𝑂𝑡) = 𝑏0 + 𝑏1𝑇𝑖𝑚𝑒 + 𝜀 𝛽3(𝐷 ∗ 𝐶𝐹𝑂𝑡) = 𝑏0 + 𝑏1𝑇𝑖𝑚𝑒 + 𝜀 𝐴𝑑𝑗.𝑅2 = 𝑏0 + 𝑏1𝑇𝑖𝑚𝑒 + 𝜀

Regression b0

(t-stat) b1

(t-stat) R2 Fitted value year 1965

Fitted value year 2011

𝛽1(𝐶𝐹𝑂𝑡) -0.701

(-39.05) 0.0068 (10.55) 0.703 -0.694 -0.381

𝛽3(𝐷 ∗ 𝐶𝐹𝑂𝑡) -0.315 (-6.07)

0.020 (10.84) 0.714 -0.295 0.625

𝐴𝑑𝑗.𝑅2 0.740

(42.09) -0.0152 (-24.09) 0.925 0.725 0.026

TACCt is total accruals measured as changes in non-cash current assets minus changes in non-debt current liabilities minus depreciation expense scaled by average total assets. CFOt is cash flows from operations measured as earnings minus total accruals, where earnings are earnings before extraordinary items scaled by average total assets. D is a dummy variable with the value of 1 if CFOt is negative. Time is the number of years since 1964. 𝛽1(𝐶𝐹𝑂𝑡), 𝛽3(𝐷 ∗ 𝐶𝐹𝑂𝑡) and 𝐴𝑑𝑗.𝑅2 are the coefficient estimates and the adjusted R2 respectively from Ball and Shivakumar (2006) model 𝑇𝐴𝐶𝐶𝑡 = 𝛽0 + 𝛽1𝐶𝐹𝑂𝑡 + 𝛽2𝐷 + 𝛽3𝐷 ∗ 𝐶𝐹𝑂𝑡 + 𝑒𝑡 estimated annually. The sample includes 228,847 firm-year observations with non-missing TACCt and CFOt from 1964 to 2012. Each year, all variables except for Time are Winsorized at 1 percent and 99 percent.

41

Table 6 Robustness checks: Time trend regressions of the adjusted R2

from the Dechow and Dichev (2002) model 𝑀𝑜𝑑𝑒𝑙: 𝐴𝑑𝑗.𝑅2 = 𝑏0 + 𝑏1𝑇𝑖𝑚𝑒 + 𝜀

Test b0

(t-stat) b1

(t-stat) Adj. R2

1 Expand the Dechow-Dichev model by including CFO from t-2 to t+2

0.809 (41.54)

-0.016 (-22.20) 0.915

2 Expand the Dechow-Dichev model by including CFO from t-3 to t+3

0.830 (43.43)

-0.017 (-22.90) 0.923

3 Top 1000 firms each year in terms of total assets

0.793 (28.00)

-0.012 (-11.22) 0.727

4 Exclude firms with sales contribution from M&As larger than 5%

0.780 (41.58)

-0.016 (-22.75) 0.917

5 Exclude firms with sales contribution from M&As larger than 1%

0.779 (41.58)

-0.016 (-22.74) 0.917

6 Exclude firms with earnings contribution from M&As larger than 5%

0.780 (41.56)

-0.016 (-22.77) 0.917

7 Exclude firms with earnings contribution from M&As larger than 1%

0.781 (41.42)

-0.016 (-22.67) 0.916

The dependent variable is the adjusted R2 from the Dechow and Dichev (2002) model or its expanded version, a proxy for the degree of accrual accounting. Time is the number of years since 1964. Dechow-Dichev model: 𝑇𝐴𝐶𝐶𝑡 = 𝛽0 + 𝛽1𝐶𝐹𝑂𝑡−1 + 𝛽2𝐶𝐹𝑂𝑡 + 𝛽3𝐶𝐹𝑂𝑡+1 + 𝑒𝑡 Expanded Dechow-Dichev model in Test 1:

𝑇𝐴𝐶𝐶𝑡 = 𝛽0 + 𝛽1𝐶𝐹𝑂𝑡−2 + 𝛽2𝐶𝐹𝑂𝑡−1 + 𝛽3𝐶𝐹𝑂𝑡 + 𝛽4𝐶𝐹𝑂𝑡+1 + 𝛽5𝐶𝐹𝑂𝑡+2 + 𝑒𝑡 Expanded Dechow-Dichev model in Test 2:

𝑇𝐴𝐶𝐶𝑡 = 𝛽0 + 𝛽1𝐶𝐹𝑂𝑡−3 + 𝛽2𝐶𝐹𝑂𝑡−2 + 𝛽3𝐶𝐹𝑂𝑡−1 + 𝛽4𝐶𝐹𝑂𝑡 + 𝛽5𝐶𝐹𝑂𝑡+1 + 𝛽6𝐶𝐹𝑂𝑡+2+𝛽7𝐶𝐹𝑂𝑡+3 + 𝑒𝑡 where TACCt is total accruals measured as changes in non-cash current assets minus changes in non-debt current liabilities minus depreciation expense scaled by average total assets. CFOt is cash flows from operations measured as earnings minus total accruals scaled by average total assets. Sales contribution is defined as sales from mergers and acquisitions (AQS) in year t divided by net sales in year t. Income contribution is defined as income from mergers and acquisitions (AQI) in year t divided by earnings before extraordinary items in year t. The sample includes 206,199 firm-year observations with non-missing TACCt, CFOt-1 , CFOt and CFOt+1 from 1965 to 2011. Each year, all variables except for Time are Winsorized at 1 percent and 99 percent.