Accounting Ethics

The ‘‘Numbers Game’’ in the Pre- and

Post-Sarbanes-Oxley Eras

ELI BARTOV* DANIEL A. COHEN**

We address two research questions in this study. First, is there a change in the prevalence of expectations management to meet or beat analysts’ earnings expectations in the aftermath of the 2001–2002 accounting scandals and the passage of the 2002 Sarbanes-Oxley Act (SOX)? Second, did the mix among the three mechanisms used for meeting or beating analysts’ earnings expectations: accrual-based earnings man- agement, real earnings management, and expectations management change in the Post-SOX period? We hypothesize and provide empirical evidence that the observed drop in the frequency of just meeting or beating analysts’ earnings expectations is associated with both (1) a decline in the use of downward expectations management and upward accrual-based earnings management in the Post-SOX period relative to the preceding seven-year period and (2) an increase in upward real earnings management activities.

1. Introduction

The accounting literature has documented that missing earnings expectations

is costly to companies, and that managing earnings and/or earnings expectations

to meet or beat analyst earnings expectations—a phenomenon often referred to

as the ‘‘numbers games’’ (Levitt [1998])—is widespread (e.g., DeGeorge, Patel,

and Zeckhauser [1999]; Bartov, Givoly, and Hayn [2002]; Matsumoto [2002]).

Earnings management may be in the form of an accounting action, also known

Keywords: Earnings Management; Real Earnings Management; Analysts’ Forecasts; Expectations Management; Sarbanes-Oxley Act

*New York University **New York University We thank Bala Balachandran (the editor), an anonymous reviewer, Yoel Beniluz, Larry Brown,

Valentin Dimitrov, Kim Dunn, Richard Frankel, Ohad Kadan, Jim Ohlson, Sugata Roychowdhury (the 2008 JAAF conference discussant), Tzachi Zach, and the seminar participants at the 2007 AAA Annual Meetings in Chicago (especially, Mikhail Pevzner, the discussant), NYU doctoral seminar, Georgia State University seminar, Rutgers University seminar, and Washington University in St. Louis seminar for useful comments and suggestions.

505

as accrual-based earnings management (e.g., Dhaliwal, Gleason, and Mills

[2004]) in which certain accruals are manipulated with no direct cash flow effect.

Examples of accrual-based earnings management include underaccruing of

expenses such as bad debt, delaying an asset write-down, or recognizing reve-

nues prematurely. Earnings management may also be in the form of a real eco-

nomic action, also known as real or transaction-based earnings management that

does affect cash flows (e.g., Roychowdhury [2006]). This activity is defined as

management actions with respect to real operating and investing activities that

deviate from normal business practices, where the primary objective is to achieve

certain reporting objectives. Examples include cutting discretionary expenses,

overproducing, or providing price discounts and lenient credit terms to boost

reported income in the short term. Earnings expectations management concerns

walking down analyst earnings expectations so as to transform an otherwise neg-

ative earnings surprise into a positive one (e.g., Bartov, Givoly, and Hayn [2002];

Matsumoto [2002]).

The purpose of this study is to examine two research questions left unan-

swered by prior research in this area. The first is to test for a change in the prev-

alence of analyst expectations management to meet or beat earnings targets

following the major accounting scandals of 2001–2002 at Enron, WorldCom, and

Global Crossing, to name just a few, and the new requirements introduced by the

2002 Sarbanes Oxley Act (SOX). We expect the prevalence of expectations man-

agement to decline in the Post-SOX period for two reasons. First, it may

decrease as a result of the significant attention earnings expectations management

received from the academic literature (e.g., Bartov, Givoly, and Hayn [2002];

Matsumoto [2002]; Jensen, Murphy, and Wruck [2004]), the financial press (e.g.,

McGee [1997]), and regulators (e.g., Levitt [1998]; Johnson [1999]) in the period

surrounding the passage of SOX. Second, corporate governance improvements

introduced by SOX may limit firms’ ability to manage analyst expectations. Still,

it is arguable that expectations management may increase due to a substitution

effect (e.g., Zang [2006]). That is, firms may substitute accrual-based earnings

management with expectations management in the Post-SOX period to meet or

beat certain performance benchmarks.

The second objective of this study is to explain any observed change in the

frequency of meeting or beating analyst earnings expectations in the Post-SOX

period. To that end, we test for a relation between the frequency of meeting or

beating earnings expectations and the prevalence of the three mechanisms used

for meeting or beating analyst earnings targets: accrual-based earnings manage-

ment, real earnings management, and earnings expectations management.

The aftermath of the passage of the Sarbanes-Oxley Act on July 30, 2002,

changed the financial reporting environment significantly. Specifically, Section

201 of SOX prohibits outside auditors from providing nine nonaudit services to

their audit clients (e.g., bookkeeping, appraisals, actuarial services, and invest-

ment-advisory work), and requires that other nonaudit services (e.g., tax services)

be approved in advance by the audit committee. These measures should increase

506 JOURNAL OF ACCOUNTING, AUDITING & FINANCE

auditors’ independence and thus result in higher-quality audit reports. In addition,

companies must provide in their annual reports an assessment of the effective-

ness of internal controls for financial reporting (Section 404), and chief executive

and financial officers (CEOs/CFOs) must certify under oath annual and quarterly

reports (Section 302) and are subject to significant penalties for false certification

(Section 906). These measures should further deter management from fraudulent

financial reporting. Given these expected effects of SOX on auditors and man-

agement, we predict that accrual-based earnings management, in general, and

ability to meet or beat analysts’ earnings expectations, in particular, to decline in

the Post-SOX period.1

With respect to changes in real earnings management, our prediction is not

straightforward. It might be argued that real earnings management to meet or

beat analyst earnings expectations is not expected to decline in the Post-SOX pe-

riod because auditors are unlikely to question this type of activity. For example,

according to CFOs interviewed by Graham, Harvey, and Rajgopal ([2005], 36),

auditors ‘‘cannot readily challenge real economic actions to meet earnings targets

that are taken in ordinary course of action.’’ Should we expect an increase in real

earnings management in the Post-SOX period? The answer to this question is

largely empirical as two conflicting forces may be at work, and it is not clear

which one dominates. On the one hand, the substitution effect (e.g., Cohen, Dey,

and Lys [2008]) suggests that real earnings management will increase to compen-

sate for the decreases in expectations management and accrual management. On

the other hand, according to Zang (2006), managers employ earnings manage-

ment strategies sequentially, that is, they select real manipulation before resorting

to accrual manipulation. This raises the possibility that real manipulation oppor-

tunities might have been exhausted in the Pre-SOX period, and thus no increase

in that activity should be expected in the Post-SOX period notwithstanding the

decreases in earnings and expectations management.

We find that the frequency of just meeting or beating analyst earnings

expectations diminished in the Post-SOX period. Consistent with the evidence in

Cohen, Dey, and Lys (2008) and Lobo and Zhou (2006), we document that

accrual-based earnings management has increased over time before the passage

of SOX and decreased significantly thereafter. More important, we find what has

not been documented before. First, the propensity to manage analyst earnings

expectations to meet or beat their earnings forecasts, which has increased over

time before SOX, has significantly decreased in the Post-SOX period. Second,

we demonstrate that this new finding, together with the findings of declined

accrual-based earnings management and increased real earnings management,

1. Two points to note: (1) Statement on Auditing Standard No. 99 guides auditors to consider the frequency a firm meets earnings expectations when evaluating the risks of fraudulent financial reporting; (2) throughout the paper we use the term Post-Scandal period and Post-SOX period inter- changeably.

507THE ‘‘NUMBERS GAME’’ IN THE PRE- AND POST-SARBANES-OXLEY ERAS

explain the drop in the tendency to meet or beat analyst earnings expectations in

the Post-SOX period.

Our findings make two contributions to the literature. First, they contribute

to the extant academic literature on the earnings expectations and guidance

game. Our results suggest that expectations management as well as the propen-

sity of just meeting or beating analysts’ expectations have both decreased signifi-

cantly in the Post-SOX period. Second and more important, we are the first to

provide an explanation for these findings. Specifically, by being the first to

simultaneously consider all three mechanisms used to just meet or beat analysts’

expectations while controlling for varying economic activities, we show that the

drop in the frequency of just meeting or beating is related to a blend shift in the

three mechanisms used to meet or beat this important earnings benchmark:

accrual-based earnings management and expectations management have both sig-

nificantly declined, whereas real earnings management activities increased. One

implication of this finding is that in the Post-SOX period, investors and other

capital market participants should pay more attention to real earnings manage-

ment activities used to meet certain earnings targets than in the Pre-SOX period.

Section 2 surveys extant literature on mechanisms used to meet or beat ana-

lyst earnings expectations and contrasts it with our work. Section 3 outlines the

sample selection procedure, defines the variables, and describes the data. Section 4

outlines the empirical tests and reports the results. Section 5 summarizes our main

findings and states our conclusions.

2. Background and Motivation

Earnings management to influence accounting appearances and meet or beat

certain benchmarks has been drawing substantial attention from accounting

researchers for decades. While early studies have sought to document the exis-

tence of real earnings management in specific settings (e.g., Hand [1989]; Bartov

[1993]) recent research provides large sample evidence on real earnings manage-

ment activities (e.g., Gunny [2006]; Roychowdhury [2006]; Zang [2006]). In

addition, while accrual-based earnings management has received significant

attention in the literature, more recent studies analyze a new tool used to meet or

beat analysts’ earnings forecasts, an important benchmark: earnings expectations

management (e.g., Bartov, Givoly, and Hayn [2002]; Matsumoto [2002]). Taken

as a whole, these studies demonstrate that companies use all three mechanisms

to meet certain earnings benchmarks.

Recent studies also investigate a possible substitution effect between

accrual-based earnings management and real earnings management. Zang (2006)

examines whether managers use real and accrual manipulations in managing

earnings as substitutes, and also the sequence in which these mechanisms are

employed. She documents a substitution effect between the two and suggests that

managers employ these strategies sequentially—that is, they select real earnings

manipulation before resorting to accrual manipulation. Cohen, Dey, and Lys (2008),

508 JOURNAL OF ACCOUNTING, AUDITING & FINANCE

using annual data, document that accrual-based earnings management increased

steadily in the Pre-SOX period, but decreased significantly thereafter, whereas real

earnings management declined in the Pre-SOX period and increased significantly in

the Post-SOX period. They interpret their results as suggesting that firms switched

from accrual to real earnings management in the Post-SOX period, a behavior

consistent with the substitution effect found by Zang (2006).

In a recent survey, Graham, Harvey, and Rajgopal (2005), report ‘‘strong

evidence’’ that managers engage in real management activities to meet account-

ing targets. In particular, 80 percent of survey participants report that they would

decrease discretionary spending on research and development (R&D), advertis-

ing, and maintenance to meet an earnings benchmark. More than half state that

they would delay starting a new project to meet an earnings target, even if such

a delay entailed a small sacrifice in value. They observe that ‘‘the aftermath of

accounting scandals at Enron and WorldCom and the certification requirements

imposed by the Sarbanes–Oxley Act may have changed managers’ preferences

for the mix between taking accounting versus real actions to manage earnings,’’

but provide little empirical evidence to support this statement (Graham, Harvey,

and Rajgopal [2005], 36).

In this paper, we address two related research questions left unanswered by

prior studies.2 First, we test for a change in the prevalence of expectations man-

agement to meet or beat analysts’ earnings expectations between the Pre- and

Post-SOX periods (our first hypothesis). Academic research (e.g., Lobo and Zhou

[2006]; Cohen, Dey, and Lys [2008]) as well as the popular press has argued that

it became more costly for managers to engage in earnings management activities

in the Post-SOX period because of increased regulatory and auditing scrutiny as

well as more stringent enforcement for securities regulation violations. Given that

earnings management has become more costly in the Post-SOX era, a substitu-

tion effect will lead to an increase in expectations management to meet or beat

analysts’ forecasts. However, an important assumption underlying this prediction

is that managers’ aspirations and thus efforts to meet or beat analysts’ expecta-

tions have not declined between the Pre- and Post-SOX periods. If management

efforts to meet or beat expectations have declined in the Post-Sox period, the

overall frequency of meeting or beating is likely to decline as well, and thus,

ceteris paribus, the prevalence of expectations management will decrease (our

first hypothesis). By carefully measuring expectations management and by divid-

ing our sample period into four subperiods, we document that the frequency

of just meeting or beating analysts’ expectations decreased significantly in the

2. By focusing on meeting or beating an important quarterly benchmark, that is, analysts’ earn- ings forecasts, our study differs from Cohen, Dey and Lys (2008) who examine only the substitution between accrual-based earnings management and real earnings management around SOX. Adding to this stream of the literature, we consider an additional tool used to meet or beat quarterly analysts’ earnings forecasts by explicitly addressing the use of expectations management. As such, our paper is the first to simultaneously consider the three tools used to meet or beat quarterly analysts’ earnings forecasts before and after the passage of SOX.

509THE ‘‘NUMBERS GAME’’ IN THE PRE- AND POST-SARBANES-OXLEY ERAS

Post-SOX period and that expectations management decreased rather than

increased in the Post-SOX period.3

Second and more important, we test for a relation between the decline in the

observed frequency of just meeting or beating expectations in the Post-SOX pe-

riod and a shift in the mix among the three mechanisms used to just meet or beat

analysts’ expectations: accrual-based earnings management, real earnings man-

agement, and earnings expectations management. We hypothesize that the

decline in the frequency of just meeting or beating is related to decreases in

accrual-based earnings management and expectations management, while real

earnings management may have remained unchanged or increased between the

Pre- and Post-SOX periods (our second hypothesis). By being the first to con-

sider all three mechanisms simultaneously while controlling for changes in eco-

nomic activities, we are able to explain the decline in the tendency of just

meeting or beating analysts’ expectations observed in the Post-SOX period.

3. Data

3.1 Sample Selection

We obtain data from the Compustat quarterly files and the Institutional Bro- kerage Estimate System (I/B/E/S) detail files. To be included in our sample, a

firm-quarter observation must first satisfy the following three criteria:4 (1) at

least two individual earnings forecasts (not necessarily by the same individual

analyst) must exist for the quarter, which are at least twenty trading days apart;

(2) the release date of the first earnings forecast we use occurs at least three trad-

ing days after the release of the previous quarter’s earnings report; and (3) the

release date of the second earnings forecast we use precedes the release of the

current quarter’s earnings report by at least three trading days.

The first criterion ensures that there is an initial forecast and a subsequent

forecast revision. These forecasts are required to be separated in time by at least

twenty days so that the second forecast is more likely to represent a true revision

reflecting news arriving to the market after the initial forecast was issued. The

purpose of the second criterion is to prevent ‘‘stale’’ forecasts (i.e., those that are

not revised following the previous quarter’s earnings announcement) from being

included in the analysis. The third criterion ensures that the latest forecast is not

‘‘contaminated’’ by knowledge of the actual earnings number. The total number

3. In a recent concurrent study, Koh, Matsumoto, and Rajgopal (2008) find that the stock mar- ket premium to meeting or just beating analyst expectations has disappeared in the Post-Sox period and that managers rely less on income-increasing discretionary accruals and more on earnings guid- ance to meet or beat analysts’ forecasts in the Post-SOX period. However, Koh, Matsumoto, and Raj- gopal (2008) do not explore the role of real earnings management activities to meet or beat analysts’ earnings forecasts in the Pre-and Post-SOX periods.

4. These selection criteria are consistent with previous research (e.g., Bartov, Givoly, and Hayn 2002).

510 JOURNAL OF ACCOUNTING, AUDITING & FINANCE

of firm-quarter observations meeting the above three criteria is 262,754, which

span the period from January 1987 to December 2006, and represent 10,874 dis-

tinct firms. We refer to this sample as the I/B/E/S Sample.

Tests in the second part of our analysis concerning accruals earnings man-

agement and real earnings management require financial statement information

in addition to the I/B/E/S data. For these tests we impose on our sample firms

three more restrictions: (1) the required financial statement information is avail-

able on the quarterly Compustat database; (2) the firm does not belong to one of

the following three industries: financial institutions (Standard Industry Code [SIC] codes 6000-6999), utilities (SIC codes 4800-4999), or other regulated

industries (SIC codes 4000-4499); and (3) the quarterly earnings surprise relative

to the latest analyst earnings forecast is non-negative.

Similar to previous studies, we imposed the second and third criteria, respec-

tively, because the empirical models we use to estimate accruals and real earn-

ings management do not apply to firms in these industries and because the

second set of tests focuses on firms that meet or beat analysts’ earnings forecasts.

The intersection of these criteria yields a second sample (the Merged Sample) of

87,697 firm-quarter observations, representing 6,186 distinct firms.

3.2 Variable Definitions

To measure the revision in the analyst earnings forecasts, REV, we identify the

first forecast and the last forecast for the quarter. The earliest earnings forecast for

the quarter, Fearliest, is the first forecast for the quarter made subsequent to the

announcement of the previous quarter’s earnings.5 The latest forecast for the quar-

ter, Flatest, is the last forecast for the quarter made before the release of the earnings

announcement for that quarter. REV is the difference between the latest earnings

forecast and the earliest earnings forecast. The earnings surprise for the quarter,

SURP, is defined as the difference between the actual earnings per share number

and the latest forecast for the quarter, EPS � Flatest, both taken from I/B/E/S. Earn-

ings forecast error for the quarter is the difference between the actual earnings per

share number and the earliest forecast for the quarter, EPS � Fearliest. To avoid

classification errors, we used the unadjusted (for stock dividends and split) analyst

earnings forecast per share and unadjusted reported earnings per share to compute

the earnings surprise and earnings forecast error, both taken from I/B/E/S.

Meeting or beating analyst earnings expectations is defined as a zero or posi-

tive (non-negative) earnings surprise, which is the difference between the actual

earnings and the latest forecast for the quarter, SURP ¼ EPS � Flatest � 0. Just

meeting or beating analyst earnings expectations are firm-quarters observations

for which the earnings surprise exceeds analysts’ expectations by a cent per share

or less—that is, $0.00 � SURP � $0.01.

5. We did not consider earnings forecasts for the current quarter made before the release of the previous quarter’s earnings report because their subsequent revision is likely to be correlated with the content of this report rather than with new information about the current quarter’s results.

511THE ‘‘NUMBERS GAME’’ IN THE PRE- AND POST-SARBANES-OXLEY ERAS

3.3 Accrual Management Proxy

We use a cross-sectional model of discretionary accruals, where for each

quarter we estimate the model for every industry classified by its two-digit SIC

code. We estimate the model cross-sectionally to control for industry-wide

changes in economic conditions that affect total accruals and to allow the coeffi-

cients to vary across time (DeFond and Jiambalvo [1994]; Kasznik [1999]).

Our primary model is the modified cross-sectional Jones model (Jones

[1991]), as described in Dechow, Sloan, and Sweeney (1995), applied for quar-

terly data. The modified Jones model is estimated for each two digit SIC-quarter

grouping as follows:

Where:

We use current cash flows from operations, excluding extraordinary items

and discontinued operations (CFO), to calculate accruals. The industry-quarter

specific parameters obtained from eq. (1) are used to estimate firm-quarter spe-

cific nondiscretionary accruals (NDA) as a percent of lagged total assets, adjust-

ing for the change in receivables, DARjq (Dechow, Sloan, and Sweeney [1995]):

Our measure of discretionary accruals, DA, is the difference between .

3.4 Real Earnings Management Proxies

We build on prior studies (Gunny [2006]; Roychowdhury [2006]; Zang

[2006]) to develop our proxies for real earnings management. We consider three

metrics: the abnormal levels of cash flow from operations (CFO), production

costs, and selling, general, and administrative (SG&A) expenses as measures of

real activities manipulation tools.6

TAjq total accruals, defined as earnings minus cash flow for firm j in quarter q;

Assetjq total assets for firm j in quarter q; DSalesjq change in sales for firm j in quarter q; DARjq change in accounts receivables for firm j in quarter q; PPEjq gross property, plant, and equipment for firm j in quarter q;

6. Roychowdhury (2006) uses overall discretionary expenses at the annual level as one of the three proxies for real activities manipulations. Since we are using quarterly data, we cannot use Com- pustat to construct this aggregate measure thus we focus on SG&A expenses that are available on a quarterly basis.

512 JOURNAL OF ACCOUNTING, AUDITING & FINANCE

Firms can accelerate the timing of sales through increased price discounts or

by providing more lenient credit terms. These activities will temporarily increase

sales volumes, which are likely to disappear once the firm reverts to the old pri-

ces. Assuming that the margins are positive, these additional sales will boost cur-

rent period reported earnings. However, for a given sales volume, both price

discounts and more lenient credit terms will result in lower cash flows from

operations in the current period. We first generate the normal levels of CFO using the model developed by Dechow, Kothari, and Watts (1998) as imple-

mented in Roychowdhury (2006) for quarterly data. We express normal CFO as

a linear function of sales and change in sales. To estimate this model, we run the

following cross-sectional regression for each industry and quarter:

Abnormal CFO (A_CFO) is actual CFO minus the normal level of CFO calcu-

lated using the estimated coefficients from eq. (3).

Our second proxy for real earnings management is abnormal production

costs. Managers can increase production more than necessary to increase earn-

ings. When managers produce more units, they can spread the fixed overhead

costs over a larger number of units, thus lowering fixed costs per unit. As long

as the reduction in fixed costs per unit is not offset by any increase in marginal

cost per unit, total cost per unit declines. This decreases reported COGS and the

firm can report higher operating margins. However, the firm will still incur other

production and holding costs that will lead to higher annual production costs rel-

ative to sales, and lower cash flows from operations given sales levels.

Production costs are defined as the sum of cost of goods sold (COGS) and

change in inventory during the quarter. We estimate the normal level of produc-

tion costs as follows:

Abnormal level of production costs (A_PROD) is defined as the residual from

the above cross-sectional regression eq. (4) estimated quarterly for each two-digit

SIC code.

The third empirical proxy for real earnings management is abnormal SG&A

expenses (A_SGA), defined as the residual from the following cross-sectional

regression estimated quarterly for each two-digit SIC code:

Reducing such expenses will boost current period earnings. It could also lead

to higher current period cash flows (at the risk of lower future cash flows) if the

firm generally paid for such expenses in cash.

513THE ‘‘NUMBERS GAME’’ IN THE PRE- AND POST-SARBANES-OXLEY ERAS

To capture the total effects of real earnings management, we combine the

above three individual tools to compute two comprehensive measures of real

earnings management activities. For our first measure, RM1, consistent with

Zang (2006), we multiply abnormal SG&A expenses by negative one (so that the

higher amount, the more likely it is that the firm is cutting SG&A expenses) and

add it to abnormal production costs.7 The higher the amount of this aggregate

measure, the more likely it is that the firm engaged in real earnings management

activities. For the second measure, RM2, again consistent with Zang (2006), we

first multiply abnormal cash flows from operations and abnormal SG&A

expenses by negative one and then aggregate them into one measure. As for

RM2, we multiply the corresponding components by negative one, so that the

higher these amounts, the more likely it is that the firm is engaging in sales

manipulations and cutting discretionary expenditures to manage reported earnings

upward.

We acknowledge that the three individual variables underlying RM1 and

RM2 may have different implications for reported earnings, which may dilute

any results using these aggregated measures. We thus report results correspond-

ing to both the aggregate measures as well as the three individual real earnings

management proxies (A_CFO, A_PROD, and A_SGA).

3.5 Descriptive Statistics

Table 1 presents descriptive statistics for the I/B/E/S Sample (Panel A) and

for the Merged Sample (Panel B). Similar to findings in previous studies, the

results in Panel A show that our sample firms are more likely to deliver a posi-

tive earnings surprise than a negative one. Specifically, while 64.7 percent of

firm-quarters meet or beat analysts’ earnings expectations, only 35.2 percent miss

expectations. In addition, firms are more likely to exhibit a negative forecast

error (41.3% of firm-quarters) than a negative earnings surprise (35.2%). Such

difference is an indication of earnings expectations management, as it is likely

achieved by walking-down expectations. The observed negative forecast revision

(mean ¼ �0.118) is a further indication of earnings expectations management:

in the absence of such activity, the average revision is expected to be zero, not

negative. The results also show that the mean, median, 25 percentile, and 75 per-

centile of firm-size (market capitalization) are, respectively (in millions),

$2,728.27, $367.28, $128.35, and $1,396.37. This indicates that our I/B/E/S Sam-

ple contains a wide range of firm sizes.

Like the results in Panel A, the results in Panel B show that the Merged

Sample is also well diversified in terms of firm-size, and that the two samples

7. We do not multiply A_PROD by negative one because higher production costs, as noted ear- lier, are indicative of overproduction to reduce cost of goods sold. We do not combine abnormal pro- duction costs and abnormal CFO (A_CFO), because in Roychowdhury (2006) the same activities that lead to abnormally high production costs also lead to abnormally low CFO; thus, adding these two amounts leads to double counting.

514 JOURNAL OF ACCOUNTING, AUDITING & FINANCE

TABLE 1

Summary Statistics of Sample Firms

Panel A: Summary statistics, I/B/E/S sample

N Mean Std. Dev. 25% Median 75%

NEG_SURP 262,754 0.352 0.474 0 0.449 1

NEG_FE 262,754 0.413 0.471 0 0 1.000

REV 262,754 �0.118 34.121 �0.023 0 0.006

MBE 262,754 0.647 0.423 0 1.000 1.000

JUSTBEAT 262,754 0.218 0.404 0 0 0

MKTVL 262,754 2728.27 21754.36 128.35 367.28 1396.37

Panel B: Summary statistics, merged sample (I/B/E/S and Compustat)

N Mean Std. Dev. 25% Median 75%

MBE 87,697 0.726 0.328 0 1.000 1.000

JUSTBEAT 87,697 0.273 0.382 0 0 0

ACCRUALS 87,697 �0.011 0.063 �0.026 �0.013 0.006

DA 87,697 0.002 0.043 �0.007 0.005 0.016

ABS_DA 87,697 0.019 0.031 0.005 0.011 0.025

C_INV 87,697 0.004 0.033 �0.003 0 0.013

MKTVL 87,697 2917.37 15967.24 137.69 374.67 1397.67

A_SGA 87,697 0.067 0.059 0.036 0.058 0.117

Variable Definitions:

NEG_SURP is a dummy variable that takes the value of one if earnings surprise for the quarter is negative,

and zero otherwise.

NEG_FE is a dummy variable that takes the value of one if forecast error is negative, and zero otherwise.

REV is forecast revision defined as the difference between the last earnings forecast and the first earnings

forecast for the quarter, Flatest� Fearliest.

MBE is a dummy variable that takes the value of one if the firm meets and/or beats analysts’ expectations

(SURP� 0), and zero otherwise.

JUSTBEAT is a dummy variable that takes the value of one if the firm beats analysts’ expectations by a cent

per share or less ($0.00� EPS� Flatest� $0.01), and zero otherwise.

MKTVL is the market value of equity calculated as the share price times the number of shares outstanding.

ACCRUALS are defined as the difference between income before extraordinary items and cash flows from

operations, adjusted for extraordinary items and discontinued operations.

DA is discretionary accruals, calculated using the modified Jones model.

ABS_DA is the absolute value of discretionary accruals.

C_INV is the change in inventory, scaled by lagged total assets.

A_SGA are SG&A expenses scaled by lagged total assets.

are quite similar in terms of this variable. The results also demonstrate that the

variables underlying the estimation of our earnings management proxies, discre-

tionary accruals, change in inventory, and SG&A—all scaled by lagged total

assets—posses well-behaved properties. That is, their distributions are symmet-

ric around the mean and an outlier observations problem is not a serious

problem.

4. Tests and Results

4.1 Subperiod Analyzed

We partition our sample period into four subperiods (see Figure 1). The first

subperiod analyzed is the Early Pre-Accounting Scandal period extending from

the beginning of the sample period (January 1987) through the end of 1993. The

second subperiod is the Late Pre-Accounting Scandal period from the beginning

of 1994 through the second quarter of 2001. The third subperiod is the Scandal

period, from the beginning of third quarter of 2001 through the second quarter of

2002, and the fourth subperiod is the Post-SOX (Post-Scandal) period, from the

beginning of the third quarter of 2002 through the end of our sample period

(December 2006).

We partition the Pre-Scandal period into two periods, the Early and Late

Pre-Scandal periods, because findings in prior research (e.g., Bartov, Givoly, and

Hayn [2002]; Brown and Caylor [2005]), as well as anecdotal evidence, suggest

that the use of analysts’ estimates as a benchmark for firm performance and the

prevalence of the ‘‘expectations game’’ both increased substantially in the mid-

1990s.8 For the purpose of testing our two hypotheses, the two subperiods of in-

terest are thus the Late Pre-Scandal period and the Post-SOX period.

FIGURE 1

Timeline of Sample Subperiods Analyzed

Note: The sample period begins in January 1987 and ends in December 2006.

8. Several sources began providing earnings benchmarks based on analysts’ forecasts on the Internet only in the mid-1990s. Perhaps the best known, First Call, introduced its service to the Web in 1994.

516 JOURNAL OF ACCOUNTING, AUDITING & FINANCE

4.2 Changes in Frequency of Just Meeting or Beating Analyst

Earnings Expectations

Before testing our two hypotheses, we examine changes in the frequency of

just meeting/beating analysts’ earnings expectations between the Late Pre-Scandal

period and the Post-SOX period. Based on arguments advanced recently (e.g.,

Jensen, Murphy, and Wruck [2004]; Graham, Harvey, and Rajgopal [2005]), we

expect the frequency of just meeting or beating analysts’ expectation to decline in

the Post-SOX period. Figure 2 plots the percentage of firms meeting or beating

analysts’ earnings forecasts by a cent per share or less over the period spanning

the first quarter of 1987 throughout the fourth quarter of 2006. The figure suggests

a decline in the percentage of firms that just meet or beat analysts’ earnings fore-

casts, especially since the early 2000s.

Next, we use both univariate and regression analysis to statistically test for

temporal changes in the frequency of just meeting or beating analysts’ earnings

forecasts. The univariate tests compare the quarterly frequency of firms that meet

or just beat analysts’ expectations across our four sample periods. The results in

Panel A of Table 2 show that the frequency of just meeting or beating increased

FIGURE 2

Percentage of Firms Meeting or Beating Analysts’ Forecasts by

a Cent per Share or Less over Time

Note: The figure represents the percentage of firms just meeting or beating analysts’ earnings

forecasts over time. The figure plots the frequency of JUSTBEAT, a dummy variable that takes the

value of one if the firm-quarter observation beats/meets analysts’ expectations by a cent per share or

less ($0.00 � EPS � Flatest � $0.01).

517THE ‘‘NUMBERS GAME’’ IN THE PRE- AND POST-SARBANES-OXLEY ERAS

between the Early and Late Pre-Scandal periods, from 18.12 percent to 26.81

percent, and that this increase of 8.69 percent is significant at the 1 percent sig-

nificance level. This result is consistent with findings in Bartov, Givoly, and

Hayn (2002) and further highlights that the overall Pre-Scandal period (1987–

2001) is not homogeneous and thus must be disaggregated into the two subper-

iods. Turning to our prediction, the frequency of just meeting or beating analysts’

earnings expectations substantially declines between the Late Pre-Scandal period

and the Post-SOX (Post-Scandal) period, from 26.81 percent to 17.23 percent,

and this decline of 9.58 percent is significant at the 1 percent significance level.

The regression tests for changes in the frequency of just meeting or beating

analysts’ earnings expectations between the four sample subperiods involve esti-

mating the following model:

Where:

FREQt, the dependent variable, is the frequency of firms just meeting or beating

analysts’ earnings expectations in quarter t; PRE94t is a dummy variable that

takes the value of one if quarter t is before the first quarter of 1994 and zero oth-

erwise; SCANt is a dummy variable that takes the value of one if quarter t falls

within the second quarter of 2001 and the second quarter of 2002, and zero oth-

erwise; POSTt is a dummy variable that takes the value of one if quarter t is after

the third quarter of 2002, and zero otherwise.

In terms of eq. (6), the intercept, b0, measures the frequency of just meeting or

beating analysts’ earnings expectations in the Late Pre-Scandal period, and the slope

coefficients, b1, b2, and b3, measure the difference in frequency between the Late

Pre-Scandal period and the Early Pre-Scandal period, and the Scandal period and

the Post-SOX period, respectively. We expect a decline in the frequency between the

Late Pre-Scandal period and the Post-SOX period, that is, b3 < 0. We estimate eq. (6)

over our full sample period, the nineteen years spanning from January 1987 through

December 2006, and thus use eighty quarterly observations. The results displayed in

Panel B of Table 2 are similar to those of the univariate tests reported in Panel A.

Specifically, as predicted, the coefficient on POST is negative, �0.096, and highly

significant (t-statistic ¼ �8.52) indicating a decline in the frequency of just meeting

or beating analysts’ earnings expectations in the Post-SOX period relative to the Late

Pre-Scandal period. Like the univariate results, the regression results also show an

increase in the frequency of just meeting or beating analysts’ expectations between

the Early and Late Pre-Scandal periods indicated by a significantly negative coeffi-

cient (�0.086) on PRE94. Overall, the results in Table 2 are consistent with the pre-

diction that in the Post-SOX (Post-Scandal) period the propensity of just meeting or

beating analysts’ expectations declined relative to the Late Pre-Scandal period.

4.3 Changes in Expectations Management

Our second set of tests for a decline in the prevalence of earnings expecta-

tions management between the Pre- and Post-SOX periods (our first hypothesis)

518 JOURNAL OF ACCOUNTING, AUDITING & FINANCE

TABLE 2

Temporal Analysis of Frequency of Firms Meeting or Beating Analysts’

Expectations by One Cent per Share or Less

Panel A: Univariate analysis of frequencies

1987:Q1–

1993:Q4

(%)

(A)

1994:Q1–

2001:Q1

(%)

(B)

2001:Q2–

2002:Q2

(%)

(C)

2002:Q3–

2006:Q4

(%)

(D)

Frequency of Firms Just

Meet or Beat

18.12 26.81 25.22 17.23

Differences

(B) � (A) 8.69%*

(C) � (B) �1.59%*

(D) � (C) �7.99%*

(D) � (B) �9.58%*

Panel B: Regression analysis of the frequencies

FREQt ¼ b0 þ b1 PRE94t þ b2 SCANt þ b3 POSTt þ et

Variable

Estimate

(t-statistic)

Intercept 0.268

(53.07)

PRE94 �0.086

(�13.08)

SCAN �0.018

(�4.56)

POST �0.096

(�8.52)

N (Quarters) 80

Adj. R2 0.89

*Significant at the 1 percent level, using the test of proportions.

Variable Definitions:

FREQ is the frequency of firms beating analysts’ expectations by a cent per share or less, that is, $0.00 � EPS� Flatest� $0.01

PRE94 is a dummy variable that takes the value of one if the observation is before the first quarter of 1994,

and zero otherwise.

SCAN is a dummy variable that takes the value of one if the observation falls within the third quarter of

2001 through the second quarter of 2002, and zero otherwise.

POST is a dummy variable that takes the value of one if the observation is after the end of the second quarter

of 2002, and zero otherwise.

concerns examining the role interim analyst forecast revisions plays in affecting

the sign of the end-of-quarter earnings surprise. To increase power, we restrict

the analysis to a subsample consisting of firm-quarters that are most likely or

least likely to be affected by expectations management. Specifically, we compare

the observed sign of an earnings surprise with the sign of the earnings surprise

that would have resulted in the absence of an interim forecast revision. In the ab-

sence of an interim revision, the sign of the quarterly earnings surprise would be

the same as the sign of the quarterly forecast error. Observing a negative forecast

error that turns into a positive earnings surprise is thus consistent with expecta-

tions management (walking-down expectations), because it must result from an

excessive downward forecast revision. Likewise, a zero or positive forecast error

that turns into a negative earnings surprise (because of an excessive upward fore-

cast revision) is inconsistent with expectations management. In the absence of

management intervention, the proportion of observations in which the interim

forecast revision offsets the sign of the earnings surprise should be identical

between cases with negative errors and cases with positive errors.

Tables 3, 4, and 5 display the results from tests for a change in earnings expecta-

tions management in the Post-SOX period. Consider the results in Table 3 first. The

TABLE 3

Relative Frequency of Negative Forecast Errors and

Negative Earnings Surprises

Percentage of

Negative Earnings

Surprises (%)

Percentage of

Negative Forecast

Errors (%)

Excess of Negative

Earnings Errors over

Negative Surprise

Cases (%)

(A) (B) (C) ¼ (B) � (A)

All years 35.24 41.35 6.11*

By Subperiods

1987:Q1–1993:Q4 (1) 48.11 51.63 3.52*

1994:Q1–2001:Q2 (2) 32.36 39.87 7.51*

2001:Q3–2002:Q2 (3) 28.64 37.06 8.42*

2002:Q3–2006:Q4 (4) 31.85 36.84 4.99*

Differences

(2) � (1) �15.75* �11.76* 3.99*

(3) � (2) �3.72* �2.81* 0.91*

(4) � (3) 3.21* �0.22 �3.43*

(4) � (2) �0.51 �3.03* �3.54*

Note: The sample consists of 267,547 firm-quarter observations for 1987–2006. Earnings surprise is

the difference between the actual earnings and the latest forecast for the quarter, EPS� Flatest. Forecast error

is the difference between the actual earnings and the earliest forecast for the quarter, EPS� Fearliest. * and ** are significant at the 1 percent and 5 percent levels, using the test of proportions, respectively.

520 JOURNAL OF ACCOUNTING, AUDITING & FINANCE

percentage of negative earnings surprises over the entire sample, 35.24 percent, is sig-

nificantly smaller at the 1 percent significance level than the percentage of negative

forecast errors, 41.35 percent. This result is consistent with expectations management

during the whole sample period whereby analyst earnings forecasts are dampened

during the quarter so as to increase the likelihood of a positive earnings surprise.

Examining the change in the frequency of negative earnings surprises across

our sample subperiods, we note a monotonic increase in the excess of negative

earnings errors over negative earnings surprises in the first three subperiods:

from 3.52 percent in the Early Pre-Scandal period, to 7.51 percent in the Late

Pre-Scandal period, and to 8.42 percent in the Scandal period. However, this

trend reverses in the Post-Scandal period, where the excess percentage of nega-

tive forecast errors over the percentage of negative earnings surprises percentage

declines, not increases, to 4.99 percent from 8.42 percent in the Scandal period.

TABLE 4

Expectation Management: Frequency of Selected

Expectations Paths, by Period

Cases Likely to

be Affected by

Expectations

Management (%)

Cases Less Likely

to be Affected by

Expectations

Management (%)

Difference in

Proportions (%)

All years 29.87 11.23 18.64*

By Subperiod

1987:Q1–1993:Q4 (1) 22.41 13.72 8.69*

1994:Q1–2001:Q2 (2) 37.03 8.50 28.53*

2001:Q3–2002:Q2 (3) 42.07 8.16 33.91*

2002:Q3–2006:Q4 (4) 30.54 11.09 19.45*

Differences

(2) � (1) 14.62* �5.22* 19.84*

(3) � (2) 5.04* �0.34 5.38*

(4) � (3) �11.53* 2.93* �14.46*

(4) � (2) �6.49* 2.59* �9.08

(4) � (1 þ 2) 0.82* 0.02 0.80*

Note: The sample consists of 267,547 firm-quarter observations for 1987–2006. Cases likely to be

affected by expectation management are cases in which the forecast revision turns a negative forecast error

into a positive or zero earnings surprise, scaled by all cases with a negative forecast error. Cases less likely

to be affected by expectation management are cases in which the forecast revision turns a positive or zero

forecast error into a negative earnings surprise, scaled by all cases with a positive or zero-forecast error. The

forecast revision is the difference between the latest forecast and the earlier forecast for the quarter, Flatest� Fearliest. The earnings surprise is the difference between the actual earnings and the latest forecast for the pe-

riod, EPS – Flatest. The forecast error is the difference between the actual earnings and the earliest forecast

for the quarter, EPS� Fearliest. * and ** are significant at the 1 percent and 5 percent levels, using the test of proportions, respectively.

521THE ‘‘NUMBERS GAME’’ IN THE PRE- AND POST-SARBANES-OXLEY ERAS

Tests for statistical significance show that the differences in the excess of nega-

tive earnings errors over negative earnings surprises between each two consecu-

tive subperiods (3.99%, 0.91%, and �3.43%), as well as between the Post-

Scandal period and the Late Pre-Scandal period (�3.54%), are all highly signifi-

cant (significance level better than 1%). This observed pattern in the excess of

negative earnings errors over negative earnings surprises throughout the sample

period is consistent with the hypothesis that earnings expectations management

has become less prevalent in the Post-SOX period.

The results in Table 4 corroborate our inference of declined expectations man-

agement in the Post-SOX period. In this table, we determine the proportion of

TABLE 5

Temporal Analysis of Expectations Management

Model: EXP_M_It ¼ a0 þ b1 PRE94t þ b2 SCANt þ b3 POSTt þ et

EXP_M_1 EXP_M_2 EXP_M_3

Intercept 0.151

(35.62)

0.353

(43.25

0.268

(58.64)

PRE94 �0.024

(�4.72)

�0.134

(�12.49)

�0.052

(�9.32)

SCAN 0.022

(2.04)

0.059

(3.18)

0.049

(5.03)

POST �0.031

(�8.35)

�0.042

(�4.65)

�0.028

(�3.65)

N (Quarters) 80 80 80

Adj. R2 0.63 0.88 0.84

Note: t-statistics are reported in parentheses.

Variable Definitions: EXP_M_1 is the percentage of firm-quarters with a zero or positive earnings surprise and a negative

forecast error, relative to total number of quarterly observations.

EXP_M_2 is the percentage of firm-quarters with a zero or positive earnings surprise and a negative

forecast error, relative to total number of quarterly observations with negative forecast errors.

EXP_M_3 is the percentage of firm-quarters with a zero or positive earnings surprise and a negative

forecast revision, relative to total number of quarterly observations.

Earnings surprise is the difference between the actual earnings and the latest forecast for the quarter,

EPS� Flatest.

Forecast error is the difference between the actual earnings and the earliest forecast for the quarter,

EPS� Fearliest.

Forecast revision is the difference between the last earnings forecast and the first earnings forecast for

the quarter, Flatest� Fearliest.

PRE94 is a dummy variable that takes the value of one if the observation is before the first quarter of

1994, and zero otherwise.

SCAN is a dummy variable that takes the value of one if the observation falls within the period, third

quarter of 2001 through the second quarter of 2002, and zero otherwise.

POST is a dummy variable that takes the value of one if the observation is after the end of the second

quarter of 2002, and zero otherwise.

522 JOURNAL OF ACCOUNTING, AUDITING & FINANCE

firm-quarters with a negative forecast error that ends with a positive or zero

earnings surprise, and the proportion of firm-quarter observations with a posi-

tive or zero forecast error that ends with a negative earnings surprise. Observa-

tions that belong to the first group are more likely to result from expectations

management than those in the second group. To test for a decline in expecta-

tions management, we examine the difference between these two proportions.

Similar to the pattern observed in Table 3, the difference in proportions

increases in the first three subperiods and decreases in the fourth. More specifi-

cally, in the Late Pre-Scandal period, 37.03 percent of the firm-quarters with a

negative forecast error ended with a positive earnings surprise (as a result of an

excessive downward revision in earnings forecasts). In contrast, only 8.5 per-

cent of observations with a positive or zero forecast error ended with a negative

earnings surprise (because of an excessively positive forecast revision that

‘‘spoiled’’ what otherwise would have been a positive earnings surprise). The

statistically significant difference between the two of 28.53 percent, shown in

the rightmost column, suggests the presence of expectations management in the

Late Pre-Scandal period. In the Post-SOX period forecast revisions are also

more likely to turn a negative forecast error into a positive or zero surprise than

to turn a positive or zero forecast error into a negative earnings surprise

(30.54% compared with 11.09%). However, the difference between the two is

only 19.45 percent, lower by 9.08 percent than the 28.53 percent difference

observed for the Late Pre-Scandal period. The last line of the rightmost column shows

that this 9.08 percent decline is statistically significant at the 1 percent level. Thus, sim-

ilar to the results displayed in Table 3, the results in Table 4 also suggest a lower

propensity to manage analysts’ expectations in the Post-SOX period.

An important point emerges from analyzing the evidence in Table 4. Recall

that we divide the Pre-Scandal period into two subperiods and compare the Post-

SOX period to the latter rather than to the whole Pre-Scandal period. We made

this choice since, as discussed above, the Early and Late Pre-Scandal periods are

inherently different (Bartov, Givoly, and Hayn [2002]; Brown and Caylor [2005]).

To see this, note that the percentage of cases likely to be affected by expectations

management has declined between the Late Pre-Scandal period and the Post-SOX

period, from 37.03 percent to 30.54 percent, consistent with a decline in expecta-

tions management. However, this percentage has increased, not decreased, between

the Early Pre-Scandal period (22.41%) and the Post-SOX period (30.54%). Simi-

larly, the percentage of cases less likely to be affected by expectations manage-

ment has decreased from 13.72 to 11.09 percent, not increased, between the Early

Pre-Scandal Period and the Post-SOX period. Given these differences between the

Early and Late Pre-Scandal periods, combining the two together and then compar-

ing them to the Post-SOX period should lead to the inference that expectations

management has increased in the Post-SOX period, rather than decreased. This

intuition is confirmed by the numbers displayed in the last two lines of the right-

most column. Although the difference in the proportions is significantly negative

(�9.08%) when the Late Pre-Scandal period is compared with the Post-SOX

period, it is significantly positive (0.80%) when the Pre-Scandal period as a whole

523THE ‘‘NUMBERS GAME’’ IN THE PRE- AND POST-SARBANES-OXLEY ERAS

is compared with the Post-SOX period.9 This analysis highlights the importance of

dividing the Pre-Scandal period into two subperiods to avoid contamination by

low expectations management frequency in the Early Pre-Scandal period, during

which time the use of analyst estimates as a benchmark for firm performance and

the ‘‘expectations game’’ were both at their infancy.

To further test for a decline in expectations management, we estimate the

following regression model:

Where:

EXP_M_Jt, the dependent variable (J ¼ 1, 2, or 3), is the proportion of firm-quar-

ters likely to be affected by expectations management in quarter t; PRE94t is a

dummy variable that takes the value of one if quarter t falls before the first quarter

of 1994 (i.e., within the Early Pre-Scandal period), and zero otherwise; SCANt is a

dummy variable that takes the value of one if quarter t falls within the second

quarter of 2001 through the second quarter of 2002 (i.e., within the Scandal

period), and zero otherwise; POSTt is a dummy variable that takes the value of one

if the quarter t is after the third quarter of 2002 (i.e., within the Post-SOX period).

In terms of eq. (7), the intercept, a0, measures the proportion of firm-quarters

likely to be affected by expectations management in the Late Pre-Scandal period,

and the slop coefficients, b1, b2, and b3, measure the difference in proportion

between the Late Pre-Scandal period and the Early Pre-Scandal period, and

between the Scandal period and the Post-SOX period, respectively. The hypothesis

of a decline in expectations management in the Post-SOX period relative to the

Late Pre-Scandal period implies: b3 < 0.

The regression results are reported in Table 5. Note that the dependent vari-

able is measured in three alternative ways. For consistency across tables,

EXP_M_2 is similar to our definition of the percentage of cases likely to be

affected by expectation management in the previous table (see Table 4). In addi-

tion, we consider two alternative measures, EXP_M_1, where the deflator is the

total number of quarterly observations, rather than total number of quarterly

observations with negative forecast errors, and EXP_M_3, which is defined as

the percentage of firm-quarters with a zero or positive earnings surprise and a

negative forecast revision, relative to total number of quarterly observations. We

estimate eq. (7) throughout the full sample period that spans the nineteen-year

period, January 1987 through December 2006, and thus use eighty quarterly

observations. The results in Table 5 reinforce the results from the univariate

9. The overall evidence reported in Table 4 helps reconcile our findings with the results docu- mented in Koh, Matsumoto, and Rajgopal (2008) who find that expectations management increased rather than decreased in the Post-SOX period. The difference in findings follows from our research design choice to divide the Pre-Scandal period into two subperiods and to compare the Post-SOX period with the latter period rather than to the whole Pre-Scandal period as Koh, Matsumoto, and Rajgopal (2008) do.

524 JOURNAL OF ACCOUNTING, AUDITING & FINANCE

results in Tables 3 and 4. As hypothesized, b3, the coefficient on POST, is signif-

icantly negative for all three specifications of the dependent variable.

In summary, the results in Tables 3, 4, and 5 are all consistent with earnings

expectations being managed so as to result in positive earnings surprises in both

the Late Pre-Scandal period and the Post-SOX period. In particular, downward

revisions are encouraged when, in their absence, the earnings surprise is expected

to be negative, while upward revisions are discouraged if they might lead to a neg-

ative earnings surprise. More important, the results in all three tables are consistent

with our first hypothesis of a significant decline in expectations management in

the Post-SOX period relative to the Late Pre-Scandal period.

4.4 Changes in the Blend among the Three Mechanisms to Just Meet or

Beat Analysts’ Expectations

What may explain the overall observed decline in the frequency to just meet

or beat analysts’ expectations? Our second hypothesis predicts that the decline in

the frequency of just meeting or beating analysts’ expectations mirrors a mix

shift among the mechanisms used to meet or just beat analysts’ earnings expecta-

tions. In the next section, we explicitly test this hypothesis by simultaneously

considering three mechanisms: accrual-based earnings management, expectations

management, and real earnings management.

We first provide a univariate analysis on the various tools used to just meet or

beat analysts’ expectations by a cent per share or less (i.e., JUSTBEAT). Table 6

reports the distribution of firms that use accrual-based earnings management,

expectations management, and real earnings management to just meet or beat ana-

lysts’ expectations. The proportion of firm-quarters that JUSTBEAT is more likely

to have been driven by accrual-based earnings management are identified by sub-

tracting the amount of unexpected accruals (calculated using the modified Jones

model) from the reported earnings and then recomputing the earnings surprise

(SURP) for all these cases to determine whether or not they remain designated as

JUSTBEAT after adjusting for unexpected accruals. Similarly, we identify JUST- BEAT firm-quarter observations that are more likely to represent expectations man-

agement, which we identify as cases with a negative forecast error that end with a

zero or positive earnings surprise. Finally, we identify the proportion of firm-quar-

ters that JUSTBEAT is more likely to have been driven by real earnings manage-

ment activities. We adjust the reported earnings of all JUSTBEAT cases by the

value of A_CFO, A_PROD, or A_SGA and then recompute the earnings surprise

(SURP) for all these cases to determine whether or not they still remain designated

as JUSTBEAT after adjusting for unexpected real earnings management activities.10

10. We use either one of the real earnings management proxies to address the possibility that a firm might be using different real earnings management strategies to meet or beat analysts’ expecta- tions. For example, if a firm remains designated as a JUSTBEAT observation after adjusting for A_PROD but not using A_SGA (and vice versa), we still classify this observation as more likely to be driven by real earnings management activities.

525THE ‘‘NUMBERS GAME’’ IN THE PRE- AND POST-SARBANES-OXLEY ERAS

The evidence in Table 6 suggests that the proportion of firm-quarter observations

that rely more on accrual-based earnings management to just meet or beat analyst

expectations has declined to 38.68 percent in the Post-SOX period compared with

46.34 percent in the Late Pre-Scandals period (w2 = 67.36, p-value < 0.001). In

addition, we document that 6.53 percent of firms rely on expectations management

to just meet or beat analyst expectation in the Post-SOX period, which is lower than

the proportion of firms in the Late Pre-Scandal period (7.64%, w2 ¼ 38.96, p-value

< 0.001). Finally, our results suggest that there is an increase in the proportion of

TABLE 6

Univariate Analysis of Firms that Just Meet or Beat Analysts’ Expectations

1987:Q1–

1993:Q4 (%)

Early

Pre-Scandals

1994:Q1–

2001:Q1 (%)

Late

Pre-Scandals

2001:Q2–

2002:Q2 (%)

Scandals

2002:Q3–

2006:Q4 (%)

Post-SOX

Proportion of Firms Relying on

Accrual-Based Earnings

Management

53.38% 46.34% 41.27% 38.68%

w2 Statistic versus Post-SOX 87.32

(<0.001)

67.36

(<0.001)

42.68

(<0.001)

Proportion of Firms Relying on

Expectations Management

4.29% 7.64% 8.73% 6.53%

w2 Statistic versus Post-SOX 96.24

(<0.001)

38.96

(<0.001)

15.39

(<0.001)

Proportion of Firms Relying on

Real Earnings Management

15.38% 15.79% 16.14% 21.16%

w2 Statistic versus Post-SOX 42.93

(<0.001)

53.67

(<0.001)

65.91

(<0.001)

Note: For JUSTBEAT firms, that is, firm-quarter observations that beat or meet analysts’ expectations

by a cent per share or less ($0.00� EPS � Flatest� $0.01), we report the proportion of firms that are more

likely to use accrual-based earnings management, expectations managements, or real earnings management

activities.

The proportion of firm-quarters that JUSTBEAT is more likely to have been driven by accrual-based

earnings management are identified by subtracting the amount of unexpected accruals (calculated using the

modified Jones model) from the reported earnings and then recomputing the earnings surprise (SURP) for

all these cases to determine whether or not they still remain designated as JUSTBEAT after adjusting for

unexpected accruals.

The proportion of firm-quarters that JUSTBEAT is more likely to represent expectations management

are identified as cases with a negative forecast error that end with a zero or positive earnings surprise.

The proportion of firm-quarters that JUSTBEAT is more likely to have been driven by real earnings

management activities are identified by adjusting the reported earnings of all JUSTBEAT cases by the value

of A_CFO, A_PROD, or A_SGA and then recomputing the earnings surprise (SURP) for all these cases to

determine whether or not they still remain designated as JUSTBEAT after adjusting for unexpected real

earnings management activities.

526 JOURNAL OF ACCOUNTING, AUDITING & FINANCE

firms that use real earnings management activities in the Post-SOX period (21.16%)

to just meet/beat analysts’ expectations compared to the Late Pre-Scandal period

(15.79%) and that this finding is statistically significant (w2 ¼ 53.67, p-value <

0.001).

Next, we perform a multivariate analysis to test for any change in the mix

of strategies used to just meet or beat analyst expectations and estimate the fol-

lowing logit model:

Where:

JUSTBEAT, the dependent variable, is a binary variable taking the value of one

if the firm-quarter observation beats or meets analyst earnings expectations by

a cent per share or less, and zero otherwise; PRE94t is a dummy variable that

takes the value of one if quarter t falls before the first quarter of 1994 (i.e.,

within the Early Pre-Scandal period), and zero otherwise; SCANt is a dummy

variable that takes the value of one if quarter t falls within the second quarter

of 2001 through the second quarter of 2002 (i.e., within the Scandal period),

and zero otherwise; POSTt is a dummy variable that takes the value of one if

the quarter t is after the end of the second quarter of 2002 (i.e., in the Post-

SOX period); DA is discretionary accruals calculated using the modified Jones

model; EXP_M is a dummy variable taking the value of one if earnings surprise

for the quarter is zero or positive and analyst earnings forecast revision is nega-

tive, and zero otherwise, where earnings surprise is the difference between the

actual earnings number and the latest earnings forecast for the quarter, and

forecast revision is the difference between the last earnings forecast and the

first earnings forecast for the quarter; REAL_EM is a proxy for real earnings

management and is either: A_CFO, A_PROD, A_SGA, RM1, or RM2 where

A_CFO are abnormal cash flow from operations, A_PROD are abnormal pro-

duction costs, A_SGA are abnormal SG&A expenses, and RM1 and RM2 are

the aggregate proxies for real earnings management activities. Dummy varia-

bles for fiscal quarters Q1, Q2, and Q3 are included (not tabulated) in each of

the estimated models.

527THE ‘‘NUMBERS GAME’’ IN THE PRE- AND POST-SARBANES-OXLEY ERAS

In testing for a mix shift among the three mechanisms used to meet or just beat

analysts’ earnings expectations between the Late Pre-Scandal period and Post-SOX

period, the variables of interests in terms of eq. (8) are as follows: b16, the coefficient

on DA*POST, b19, the coefficient on EXP_M*POST, and b22, the coefficient on

REAL_EM*POST. Our second hypothesis predicts: b16 < 0, b19 < 0, and b22� 0.

4.5 Control Variables

We include numerous control variables in eq. (8) consistent with prior

research (e.g., Matsumoto [2002]; Skinner and Sloan [2002]): varying macroeco-

nomic conditions, firm size, growth opportunities, unexpected shocks to earnings,

and the uncertainty in the forecasting environment. As proxies for unexpected

macroeconomic shocks, we include GDP and IND_ROA, which may affect earn-

ings management and expectations management strategies. GDP is the percent-

age change in seasonally adjusted Gross Domestic Products over the previous

quarter (a proxy of overall economic activity); and IND_ROAjq is the average

return on assets of firm i’s two-digit industry (a proxy for industry-specific eco-

nomic activity), computed after excluding the return on assets of firm i. Guenther

and Young (2000) provide evidence of a high association between ROA and eco-

nomic growth rate, indicating that ROA reflects real economic activity in a

timely manner. We exclude the firm in calculating the average industry ROA to

avoid any mechanical associations among the variables in the regression. We

define P_UE, a dummy variable that takes the value of one where we observe a

positive seasonal change in quarterly earnings before extraordinary items. This

variable controls for the association between the change in earnings and the

probability of just meeting or beating the analyst forecast.

To control for any systematic variation in the frequency of just meeting or beat-

ing analysts’ expectations with growth opportunities and firm size we include SIZE and M_B, where SIZE is the natural logarithm of the market value of equity at the be-

ginning of the quarter and M_B is the ratio of market value of equity to book value of

equity. We control for uncertainty in the forecasting environment by including the

absolute value of the forecast error, ABS_FE, where the forecast error is defined as

before, that is, the difference between the actual earnings per share number and the

earliest forecast for the quarter, EPS—Fearliest. Finally, we include the number of

analysts following the firm in a given quarter, ANALYST, as prior research (e.g.,

Kasznik and McNichols [2002]) suggests that this variable might be associated with

the probability of meeting or beating analysts’ forecasts.

Table 7 documents the estimation results for the logistic model specified in

eq. (8). Since a firm can enter our sample numerous times, there might be some

within-firm autocorrelation of the errors terms. To address this concern, we esti-

mate eq. (8) using a generalized linear model to correct for time dependence

errors within firms based on Liang and Zeger (1986).11 Reading across Table 7,

11. Our conclusions remain the same once we use standard logistic regressions.

528 JOURNAL OF ACCOUNTING, AUDITING & FINANCE

we note three salient points. First, the results reported in all five columns are

similar, indicating they are robust to the proxy used for real earnings manage-

ment. Second, the coefficients on DA (b11), EXP_M (b12), and REAL_EM (b13)

are all significant in the predicted direction with the exception of the coefficient

estimate on abnormal cash flows from operations (column 1, p-value of 0.139).

This suggests that in the Late Pre-Scandal period all three mechanisms—accrual-

based earnings management, expectations management, and real earnings man-

agement—were used to just meet or beat analysts’ earnings expectations. This

implies that these tools to just meet or beat analysts’ forecasts are used as com-

plements. Third, as predicted, the coefficient b16 (DA*POST), ranging from

�1.124 to �1.243 depending on the proxy used for real earnings management,

and the coefficient b19 (EXP_M*POST), ranging from �0.156 to �0.173, are

both statistically significantly negative, and b22 (REAL_EM*POST) is significant

for increasing abnormal production costs and cutting SG&A expenses, as well as

for the two aggregate real earnings management proxies. These results suggest

that consistent with our second hypothesis the drop in the frequency of just meet-

ing or beating in the Post-SOX period relative to the preceding seven-year period

is associated with an overall decline in the use of earnings expectations manage-

ment and accrual-based earnings management, whereas real earnings manage-

ment increased as manifested in higher abnormal production costs and decreasing

SG&A expenses.12

Overall, the reported evidence is consistent with a substitution between

accrual-based, real earnings management activities and downward earnings guid-

ance used to meet or beat an important quarterly benchmark such as analysts’

forecasts in the Post-SOX period.

4.6 Robustness Tests

In this section, we assess the reliability of our findings by considering two

types of sensitivity tests. First, a criticism of discretionary accruals models is

their classification of nondiscretionary accruals as discretionary. To address this

concern, we assess the sensitivity of our findings in Table 7 after computing dis-

cretionary accruals using two alternative models. First, previous research has

shown that measures of unexpected accruals are more likely to be incorrectly

specified for firms with extreme levels of performance. In particular, Dechow,

Sloan, and Sweeney (1995) and Kasznik (1999) document that estimated discre-

tionary accruals are negative for firms with low earnings and positive for firms

with high earnings. To address this concern, we adjust the modified Jones model

12. Although we find abnormal cash flows from operations to be negatively associated with the probability of just meeting or beating analysts’ forecasts, the coefficient estimate is not statistically significant at conventional levels. One explanation might be that some real activities manipulations, such as channel stuffing, might be captured by discretionary accruals. When we run a specification in which we exclude discretionary accruals as an explanatory variable and include only abnormal cash flows as an independent variable, we find it to load significantly in the predicted direction consistent with this explanation.

529THE ‘‘NUMBERS GAME’’ IN THE PRE- AND POST-SARBANES-OXLEY ERAS

TABLE 7

Logit Analysis of Firms that Just Meet or Beat Analysts’ Expectations

Variable A_CFO A_PROD A_SGA RM1 RM2

Intercept 0.276

(<0.0001)

0.279

(<0.0001)

0.288

(<0.0001)

0.251

(<0.0001)

0.243

(<0.0001)

PRE94 �0.093

(<0.0001)

�0.096

(<0.0001)

�0.097

(0.019)

�0.092

(<0.0001)

�0.089

(0.008)

SCAN �0.006

(0.154)

�0.008

(0.132)

�0.009

(0.131)

�0.006

(0.124)

�0.005

(0.121)

POST �0.019

(0.039)

�0.018

(0.027)

�0.011

(0.021)

�0.017

(0.026)

�0.016

(0.039)

GDP 0.037

(<0.0001)

0.042

(<0.0001)

0.039

(<0.0001)

0.036

(<0.0001)

0.041

(<0.0001)

IND_ROA 0.127

(<0.0001)

0.134

(<0.0001)

0.142

(<0.098)

0.124

(<0.0001)

0.129

(<0.0001)

SIZE 0.041

(<0.0001)

0.038

(<0.0001)

0.042

(<0.0001)

0.029

(<0.0001)

0.034

(<0.0001)

M_B 0.012

(<0.0001)

0.014

(<0.0001)

0.011

(<0.0001)

0.012

(<0.0001)

0.013

(<0.0001)

P_UE 1.241

(<0.0001)

1.184

(<0.0001)

1.226

(<0.0001)

1.176

(<0.0001)

1.217

(<0.0001)

ABS_FE �18.351

(<0.0001)

�12.354

(<0.0001)

�14.324

(<0.0001)

�13.674

(<0.0001)

�14.329

(<0.0001)

ANALYST 0.607

(<0.0001)

0.591

(<0.0001)

0.619

(<0.098)

0.584

(<0.0001)

0.596

(<0.0001)

DA 2.109

(<0.0001)

2.206

(<0.0001)

2.131

(<0.0001)

2.224

(<0.0001)

2.154

(<0.0001)

EXP_M 0.239

(<0.0001)

0.252

(<0.0001)

0.249

(<0.0001)

0.281

(<0.0001)

0.267

(<0.0001)

REAL_EM �0.072

(0.139)

0.423

(0.003)

�0.129

(0.081)

0.213

(0.073)

0.174

(0.049)

DA*PRE94 1.452

(<0.0001)

1.568

(<0.0001)

1.436

(<0.0001)

1.549

(<0.0001)

1.517

(<0.0001)

DA*SCAN �0.804

(<0.0001)

�0.824

(0.021)

�0.842

(0.008)

�0.851

(0.041)

�0.829

(0.004)

DA*POST �1.124

(0.013)

�1.213

(0.012)

�1.235

(0.004)

�1.209

(0.021)

�1.243

(0.005)

EXP_M*PRE94 �0.134

(<0.0001)

�0.141

(<0.0001)

�0.146

(<0.0001)

�0.152

(<0.0001)

�0.155

(<0.0001)

EXP_M*SCAN 0.221

(0.006)

0.216

(0.001)

0.214

(0.006)

0.212

(0.003)

0.216

(0.004)

EXP_M*POST �0.156

(0.003)

-0.162

(0.007)

�0.168

(0.003)

�0.173

(0.004)

�0.163

(0.006)

REAL_EM*PRE94 0.017

(0.334)

0.024

(0.524)

�0.031

(0.236)

0.039

(0.227)

0.032

(0.353)

REAL_EM*SCAN 0.014

(0.328)

0.037

(0.241)

�0.114

(0.460)

0.034

(0.181)

0.074

(0.243)

REAL_EM*POST �0.114

(0.113)

0.249

(0.038)

�0.381

(0.046)

0.192

(0.021)

0.114

(0.036)

No. of

Observations

87,697 87,697 82,274 82,274 82,274

Log-Likelihood

Ratio

2371.58 2463.24 2247.96 2687.72 2703.34

Note: The logistic model is estimated using the generalized linear model method to correct for within-

firm time dependence. P-values are reported in parentheses.

JUSTBEAT, the dependent variable is a binary variable taking the value of one if the firm-quarter ob-

servation beats or /meets analysts’ expectations by a cent per share or less ($0.00� EPS-Flatest� $0.01).

PRE94 is a dummy variable that takes the value of one if the observation is before the first quarter of

1994, and zero otherwise.

SCAN is a dummy variable that takes the value of one if the observation falls within the third quarter

of 2001 through the second quarter of 2002, and zero otherwise.

POST is a dummy variable that takes the value of one if the observation is after the end of the second

quarter of 2002, and zero otherwise.

GDP is percentage change in seasonally adjusted GDP over the previous quarter.

IND_ROA is the industry average ROA for the quarter, calculated for each two-digit SIC code.

SIZE is the natural logarithm of the market value of equity at the beginning of the quarter.

M_B is the ratio of market value of equity to book value of equity.

P_UE is a dummy variable that takes the value of one if there is a positive seasonal change in quarterly

earnings before extraordinary items.

ABS_FE is the absolute value of the forecast error, where the forecast error is defined as the difference

between the actual earnings per share number and the earliest forecast for the quarter, EPS� Fearliest.

ANALYST is the number of analysts following the firm in a given quarter.

DA is defined as discretionary accruals calculated using the modified Jones-model.

EXP_M is a dummy variable taking the value of one if earnings surprise for the quarter is zero or posi-

tive and analysts’ forecast revision is negative.

Earnings surprise is the difference between the actual earnings and the latest forecast for the quarter,

EPS� Flatest.

Forecast error is the difference between the actual earnings and the earliest forecast for the quarter,

EPS� Fearliest.

Forecast revision is the difference between the last earnings forecast and the first earnings forecast for

the quarter, Flatest� Fearliest.

REAL_EM is a proxy for real earnings management activities and is either: A_CFO, A_PROD, A_SGA, RM1, or RM2.

A_CFO is abnormal cash flow from operations, measured as the deviations from the predicted values

from the corresponding industry-quarter regression:

A_PROD is abnormal production costs, measured as the deviations from the predicted values from the

corresponding industry-quarter regression:

A_SGA is abnormal sales, general and administration expense measured as deviations from the pre-

dicted values from the corresponding industry-quarter regression:

RM1 is an aggregate measure of real earning management activities and is calculated as the sum of

abnormal SG&A expenses multiplied by negative one and abnormal production costs.

RM2 is an aggregate measure of real earnings management activities and is the sum of abnormal cash

flows and abnormal SG&A expenses, both multiplied by negative one.

Dummy variables for fiscal quarters Q1, Q2, Q3 are included (not tabulated) in each of the estimated

models.

TABLE 7 (Continued )

by including a measure of current operating performance—that is, the current

cash flows from operations excluding extraordinary items, as a control variable.

Our discretionary accrual model becomes:

A second alternative builds on the discussion in McNichols (2002), Dechow,

Richardson, and Tuna (2003), and Larcker and Richardson (2003). Since accruals

are changes in working capital accounts, one would expect fast-growing firms to

have larger accruals (McNichols [2002]). In line with this prediction, we include

the book-to-market ratio (BM) as a proxy for expected growth in firm’s opera-

tions. BM is measured as the ratio of the book value of common equity to the

market value of common equity:

The industry-quarter specific parameters obtained from eqs. (9) and (10),

respectively, are used to estimate firm-quarter specific nondiscretionary accruals

as a percent of lagged total assets, as in the first model specified in eq. (1),

which we used throughout the analysis. The results of these sensitivity checks

(not tabulated for parsimony) show that the results in Table 7 are robust to alter-

native measures of discretionary accruals.

As a second sensitivity test, we use within-industry-quarter ranks for the

accrual-based and real earnings management variables to avoid any influence of out-

liers. In untabulated results, we find evidence consistent with the findings reported

in Table 7. This alternative specification echoes our main findings of a substitution

between accrual-based earnings management, real earnings management activities,

and downward earnings guidance used to meet or beat an important quarterly bench-

mark such as analysts’ earnings expectations in the Post-SOX period.

5. Conclusion

In this study, we test for changes in the strategies used to meet or beat ana-

lysts’ earnings forecasts following the major accounting scandals of 2001–2002

and the regulatory reforms introduced by the SOX Act of 2002. First, we find

that the overall frequency of just meeting or beating analysts’ earnings expecta-

tion has declined in the Post-SOX period. With regards to the different tools used

to meet or beat analysts’ expectations, we report new evidence that has not been

documented to date. We observe a statistically significant decline in expectations

management in the Post-SOX period compared with the late 1990s. This suggests

that managers have reduced their reliance on downward guidance as a mecha-

nism to just meet or beat analysts’ earnings expectations.

532 JOURNAL OF ACCOUNTING, AUDITING & FINANCE

Second, we provide an explanation for the observed decline in the tendency

to just meet or beat analysts’ expectations. We acknowledge that managers can

simultaneously use a mix of actions to just meet or beat analysts’ earnings

expectations, namely, accrual-based earnings management, expectations manage-

ment, and real earnings management. In fact, one feature underlying our research

design that distinguishes our work from related studies is our use of an empirical

specification that considers simultaneously these three different mechanisms. Our

results suggest that while all three mechanisms are used to just meet or beat ana-

lysts’ earnings expectations, the decline in the frequency of just meeting or beat-

ing between the Post-SOX period and the Late-Pre-Scandal period is related to a

relative decline in both accruals management and expectations management,

whereas real earnings management seems to have overall increased.

Our findings make two main contributions to the existing literature. First,

they contribute to the extant academic literature on the earnings expectations

game by showing that expectations management as well as the propensity of just

meeting or beating analysts’ expectations has decreased significantly in recent

years, namely, in the Post-SOX period. Second and more important, our results

explain the observed decline in the frequency of just meeting or beating analysts’

expectations in the Post-SOX period. Specifically, by being the first study to

simultaneously consider all three mechanisms used to meet or beat analysts’

earnings expectations while controlling for varying economic activities, we show

that the drop in the frequency of just meeting or beating is related to a blend

shift in the three mechanisms used to meet or beat analysts’ earnings expecta-

tions. One implication of our evidence is that in the Post-SOX period investors

and other capital market participants should pay more attention to real earnings

management activities than they did in the Pre-SOX period.

REFERENCES

Bartov, E. 1993. ‘‘The Timing of Asset Sales and Earnings Manipulation.’’ The Accounting Review 68: 840–855.

Bartov, E., D. Givoly, and C. Hayn. 2002. ‘‘The Rewards to Meeting or Beating Analysts’ Fore- casts.’’ Journal of Accounting and Economics 33: 173–204.

Brown, L. D., and M. L. Caylor. 2005. ‘‘A Temporal Analysis of Quarterly Earnings Thresholds: Pro- pensities and Valuation Consequences.’’ The Accounting Review 80 (April): 423–440.

Cohen, D., A. Dey, and T. Lys. 2008. ‘‘Real and Accrual Based Earnings Management in the Pre and Post Sarbanes Oxley Periods.’’ The Accounting Review 83 (May): 757–787.

Dechow, P. M., R. G. Sloan, and A. P. Sweeney. 1995. ‘‘Detecting Earnings Management.’’ The Accounting Review 70: 193–225.

Dechow, P. M., S. P. Kothari, and R. Watts. 1998. ‘‘The Relation between Earnings and Cash Flows.’’ Journal of Accounting and Economics 25: 133–168.

Dechow, P. M., S. A. Richardson, and A. I. Tuna. 2003. ‘‘Why Are Earnings Kinky? An Examination of the Earnings Management Explanation.’’ Review of Accounting Studies 8: 355–384.

DeFond, M. L., and J. Jiambalvo. 1994. ‘‘Debt Covenant Effects and the Manipulation of Accruals.’’ Journal of Accounting and Economics 17: 145–176.

DeGeorge, F., J. Patel, and R. Zeckhauser. 1999. ‘‘Earnings Management to Exceed Thresholds.’’ Journal of Business 72 (January): 1–33.

Dhaliwal, D. S., C. A. Gleason, and L. F. Mills. 2004. ‘‘Last Chance Earnings Management: Using the Tax Expense to Achieve Earnings Targets.’’ Contemporary Accounting Research 21: 431–459.

533THE ‘‘NUMBERS GAME’’ IN THE PRE- AND POST-SARBANES-OXLEY ERAS

Graham, J. R., C. R. Harvey, and S. Rajgopal. 2005. ‘‘The Economic Implications of Corporate Fi- nancial Reporting.’’ Journal of Accounting and Economics 40: 3–73.

Guenther, D., and D. Young. 2000. ‘‘The Association between Financial Accounting Measures and Real Economic Activity: A Multinational Study.’’ Journal of Accounting and Economics 29: 53–72.

Gunny, K. 2006. ‘‘What are the Consequences of Real Earnings Management?’’ Working Paper, Uni- versity of Colorado at Boulder.

Hand, J. R. 1989. ‘‘Did Firms Undertake Debt-Equity Swaps for an Accounting Paper Profit or True Financial Gain?’’ The Accounting Review 64: 587–623.

Jensen, M. C., K. J. Murphy, and E. G. Wruck. 2004. ‘‘Remuneration: Where We’ve Been, How We Got to Here, What Are the Problems, and How to Fix Them.’’ Harvard Business School NOM Research Paper No. 04-28.

Johnson, N. S. 1999. Remarks on Current SEC Developments Titled, ‘‘Managed Earnings,’’ and ‘‘The Year of the Accountant.’’ Paper presented at the Utah State Bar Mid-Year Convention, St. George, UT, March 6, 2001. http://www.sec.gov/news/speech/speecharchive/1999/spch264.htm (accessed December).

Jones, J. 1991. ‘‘Earnings Management during Import Relief Investigations.’’ Journal of Accounting Research 29: 193–228.

Kasznik, R. 1999. ‘‘On the Association between Voluntary Disclosure and Earnings Management.’’ Journal of Accounting Research 33: 353–367.

Kasznik, R., and M. McNichols. 2002. ‘‘Does Meeting Earnings Expectations Matter? Evidence from Analyst Forecast Revisions and Share prices.’’ Journal of Accounting Research 40: 727–759.

Koh, K., D. Matsumoto, and S. Rajgopal. 2008. ‘‘Meeting Analyst Forecast in the Post-Enron World: Early Evidence on Stock Market Rewards and Managerial Actions.’’ Contemporary Account- ing Research 25: 1067–1098.

Larcker, D. F., and S. A. Richardson. 2003. ‘‘Corporate Governance, Fees for Non-Audit Services and Accrual Choices.’’ Journal of Accounting Research 42: 625–658.

Levitt, A. 1998. ‘‘The Numbers Game.’’ Securities and Exchange Commission. Remarks delivered at the NYU Center for Law and Business, September 28, 1998 http://www.sec.gov/news/speech/ speecharchive/1998/spch220.txt (accessed July).

Liang, K., Zeger, S., 1986. ‘‘Longitudinal Data Analysis Using Generalized Linear Models.’’ Biome- trika 73: 13–22.

Lobo, J. L., and J. Zhou. 2006. ‘‘Did Conservatism in Financial Reporting Increase after the Sar- banes-Oxley Act? Initial Evidence.’’ Accounting Horizons 20: 57–73.

Matsumoto, D. 2002. ‘‘Management Incentives to Avoid Negative Earnings Surprises.’’ The Account- ing Review 77: 483–514.

McGee, S., 1997. ‘‘As Stock Market Surges Ahead, Predictable Profits Are Driving It.’’ The Wall Street Journal, May 5, 1997, C1.

McNichols, M. F. 2002. ‘‘Discussion of the Quality of Accruals and Earnings: The Role of Accrual Estimation Errors.’’ The Accounting Review 77 (Supplement): 61–69.

Roychowdhury, S. 2006. ‘‘Earnings Management through Real Activities Manipulation.’’ Journal of Accounting and Economics 42: 335–370.

Skinner, D., and R. Sloan. 2002. ‘‘Earnings Surprises, Growth Expectations, and Stock Returns or Don’t Let an Earnings Torpedo Sink your Portfolio.’’ Review of Accounting Studies 7: 289– 312.

Zang, A. Z. 2006. ‘‘Evidence on the Tradeoff between Real Manipulation and Accrual Manipula- tion.’’ Working paper, University of Rochester.

534 JOURNAL OF ACCOUNTING, AUDITING & FINANCE

Copyright of Journal of Accounting, Auditing & Finance is the property of Greenwood Publishing and its

content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's

express written permission. However, users may print, download, or email articles for individual use.