The use of advertising activities to meet earnings benchmarks: evidence from monthly data
-
Upload
daniel-cohen -
Category
Documents
-
view
213 -
download
0
Transcript of The use of advertising activities to meet earnings benchmarks: evidence from monthly data
The use of advertising activities to meet earningsbenchmarks: evidence from monthly data
Daniel Cohen Æ Raj Mashruwala Æ Tzachi Zach
Published online: 7 July 2009
� Springer Science+Business Media, LLC 2009
Abstract Using a unique database of monthly media advertising spending, we
examine whether managers engage in real earnings management to meet quarterly
financial reporting benchmarks. We extend prior literature by (1) separately ana-
lyzing advertising activities, allowing us to explore the possibility that managers
could reduce or boost advertising to meet benchmarks; (2) analyzing actual activ-
ities as opposed to inferring them from reported expenses, which are also subject to
accrual choices; (3) investigating the timing, within a quarter, of altered advertising
spending; and (4) examining quarterly earnings benchmarks. We find that managers,
on average, reduce advertising spending to avoid losses and earnings decreases.
However, we also report that firms in the late stages of their life cycle increaseadvertising to meet earnings benchmarks. Finally, we find some evidence that firms
increase advertising in the third month of a fiscal quarter and in the fourth quarter to
beat prior year’s earnings.
Keywords Real earnings management � Advertising � Earnings benchmarks
JEL Classification M41 � M432 � G38
D. Cohen
Stern School of Business, New York University, 44 West 4th Street, New York, NY 10012, USA
e-mail: [email protected]
R. Mashruwala
College of Business Administration, University of Illinois at Chicago, Chicago, IL 60607, USA
e-mail: [email protected]
T. Zach (&)
Fisher College of Business, The Ohio State University, 450 Fisher Hall, 2100 Neil Ave., Columbus,
OH 43210, USA
e-mail: [email protected]
123
Rev Account Stud (2010) 15:808–832
DOI 10.1007/s11142-009-9105-8
1 Introduction
Extensive empirical evidence shows that managers engage in earnings manage-
ment (for reviews, see Healy and Wahlen 1999; Kothari 2001; and Fields et al.
2001). The literature to date identifies two main strategies to manage reported
earnings: (1) accounting method changes and accrual-based earnings management
that do not affect cash flows and (2) real actions that affect cash flows. The early
evidence on earnings management concentrates on accrual-based strategies. More
recently, Graham et al. (2005), in a survey of top executives, report that 80% of
respondents would decrease discretionary spending to achieve financial reporting
objectives. Consistent with the survey, large sample evidence suggests that
managers engage in real earnings management activities to avoid reporting
annual losses (Roychowdhury 2006). These actions have implications for future
firm performance (Gunny 2005) and are viewed as substitutes to accrual-based
strategies (Zang 2006).
In this paper, we address the following research questions: (1) do managers use
advertising to meet certain earnings benchmarks? (2) what are some factors that
determine whether managers increase or decrease advertising expenditures to meet
these benchmarks? and (3) is there a specific time during a fiscal period in which
altering advertising spending to achieve financial reporting goals is more likely to
occur?
Because these questions cannot be answered with data available on
Compustat, we address them by employing a proprietary database of monthly
advertising activities over a six-year period (2001 through 2006). The database
contains information on advertising spending in multiple media channels (for
example, television, radio, newspaper, etc.). It is constructed by a media-tracking
company that records individual advertisements for a large number of media
outlets.
Prior large-sample studies generally aggregate annual advertising expenses
together with other discretionary expenses such as research and development and
maintenance or include them as part of selling and general expenses (for example
Roychowdhury 2006; Zang 2006; Gunny 2005). Our separate analysis of advertising
enables us to contribute to the literature in two important ways.
First, while an implied conclusion from prior evidence is that the documented
aggregate effects can be attributed to each of the components equally, there is no
direct evidence of this. We provide new and direct evidence on whether advertising
is used as a tool for real earnings management activities.
Second, studying advertising separately allows us to focus on its unique costs and
effects on short-term earnings, which may differ from those of other expenditures,
such as research and development (R&D). Compared with R&D, advertising might
have a more immediate impact on sales, leading to the possibility that managers
may increase advertising to generate a positive short-term response in revenues and
earnings. Thus, we investigate factors that determine the direction (increase or
decrease) in which managers alter their advertising spending to attain accounting
benchmarks.
The use of advertising activities to meet earnings benchmarks 809
123
Our third contribution stems from a unique feature of our data, which is based on
the observation of advertising activities. This enables us to directly examine
managerial actions that have immediate financial statement consequences. In
contrast, prior studies of real earnings management rely on income statement
expenses. These studies implicitly assume a one-to-one correspondence between the
observed expenses and the real actions underlying them. These expenses, however,
could be affected by accrual choices, especially in samples of firms that have
incentives to achieve financial reporting objectives. Thus, drawing inferences about
real earnings management from expenses is problematic, because these expenses
may be a function not only of real actions but also of accrual choices. Our analysis is
free of this inferential problem.
Fourth, our monthly data allow us to study the timing of managerial choices
regarding real activities within the fiscal period. This aspect could not be achieved
with Compustat data. Understanding timing is important because it illuminates the
considerations that managers face when altering real activities. Before taking
actions, managers would like to learn about the firm’s expected financial
performance to minimize the magnitude of the costly real action needed to meet
benchmarks. The expected periodic performance becomes clearer as the period
advances. At the same time, however, managers become more limited in their
ability to take actions that would have the intended effects on financial statements
before the fiscal period ends. By examining high-frequency data, we address these
tradeoffs and attempt to identify the month in which real earnings management
takes place.
Finally, using our high-frequency data on advertising we can comment on
managerial decisions about advertising expenses to meet quarterly benchmarks,
which was not possible before. This is important because quarterly benchmarks
have been the focus of the extant literature on meeting financial reporting objectives
(DeGeorge et al. 1999; Brown and Caylor 2005). Survey evidence in Graham et al.
(2005) confirms that managers attend to quarterly performance and benchmark
beating. Furthermore, the relative costs of real versus accrual management vary
across quarters, and therefore studying the quarterly behavior is interesting.
Summary of the results. We find strong evidence that firms reduce their
advertising spending to meet two of the three financial reporting objectives that we
examine: avoiding reporting a quarterly loss and meeting earnings from the same
quarter in the previous year. Also, we find evidence that firms in the late stages of
their life cycle increase advertising spending to meet earnings benchmarks.
Furthermore, we find evidence that firms increase their advertising in the third
month of the fiscal quarter to meet earnings from the same quarter in the previous
year. Finally, we report that suspect firms are more likely to increase their
advertising in the fourth fiscal quarter compared with other quarters.
The paper proceeds as follows. In Sect. 2 we survey the literature on real earnings
management and develop our hypotheses. In Sect. 3 we describe our research
design. Section 4 describes our data. In Sect. 5 we report the results. Section 6
concludes.
810 D. Cohen et al.
123
2 Hypotheses development
2.1 Literature overview
Accounting researchers have contemplated that firms manage earnings by cutting
discretionary expenses to achieve certain earnings benchmarks such as (1) avoiding
negative earnings, (2) avoiding earnings decreases, and (3) meeting or beating
analysts’ forecasts (for example, Burgstahler and Dichev 1997; Roychowdhury
2006). Studies have referred to such activities as ‘‘real’’ earnings management,
which differ from accrual manipulations in that they generally have direct cash flow
consequences. Graham et al. (2005, p. 32) find ‘‘strong evidence that managers take
real economic actions to maintain accounting appearances.’’Roychowdhury (2006) defines real earnings management as ‘‘actions that deviate
from normal business practices, undertaken with the primary objective to mislead
certain stakeholders into believing that earnings benchmarks have been met in the
normal course of operations.’’ Examining annual data, he finds evidence consistent
with firms trying to avoid reporting losses in several ways: (1) boosting sales
through accelerating their timing; (2) generating additional unsustainable sales
through price discounts or more lenient credit terms1; (3) overproducing and thereby
allocating less overhead to cost of goods sold; and (4) aggressively reducing
aggregate discretionary expenses (defined as the sum of research and development,
advertising, and SG&A expenses). This is most likely to occur when such expenses
do not generate immediate revenues and income. In related research, Gunny (2005)
examines the consequences of real earnings management and finds that it has a
significant negative impact on future operating performance. In a different context,
Darrough and Rangan (2005) report a negative association between R&D spending
before initial public offerings (IPO’s) and the number of shares that insiders sell in
the offerings. They conclude that managers reduce R&D before an IPO, to achieve a
higher offering price.
Financial executives indicate a greater willingness to manage earnings through
real activities than through accruals (see surveys by Bruns and Merchant 1990; and
Graham et al. 2005). Roychowdhury (2006) argues that there are at least two
reasons for this preference. First, accrual-based earnings management is more likely
to draw auditor or regulatory scrutiny than real decisions. Second, relying on accrual
manipulation alone is risky. The realized shortfall between unmanaged earnings and
the desired threshold can exceed the amount by which it is possible to manipulate
accruals after the end of the fiscal period. If reported income falls below the
threshold and all accrual-based strategies to meet it are exhausted, managers are left
with no options because real activities cannot be adjusted at or after the end of the
fiscal reporting period. Zang (2006) analyzes the tradeoffs between accrual
manipulations and real earnings management. She suggests that decisions to manage
earnings through real actions precede decisions to manage earnings through
1 Chapman (2008) examines the pricing behavior of soup manufacturers utilizing a unique data set of
supermarket price data. He finds some evidence of meeting earnings benchmarks through the provision of
price promotions during the final month of the quarter.
The use of advertising activities to meet earnings benchmarks 811
123
accruals. Her results show that real manipulation is positively correlated with the
costs of accrual manipulation and that accrual and real manipulations are negatively
correlated. She concludes that managers treat the two strategies as substitutes.
2.2 The effects of advertising on reported earnings
2.2.1 The advertising setting
While evidence in the literature seems consistent with managers altering their
operating decisions to achieve short-term earnings goals, there are at least two
limitations to such an interpretation. First, because prior studies pool all
discretionary expenses together, by adding R&D, advertising, and SG&A expenses
(for example, Roychowdhury 2006), it is not clear which types of discretionary
expenses are subject to change and in which direction. The activities may differ in
their costs of manipulation. For example, altering advertising may be easier and
quicker to execute than changing R&D, because it does not involve adjustment
costs, such as the disposal of assets, laying off workers or both (Anderson et al.
2003). In addition, the activities may have different implications on short-term
earnings. For example, prior studies have assumed that reducing discretionary
expenses maps into a contemporaneous increase in reported earnings, because of the
direct effect of immediate expensing. However, compared with R&D, advertising
might have a more immediate impact on sales, leading to the possibility that
managers may increase advertising to generate a positive short-term response in
revenues and earnings. Thus, the decision to decrease advertising might also depend
on the relation between advertising and sales. When studying aggregate measures of
discretionary expenses, this richness cannot be captured empirically. Therefore, by
separately analyzing advertising spending and their unique costs and effects on
short-term earnings, we minimize the heterogeneity in manipulation costs and
benefits across various discretionary expense items that are pooled in an aggregate
measure.
The second limitation to interpreting existing evidence as supportive of real
earnings management is that it is based on reported expenses in the income
statement. An analysis of expenses assumes a one-to-one mapping between
recorded expenses in the income statement and the real activities underlying them.
This may not necessarily hold if expenses are affected by managers’ accrual
choices. Thus, an observed reduction in expenses may be due to accrual earnings
management rather than real earnings management. By studying reported expenses,
researchers cannot unambiguously draw inferences about real earnings management
because the alternative explanation of accrual earnings management cannot be ruled
out. In contrast, we examine actual advertising activities that are observed in the
marketplace using a database that tracks firms’ advertising at its source. According
to GAAP, advertising spending has to be expensed as incurred, and therefore our
database includes all media-related advertising that should have been expensed in
the period in which it was aired or printed. (For a detailed description of the
accounting treatment of advertising under GAAP, please see Appendix 1). As a
812 D. Cohen et al.
123
result, we comment on real activities and not on reported expenses, which could be
affected by both real activities and accrual manipulations.2
A question that arises is whether managers have the flexibility to engage in real
earnings management using advertising. The firm may be pre-committed to
advertising contracts in various media outlets and may not be able to make any
changes to advertising under these contracts. However, from our conversations with
a sales manager in a television station and with an advertising director and a
marketing vice president in two large corporations, it seems that altering media
spending is certainly feasible. In most media outlets (newspapers, radio,
magazines), there is ample flexibility, and a large portion of the activities are not
contract-based. The most rigid contracts are in television. Normally, the contracts of
television advertising are on an annual basis and contain cancellation clauses for
specific spots that require a four-week notice. However, according to the sales
manager, it is customary to honor cancellation requests even within the four-week
period. In general, in such circumstances the advertising is shifted within the
contract period or even outside of it. In addition, about 30% of the television market
is ‘‘scatter’’ where purchases are made on a spot basis. Moreover, several media
buyers operate an ongoing market through which firms can sell their reserved spots
at the last minute to other firms. Hence, we believe that altering of media-related
advertising activities is possible with respect to the flexibility of contracts between
firms and media outlets.3
2.2.2 Altering advertising activities to affect earnings
Managers who are interested in increasing short-term earnings may choose to either
decrease or increase advertising to achieve their goals. Our maintained assumption
is that before managers alter current advertising to influence financial reporting
outcomes, they were committed to spending an optimal predetermined level of
advertising (denoted by A*), given the prevailing business conditions and
information. This optimal level maximizes the total expected lifetime earnings of
the firm (for example, P1 ? P2, where P1 is current period’s earnings, and P2 is
the sum of earnings in future periods beyond the first period). Given that managers’
objective is to increase current reported earnings (P1) and our maintained
assumption, real earnings management results in a reduction in lifetime earnings
and firm value.4
Consider first the case of reducing advertising to meet financial reporting
benchmarks. For example, Netflix’s CFO said during the 2005 third-quarter
conference call, ‘‘we plan to cap our marketing spending just like we did in Q3 and
2 In the event that the altering of advertising activities does not get reflected in the financial statements in
the current period, then one does not expect to find a relation between these advertising activities and
financial reporting objectives. This correlation is expected to arise when managers alter their activities
and anticipate that their actions would have a direct impact on reported expenses and financial results.3 The feasibility of altering advertising expenditures is also related to the organizational structure of the
firm, the incentives provided to the marketing staff and the degree of control over the staff by the CEO.4 This is consistent with the definitions and interpretations of real earnings management in the accounting
literature, as advanced recently by Graham et al. (2005) and Roychowdhury (2006).
The use of advertising activities to meet earnings benchmarks 813
123
meet our earnings guidance’’ (emphasis added).5 Such a deviation from A* to ADEC
(ADEC \ A*) has two effects. First, the direct effect is a decrease in operating
expenses, which increases current reported earnings. Second, a reduction in
advertising could translate into a decrease in sales in the current period, possibly
leading to a decrease in reported income. Thus, a manager interested in increasing
current reported earnings would reduce advertising only if the decrease in earnings
resulting from declining sales does not offset the increase in earnings from the direct
effect. This will occur if the time lag between advertising and the reaction in sales is
relatively large or if the elasticity of sales to advertising is not high or both. Also,
since the firm was at an optimal level of advertising (A*) and since P1 increased, it
must be that P2 decreased by more than the increase in P1, resulting in lower
lifetime earnings.
Next, consider the case of increasing advertising to meet accounting benchmarks.
The direct effect of increased advertising is a reduction in short-term earnings. Thus,
to achieve a net increase in current earnings, the impact of advertising on sales and
earnings has to offset the direct reduction in earnings. This can be achieved if the
lag between advertising and sales is short or if the elasticity of advertising is high or
both. However, if managers choose a higher, but suboptimal, level of advertising
(AINC, where AINC [ A*), then the increase in current reported earnings resulting
from AINC must be offset by a larger reduction in future earnings. Otherwise, an
increase in current period earnings that is accompanied by an increase in lifetime
earnings is not value decreasing and may be optimal, contradicting the definition of
real earnings management. Because lifetime profits are higher under A* than under
AINC, the increase in current reported earnings will be smaller than the reduction in
future earnings. This can occur if the increase in current earnings, through an
increase in sales, comes at the expense of future sales.
An increase in advertising could also be motivated by meeting or beating
benchmarks that are not based on earnings. For example, Blockbuster significantly
increased its customer acquisition and marketing activities towards the end of the
fourth quarter of 2006 to achieve the milestone target of 2 million subscribers to its
online offering.6 In this case, Blockbuster was concerned with meeting a sales-based
benchmark.
Our above discussion suggests that a manager’s decision to either decrease or
increase advertising spending to boost short-term earnings depends on the relation
5 See the full transcript at http://internet.seekingalpha.com/article/3995.6 ‘‘Blockbuster began offering free in-store rentals to subscribers of online rival Netflix Inc. on Tuesday
as it tries to take market share from Netflix and reach its year-end subscriber goal. Netflix subscribers can
get one free rental at Blockbuster stores for each address label they bring in from their Netflix mailing
envelopes. The offer to Netflix subscribers, which runs until December 21, comes after Blockbuster began
allowing subscribers to its own online rental service to swap DVDs they received in the mail at
Blockbuster stores. The program, called Total Access, highlights the main difference between the two
competitive services: Blockbuster has stores where consumers can immediately satisfy a desire for a
particular movie. Blockbuster forecasted that it will add about 500,000 subscribers in the current quarter,
to reach its year-end goal of 2 million total subscribers.’’ (Financial Wire, December 6, 2006).
814 D. Cohen et al.
123
between advertising and sales. The marketing literature studies this relation along
two dimensions: (1) actual response of sales to advertising (advertising elasticity)
and (2) the lasting effect of advertising on future sales (the carryover effect).7 This
literature proposes several characteristics of firms and products that determine
elasticity and carryover. Hence, we can use these characteristics to distinguish
between firms that are more likely to increase, rather than decrease, advertising
activities to meet financial reporting objectives. For example, we expect firms with
high (low) advertising elasticity to increase (decrease) advertising to meet earnings
targets. In addition, the decision to reduce advertising will only result in an increase
in earnings if the impact on sales can be deferred to future periods. Therefore, firms
with high carryover are more likely to decrease advertising to meet the current
period’s targets.
Differences in elasticities and carryover can be observed across product
categories based on the stage of market development. For instance, elasticities are
higher during the early growth phase of a product market, when many new
customers are added, than during the maturity phase, when most customers already
have substantial experience with the product (Parsons 1975; Arora 1979; Winer
1979; Parker and Gatignon 1996). The carryover effect is also likely to vary with a
firm’s life cycle. In the early stages, a firm uses more awareness advertising, whose
objective is to increase sales in the long run and not necessarily to generate an
immediate impact. When firms mature, their advertising shifts from awareness to
action advertising, whose objective is to increase short-term sales. The combined
effect of elasticity and carryover and its relation to the firm’s life cycle is
ambiguous, because each element predicts a different behavior in the early and late
stages of the life cycle.
To distinguish between firms that are more likely to increase or decrease
advertising to meet financial reporting objectives, we construct a composite variable
to proxy for a firm’s life cycle.8 The proxy is based on the procedures outlined in
Anthony and Ramesh (1992) and is a function of four underlying variables: (1)
growth in sales, (2) dividend rate, (3) capital expenditures, and (4) firm age.
The above discussion leads to our first two hypotheses:
H1a Firms that are suspected to have managed earnings to achieve earnings
benchmarks have abnormal advertising expenses that are different from firms that
are not suspected to have managed earnings.
H1b The direction of alterations in advertising in firms that are suspected to have
managed earnings to achieve earnings benchmarks is expected to vary along the
stages of firms’ life cycles.
7 See Assmus et al. (1984) and Bagwell (2007) for a review of this literature.8 Ideally we would like to empirically capture the life cycle of the product. However, we are limited by
the fact that sample firms have multiple products, possibly in different life cycles. We assume that there is
a positive correlation between a firm’s life cycle and its products’ average life cycle.
The use of advertising activities to meet earnings benchmarks 815
123
2.3 Timing of advertising activities
By using monthly advertising data, we can examine the timing of altering
discretionary expenses.9 When managing earnings using advertising, firms face a
tradeoff between the costs of deviating from an optimal pre-determined advertising
strategy and the benefits of achieving an earnings target. To minimize the deviation
costs, managers must have as much information as possible about the difference
between current firm performance and the desired earnings benchmark. This
information is more complete towards the end of a fiscal period, when expected pre-
managed periodic performance is better gauged. By waiting for a later point in the
period, managers obtain more information about how much they need to alter their
real activities. However, since there may be a lag between the time a decision is
made and when it is executed, managers cannot wait too long, because taking real
actions may no longer be possible.
To address this tradeoff, managers choose the real actions that require the least
amount of lag time for execution. Advertising activities require shorter lag time
between the decision and execution time, compared with, for example, R&D.
Therefore, advertising is an appropriate setting in which to examine the tradeoffs
because managers potentially have sufficient time both to learn about current
performance and to alter advertising before the end of the quarter.
To gauge whether there is any indication that altering advertising occurs later in
the quarter, we propose our second hypothesis:
H2 Firms that are suspected to have managed earnings to achieve earnings
benchmarks will have abnormal advertising expenses (either positive or negative)
during the last month of the quarter.
2.4 The fourth fiscal quarter
The accounting profession has followed an integral approach for interim reporting
as mandated in APB Opinion No. 28 (APB 1973), SFAS No. 3 (FASB 1974), and
FASB Interpretation No. 18 (FASB 1977). The premise of this approach is that the
fiscal year is the main reporting period, while the quarterly financial reports are
integral parts of the annual reporting process. The standards underlying the integral
approach require the reporting firms to estimate many of the annual operating
expenses and then allocate these estimates to interim quarters. Managers can defer
and accrue certain costs depending on their expectations about the operations of the
business for the fiscal year as well as other reporting incentives. In contrast,
quarterly sales revenues can be recognized on the same basis as for the entire fiscal
year period.
The principles of the integral approach will result in less precise interim
estimates of cost accruals and deferrals in the first three fiscal quarters as compared
9 Zang (2006) argues that there is a pecking order in the tradeoff between earnings management through
real activities versus through accrual decisions. Specifically, managers exhaust the options for real
earnings management before resorting to earnings management through accruals. In contrast, in our
analysis we study the timing of real activities, which may occur before any accrual decisions are made.
816 D. Cohen et al.
123
with the fourth quarter. These cost estimates provide managers with an opportunity
to exert greater control over interim reported accounting figures relative to fourth
quarter earnings. Moreover, in the interim quarters, the figures are not necessarily
audited.
The relative costs of accrual management are likely to be higher in the fourth
quarter since managers have less discretion over costs than in interim periods and
due to elevated auditor oversight in the fourth quarter. As a result, we anticipate that
firms would engage in more real earnings management in the fourth quarter. This
leads to our third hypothesis:
H3 Firms that are suspected to have managed earnings to achieve earnings
benchmarks will have abnormal advertising activities during the last quarter of the
fiscal year.
3 Research design
In this section, we describe how we measure (1) the incentives to manage earnings
and meet financial reporting benchmarks and (2) the abnormal advertising activity,
which is the hypothesized real activity used by managers to achieve their financial
reporting objectives.
3.1 Suspect firms
According to the survey by Graham et al. (2005, p. 21), ‘‘Several performance
benchmarks have been proposed in the literature… such as previous years’ or
seasonally lagged quarterly earnings, loss avoidance, or analysts’ consensus
estimates.’’ Following the standard approach in the literature, we define firms that
are suspected to have managed earnings as those that fall in the areas immediately to
the right of zero, in the cross-sectional distribution of (1) earnings before
extraordinary items, (2) changes in earnings relative to the same quarter in the
previous year, and (3) analysts’ forecast errors, defined as actual earnings minus
analysts’ forecasts. The presumption is that, in these areas of the cross-sectional
distribution, there is a higher frequency of firms that have managed earnings,
including firms that managed earnings through real activities (for example,
Burgstahler and Dichev 1997).
Focusing on these specific ‘‘suspect’’ firms enables us to increase the power of
our tests. We classify suspect firms as those (1) whose quarterly earnings before
extraordinary items (scaled by total assets) are between 0 and 0.00125 (labeled
SUSPECT)10; (2) whose seasonal quarterly change in earnings, compared with the
same quarter in the previous year, is between 0 and 0.005 (labeled SUSPECT_Q4);
and (3) that just met the consensus analysts’ forecast by one cent (labeled MBE). For
MBE we use the outstanding consensus forecast in the first month of the quarter
10 Since we use quarterly data we divide the top of the range defined in Roychowdhury (2006) of 0.005
by a scale of 4.
The use of advertising activities to meet earnings benchmarks 817
123
because it is observable at the time managers make the real earnings management
decisions.
We acknowledge that classifying firm-quarter observations into the ‘‘suspect’’
categories has its drawbacks. Since this classification is based on ex post
realizations, we are likely to overlook some firms that managed earnings through
varying their advertising activities but did not fall into the narrow range
immediately to the right of the relevant yardstick.
3.2 Modeling abnormal advertising activities
To capture the ‘‘normal’’ level of advertising spending, two potential approaches
could be used: cross-sectional (industry model) and time-series. An industry model
would examine the average industry spending in a particular month and attribute
any deviation from the average as abnormal advertising. In the specific context of
advertising, this may be problematic because these deviations may reflect the
strategy of a particular firm to intentionally deviate from its competitors, rather than
real earnings management. Furthermore, an industry model assumes uniform
advertising determinants across all firms in the industry. On the other hand, a time-
series model assumes uniform determinants at the firm level by employing the firm
as its own control and collapsing all firm-level determinants into one model. The
model assumes that the effect of changes in these determinants is captured in the
lagged values of the dependent variable (that is, advertising outlays), which are used
as explanatory variables. We believe that the assumption of uniform firm-level
determinants is less restrictive than the assumption of uniform industry-level
determinants. Finally, under the time-series approach, the residuals are less likely to
contain measurement errors that are correlated with omitted variables, unlike the
residuals obtained from industry-based cross-sectional models.
Because of its potential advantages in the advertising context, we employ a firm-
specific time-series model to isolate the abnormal portion of advertising expense
and to evaluate whether managers change advertising spending in response to
accounting-based incentives. In this model, we use up to 72 months of advertising
data. We employ a model utilized by Foster (1977) to characterize the time-series of
quarterly earnings. The quarterly earnings expectation from the Foster model is:
EðQtÞ ¼ h1 � Qt�4 þ h2 � ðQt�1 � Qt�5Þ þ et ð1Þ
where Qt is earnings in quarter t; Qt-4 is an autoregressive term; and (Qt-1 - Qt-5)
is a drift term. The Foster model assumes that earnings follow a first-order auto-
regressive process in seasonal differences. We adapt the Foster model to estimate a
model for the monthly series of advertising outlays, as follows:
E ADStð Þ ¼ h1 � ADSt�12 þ h2 � ADSt�1 � ADSt�13ð Þ þ et ð2Þ
where ADSt is the monthly advertising outlay scaled by the annual sales during the
prior year. We use the residuals from this model as proxies for abnormal advertising
activities and attribute them to real earnings management.
818 D. Cohen et al.
123
4 Data
4.1 Advertising data
We obtain monthly advertising information from a proprietary database constructed
by a large media-tracking company. Our sample covers the years 2001 through
2006. The data are gathered by continuously monitoring multiple media channels
and collecting information about observed advertisements. The media channels
covered include (1) television (cable and network), (2) print (newspapers, local
magazines, consumer magazines and business magazines), (3) radio (local and
national), and (4) the Internet.11 The translation of media spots to monetary amounts
is done by surveying media channels for their rates at various points during the
quarter. For example, for network television, the rates used to estimate advertising
expenditures are supplied primarily by the networks and, in some instances, by
advertising agencies and advertisers. The networks provide average 30-second
program rates, by program title, on a monthly basis after the completion of the
monitored month. Once this information is received, rates specialists check it for
consistency, completeness, and special episode titling.
The database covers all media spending that appears in the tracked channels and
is reported by brand. For example, the media spending by Johnson & Johnson is
reported separately by its brands, which include Band-Aid, Tylenol, etc. We
aggregate the advertising outlays of all brands that belong to a particular firm. We
emphasize that these data cover only media spending and do not include other forms
of advertising, such as the production costs of advertising, promotional materials,
and direct-response advertising (for example, catalogs sent directly to consumers).
4.2 Sample selection
Table 1 reports the steps we took to construct our sample. To make sure that we
include only firms that have substantial advertising budgets, we collect information
about actual advertising activities and media spending information for firms whose
average annual advertising expense over our sample period (2001 through 2006)
exceeds $100,000. We perform this initial screen using the annual advertising
expense figure in Compustat (data#45).12 Of those firms (2,312), we identify 1,562
firms that are covered in our proprietary database. Thus, the database covers about
67% of Compustat firms with annual advertising expenses exceeding $100,000.
The main data requirement is the ability to estimate abnormal advertising
activities using the time-series model described in Eq. (2). This limits our sample to
1,156 firms covering 41,960 monthly observations. Next, we separately construct
11 See Appendix 2 for a detailed description of each media channel.12 We begin the sample construction with an initial screen in Compustat because the search of our
proprietary database requires a manual input of firms and does not allow a restricted search based on the
advertising number that is included in the database. As a result, we have to prepare a list of firms from an
external source (Compustat) that will form our initial sample. A check of the data indicates that, by
employing this cutoff, we include in our initial sample more than 95% of firms with non missing annual
advertising expenses in Compustat.
The use of advertising activities to meet earnings benchmarks 819
123
three samples based on the variable that we use to proxy for financial reporting
incentives. As described in Table 1, after requiring sufficient data on some of the
control variables, our samples cover between 919 and 1,110 firms, and the number
of monthly observations varies from 28,492 to 34,889.
4.3 Summary statistics
Table 2 reports summary statistics for our sample firms. On average, our sample
firms’ revenues are $3.8 billion, with a median of $431 million. As indicated by the
difference between the mean and median, the sample is quite skewed with respect to
size. Similar patterns emerge in total assets (mean of $4.6 billion and median of
$432 million) and market capitalization (mean of $5.6 billion and median of $496
million). The average book-to-market ratio of firms in our sample is 0.74, with a
median of 0.42.
Table 1 Sample construction
Number
of firms
Number of
monthly
observations
Firms with material advertising expense 2,312 –
Firms covered in the proprietary database 1,562 –
Firms and observations with sufficient information to compute abnormal
advertising measures
1,156 41,960
Sample 1
Observations with sufficient information to calculate the SUSPECT variable,
based on quarterly income
1,156 40,761
Observations with sufficient information to calculate INDUSTRY_ADS,
SIZE, INCOME, and BM (after truncating extreme 1%)
1,110 34,889
Sample 2
Observations with sufficient information to calculate the SUSPECT_4Q
variable, based on change in income relative to the same quarter
in the prior year
1,084 36,902
Observations with sufficient information to calculate INDUSTRY_ADS,
SIZE, INCOME, and BM (after truncating extreme 1%)
1,044 32,326
Sample 3
Observations with sufficient information to calculate the MBE variable,
based on analysts’ forecast errors
938 31,700
Observations with sufficient information to calculate INDUSTRY_ADS,
SIZE, INCOME, and BM (after truncating extreme 1%)
919 28,492
This table describes the sample construction procedure. Each of our three samples is based on the same
information on monthly advertising activities. The monthly advertising outlays are drawn from a pro-
prietary database and are for the period 2001 through 2006. The samples differ on the type of financial
reporting benchmark hypothesized to affect managers’ incentives to alter their advertising activities
820 D. Cohen et al.
123
Average annual advertising expenses as reported in Compustat (data#45) for our
sample firms is about $97 million, with a median of $8.8 million. Advertising
expenses constitute about 2.5% of annual sales. Media-related advertising outlays
tracked by the proprietary database are $43.1 million, on average, for the same set of
firms. This suggests that media-related outlays constitute about 45% of advertising
expenses reported in the financial statements. The reason for the difference is that,
as noted earlier, our database tracks only advertising activities related to media
spending. It does not include other forms of advertising, like production costs,
promotional materials, and direct-response advertising (for example, catalogs sent
directly to consumers), all of which are included as part of the advertising expense
in Compustat. The correlation between the annual advertising expenses reported in
Compustat and the annual media-related advertising in our database is 0.85.
5 Results
5.1 Abnormal advertising activities
We first report the results of estimating the time-series model of monthly advertising
outlays as described in Eq. (2). The purpose of this model is to isolate the abnormal
(discretionary) portion of advertising outlays. Table 3 reports the estimation results
of Eq. (2). The table reports summary statistics of the model’s coefficients, h1 and
h2, for all 1,110 firm-specific models. The average number of months used in each
firm-specific model is about 36. We observe that monthly advertising media
spending has a strong relation to the spending in the same month of the prior year.
The average of h1 is about 0.80, and this coefficient is statistically significant in over
85% of the firm-specific models. The average of h2 is 0.26 and is significant in over
half of the models.
Table 2 Descriptive statistics for sample firms
Mean Median SD 25% 75%
Sales (Mil $) 3,763 430.7 14,258 117.9 1,646.7
Assets (Mil $) 4,590 431.8 20,285 122.3 1,724.4
Market capitalization (Mil $) 5,564 495.8 22,571 140.4 1,951.2
BM 0.74 0.42 3.95 0.24 0.70
Advertising expenses (Mil $) 96.7 8.75 347.2 1.7 40.2
Advertising expenses/Sales 3.97% 1.91% 7.97% 0.82% 4.56%
Advertising outlays (Mil $) 43.1 1.60 181.2 0.24 12.5
Advertising outlays/Sales 2.32% 0.46% 8.85% 0.11% 1.76%
This table reports summary statistics for our final sample of 1,110 firms. The numbers reported are the
summary of the firm-average across the six sample years. Sales is defined as the annual figure of total
sales in millions of dollars from Compustat (data#12). Assets is defined as total assets from Compustat
(data#6). BM is the book-to-market ratio. Advertising expenses is the annual advertising expense from
Compustat (data#45). Advertising outlays is the annual sum of all monthly media-related advertising
outlays tracked by our proprietary database
The use of advertising activities to meet earnings benchmarks 821
123
5.2 Testing the hypotheses
To test our hypotheses of managers’ altering advertising activities to attain financial
reporting objectives, we estimate variations of the following regression model using
a panel data set:
ABN ADSt ¼ aþ b1 � INCENTIVE þ b2 � Y þ b3 � INCENTIVE � Yþ Controlsþ et ð3Þ
The dependent variable, ABN_ADSt, is the residual obtained from firm-specific
time-series models of monthly advertising spending described in the previous
section.13 INCENTIVE is an indicator variable that is a function of one of the three
financial reporting benchmarks as defined in Sect. 3.1: SUSPECT, SUSPECT_4Qand MBE.
We interact the incentive variables with several variables (Y’s) that help us test
H1b, H2 and H3. For hypothesis 1b, we use an indicator variable (CYCLE) that is
equal to one if the firm is at a late stage of its life cycle. We determine the stage by
using the algorithm outlined in Anthony and Ramesh (1992). Each year, we rank
firms into three groups based on each of the following variables: (1) dividend payout
ratio, (2) annual sales growth, (3) capital expenditures, and (4) firm age. We then
sum the group numbers to which a firm belongs, resulting in a measure that ranges
from 4 to 12. Firms in the earliest stage of their life cycle receive a score of 4 (that
is, these firms are ranked in the lowest third of dividend payout, the highest third of
sales growth, the highest third of capital expenditures and the lowest third of firm
Table 3 Results of time-series models of monthly advertising expenditures
H1 H2
N 1,110 1,110
Mean 0.7821 0.2640
Standard deviation 0.5774 0.3154
10% 0.2753 -0.0900
25% 0.5604 0.0254
Median 0.7937 0.2563
75% 0.9317 0.4849
90% 1.0189 0.6568
% Significant (5%) 86.6 55.6
Average (median) adjusted R2 27.8%
(27.7%)
This table reports summary statistics of coefficients from estimating a firm-specific time-series model of
monthly advertising expenditures. The average (median) number of observations per firm is 36.2 (40.0).
The model is an adaptation of the Foster (1977) model for quarterly earnings:
E ADStð Þ ¼ h1 � ADSt�12 þ h2 � ADSt�1 � ADSt�13ð Þ þ et
13 While this approach tries to minimize the measurement error in our dependent variable, it is certainly
not free from such error. However, since this measurement error appears in our dependent variable, its
only effect would be manifested in a lower R2. Clearly, this is not likely to bias the inferences we draw
based on the reported results.
822 D. Cohen et al.
123
age). We set CYCLE equal to one if the composite score is greater than or equal to
10. For hypothesis 2, we use an indicator variable (MONTH3) that equals one if the
advertising spending occurred during the last month of a fiscal quarter. For
hypothesis 3, we use an indicator variable (QTR4) that equals one if the advertising
spending occurred during the fourth fiscal quarter.
We control for industry-wide shocks that may have influenced the advertising
levels of a firm in a particular month and that were not captured by the time-series
model. To do so, we include in the regressions the average abnormal advertising of
all firms in the same industry during the same month (based on Fama-French 48
industry groups), excluding the firm being studied (INDUSTRY_ADS). Additional
control variables include SIZE, the natural logarithm of market capitalization at the
beginning of the fiscal year; INCOME, quarterly earnings before extraordinary items
deflated by beginning of the year’s assets; and BM, the book-to-market ratio at the
beginning of the fiscal year.
5.2.1 Hypotheses 1a and 1b
Panels A through C of Table 4 report the results of estimating several versions of Eq.
(3), after truncating 1% of the smallest and largest extreme observations of each
variable. The panels vary by the financial reporting benchmark that is being tested. To
account for potential dependence among the monthly observations, all our statistical
tests are based on clustered standard errors at the firm level (Petersen 2008).
Beginning with Panel A of Table 4, where the financial reporting objective is to
avoid reporting a loss (SUSPECT), we find that in all specifications, the coefficient
on SUSPECT is negative and highly significant (p-values of less than 0.01). This is
consistent with H1a, which suggests that firms manage their advertising spending to
report earnings that just meet the zero earnings benchmark. Moreover, the negative
sign indicates that, on average, suspect firms choose to reduce advertising.14 Recall
that, in the case of advertising spending, firms could boost advertising to increase
sales and earnings.
Specification (3) in Panel A of Table 4 examines hypothesis 1b, which considers
the possibility of managers either decreasing or increasing advertising to achieve
certain earnings benchmarks. This depends on the length of the lead-lag relationship
between advertising and its sales response. We capture these effects using the
variable CYCLE. It is argued that firms in the earlier stages of their life cycle will
tend to have a higher elasticity of advertising-to-sales and a lower carryover effect.
Therefore, our prediction on CYCLE is ambiguous.
The coefficient on CYCLE is negative and significant (p-value = 0.003)
suggesting that, on average, firms in the later stages of their life cycle tend to
have lower abnormal advertising. However, the coefficient on the interaction term
(SUSPECT*CYCLE) is positive and significant (p-value = 0.041), suggesting that
later-stage firms whose aim is to avoid losses are more likely to increase their
14 Since we capture only media-related advertising, and not all advertising activities, our results are also
consistent with a shift from media spending to other advertising activities. However, we believe that such
an interpretation is still consistent with real earnings management, under our maintained assumption that
the ‘‘mix’’ of advertising activities was initially optimal.
The use of advertising activities to meet earnings benchmarks 823
123
Table 4
(1) (2) (3)
Panel A
INTERCEPT 0.0002
(0.000)***
0.0009
(0.000)***
0.0008
(0.000)***
SUSPECT -0.0001
(0.000)***
-0.0001
(0.007)***
-0.0002
(0.003)***
CYCLE -0.0001
(0.003)***
SUSPECT*CYCLE 0.0001
(0.041)**
INDUSTRY_ADS 0.0126
(0.063)*
0.0089
(0.182)
SIZE -0.0001
(0.000)***
-0.0001
(0.000)***
INCOME -0.0001
(0.087)*
-0.0007
(0.277)
BM -0.0002
(0.000)***
-0.0002
(0.000)***
Observations 39,953 34,889 27,713
Adjusted R2 0.0002 0.0132 0.0120
F-statistics (P-values)
SUSPECT = SUSPECT*CYCLE 6.96
(0.0085)***
Panel B
INTERCEPT 0.0003
(0.000)***
0.0009
(0.000)***
0.0008
(0.000)***
SUSPECT_4Q -0.0001
(0.000)***
-0.0001
(0.000)***
-0.0001
(0.000)***
CYCLE -0.0001
(0.008)***
SUSPECT_4Q*CYCLE 0.0001
(0.054)*
INDUSTRY_ADS 0.0129
(0.067)*
0.0089
(0.188)
SIZE -0.0001
(0.000)***
-0.0001
(0.000)***
INCOME -0.0010
(0.089)*
-0.0009
(0.189)
BM -0.0002
(0.000)***
-0.0002
(0.000)***
Observations 36,146 32,326 26,029
Adjusted R2 0.0018 0.0137 0.0107
F-statistics (P-values)
SUSPECT_4Q = SUSPECT_4Q*CYCLE 8.36
(0.0039)***
824 D. Cohen et al.
123
advertising. To formally evaluate whether later-stage suspect firms behave
differently compared with suspect firms in their earlier stages, we perform an
F-test comparing the coefficients on SUSPECT and the interaction term (SUS-PECT*CYCLE). The results strongly suggest that later-stage suspect firms have
higher abnormal advertising spending than suspect firms in the early stages
(F-statistic = 6.96, p-value = 0.0085). This is consistent with a dominant carryover
effect in mature firms that use action advertising to influence short-term sales.
Finally, INDUSTRY_ADS, which controls for contemporaneous industry-wide
shocks, is positively associated with abnormal advertising spending in specification
(2), as expected. SIZE and BM also load significantly. Larger firms and firms with
Table 4 continued
(1) (2) (3)
Panel C
INTERCEPT 0.0002
(0.000)***
0.0010
(0.000)***
0.0009
(0.000)***
MBE -0.0001
(0.900)
-0.0001
(0.574)
-0.0001
(0.485)
CYCLE -0.0001
(0.025)**
MBE*CYCLE 0.0001
(0.234)
INDUSTRY_ADS 0.0160
(0.025)**
0.0128
(0.057)*
SIZE -0.0001
(0.000)***
-0.0001
(0.000)***
INCOME -0.0004
(0.525)
-0.0004
(0.517)
BM -0.0003
(0.000)***
-0.0002
(0.000)***
Observations 31,179 28,492 22,911
Adjusted R2 0.0000 0.0129 0.0109
F-statistics (P-values)
MBE = MBE*CYCLE 1.11
(0.2929)
The table reports estimation results of models whose dependent variable is monthly abnormal advertising expenditures.These are calculated based on the residuals from firm-specific time-series models estimated with up to 72 months (anaverage of 36 months) of data. SUSPECT (Panel A) is an indicator variable equal to one if quarterly income beforeextraordinary items deflated by total assets is above 0 and less than 0.0125%. SUSPECT_4Q (Panel B) is an indicatorvariable equal to one if the changes in quarterly income before extraordinary items deflated by total assets is above 0 andless than 0.05%. MBE (Panel C) is an indicator variable that takes the value of one if the firm’s earnings per share for thequarter met the outstanding analyst consensus forecast during the first month of the quarter by one cent per share or less.CYCLE is an indicator variable equal to one if the firm is at the late stages of its life cycle. The variable is computedbased on the procedures in Anthony and Ramesh (1992). INDUSTRY_ADS is the average abnormal advertising inmonth t of all firms in the same industry. SIZE is the natural logarithm of market capitalization. INCOME is incomebefore extraordinary items deflated by total assets at the beginning of the year. BM is the book-to-market ratio. P-valuesin parentheses are based on robust firm-clustered standard errors (e.g., Rogers 1993; Petersen 2008). ***, **, and *indicate significance at 1%, 5%, and 10% levels
The use of advertising activities to meet earnings benchmarks 825
123
higher book-to-market ratios are more likely to have lower abnormal advertising
spending.
Panel B of Table 4 reports the results of similar models to those reported in Panel
A. The difference is that the incentive variable, SUSPECT_4Q, is defined based on
the changes in earnings before extraordinary items relative to the same quarter in the
previous year. Similar to Panel A, the coefficient on SUSPECT_4Q is consistently
negative and significant. Also, based on the F-test that compares SUSPECT_4Q with
SUSPECT_4Q*CYCLE, there is evidence of a relation between the firm life cycle
and the likelihood to meet the financial reporting benchmark through increasing
advertising activities (F-statistic = 8.36, p-value = 0.0039).
In Panel C of Table 4, we report results using the third proxy, MBE, which uses
analysts’ forecasts as the benchmark. The coefficient on MBE is not significant in all
specifications, and there is no evidence that later-stage suspect firms have higher
abnormal advertising than other suspect firms (F-statistic = 1.11, p-value =
0.2929).15
5.2.2 Hypotheses 2 and 3
In Table 5 we report the results for hypotheses 2 and 3. In hypothesis 2, we examine
when real earnings management may occur during the fiscal quarter, using
MONTH3 and the interaction term INCENTIVE*MONTH3. Table 5 reports the
results for all three variables that proxy for earnings management incentives. In
specification (1), where SUSPECT is used as a proxy for incentives, we find that the
coefficient on MONTH3 is not significant. To compare the advertising of suspectfirms in the third month of the quarter, relative to their advertising in the first and
second month, we perform an F-test for the difference between the coefficients on
SUSPECT and SUSPECT*MONTH3. These coefficients are not statistically
different from each other (F-statistic = 0.56, p-value = 0.4562). Thus, we do not
find evidence that suspect firms are more likely to reduce their advertising in the
third month of a fiscal quarter relative to the first and second months.
The second column in Table 5 addresses hypothesis 3. It reports results,
including the indicator variable QTR4 and the interaction term SUSPECT*QTR4, to
examine whether the general effect of SUSPECT is different during the fourth
quarter. Based on the F-test that compares the coefficients on SUSPECT and
SUSPECT*QTR4 (F-statistic = 6.00, p-value = 0.0145), we conclude that suspect
firms are more likely to increase their advertising in the fourth fiscal quarter.
The remaining columns of Table 5 report the results of testing hypotheses 2 and
3 for the two additional financial reporting benchmarks (SUSPECT_Q4 and MBE).
The results for SUSPECT_4Q suggest that advertising spending is higher in the third
month of the fiscal quarter (F-statistic = 7.99, p-value = 0.0048) and in the fourth
quarter (F-statistic = 4.19, p-value = 0.0409). The results for MBE are weaker,
similar to Panel C of Table 4.
15 The results on the MBE variable are stronger when all observations are included in the analysis. The
weaker results for MBE are consistent with the notion that managing real activities towards a moving
benchmark, such as analysts’ forecasts, is inherently more difficult.
826 D. Cohen et al.
123
Table 5
X = SUSPECT X = SUSPECT_4Q X = MBE
(1) (2) (3) (4) (5) (6)
INTERCEPT 0.0009
(0.000)***
0.0009
(0.000)***
0.009
(0.000)***
0.0009
(0.000)***
0.0010
(0.000)***
0.0010
(0.000)***
X -0.0001
(0.029)**
-0.0002
(0.000)***
-0.0001
(0.000)***
-0.0001
(0.000)***
-0.0001
(0.447)
0.0001
(0.953)
MONTH3 0.0000
(0.914)
-0.0001
(0.692)
-0.0001
(0.729)
X*MONTH3 -0.0000
(0.645)
0.0001
(0.400)
0.0001
(0.581)
QTR4 -0.0001
(0.920)
0.0001
(0.726)
-0.0001
(0.533)
X*QTR4 0.0002
(0.088)*
-0.0001
(0.808)
-0.0001
(0.146)
INDUSTRY_ADS 0.0126
(0.063)*
0.0126
(0.063)*
0.0129
(0.067)*
0.0129
(0.067)*
0.0160
(0.025)**
0.0160
(0.026)**
SIZE -0.0001
(0.000)***
-0.0001
(0.000)***
-0.0001
(0.000)***
-0.0001
(0.000)***
-0.0001
(0.000)***
-0.0001
(0.000)***
INCOME -0.0009
(0.087)*
-0.0009
(0.088)*
-0.0010
(0.088)*
-0.0010
(0.089)*
-0.0004
(0.524)
-0.0004
(0.523)
BM -0.0002
(0.000)***
-0.0002
(0.000)***
-0.0002
(0.000)***
-0.0002
(0.004)***
-0.0003
(0.000)***
-0.0003
(0.000)***
Observations 34,889 34,889 32,326 32,326 28,492 28,492
Adjusted R2 0.0132 0.0133 0.0138 0.0137 0.0129 0.0129
F-statistics (P-values)
X = X*MONTH3 0.56
(0.4562)
7.99
(0.0048)***
0.59
(0.4411)
X = X*QTR4 6.00
(0.0145)**
4.19
(0.0409)**
1.01
(0.3145)
The table reports estimation results of models whose dependent variable is monthly abnormal advertising
expenditures. These are calculated based on the residuals from firm-specific time-series models estimated
with up to 72 months (an average of 36 months) of data. SUSPECT is an indicator variable equal to one if
quarterly income before extraordinary items deflated by total assets is above 0 and less than 0.0125%.
SUSPECT_4Q is an indicator variable equal to one if the changes in quarterly income before extraor-
dinary items deflated by total assets is above 0 and less than 0.05%. MBE is an indicator variable that
takes the value of one if the firm’s earnings per share for the quarter met the outstanding analyst
consensus forecast during the first month of the quarter by one cent per share or less. MONTH3 is an
indicator variable equal to one if the month is the third month of a fiscal quarter. QTR4 is an indicator
variable equal to one if the quarter is the fourth fiscal quarter. INDUSTRY_ADS is the average abnormal
advertising in month t of all firms in the same industry. SIZE is the natural logarithm of market capi-
talization. INCOME is income before extraordinary items deflated by total assets at the beginning of the
year. BM is the book-to-market ratio. p-values in parentheses are based on robust firm-clustered standard
errors (e.g., Rogers 1993; Petersen 2008)
***, **, * Significance at 1, 5, and 10% levels
The use of advertising activities to meet earnings benchmarks 827
123
Overall, the results in this section indicate that on average firms reduce their
advertising spending to avoid losses and to beat earnings from the same quarter last
year. However, we find that firms in the later stages of their life cycle increase their
advertising to meet benchmarks, possibly through the use of action advertising
whose impact on sales is relatively quick. Finally, we find some evidence that
advertising spending is higher in the third month of each quarter and in the fourth
quarter.
6 Conclusion
We study whether managers alter their advertising spending to achieve financial
reporting objectives, such as avoiding reporting a loss, avoiding reporting a decrease in
earnings, and meeting analysts’ consensus forecasts. We utilize a unique database of
monthly advertising spending in media outlets. This database enables us to analyze
actual real managerial activities as opposed to reported expenses, which are also
subject to accrual manipulations. Thus, our approach is different from that of prior
studies in which reported expenses serve as an indirect proxy for inferring real
earnings management. Our data on advertising also allow us to focus on a specific item
that is subject to managerial discretion as opposed to an aggregate proxy of several
discretionary expenses activities. Finally, the high-frequency monthly data provides
us with a powerful setting to investigate the incentives to meet quarterly earnings
benchmarks as well as to examine the timing of real activities’ manipulations.
We find evidence that firms reduce their advertising spending to meet two of the
three financial reporting objectives that we examine. In addition, we uncover
circumstances in which firms have incentives to increase advertising spending to
meet earnings benchmarks. Specifically, more mature firms, which are likely to use
action advertising, exhibit patterns consistent with increasing advertising to meet
earnings benchmarks. Finally, we find some evidence of an increase in abnormal
advertising activities in the third month of a fiscal quarter and in the fourth quarter.
Overall, this study extends the literature on real earnings management activities.
The evidence we provide enhances our understanding of the interplay between
managers’ incentives and the real actions managers take to achieve financial
reporting goals.
Acknowledgments We gratefully acknowledge the financial support of the Center for Research in
Economics and Strategy at the Olin Business School. We have benefited from the comments of two
anonymous referees, Eli Bartov, Anne Beatty, Messod Beneish, Larry Brown, Donal Byard, Sandra
Chamberlain (FARS discussant), Masako Darrough, Richard Dietrich, Nick Dopuch, Peter Easton,
Richard Frankel, James Hansen, Steve Huddart, Bjorn Jorgensen, April Klein, Andy Leone, Fred
Lindahl, Thomas Lys, Christina Mashruwala, Shamin Mashruwala, Brian Miller, Doron Nissim, Shail
Pandit, Sugata Roychowdhury, Kristi Schneck, Ram Venkataraman, Xin Wang, Joe Weber, TJ Wong,
Minlei Yei (AAA discussant), Amy Zang (CFEA discussant), Paul Zarowin, Jerry Zimmerman,
participants at the 2008 FARS conference, the 2008 AAA annual meeting, the 2007 Conference of
Financial Economics and Accounting, the 2007 Nick Dopuch Conference, and workshop participants at
University of Arizona, Baruch College, University of Chicago, Chinese University of Hong Kong,
Columbia University, Georgia State University, George Mason University, George Washington
University, IDC Herzliya, INSEAD, University of Illinois at Chicago, University of Illinois at Urbana-
828 D. Cohen et al.
123
Champaign, University of Maryland, University of Miami, The Ohio State University, Penn State
University, Southern Methodist University, Stanford University, University of Technology Sydney, and
Washington University in St. Louis.
Appendix 1
Our database records advertising activities at the time they appear in a media outlet
(for example, when they are aired on television or appear in print). According to
Statement of Position (SOP) 93-7: Reporting on Advertising Costs, firms are
required to expense advertising activities as they are incurred. Thus, if firms follow
GAAP, our database will include all media-related advertising that should have
been expensed in the period in which it was aired or printed.
The following paragraphs from SOP 93-7 describe the accounting treatment in
detail:
Paragraph 26
The costs of advertising should be expensed either as incurred or the first time theadvertising takes place… except for
1. Direct-response advertising (1) whose primary purpose is to elicit sales tocustomers who could be shown to have responded specifically to the advertisingand (2) that results in probable future economic benefits.
2. Expenditures for advertising costs that are made subsequent to recognizingrevenues related to those costs.
Paragraph 42
Advertising activities may have several component costs. Two primary components,which are made up of other components, are the costs of (a) producing advertisements,such as the costs of idea development, writing advertising copy, artwork, printing,audio and video crews, actors, and other costs, and (b) communicating advertisementsthat have been produced, such as the costs of magazine space, television airtime,billboard space, and distribution (postage stamps, for example).
Paragraph 43
Costs of producing advertising are incurred during production rather than when theadvertising takes place.
Paragraph 44
Costs of communicating advertising are not incurred until the item or service hasbeen received and should not be reported as expenses before the item or service hasbeen received… For example
The use of advertising activities to meet earnings benchmarks 829
123
• The costs of television airtime should not be reported as advertising expensebefore the airtime is used. Once it is used, the costs should be expensed, unlessthe airtime was used for direct-response advertising activities that meet thecriteria for capitalization under this SOP.
• The costs of magazine, directory, or other print media advertising space shouldnot be reported as advertising expense before the space is used. Once it is used,the costs should be expensed, unless the space was used for direct-responseadvertising activities that meet the criteria for capitalization under this SOP.
Our database contains only media-related advertising that cannot be treated as direct-
response advertising and therefore has to be expensed in the period of appearance.
Suppose that an advertisement appeared on television on January 15, 2007. This
advertisement will be captured in our database. If the firm follows GAAP, the
advertisement will appear as an expense in the financial statements for the quarter
ending on March 31, 2007. Thus, our database captures all media-related advertising
activities during the period in which they are supposed to be expensed in the income
statement.
Appendix 2
This appendix describes the various media outlets from which the advertising data
are collected.
Television
Cable television: Commercial occurrences and advertising activities information for
52 cable television networks. Cable television is monitored via satellite 24 hours a
day, 365 days a year.
Spot television: Commercial occurrence and expenditure information for more than
600 English-speaking stations and 35 Spanish-speaking stations in 100 major
markets. Spot television is monitored 21 hours a day (5 a.m.–2 a.m.). The monitored
stations constitute the principal stations in each market and typically include the
network affiliates, major independents, and Spanish affiliates. Public broadcasting
stations are not monitored.
Network television: Commercial occurrence and expenditure information for seven
broadcast networks: ABC, CBS, FOX, NBC, PAX/I, MNTV, and CW.
Magazines: All paid advertising space and expenditure data in over 350 consumer
magazines and over 30 local magazines.
Newspapers: All advertising space in over 250 daily and Sunday newspaper editions
and Sunday magazines, including national newspapers such as The New York Times,
830 D. Cohen et al.
123
USA Today, and The Wall Street Journal. In national newspapers, all national and
regional editions are measured.
Radio
Local radio: Data are based on the tracking of radio billings by stations in more than
100 radio markets.
National spot radio: Nationally placed spot radio data for approximately 4,000
stations in more than 225 markets. Reported expenditures are based on audited
billings from contract information provided by the following major national station
representative organizations: ABC, Allied, Christal, Clear Channel, Cumulus, D&R,
Infinity, Katz, KHM, Lotus, and McGavren.
Internet
Advertising expenditure information is measured for over 2,500 sites and 90,000
brands in the United States and Canada. A proprietary spider probes the sites on an
ongoing basis and captures and stores every image on a page.
References
Accounting Principles Board. (1973). Interim Financial Reporting. APB Opinion No. 28. New York.
Anderson, M., Banker, R., & Janakiraman, S. (2003). Are selling, general, and administrative costs
‘sticky’? Journal of Accounting Research, 41, 47–63.
Anthony, J., & Ramesh, K. (1992). Association between accounting performance measures and stock
prices: A test of the life cycle hypothesis. Journal of Accounting and Economics, 15, 203–227.
Arora, R. (1979). How promotion elasticities change. Journal of Advertising Research, 19, 57–62.
Assmus, G., Farley, J., & Lehmann, D. R. (1984). How advertising affects sales: Meta-analysis of
econometric results. Journal of Marketing Research, 21, 65–74.
Bagwell, K. (2007). The economic analysis of advertising. In Mark Armstrong & Rob Porter (Eds.),
Handbook of industrial organization (Vol. 3, pp. 1701–1844). Amsterdam: North-Holland.
Brown, L., & Caylor, M. (2005). A temporal analysis of quarterly earnings thresholds: Propensities and
valuation consequences. The Accounting Review, 80, 423–440.
Bruns, W., & Merchant, K. (1990). The dangerous morality of managing earnings. ManagementAccounting, 72, 22–25.
Burgstahler, D., & Dichev, I. (1997). Earnings management to avoid earnings decreases and losses.
Journal of Accounting and Economics, 24, 99–126.
Chapman, C. (2008). The effects of real earnings management on the firm, its competitors and subsequent
reporting periods. Working paper, Northwestern University.
Darrough, M., & Rangan, S. (2005). Do insiders manipulate earnings when they sell their shares in an
initial public offering? Journal of Accounting Research, 45, 1–33.
DeGeorge, F., Patell, J., & Zeckhauser, R. (1999). Earnings management to exceed thresholds. Journal ofBusiness, 72, 1–33.
Fields, T. D., Lys, T. Z., & Vincent, L. (2001). Empirical research on accounting choice. Journal ofAccounting and Economics, 31, 255–307.
Financial Accounting Standards Board. (1974). Reporting accounting changes in interim financial
statements. Statement of financial accounting standards No. 3. Stamford, CT.
Financial Accounting Standards Board. (1977). FASB interpretation No. 18. Accounting for income taxes
in interim periods. Stamford, CT.
The use of advertising activities to meet earnings benchmarks 831
123
Foster, G. (1977). Quarterly accounting data: Time-series properties and predictive ability results. TheAccounting Review, 52, 1–21.
Graham, J. R., Harvey, C. R., & Rajgopal, S. (2005). The economic implications of corporate financial
reporting. Journal of Accounting and Economics, 40, 3–73.
Gunny, K. (2005). What are the consequences of real earnings management? Working paper, University
of Colorado. http://papers.ssrn.com/sol3/papers.cfm?abstract_id=816025.
Healy, P., & Wahlen, J. (1999). A review of the earnings management literature and its implications for
standard setting. Accounting Horizons, 13, 365–384.
Kothari, S. P. (2001). Capital market research in accounting. Journal of Accounting and Economics, 31,
105–231.
Parker, P. M., & Gatignon, H. (1996). Order of entry, trial diffusion, and elasticity dynamics: An
empirical case. Marketing Letters, 7, 95–109.
Parsons, L. (1975). The product life cycle and time-varying advertising elasticities. Journal of MarketingResearch, 12, 476–480.
Petersen, M. (2008). Estimating standard errors in finance panel data sets: Comparing approaches. Reviewof Financial Studies, 22, 435–480.
Rogers, W. (1993). Regression standard errors in clustered samples. Stata Technical Bulletin, 13, 19–23.
Roychowdhury, S. (2006). Earnings management through real activities manipulation. Journal ofAccounting and Economics, 42, 335–370.
Winer, R. S. (1979). An analysis of the time-varying effects of advertising: The case of Lydia Pinkham.
Journal of Business, 52, 563–576.
Zang, A. (2006). Evidence on the tradeoff between real manipulation and accrual manipulation. Working
paper, University of Rochester. http://papers.ssrn.com/sol3/papers.cfm?abstract_id=961293.
832 D. Cohen et al.
123