-
Accounting Horizons American Accounting AssociationVol. 27, No.
1 DOI: 10.2308/acch-502912013pp. 91112
Revenue Recognition, Earnings Management,and Earnings
Informativeness in the
Semiconductor Industry
Stephanie J. Rasmussen
SYNOPSIS: Manufacturers that sell products to distributors
experience product returnand pricing adjustment uncertainties until
the products are resold to end-customers.Such manufacturers
recognize revenue when products are delivered to distributors
(sell-in), when distributors resell products (sell-through), or
under some combination of thesemethods (sell-in for some
distributor sales and sell-through for others). This studyexamines
the implications of these revenue recognition methods for a sample
ofsemiconductor firms during 20012008. Semiconductor firms face
rapid productobsolescence, declining prices over product life
cycles, and unexpected industrydownturns, which naturally lead to
product return and pricing adjustment uncertainties. Ifind that
sell-through and combination firms are less likely to manage
earnings comparedto sell-in firms. I also find that earnings are
more informative for sell-through firmscompared to both sell-in and
combination firms. These findings suggest thatmanufacturers that
sell products through the distribution channel should defer
revenuerecognition until product return and pricing adjustment
uncertainties are resolved.
Keywords: revenue recognition; earnings management; earnings
informativeness;distributors; manufacturers.
JEL Classications: M41.
Data Availability: Data are available from the sources identied
in the text.
Stephanie J. Rasmussen is an Assistant Professor at The
University of Texas at Arlington.
I gratefully acknowledge helpful comments and suggestions
offered by Terry Shevlin (editor), two anonymous referees,Anwer
Ahmed (dissertation chair), Kris Allee, Cory Cassell, Mike Drake,
Jap Efendi, Rebecca Files, Tom Omer, DudleyPoston, Jaime Schmidt,
Mary Stanford, Senyo Tse, Connie Weaver, and workshop participants
at Texas A&MUniversity, University of Houston, The University
of Texas at Arlington, and the 2009 AAA Annual Meeting. I
amindebted to the following practitioners for willingly sharing
insights from their own experiences with revenue recognitionin the
semiconductor industry: Sanjoy Chatterji, Wendy Clancy, Bernard
Gutmann, Linda King, Carl Mangine, andLaurie Martens. I thank
Linying Zhou for research assistance. Financial support from Texas
A&Ms Mays BusinessSchool and The University of Texas at
Arlington is greatly appreciated.
This paper is based, in part, on my dissertation completed at
Texas A&M University.
Submitted: October 2011Accepted: June 2012
Published Online: September 2012Corresponding author: Stephanie
J. Rasmussen
Email: [email protected]
91
-
INTRODUCTION
Revenue is one of the most important earnings components,
usually the largest item on the
income statement, and a strong indicator of firm performance
(e.g., Turner 2001). Revenue
is also very complex as evidenced by more than 200 publications
(statements, opinions,
bulletins) that provide revenue recognition guidance under U.S.
GAAP (FASB 2005). Given the
importance and complexity of revenue, it is imperative for
financial statement users to have a strong
understanding of revenue recognition and its implications for
evaluating firm performance and
valuation.
Standard-setters and academic researchers note that trade-offs
exist between the relevance and
reliability of different accounting practices (see e.g., FASB
1980; Schipper 2003). Prior research
examines the implications of revenue recognition when
uncertainties exist related to product
delivery (Altamuro et al. 2005; Zhang 2005) and the pricing of
undelivered contract elements
(Srivastava 2011). These studies find that earnings are more
informative, yet more likely to be
managed, when revenue recognition occurs before the
uncertainties are resolved. I extend these
studies by examining the implications of revenue recognition for
manufacturers that sell their
products to distributors. Such manufacturers face product return
and pricing adjustment
uncertainties until the distributors resell products to
end-customers. In addition, these manufacturers
have opportunities for real earnings management through channel
stuffing. Channel stuffing
refers to either pulling in distributor orders from a future
period or shipping large, unusual orders to
distributors in order to boost revenue (Penman 2007). Therefore,
it is not clear whether prior
studies findings with respect to the implications of revenue
recognition will hold for firms with
product return and pricing adjustment uncertainties.
Three revenue recognition methods exist for manufacturers sales
to distributors. Under the
sell-in method, manufacturers recognize revenue upon delivery of
product to the distributor (i.e.,sales into the distribution
channel) and maintain product return and pricing adjustment
accrualsuntil distributor rights have lapsed at resale. Under the
sell-through method, manufacturersrecognize revenue when the
distributor resells product to an end-customer and all
uncertainties have
been resolved (i.e., sales through the distribution channel).
Manufacturers exclusively use one ofthese methods to recognize
revenue for sales to all distributors, or they use the sell-in
method for
some sales to distributors and the sell-through method for other
sales (hereafter, the combinationmethod).
Consistent with prior research (Altamuro et al. 2005; Srivastava
2011), I examine the trade-offs
between managerial discretion and earnings informativeness for
differing revenue recognition
methods. Since the sell-in method recognizes revenue before
product return and pricing adjustment
uncertainties are resolved, managers must estimate the
likelihood of those events. Managers at
sell-in firms also have opportunities to intentionally
manipulate their estimates or stuff the
distribution channel in order to meet or beat earnings
benchmarks (Glass, Lewis & Co. 2004;
Greenberg 2006; Schilit and Perler 2010). In contrast,
combination firms only have opportunities to
exercise discretion for a portion of all distributor revenue,
and managerial discretion does not exist
for sell-through firms that defer revenue recognition until
product resale by distributors. Due to the
differing opportunities for managers to exercise discretion, I
expect that the incidence of earnings
management is less likely for sell-through and combination firms
compared to sell-in firms.
While I expect earnings management differences among the revenue
recognition methods for
sales to distributors, it is not clear whether the
informativeness of manufacturers earnings differs
among the methods. On one hand, the sell-in method provides
financial statement users with more
timely information than the combination and sell-through methods
about the business transactions
between manufacturers and distributors. On the other hand,
estimates of product returns and pricing
adjustments required under the sell-in method are susceptible to
both intentional and unintentional
92 Rasmussen
Accounting HorizonsMarch 2013
-
estimation errors. In addition, sell-in firms receive a greater
benefit from channel stuffing compared
to combination and sell-through firms. Thus, the sell-in method
would not reflect a firms actual
revenue-generating performance as well as the combination or
sell-through method if these issues
prevail.
I examine revenue recognition methods for sales to distributors
and their trade-offs using a
sample of 1,572 firm-quarters for 80 semiconductor manufacturers
during 20012008. Thirty-two
percent of the sample uses the sell-in method, 20 percent of the
sample uses the sell-through
method, and 48 percent of the sample uses a combination of the
two revenue recognition methods. I
limit the sample to semiconductor firms because this industry
experiences rapid product
obsolescence, declining prices over product life cycles, and
unexpected industry downturns. These
issues contribute to channel stuffing, product return, and
pricing adjustment uncertainties for
manufacturers with distributor customers. Semiconductor firms
are also more likely to disclose
information about their relationships with distributors and less
likely to have significant service- or
retail-related revenue (which could add noise to empirical
analyses) than other firms that sell to
distributors. I begin the sample period in 2001 because revenue
recognition principles have been
consistent since that year (first under Staff Accounting
Bulletin [SAB] 101 and later under SAB
104).
I first test whether earnings management is less likely for
sell-through and combination firms
compared to sell-in firms. Consistent with prior research (e.g.,
Barton and Simko 2002; Cheng and
Warfield 2005), I use the incidence of meeting or beating
analysts consensus quarterly earnings
forecast as a proxy for earnings management to a benchmark. I
find that sell-through and
combination firms are significantly less likely to meet or beat
analysts consensus quarterly earnings
forecast compared to sell-in firms after controlling for other
determinants of meeting or beating that
have been identified by prior work. The likelihood of meeting or
beating analysts consensus
earnings forecast does not differ between sell-through and
combination firms. These findings
suggest that earnings management is more likely for firms that
recognize all revenue before product
return and pricing adjustment uncertainties are resolved.
Next, I test for earnings informativeness differences among
sell-in, sell-through, and
combination firms. Specifically, I examine whether the earnings
response coefficient, an indicator
of earnings informativeness, varies for firms with different
revenue recognition methods (e.g.,
Altamuro et al. 2005; Srivastava 2011). I define unexpected
earnings as the difference between
actual earnings per share and analysts last consensus earnings
forecast prior to the quarterly
earnings announcement (scaled by stock price at quarter-end),
and unexpected returns are
cumulative market-adjusted stock returns for two trading days
beginning on the quarterly earnings
announcement date. I also control for many determinants of
earnings response coefficients
identified by prior research. I find that the earnings response
coefficient (the coefficient on
unexpected earnings) is significantly higher for sell-through
firms compared to both sell-in and
combination firms. This finding suggests that earnings are more
informative for firms that defer
revenue recognition until all product return and pricing
adjustment uncertainties are resolved.
Collectively, this study suggests that manufacturers with
product return and pricing adjustment
uncertainties should recognize revenue from sales to
distributors after the uncertainties are resolved.Although this
conclusion differs from prior studies, which suggest that earnings
are more
informative if firms recognize revenue before uncertainties are
resolved (Altamuro et al. 2005;Srivastava 2011), important
differences exist between my setting and those examined in prior
work.
First, the revenue recognition settings examined by Altamuro et
al. (2005), and Srivastava (2011)
allow for manager manipulation of accounting estimates while my
setting allows for both
manipulation of accounting estimates and real earnings
management. Channel stuffing is a seriousconcern of regulators,
forensic accountants, and others (Glass, Lewis & Co. 2004;
Greenberg 2006;
Schilit and Perler 2010), and egregious channel stuffing schemes
have resulted in Securities and
Revenue Recognition, Earnings Management, and Earnings
Informativeness 93
Accounting HorizonsMarch 2013
-
Exchange Commission (SEC) enforcement actions. It is likely that
public awareness of channel
stuffing in my setting contributes to the finding that earnings
are more informative when revenue
recognition is deferred until distributors have resold product
to end-customers.
Second, I examine sample firms that consistently use the same
revenue recognition method
throughout the sample period while Altamuro et al. (2005) and
Srivastava (2011) do not. Forester
(2008) suggests that the cumulative-effect adjustment of an
accounting change impacts earnings
informativeness in the transitional period following firms
adoption of a new revenue recognition
method. When he excludes the transitional period from his
analysis of Altamuro et al.s (2005)
setting, he finds that earnings are more informative for firms
deferring revenue recognition under
SAB 101 compared to firms that accelerated revenue recognition
prior to SAB 101. This result is
comparable to what I find with respect to the earnings
informativeness of sell-through and sell-in
firms with distributor customers. Srivastavas (2011) results
would not be affected by a transitional
period because firms in his setting were not required to report
a cumulative-effect adjustment when
changing revenue recognition methods.
I expect that my study will interest students, managers, and
other financial statement users
because it contributes to a growing literature that examines
revenue recognition in specific
industries (Bowen et al. 2002; Zhang 2005; Srivastava 2011).
Differences in revenue recognition
practices often make it difficult for financial statement users
to compare revenue and earnings
among entities and industries (Schipper et al. 2009), and it is
important to examine many settings in
order to improve our understanding of the implications of
revenue recognition for firms. This study
is also potentially useful to investors, analysts, auditors, and
regulators who monitor semiconductor
and other high-technology manufacturers that sell to
distributors. Since the results suggest that
earnings management concerns about the sell-in method are
warranted and earnings informative-
ness differs based on firms revenue recognition methods, these
factors should be considered when
interpreting the financial statements and stock returns of
manufacturers that sell to distributors and
comparing them to other firms.
The next section discusses background and develops hypotheses. I
then discuss the empirical
models and describe the sample in the third section. Empirical
evidence is presented the fourth
section and the final section summarizes and concludes.
BACKGROUND AND HYPOTHESES
Background
Distributors purchase products from manufacturers and later
resell the products. This
arrangement benefits manufacturers because distributors (1) act
as an additional sales force, (2)aggregate and service small orders
that manufacturers are otherwise unwilling to fulfill, and (3)
reduce manufacturers collection risk (Credit Suisse 2007). Many
manufacturers rely heavily on
distributors. For example, research suggests that distributors
service more than 25 percent of global
semiconductor/electronic component sales (Credit Suisse 2007),
and some semiconductor
manufacturers indicate that at least 50 percent of their sales
are to distributors (e.g., Cypress
Semiconductor 2006 10-K filing; Fairchild Semiconductor 2008
10-K filing).
In order for manufacturers to recognize revenue from sales to
distributors, SAB 101 and later
SAB 104, both require that (1) persuasive evidence of an
arrangement exists, (2) delivery has
occurred, (3) the final selling price is fixed or determinable,
and (4) collectability is reasonably
assured (SEC 1999, 2003). Although manufacturers sales to
distributors easily meet the
arrangement, delivery, and collectability requirements, the
manufacturer must decide if the final
selling price is fixed or determinable. A conservative
interpretation of the revenue recognition
standard suggests that the final selling price is indeterminable
for sales subject to pricing
adjustments or rights-of-return. However, interpretive guidance
suggests that a selling price is
94 Rasmussen
Accounting HorizonsMarch 2013
-
determinable if product returns and pricing adjustments can be
reasonably estimated.1 Thus, the
revenue recognition standard allows managers some discretion to
communicate private, value-
relevant information to financial statement users but there is
also the risk that managers will use the
discretion to manipulate earnings (Healy and Wahlen 1999).
Depending on the fixed or determinable nature of the final
selling price, a manufacturer will
recognize revenue from sales to distributors under one of three
revenue recognition methods: sell-
in, sell-through, or combination. Under the sell-in method, the
manufacturer records accounts
receivable, reduces inventory, and recognizes both revenue and
cost of goods sold when product
is delivered to the distributor. This accounting method provides
a timely reflection of product
transfer between the two parties. Since revenue is recognized at
delivery, the manufacturer
maintains product return and pricing adjustment accruals for
limited return rights on regular
purchases2 and pricing adjustments intended to compensate for
falling market prices or
incentivize sales of certain products (Lee et al. 2000; Credit
Suisse 2007). These accrual estimates
are typically based on historical distributor return and pricing
adjustment data. Manufacturers
revenue recognition disclosures suggest that the sell-in method
is used when (1) distributors do
not have product return and pricing adjustment rights (i.e.,
selling prices are fixed at the time of
sale), or (2) distributors product returns and pricing
adjustments can be reasonably and reliably
estimated (see Appendix A).
Under the sell-through method, the manufacturer reduces
inventory and records accounts
receivable, deferred revenue, and deferred cost of goods sold
when product is delivered to the
distributor. The manufacturer recognizes revenue and cost of
goods sold once notification is
received from the distributor that the product has been resold.3
This method more accurately reflects
end-customer demand, and product return and pricing adjustment
accruals are not needed since
revenue is deferred until distributor rights have lapsed.
Although distributors provide inventory and
resale data to manufacturers, challenges exist regarding data
reliability and format. Chipalkatti et al.
(2007) note that once data are received from multiple
distributors, manufacturers must remove
errors, validate the data, and convert the data into one
consistent format.4 In order to address these
issues, sell-through revenue recognition requires additional
internal controls beyond those needed
for other sales. Revenue recognition disclosures suggest that
manufacturers typically use the sell-
through method to recognize revenue when they believe they are
unable to accurately estimate
distributors product returns and pricing adjustments (see
Appendix A).
A combination of the two methods occurs when a manufacturer
recognizes revenue under the
sell-in method for some distributors and under the sell-through
method for other distributors. Under
this method, product return and pricing adjustment accruals are
maintained only for those sales to
distributors that are recognized under the sell-in method.
Manufacturers disclosures suggest that
they use a combination of the sell-in and sell-through methods
if the firm is able to reasonably and
1 SAB 104 interpretive guidance refers to Statement 48, }6 and
8, which state that revenue cannot be recognized if afirm is unable
to make a reasonable estimate of product returns (FASB 1981). SAB
104 also directs users to SOP97-2, }26 and 30-33, which state that
prices on products sold to distributors are not fixed and
determinable if theseller is unable to make reasonable estimates of
pricing adjustments (AICPA 1997).
2 Manufacturers often give distributors the right to return a
certain percentage of their inventory or exchange oldinventory for
new inventory (i.e., stock rotation) (Chipalkatti et al. 2007). In
addition, distribution agreementstypically include clauses that
allow distributors to return any product on hand if the
relationship with themanufacturer is terminated (e.g., Arrow
Electronics 2008 10-K filing; Ingram Micro 2008 10-K filing).
3 Practitioners indicated that, regardless of the revenue
recognition method, manufacturers regularly receive resaleand
inventory data from distributors. These data are used to understand
end-customer demand and planproduction.
4 Texas Instruments cites lack of confidence in distributor data
as one reason it uses the sell-in method (Greenberg2006). KPMG LLP
(2006) finds that at least 20 percent of resale reports from
channel partners contain errors. Dataissues also affect accrual
estimation under the sell-in method.
Revenue Recognition, Earnings Management, and Earnings
Informativeness 95
Accounting HorizonsMarch 2013
-
reliably estimate product return and pricing adjustment accruals
for some distributors but not for
others. Estimation abilities can differ among distributors based
on (1) the availability of historical
data needed to predict future returns and pricing adjustments,
or (2) differing inventory levels that
could influence distributor power over the manufacturer.
Alternatively, manufacturers disclosures
suggest that some firms use a combination of the two methods if
product return and pricing
adjustment rights are only given to some distributors.5
Manufacturers in many industries exhibit some use of the sell-in
and sell-through methods
for sales to distributors (Glass, Lewis & Co. 2004;
Greenberg 2006; Chipalkatti et al. 2007). I
focus my study on firms in the semiconductor industry for a
variety of reasons. First, the
semiconductor industry experiences rapid product obsolescence,
declining prices over product
life cycles, and unexpected industry downturns. These issues all
contribute to product return and
pricing adjustment uncertainties for sales to distributors.
Second, my review of SEC filings
indicates that semiconductor firms are more likely than firms in
other industries to disclose
information about their distributor relationships. Specifically,
more semiconductor firms disclose
the percentage of revenue attributable to all or some of their
distributor customers, which
indicates firm reliance on the distribution channel. Third, I
find that semiconductor firms generate
most of their revenue from product sales, while manufacturers
selling to distributors in other
industries often have significant service-related revenue.
Examining firms with a significant
amount of service revenue could add noise to my empirical
analyses if revenue recognition
methods differ for products and services. Fourth, semiconductor
firms are less likely than other
manufacturers to sell to retailers because semiconductor
products are typically components used
in the assembly of a product. Sales to retailers could also add
noise to my empirical analyses.
Finally, only examining the semiconductor industry allows me to
focus on a set of manufacturers
with relatively homogeneous characteristics.
Hypotheses
The sell-in method offers opportunities for managers to
manipulate earnings using both real
activities and accounting accruals. For example, managers can
ship excess product to distributors at
the end of an accounting period in order to increase earnings
(i.e., channel stuffing) (Penman 2007).
Channel stuffing boosts revenue of sell-in firms because revenue
recognition occurs when product
is delivered to distributors. The SEC has investigated channel
stuffing and brought enforcement
actions against firms with egregious channel stuffing activities
(e.g., Vitesse Semiconductor). Lynn
Turner, former SEC chief accountant, summarizes concerns about
these activities as follows: Ifound nothing good about revenue
recognition upon sell-in. Sooner or later, the urge to stuff
the
channel, especially when things are not going well and numbers
for the next quarter are short, is
very tempting (Greenberg 2006). In contrast, managers using the
sell-through method are lesslikely to stuff the distribution
channel because revenue recognition is deferred until
distributors
resell products to end-customers.
Managers at sell-in firms can also manage earnings through
accrual manipulations. As
discussed earlier, sell-in firms maintain product return and
pricing adjustment accruals until
distributor rights have lapsed. Estimation of these accruals is
subject to managerial discretion, and
extensive accounting research suggests that managers use
accruals to manage earnings (see Healy
5 Changes in a manufacturers ability to estimate product returns
and pricing adjustments or contractual rightsoffered to a
distributor over time could lead the manufacturer to change the
revenue recognition method for thedistributor in question. Such a
change in the revenue recognition method necessitates that the firm
report theaccounting change and a cumulative-effect adjustment,
which should discourage opportunistic revenuerecognition
changes.
96 Rasmussen
Accounting HorizonsMarch 2013
-
and Wahlen 1999; Beneish 2001; Fields et al. 2001 for surveys).6
In contrast, sell-through firms do
not have the opportunity to manipulate product return and
pricing adjustment accruals for
distributor customers since these accruals are not
maintained.
Prior research suggests that earnings management occurs through
both real activities
manipulation and accrual manipulation (e.g., Healy and Wahlen
1999; Fields et al. 2001; Graham
et al. 2005; Roychowdhury 2006).7 Sell-in firms have
opportunities to manage earnings from
distributor sales using both types of manipulation while
sell-through firms do not. In addition, sell-
in firms have more manipulation opportunities than combination
firms since the combination
method recognizes revenue at product delivery for only a portion
of a firms sales to distributors.
My first hypotheses (stated in the alternative form) are as
follows:
H1a: The incidence of earnings management is less likely for
sell-through firms compared to
sell-in firms.
H1b: The incidence of earnings management is less likely for
combination firms compared to
sell-in firms.
Earnings informativeness could also differ based on a firms
revenue recognition method for
sales to distributors. Earnings that provide new information to
the market about expectations of
future cash flows, as evidenced by changes in stock prices, are
considered to be informative
(Kothari 2001). On one hand, the sell-in method provides a
timely reflection of product transfer
between manufacturers and distributors. New accounting
information is more quickly incorporated
into the financial statements under this method compared to the
sell-through and combination
methods and should be useful to the market assuming that accrual
estimates are accurate and
manufacturers do not stuff the distribution channel. When
sell-in revenue recognition results in
accurate and reliable financial statements, earnings should be
more informative for sell-in firms
compared to sell-through and combination firms.
On the other hand, if sell-in firms have intentional performance
manipulations (i.e., channel
stuffing, accrual manipulation), earnings reported by these
firms are not earned and earnings
informativeness should suffer. In addition, unintentional
estimation errors can reduce the
informativeness of earnings. Marketing theory suggests that
powerful distributors have the ability
to heavily influence trade terms with manufacturers (e.g., Tsay
2002). Since manufacturers often
sell a large amount of product to distributors and 75 percent of
semiconductor/electronic component
distributors purchases represent speculation and forecasts of
future end-customer orders (Credit
Suisse 2007), distributors have leverage to pressure
manufacturers into accepting special return or
pricing adjustment requests if end-customer demand does not
materialize. For example, BCD
Semiconductor responded to distributor requests following a
recent industry downturn by allowing
distributors to return nearly four times more product than what
was required under the companys
standard return rights (BCD Semiconductor 2008 Registration
Statement). Distributors are also
likely to request special returns for product that was
previously stuffed into the channel. Thus, if
6 Use of product return and pricing adjustment accruals under
the sell-in method also increases the risk ofunintentional accrual
estimation errors by management.
7 Earnings management opportunities also exist for sales to
non-distributor customers. Revenue recognitiondisclosures for
sample firms suggest that manufacturers typically recognize revenue
from non-distributorcustomers at the time of product delivery and
sales returns are only allowed for defective products.
Thus,managers have opportunities to stuff non-distributor channels
and manipulate warranty accruals. Since sell-in,sell-through, and
combination firms all have the opportunity to stuff the
non-distribution channel and manipulatenon-distributor sales
accruals, I attempt to control for these activities in my empirical
test of earningsmanagement.
Revenue Recognition, Earnings Management, and Earnings
Informativeness 97
Accounting HorizonsMarch 2013
-
sell-in revenue recognition results in inaccurate earnings or
product returns estimates, earnings
should be less informative for sell-in firms compared to
sell-through and combination firms.
Because it is unclear whether earnings informativeness differs
based on a manufacturers
revenue recognition method for sales to distributors, my final
hypothesis (stated in the null form) is
as follows:
H2: Earnings informativeness does not differ among sell-in,
sell-through, and combinationfirms.
METHOD
Revenue Recognition and Earnings Management
I use the following logistic regression model to examine if
sell-through and combination firms
manage earnings less than sell-in firms (H1a and H1b):
MeetBeatit a0 a1SellThroughit a2Comboit a3Sales Growthit a4Rank
of MTBit a5NOAit a6Rank of Sharesit a7Rank of Sizeit a8Reportit
a9Report3Avg Disty Revenueit a10Bonus%it a11Options%it
a12Analystsit a13CVAFit a14Revise Downit axTime Indicators e: 1
Degeorge et al. (1999) find that a disproportionate number of
firms meet or beat analysts earnings
forecasts. Consistent with prior literature (e.g., Barton and
Simko 2002; Cheng and Warfield 2005),
I use the incidence of meeting or beating analysts consensus
earnings forecast as a proxy for
managing earnings to a benchmark. MeetBeat is an indicator
variable that equals 1 if the firm meetsthe last I/B/E/S consensus
earnings forecast prior to the quarterly earnings announcement or
beats it
by any amount, and 0 otherwise. I examine the incidence of
meeting or beating analysts consensus
earnings forecast and not analysts consensus revenue forecast
because (1) revenue recognitionmethods for sales to distributors
affect both revenues and cost of goods sold, and (2) executives
report that earnings is a more important performance metric than
revenues (Graham et al. 2005).
The main variables of interest in Model 1 are Sell-Through and
Combo, which are indicators equalto 1 if the firm uses the
sell-through or combination revenue recognition method for sales
to
distributors, and 0 otherwise. Consistent with H1a and H1b, I
expect negative coefficients on Sell-Through and Combo,
respectively, indicating that earnings management is less likely
under the sell-through and combination methods compared to the
sell-in method.
Model 1 includes a variety of control variables that prior
research suggests are associated with
meeting or beating analysts consensus earnings forecast.
Consistent with prior research (e.g., Das
et al. 1998; Barton and Simko 2002; Cheng and Warfield 2005), I
control for the following firm
characteristics: growth opportunities (Sales Growth, Rank of
MTB), constraints on earningsmanagement (net operating assets
[NOA], shares outstanding [Rank of Shares]), and firm size (Rankof
Size). NOA is an accrual-based measure of net assets, and Barton
and Simko (2002) find thatprior abnormal accruals are positively
associated with NOA. High NOA suggests overstated assetsand
previous earnings management through abnormal accruals, which
should make it more difficult
for managers to manipulate earnings in the current quarter.
I control for the percentage of revenue attributable to
distributor customers8 because reliance
on the distribution channel likely influences firms revenue
recognition practices and opportunities
8 Revenue attributable to distributor customers (if reported) is
disclosed in the 10-K filing. However, not all firmsfollow
consistent reporting practices. For instance, some firms report
revenue attributable to all distributorcustomers while other firms
only report revenue attributable to their top one or two
distributor customers. Inaddition, some firms do not report revenue
attributable to distributor customers every year while other firms
neverreport this information.
98 Rasmussen
Accounting HorizonsMarch 2013
-
to stuff non-distributor channels and/or manipulate
non-distributor sales accruals.9 Report controlsfor the fact that
not all firms report how much of their revenue is attributable to
distributor
customers. This variable is an indicator set to 1 for those
firms reporting revenue attributable to
distributor customers at least once during the sample period and
0 for those firms never reporting
this information. Report3 Avg Disty Revenue is the interaction
between Report and the average ofall annual revenue attributable to
distributor customers reported by the firm during the sample
period. This variable is set to 0 for firms that never report
revenue attributable to distributor
customers. I control for the percentage of CEO compensation
attributable to cash-based incentives
(Bonus%) and stock options (Options%) because prior research
suggests that management incentivecompensation is associated with
earnings management activities (e.g., Cheng and Warfield 2005;
Bergstresser and Philippon 2006; Cornett et al. 2008) and
accounting choices (e.g., Aboody et al.
2000; Aboody et al. 2004; Efendi et al. 2007).
Consistent with prior research (e.g., Johnson 1999; Payne and
Robb 2000; Barton and Simko
2002; Cheng and Warfield 2005), I also control for
characteristics of the analysts forecasts
including the number of analysts following a firm (Analysts),
variation among analysts forecasts(CVAF), and recent downward
revisions in analysts forecasts (Revise Down). Finally, Model
1includes calendar-quarter fixed effects and clusters standard
errors by firm (Petersen 2009). All
variables included in Model 1 are defined in either Table 1 or
Table 2.
Revenue Recognition and Earnings Informativeness
I use the following regression model to examine if earnings
informativeness differs among sell-
in, sell-through, and combination firms (H2):
URit b0 b1UEit b2SellThroughit b3Comboit b4UEit3 SellThroughit
b5UEit3Comboit bxControlsit byUEit3Controlsit bzTime Indicators
e:
2Prior accounting research models stock price as a function of
future dividends (which are assumed
to be related to future earnings), and derivations of this model
lead to an association between
unexpected stock returns and unexpected earnings (see e.g.,
Collins and Kothari 1989; Lev 1989;
Kothari 2001). The coefficient on unexpected earnings (i.e., the
earnings response coefficient) is
considered to be an indicator of earnings informativeness (e.g.,
Francis et al. 2006), and Model 2
examines whether the earnings response coefficient varies based
on firms revenue recognition
methods for sales to distributors.
Consistent with the recommendations of Berkman and Truong
(2009), I measure
unexpected returns (UR) as the cumulative market-adjusted stock
return for two trading daysbeginning on the quarterly earnings
announcement date. Unexpected earnings (UE) equals thedifference
between actual earnings per share and analysts last consensus
earnings forecast prior
to the quarterly earnings announcement (both reported by
I/B/E/S), scaled by stock price at
quarter-end. This definition of UE is similar to prior studies
(see e.g., Lopez and Rees 2002;Nelson et al. 2008; Wilson 2008;
Chen et al. 2011) and is consistent with MeetBeat in Model1.10
Sell-Through and Combo are as previously defined.
9 NOA should help to control for non-distributor accrual
manipulation. However, a measure of revenue attributableto
distributor customers further controls for any managerial
discretion involving non-distributor sales that affectsthe
likelihood of a firm meeting or beating analysts consensus earnings
forecast.
10 Prior studies calculate unexpected earnings as the difference
between actual earnings and either analysts consensusearnings
forecast or actual earnings from a prior period (see Lev [1989] and
Kothari [2001] for surveys of the literature).However, recent
research suggests that managers use earnings guidance to influence
analysts forecasts (e.g., Cotter et al.2006), which could affect
the UE measure used in my main test. As a sensitivity test, I use
an unexpected earningsmeasure based on the earnings time series,
and inferences are unchanged.
Revenue Recognition, Earnings Management, and Earnings
Informativeness 99
Accounting HorizonsMarch 2013
-
The main variables of interest in Model 2 are UE, UE3
Sell-Through, and UE3Combo. UErepresents the earnings response
coefficient (ERC) for sell-in firms. UE3 Sell-Through and UE3Combo
indicate the incremental ERC for sell-through and combination
firms, respectively,compared to sell-in firms. Positive and
significant coefficients for UE 3 Sell-Through and UE 3Combo will
indicate that earnings are more informative (i.e., ERCs are higher)
for sell-through andcombination firms compared to sell-in firms.
Negative and significant coefficients for the
interactions will indicate that earnings are less informative
for sell-through and combination firms
compared to sell-in firms.
Model 2 also controls for factors that are expected to affect
the relationship between unexpected
earnings and unexpected returns. Prior research finds that ERCs
are affected by firm size (Collins and
Kothari 1989), the market-to-book ratio (Collins and Kothari
1989), earnings persistence (Kormendi
and Lipe 1987), equity beta (Collins and Kothari 1989; Easton
and Zmijewski 1989), incidence of a
net loss (Hayn 1995), fourth-quarter earnings announcements
(Salamon and Stober 1994), and the
number of analysts following the firm (Shores 1990). Prior
research also suggests that managers are
incentivized to exercise accounting discretion when their
incentive compensation is based on earnings
performance (e.g., Bowen et al. 2003). Measures of executive
compensation can control for
managerial discretion (other than that exercised under the
revenue recognition method) that affects
ERCs. I also expect that ERCs are affected by a manufacturers
over or under reliance on distributors
compared to other customers. Thus, I include the following
control variables in the model: Rank ofSize, Rank of MTB, Persist,
Beta, Loss, Fourth Quarter, Analysts, Bonus%, Options%, Report,
andReport3 Avg Disty Revenue; and, I interact each of these control
variables with UE. All variablesincluded in Model 2 are defined in
Table 1, Table 2, or Table 3.
Sample
In order to develop my sample, I first obtain quarterly data for
all semiconductor firms (SIC
3674) in the Compustat Fundamentals Quarterly database during
20012008. I begin the sample
period in 2001 because SAB 101, which offered additional
guidance on revenue recognition
disclosures, became effective in that year. SAB 104 later
rescinded guidance in SAB 101 that was
superseded by the FASBs Emerging Issues Task Force (EITF) 0021,
but SAB 104 did not change
the revenue recognition principles in SAB 101 (SEC 1999, 2003).
Next, I hand collect all available
annual SEC filings during 20012008 for the semiconductor firms
with Compustat data. I exclude a
firm from the sample if it did not file at least one annual
report (10-K or 20-F) with the SEC during
the sample period, or if the annual SEC filings indicate that
the firm (1) is not a semiconductor
manufacturer, (2) does not sell products to distributors, (3)
has significant consignment agreements,
(4) generates more than half of its revenue from service
activities, (5) sells to retailers,11 (6) does
not offer product return privileges or pricing adjustments to
distributors,12 or (7) changed revenue
recognition methods during the sample period.13 I also exclude
firms lacking the Compustat, CRSP,
I/B/E/S, and ExecuComp data required for my analyses. The final
sample includes 80 unique
semiconductor firms with required data for 1,572
firm-quarters.
I classify the sample firms as sell-in, sell-through, or
combination based on the revenue
recognition disclosures presented in their annual SEC filings
(10-K or 20-F). Sell-in firms are those
firms that use the sell-in revenue recognition method for all
sales to distributors. Sell-through firms
11 I exclude firms with significant consignment agreements,
service revenue, and retail sales because revenuegenerated from
these activities could add noise to the empirical analyses.
12 This restriction ensures that all firms in my sample have
product return and pricing adjustment uncertaintiesrelated to sales
to distributors.
13 Fourteen firms appeared to switch revenue recognition methods
during the sample period. Because this subsamplewas so small, I was
unable to perform empirical tests that examined only these
firms.
100 Rasmussen
Accounting HorizonsMarch 2013
-
use the sell-through revenue recognition method for all sales to
distributors. Combination firms use
the sell-in revenue recognition method for some distributor
sales and the sell-through revenue
recognition method for other distributor sales.
EMPIRICAL EVIDENCE
Descriptive Statistics
Table 1, Panel A presents descriptive statistics for the full
sample of 1,572 firm-quarters, while
Panel B presents descriptive statistics for the sell-in,
combination, and sell-through subsamples.
Twenty percent of the observations represent use of the
sell-through method exclusively (Sell-Through), 48 percent of the
observations represent a combination of both the sell-in and
sell-through methods (Combo), and 32 percent of the observations
represent use of the sell-in methodexclusively. As expected,
deferred revenue is smallest for firms recognizing revenue for
all
distributors when products are delivered (sell-in) and largest
for firms deferring revenue recognition
for all distributors until the products have been resold
(sell-through). Mean (median) current
deferred revenue scaled by total assets at quarter-end (Deferred
Revenue) is 0.00 (0.00), 0.01 (0.00),and 0.02 (0.01) for sell-in,
combination, and sell-through firms, respectively, and the
means
(distributions) significantly differ at p , 0.000 for all three
types of firms. Actual reported revenueattributable to distributor
customers (Raw Disty Revenue) is available for two-thirds of the
firm-quarters, and significant differences exist among the firms
reporting this information. Raw DistyRevenue is highest, on
average, for sell-through firms (54 percent) followed by
combination firms(34 percent) and sell-in firms (30 percent).
Meanwhile, 82 percent of the firm-quarters represent
firms that report revenue attributable to distributor customers
at least once during the sample period
(Report). Mean Report 3 Avg Disty Revenue, the distributor
revenue variable included in theempirical tests, is 29 percent for
the full sample, 51 percent for sell-through firms, 25 percent
for
combination firms, and 23 percent for sell-in firms. Report 3
Avg Disty Revenue is significantlyhigher for sell-through firms
compared to both combination and sell-in firms.
With respect to the other variables included in the analyses,
firms meet or beat analysts
consensus earnings forecast (MeetBeat) in 79 percent of the full
sample quarters, 82 percent of thesell-in quarters, 77 percent of
the combination quarters, and 79 percent of the sell-through
firm-
quarters. Mean and median tests suggest that the incidence of
meeting or beating analysts
consensus earnings forecast significantly differs only between
sell-in and combination firms.
Meanwhile, mean unexpected stock returns (UR) are 0.002 for the
full sample, 0.002 for sell-infirms, 0.002 for combination firms,
and 0.007 for sell-through firms, but there is little evidence
that
these returns significantly differ among the three types of
firms. Table 1, Panel B also suggests that
many of the control variables used in the empirical analyses
significantly differ among sell-in,
combination, and sell-through firms.14
Revenue Recognition and Earnings Management
Table 2 presents estimation results for Model 1, with examines
the incidence of earnings
management for sell-in, sell-through, and combination firms. The
sample used to estimate this
model consists of the 1,572 firm-quarters during 20012008 with
required data for all analyses. The
Pseudo R2 is 10 percent and the area under the ROC curve is 72
percent, suggesting that the model
14 Recent studies examining the earnings-returns association
(Hirshleifer et al. 2009; Drake et al. 2012) use decileranks of
firm size and market-to-book measures in their empirical models
instead of raw values. I follow thisapproach and include Rank of
Size and Rank of MTB in Model 1 and Model 2. I also correct for
skewness ofShares by including the decile rank of this measure
(Rank of Shares) in Model 1.
Revenue Recognition, Earnings Management, and Earnings
Informativeness 101
Accounting HorizonsMarch 2013
-
TA
BL
E1
Des
crip
tiv
eS
tati
stic
s
Pa
nel
A:
Fu
llS
am
ple
Fu
llS
am
ple
(n
1,5
72
)
Mea
n2
5%
Qu
art
ile
Med
ian
75
%Q
ua
rtil
e
Sell
-Thr
ough
0.2
00
00
Com
bo0
.48
00
1
Def
erre
dR
even
ue0
.01
0.0
00
.00
0.0
1
Raw
Dis
tyR
even
ue0
.37
0.1
90
.36
0.5
3
Rep
ort
0.8
21
11
Rep
ort3
Avg
Dis
tyR
even
ue0
.29
0.1
10
.27
0.4
6
Ana
lyst
s1
25
10
17
Bet
a2
.66
1.6
92
.61
3.5
3
Bon
us%
0.1
00
.00
0.0
50
.16
CV
AF
0.0
30
.00
0.0
30
.10
Los
s0
.23
00
0
Mee
tBea
t0
.79
11
1
MT
B2
.93
1.4
32
.39
3.9
3
NO
A4
.26
2.6
33
.67
5.3
1
Opt
ions
%0
.48
0.0
70
.56
0.7
9
Per
sist
0.4
00
.16
0.4
00
.62
Rev
ise
Dow
n0
.82
11
1
Sale
sG
row
th0
.11
0.0
80
.08
0.2
6
Size
($M
)1
,82
0.4
12
28
.19
71
8.9
32
,12
2.7
1
Shar
es($
M)
18
6.9
83
0.9
68
7.2
92
55
.75
UE
0.0
00
50
.00
00
0.0
00
40
.00
20
UR
0.0
02
0.0
54
0.0
02
0.0
59
(con
tinu
edo
nn
ext
pa
ge)
102 Rasmussen
Accounting HorizonsMarch 2013
-
TA
BL
E1
(co
nti
nu
ed)
Pa
nel
B:
Sel
l-In
,C
om
bin
ati
on
,a
nd
Sel
l-T
hro
ug
hS
ub
sam
ple
s
Sel
l-In
(n
50
7)
Co
mb
o(n
75
0)
Sel
l-T
hro
ug
h(n
31
5)
Mea
n2
5%
Qrt
l.M
ed.
75
%Q
rtl.
Mea
n2
5%
Qrt
l.M
ed.
75
%Q
rtl.
Mea
n2
5%
Qrt
l.M
ed.
75
%Q
rtl.
Def
erre
dR
even
uea,b
,c,d
,e,f
0.0
00
.00
0.0
00
.00
0.0
10
.00
0.0
00
.02
0.0
20
.00
0.0
10
.04
Raw
Dis
tyR
even
uea,b
,c,d
,e,f
0.3
00
.15
0.2
90
.47
0.3
40
.17
0.3
20
.48
0.5
40
.30
0.5
50
.73
Rep
orta
,b,c
,d,e
,f0
.82
11
10
.75
11
10
.97
11
1
Rep
ort3
Avg
Dis
tyR
even
uec,d
,e,f
0.2
30
.03
0.2
30
.36
0.2
50
.10
0.2
50
.43
0.5
10
.32
0.5
10
.70
Ana
lyst
sc,d
,e,f
11
61
01
61
15
10
16
14
51
42
1
Bet
aa,b
,e,f
2.5
11
.53
2.4
83
.30
2.8
11
.86
2.7
53
.75
2.5
51
.48
2.5
73
.43
Bon
us%
a,b
,e,f
0.1
10
.00
0.0
70
.19
0.0
90
.00
0.0
40
.14
0.1
10
.01
0.0
70
.16
CV
AF
c,d
,e,f
0.0
10
.00
0.0
30
.09
0.0
40
.00
0.0
20
.11
0.0
70
.00
0.0
40
.09
Los
sc,d
,e,f
0.2
50
01
0.2
60
01
0.1
30
00
Mee
tBea
ta,b
0.8
21
11
0.7
71
11
0.7
91
11
MT
Ba,b
,c,d
,f2
.47
1.3
81
.93
3.1
13
.13
1.4
32
.50
4.1
63
.20
1.6
23
.29
4.4
6
NO
Aa,b
,c,d
4.5
03
.04
4.0
55
.37
4.2
22
.27
3.3
15
.56
3.9
82
.42
3.4
75
.05
Opt
ions
%a,b
,c,d
0.4
50
.12
0.5
00
.70
0.5
00
.10
0.5
60
.80
0.5
10
.00
0.6
50
.82
Per
sist
b,c
,d,e
,f0
.36
0.1
00
.33
0.5
90
.38
0.1
70
.38
0.5
80
.51
0.2
70
.49
0.7
6
Rev
ise
Dow
na,b
,e,f
0.8
41
11
0.7
91
11
0.8
61
11
Sale
sG
row
tha,e
0.0
90
.09
0.0
90
.26
0.1
30
.09
0.0
80
.30
0.0
80
.04
0.0
90
.21
Size
($M
)a,c
,d,f
2,2
85
.03
21
1.0
46
81
.85
2,0
12
.20
1,5
77
.92
20
7.9
26
48
.15
2,0
88
.11
1,6
49
.93
31
3.0
81
,45
8.7
82
,51
2.3
1
Shar
es($
M)a
,b,c
,d,f
21
5.9
63
0.6
37
2.7
51
52
.55
16
8.6
93
5.8
99
1.3
42
91
.89
18
3.8
62
9.8
61
75
.79
33
5.8
5
UE
b,d
,e0
.00
08
0.0
00
00
.00
06
0.0
02
10
.00
01
0.0
00
00
.00
03
0.0
02
10
.00
09
0.0
00
00
.00
03
0.0
01
3
UR
d0
.00
20
.06
00
.00
50
.05
90
.00
20
.05
80
.00
40
.06
10
.00
70
.03
80
.00
80
.05
1
Pan
elA
pre
sents
des
crip
tive
stat
isti
csfo
ra
sam
ple
of
1,5
72
quar
terl
yobse
rvat
ions
for
80
uniq
ue
sem
iconduct
or
firm
sduri
ng
20012008.
Pan
elB
pre
sents
des
crip
tive
stat
isti
csfo
rth
ese
ll-i
n,
com
bin
atio
n,
and
sell
-thro
ugh
subsa
mple
s.M
ean
(med
ian)
dif
fere
nce
sat
p,
0.1
0usi
ng
atw
o-t
aile
dt-
(Wil
coxon
Sum
-Ran
k)
test
are
den
ote
dby
a(b)
for
sell
-in
firm
sver
sus
com
bin
atio
nfi
rms,
c(d)
for
sell
-in
firm
sver
sus
sell
-thro
ugh
firm
s,an
de(
f)fo
rco
mbin
atio
nfi
rms
ver
sus
sell
-thro
ugh
firm
s.A
llco
nti
nuous
var
iable
sar
ew
inso
rize
dat
the
1st
and
99th
per
centi
les.
(con
tinu
edo
nn
ext
pa
ge)
Revenue Recognition, Earnings Management, and Earnings
Informativeness 103
Accounting HorizonsMarch 2013
-
TA
BL
E1
(co
nti
nu
ed)
Var
iable
Defi
nit
ions:
Sell
-Thr
ough
in
dic
ator
var
iable
equal
to1
ifth
efi
rmuse
sth
ese
ll-t
hro
ugh
reven
ue
reco
gnit
ion
met
hod
for
sale
sto
dis
trib
uto
rs,
and
0oth
erw
ise;
Com
bo
indic
ator
var
iable
equal
to1
ifth
efi
rmuse
sth
eco
mbin
atio
nre
ven
ue
reco
gnit
ion
met
hods
for
sale
sto
dis
trib
uto
rs,
and
0oth
erw
ise;
Def
erre
dR
even
ue
curr
ent
def
erre
dre
ven
ue
scal
edby
tota
las
sets
atquar
ter-
end;
Raw
Dis
tyR
even
ue
per
centa
ge
of
annual
reven
ue
attr
ibuta
ble
todis
trib
uto
rcu
stom
ers
and
isas
signed
toal
lof
the
firm
squar
terl
yobse
rvat
ions
duri
ng
the
fisc
alyea
r(t
hes
edat
aar
eav
aila
ble
for
1,0
58
of
the
sam
ple
firm
-quar
ters
);R
epor
t
indic
ator
set
var
iable
set
to1
ifth
efi
rmre
port
san
nual
reven
ue
attr
ibuta
ble
todis
trib
uto
rcu
stom
ers
atle
ast
once
duri
ng
the
sam
ple
per
iod,
and
0oth
erw
ise;
Rep
ort3
Avg
Dis
tyR
even
ue
aver
age
of
all
annual
reven
ue
attr
ibuta
ble
todis
trib
uto
rcu
stom
ers
report
edby
the
firm
duri
ng
the
sam
ple
per
iod
and
isas
signed
toal
lof
the
firm
squar
terl
yobse
rvat
ions,
or
isse
tto
0fo
rfi
rms
that
do
not
report
this
info
rmat
ion
inan
yof
the
sam
ple
yea
rs;
Ana
lyst
s
num
ber
of
anal
yst
earn
ings
fore
cast
sin
cluded
inth
ela
stI/
B/E
/Sco
nse
nsu
sfo
reca
stpri
or
toth
equar
terl
yea
rnin
gs
announce
men
t;B
eta
syst
emat
icri
skfr
om
the
mar
ket
model
for
the
twel
ve-
month
per
iod
endin
gbef
ore
the
star
tof
the
quar
ter;
Bon
us%
CE
Os
cash
-bas
edin
centi
ve
com
pen
sati
on
scal
edby
the
CE
Os
tota
lan
nual
com
pen
sati
on
and
isas
signed
toal
lof
the
firm
squar
terl
yobse
rvat
ions
duri
ng
the
fisc
alyea
r;C
VA
F
coef
fici
ent
of
var
iati
on
(the
stan
dar
ddev
iati
on
scal
edby
the
mea
n)
of
the
last
I/B
/E/S
conse
nsu
sea
rnin
gs
fore
cast
pri
or
toth
equar
terl
yea
rnin
gs
announce
men
t;L
oss
indic
ator
var
iable
equal
to1
ifac
tual
quar
terl
yea
rnin
gs
report
edby
I/B
/E/S
are
neg
ativ
e,an
d0
oth
erw
ise;
Mee
tBea
tin
dic
ator
var
iable
equal
to1
ifth
efi
rmm
eets
the
last
I/B
/E/S
conse
nsu
sea
rnin
gs
fore
cast
pri
or
toth
equar
terl
yea
rnin
gs
announce
men
tdat
eor
bea
tsit
by
any
amount,
and
0oth
erw
ise;
MT
B
mar
ket
val
ue
of
equit
yat
quar
ter-
end
scal
edby
book
val
ue
of
equit
yat
quar
ter-
end;
NO
A
net
oper
atin
gas
sets
atth
een
dof
the
pri
or
quar
ter,
scal
edby
sale
sfo
rth
epri
or
quar
ter;
Opt
ions
%
doll
arval
ue
of
new
opti
ons
gra
nte
dto
the
CE
Oduri
ng
the
yea
rsc
aled
by
the
CE
Os
tota
lan
nual
com
pen
sati
on
and
isas
signed
toal
lof
the
firm
squar
terl
yobse
rvat
ions
duri
ng
the
fisc
alyea
r.(I
foll
ow
the
appro
ach
use
dby
Chen
gan
dF
arber
[2008]
and
Efe
ndi
etal
.[2
013]
tocr
eate
this
var
iable
);P
ersi
st
firs
t-ord
erau
toco
rrel
atio
nco
effi
cien
tof
quar
terl
yea
rnin
gs
esti
mat
edover
the
pas
tfo
ur
yea
rs;
Rev
ise
Dow
n
indic
ator
var
iable
equal
to1
ifat
leas
tone
anal
yst
revis
eddow
nhis
/her
earn
ings
fore
cast
for
the
firm
duri
ng
the
per
iod
bet
wee
nth
epri
or
quar
ter
sea
rnin
gs
announce
men
tan
dth
ecu
rren
tquar
ter
sea
rnin
gs
announce
men
t,an
d0
oth
erw
ise;
Sale
sG
row
th
sale
sfo
rth
equar
ter
div
ided
by
sale
sfr
om
the
sam
equar
ter
inth
epri
or
yea
r,le
ssone;
Size
tota
las
sets
atquar
ter-
end;
Shar
es
aver
age
num
ber
of
com
mon
shar
esouts
tandin
gduri
ng
the
quar
ter;
UE
dif
fere
nce
bet
wee
nac
tual
earn
ings
per
shar
ean
dth
ela
stco
nse
nsu
sea
rnin
gs
fore
cast
pri
or
toth
equar
terl
yea
rnin
gs
announce
men
tdat
e(b
oth
report
edby
I/B
/E/S
),sc
aled
by
stock
pri
ceat
quar
ter-
end
(rep
ort
edby
CR
SP
);an
dU
R
cum
ula
tive
stock
retu
rnfo
rtw
otr
adin
gday
sbeg
innin
gon
the
quar
terl
yea
rnin
gs
announce
men
tdat
e,ad
just
edby
the
val
ue-
wei
ghte
dm
arket
index
.
104 Rasmussen
Accounting HorizonsMarch 2013
-
has acceptable predictive power of the incidence of meeting or
beating analysts consensus earnings
forecast (Hosmer and Lemeshow 2000). Although the model controls
for calendar-quarter fixed
effects, I do not tabulate the fixed effects coefficients for
parsimony.
Consistent with H1a and H1b, the Sell-Through and Combo
coefficients are negative andsignificant (a1 0.457, p 0.067; a2
0.493, p 0.015, respectively), suggesting that sell-through and
combination firms are both less likely to meet or beat analysts
consensus earnings
forecast than sell-in firms. Since combination firms have more
opportunity to manage earnings
compared to sell-through firms, it is surprising to find that
the Combo coefficient appears larger inmagnitude and has a stronger
statistical significance compared to the Sell-Through
coefficient.However, an untabulated v2 test indicates that the
Sell-Through and Combo coefficients do notsignificantly differ. In
order to quantify the impact of the revenue recognition method on
the
likelihood of a firm meeting or beating analysts consensus
earnings forecast, I calculate average
partial effects (untabulated).15 The average partial effect of
the sell-through (combination) method
decreases the likelihood of a firm meeting or beating analysts
consensus earnings forecast by 6.9
(7.0) percent, and this decrease is significant at the p0.041
(0.017) level.16 Since sell-through andcombination firms are less
likely to meet or beat analysts consensus earnings forecast (the
proxy
for earnings management), these findings suggest that sell-in
firms exercise discretion under their
revenue recognition method to manage earnings. With respect to
control variables, Sales Growth,Rank of MTB, Bonus%, and Analysts
are positively associated with MeetBeat, while Revise Down
isnegatively associated with MeetBeat.
In an untabulated test, I use the Heckman procedure (Heckman
1979; Wooldridge 2002) to
control for the possibility that determinants of the revenue
recognition method are correlated with
the likelihood that a firm meets or beats analysts consensus
earnings forecast. I estimate two
selection equations, each predicting use of either the
sell-through or the combination method. Each
selection equation includes all control variables from Model 1
plus two instrument variables: the
decile rank of firm age and an indicator representing use of an
industry specialist auditor (auditor
with the highest market share in the semiconductor industry).17
I then calculate the inverse Mills
ratios from the selection equations and include these inverse
Mills ratios in the outcome equation
predicting the likelihood of meeting or beating analysts
consensus earnings forecast. Sell-Throughand Combo remain
negatively and significantly associated with MeetBeat in the
outcome equation(results untabulated), consistent with Table
2.18
15 Average partial effects equal the sample average of marginal
effects computed for each observation (Wooldridge2002, 2224). Since
Sell-Through and Combo are indicator variables representing
different categories of a singleunderlying variable (revenue
recognition), I follow Bartus (2005) and restrict the observations
used to computeaverage partial effects to the category of interest
and the reference group. When calculating the average partialeffect
of Sell-Through, observations are restricted to sell-through and
sell-in firms. Similarly, when calculatingthe average partial
effect of Combo, observations are restricted to combination and
sell-in firms.
16 The average partial effect of Sell-Through (Combo) decreases
the likelihood of a firm meeting or beating analystsconsensus
earnings forecast from 83.3 (82.7) percent to 76.4 (75.8) percent.
These likelihoods can be comparedto the 78.8 percent unconditional
mean for the full sample (1,239 firm-quarter observations meeting
or beatinganalysts consensus earnings forecast versus 1,572 sample
observations).
17 Heckmans procedure requires at least one instrument variable
in the selection model. Untabulated results of thelogistic
regression selection models indicate that the decile rank of firm
age is negatively and significantlyassociated with Sell-Through (p
, 0.000). Use of the industry specialist auditor is positively and
significantlyassociated with Combo (p , 0.000).
18 Correlation tests indicate that the inverse Mills ratios
calculated from the two selection equations have a strongnegative
correlation (0.575, p , 0.000). As an alternative to including both
inverse Mills ratios in the outcomeequation, I estimate two outcome
equations. The first outcome equation is estimated using the
subsample of sell-in and sell-through firms and includes
Sell-Through and the inverse Mills ratio calculated when predicting
Sell-Through. The second outcome equation is estimated using the
subsample of sell-in and combination firms andincludes Combo and
the inverse Mills ratio calculated when predicting Combo. The
coefficient of interest (Sell-Through or Combo), is negatively and
significantly associated with MeetBeat in each of these outcome
equations.
Revenue Recognition, Earnings Management, and Earnings
Informativeness 105
Accounting HorizonsMarch 2013
-
Collectively, the results presented in Table 2 and the
additional analysis indicate that earnings
management to a benchmark does differ based on firms revenue
recognition methods for sales to
distributors. Specifically, the results suggest that earnings
management is more likely for firms that
recognize all revenue before product return and pricing
adjustment uncertainties are resolved. This
finding implies that earnings management concerns about the
sell-in method are warranted.
Revenue Recognition and Earnings Informativeness
Table 3 presents estimation results for Model 2, which examines
the ERCs of firms with
different revenue recognition methods. The sample used to
estimate this model consists of the 1,572
firm-quarters during 20012008 with required data for all
analyses. The model specification
explains 3.5 percent of unexpected returns, which is consistent
with prior studies that examine the
earnings-returns association (Lev 1989). For parsimony, I do not
tabulate the coefficients for the
control variables or their interactions with UE.Unexpected
earnings (UE) are positively and significantly associated with
unexpected returns
(UR) (b1 4.371; p 0.003), indicating a positive ERC for sell-in
firms. The main effects of Sell-Through and Combo and the UE3 Combo
coefficient are insignificant. However, the UE3 Sell-Through
coefficient is positive and significant (b4 2.302; p 0.065).
Because UE3 Sell-Throughindicates the incremental ERC for
sell-through firms compared to sell-in firms, this finding
suggests
TABLE 2
The Association between Earnings Management, Revenue Recognition
Methods for Sales toDistributors, and Control Variables
Variable Prediction Coeff. p-value
Intercept / 0.334 (0.559)Sell-Through 0.457 (0.067)Combo 0.493
(0.015)Sales Growth 0.459 (0.042)Rank of MTB 1.018 (0.004)NOA 0.010
(0.393)Rank of Shares / 0.055 (0.945)Rank of Size / 0.241
(0.703)Report / 0.449 (0.202)Report 3 Avg Disty Revenue / 0.624
(0.311)Bonus% 1.716 (0.024)Options% 0.075 (0.395)Analysts 0.044
(0.033)CVAF 0.043 (0.428)Revise Down / 0.754
(0.001)Calendar-quarter fixed effects Yes
n 1,572
Pseudo R2 0.10
Area under ROC curve 0.72
This table presents the results of Model 1 where the dependent
variable is MeetBeat. Variables are defined in Table 1with the
following exceptions. Rank of MTB, Rank of Shares, and Rank of Size
are decile ranks of MTB, Shares, and Size,scaled to range between 0
and 1. All standard errors are clustered by firm (Petersen 2009).
Bold coefficients and p-valuesindicate statistical significance at
the 0.10 level or less. One-tailed tests are used when a direction
is predicted, and two-tailed tests are used when there is no
prediction.
106 Rasmussen
Accounting HorizonsMarch 2013
-
that the unexpected earnings of sell-through firms are more
informative (have a stronger association
with unexpected stock returns) than the unexpected earnings of
sell-in firms. In addition, the
insignificant UE 3 Combo coefficient indicates that sell-through
firms also have a significantlyhigher ERC compared to combination
firms. Untabulated coefficients for the interactions between
UE and control variables indicate that ERCs are lower during the
firms fourth quarter and when the
firm reports a loss. Meanwhile, ERCs are higher as analyst
following increases.
I perform a variety of untabulated tests to assess the
robustness of the results presented in Table
3. First, I use the Heckman procedure (Heckman 1979; Wooldridge
2002) to control for potential
endogeneity of the revenue recognition method. The selection
equations predict use of the sell-
through and combination methods and include all controls from
Model 2 plus the decile rank of
firm age and an indicator representing use of the industry
specialist auditor.19 Inverse Mills ratios
generated from the selection equations are included in the
outcome equation.20 Second, instead of
measuring unexpected returns over a short window surrounding the
earnings announcement, I
measure UR over the window starting two days after the prior
quarters earnings announcement
date and ending one day after the current quarters earnings
announcement date. Since many of the
sample firm-quarter return windows overlap in this untabulated
test, I include calendar-quarter fixed
effects. Third, I use an alternate measure of UE based on the
earnings time series. Specifically, I
calculate UE as the difference between actual earnings per share
for the current quarter and actual
earnings per share for the same quarter in the prior year (both
reported by I/B/E/S), scaled by stock
price at quarter-end. Untabulated results for all of these
independent tests indicate that UE3 Sell-Through is positively and
significantly associated with unexpected returns while the UE3
Combocoefficient is insignificantly different from zero.21
Finally, I examine the effect of the revenue recognition method
on the association between
unexpected returns and unexpected gross margin. For this
analysis, unexpected gross margin
(UGM) equals the difference between actual gross margin
percentage and analysts last consensus
gross margin percentage forecast prior to the quarterly earnings
announcement (both reported by I/
B/E/S), scaled by analysts last consensus gross margin
percentage forecast.22 There are 411 firm-
quarter observations during 20062008 with data available for
UGM. When UGM replaces UE in
Model 2 and is interacted with the revenue recognition
indicators and control variables, the UGM3Sell-Through coefficient
is positive and significant (p 0.004) and UGM 3 Combo
isinsignificantly different from zero.
In sum, the results presented in Table 3 and the additional
analyses suggest that unexpected
earnings are more strongly associated with unexpected returns
for sell-through firms compared to
both sell-in and combination firms. Stated differently, the
results suggest that earnings are more
informative for firms that defer revenue recognition until all
product return and pricing adjustment
uncertainties are resolved.
19 Untabulated results of the logistic selection models indicate
that the decile rank of firm age is negatively andsignificantly
associated with Sell-Through (p , 0.000) and use of the industry
specialist auditor is positively andsignificantly associated with
Sell-Through and Combo (p 0.10 and p , 0.000, respectively).
20 As with the MeetBeat analysis, correlation tests indicate
that the inverse Mills ratios calculated from the twoselection
equations have a strong negative correlation (0.543, p , 0.000).
Untabulated results indicate thatinferences with respect to
UE3Sell-Through and UE3Combo are the same for a full-sample
specification of theoutcome equation as well as separate outcome
equations estimated using either the sell-through/sell-in
orcombination/sell-in subsample of firms.
21 The statistical significance of the UE3 Sell-Through
coefficient using a two-tailed test is marginal (0.11) for
thequarterly returns test and less than 0.10 for all other
untabulated tests.
22 Since gross margin forecasts are percentages, I use analysts
last consensus gross margin percentage forecast asthe scalar
instead of the stock price at quarter-end.
Revenue Recognition, Earnings Management, and Earnings
Informativeness 107
Accounting HorizonsMarch 2013
-
SUMMARY AND CONCLUSIONS
Although revenue is one of the most important figures reported
on the income statement, little
research exists on specific revenue recognition methods and
their implications for firms. This study
examines the revenue recognition methods of semiconductor firms
and their implications for
earnings management and earnings informativeness. Semiconductor
firms sell a significant amount
of product to distributors and face product return and pricing
adjustment uncertainties until the
distributors resell the product to end-customers. I find that
firms deferring revenue recognition until
product return and price adjustment uncertainties are partly or
fully resolved are less likely to meet
or beat analysts consensus earnings forecast than firms
immediately recognizing revenue for sales
to distributors. This finding suggests that earnings management
is more likely when firms recognize
revenues before all uncertainties are resolved. I also find that
earnings are more informative (the
earnings-returns association is stronger) for firms that defer
revenue recognition until products are
resold to end-customers (i.e., all product return and pricing
adjustment uncertainties have been
resolved).
This study extends a growing stream of research that examines
the implications of revenue
recognition for firms in different industries (Bowen et al.
2002; Zhang 2005; Srivastava 2011) and
can help students, practitioners, and other financial statement
users better understand revenue
recognition methods and their associations with earnings
management and earnings informative-
ness. This study is also potentially useful for investors,
analysts, and auditors who monitor
high-technology manufacturers because the results imply that
earnings management concerns about
the sell-in method are warranted and earnings are less
informative when revenue is immediately
recognized for sales to at least some distributors. This study
is also potentially informative for
regulators and standard-setters because the findings suggest
that manufacturers with significant
product return and pricing adjustment uncertainties should only
recognize revenue from sales to
distributors once all of the uncertainties are resolved.
TABLE 3
The Association between Unexpected Returns, Unexpected Earnings,
Revenue RecognitionMethods for Sales to Distributors, and Control
Variables
Variable Prediction Coeff. p-value
Intercept / 0.025 (0.012)UE 4.371 (0.003)Sell-Through / 0.004
(0.639)Combo / 0.007 (0.168)UE 3 Sell-Through / 2.302 (0.065)UE
3Combo / 0.264 (0.714)Controls YesUE 3Controls Yes
n 1,572
Adjusted R2 0.035
This table presents the results of Model 2 where the dependent
variable is UR. The control variables include: Rank ofSize, Rank of
MTB, Persist, Beta, Loss, Fourth Qtr, Analysts, Bonus%, Options%,
Report, and Report 3 Avg DistyRevenue. Variables are defined in
Table 1 and Table 2 with the following exception. Fourth Qtr is an
indicator variableequal to 1 if the firm is announcing
fourth-quarter earnings, and 0 otherwise. Bold coefficients and
p-values indicatestatistical significance at the 0.10 level or
less. One-tailed tests are used when a direction is predicted, and
two-tailed testsare used when there is no prediction.
108 Rasmussen
Accounting HorizonsMarch 2013
-
This study is subject to limitations. First, because I limit my
sample to semiconductor firms, I
have not examined if the findings generalize to all industries
and revenue recognition methods.
Second, although I have attempted to control for the
characteristics of firms with different revenue
recognition methods in the empirical tests, it is possible that
the results reflect inherent differences
among sell-in, sell-through, and combination firms. Even with
these limitations, the study provides
insight into the observed effects and improves our understanding
of firms with different revenue
recognition methods.
REFERENCES
Aboody, D., R. Kasznik, and M. Williams. 2000. Purchase versus
pooling in stock-for-stock acquisitions:
Why do firms care? Journal of Accounting and Economics 29 (3):
261286.Aboody, D., M. E. Barth, and R. Kasznik. 2004. Firms
voluntary recognition of stock-based compensation
expense. Journal of Accounting Research 42 (2): 123150.American
Institute of Certified Public Accountants (AICPA). 1997. Statement
of Position 97-2: Software
Revenue Recognition. New York, NY: AICPA.Altamuro, J., A. L.
Beatty, and J. Weber. 2005. The effects of accelerated revenue
recognition on earnings
management and earnings informativeness: Evidence from SEC Staff
Accounting Bulletin No. 101.
The Accounting Review 80 (2): 373401.Barton, J., and P. J.
Simko. 2002. The balance sheet as an earnings management
constraint. The Accounting
Review 77 (1): 127.Bartus, T. 2005. Estimation of marginal
effects using margeff. The Stata Journal 5 (3): 309329.Beneish, M.
D. 2001. Earnings management: A perspective. Managerial Finance 27
(12): 317.Bergstresser, D., and T. Philippon. 2006. CEO incentives
and earnings management. Journal of Financial
Economics 80 (3): 511529.Berkman, H., and C. Truong. 2009. Event
day 0? After-hours earnings announcements. Journal of
Accounting Research 47 (1): 71103.Bowen, R. M., A. K. Davis, and
S. Rajgopal. 2002. Determinants of revenue-reporting practices for
internet
firms. Contemporary Accounting Research 19 (4): 523562.Bowen, R.
M., S. Rajgopal, and M. Venkatachalam. 2003. Accounting discretion,
corporate governance,
and firm performance. Contemporary Accounting Research 25 (2):
351405.Chen, S., D. Matsumoto, and S. Rajgopal. 2011. Is silence
golden? An empirical analysis of firms that stop
giving quarterly earnings guidance. Journal of Accounting and
Economics 51: 134150.Cheng, Q., and T. D. Warfield. 2005. Equity
incentives and earnings management. The Accounting Review
80 (2): 441476.
Cheng, Q., and D. B. Farber. 2008. Earnings restatements,
changes in CEO compensation, and firm
performance. The Accounting Review 83 (5): 12171250.Chipalkatti,
N., S. Chatterji, and S. Bee. 2007. Effective controls for sales
through distribution channels.
The CPA Journal 77 (9): 6066.Collins, D., and S. Kothari. 1989.
An analysis of intertemporal and cross-sectional determinants of
earnings
response coefficients. Journal of Accounting and Economics 11:
143181.Cornett, M. M., A. J. Marcus, and H. Tehranian. 2008.
Corporate governance and pay-for-performance: The
impact of earnings management. Journal of Financial Economics 87
(2): 357373.Cotter, J., I. Tuna, and P. D. Wysocki. 2006.
Expectations management and beatable targets: How do
analysts react to explicit earnings guidance? Contemporary
Accounting Research 23 (3): 593624.Credit Suisse. 2007. Component
Distributors: ARW and AVT. November 26. New York, NY: Credit
Suisse.Das, S., C. B. Levine, and K. Sivaramakrishnan. 1998.
Earnings predictability and bias in analysts earnings
forecasts. The Accounting Review 73 (2): 277294.Degeorge, F., J.
Patel, and R. Zeckhauser. 1999. Earnings management to exceed
thresholds. The Journal of
Business 72 (1): 133.Drake, M. S., D. T. Roulstone, and J. R.
Thornock. 2012. Investor information demand: Evidence from
Google searches around earnings announcements. Journal of
Accounting Research 50 (4): 10011040.
Revenue Recognition, Earnings Management, and Earnings
Informativeness 109
Accounting HorizonsMarch 2013
-
Easton, P., and M. Zmijewski. 1989. Cross-sectional variation in
the stock market response to accounting
earnings announcements. Journal of Accounting and Economics 11:
117141.
Efendi, J., A. Srivastava, and E. P. Swanson. 2007. Why do
corporate managers misstate financial
statements? The role of option compensation and other factors.
Journal of Financial Economics 85(3): 667708.
Efendi, J., R. Files, B. Ouyang, and E. P. Swanson. 2013.
Executive turnover following option backdating
allegations. The Accounting Review 88 (1): 75105.
Fields, T. D., T. Z. Lys, and L. Vincent. 2001. Empirical
research on accounting choice. Journal ofAccounting and Economics
31: 255307.
Financial Accounting Standards Board (FASB). 1980. Qualitative
Characteristics of AccountingInformation (As Amended). Statement of
Financial Accounting Concepts No. 2. Norwalk, CT: FASB.
Financial Accounting Standards Board (FASB). 1981. Revenue
Recognition When Right of Return Exists.Statement of Financial
Accounting Standards No. 48. Norwalk, CT: FASB.
Financial Accounting Standards Board (FASB). 2005. Financial
Accounting Standards Advisory CouncilMeeting: Revenue Recognition
Project. Notes for June 21, 2005. Norwalk, CT: FASB.
Forester, C. 2008. Does More Conservative Revenue Recognition
Improve the Informativeness of Earnings?
Working paper, University of Minnesota.
Francis, J., P. Olsson, and K. Schipper. 2006. Earnings quality.
Foundations and Trends in Accounting 1(4): 259340.
Glass, Lewis & Co. 2004. Revenue Recognition Policies,
Practices, and DisclosuresPharmaceuticalIndustry. Yellow Card
Industry Alert, May 5. New York, NY: Glass, Lewis & Co.
Graham, J. R., C. R. Harvey, and S. Rajgopal. 2005. The economic
implications of corporate financial
reporting. Journal of Accounting and Economics 40: 373.
Greenberg, H. 2006. A shift to sell-in accounting could be clue
to brewing trouble. The Wall StreetJournal, June 17.
Hayn, C. 1995. The information content of losses. Journal of
Accounting and Economics 20: 125153.
Healy, P. M., and J. M. Wahlen. 1999. A review of the earnings
management literature and its implications
for standard setting. Accounting Horizons 13 (4): 365383.
Heckman, J. J. 1979. Sample selection bias as a specification
error. Econometrica 47 (1): 153161.
Hirshleifer, D., S. Lim, and S. Teoh. 2009. Driven to
distraction: Extraneous events and under reaction to
earnings news. Journal Finance 64 (5): 22892325.
Hosmer, D. W., and S. Lemeshow. 2000. Applied Logistic
Regression. 2nd Ed. Upper Saddle River, NJ:Prentice Hall.
Johnson, N. S. 1999. Current SEC DevelopmentsManaged Earnings
and the Year of the Accountant.Remarks delivered at the Utah State
Bar Mid-Year Convention, St. George, UT, March 6.
Kormendi, R., and R. Lipe. 1987. Earnings innovations, earnings
persistence, and stock returns. TheJournal of Business 60 (3):
323345.
Kothari, S. P. 2001. Capital markets research in accounting.
Journal of Accounting and Economics 31: 105231.
KPMG LLP. 2006. Improved point-of-sale reporting benefits
manufacturers, channels, and end users: A
study of the high-tech indu