Cost Behavior in a Period of Economic Crisis and the Effect of the

Cost Behavior in a Period of Economic Crisis
and the Effect of the Frequency of Updating
Information on Cost Behavior
Wouter van Zuijlen
ANR: 959484
Master Thesis
Program: Accounting
Track: Accountancy
Supervisor: Dr. B. Dierynck
Second reader: Dr. K. Law
Defense date: 18-09-2012
- 2012-
Cost Behavior in a Period of Economic Crisis and
the Effect of the Frequency of Updating Information on Cost Behavior
Abstract
In this study I examine the behavior of SG&A costs for a large sample of firms from 21 countries.
First, I hypothesize that firms in a crisis period exhibit more symmetric cost behavior compared to
firms in a non-crisis period. My results show that both in times of crisis and in non-crisis times, firms
exhibit significant asymmetrical SG&A cost behavior. Second, I hypothesize that firms that update
their internal information more frequent exhibit more symmetric cost behavior. My results show
that, using Compustat data for the interim reporting frequency, firms that update their internal
information less frequent exhibit more symmetrical SG&A cost behavior. Using the method of Fu for
the interim reporting frequency, firms that update their internal information more frequent exhibit
the same asymmetrical SG&A cost behavior as firms that update internal information less frequent.
Keywords: Cost asymmetry; Selling, General, and Administrative Costs (SG&A costs); economic crisis;
updating information.
Data availability: Data is available from the sources mentioned in the text.
2
I. INTRODUCTION
The traditional cost model assumes that variable costs change proportionally and symmetrically with
changes in activity of a firm. Many studies have examined the empirical validity of this key
assumption underlying the traditional cost model (including Noreen and Soderstrom 1997; Anderson
et al. 2003; Balakrishnan et al. 2004; Chen et al. 2012; Dierynck et al. 2012). All of these studies found
that costs behave asymmetrically instead of symmetrically. Asymmetric cost behavior means that
costs increase more with an activity increase than they decrease with an activity decrease. Anderson
et al. (2003) find that an increase of one percent in sales results on average in a 0.55 percent increase
of Selling, General and Administrative (SG&A) costs, but a decrease of one percent in sales results on
average in a 0.35 percent decrease of SG&A costs. They label this type of cost behavior as “sticky”.
In this study I examine two topics. First, I will examine whether SG&A costs of firms behave
more symmetric in a period of economic crisis compared to a non-crisis period. Second, I examine
the effect of the frequency of updating information on cost asymmetry. Mankiw and Reis (2002)
demonstrate analytically that firms - on aggregate conditions - infrequently update their information,
because it is costly to acquire, absorb, and process information. I expect that the degree of cost
stickiness is determined by the frequency by which the firm updates its financial information.
I examine my two research questions by using a sample of 77,176 firm-year observations
over the period 2001-2010. This sample is constructed by using firms from 21 countries all over the
world. Controlling for economic determinants of cost asymmetry as identified in prior research
(Anderson et al. 2003; Pinnuck and Lillis 2007; Dierynck et al. 2012), I find that firms, on average,
exhibit significant asymmetrical SG&A cost behavior. Specifically, SG&A costs increase with 0.59
percent following a one percent increase in activity and decrease with 0.37 percent following a one
percent decrease in activity.
My results further show that there is no consistent support for the hypothesis that the crisis
leads to more symmetrical SG&A cost behavior. Specifically, both in times of crisis and in non-crisis
periods, significant asymmetrical cost behavior is detected. Next, my results do not provide any
3
support for the hypothesis that firms that update their internal information more frequently exhibit
symmetric cost behavior. Using Compustat interim reporting frequency data, I show that firms
updating their internal information more frequently exhibit more asymmetric SG&A cost behavior.
Using interim reporting frequency data following the method of Fu (2012), I show that firms updating
their internal information more frequently exhibit the same asymmetrical SG&A cost behavior as
firms updating their information less frequent. So, both results are in contrast with my expectations.
However, these results do not seem to be entirely robust. Controlling for country effects
leads in multiple cases to different results compared to the main results, which indicates that the
United States is a very dominant country in obtaining the main results.
This study contributes to the existing literature in two ways. First, I test cost stickiness on a
large dataset containing observations from countries spread all over the world, both for a period of
economic crisis and a non-crisis period. Second, this study also provides some first large-sample
evidence with regard to the effect of updating information on cost asymmetry, although the
evidence is seriously limited.
The remainder of this study paper is organized as follows. Section 2 gives an overview of
prior literature, while in section 3 the hypotheses are developed. Section 4 describes the data
collection and the research methods used. Section 5 presents the findings and section 6 contains the
conclusions of this study.
II. LITERATURE OVERVIEW
This section provides an overview of all relevant prior literature related to the definition of cost
stickiness, to the reasons for cost stickiness and to the determinants of cost stickiness.
Cost Stickiness
Anderson et al. (2003) mention that the traditional cost model assumes that variable costs change
proportionally and symmetrically with changes in the activity driver. Many studies have examined
4
the empirical validity of this key assumption underlying the traditional cost model (including Noreen
and Soderstrom 1997; Anderson et al. 2003; Balakrishnan et al. 2004; Chen et al. 2012; Dierynck et
al. 2012). All these studies found that costs behave asymmetrically instead of symmetrically.
Asymmetric cost behavior means that costs increase more with an activity increase than they
decrease with an activity decrease. For instance, if the activity increases with 10% then costs increase
with 6.5%, while an activity decrease of 10% leads to a cost decrease of only 3.5%. Anderson et al.
(2003) label this type of cost behavior as “sticky”.
Prior research focused on behavior of different types of costs. In the study of Noreen and
Soderstrom (1997) the overhead costs of service departments of hospitals are investigated. Dierynck
et al. (2012) focus on labor costs in Belgian firms. Although several type of costs are examined, most
prior research focuses on Selling, General and Administrative (SG&A) costs when determining the
stickiness of costs (Anderson et al. 2003; Medeiros and Costa 2004; Chen et al. 2012). Banker and
Chen (2006) and Calleja et al. (2006) zoomed in on the behavior of operating costs, which consists of
SG&A costs and cost of goods sold.
There are several reasons for using SG&A costs as a proxy for costs. First of all, sales volume
drives many of the components of SG&A costs (Cooper and Kaplan 1998, 341). A second argument is
that SG&A costs are a significant part of sales revenue. For instance in the study of Anderson et al.
(2003), SG&A costs make up 26.4% of sales revenue. A final argument for examining SG&A costs is
that the ratio of SG&A costs to sales is closely monitored by investors and analysts (Palepu, Healy and
Bernard 2000, 9-9).
Reasons for Cost Stickiness
The main reason for the occurrence of asymmetric cost behavior is that managers make rational
decisions with regard to trading off the costs of resource adjustments against the costs of retaining
unutilized resources (Anderson et al. 2003; Noreen and Soderstrom 1997). Adjustment costs are
costs to reduce or restore committed resources, including severance pay when employees are
5
dismissed in case of an activity decrease, search and training costs to hire new employees when the
activity increases, and organizational costs such as the loss of morale among remaining employees
when colleagues are fired. The phenomenon of cost stickiness can also arise because managers
perceive the costs of downward adjustments to be higher than those of upward adjustments
(Anderson et al. 2003; Noreen and Soderstrom 1997). However, in the paper of Anderson & Lanen
(2007) it is discussed whether managerial influence is the main reason for cost stickiness. They find
no consistency for other types of costs that are also subject to managerial discretion.
Economic Determinants of Cost Stickiness
Besides managerial intent, prior studies have identified a large number of other factors that affect
the degree of cost stickiness. Chen et al. (2012) discuss the influence of the agency problem, which
arises as a consequence of misalignment of interests between managers and shareholders. The
authors provide convincing evidence that SG&A cost asymmetry is positively associated with
managers’ empire building incentives due to the agency problem. Empire building is described as
managers’ tendencies to grow the firm beyond its optimal size or to maintain unutilized resources
with the purpose of increasing personal utility from status, power, compensation, and prestige.
Another finding of Chen et al. (2012) is that good corporate governance mitigates the positive
association between the agency problem and cost asymmetry, which suggests that corporate
governance mechanisms play an important role in mitigating cost stickiness.
Several studies (Anderson et al. 2003; Subramaniam & Weidenmier 2003; Balakrishnan et al.
2004; Calleja et al. 2006) provide evidence that cost behavior is also determined by firm specific
characteristics that capture organizational constraints on resource adjustment such as asset intensity
and employee intensity. The degree of cost asymmetry increases with the asset intensity and
employee intensity, because it is costly to dispose assets and to dismiss employees if the demand
drops and it also costly to invest once the demand picks up again. Balakrishnan and Gruca (2008)
examine a sample of Canadian hospitals and note that the extent to which a function represents the
6
organization’s core competency influences the stickiness of costs. Their results highlight the
importance of considering organizational characteristics when understanding cost behavior.
Cost asymmetry is also determined by the timing of the activity decrease. Anderson et al.
(2003) for instance find that a successive decrease in revenues is a determinant of cost stickiness.
They accumulated evidence that managers’ assessments of the permanence of a demand reduction
are likely to get stronger as revenue declines continues. Therefore, managers are likely to consider a
revenue decline to be more permanent when it occurs in a second consecutive period of revenue
losses, which results in less cost stickiness. In the same study, Anderson et al. (2003) also provide
evidence for the conjecture that the economic growth of a firm is a determinant of cost asymmetry.
The authors find that managers are less willing to reduce committed resources in periods of
macroeconomic growth than in other periods, resulting in more cost asymmetry. The argument for
this is that managers consider a revenue decline in a period of macroeconomic growth to be more
transitory than a revenue decline in a period of macroeconomic contraction.
Pinnuck and Lillis (2007) and Dierynck et al. (2012) focus on the role of an earnings
benchmark for cost asymmetry. Pinnuck and Lillis (2007) find that firms that report an accounting
loss experience a significantly lower level of investment in employees than firms reporting an
accounting profit. This means that when a firm makes a loss in the prior year, they will cut costs
which leads to more symmetric cost behavior. Dierynck et al. (2012) examine the influence of
incentives to meet or beat the zero earnings benchmark on cost asymmetry of labor costs. The
findings are that the presence of these incentives limits the increase in labor costs following an
activity increase and that managers are more willing to cut labor costs following an activity decrease,
which results in more symmetric cost behavior.
In the study of Banker and Chen (2006) the moderating influence of structural factors on the
degree of cost asymmetry is investigated. They find that country specific labor market characteristics
play a role when determining the degree of stickiness of labor costs.
7
III. HYPOTHESES
The Economic Crisis and Cost Stickiness
This study focuses on comparing cost stickiness in a period of economic crisis to cost stickiness in a
non-crisis period. I expect that the cost stickiness will be lower in a crisis period than in a non-crisis
period. First, the economic crisis forces firms to cut costs, as is shown by the European Central Bank
(2010). They have conducted a survey among Austrian firms just before (in 2007) and shortly after
the heaviest period of the economic crisis (in 2009). This survey signals that during this crisis some 85
percent of companies considered cost cutting a ‘very relevant’ or ‘relevant’ measure in their specific
situation, which makes it the most widespread response of enterprises to the crisis. The preference
for cost cutting during the economic crisis indicates that the trade-off between adjusting resources
and maintaining unutilized resources is more likely to be solved in favor of adjusting the resources.
This cost cutting strategy during the economic crisis results in the expectation that in such a period
cost stickiness will decrease.
Further, it is argued that in a period of economic crisis the maturity of contracts of the firm
decreases, because firms try to avoid long-term commitment to costs. For instance, Caggese and
Cunat (2008) provide evidence that financially constrained firms use fixed-term workers more
intensely and make them absorb a larger fraction of the total employment volatility than financially
unconstrained firms do. The use of fixed-term contracts in a period of an economic crisis is expected
to result in less cost stickiness, because it is less costly to dismiss these employees and thus less
unutilized employees are maintained. This results in the following hypothesis:
H1: firms exhibit less cost stickiness in a period of economic crisis than in a non-crisis period.
Interim Reporting Frequency and Cost Stickiness
Besides examining the overall behavior of costs in a period of economic crisis, this study also focuses
on the effect of the frequency of updating information on costs. Mankiw and Reis (2002)
demonstrate analytically that firms - on aggregate conditions - infrequently update their information,
8
because it is costly to acquire, absorb, and process information. I expect that the degree of cost
stickiness is determined by the frequency by which the firm updates its financial information. In this
study I will use the frequency of interim financial reporting to measure the frequency with which
firms are updating their information about costs and revenues.
Empirical evidence on interim reporting frequency is relatively scarce. Fu (2012) suggests that
the difficulty of gathering a sample of firms that exhibit cross-sectional or time-series variation in
interim reporting frequency causes this scarceness. The study of Butler et al. (2007) shows that there
is little evidence to support the claim that regulation forcing firms to report more frequently
improves earnings timeliness; information in annual earnings is impounded into share price more
quickly for voluntary switches from semiannual reporting to quarterly reporting, but not for
mandatory switches. Fu (2012) examines the effect of an increase of mandatory interim financial
reporting on the cost of equity. Overall, his evidence strongly suggests that the mandatory increase
in interim reporting frequency reduces the cost of equity.
There are also some studies that have focused on the effect of an increase of mandatory
interim reporting on information aspects. Cuijpers & Peek (2010) examine voluntary quarterly
reporting in the U.K., the Netherlands, and Denmark, where only semiannual reporting is required.
They find that an increase in a firm's financial reporting frequency reduces investors' incentives to
acquire private information, and consequently reduces information asymmetry among investors. Fu
et al. (2011) find that more frequent interim reporting is associated with lower levels of information
asymmetry because it increases the amount of information available to the public. Fu (2012) also
mentions that more information is brought to the market after the mandatory switch from
semiannually to quarterly interim reporting. To summarize, prior research indicates that more
information is available to external parties when the frequency of interim reporting increases.
The relationship between the frequency of updating information and the provision of
information to external parties can be explained in two steps. First, managers are instigated to take
more timely decisions if internal information is updated more often. Chenhall and Morris (1986)
9
state that the ability of a manager to respond quickly to events is likely to be influenced by the
timeliness of the Management Accounting System (hereafter, MAS). They specify timeliness in terms
of the provision of information on request and the frequency of reporting systematically collected
information, and state that timely information enhances the facility of MAS to report upon the most
recent events. This indicates that managers are able to respond quickly to events if the information is
timely. However, it does not tell whether they actually will respond quickly if information is updated
more often. Chia (1995) does provide this evidence: she finds that timely information is positively
related to the performance of managers. This means that when information is updated more often,
managers are really instigated to actually respond quickly, because managers that have more timely
information perform better. So, this better performance must be due to a quicker response.
The second step is that there is a link between the MAS and the financial reporting system.
Joseph et al. (1996) find that in practice the relation between financial reporting requirements and
managerial decision making is, if anything, getting stronger and more explicit. Hemmer and Labro
(2008) explore the optimal relation between financial reporting systems and the quality of
management accounting information systems in an economic model where information is
(potentially) useful for both decision making and control. They provide an explicit and significant
connection between exogenous and observable properties of a firm’s financial reporting system and
the quality of the managerial accounting system on which manager(s) base real economic decisions.
As more frequent updated information is more timely, this means that more timely internal
information is linked to more timely external information. So, via these two steps I motivate that
updating internal information more often leads to more timely decisions, both internally and
externally. Because information is more timely, less cost stickiness is expected; Anderson et al.
(2003) note that waiting to obtain information leads to sticky costs because unutilized resources are
maintained. This expectation is reflected in the following hypothesis:
H2: firms updating their information more frequent exhibit less cost stickiness than firms updating
their information less frequent.
10
IV. METHOD
Sample Selection
I focus on a sample that covers the years 2001 through 2010. The years 2008, 2009, and 2010 are
marked as crisis-years; all earlier years are considered non-crisis-years. To examine my two
hypotheses, data about SG&A costs and sales are needed. Also, data about other financial variables,
which will be explained in the model-specification section, are needed. I collected these data from
the Compustat Global and Compustat North-America databases. To construct my economic growthvariable, I used data from The Worldbank.1
The sample contains firms located in the United States and Canada, the European countries
Austria, Belgium, Denmark, Finland, France, Germany, Ireland, Italy, Luxembourg, Portugal, The
Netherlands, Spain, Sweden, and the United Kingdom, the Asian countries Japan, Hong Kong, and
Singapore, and the Oceanic countries Australia and New Zealand. The United States and Canada are
two leading countries that require firms to report every quarter, while almost all firms in European
countries are mandated to report only semi-annually in my sample period. Fu (2012) mentions that
some firms in Germany, Portugal and Sweden are exceptions to this; listed firms and firms of a
certain size have to report quarterly.2 Firms located in Japan, Hong Kong, Singapore, Australia, and
New Zealand are also examined in order to construct a worldwide sample. In Hong Kong, Australia,
and New Zealand, semi-annual interim reporting is mandatory during my sample period. Fu (2012)
notes that since 2003, Singapore has mandated quarterly reporting for firms with a market
capitalization of more than 75 million Singapore dollars (about US$ 58,3 million); all other firms
1
Source: data.worldbank.org. I used nominal Gross National Income (GNI, Atlas method) and Consumer Price Index (CPI) in
each year for all countries in order to calculate the real growth in GNI.
2
In Germany, firms listed in the primary standard (called “new market before 2000”) of its largest stock exchange
(Deutsche Börse) have to report quarterly; for all other firms only semi-annual interim reporting is mandatory. In Portugal,
firms that satisfy conditions related to firm size are mandated to provide quarterly interim reporting; for all other firms only
semi-annual interim reporting is mandatory (see Fu, 2012). In Sweden, quarterly interim reporting is mandated for firms
listed on the Stockholm Stock Exchange; for all other firms only semi-annual interim reporting is mandatory. These
reporting requirements for Germany, Portugal and Sweden apply in my whole sample period.
11
report semi-annually. Before, all firms reported semi-annually. Fu (2012) states that in Japan, the
Tokyo stock exchange has mandated quarterly reporting since March 2004, while other stock
exchanges such as the Jasdaq Securities Exchange retained the semiannual reporting requirement
until 2007. Since April 2008, all listed firms in Japan have had to report quarterly.
Table 1, Panel A, describes how the final sample is constructed. Based on prior research
(Anderson et al. 2003; Chen et al. 2012; Dierynck et al. 2012) I require SG&A costs and sales revenues
to be available in the current year and previous year. Observations with missing data on either sales
revenue or SG&A costs for the current or preceding year are deleted. Further, observations with
missing data on other variables are omitted. Next, I require SG&A costs to be smaller than sales.
Finally, following Chen et al. (2012), I trim the top and the bottom 0.5% of the observations with
extreme values in the change of SG&A costs and change of sales revenue. These sample selection
procedures result in a final sample that consists of 77.176 firm-year observations between 2001 and
2010.
Table 1, Panel B presents the distribution of the firm-year observations over the countries for
the sample of firms providing quarterly interim reporting and firms providing semi-annual interim
reporting (both for the Compustat data and Fu-method, which will be explained later on). This
distribution is provided on a yearly basis.
Insert Table 1 about here
Model Specification and Variable Definitions
Asymmetry of SG&A Costs
I test cost asymmetry by using both a basic and an extended model. The basic model is identical to
the first model of Anderson et al. (2003) and is given by equation (1):
12
SG&Ait denotes selling, general and administrative costs of firm i in year t, Salesit are sales revenues,
and Decrease_Dummyit is an indicator variable set equal to one when sales in year t are smaller than
sales in year t-1, and 0 otherwise.
Because the value of Decrease_Dummy is 0 when revenue increases, the coefficient β1
measures the percentage increase in SG&A costs for a one percent increase in sales. Because the
value of Decrease_Dummy is one when revenue decreases, the sum of β1 and β2 measures the
percentage decrease in SG&A costs for a one percent decrease in sales. Based on prior literature, my
expectations are that β1 will be positive, β2 will be negative and the absolute value of β1 will be
greater than the absolute value of β2. This would indicate that SG&A costs are sticky.
Compared to the basic model, the extended model given by equation (2) includes controls for
economic determinants of cost asymmetry as identified earlier in this study. These controls are based
on prior research (Anderson et al. 2003; Pinnuck and Lillis 2007; Dierynck et al. 2012).
13
Successive_Decrease equals one if sales have decreased in two consecutive years (i.e., sales
t-2
>
salest-1 > salest), and zero otherwise. Employee_Intensity is the ratio of total number of employees
over sales, while Asset_intensity is the ratio of total assets over sales. Economic_Growth is growth in
real GNI in year t. Loss_Prior_Year indicates whether the firm reported a loss in the prior year, or not.
All other variables are as previously defined.
Cost Asymmetry and Economic Crisis (Hypothesis 1)
I use two approaches to investigate whether “economic-crisis-observations” exhibit a lower degree
of SG&A cost asymmetry than “non-crisis-observations”. First, I add to equation (2) an indicator
variable, Ecrisis_Dummy, which is set equal to one if year t is a year of economic crisis, and 0 if not,
and include
two-way (Log(Salesit/Salesit-1)*Ecrisis_Dummy) and three-way (Decrease_Dummyit
*Log(Salesit/Salesit-1)*Ecrisis_Dummy) interaction terms. This enables these equations to be
estimated using the whole sample while enabling coefficients to differ for observations in times of
economic crisis and observations in non-crisis times. Specifically, the two-way interaction term
reflects the difference in SG&A cost increase following an activity increase of firms that are in
economic crisis relative to non-crisis firms. If it equals for instance -0,08, this means that the increase
in SG&A cost for an activity increase is 8% lower for firms that are in a period of economic crisis
relative to firms that are in a non-crisis period (for an equal activity increase). The three-way
interaction term represents the difference in SG&A cost decrease following an activity decrease of
firms that are in economic crisis relative to non-crisis firms. If it equals for instance 0,08, this means
that the decrease in SG&A cost for an activity decrease is 8% higher for firms that are in a period of
economic crisis relative to firms that are in a non-crisis period (for an equal activity decrease). For the
first hypothesis, I expect that SG&A costs are less sticky in a crisis period compared to a non-crisis
period, which results in the expectation that β3 will be significantly negative and β4 will be
significantly positive.
14
Second, I make use of split-sample regressions, which means that the basic and extended
model are separately estimated for the non-crisis (years 2001-2007) and crisis (2008-2010) sample. I
predict that firms that are in economic crisis exhibit less cost asymmetry than firms that are not in
economic crisis. Relative to non-crisis observations, I expect β1 to be smaller and β2 to be less
negative for crisis observations. Reasons for these expectations are that the economic crisis forces
firms to cut costs and that the maturity of contracts decreases, which lead to lower cost asymmetry.
Cost Asymmetry and Interim Financial Reporting (Hypothesis 2)
I use the same two approaches to investigate whether firms that report quarterly exhibit a lower
degree of SG&A cost asymmetry than semi-annual reporters. First, instead of using Ecrisis_Dummy in
equation (2) as an indicator variable, I use Ifr_Dummy, which is set equal to one if a firm in year t is
mandated to provide quarterly interim financial reporting, and 0 if semi-annual interim financial
reporting is mandated. I then include two-way (Log(Salesit/Salesit-1)*Ifr_Dummy) and three-way
(Decrease_Dummyit *Log(Salesit/Salesit-1)*Ifr_Dummy) interaction terms. The two-way interaction
term reflects the difference in SG&A cost increase following an activity increase of firms reporting
quarterly relative to firms reporting semi-annually. If it equals for instance -0,08, this means that the
increase in SG&A costs for an activity increase is 8% lower for firms reporting quarterly relative to
firms reporting semi-annually (for an equal activity increase). The three-way interaction term
represents the difference in SG&A cost decrease following an activity decrease of firms providing
quarterly interim reporting relative to firms providing interim reporting semi-annually. If it equals for
instance 0,08, this means that the decrease in SG&A costs for an activity decrease is 8% higher for
firms reporting quarterly compared to firms reporting semi-annually (for an equal activity decrease).
For the second hypothesis, I expect that firms experience less sticky SG&A costs when they update
their internal information more frequent, which results in the expectation that β3 will be significantly
negative and β4 will be significantly positive.
15
Second, I again make use of split-sample regressions, so the basic and extended model are
separately estimated for firms reporting quarterly and firms reporting semi-annually. I predict that
firms that provide more frequent interim reporting update their internal information more often,
which leads to quicker responses by managers; lower cost asymmetry is expected for quarterly
reporters. Therefore, relative to semi-annual observations, I expect β1 to be smaller and β2 to be less
negative for firms reporting quarterly.
V. RESULTS
Descriptive Results
The descriptive statistics for the different subsamples are presented in Table 2. The focus is on the
descriptive statistics of six subsamples: the economic crisis sample and the non crisis-sample, the
quarterly interim reporting sample and the semi-annual interim reporting sample using Compustat
data, and the quarterly interim reporting sample and the semi-annual interim reporting sample using
data based on the method of Fu. On average, the sales revenues of the semi-annual interim reporting
sample using Compustat data ($199,000,000,000) are enormous compared to the quarterly interim
reporting sample using Compustat data ($4,285,665,730). This is due to the presence of huge
Japanese firms in the semi-annual reporting sample. Also the SG&A costs for the semi-annual interim
reporting sample ($31,972,526,070) are extremely high compared to quarterly interim reporting
sample ($682,729,310). The dispersion in the average ratio of SG&A costs to sales is small: all
samples have a ratio between 21 and 28 percent. These high percentages indicate that SG&A costs
are a very significant cost category for firms. The number of employees and the beginning total
assets of the firm give an indication about the size of the firms in the different subsamples. Based on
the number of employees, the largest firms are in the quarterly interim reporting sample using data
based on the method of Fu (12.099 employees) and in the quarterly interim reporting sample using
Compustat data (11.282). Based on the beginning total assets, the largest firms are in the semiannual interim reporting sample using Compustat data ($207,000,000,000). The gigantic Japanese
16
firms are present in this sample. The economic crisis sample has the highest probability of a decrease
in sales (42 percent), as well as the highest probability of a successive decrease in sales revenue (20
percent). Especially in the case of a decrease in sales, this result is very significant compared to the
other samples.
Insert Table 2 about here
Hypothesis Tests
Asymmetry of SG&A Costs
Column I and II of Table 3 present the regression summary statistics for the basic and extended
models, Equations (1) and (2), for the full sample consisting of 77,176 firm-year observations. The
results of the basic model show significant SG&A cost asymmetry with β1 equal to 0.59 (t = 130.722)
and β2 equal to -0.14 (t = -15.993). After controlling for economic determinants of cost asymmetry by
using the extended model, SG&A costs keep behaving significantly asymmetrical (β1 = 0.59, t =
128.701; β2 = -0.22, t = -19.209). This means that an increase in sales revenues of one percent leads
to an increase in SG&A costs of 0.60 percent, while a decrease in sales revenues of one percent
results in a decrease in SG&A costs of 0.37 percent.
Insert Table 3 about here
In the extended model, the control variable successive decrease is significant (β5 = 0,10; t = 7.418). It
shows that firms with a revenue decline in a second consecutive period exhibit lower SG&A cost
stickiness. This is in line with the view that managers adjust their behavior to this news of declining
revenue, as they are more likely to consider this to be more permanent.
17
Cost Asymmetry and Economic Crisis
One important prediction in this study is that firms exhibit a lower degree of SG&A cost asymmetry in
times of economic crisis than in a non-crisis period. In other words, I expect that so called “crisisobservations” will exhibit a smaller SG&A cost increase following an activity increase and a larger
SG&A cost decrease following an activity decrease.
Table 3, Column III, present findings from the first test of this hypothesis, which is based on
estimating Equation (2) including the crisis indicator variable and the corresponding interaction
terms. Consistent with my prediction, I find a negative coefficient on the two-way interaction term
(β3 = -0.03; t =-2.927). So, in comparison to firms that are not in times of economic crisis, the SG&A
cost increase of crisis firms for a one percent increase in sales is 0.03 percent smaller. The three-way
interaction term, β4, shows that the decrease in SG&A cost following an activity decrease is not
higher for crisis-firms compared to non-crisis firms (β4 = 0.00; t = 0.043), which is inconsistent with
my prediction.
Table 4 presents the result for my split sample tests. These tests are conducted by estimating
the basic and extended models separately for the crisis and non-crisis observations. Findings for the
non-crisis observations are similar to those of the full sample. In other words, I find strong evidence
for asymmetric cost behavior of non-crisis observations for the basic model (β1 = 0.60, t = 119.966; β2
= -0.12, t = -11.103) as well as for the extended model (β1 = 0.60, t = 117.014; β2 = -0.23, t = -14.041).
However, the crisis observations show nearly the same asymmetrical behavior, both for the basic
model (β1 = 0.56, t = 64.155; β2 = -0.14, t = -9.428) as well as for the extended model (β1 = 0.57, t =
63.619; β2 = -0.22, t = -11.847), which is in contrast with my prediction. I predicted that crisis
observations would show more symmetric behavior compared to non-crisis observations. Taken
together, the results for the split-sample regressions provide no support for the hypothesis that firms
exhibit more symmetric cost behavior in a crisis period compared to a non-crisis period.
Insert Table 4 about here
18
Overall, the findings in Table 3 and 4 do not provide consistent support for my first hypothesis that
crisis observations exhibit more symmetrical SG&A cost behavior compared to non-crisis
observations. I do find a negative coefficient on the two-way crisis interaction term in column III of
Table 3. All other findings in Table 3 and 4 show that the degree of cost asymmetry is not significantly
different for crisis observations and non-crisis observations; both in times of crisis and in non-crisis
periods, asymmetrical cost behavior is detected.
Cost Asymmetry and Updating Information
The second important prediction in this study is that firms that update their internal information
system more frequent exhibit a lower degree of SG&A cost asymmetry. The frequency of updating
internal information is measured by the frequency of interim financial reporting. More specifically, I
expect that firms that report every quarter exhibit a smaller SG&A cost increase following an activity
increase and a larger SG&A cost decrease following an activity decrease, compared to
firms
reporting semi-annually.
My results concerning this hypothesis are based on two sources. In the first place, the both
Compustat-databases provide data about the reporting periodicity of a firm. In other words, these
data should represent how often a firm provides interim financial reporting. However, this source
seems to be not totally reliable. For instance, Fu (2012) mentions that since April 2008, all listed firms
in Japan are mandated to report quarterly. This is in contrast with the data that Compustat provides,
because Compustat states all Japanese firms in my sample report semi-annually.3 Because I question
the correctness of all the data Compustat provides, I use a second method of distinguishing semiannual and quarterly reporting firms. This method comes down to following Fu’s (2012)
requirements for the countries in my sample, and manually determine the interim reporting
3
I have tested the reporting periodicity (RP) variable that Compustat provides by investigating some major Japanese listed
firms in my sample. The websites of Nippon Telegraph & Telephone (NTT), Mitsubishi Corporation, Nippon Steel
Corporation (NSC), Nissan Motors, and Fujitsu Limited all show that they, in contrast to what Compustat shows, report
quarterly.
19
frequency based on these requirements. Unfortunately, it is not possible to include Japan in the
sample; the high number of observations in Japan makes it impossible to manually determine the
interim reporting frequency for each individual firm. This is a severe limitation, because Japan is an
important country in my sample, also for its changes in interim reporting regime over the years.
Table 5, column I, presents findings from the hypothesis test using interim reporting
frequency-data from Compustat. This is based on estimating Equation (2) including the interim
financial reporting frequency and corresponding interaction terms. In contrast with my predictions, I
find a positive coefficient on the two-way interaction term (β3 = 0.03; t = 2.845) and a significantly
negative coefficient on the three-way interaction term (β4 = -0.08; t = -3.973). This indicates that the
SG&A cost increase of firms reporting quarterly for a one percent increase in sales is 0.03 percent
larger compared to firms reporting semi-annually, and that the SG&A cost decrease of firms
reporting quarterly for a one percent decrease in sales is 0.08 percent smaller compared to firms
reporting semi-annually.
Insert Table 5 about here
Table 6 presents the result for my split sample tests based on Compustat interim reporting
frequency data. These tests are conducted by estimating the basic and extended models separately
for the quarterly- and semi-annual-reporting observations. I find strong evidence for asymmetric cost
behavior of quarterly reporting observations for the basic model (β1 = 0.58, t = 96.066; β2 = -0.14, t =
-12.406) as well as the extended model (β1 = 0.58, t = 94.832; β2 = -0.24, t = -15.227). The semiannual reporting observations exhibit far less asymmetric SG&A cost behavior for the basic model (β1
= 0.56, t = 88.078; β2 = -0.03, t = -3.048) and for the extended model (β1 = 0.55, t = 86.522; β2 = -0.15,
t = -10.110). This is inconsistent with my predictions; I expected that quarterly reporting firms would
exhibit more symmetric behavior compared to semi-annual reporting firms. Taken together, results
for the split-sample regressions reject my hypothesis, because firms updating their internal
20
information less exhibit more symmetric SG&A cost behavior compared to firms updating internal
information more frequent.
Insert Table 6 about here
Table 5, column II, presents findings from the hypothesis test using interim reporting frequency-data
from the Fu-method. I find a very small negative coefficient on the two-way interaction term (β3 =
-0.01; t = -0.564), but this coefficient is far from significant. On the three-way interaction term, I find
a very small positive coefficient (β4 = -0.01; t = 0.470), but again this coefficient is insignificant. This
means my predictions are not supported, because the difference in SG&A cost increase between
firms reporting quarterly and firms reporting for a one percent increase in sales is close to zero. The
same holds for the difference in SG&A cost decrease between quarterly- and semi-annual reporters
for a one percent activity decrease: it is close to zero. So, these results provide no basis to think that
quarterly reporters exhibit more symmetric SG&A cost behavior.
Table 7 presents the result for my split sample tests based on interim reporting frequency
data following the method of Fu. Again, the basic and extended models are separately estimated for
the quarterly and semi-annual reporting observations. I find evidence for asymmetric cost behavior
of quarterly reporting observations for the basic model (β1 = 0.59, t = 108.808; β2 = -0.16, t = -15.486)
as well as the extended model (β1 = 0.60, t = 109.755; β2 = -0.28, t = -20.602). The semi-annual
observations exhibit the same asymmetrical SG&A cost behavior for the basic model (β1 = 0.60, t =
27.514; β2 = -0.18, t = -4.510) and for the extended model (β1 = 0.61, t = 27.170; β2 = -0.30, t = -5.649).
This is inconsistent with my predictions; my prediction was that quarterly reporting firms would
exhibit more symmetric behavior compared to semi-annual reporting firms, but the split-sample
shows that more frequent updating of internal information has no influence on SG&A cost behavior.
21
Taken together, results for the split-sample regressions reject my hypothesis, because firms updating
internal information less frequent exhibit less asymmetric SG&A cost behavior compared to firms
updating internal information more frequent.
Insert Table 7 about here
Overall, the findings in Table 5, 6, and 7 do not provide any support for my hypothesis that firms that
update their internal information more frequently exhibit less asymmetric cost behavior. Using
Compustat interim reporting frequency data, Table 5, column I, and Table 6 show that firms updating
their internal information more frequently exhibit more asymmetric SG&A cost behavior. Using
interim reporting frequency data following the method of Fu, Table 5, column II, and Table 7 show
that firms updating their internal information more frequently exhibit the same asymmetrical SG&A
cost behavior as firms updating their information less frequent.
Robustness Check
In this section, I perform some additional testing to determine the sensitivity of my findings
regarding asymmetry of SG&A costs in general, as well as the influence of the crisis and the influence
of the frequency of updating information on asymmetry of SG&A costs. I do this by using country
dummies, which control for possible country effects. A country dummy is set equal to one if an
observation belongs to country X, and all observations belonging to other countries are set equal to
0. This is done for all countries in my sample. The United States serves as base level and are therefore
not taken into consideration.
Table 8 presents the results of the extended models. These results are after controlling for
economic determinants of cost asymmetry and controlling for country effects. The results show
significant asymmetrical SG&A cost behavior for the extended model with Ecrisis_Dummy (β1 = 0.58,
t = 56.175; β2 = -0.23, t = -8.949) as well for the extended models with Ifr_Dummy (Compustat: β1 =
22
0.55, t = 49.207; β2 = -0.15, t = -6.508 and Fu: β1 = 0.60, t = 29.904; β2 = -0.27, t = -5.915). This is very
similar to the SG&A cost behavior detected in the main results. Further, the extended model with
Ecrisis_Dummy shows that the SG&A cost increase of crisis firms is smaller compared to non-crisis
firms (β3 = -0.03, t = -2.781) and that the SG&A cost decrease of crisis firms is higher (β4 = 0.04, t =
1.516). This is slightly different from the main results, where there was no difference detected in the
decrease in SG&A costs between crisis and non-crisis firms. Next, the extended model with
Ifr_Dummy using Compustat data shows that the SG&A cost increase is higher for firms reporting
quarterly relative to firms reporting semi-annually (β3 = 0.03, t = 1.626) and that the SG&A cost
decrease is lower for firms reporting quarterly compared to firms reporting semi-annually (β4 = -0.13,
t = -4.258). This differs from the main results, where the difference in SG&A cost decrease between
semi-annual and quarterly reporting firms was far less pronounced. Finally, the extended model with
Ifr_Dummy applying Fu’s method shows that the SG&A cost increase is lower for firms reporting
quarterly compared to firms reporting semi-annually (β3 = -0.11, t = -2.958) and that the SG&A cost
decrease is higher for firms reporting quarterly relative to firms reporting semi-annually (β4 = 0.10, t
= 4.493). This is totally different from the main results, which provided no evidence that there is a
significant difference in SG&A cost asymmetry between semi-annual and quarterly reporting firms.
In summary, the main results do not seem to be entirely robust. Controlling for country
effects leads in multiple cases to different results compared to the main results reported earlier,
which indicates that the United States is a very dominant country in obtaining the main results.
Especially in the extended model with Ifr_Dummy applying Fu’s method, controlling for country
effects leads to results that are significantly different from the main findings. The results of the
extended model with Ifr_Dummy using Compustat data differ less from the main results, possibly
because Japan is present in that sample but not in the sample using Fu’s method.
Insert Table 8 about here
23
VI. CONCLUSION
In this study I examine whether SG&A costs of firms behave more symmetric in times of economic
crisis (2008-2010) compared to a non-crisis period (2001-2007). To do so, I constructed a sample that
consists of countries spread all over the world. My results do not provide consistent support for this
hypothesis; they show that the degree of cost asymmetry is not significantly different for crisis
observations and non-crisis observations. That is, firms exhibit asymmetrical cost behavior both in
times of crisis and in non-crisis periods.
Second, I examine the effect of the frequency of updating internal information on SG&A cost
asymmetry, by looking at firms’ interim reporting frequency. My findings do not provide any support
for this hypothesis. Using Compustat interim reporting frequency data, I find that firms updating
their internal information more frequently exhibit more asymmetric SG&A cost behavior. Using
interim reporting frequency data following the method of Fu, I find that firms updating their internal
information more frequently exhibit the same asymmetrical SG&A cost behavior as firms updating
their information less frequent.
For both hypotheses, the results do not meet my predictions. I do not have a proper
explanation for why the crisis-hypothesis has not led to the expected result. I do have some possible
arguments for why the hypothesis concerning the updating of internal information has not led to the
expected result. With regard to the Compustat data, I question whether these data are proper and
reliable, especially for Japan. The database mentions that all Japanese firms report semi-annually in
my sample period, which is not in accordance with the requirements of Fu. Japan has the largest
firms in the sample and often these firms report quarterly, so the effect of updating information on
cost asymmetry might have shifted from the quarterly reporting sample to the semi-annual reporting
sample. This would explain why the semi-annual sample, where Japanese firms are a big part of the
total sample, exhibits far less asymmetric SG&A behavior. With regard to the data following the Fu
method, the lack of significance might also be explained by Japan, because Japan is excluded there.
24
Also, the robustness check shows that the main results are not entirely robust. Controlling for
country effects leads in multiple cases to different results compared to the main results, which
indicates that the United States is a very dominant country in the obtaining the main results. So, the
results would have been different if no country effects were present.
This study contributes to the existing literature in two ways. First, I test cost stickiness on a
large dataset containing observations from countries spread all over the world, both for a period of
economic crisis and a non-crisis period. Second, this study also provides some first large-sample
evidence with regard to the effect of updating information on cost asymmetry, although the
evidence is seriously limited.
The limitation arises from the problems of determining the interim reporting frequency of
firms. I question the correctness of the output of Compustat, and applying Fu’s interim reporting
requirements has many limitations. To start, here, Japan is excluded from the sample. This is a very
important country in my study, looking at the number of observations and the changes in interim
reporting requirements within my sample period. Also, the method of Fu does not make a distinction
for large firms voluntarily reporting more frequent than mandated; some European multinationals
choose to report quarterly, while only semi-annual reporting is mandated. Next, there are limitations
regarding the interim reporting frequency of firms in Sweden, Singapore, and Portugal. For firms
reporting in these countries there was a lack of data, which forced me to use outdated information.
This might lead to errors in the interim reporting frequency of firms. Future research using exact
interim reporting frequency-data will give more precise results and will be far less limited.
A final limitation is that SG&A costs are not finely disaggregated. The use of these aggregated
data makes it difficult to infer which components of SG&A costs behave asymmetrically. As sticky
cost behavior is consistent with deliberate decision-making by managers, focusing on SG&A costs
makes it difficult to address how specific costs are managed and which managerial decisions
contribute to cost asymmetry. Future research using finer data might provide information on cost
25
behavior for different components of SG&A costs. This knowledge would develop a greater
understanding of the managers’ process of decision making.
26
REFERENCES
Anderson, M., R. Banker, and S. Janakiraman. 2003. Are selling, general, and administrative costs
“sticky”? Journal of Accounting Research 41 (1): 47-63.
Anderson, S.W., and W. Lanen. 2007. Understanding cost management: What can we learn from the
evidence on “sticky” costs. Working paper, Rice University, University of Michigan.
Balakrishnan, R., M. Peterson, and N. Soderstrom. 2004. Does capacity utilization affect the
“stickiness” of cost? Journal of Accounting Auditing and Finance 19 (3): 283-299.
Balakrishnan, R., and T. Gruca. 2008. Cost stickiness and core competency: A note.
Contemporary Accounting Research 25 (4): 993-1006.
Banker, R., and T.L. Chen. 2006. Labor Market Characteristics and Cross-Country Differences in Cost
Stickiness. Working paper, Temple University, Georgia State University.
Butler, M., A. Kraft, and I. Weiss. 2007. The Effect of Reporting Frequency on the Timeliness of
Earnings: The Cases of Voluntary and Mandatory Interim Reports. Journal of Accounting and
Economics 44: 181-217.
Caggese A., and V. Cunat. 2008. Financing constraints and fixed term employment contracts.
The Economic Journal 118: 2013–2046.
Calleja, K., M. Steliaros, and D.C. Thomas. 2006. A note on cost stickiness: Some
international comparison. Management Accounting Research 17 (2): 127-140.
Chen, C.X., H. Lu, and T. Sougiannis. 2012. The agency problem, corporate governance, and the
asymmetrical behavior of selling, general, and administrative costs. Contemporary
Accounting Research (forthcoming).
Chenhall, R.H., and D. Morris. 1986. The Impact of Structure, Environment, and Interdependence on
the Perceived Usefulness of Management Accounting Systems. The Accounting Review 61(1):
16-35.
27
Chia, Y.M. 1995. Decentralization, Management Accounting System (MAS) Characteristics and Their
Interaction Effects on Managerial Performance: A Singapore Study. Journal of Business
Finance & Accounting 22(6): 811-830.
Cooper R., and R. Kaplan. 1998. The Design of Cost Management Systems: Text, Cases and Readings.
Upper Saddle River, NJ: Prentice Hall.
Cuijpers, R., and E. Peek. 2010. Reporting Frequency, Information Precision and Private Information
Acquisition. Journal of Business Finance & Accounting 37: 27-59.
Dierynck, B., W.R. Landsman, and A. Renders. 2012. Do managerial incentives drive cost behavior?
Evidence about the role of the zero earnings benchmark for labor cost behavior in Belgian
private firms. The Accounting Review (forthcoming).
Ely, K. M. 1991. Inter-industry differences in relation between compensation and firm performance
variables. Journal of Accounting Research 29 (Spring): 37-58.
European Central Bank. 2010. Firms’ reactions to the crisis and their consequences for the labour
market. Working Paper Series No 1274.
Fu, R. 2012. Does the mandatory increase in interim reporting frequency reduce the cost of equity?
International evidence. Working paper, Erasmus University Rotterdam.
Fu, R., A. Kraft, and H. Zhang. 2011. Financial Reporting Frequency, Information Asymmetry and the
Cost of Equity. Working paper, Erasmus University, City University London, and Nanyang
Technological University.
Hemmer, T., and E. Labro. 2008. On the Optimal Relation between the Properties of Managerial and
Financial Reporting Systems. Journal of Accounting Research 46 (5): 1209-1240.
Joseph, N., S. Turley, J. Burns, L. Lewis, R. Scapens, and A. Southworth. 1996. External Financial
Reporting and Management Information: A Survey of UK Management Accountants.
Management Accounting Research 7: 73-93.
Mankiw, N. G. and R. Reis. 2002. Sticky Information Versus Sticky Prices: A Proposal to Replace the
New Keynesian Phillips Curve. Quarterly Journal of Economics 117 (4): 1295-1328.
28
Medeiros, O.R.D., and P.D.S. Costa. 2004. Cost Stickiness in Brazilian Firms. Working paper.
Noreen, E., and N.S. Soderstrom. 1997. The accuracy of proportional cost models: Evidence from
hospital service departments. Review of Accounting Studies 2 (1): 89-114.
Palepu, K., P. Healy, and V. Bernard. 2000. Business Analysis and Valuation Using Financial
Statements, Mason OH: South-Western.
Pinnuck, M., and A.M. Lillis. 2007. Profits versus losses: Does reporting an accounting loss act as
a heuristic trigger to exercise the abandonment option and divest employees? The
Accounting Review 82 (4): 1031-1053.
Subramaniam, C., and M. Weidenmier. 2003. Additional evidence on the sticky behavior of costs.
Working Paper, Texas Christian University.
29
TABLE 1
Sample selection
PANEL A: Full Sample
Number of Firm-Years
Potential sample
188,631
Less: Observations with missing data on either sales revenue or
SG&A costs for the current or preceding year
(83,493)
Less: Observations with missing data on other variables
(21,770)
Less: Observations for which SG&A costs exceed sales revenue
for the current year
(4,610)
Less: Trim top and bottom of change in sales and change in SG&A
costs at 0.5% level
(1,582)
Final sample
77,176
30
PANEL B: Distribution of Firm-Year Observations over Sample-Years and over Quarterly and
Semi-Annual Reporting
Conform Compustat
Quarterly interim reporting:
Australia
Austria
Belgium
Canada
Denmark
Finland
France
Hong Kong
Ireland
Italy
Luxembourg
The Netherlands
New-Zealand
Spain
United Kingdom
the United States
Germany
Portugal
Singapore
Sweden
2001
3
0
0
95
0
0
1
1
0
0
0
1
0
0
2
2.872
0
0
0
0
2.975
2002
5
0
0
99
0
0
1
1
0
0
0
1
1
0
1
3.072
0
0
0
0
3.181
2003
11
0
0
99
1
0
3
2
0
0
0
1
0
1
5
3.272
0
1
7
1
3.404
2004
9
6
6
170
20
11
51
6
3
2
5
18
0
2
21
3.454
114
0
69
71
4.038
2005
14
4
8
260
18
11
37
26
9
2
7
20
1
2
145
3.590
34
1
33
69
4.291
2006
241
27
37
314
46
49
91
139
25
23
10
47
26
19
618
3.719
198
7
83
72
5.791
2007
230
36
42
378
47
61
85
157
28
24
10
48
23
23
717
3.862
219
7
87
85
6.169
2008
211
39
44
409
56
83
77
186
27
41
10
56
23
31
742
3.911
228
12
75
79
6.340
2009
195
34
48
419
56
79
81
194
30
42
12
57
24
26
767
3.999
180
15
91
101
6.450
2010
199
37
47
429
58
72
139
250
33
30
12
57
19
28
786
4.128
248
19
81
105
6.777
Total
1.118
183
232
2.672
302
366
566
962
155
164
66
306
117
132
3.804
35.879
1.221
62
526
583
49.416
2001
2
4
6
0
15
8
29
13
17
1
1.941
5
16
0
297
81
92
1
80
50
2.658
2002
14
8
9
0
20
8
42
30
16
1
2.134
6
17
1
316
79
116
0
132
69
3.018
2003
42
6
10
0
23
10
47
49
19
2
2.211
7
20
0
336
52
121
0
172
74
3.201
2004
44
1
6
0
3
0
4
51
15
0
2.304
1
7
0
314
45
9
0
137
0
2.941
2005
0
2
1
0
1
0
2
31
0
0
2.404
0
2
0
23
44
3
1
44
1
2.559
2006
3
0
0
0
1
0
1
14
3
0
2.472
0
0
1
11
55
3
0
5
1
2.570
2007
1
0
1
0
0
0
1
0
2
0
2.599
0
0
0
1
57
2
0
1
1
2.666
2008
0
1
2
1
0
0
0
0
2
0
2.630
0
0
0
0
45
4
0
0
2
2.687
2009
0
0
1
0
1
0
1
0
1
0
2.638
0
0
0
4
69
3
0
0
1
2.719
2010
2
0
1
0
1
0
7
1
2
0
2.661
0
0
0
4
58
1
0
2
0
2.740
Total
108
22
37
1
65
26
134
189
77
4
23.994
19
62
2
1.306
585
354
2
573
199
27.759
Semi-annual interim reporting:
Australia
Austria
Belgium
Canada
Denmark
Finland
France
Hong Kong
Ireland
Italy
Japan
Luxembourg
The Netherlands
Spain
United Kingdom
the United States
Germany
Portugal
Singapore
Sweden
31
Conform Fu-method
Quarterly interim reporting:
Canada
the United States
Germany
Portugal
Singapore
Sweden
2001
95
2.953
26
0
0
49
3.123
2002
99
3.151
33
0
0
65
3.348
2003
99
3.324
40
1
58
70
3.592
2004
170
3.499
43
0
96
67
3.875
2005
260
3.634
23
1
33
65
4.016
2006
314
3.774
76
7
42
66
4.279
2007
378
3.919
87
7
55
79
4.525
2008
410
3.956
90
12
48
77
4.593
2009
419
4.068
84
15
45
94
4.725
2010
429
4.186
107
19
45
98
4.884
Total
2.673
36.464
609
62
422
730
40.960
2001
5
4
6
15
8
30
14
17
1
5
17
0
0
299
66
1
80
1
569
2002
19
8
9
20
8
43
31
16
1
6
18
1
1
317
83
0
132
4
717
2003
53
6
10
24
10
50
51
19
2
7
22
0
1
341
81
0
121
5
803
2004
53
7
12
23
11
55
57
18
2
6
25
0
2
335
80
0
110
4
800
2005
14
6
9
19
11
39
57
9
2
7
22
1
2
168
14
1
44
5
430
2006
244
27
37
47
49
92
153
28
23
10
47
26
20
629
125
0
46
7
1.610
2007
231
36
43
47
61
86
157
30
24
10
48
23
23
718
134
0
33
7
1.711
2008
211
40
46
56
83
77
186
29
41
10
56
23
31
742
142
0
27
4
1.804
2009
195
34
49
57
79
82
194
31
42
12
57
24
26
771
99
0
46
8
1.806
2010
201
37
48
59
72
146
251
35
30
12
57
19
28
790
142
0
38
7
1.972
Total
1.226
205
269
367
392
700
1.151
232
168
85
369
117
134
5.110
966
2
677
52
12.222
Semi-annual interim reporting:
Australia
Austria
Belgium
Denmark
Finland
France
Hong Kong
Ireland
Italy
Luxembourg
The Netherlands
New-Zealand
Spain
United Kingdom
Germany
Portugal
Singapore
Sweden
Notes:
* I label ‘quarterly reporting’ firm years as those reporting four times per year. ‘Semi-annual reporting’ firm- years are those
firm-years for which financial information is reported two times per year.
* The years 2001 through 2007 are marked as ‘non-crisis years’; the years 2008 through 2010 are marked as ‘crisis years’.
32
TABLE 2
Sample Descriptive Statistics
Variable
Sales
(in $000)
SGA
(in $000)
Change_Sales
(in $000)
Change_SGA
(in $000)
Relative_Change_Sales
(in %)
Relative_Change_SG&A
(in %)
SG&A / Sales
(in %)
Number of emp
Beginning Total Assets
(in $000)
Decline observations
Successive decrease
mean
median
std dev
mean
median
std dev
mean
median
std dev
mean
median
std dev
mean
median
std dev
mean
median
std dev
mean
median
std dev
mean
median
std dev
mean
median
std dev
mean
median
std dev
mean
median
std dev
Full sample
Non-crisis sample
Crisis sample
74.465.040,13
1.157.600,00
547.500.000,00
11.940.017,81
200.372,50
84.025.826,26
1.386.586,88
12.082,00
69.513.820,08
189.449,56
2.310,00
8.024.338,91
3,15
2,68
31,57
0,59
0,35
8,32
0,25
0,21
0,38
9.412,36
996,00
39.654,21
80.789.360,09
1.620.413,50
612.000.000,00
0,29
0,00
0,45
0,13
0,00
0,34
78.067.652,73
1.199.373,00
564.400.000,00
12.402.520,40
207.983,00
85.916.949,70
4.189.547,05
25.511,50
51.482.517,03
437.029,20
4.056,00
7.592.467,62
5,62
3,97
29,91
1,00
0,50
8,20
0,25
0,21
0,19
9.261,56
992,00
39.387,38
83.473.766,40
1.639.157,00
604.800.000,00
0,22
0,00
0,41
0,10
0,00
0,30
68.034.756,25
1.084.505,50
515.900.000,00
11.114.529,21
185.803,00
80.535.136,53
-3.327.036,16
1.097,00
92.010.355,77
-226.888,16
575,50
8.686.659,19
-0,99
0,45
33,78
-0,10
0,12
8,47
0,26
0,21
0,58
9.681,52
1.000,00
40.125,34
75.998.150,58
1.589.924,00
624.500.000,00
0,42
0,00
0,49
0,20
0,00
0,40
Quarterly Rep. Sample
(Compustat)
4.285.665,73
255.908,00
64.007.320,00
682.729,31
50.610,50
6.871.467,70
272.695,24
8.222,50
8.298.943,49
40.317,45
1.861,00
665.271,61
4,31
3,50
36,34
0,80
0,51
9,80
0,28
0,23
0,46
11.282,02
1.016,00
46.129,32
9.694.911,00
444.986,50
189.800.000,00
0,26
0,00
0,44
0,11
0,00
0,31
Semi-Annual Rep. Sample
(Compustat)
199.000.000,00
30.536.000,00
895.300.000,00
31.972.526,07
5.037.500,00
137.500.000,00
439.896,95
16.000,00
13.102.877,97
439.896,95
16.000,00
13.102.877,97
1,21
1,71
21,14
0,23
0,17
4,89
0,21
0,18
0,14
6.085,29
971,00
23.806,74
207.000.000,00
29.747.000,00
957.600.000,00
0,34
0,00
0,47
0,18
0,00
0,38
Quarterly Rep. Sample
(Fu)
4.597.390,44
279.552,00
69.584.760,73
764.313,68
56.178,50
7.568.137,64
43.397,33
2.006,00
643.984,53
43.397,33
2.006,00
643.984,00
4,15
3,17
35,16
0,74
0,47
9,74
0,28
0,24
0,20
12.099,21
1.114,00
48.980,04
11.845.976,71
512.776,50
209.900.000,00
0,29
0,00
0,45
0,12
0,00
0,32
Semi-Annual Rep. Sample
(Fu)
3.653.108,84
191.088,00
284.744.885,96
523.083,31
32.003,50
3.698.119,47
290.861,75
5.039,50
10.450.801,86
34.824,46
998,00
1.005.265,10
4,76
4,44
39,13
0,92
0,64
9,30
0,26
0,21
0,85
9.847,72
870,50
35.106,44
5.882.575,29
194.278,50
72.986.145,73
0,24
0,00
0,43
0,08
0,00
0,27
33
Employee intensity
Asset intensity
Economic growth
Loss Prior Year
mean
median
std dev
mean
median
std dev
mean
median
std dev
mean
median
std dev
0,00
0,00
0,00
2,47
1,07
4,89
0,11
0,01
0,36
0,21
0,00
0,41
77,176
Observations
0,00
0,00
0,00
2,42
1,06
4,56
0,03
0,01
0,05
0,18
0,00
0,39
49,463
0,00
0,00
0,00
2,56
1,10
5,43
0,07
-0,03
0,50
0,25
0,00
0,43
27,713
0,00
0,00
0,00
3,16
1,15
5,72
0,02
0,01
0,05
0,22
0,00
0,42
49,416
0,00
0,00
0,00
1,25
1,00
2,41
0,03
0,01
0,06
0,18
0,00
0,39
27,760
0,00
0,00
0,00
3,44
1,18
5,69
0,02
0,01
0,04
0,23
0,00
0,42
40,960
0,00
0,00
0,00
1,89
1,06
5,85
0,03
0,04
0,07
0,17
0,00
0,37
12,222
Variable definitions:
SG&AC
=
SG&A costs;
Sales
=
sales revenue;
Change_SG&A
=
change in SG&A costs between year t and year t-1;
Relative_Change_SG&A
=
ratio of change in SG&A costs between year t and year t-1 to beginning total assets of year t;
Relative_Change_Sales
=
ratio of change in sales between year t and year t-1 to beginning total assets of year t;
SG&AC/Sales
=
ratio of SG&A costs to sales revenue;
Number_Of_Employees
=
number of employees in year t;
Beginning_Total_Assets
=
total assets of year t-1;
Decline_Observations
=
percentage of observations with a decrease in sales revenue between year t and year t-1;
Successive_Decrease
=
equals 1 when salest-2 > salest-1 > salest, and 0 otherwise;
34
Employee_Intensity
=
ratio of total number of employees to sales revenue;
Asset_Intensity
=
ratio of total assets to sales revenue;
Economic_Growth
=
growth in real gross national income (GNI)
Loss_Prior_Year
=
equals 1 when the firms reports a loss in the previous year, and 0 otherwise;
This table presents descriptive statistics for the full sample and subsamples over the period 2001-2010.
35
TABLE 3
Summary Statistics from Regressions with Log(SG&ACit/SG&Ait-1) as Dependent Variable
I
Variable
Predicted Sign
Basic Modela
II
Extended Modelb
III
Extended Model with
Ecrisis_Dummyc
β0: Constant
β1: Log (Salesit/Salesit-1)
+
Two-Way Interaction Terms
β2: Decrease_Dummy* Log(Salesit/Salesit-1)
-
β3: Log(Salesit/Salesit-1)*Ecrisis_Dummy
Three-Way Interaction Terms
β4: Decrease_Dummy *Log(Salesit/Salesit-1)
*Ecrisis_Dummy
β5: Decrease_Dummy *Log(Salesit/Salesit-1)
*Successive_Decrease
Coeff
(t-stat)
<significance>
0.02
(19.165)
<.000>
0.59
(130.722)
<.000>
Coeff
(t-stat)
<significance>
0.03
(28.329)
<.000>
0.59
(128.701)
<.000>
Coeff
(t-stat)
<significance>
0.03
(27.256)
<.000>
0.60
(107.991)
<.000>
-
-0.14
(-15.993)
<.000>
-
-0.22
(-19.309)
<.000>
-
-0.22
(-15.450)
<.000>
-0.03
(-2.927)
<.003>
+
-
-
+
-
0.10
(7.418)
<.000>
0.00
(0.043)
<.965>
0.11
(7.680)
<.000>
36
β6: Decrease_Dummy *Log(Salesit/Salesit-1)
*Employee_Intensity
-
-
β7: Decrease_Dummy *Log(Salesit/Salesit-1)
* Asset_Intensity
-
-
β8: Decrease_Dummy *Log(Salesit/Salesit-1)
*Economic_Growth
-
-
β9: Decrease_Dummy *Log(Salesit/Salesit-1)
*Loss_Prior_Year
+
-
Main Terms
β10: Ecrisis_Dummy
?
-
β11: Successive_Decrease
?
-
β12: Employee_Intensity
?
-
β13: Asset_Intensity
?
-
β14: Economic_Growth
?
-
β15: Loss_Prior_Year
?
-
355.03
(0.808)
<.419>
-0.01
(-8.026)
<.000>
0.09
(4.343)
<.000>
0.02
(1.405)
<.160>
-
Number of observations
77,176
-0.02
(-7.072)
<.000>
100.78
(2.889)
<.004>
0.00
(11.739)
<.000>
-0.02
(-8.026)
<.000>
-0.07
(-40.646)
<.000>
77,176
Adjusted R-Square
30,6%
33,7%
396.17
(0.900)
<.368>
-0.01
(-7.973)
<.000>
0.09
(4.295)
<.000>
0.02
(1.678)
<.093>
-0.01
(-4.614)
<.000>
-0.02
(-6.502)
<.000>
99.73
(2.860)
<.004>
0.00
(11.665)
<.000>
-0.02
(-7.840)
<.000>
-0.07
(-40.004)
<.000>
77,176
33,8%
37
a
Log(SG&ACit/SG&ACit-1) = β0 + β1*Log(Salesit/Salesit-1) + β2*Decrease_Dummy*Log(Salesit/Salesit-1) + ∈it
b
Log (SG&ACit/SG&ACit-1) = β0 + β1*Log(Salesit/Salesit-1) + β2*Decrease_Dummy*Log(Salesit/Salesit-1) + β5*Decrease_Dummy*Log(Salesit/Salesit-1)
*Successive Decrease + β6*Decrease_Dummy*Log(Salesit/Salesit-1)*Employee_Intensity + β7*Decrease_Dummy*Log(Salesit/Salesit-1)
*Asset_Intensity + β8*Decrease_Dummy*Log(Salesit/Salesit-1)*Economic_Growth + β9*Decrease_Dummy*Log(Salesit/Salesit-1)*Loss_Prior_Year
+ Main Terms + ∈it
c
Log(SG&ACit/SG&ACit-1) = β0 + β1*Log(Salesit/Salesit-1) + β2*Decrease_Dummy*Log(Salesit/Salesit-1) + β3*Log(Salesit/Salesit-1)*Ecrisis_Dummy +
β4*Decrease_Dummy*Log(Salesit/Salesit-1)*Ecrisis_Dummy + β5*Decrease_Dummy*Log(Salesit/Salesit-1)*Successive Decrease +
β6*Decrease_Dummy*Log(Salesit/Salesit-1)*Employee_Intensity + β7*Decrease_Dummy*Log(Salesit/Salesit-1)*Asset_Intensity +
β8*Decrease_Dummy*Log(Salesit/Salesit-1)*Economic_Growth + β9*Decrease_Dummy*Log(Salesit/Salesit-1)*Loss_Prior_Year +
Main Terms + ∈it
38
TABLE 4
Summary Statistics for Split Sample Regressions on The Effect of Economic Crisis
with Log (SG&ACit/SG&ACit-1) as the Dependent Variable
Economic Crisis Samplea
Variable
Basic
Modelc
Extended
Modeld
β0: Constant
0.01
(5.162)
<.000>
0.56
(64.155)
<.000>
Non-Crisis Sampleb
Basic
Modelc
Extended
Modeld
0.02
(9.823)
<.000>
0.57
(63.619)
<.000>
0.02
(21.272)
<.000>
0.60
(119.966)
<.000>
0.03
(26.105)
<.000>
0.60
(117.014)
<.000>
-0.14
(-9.428)
<.000>
-0.22
(-11.847)
<.000>
-0.12
(-11.103)
<.000>
-0.23
(-14.041)
<.000>
β3: Decrease_Dummy*Log(Salesit/Salesit-1)
*Successive_Decrease
-
-
β4: Decrease_Dummy*Log(Salesit/Salesit-1)
*Employee_Intensity
-
β5: Decrease_Dummy*Log(Salesit/Salesit-1)
*Asset_Intensity
-
β6: Decrease_Dummy*Log(Salesit/Salesit-1)
*Economic_Growth
-
β7: Decrease_Dummy*Log(Salesit/Salesit-1)
*Loss_Prior_Year
-
0.10
(4.756)
<.000>
1231.71
(1.771)
<.077>
-0.01
(-4.050)
<.000>
-0.02
(-0.829)
<.407>
-0.01
(-0.502)
<.616>
-0.02
(-4.254)
<.000>
563.90
(3.659)
<.000>
0.00
(8.175)
<.000>
-
β1: Log (Salesit/Salesit-1)
Two-Way Interaction Terms
β2: Decrease_Dummy*Log(Salesit/Salesit-1)
Three-Way Interaction Terms
Main Terms
β8: Successive_Decrease
-
β9: Employee_Intensity
-
β10: Asset_Intensity
-
-
-
-
-
-
-
0.14
(6.707)
<.000>
-367.361
(-0.592)
<.554>
-0.01
(-5.821)
<.000>
-0.04
(-0.209)
<.835>
0.05
(2.798)
<.005>
-0.01
(-4.840)
<.000>
63.63
(1.937)
<.053>
0.00
(9.271)
<.000>
39
β11: Economic_Growth
-
Number of observations
27,713
-0.01
(-3.480)
<.001>
-0.08
(-23.590)
<.000>
27,713
β12: Loss_Prior_Year
-
Adjusted R-Square
26,3%
29,6%
-
49,463
-0.05
(-3.060)
<.002>
-0.06
(-32.618)
<.000>
49,463
33,2%
36,1%
-
a
The economic crisis sample consists of observations from the years 2008 through 2010.
b
The non-crisis sample consists of observations from the years 2001 through 2007.
c
Log(SG&ACit/SG&ACit-1) = β0 + β1*Log(Salesit/Salesit-1) + β2*Decrease_Dummy*Log(Salesit/Salesit-1) +
∈it
d
Log(SG&ACit/SG&ACit-1) = β0 + β1*Log(Salesit/Salesit-1) + β2*Decrease_Dummy*Log(Salesit/Salesit-1) +
β5*Decrease_Dummy*Log(Salesit/Salesit-1)*Successive Decrease + β6*Decrease_Dummy
*Log(Salesit/Salesit-1)*Employee_Intensity + β7*Decrease_Dummy*Log(Salesit/Salesit-1)
*Asset_Intensity + β8*Decrease_Dummy*Log(Salesit/Salesit-1)*Economic_Growth +
β9*Decrease_Dummy*Log(Salesit/Salesit-1)*Loss_Prior_Year + Main Terms + ∈it
40
TABLE 5
Summary Statistics from Regressions on the Updating of Information with Log(SG&ACit/SG&Ait-1) as Dependent Variable
I
Variable
Predicted Sign
Extended Model with
Ifr_Dummy (Compustat)a
β0: Constant
β1: Log (Salesit/Salesit-1)
+
Two-Way Interaction Terms
β2: Decrease_Dummy* Log(Salesit/Salesit-1)
-
β3: Log(Salesit/Salesit-1)*Ifr_Dummy
-
Three-Way Interaction Terms
β4:Decrease_Dummy *Log(Salesit/Salesit-1)
*Ifr_Dummy
+
β5: Decrease_Dummy *Log(Salesit/Salesit-1)
*Successive_Decrease
+
Coeff
(t-stat)
<significance>
0.02
(13.589)
<.000>
0.55
(57.097)
<.000>
II
Extended Model with
Ifr_Dummy (Fu-method)a
Coeff
(t-stat)
<significance>
0.04
(12.462)
<.000>
0.61
(46.359)
<.000>
-0.15
(-8.144)
<.000>
0.03
(2.845)
<.004>
-0.28
(-10.757)
<.000>
-0.01
(-0.564)
<.573>
-0.08
(-3.973)
<.000>
0.09
(6.724)
<.000>
0.01
(0.470)
<.639>
0.10
(5.193)
<.000>
41
β6: Decrease_Dummy *Log(Salesit/Salesit-1)
*Employee_Intensity
-
β7: Decrease_Dummy *Log(Salesit/Salesit-1)
* Asset_Intensity
-
β8: Decrease_Dummy *Log(Salesit/Salesit-1)
*Economic_Growth
-
β9: Decrease_Dummy *Log(Salesit/Salesit-1)
*Loss_Prior_Year
+
Main Terms
β10: Ifr_Dummy
?
β11: Successive_Decrease
?
β12: Employee_Intensity
?
β13: Asset_Intensity
?
β14: Economic_Growth
?
β15: Loss_Prior_Year
?
Number of observations
Adjusted R-Square
917.30
(2.057)
<.004>
-0.01
(-8.256)
<.000>
0.08
(3.881)
<.000>
0.02
(1.904)
<.057>
1078.39
(2.130)
<.033>
-0.01
(-8.094)
<.000>
-0.07
(-3.103)
<.002>
0.05
(3.211)
<.001>
0.02
(10.336)
<.000>
-0.02
(-6.417)
<.000>
41.10
(1.172)
<.241>
0.00
(8.359)
<.000>
-0.01
(-2.760)
<.006>
-0.07
(-41.475)
<.000>
77,176
-0.00
(-0.398)
<.691>
-0.02
(-4.449)
<.000>
25.83
(0.628)
<.530>
0.00
(6.666)
<.000>
-0.00
(-1.087)
<.277>
-0.08
(-32.907)
<.000>
53,182 (Japan excluded)
34,0%
30,5%
42
a
Log(SG&ACit/SG&ACit-1) = β0 + β1*Log(Salesit/Salesit-1) + β2*Decrease_Dummy*Log(Salesit/Salesit-1) + β3*Log(Salesit/Salesit-1)*Ifr_Dummy +
β4*Decrease_Dummy*Log(Salesit/Salesit-1)*Ifr_Dummy + β5*Decrease_Dummy*Log(Salesit/Salesit-1)*Successive Decrease +
β6*Decrease_Dummy*Log(Salesit/Salesit-1)*Employee_Intensity + β7*Decrease_Dummy*Log(Salesit/Salesit-1)*Asset_Intensity +
β8*Decrease_Dummy*Log(Salesit/Salesit-1)*Economic_Growth + β9*Decrease_Dummy*Log(Salesit/Salesit-1)*Loss_Prior_Year +
Main Terms + ∈it
43
TABLE 6
Summary Statistics for Split Sample Regressions on The Effect of Updating Information
(Compustat) with Log (SG&ACit/SG&ACit-1) as the Dependent Variable
Quarterly IR Samplea
Semi-Annual IR Sampleb
Variable
Basic
Modelc
Extended
Modeld
β0: Constant
0.03
(20.203)
<.000>
0.58
(96.066)
<.000>
0.04
(25.103)
<.000>
0.58
(94.832)
<.000>
0.01
(6.259)
<.000>
0.56
(88.078)
<.000>
0.02
(13.066)
<.000>
0.55
(86.522)
<.000>
-0.14
(-12.406)
<.000>
-0.24
(-15.227)
<.000>
-0.03
(-3.048)
<.002>
-0.15
(-10.110)
<.000>
β3: Decrease_Dummy*Log(Salesit/Salesit-1)
*Successive_Decrease
-
-
β4: Decrease_Dummy*Log(Salesit/Salesit-1)
*Employee_Intensity
-
β5: Decrease_Dummy*Log(Salesit/Salesit-1)
*Asset_Intensity
-
β6: Decrease_Dummy*Log(Salesit/Salesit-1)
*Economic_Growth
-
β7: Decrease_Dummy*Log(Salesit/Salesit-1)
*Loss_Prior_Year
-
0.09
(4.550)
<.000>
939.53
(1.730)
<.084>
-0.01
(-7.850)
<.000>
-0.76
(-3.952)
<.000>
0.04
(2.078)
<.038>
-0.02
(-5.443)
<.000>
31.69
(0.755)
<.450>
0.00
(6.489)
<.000>
-
β1: Log (Salesit/Salesit-1)
Basic
Modelc
Extended
Modeld
Two-Way Interaction Terms
β2: Decrease_Dummy*Log(Salesit/Salesit-1)
Three-Way Interaction Terms
Main Terms
β8: Successive_Decrease
-
β9: Employee_Intensity
-
β10: Asset_Intensity
-
-
-
-
-
-
-
0.08
(5.005)
<.000>
142.95
(0.139)
<.890>
0.01
(2.545)
<.011>
1.013
(7.365)
<.000>
0.00
(0.077)
<.938>
-0.01
(-5.338)
<.000>
172.00
(1.769)
<.077>
0.00
(5.431)
<.000>
44
β11: Economic_Growth
-
Number of observations
49,416
0.01
(6.489)
<.000>
-0.08
(-32.029)
<.000>
49,416
β12: Loss_Prior_Year
-
Adjusted R-Square
27,3%
30,4%
-
27,759
-0.06
(-4.215)
<.000>
-0.06
(-30.003)
<.000>
27,759
40,0%
43,5%
-
a
The quarterly IR sample consists of observations that provide interim reporting every quarter.
b
The semi-annual IR sample consists of observations that provide interim reporting every half year.
c
Log(SG&ACit/SG&ACit-1) = β0 + β1*Log(Salesit/Salesit-1) + β2*Decrease_Dummy*Log(Salesit/Salesit-1) +
∈it
d
Log(SG&ACit/SG&ACit-1) = β0 + β1*Log(Salesit/Salesit-1) + β2*Decrease_Dummy*Log(Salesit/Salesit-1) +
β5*Decrease_Dummy*Log(Salesit/Salesit-1)*Successive Decrease + β6*Decrease_Dummy
*Log(Salesit/Salesit-1)*Employee_Intensity + β7*Decrease_Dummy*Log(Salesit/Salesit-1)
*Asset_Intensity + β8*Decrease_Dummy*Log(Salesit/Salesit-1)*Economic_Growth +
β9*Decrease_Dummy*Log(Salesit/Salesit-1)*Loss_Prior_Year + Main Terms + ∈it
45
TABLE 7
Summary Statistics for Split Sample Regressions on The Effect of Updating Information
(Fu Method) with Log (SG&ACit/SG&ACit-1) as the Dependent Variable
Quarterly IR Samplea
Variable
Basic
Modelc
Extended
Modeld
0.03
(20.615)
<.000>
0.60
(109.755)
<.000>
0.02
(4.737)
<.000>
0.60
(27.514)
<.000>
0.04
(6.804)
<.000>
0.61
(27.170)
<.000>
-0.16
(-15.486)
<.000>
-0.28
(-20.602)
<.000>
-0.18
(-4.510)
<.000>
-0.30
(-5.649)
<.000>
β3: Decrease_Dummy*Log(Salesit/Salesit-1)
*Successive_Decrease
-
-
β4: Decrease_Dummy*Log(Salesit/Salesit-1)
*Employee_Intensity
-
β5: Decrease_Dummy*Log(Salesit/Salesit-1)
*Asset_Intensity
-
β6: Decrease_Dummy*Log(Salesit/Salesit-1)
*Economic_Growth
-
β7: Decrease_Dummy*Log(Salesit/Salesit-1)
*Loss_Prior_Year
-
0.12
(6.895)
<.000>
1652.40
(3.532)
<.000>
-0.01
(-9.886)
<.000>
-0.73
(-3.519)
<.000>
0.04
(3.013)
<.003>
0.07
(0.984)
<.325>
2228,84
(1,258)
<.208>
-0.01
(-1.029)
<.303>
-0.79
(-2.083)
<.037>
0.06
(1.087)
<.277>
-0.02
(-5.038)
<.000>
576.08
(5.839)
<.000>
0.00
(8.487)
<.000>
-
β0: Constant
β1: Log (Salesit/Salesit-1)
Basic
Modelc
0.02
(19.456)
<.000>
0.59
(108.808)
<.000>
Extended
Modeld
Semi-Annual IR Sampleb
Two-Way Interaction Terms
β2: Decrease_Dummy*Log(Salesit/Salesit-1)
Three-Way Interaction Terms
Main Terms
β8: Successive_Decrease
-
β9: Employee_Intensity
-
β10: Asset_Intensity
-
-
-
-
-
-
-
-0.02
(-1.399)
<.0162>
-31.02
(-0.429)
<.668>
0.00
(2.300)
<.021>
46
β11: Economic_Growth
-
Number of observations
40,960
0.00
(0.031)
<.976>
-0.08
(-37.433)
<.000>
40,960
β12: Loss_Prior_Year
-
Adjusted R-Square
37,7%
42,1%
-
12,222
-0.04
(-0.746)
<.456>
-0.10
(-10.266)
<.000>
12,222
13,6%
15,3%
-
a
The quarterly IR sample consists of observations that provide interim reporting every quarter.
b
The semi-annual IR sample consists of observations that provide interim reporting every half year.
c
Log(SG&ACit/SG&ACit-1) = β0 + β1*Log(Salesit/Salesit-1) + β2*Decrease_Dummy*Log(Salesit/Salesit-1) +
∈it
d
Log(SG&ACit/SG&ACit-1) = β0 + β1*Log(Salesit/Salesit-1) + β2*Decrease_Dummy*Log(Salesit/Salesit-1) +
β5*Decrease_Dummy*Log(Salesit/Salesit-1)*Successive Decrease + β6*Decrease_Dummy
*Log(Salesit/Salesit-1)*Employee_Intensity + β7*Decrease_Dummy*Log(Salesit/Salesit-1)
*Asset_Intensity + β8*Decrease_Dummy*Log(Salesit/Salesit-1)*Economic_Growth +
β9*Decrease_Dummy*Log(Salesit/Salesit-1)*Loss_Prior_Year + Main Terms + ∈it
47
TABLE 8
Summary Statistics from Regressions with Log(SG&ACit/SG&Ait-1) as Dependent Variable, including Country Dummies
I
Variable
Predicted Sign
β0: Constant
β1: Log (Salesit/Salesit-1)
+
Two-Way Interaction Terms
β2: Decrease_Dummy* Log(Salesit/Salesit-1)
-
β3: Log(Salesit/Salesit-1)*Ecrisis_Dummy
resp. Ifr_Dummy
Three-Way Interaction Terms
β4: Decrease_Dummy *Log(Salesit/Salesit-1)
*Ecrisis_Dummy resp. Ifr_Dummy
β5: Decrease_Dummy *Log(Salesit/Salesit-1)
*Successive_Decrease
-
+
+
II
III
Extended Model with
Extended Model with
Extended Model with
Ecrisis_Dummya
Ifr_Dummy (Compustat)b
Ifr_Dummy (Fu-method)b
Coeff
(t-stat)
<significance>
0.02
(7.594)
<.000>
0.58
(56.175)
<.000>
Coeff
(t-stat)
<significance>
0.02
(8.527)
<.000>
0.55
(49.207)
<.000>
Coeff
(t-stat)
<significance>
0.04
(6.050)
<.000>
0.60
(29.904)
<.000>
-0.23
(-8.949)
<.000>
-0.04
(-2.781)
<.005>
-0.15
(-6.508)
<.000>
0.03
(1.626)
<.104>
-0.27
(-5.915)
<.000>
-0.11
(-2.958)
<.003>
0.04
(1.516)
<.130>
0.11
(5.245)
<.000>
-0.13
(-4.258)
<.000>
0.09
(4.207)
<.000>
0.10
(1.493)
<.136>
0.07
(1.300)
<.194>
48
β6: Decrease_Dummy *Log(Salesit/Salesit-1)
*Employee_Intensity
-
β7: Decrease_Dummy *Log(Salesit/Salesit-1)
* Asset_Intensity
-
β8: Decrease_Dummy *Log(Salesit/Salesit-1)
*Economic_Growth
-
β9: Decrease_Dummy *Log(Salesit/Salesit-1)
*Loss_Prior_Year
+
Main Terms
β10: Ecrisis_Dummy resp. Ifr_Dummy
?
β11: Successive_Decrease
?
β12: Employee_Intensity
?
β13: Asset_Intensity
?
β14: Economic_Growth
?
β15: Loss_Prior_Year
?
Country dummies
?
821.64
(0.822)
<.411>
-0.00
(-0.535)
<.593>
0.13
(5.424)
<.000>
0.01
(0.565)
<.572>
1836.63
(1.778)
<.075>
0.00
(0.003)
<.998>
0.10
(4.272)
<.000>
0.01
(0.555)
<.579>
1070.42
(0.715)
<.475>
-0.01
(-1.618)
<.106>
-0.07
(-1.920)
<.055>
0.05
(1.147)
<.252>
-0.01
(-1.950)
<.051>
-0.01
(-2.439)
<.015>
-47.92
(-1.163)
<.245>
0.00
(4.582)
<.000>
-0.01
(-2.838)
<.005>
-0.07
(-23.850)
<.000>
included
0.00
(-0.071)
<.943>
-0.01
(-2.935)
<.003>
-37.367
(-0.906)
<.365>
0.00
(4.633)
<.000>
-0.01
(-3.044)
<.002>
-0.07
(-24.251)
<.000>
included
-0.01
(-0.607)
<.544>
-0.02
(-1.359)
<.174>
-34.79
(-0.536)
<.592>
0.00
(2.428)
<.015>
-0.01
(-1.426)
<.154>
-0.09
(-13.131)
<.000>
included
49
Number of observations
40,712
40,712
16,718
Adjusted R-Square
26,1%
26,1%
17,6%
a
Log(SG&ACit/SG&ACit-1) = β0 + β1*Log(Salesit/Salesit-1) + β2*Decrease_Dummy*Log(Salesit/Salesit-1) + β3*Log(Salesit/Salesit-1)*Ecrisis_Dummy +
β4*Decrease_Dummy*Log(Salesit/Salesit-1)*Ecrisis_Dummy + β5*Decrease_Dummy*Log(Salesit/Salesit-1)*Successive Decrease +
β6*Decrease_Dummy*Log(Salesit/Salesit-1)*Employee_Intensity + β7*Decrease_Dummy*Log(Salesit/Salesit-1)*Asset_Intensity +
β8*Decrease_Dummy*Log(Salesit/Salesit-1)*Economic_Growth + β9*Decrease_Dummy*Log(Salesit/Salesit-1)*Loss_Prior_Year +
Main Terms (including Country Dummies) + ∈it
b
Log(SG&ACit/SG&ACit-1) = β0 + β1*Log(Salesit/Salesit-1) + β2*Decrease_Dummy*Log(Salesit/Salesit-1) + β3*Log(Salesit/Salesit-1)*Ifr_Dummy +
β4*Decrease_Dummy*Log(Salesit/Salesit-1)*Ifr_Dummy + β5*Decrease_Dummy*Log(Salesit/Salesit-1)*Successive Decrease +
β6*Decrease_Dummy*Log(Salesit/Salesit-1)*Employee_Intensity + β7*Decrease_Dummy*Log(Salesit/Salesit-1)*Asset_Intensity +
β8*Decrease_Dummy*Log(Salesit/Salesit-1)*Economic_Growth + β9*Decrease_Dummy*Log(Salesit/Salesit-1)*Loss_Prior_Year +
Main Terms (including Country Dummies) + ∈it
50