Additional Evidence on the Sticky Behavior of Costs

Chandra Subramaniam
Marcia L. Weidenmier*

M. J. Neeley School of Business

Texas Christian University

February 10, 2003

*
Corresponding author:
Marcia L. Weidenmier
M.J. Neeley School of Business
Texas Christian University
TCU Box 298530
Fort Worth, TX 76129
Email: M.Weidenmier@tcu.edu

Financial support from CTAEC is greatly acknowledged.

 Additional Evidence on the Sticky Behavior of Costs

Abstract

Recent studies in cost behavior find that selling, general, and administrative costs

(SG&A) and cost of good sold (CGS) are sticky, i.e., costs increase more with activity

increases than they decrease in response to equivalent activity decreases. We examine

whether the magnitude of the activity changes is the driving force behind this “sticky”

phenomenon. Specifically, we explore how costs behave for different ranges of sales

activity changes. SG&A and CGS costs do not exhibit sticky cost behavior for small

revenue changes, implying linearity of costs with respect to revenue. However, when

revenue changes by more than ten percent, costs exhibit sticky behavior. We also find

inter-industry differences in the stickiness behavior of both SG&A and CGS costs as well

as in the determinants of sticky behavior.

 Additional Evidence on the Sticky Behavior of Costs

I. INTRODUCTION

The traditional model of cost behavior assumes that costs are strictly proportional to

activity levels (Noreen 1991). Recent research, however, has shown that not only do costs not

change proportionately with activity, but costs may also be sticky, i.e., they respond differentially

to upward and downward changes in activity. To gain a better understanding of the prevalence

of sticky behavior, this paper examines whether this sticky cost behavior is dependent on the

magnitude of activity change. We test this proposition on selling, general, and administrative

costs (SG&A) and cost of good sold (CGS). This study also explores how industry affiliation

may differentially affect the sticky cost behavior.

Empirical research consistently reports that costs do not change proportionately with

activity (Banker and Johnston, 1993; Noreen and Soderstrom, 1994, 1997). In addition,

Anderson, Banker, and Janakiraman (2003) and Anderson, Banker, Chen, and Janakiraman

(2001) (hereafter, ABJ and ABCJ, respectively) find the evidence of stickiness for SG&A and

CGS costs in large samples of firms from multiple industries.

However, studies examining small samples of firms from a single industry report mixed

results. Noreen and Soderstrom (1997) generally report no evidence of sticky cost behavior for a

sample of hospitals. On the other hand, Balakrishnan, Peterson, and Soderstrom (2003)

(hereafter, BPS) do find that costs exhibit stickiness but only when resources are strained at

therapy clinics. In fact, when there is excess capacity, the response to large decreases in activity

is greater than the response to large increases in activity-the opposite response predicted by

stickiness. Furthermore, the authors suggest caution when applying ABJ’s conclusion that costs

are sticky (p. 3).

1
 This paper attempts to resolve the conflicting results of prior literature. We first

hypothesize that sticky behavior is a function of the magnitude of the activity change. Unlike

small activity changes, large activity changes force managers to alter the cost structure of the

firm shifting the total cost line. Moreover, managers are more likely to change costs for activity

increases than for activity decreases (Cooper and Kaplan, 1998; BPS, 2003). We hypothesize

that stickiness is the result of managers’ asymmetrical response to large demand changes.

BPS (2003) examines the effect of the magnitude of the change on stickiness, but do not

find stickiness for sample as a whole. Their ability to examine magnitude, however, is limited

due to the size of the activity changes in their sample (a large change is defined as ± 3 percent).

Our sample of firms, with activity changes as large as ± 30 percent, allows us to make a better

assessment of whether (and how) activity magnitude affects sticky behavior.

The strongest evidence of sticky behavior is found in samples consisting of firms from

multiple industries (ABJ, 2003; ABCJ, 2001), while samples from one industry (Noreen and

Soderstrom, 1994, 1997) exhibit limited sticky cost behavior. Prior research shows that each

industry operates in a different production environment causing accounting variables to be

industry specific, making the relationship of costs to activity changes industry-specific (Ely,

1991). Accordingly, we also hypothesize that industries exhibit different patterns of sticky

behavior.

To evaluate if magnitude of activity change is driving sticky cost behavior, we examine

the level of stickiness as a function of the absolute magnitude of activity changes by stratifying

the activity changes into six bands. We also determine the magnitude of activity changes at

which costs exhibit sticky behavior in different industries. Finally, we study inter-industry

differences in the determinants of stickiness, by examining five determinants of sticky cost

2
behavior, including fixed assets, employees, inventory, and interest expense, which are likely to

vary by industry.

The results show that SG&A and CGS costs do not exhibit sticky cost behavior for small

revenue changes. However, when revenue changes by more than ten percent, SG&A and CGS

costs exhibit sticky behavior. In addition, we find that manufacturing is the “stickiest” industry

due to high levels of fixed assets and inventory, while merchandising is the “least sticky”

industry due its highly competitive environment. The financial and service industries do exhibit

some level of stickiness with interest expense driving stickiness in the financial industry, and

employee and inventory intensity driving stickiness in the service industry.

These findings help reconcile the results of prior studies. In a large sample from multiple

industries, we do find stickiness in the aggregate, supporting ABJ and ABCJ. Yet, a detailed

examination reveals that stickiness is exhibited for only larger magnitudes of activity changes as

hypothesized by BPS. We also find that industry affiliation affects the presence of stickiness.

These results suggest that the magnitude of the activity changes as well as industry affiliation

must be considered when studying cost stickiness.

The remainder of the paper is organized as follows. Section 2 discusses relevant prior

literature and develops the hypotheses. Section 3 describes the sample and the models used to

perform hypothesis testing. Section 4 presents the results. Section 5 concludes by discussing the

implications of the findings for researchers and managers.

II. Background and Overview

The traditional model of cost behavior assumes proportionality between costs and activity

as long as costs are in the relevant range, implying that a one percent increase in activity is

3
accompanied by a one percent increase in cost. This assumption of strict cost proportionality is

increasingly coming under scrutiny. Several empirical studies report that costs do not change

proportionately with activity and exhibit increasing returns to scale1 (Banker and Johnston, 1993;

Noreen and Soderstrom, 1994, 1997; BPS, 2003). Noreen and Soderstrom (1997) find that a

fixed cost model predicts future costs better than an activity-based cost prediction model. They

also test the “folklore” that costs increase more in response to activity increases than they

decrease in response to activity decreases, but find insignificant results for a sample of

Washington state hospitals.

ABJ (2003) and ABCJ (2001) find support for this “folklore,” which they call stickiness,

in a large sample of firms in different industries. Both papers argue that this phenomenon is

created by the asymmetrical response of managers. Specifically, if the demand is higher than

expected, resources are immediately strained (if the firm is currently operating close to capacity)

and managers are likely to increase committed resources (and related costs) to fill customer

orders. If demand is lower than expected, slack resources exist which managers may be slow to

eliminate from the system (ABJ, p. 6).

BPS extends ABJ and ABCJ by examining the effect of magnitude of activity change and

capacity utilization on sticky behavior for a sample of therapy clinics. They find that cost

stickiness depends on capacity utilization. In fact, costs exhibit stickiness only when the

resources are strained. However, their ability to examine different levels of activity changes is

limited due to the size of the activity change in their sample (a large change is defined as greater

than ± 3 percent).2

1
Increasing returns to scale implies that a one percent increase in activity will be accompanied by less than one
percent increase in cost.
2
75 percent of all observations in their sample change less than ± 7 percent.
4
 In this paper, we argue, similar to BPS, that stickiness is the result of the magnitude of

the activity changes. While small activity changes can be managed using the firm’s existing

resources, large activity changes force managers to alter the cost structure of the firm shifting the

total cost line. Upward shifts occur when managers increase costs and firm capacity to

accommodate the activity increase, while downward shifts occur when managers reduce costs

and surplus capacity to accommodate the activity decrease. While large activity increases

usually result in immediate cost increases, large activity decreases may not result in immediate

cost reductions (even if significant slack exists), because firms may not want to or can’t reduce

assets, employees, and/or other costs in the short term. In fact, Cooper and Kaplan (1998) and

BPS (2003) suggest that managers are more likely to change costs when activity increases than

when activity decreases due to the presumption (and hope) that the activity decrease is

temporary. Therefore, we hypothesize that stickiness is the result of managers’ asymmetrical

response to large demand changes. Therefore, our first hypothesis is as follows:

Hypothesis 1: Sticky cost behavior exists for large changes in the absolute magnitude of sales

activity.

Cost behavior is likely to differ between industries because each industry has its own

production and operational environment with its associated technology, product markets, and

regulatory environments, causing accounting variables to capture/reflect the environment

differently in each industry (Ely, 1991). Each industry has its own characteristic level of

inventory, property, plant and equipment, and labor costs, as well as operating ratios. For

example, Lazere (1996) finds the SG&A-to-revenue ratio is industry-specific. Moreover, due to

5
the lack of authoritative guidance, the components of SG&A and CGS vary dramatically from

one company to another, i.e., one company’s SG&A may be another company’s CGS (Mintz,

1994; Lazere, 1995). Therefore, we expect firms in different industries to not only have different

relevant ranges (range where costs change linearly with activity) and levels of unused capacity or

slack, but also inter-industry differences in the stickiness in SG&A and CGS. To understand the

sticky phenomenon, we examine the stickiness of SG&A and CGS costs individually as well as

jointly by industry. Accordingly, we make the following hypothesis:

Hypothesis 2A: The magnitude of the sticky cost behavior of SG&A and CGS to changes in the

absolute magnitude of sales activity is different across industries.

ABCJ (2001) provide evidence that fixed cost intensity, employee intensity, future sales

prospects, revenue declines in the previous period, GNP growth, industry concentration ratio,

and revenue volatility are among the determinants that affect the sticky behavior of CGS costs

for service firms. However, their sample includes both financial and service firms. We believe

that financial firms, operating in a regulatory environment, have different production

environments and therefore different cost behaviors than service firms. Our interest is in

identifying whether different determinants drive the sticky behavior of costs in different

industries. Accordingly, we make the following hypothesis:

Hypothesis 2B: The determinants of the sticky cost behavior of SG&A and CGS are different

across industries.

6
III. SAMPLE SELECTION AND EMPIRICAL DESIGN

Sample Selection

The data set includes 22 years of annual data from Compustat during the period 1979-

2000. The data items include net sales revenue (annual item #12), CGS (#41), and SG&A

(#189). Industry classification is implemented using one-digit SIC code industry groups for the

following four groups: manufacturing (SIC codes 2 and 3), merchandising (SIC code 5), service

firms (SIC codes 4, 7, and 8), and financial services (SIC code 6).3

To remain in the sample, observations must have SG&A, CGS, and sales revenue in the

current and preceding year. We exclude observations in which the SG&A or CGS exceeds sales

revenue and when sales revenue changed by more than 30 percent to remove effects of

acquisitions, mergers, divestitures, and bankruptcy. The sample selection process above is

consistent with ABJ (2003) and ABCJ (2001). In addition, each firm must have at least three

usable observations to be included. This criterion was included to eliminate firms that are

relatively new or are not consistently maintained by Compustat. Using these criteria, the sample

consists of 82,118 observations for 9,592 firms with an average of 8.6 observations per firm.4

Table 1, Panel A presents descriptive statistics for sales revenue, SG&A, CGS and Total

Costs (sum of SG&A and CGS). Mean (median) sales revenue is $1,294 ($119) million. The

mean (median) of SG&A-to-revenue is 24.43 (21.54) percent and CGS-to-revenue is 63.77

(66.42) percent. The magnitude of sales revenue and SG&A in our sample is similar to ABJ

(2003). Panel B provides information on the percentage of firm-years when the sales revenue,

3
Within the one-digit SIC code, the production environments, including the technology, product markets, and
regulatory environment, are assumed to be highly correlated. While two, three, and four-digit SIC codes may
provide finer partitions in classification, they 1) significantly increase the number of the industries to be studied
without substantial addition to our understanding and 2) decrease the sample size for each industry. In addition, we
exclude utilities (SIC 4900-4999) from the sample of service firms because they had been regulated during the
1980s and most of the 1990s.
4
This sample is 27 percent larger than the sample used in ABJ (2003).
7
SG&A, CGS, and Total Costs declined from the previous year. Revenue declined in 30.36

percent of firm-year observations by an average of 10.01 percent. SG&A (CGS) declined in

27.13 (31.70) percent of firm-year observations by an average of 11.88 (11.66) percent.

Table 2, Panel A presents descriptive statistics for sales revenue, SG&A, CGS, and Total

Costs by industry. Consistent with the population of firms on Compustat, more than half of our

sample is manufacturing firms with the remainder of our sample evenly distributed among the

remaining three industries. On average, manufacturing and merchandising firms are larger than

service and financial firms. The proportion of SG&A-to-revenues is highest, as expected, for

service firms at 28.03 percent and lowest for merchandising firms at 22.57 percent. The

proportion of CGS-to-revenues is highest for merchandising firms at 71.44 percent and lowest

for financial firms at 52.28 percent. The CGS for financial firms represents the interest expense

and provision for loan losses as reported by banks and savings and loan firms. Panel B shows

negative changes in revenues occurred more frequently for manufacturing firms during this

period (31.81 percent), while financial services firms reported the lowest number of negative

changes in revenues (22.85 percent).

Model Specification

Hypotheses 1 suggests that sticky cost behavior is observed only with large changes in

the absolute magnitude in sales activity. To test this hypothesis, we stratify the activity changes

into six bands in order to identify the level of change in activity that generates sticky behavior.

We first determine how cost responds to absolute changes in revenue and then test for a different

response to revenue decreases. Accordingly, we estimate Equation 1, an expanded version of the

(log) model used by ABJ and ABCJ, for SG&A, CGS, and TOTAL (SG&A + CGS) for all firm-

years:

8
  DVi,t  6
 Re venuei, t  6  Revenuei, t 
log  
 DVi,t − 1 
= β 0 +
∑
S =1
β S * RSi, t * log
  + ∑ β (S + 6) * DSi, t * RSi, t * log 
 Revenuei, t - 1  S =1
 + εi , t (1)
 Revenuei, t - 1 

where:

DV = SG&A, CGS, or TOTAL (CGS + SG&A)

R1it = 1 if percent change in sales revenue is [-0.05, 0.05]

R2it = 1 if percent change in sales revenue is [-0.10, -0.05) or (0.05, 0.10]

R3it = 1 if percent change in sales revenue is [-0.15, -0.10) or (0.10, 0.15]

R4it = 1 if percent change in sales revenue is [-0.20, -0.15) or (0.15, 0.20]

R5it = 1 if percent change in sales revenue is [-0.25, -0.20) or (0.20, 0.25]

R6it = 1 if percent change in sales revenue is [-0.30, -0.25) or (0.25, 0.30]

D1it = 1 if percent change in sales revenue is [-0.05, 0)

D2it = 1 if percent change in sales revenue is [-0.10, -0.05)

D3it = 1 if percent change in sales revenue is [-0.15, -0.10)

D4it = 1 if percent change in sales revenue is [-0.20, -0.15)

D5it = 1 if percent change in sales revenue is [-0.25, -0.20)

D6it = 1 if percent change in sales revenue is [-0.30, -0.25)

The above model is interpreted in the following way. Changes in revenue are stratified

into six bands of increasing amounts of absolute change in revenue.5 The coefficient β1

measures the percentage increase in the dependent variable with a one percent increase in sales

revenue when the one-year absolute change in revenue is between zero and five percent. A

5
Using 5 percent changes in revenue is an arbitrary choice. However, choosing other ranges gives similar
qualitative results. For the full sample, 24.86 percent of the observations have absolute change in revenues of less
than 5 percent, 23.12 percent have absolute change in revenues of between 5 percent and 10 percent, 19.21 percent
between 10 percent and 15 percent, 14.46 percent between 15 percent and 20 percent, 10.61 percent between 20
percent and 25 percent, and 7.78 percent between 25 percent and 30 percent. Within each range, on average, 30.3
percent of the observations are negative.
9
negative coefficient on the corresponding term β7 indicates that the dependent variable exhibits

stickiness in this range (zero and five percent). The remaining coefficients β2 through β6 and

their corresponding “sticky” coefficients β8 through β12 follow the same pattern for the indicated

revenue changes.

The “sticky” coefficients β7 through β12 indicate the magnitude of the activity change that

causes managers to change the cost structure of the firm. Alternatively, they indicate how much

surplus capacity firms generally maintain in the given industry. Hypothesis 2A states that the

sticky cost behavior between SG&A and sales revenue as well as between CGS and sales

revenue differs significantly across industries. To test this hypothesis, we examine Equation 1

by industry.

Finally, we examine whether different determinants drive the sticky behavior of costs

(SG&A, CGS and Total Costs) in manufacturing, merchandising, service, and financial firms.

We selected fixed asset intensity, employee intensity, decrease in revenue for two consecutive

years, inventory intensity, and interest ratio as our determinants of choice. For example, we

hypothesize that while fixed asset intensity may be driving the sticky behavior of all costs in

manufacturing firms, the same cannot be said for less asset-intensive service firms. Similarly,

interest expense should drive the sticky behavior of costs for financial firms, but not in the other

three industries. To test whether the selected determinants affect sticky cost behavior differently

across industries (Hypothesis 2B), we implement the following model:

10
  DVi,t   Re venuei, t 
log   = γ 0 + γ 1 * log 
 DVi,t − 1   Revenuei, t - 1 
γ 3 * Decrease_Dummy * log 
γ 4 * Decrease_Dummy * log 
γ 5 * Decrease_Dummy * log 
γ 6 * Decrease_Dummy * log 
γ 7 * Decrease_Dummy * log 
 Revenuei, t 
 + γ 2 * Decrease_Dummy * log  +
 Revenuei, t - 1 
 Revenuei, t 
* Fixed Asset Intensity +
 Revenuei, t - 1 
 Revenuei, t 
* Employee Intensity +
 Revenuei, t - 1 
 Revenuei, t 
* Second Decrease +
 Revenuei, t - 1 
 Revenuei, t 
* Inventory Intensity +
 Revenuei, t - 1 
 Revenuei, t 
* Interest Ratio + εi, t (2)
 Revenuei, t - 1 

As in Equation 1, DV equals SG&A, CGS, or Total Costs (sum of SG&A and CGS) and

Decrease_Dummy equals one if revenue in period t is less than revenue in period t-1. We define

Fixed Asset Intensity as the log of the ratio of the net book value of fixed assets (Compustat #8)

to sales revenue and Employee Intensity as the log of the ratio of the number of employees (#29)

to sales revenue, consistent with ABCJ.6 The Second Decrease variable equals one if revenue

declined in the previous period as well. Inventory Intensity is the log of the ratio of the

inventory (#42) to sales revenue. Finally, Interest Ratio is the log of the ratio of interest expense

(#15) to sales revenue for manufacturing, merchandising, and service firms. For financial firms,

however, the Interest Ratio equals the log of the ratio of interest expense (#339) to sales revenue.

IV. EMPIRICAL RESULTS

Test of stickiness of costs for the full sample

For consistency with prior literature, we first test for the presence of stickiness with our

model (Equation 1) by collapsing the revenue change bands - setting R1 = R2 = ....= R6 and D1 =

6
Since Asset Intensity is also usually defined as the log of the ratio of the net book value of fixed assets (Compustat
#8) to total asset (#6), we repeated our tests with this definition with no qualitative change in the results.
11
D2 = …. = D6. Table 3, Panel A provides the results of this simplified regression model.7 When

the dependent variable is SG&A this regression is similar to the study by ABJ (2003 Table 2).

The estimated coefficient for α1 is 0.6970 (t-value = 81.59) implying that SG&A increases by

0.7 percent for every 1 percent increase in revenues. The estimated coefficient for α2 is –0.1210

(t-value = 7.79). In essence, SG&A decreases by 0.58 percent for every 1 percent decrease in

revenues. The coefficients α1 and α1+ α2 are both also significantly lower than one implying

that SG&A costs are not proportional to changes in revenues (and exhibit increasing returns to

scale). These results are consistent with ABJ (2003) showing sticky behavior of SG&A costs.

When the dependent variable is CGS, the estimated coefficient for α1 is 1.0139 (t-value =

155.39) implying that CGS increases by 1.01 percent for every 1 percent increase in revenues.

The estimated coefficient for α2 is –0.0715 (t-value = 6.03). In essence, CGS decreases by 0.94

percent for every 1 percent decrease in revenues. This result provides evidence of the sticky

behavior of CGS costs. Lazere (1995, 1996) states that firms inconsistently classify some

expenses sometimes as SG&A and other times as CGS costs. Therefore, we repeat the

regression model using Total Cost changes (SG&A + CGS) as the dependent variable. The

estimated coefficient for α1 is 0.9266 (t-value = 191.53) and the estimated coefficient for α2 is

–0.0814 (t-value = 9.27), supporting the sticky behavior of Total Costs. Hence, the results in

Table 3, Panel A support the sticky behavior of SG&A and CGS, individually and jointly.

To test our first hypothesis, H1, we estimate Equation 1 in its original form, which

stratifies the revenue changes into absolute five percent intervals. The results of this OLS

regression for the pooled sample are shown in Table 3, Panel B. For SG&A, the coefficients β1

7
To minimize the effect of extreme observations, observations in the top or bottom 0.5 percent distribution of any
variable are excluded (Chen and Dixon, 1972). This trimming procedure results in a sample of 80,778 firm-year
observations.

12
 through β6, which relate percentage increases in costs to percentage increase in revenue activity,

show an upward trend. Specifically, the coefficient β1 is 0.5252 (t-value = 6.21), β2 is 0.5969 (t-

value = 16.94) whereas β6 is 0.7018 (t-value = 52.31) implying that a one percent increase in

sales revenue results in only 0.52 percent increase in SG&A costs when revenue increases by

less than five percent. However, when revenue increases between 25 and 30 percent, a one

percent increase in revenue results in 0.70 percent increase in SG&A costs.

Coefficients β7 through β12 measure the sticky behavior of costs for increasing levels of

revenue changes. For SG&A costs, the estimates for the coefficients β7 and β8 are positive, but

not significant, implying that when revenues change by less than ten percent, SG&A costs are

not sticky. However, beyond a ten percent change in revenue, SG&A exhibits sticky behavior as

shown by the significantly negative coefficients. This result would suggest that SG&A costs

vary proportionally with activity as long as revenue does not change by more than ten percent.

Beyond a ten percent increase in revenues, managers must increase the capacity of the firm by

increasing the level of committed resources. On the other hand, when revenues decline by more

than ten percent managers are unlikely or unable to proportionally decrease the SG&A costs.

Similar results are found for CGS and Total Costs in support of hypothesis H1.8

8
We perform two robustness tests. First, to reduce the effects of serial correlation, the model is estimated separately for
each year. To test the significance of these parameters, we aggregate the standardized firm-specific t-statistics associated
with the parameter estimates assuming cross-sectional independence and construct a Z statistic. The statistic has a standard
N
1 tj
normal distribution and is computed as Z =
N
∑j=1 k j / k j -2
where tj and kj are the t-statistics and the degrees of

freedom from the ith firm and where N is the number of years (see Healy, Kang, and Pelepu, 1987). Next, since we use a
large panel of data from different industries and years, we use a fixed effects model with separate dummies for each year.
The results are qualitatively unchanged.

13
Inter-industry Differences

Table 4, Panel A provides the results of regressing changes in SG&A, CGS, and Total

Costs on annual changes in sales revenue for the four selected industries-manufacturing,

merchandising, service, and financial services. For all four industries, the estimated coefficient

α1 is significantly positive for SG&A costs and significantly differently from one. Interestingly,

there is no such consistency for the stickiness variable. Manufacturing, merchandising, and

service firms all show evidence of sticky costs behavior. In fact, manufacturing firms show the

highest level of stickiness in SG&A costs with the estimated coefficient α2 being –0.1544 (t-

value = 8.60), followed by service firms (α2 = –0.1051, t-value = 2.20) and merchandising firms

(α2 = –0.0952, t-value = 2.42). However, financial firms do not show any stickiness (α2 is

positive and not significant) for SG&A costs.

For CGS costs, the estimated coefficient α1 is not different from one for manufacturing,

merchandising, and service firms, whereas financial firms have a α1 significantly greater than

one. When the market is improving and revenues are increasing, borrowing costs appear to rise

faster for financial firms. Ma (1988) provides a potential explanation for this result. He shows

that bank managements tend to raise (lower) loan loss provisions in periods of high (low)

operating income to smooth earnings. In fact, the SEC has investigated a number of banks for

manipulating earnings by increasing loan loss provisions during good financial times and

reducing them during bad financial times (Weidner et al. 1998, Moyer 1999). More importantly,

our results show that CGS costs exhibit sticky behavior for manufacturing and financial firms

14
and marginally for service firms (at ten percent).9 This implies that strict proportionality of CGS

costs with sales revenue cannot be rejected for merchandising firms.

Table 4, Panel B presents the results after stratifying the revenue changes into absolute

five percent intervals for the selected industries. For manufacturing firms, when annual changes

in sales revenue are below ten percent, the linearity of SG&A costs to revenue changes cannot be

rejected. On the other hand, when the absolute change in sales revenue is greater than ten

percent, SG&A costs are sticky. For merchandising firms, sticky cost behavior for SG&A costs

is observed only when the absolute change in sales revenue is greater than 25 percent, but only

marginally at the ten percent level. Service and financial firms do not show sticky SG&A cost

behavior at any level.

Examining the CGS results, only manufacturing and financial firms exhibit sticky

behavior. Specifically, for manufacturing firms, CGS costs are sticky beyond a fifteen percent

absolute change in sales revenue. For financial firms, CGS costs are sticky when interest

revenue changes by more than ten percent. The Total Cost results are similar. Specifically,

manufacturing and financial firms manifest sticky behavior beyond ten and fifteen percent

changes of revenue, respectively. Service firms exhibit some sticky behavior beyond a twenty

percent change of revenue. On the other hand, merchandising firms are not sticky at all.

In general, the results support our hypothesis that costs are sticky for large activity

changes (potentially outside the firm’s relevant range). The results also show that cost behavior

varies across industries with some industries exhibiting sticky behavior and others not, implying

that industry affiliation is important when analyzing sticky behavior. For example, ABCJ (2001)

examines the sticky cost behavior jointly for service and financial firms. They find that both

9
As discussed earlier, Compustat identifies CGS for financial firms as interest expense and provision for loan
losses. Stickiness is found for financial firms because they have to continue to pay their borrowing costs even as
revenues decline.
15
SG&A and CGS costs are sticky. By examining each industry separately, we show that (1)

SG&A costs are sticky for service firms, but not for financial firms; and (2) CGS costs are sticky

for financial firms, but only marginally for service firms.10

Inter-industry determinants of sticky cost behavior

In this section, we present an empirical test of the determinants of sticky cost behavior.

We include only those determinants that we believe may differentially affect the four selected

industries. Table 5 presents the results for the full sample ignoring industry classification. We

find that high levels of fixed asset intensity increase the stickiness of SG&A costs, while high

levels of employee intensity and when revenue decreases for two consecutive years decrease

stickiness of SG&A costs. In contrast, we find that higher levels of fixed asset intensity and

employee intensity increase the stickiness of CGS costs but revenue decreases for two

consecutive years decrease stickiness. The results for Total Costs are similar to the CGS costs.

These results suggest that, in the short term, it may be difficult to reduce capacity costs

related to fixed assets. Interestingly, employee intensity, measured by number of employees-to-

sales revenue ratio, is shown to increase sticky behavior for CGS costs, but decrease sticky

behavior of SG&A costs-implying that CGS stickiness may be a function of the difficulties

related to reducing direct labor. To obtain a more precise measure of employee intensity, we

attempted to use the actual labor (and related) expense (Compustat #42). Unfortunately, this

measure is not given for most observations and reduces our sample by more than 90 percent,

making estimation unreliable.

Finally, in Table 6, we test the determinants of sticky behavior by industry. For

manufacturing firms, fixed asset intensity appears to drive the stickiness of each type of costs

10
ABCJ (2001) combine service and financial firms in their study and find that both SG&A and CGS costs are
sticky. We find a similar result when the two industries are combined.
16
(SG&A, CGS, and Total Costs). However, that is not the case for the other three industries,

which are less asset-intensive. Not surprisingly, employee intensity appears to drive stickiness

of CGS costs in merchandising and service firms, whose sales are generated by employees, but

not in manufacturing and financial firms.

Inventory intensity increases the stickiness of SG&A and CGS costs in manufacturing,

merchandising, and service firms. This result implies that while inventory increases when

revenue increases, firms are unable or are reluctant to reduce inventory by the same proportion

when revenue is declining. The interest ratio, representing interest expense on deposits, long-

term debt and all other borrowings, increases the stickiness of both SG&A and CGS costs for

financial services firms. However, as one would expect interest ratio does not appear to increase

stickiness of costs in manufacturing, merchandising, and service firms. Finally, the positive

coefficient on the second decrease variable, representing two consecutive years of revenue

reduction, shows that over the long-term managers do reduce costs when revenue drops-for each

type of cost (SG&A, CGS and Total costs) in every industry, implying stickiness is a short run

phenomena.11

V. CONCLUSION

This study attempts to reconcile the conflicting results of extant research by examining

sticky behavior in four new ways. First, the study examines if the sticky behavior of costs is

dependent on the absolute magnitude of revenue changes. Second, the study examines the sticky

behavior of SG&A and CGS costs, individually and jointly. Third, the study examines stickiness

for four different industries - manufacturing, merchandising, service, and financial services.

11
The only exception is in the financial services industry where the β5 coefficient is positive, but not significant for
SG&A costs.
17
Finally, the study examines how selected determinants affect the sticky behavior of costs

differently across industries.

We find that SG&A and CGS, individually and jointly, exhibit sticky behavior when

tested for all firms/industries. We also find that costs are sticky when revenues change by more

than ten percent. For revenue increases greater than ten percent, managers must expand the

capacity of the firm by changing the firm’s committed resources. If revenues decrease by more

than ten percent, managers may not want or be able to reduce the capacity of the firm, causing

the sticky behavior of costs. However, for activity changes of less than ten percent linearity of

costs cannot be rejected subject to capacity utilization as suggested by BPS (2003).

Exploring each industry separately gives us greater insights into cost behavior and the

different impact of magnitude on stickiness. The industry exhibiting the highest level of

stickiness is manufacturing. SG&A and CGS are individually and jointly sticky for activity

changes greater than ten to fifteen percent. The results show that manufacturing is the “stickiest”

industry due to its high levels of fixed assets and inventory.

Merchandising and service firms exhibit low levels of stickiness due to their low levels of

fixed assets and use of temporary help that allows firms to respond immediately to major

changes in revenues. In fact, a high inventory level is the only determinant consistently

increasing the sticky behavior of costs in these two industries. The merchandising industry also

has the lowest profit margin of the four industries studied implying that it is a competitive

industry, so firms must be able to react quickly to the economic environment to remain viable.

For service firms, a simple analysis without stratifying revenue changes shows that

SG&A and CGS are individually and jointly sticky. However, only Total Costs are sticky

beyond twenty percent when the revenue stratification analysis is performed. Appropriately,

18
employee and inventory intensity are the main drivers of sticky behavior in this industry, where

success is determined by employees’ ability to satisfy customers with the right product/service.

The financial industry manifests some interesting behavior. CGS costs, representing

interest expense and provision of loan losses, is sticky beyond a ten percent change in interest

revenue. The sticky behavior of CGS drives the stickiness of Total Costs (beyond a fifteen

percent change in revenues). While employee intensity increases the stickiness of costs, the

main driver of sticky behavior appears to be interest expense and loan loss provisions. These

results are not surprising given that financial firms are obligated to continue to pay borrowing

costs and maintain loan loss provisions (at some level) even during periods of revenue declines.

Interestingly, when markets are improving and interest revenues are increasing, it appears that

borrowing costs are rising faster than revenues. These facts may signal that financial firms plan

for “rainy days” by increasing the provision for loan losses when interest revenues are high as

Ma (1988) suggests.

This paper reconciles the conflicting results of prior literature showing that the stickiness

is true in the aggregate. On the other hand, our study also shows that stickiness is not present for

all magnitudes of changes and in all industries. Our results support BPS’s statement that caution

must be applied when applying ABJ’s conclusion that costs are sticky (BPS, p. 3). Future

research can explore additional characteristics that affect cost behavior in each industry to better

understand stickiness.

19
 REFERENCES

Anderson, M. C., Banker, R. D., and S. Janakiraman. 2003. Are selling, general, and
administrative costs “sticky”? Journal of Accounting Research (forthcoming).

Anderson, M. C., Banker, R. D., Chen T. L., and S. Janakiraman. 2001. Drivers of stickiness in
the cost of sales at service firms. Working paper, The University of Texas at Dallas.

Balakrishnan, R., M. Peterson, and N. Soderstrom. 2003. Does capacity utilization affect the
“stickiness” of cost? Working paper, The University of Iowa and the University of
Colorado at Denver.

Banker, R. D. and H. H. Johnston. 1993. An empirical study of cost drivers in the U.S. airline
industry. The Accounting Review 68: 576-601.

Belsley, D., E. Kuh, and R. Welsch. 1980. Regression diagnostics: Identifying influential data
and collinearity. New York: Wiley.

Chen, E. and W. Dixon. 1972. Estimates of parameters of a censored regression sample.

Journal of the American Statistical Association 67: 664-671.

Christensen, L. and D. Sharp. 1993. How ABC can add value to decision making: Refinements
can make an accounting system more accurate and more flexible. Management
Accounting (May): 38-42.

Cooper, R. and R. S. Kaplan. 1992. Activity-based systems: Measuring the costs of resource
usage. Accounting Horizons 6(September): 1-13.

Cooper, R. and R. S. Kaplan. 1998, The Design of Cost Management Systems: Text, Cases and
Readings (2nd Edition) Prentice Hall, Upper Saddle River, N.J.

Durbin, J. and G. Watson. 1951. “Testing for serial correlation in least-squares regression.”
Biometrika 38: 159-177.

Ely, K. M. 1991, Inter-industry differences in relation between compensation and firm

performance variables. Journal of Accounting Research 29(Spring): 37-58.

Hansen, D. R. and M. M. Mowen. 2003. Management Accounting 6th Ed. Cincinnati, OH:
SouthWestern.

Healy, P., S. Kang and K. Palepu. 1987. The effect of accounting procedure changes on CEOs'
cash salary and bonus compensation. Journal of Accounting and Economics (9): 7-34.

Horngren, C. T. and G. Foster. 1991. Annotated Instructor’s Edition Cost Accounting: A

Managerial Emphasis 7th Ed. Englewood Cliffs, NJ: Prentice-Hall.

20
Lazere, C. 1995. Spotlight on SG&A. CFO 11(December): 39-45.

Lazere, C. 1996. Spotlight on SG&A. CFO 12(December): 28-34.

Ma, C. K. 1988. Loan loss reserves and income smoothing: The experience in the U.S. banking
industry. Journal of Business Finance and Accounting 15(4): 487-497.

Mintz, S. L. 1994. Spotlight on SG&A. CFO 10(December): 63-65.

Moyer, L. 1999. SEC Reviews Bankers Trust’s loan loss reserve practices. American Banker
164 (95) May 19 p. 30.

Noreen, E. 1991. Conditions under which activity-based costs provide relevant costs. Journal of
Management Accounting Research 3: 159-168.

Noreen, E. and N. Soderstrom. 1994. Are overhead costs strictly proportional to activity?
Journal of Accounting and Economics 17 (1/2): 255-278.

Noreen, E. and N. Soderstrom. 1997. The accuracy of proportional cost models: Evidence from
hospital service departments. Review of Accounting Studies 2: 89-114.

Weidner, D., N. Ring, and S. Barancik. 1998. ‘Earnings management’ found in all businesses,
SEC chief says. American Banker 163(215): 1-2.

White, H. 1980. A heteroskedasticity-consistent covariance matrix estimator and a direct test

for heteroskedasticity. Econometrica 48: 817-838.

21
 Table 1

Panel A: Distribution of annual revenue, SG&A, CGS, and Total Costs during the period 1979-
2000.a

Mean Standard Median Lower Upper quartile

dev quartile
Sales revenue 1,294.07 5,839.23 118.74 26.187 530.95
SG&A 234.43 1,027.48 21.93 5.53 94.46
SG&A % of revenue 24.43 15.48 21.54 13.47 31.43
CGS 885.48 4,420.75 73.64 15.31 341.76
CGS % of revenue 63.77 17.37 66.42 52.90 76.49
Total Costs 1,119.91 5,190.06 102.23 22.84 454.90
Total Costs % of 88.20 14.13 89.81 82.75 95.16
revenue

Panel B: Periodic fluctuations in revenue, SG&A, CGS, and Total Costs during the period 1979-
2000.

Percentage Mean Standard Median Lower Upper

of firm- percentage deviation of percentage quartile of quartile of
years with decrease percentage decrease percentage percentage
negative across decreases across decreases decreases
percentage periods across periods across across
change from periods periods periods
previous
period
Sales 30.36 10.01 7.85 8.08 15.14 3.42
revenue
SG&A 27.13 11.88 12.12 8.01 16.49 3.28
CGS 31.70 11.66 11.25 8.61 16.43 3.60
Total Costs 29.35 10.44 9.89 7.74 14.84 3.22

a
All numbers are reported in millions of dollars. The distribution is for the population of 82,118 firm-year
observations from 9,592 firms from the 2001 Compustat database. The observations satisfy the following criteria: no
missing observation for the current and preceding year for each variable, no firm years in which selling, general, and
administrative (SG&A) or cost of goods sold (CGS) costs exceeded revenues, no firm years in which revenues
changed by more than thirty percent, and firms with at least three observations during the twenty-two year period.

22
 Table 2
Panel A: Industry means (medians in parenthesis) of annual revenue, SG&A, CGS, and Total Costs
during the period 1979-2000 by industry.

Manufacturing Merchandising Service Financial

Services
Number of firm-years 43,259 12,113 11923 8,506
Sales revenue 1,544.95 1,516.17 635.32 994.44
(126.57) (233.60) (65.47) (106.88)
SG&A 279.34 251.91 147.44 202.65
(24.28) (41.00) (14.23) (22.17)
SG&A % of revenue 24.71 22.57 28.65 23.47
(21.81) (21.16) (23.78) (22.42)
CGS 1,059.66 1,172.31 356.26 569.29
(81.89) (162.54) (35.77) (53.70)
CGS % of revenue 65.69 71.44 58.87 52.28
(67.84) (72.96) (61.51) (51.82)
Total costs 1,339.00 1,424.22 503.70 771.94
(113.55) (216.83) (56.08) (79.12)
Total costs % of revenue 90.39 94.02 87.52 75.76
(89.91) (94.44) (89.54) (74.63)

Panel B: Mean (median in parenthesis) fluctuations in revenue, SG&A, CGS, and Total Costs during the
period 1979-2000.

Manufacturing Merchandising Service Financial

Services
Percentage of firm-years with 31.81 26.00 29.54 22.85
negative percentage change in
sales revenue from previous
period
Percentage decrease in sales 9.92 9.08 10.40 8.97
revenue across periods (7.99) (7.10) (8.47) (6.61)
Percentage of firm-years with 27.63 23.69 29.33 18.06
negative percentage change in
SG&A from previous period
Percentage decrease in SG&A 10.74 11.11 13.32 11.42
across periods (7.42) (7.16) (9.46) (6.51)
Percentage of firm-years with 32.41 25.91 30.98 30.83
negative percentage change in
CGS from previous period
Percentage decrease in CGS 10.45 10.00 12.54 13.92
across periods (8.06) (7.27) (9.31) (10.25)
Percentage of firm-years with 29.95 24.91 28.30 26.29
negative percentage change in
total cost from previous period
Percentage decrease in total 9.60 9.19 10.76 11.08
costs across periods (7.37) (6.80) (8.02) (8.06)
SG&A represents selling, general, and administrative expense. CGS represents cost of goods sold. Total Costs
represents the sum of SG&A and CGS.

23
 Table 3
Panel A
Results of regressing annual changes in SG&A, CGS, and Total Costs on annual changes in sales
revenues for the twenty-two year period 1979-2000.
Number of observations = 80,778.

 DVi,t   Re venuei, t   Revenuei, t 

log   = α 0 + α 1 * log   + α 2 * Decrease_Dummy * log   + εi, t
 DVi,t − 1   Revenuei, t - 1   Revenuei, t - 1 

Pooled estimation coefficient estimates (t-statistics)

SG&A CGS Total costs
α0 0.0273 -0.0029 0.0049
(25.75) (3.32) (8.11)
α1 0.6970 1.0139 0.9266
(81.59) (155.39) (191.53)
α2 -0.1210 -0.0715 -0.0814
(7.79) (6.03) (9.27)
Adjusted R2 0.1944 0.4914 0.5908

DV equals selling, general, and administrative expense (SG&A), cost of goods sold (CGS), or Total Costs (sum of
SG&A and CGS). Decrease_Dummy equals one if revenue in period t is less than revenue in period t-1, zero
otherwise.

24
Panel B
Results of regressing annual changes in SG&A, CGS, and Total Costs on annual changes in sales
revenues for the twenty-two year period 1979-2000 for all firm-years. Annual changes are
partitioned into six groups, based on the absolute magnitude of revenue change from the
previous year given in five percent increments.
Number of observations = 80,778

Equation 1:a
 DVi,t  6
 Re venuei, t  6  Revenuei, t 
log  
 DVi,t − 1 
= β 0 +
∑
S =1
β S * R Si, t * log   + ∑ β (S + 6) * DSi, t * RSi, t * log 
 Revenuei, t - 1  S =1
 + εi , t (1)
 Revenuei, t - 1 

Pooled estimation coefficient estimates (t-statistics)

SG&A CGS Total costs
β0 0.0314 -0.0022 0.0063
(14.57) (1.31) (5.18)
β1 0.5252 1.0141 0.8907
(6.21) (15.70) (18.60)
β2 0.5969 0.9798 0.8802
(16.94) (36.41) (44.12)
β3 0.6540 1.0090 0.9133
(28.99) (58.57) (71.50)
β4 0.6737 1.0007 0.9121
(38.91) (75.68) (93.03)
β5 0.6701 1.0196 0.9227
(45.60) (90.86) (110.90)
β6 0.7018 1.0183 0.9288
(52.31) (99.39) (122.26)
β7 0.1405 -0.0804 -0.0121
(0.90) (0.68) (0.14)
β8 0.0103 -0.0086 -0.0064
(0.16) (0.17) (0.17)
β9 -0.0830 -0.0919 -0.0844
(1.99) (2.88) (3.57)
β10 -0.4747 -0.1006 -0.0678
(1.49) (4.15) (3.77)
β11 -0.1069 -0.0558 -0.0755
(4.06) (2.77) (5.06)
β12 -0.0846 -0.0513 -0.0600
(3.68) (2.92) (4.61)
Adjusted R2 0.1946 0.4915 0.5909

25
a
where:

DV = selling, general, and administrative expense (SG&A), cost of goods sold (CGS), or

TOTAL (CGS + SG&A)

R1it = 1 if percent change in sales revenue is [-0.05, 0.05]

R2it = 1 if percent change in sales revenue is [-0.10, -0.05) or (0.05, 0.10]

R3it = 1 if percent change in sales revenue is [-0.15, -0.10) or (0.10, 0.15]

R4it = 1 if percent change in sales revenue is [-0.20, -0.15) or (0.15, 0.20]

R5it = 1 if percent change in sales revenue is [-0.25, -0.20) or (0.20, 0.25]

R6it = 1 if percent change in sales revenue is [-0.30, -0.25) or (0.25, 0.30]

D1it = 1 if percent change in sales revenue is [-0.05, 0)

D2it = 1 if percent change in sales revenue is [-0.10, -0.05)

D3it = 1 if percent change in sales revenue is [-0.15, -0.10)

D4it = 1 if percent change in sales revenue is [-0.20, -0.15)

D5it = 1 if percent change in sales revenue is [-0.25, -0.20)

D6it = 1 if percent change in sales revenue is [-0.30, -0.25)

26
 Table 4
Panel A
Results of regressing annual changes in SG&A, CGS, and Total Costs on annual changes in sales
revenues for the twenty-two year period 1979-2000 for four different industries including
manufacturing, merchandising, service, and financial services.

 DVi,t   Re venuei, t   Revenuei, t 

log   = α 0 + α 1 * log   + α 2 * Decrease_Dummy * log   + εi, t
 DVi,t − 1   Revenuei, t - 1   Revenuei, t - 1 

Pooled estimation coefficient estimates (t-statistics)

Manufacturing firms Merchandising firms
SG&A CGS Total Costs SG&A CGS Total Costs
α0 0.0256 0.0012 0.0072 0.0217 0.0002 0.0049
(20.99) (1.69) (12.11) (8.60) (0.14) (6.15)
α1 0.7092 0.9657 0.9013 0.8126 1.0081 0.9621
(70.80) (168.05) (183.85) (39.27) (104.76) (146.44)
α2 -0.1544 -0.0383 -0.0771 -0.0952 -0.0008 -0.0118
(8.60) (3.72) (8.77) (2.42) (0.04) (0.95)
Adjusted R2 0.2523 0.6935 0.7216 0.2576 0.7332 0.8410
Number of 42,642 42,642 42,642 12,027 12,027 12,027
observations

Panel A (continued)

Pooled estimation coefficient estimates (t-statistics)

Service firms Financial firms
SG&A CGS Total Costs SG&A CGS Total Costs
α0 0.0294 0.0039 0.0095 0.0453 -0.0363 -0.0112
(8.70) (1.34) (5.30) (13.86) (11.67) (5.67)
α1 0.7385 0.9749 0.9044 0.5108 1.3666 1.0928
(28.27) (43.29) (66.44) (20.35) (57.20) (71.89)
α2 -0.1051 -0.0797 -0.1064 0.0635 -0.28631 -0.1496
(2.20) (1.93) (4.27) (1.25) (5.90) (4.84)
Adjusted R2 0.1706 0.3383 0.5379 0.1289 0.4738 0.6012
Number of 11,737 11,737 11,737 8,335 8,335 8,335
observations
DV equals selling, general, and administrative expense (SG&A), cost of goods sold (CGS), or Total Costs (sum of
SG&A and CGS). Decrease_Dummy equals one if revenue in period t is less than revenue in period t-1, zero
otherwise.

27
Panel B –Results of regressing annual changes in SG&A, CGS, and Total Costs on annual
changes in sales revenues for the twenty-two year period 1979-2000 for the following four
different industries: manufacturing, merchandising, service, and financial services. Annual
changes are partitioned into six groups, based on the absolute magnitude of revenue change from
the previous year, in five percent increments. Variables are identified in Table 3.

Pooled estimation coefficient estimates (t-statistics)

Manufacturing firms N=42,642 Merchandising N= 12,027
SG&A CGS Total SG&A CGS Total
Costs Costs
β0 0.0320 0.0013 0.0095 0.0207 0.0020 0.0051
(12.99) (0.93) (7.90) (4.09) (0.83) (3.14)
β1 0.4617 0.9298 0.8056 0.7916 0.9603 0.9541
(4.75) (16.69) (16.95) (4.01) (10.46) (15.22)
β2 0.5839 0.9459 0.8483 0.8098 0.9593 0.9529
(14.39) (40.63) (42.71) (9.94) (25.31) (36.82)
β3 0.6555 0.9556 0.8772 0.7941 0.9882 0.9489
(25.22) (64.09) (68.96) (15.15) (40.52) (56.98)
β4 0.6463 0.9636 0.8795 0.8601 1.0027 0.9745
(32.31) (83.97) (89.83) (21.15) (52.98) (75.42)
β5 0.6927 0.9662 0.8943 0.7490 1.0151 0.9539
(40.20) (97.73) (106.04) (21.47) (62.53) (86.06)
β6 0.7002 0.9766 0.9038 0.8761 0.9891 0.9680
(44.27) (107.63) (116.76) (26.71) (64.80) (92.88)
β7 0.3252 -0.0860 0.0669 -0.3364 0.1201 -0.0626
(1.83) (0.84) (0.77) (0.90) (0.69) (0.53)
β8 0.0682 -0.0275 0.0064 -0.0957 0.0662 0.0005
(0.90) (0.64) (0.17) (0.61) (0.91) (0.01)
β9 -0.1117 -0.0356 -0.0598 -0.1324 0.0336 0.0110
(2.34) (1.30) (2.56) (1.32) (0.72) (0.35)
β10 -0.0456 -0.0584 -0.0518 -0.1007 -0.0244 -0.0304
(1.25) (2.80) (2.91) (1.28) (0.67) (1.21)
β11 -0.1269 -0.0117 -0.0482 -0.0971 0.0189 -0.0139
(4.19) (0.67) (3.25) (1.45) (0.61) (0.66)
β12 -0.0985 -0.454 -0.0651 -0.1089 0.0333 -0.0059
(3.71) (2.98) (5.01) (1.86) (1.22) (0.32)
Adjusted 0.2526 0.6936 0.7216 0.2585 0.7322 0.8410
R2

DV equals selling, general, and administrative expense (SG&A), cost of goods sold (CGS), or Total Costs (sum of
SG&A and CGS). Decrease_Dummy equals one if revenue in period t is less than revenue in period t-1, zero
otherwise.

28
Panel B (continued)

Pooled estimation coefficient estimates (t-statistics)

Service firms N= 11,737 Financial Firms N= 8,335
SG&A CGS Total SG&A CGS Total
Costs Costs
β0 0.0387 0.0102 0.0126 0.0435 -0.0431 -0.0130
(5.57) (1.70) (3.47) (6.49) (6.76) (3.19)
β1 0.1415 0.7749 0.7352 0.5759 1.7318 1.2741
(0.51) (3.26) (5.12) (2.27) (7.18) (8.30)
β2 0.5617 0.7876 0.7876 0.4865 1.5592 1.1557
(4.97) (8.08) (13.38) (4.56) (15.36) (17.88)
β3 0.6209 0.9213 0.8695 0.4465 1.4606 1.0995
(8.57) (14.74) (23.02) (6.60) (22.69) (26.82)
β4 0.7246 0.9251 0.9002 0.5150 1.4002 1.1001
(13.17) (19.49) (31.39) (9.96) (28.45) (35.11)
β5 0.6926 0.9437 0.8860 0.5226 1.4191 1.1122
(15.21) (24.02) (37.33) (12.18) (34.77) (42.80)
β6 0.7054 0.9781 0.9067 0.5751 1.3337 1.0829
(17.41) (27.99) (42.95) (14.70) (35.84) (45.71)
β7 0.8991 0.2169 0.1104 -0.4330 -0.6848 -0.3009
(1.79) (0.50) (0.42) (0.91) (1.51) (1.04)
β8 0.0202 0.2063 -0.0143 -0.2667 -0.2246 -0.1344
(0.10) (1.13) (0.13) (1.28) (1.13) (1.06)
β9 -0.0304 -0.0632 -0.1340 0.3122 -0.3250 -0.0361
(0.23) (0.54) (1.90) (2.29) (2.51) (0.44)
β10 0.0423 -0.0384 -0.0534 0.0371 -0.5573 -0.2955
(0.42) (0.44) (1.02) (0.36) (5.67) (4.72)
β11 -0.1019 0.0310 -0.1061 0.03827 -0.3678 -0.1693
(1.23) (0.43) (2.45) (0.44) (4.47) (3.23)
β12 0.0385 -0.0474 -0.492 0.0351 -0.2900 -0.1585
(0.55) (0.78) (1.34) (0.45) (3.87) (3.32)
Adjusted 0.1716 0.3384 0.5385 0.1312 0.4757 0.6021
R2

DV equals selling, general, and administrative expense (SG&A), cost of goods sold (CGS), or Total Costs (sum of
SG&A and CGS). Decrease_Dummy equals one if revenue in period t is less than revenue in period t-1, zero
otherwise.

29
 Table 5
Results of regressing annual changes in SG&A, CGS, and Total Costs on annual changes in sales
revenues for the twenty-two year period 1979-2000.
Number of observations = 71,212.

Equation 2:

 DVi,t   Re venuei, t   Revenuei, t 

log   = γ 0 + γ 1 * log   + γ 2 * Decrease_Dummy * log  +
 DVi,t − 1   Revenuei, t - 1   Revenuei, t - 1 
 Revenuei, t 
γ 3 * Decrease_Dummy * log   * Fixed Asset Intensity +
 Revenuei, t - 1 
 Revenuei, t 
γ 4 * Decrease_Dummy * log   * Employee Intensity +
 Revenuei, t - 1 
 Revenuei, t 
γ 5 * Decrease_Dummy * log   * Second Decrease + εi, t
 Revenuei, t - 1 

Pooled estimation coefficient estimates (t-statistics)

SG&A CGS Total costs
γ0 0.02643 -0.0026 0.0045
(24.00) (3.13) (7.40)
γ1 0.7044 1.0054 0.9265
(78.68) (151.70) (188.16)
γ2 -0.2617 -0.4106 -0.4228
(5.33) (11.31) (15.67)
γ3 -0.0547 -0.0114 -0.0170
(8.81) (2.49) (4.98)
γ4 0.0292 -0.0516 -0.0411
(3.10) (7.41) (7.93)
γ5 0.3725 0.169 0.2255
(23.78) (13.97) (26.17)

Adjusted R2 0.2066 0.5150 0.6140

DV equals selling, general, and administrative expense (SG&A), cost of goods sold (CGS), or Total Costs (sum of
SG&A and CGS). Decrease_Dummy equals one if revenue in period t is less than revenue in period t-1, zero
otherwise. Fixed Asset Intensity is the log of the ratio of the net book value of fixed assets (Compustat #8) to sales
revenue. Employee Intensity is the log of the ratio of the number of employees (Compustat #29) to sales revenue.
Second Decrease takes the value of one if revenue also declined in the previous period.

30
 Table 6
Results of regressing annual changes in SG&A, CGS, and Total Costs on annual changes in sales
revenues for the twenty-two year period 1979-2000 for four different industries including
manufacturing, merchandising, service, and financial services.
Equation 2:

 DVi,t   Re venuei, t   Revenuei, t 

log   = γ 0 + γ 1 * log   + γ 2 * Decrease_Dummy * log  +
 DVi,t − 1   Revenuei, t - 1   Revenuei, t - 1 
 Revenuei, t 
γ 3 * Decrease_Dummy * log  * Fixed Asset Intensity +
 Revenuei, t - 1 
 Revenuei, t 
γ 4 * Decrease_Dummy * log  * Employee Intensity +
 Revenuei, t - 1 
 Revenuei, t 
γ 5 * Decrease_Dummy * log  * Second Decrease +
 Revenuei, t - 1 
 Revenuei, t 
γ 6 * Decrease_Dummy * log  * Inventory Intensity +
 Revenuei, t - 1 
 Revenuei, t 
γ 7 * Decrease_Dummy * log  * Interest Ratio + εi, t
 Revenuei, t - 1 

Pooled estimation coefficient estimates (t-statistics)

Manufacturing firms Merchandising firms
SG&A CGS Total SG&A CGS Total
Costs Costs
γ0 0.0238 0.0006 0.0064 0.0224 0.0002 0.0043
(18.98) (0.89) (10.63) (8.46) (0.02) (5.78)
γ1 0.7166 0.9691 0.9073 0.8183 1.0109 0.9696
(68.34) (162.54) (180.64) (37.25) (115.39) (157.04)
γ2 -0.3641 -0.2567 -0.2833 0.1550 -0.2203 -0.2963
(5.49) (6.81) (8.92) (1.01) (3.59) (6.86)
γ3 -0.0946 -0.0542 -0.0543 -0.1106 0.0463 -0.0120
(9.48) (9.56) (11.37) (4.37) (4.58) (1.69)
γ4 -0.0047 -0.0004 -0.0026 0.1529 -0.0566 -0.0084
(0.35) (0.05) (0.40) (5.49) (5.08) (1.07)
γ5 0.3292 0.099 0.1661 0.2627 0.1335 0.1385
(18.36) (9.80) (19.34) (5.74) (7.29) (10.76)
γ6 -0.1232 -0.0713 -0.1026 -0.0685 -0.0307 -0.0448
(9.26) (9.42) (16.11) (3.32) (3.72) (10.76)
γ7 0.0838 0.0108 0.0368 0.0088 0.0233 -0.0074
(11.66) (2.64) (10.70) (0.41) (2.71) (1.22)

Adjusted R2 0.2792 0.7128 0.7483 0.2690 0.7906 0.8732

# of 36,578 36,578 36,578 10,348 10,348 10,348
observations
31
Table 6 (continued)

Pooled estimation coefficient estimates (t-statistics)

Service firms Financial firms
SG&A CGS Total SG&A CGS Total
Costs Costs
γ0 0.0252 0.0041 0.0084 0.5665 -0.0598 -0.0238
(6.12) (1.35) (4.19) (20.92) (14.97) (10.57)
γ1 0.7445 0.9577 0.9060 0.4585 1.5350 1.1953
(22.52) (39.92) (56.43) (22.75) (51.56) (71.10)
γ2 -0.1214 -0.6331 -0.4987 -1.5181 -2.3073 -2.0653
(0.63) (4.51) (5.31) (3.25) (3.35) (5.31)
γ3 -0.1109 0.0370 -0.0227 0.1904 0.0294 0.0328
(4.19) (1.93) (1.76) (3.33) (0.35) (0.69)
γ4 0.0663 -0.1174 -0.0702 -.2973 -0.1288 -0.2377
(1.90) (4.65) (4.15) (3.43) (1.01) (3.29)
γ5 0.3790 0.0827 0.1926 0.0631 0.5630 0.3286
(6.47) (1.94) (6.76) (1.04) (6.28) (6.49)
γ6 -0.0806 -0.0492 -0.0538
(3.82) (3.21) (5.24)
γ7 0.0890 0.0071 0.0540 -0.3035 -1.4005 -0.6523
(4.03) (0.44) (5.02) (3.23) (10.08) (8.31)

Adjusted R2 0.1853 0.4604 0.6068 0.2428 0.5998 0.7435

Number of 6537 6537 6537 3,719 3,719 3,719
observations

DV equals selling, general, and administrative expense (SG&A), cost of goods sold (CGS), or Total Costs (sum of
SG&A and CGS). Decrease_Dummy equals one if revenue in period t is less than revenue in period t-1, zero
otherwise. Fixed Asset Intensity is the log of the ratio of the net book value of fixed assets (Compustat #8) to sales
revenue. Employee Intensity is the log of the ratio of the number of employees (Compustat #29) to sales revenue.
Second Decrease takes the value of one if revenue also declined in the previous period. Inventory Intensity is log of
the ratio of the inventory (Compustat #42) to sales revenue. Interest ratio is log of the ratio of interest expense
(Compustat #15) to sales revenue for manufacturing, merchandising and service firms. For financial firms, interest
ratio is log of the ratio of interest expense (Compustat #339) to sales revenue.

32