SSRN Id975135

Understanding Cost Management:
What Can We Learn from the Evidence on Sticky Costs?
Shannon W. Anderson
Jesse H. Jones Graduate School of Management; Rice University
6100 Main Street; 239 McNair Hall MS 531
Houston, TX 77005
and
University of Melbourne
Department of Accounting and Business Information Systems
and
William N. Lanen*
The Stephen M. Ross School of Business
at the University of Michigan
701 Tappan St.
Ann Arbor, MI 48109-1234
(lanen@umich.edu)
January 2007
*Corresponding author. We are grateful to Xiaoling Chen, Harry Evans, Jason Schloetzer, Eelke
Wiersma, S. Mark Young, and workshop participants at Southern Methodist University for comments on
earlier versions of the paper.
Electronic
Electronic
copy of copy
this paper
available
is available
at: http://ssrn.com/abstract=975135
ABSTRACT
Recent research (Anderson, Banker and Janakiraman [ABJ], 2003) provides evidence that the absolute
change in SG&A cost that is associated with decreased sales activity is systematically less than that
associated with increased sales activity (so-called sticky costs). They view an asymmetric cost response
as incompatible with traditional cost accounting models in which costs are strictly fixed or variable with
respect to activity. Consequently, the authors interpret their finding as evidence of overt cost management
by managers. In this paper we revisit and reject the conclusion that ABJs evidence of stickiness
implies managerial action in response to demand shifts as distinct from a mechanical alternative. Indeed,
we argue that the mechanical model is an empty theory because all cost realizations are logically the
result of management decision making in some time period. That said, what motivates managers to adjust
costs and how they do so are important questions. Thus we consider whether the empirical tests and data
employed in prior studies are appropriate for testing any theory of managerial cost adjustments
(asymmetric or otherwise) to activity changes. We first illustrate the fragility of empirical results related
to the characterization of SG&A costs as sticky. Although we find weak evidence consistent with sticky
SG&A costs, the results are quite sensitive to assumptions about what managerial behavior is implied by
the sticky cost hypothesis. Specifically, if we constrain our inquiry to cases in which SG&A changes
move in the same direction as sales activity, as implied by the theory of sticky cost behavior, we find only
limited evidence of sticky costs. Further, we find no consistent mode of cost behavior when we test the
sticky cost model using other types of cost (e.g., labor costs, R&D costs, PP&E costs) that are as likely as
SG&A to be subject to managerial discretion. The instability and general lack of significance of the tests
for sticky costs demands that we consider whether the empirical test and the data are suited to the task of
testing the hypothesis. Upon revisiting the economic theory of production and comparing the empirical
test implied by theory to that which has been employed, we conclude that the methods and data employed
to date cannot be used to distinguish between alternative theories of cost adjustment. We conclude by
offering suggestions for how future research might more meaningfully identify the theories of
management decision making and cost management that are most consistent with observed cost behavior.
Electronic
Electronic
copy of copy
this paper
available
is available
1.
Introduction
Cost accounting models represent relations among the activities of the firm, the inputs consumed
in enacting these activities, and the costs of providing these inputs. In the traditional model of cost
accounting, costs are characterized as either fixed or variable with respect to activity; thus changes in cost
depend only on the change in activity, not on whether activity is increasing or decreasing (Anderson,
Banker, and Janakiraman [2003]hereafter ABJp. 47-8). However, many researchers argue that this
characterization of cost behavior is inconsistent with the way that managers manage costs (e.g., Cooper
and Kaplan [1998]; Noreen and Soderstrom [1997]). Combined with information on the value that
customers place on the firms activities, managers use cost accounting data to make decisions about the
levels of inputs and outputs that maximize the long-run profits of the firmwhat we term, cost
management. A central component of cost management is managers response to exogenous shocks to
output demand or to input supply. Exogenous shocks are posited to cause managers to revisit the relations
between activity and input levelsand between revenues and costsand to evaluate costly adjustments
as compared to the status quo. Cooper and Kaplan (1998) argue that cost management renders the
traditional model of fixed and variable cost obsolete.
In a recent paper, ABJ provide the first large-sample evidence that changes in cost depend on the
direction of the change in activity. Using a broad sample of firms over a 20-year period, ABJ document
that selling, general, and administrative (SG&A) costs exhibit a phenomenon they describe as
stickiness.1 Specifically, SG&A costs tend to fall in proportion to a decline in sales revenue in the
previous year at a lower rate than that at which they rise in response to an increase in revenue in the
In the management accounting literature, sticky has taken on the meaning of asymmetric reactions to activity
changes. In the economics literature, sticky prices or wages are merely slow to adjust in either direction.
Hamermesh and Pfann (1996, p. 1271) provide a comprehensive discussion of the econometric implications of
asymmetric adjustment costs for production inputs. Thus, roughly speaking, economists are concerned with the lag
between new information and adjustments to the new information while management accountants are concerned
with how the rate of adjustment depends on the nature of the new information. We will continue to use sticky
throughout this paper in the sense of the management accounting literature; that is, sticky costs are those that
respond asymmetrically to activity changes.
Electronic
Electronic
copy of copy
this paper
available
is available
Understanding Cost Management
previous year. This is not consistent with costs being either fixed or variable when, as is typically
assumed in managerial accounting, unit variable costs are constant over a range of activity. Consequently,
ABJ conclude that their finding of a widespread, time-invariant, asymmetric relation between changes in
SG&A costs and changes in sales levels provides evidence of active cost management. Combined with
evidence that stickiness reverses over longer time periods and that the degree of stickiness is positively
associated with adjustment costs that would enter into managers calculus of cost management, they
conclude that their results allow us to reject the traditional theory of cost behavior in which costs respond
mechanically to activity volume. Thus, ABJ do more than simply documenting sticky costs; they
interpret sticky costs as implying active cost management. Follow-on studies have accepted this
conclusion, focusing on identifying settings in which managers are more or less likely to engage in active
cost management (e.g., Subramaniam and Weidenmier [2003]) and on using information about the firms
cost structure to evaluate the firm (e.g., Banker and Chen [2006]).
This paper has two objectives. First, we argue that the mechanical model of cost behavior is not a
suitable benchmark for conclusively establishing the presence of managerial actions aimed at cost
management. It is not difficult to construct a thought experiment in which costs that are not sticky are
nonetheless the result of managerial decisions. Nor is it difficult to construct examples in which
stickiness, as evidenced by the tests of ABJ, is not the result of short-term active cost management by
managers. Cost management theories cannot be evaluated by examining observed cost behavior of firms
where managers make decisions (including decisions about resource utilization) that can produce a record
of costs that is indistinguishable from that produced by a mechanical relationship. If all decisions are
taken by a manager in some time period, even a mechanical cost relation such as that associated with
technical or engineering production specifications can be traced to past managerial decisions. The
mechanical model of costs that is hypothesized in counterpoint to managerially-determined sticky costs
requires one to draw arbitrary distinctions about what constitutes cost management. In this paper we argue
that the absence of distinct theories of cost management is impeding progress in understanding the
phenonmenon.
2
Although we reject the mechanical model as a meaningful benchmark, we believe that it is

important to develop and test alternative theories of cost management. Economic theories about how
adjustment costs (and asymmetries thereof) influence managers response to exogenous shifts of demand
or input supply are important candidates for consideration. Indeed, there is a large economics literature on
adjustment processes (e.g., Hamermesh and Pfann [1996]; Blinder [1991], Blinder et al. [1998]). The
question then is how management accounting research can contribute new insights about the work of
managers in adjusting resources and managing costs. The second objective of this paper is to consider this
question. In research on labor cost adjustments, Hamermesh and Pfann (1996, p. 1279) report that no
universal nonconvex or asymmetric specification describes adjustment of employment demand because in
all likelihood firms are distinguished by the skills of their workers and the nature of shocks to demand
have underlying adjustment costs with different structures. In counterpoint, ABJ found a widespread,
time-invariant asymmetric relation between changes in SG&A costs and changes in sales levels. Thus one
possible answer to the question is simply that accounting researchers may contribute to the understanding
of cost management by bringing different data to bear on the question. We consider this as well as other
opportunities in our discussion of directions for future research.
The paper proceeds in stages that roughly reverse the development sequence of the ABJ paper.
ABJ start with two competing theories of cost structure, a traditional cost structure of strictly fixed and
variable costs and an alternative sticky cost structure in which cost adjustments vary in magnitude with
the direction of activity change. The authors implicitly assume that the two cost structures are uniquely
associated with two distinct modes of cost management, a mechanical mode in which costs adjust
without managerial intervention according to the schedule of production, and a managerial (sticky)
mode in which managers take decisions about the mix of resources (costs) needed to support the
production schedule. An empirical test of the relation between cost changes and activity changes is
developed and tested using accounting SG&A cost data and sales data (a proxy for production activity)
for a large sample of firms and time periods.
In this paper we reverse the progression of ABJ. We first revisit the results of the basic test of
stickiness using similar methods, data and sample selection procedures. Upon finding that the results are
fragile in general and not found in several subsamples, and having ruled out errors in the execution of the
test, we revisit the design of the test in relation to the economic theory from which it derives. Since
economists have also failed to find a robust universal model of cost adjustments for labor costs, we
reconsider whether the test and the data are capable of providing an appropriate test of whether costs are
sticky. We find many reasons to doubt this. Revisiting the way that the economic theory of production
is made operational in the empirical tests as well as the sample selection and the choice of costs that are
studied, we conclude that the data and the test for stickiness do not have sufficient fidelity with the
underlying theory. Consequently, even if the empirical results indicated a robust finding of stickiness, we
do not believe that any conclusion about asymmetric adjustment costs can be drawn. We then consider
whether a perfectly executed test would discriminate between the two hypothesized cost models. We
conclude that while a perfectly executed test would detect whether costs respond asymmetrically to
changes in activity, the test would not allow us to conclude anything about how managers manage costs.
The problem lies in the assertion that stickiness is uniquely associated with managerial action, or
conversely, that the traditional fixed and variable cost model is incompatible with managerial action. We
conclude the paper with a discussion of how future research in management accounting can advance our
knowledge about how managers adjust resources. In particular we discuss the need for articulating
distinct theories of cost management and using appropriate methods and data to determine which theories
are most consistent with the data.
The paper is organized as follows. In the next section, we update the ABJ analysis of SG&A costs
and investigate the robustness of findings about stickiness in economically meaningful sub-samples and
for different categories of cost. The purpose of these extensions is to understand where, and to what
extent, stickiness occurs. In Section 3, we discuss the theoretical and empirical reasons why the data and
tests of Section 2 might not be appropriate for determining whether costs adjust asymmetrically to activity
changes. Further, we discuss why these data and tests cannot discriminate between active cost
4
management and what ABJ term, a mechanistic model relating costs and activity. In Section 4 we
return to the basic research question and consider how to improve upon our research methods to better
address questions about cost management. Recent economic research on sticky prices (Blinder 1991) and
on adjustment costs associated with changing input factors (Hamermesh [1995], Hamermesh and Pfann
[1996]) is instructive in this endeavor. We conclude in Section 5 with a summary of our findings and their
implications for future research.
2.
ABJ RevisitedTesting for Cost Stickiness

In this section, we consider first the basic result of cost stickiness documented in ABJ. We then
examine the sticky cost hypothesis in various sub-samples to determine how robust this finding is. Our
objective is to continue the inquiry begun by ABJ into the prevalence of cost stickiness and explanations
for this cost behavior.
2.1. The Basic Sticky Cost Model

ABJ specify SG&A costs (SG&A) as a function of sales revenue (Revenue). They start by
calculating the ratios of current SG&A costs (revenues) to previous period SG&A costs (revenues). They
then transform these variables by taking logarithms. To test for stickiness, ABJ introduce an indicator
variable (Decline_Dummy), which takes on the value 1 when current period revenues are lower than
previous period revenues. This indicator variable is then multiplied by the log of the ratio of current to
previous period revenue. Their regression model, then, is (ABJ, p. 52):
SG & Ai,t
Revenuei,t
Revenuei,t
log
= 0 + 1 log
+ 2 * Decrease _ Dummy * log
+ i,t
SG & Ai,t 1
Revenuei,t 1
Revenuei,t 1
As ABJ note, the change in SG&A costs associated with a 1% change in sales revenue is measured by the
coefficient 1 when sales have increased from the previous period and by (1 + 2) when sales revenue
has declined. Therefore, a negative value for 2 indicates sticky costs. In the remainder of this section, we
focus on the estimate of 2 although we report and discuss the value of 1 as well.
2.2 The Sample

We start by gathering an updated sample that is similar to that used by ABJ and screening data
according to the process described in ABJ. Specifically, the data are from the Compustat Annual
Industrial Tapes for 2004 as maintained by Wharton Research Data Services (WRDS). We first collect
data on Sales Revenue (Compustat Data Item #12) and SG&A (Compustat Data Item #189) for the period
1978 2004 for all active and inactive firms. Thus, our sample includes one year prior to and six years
subsequent to the sample employed by ABJ. Our primary motivation for extending the already-sizeable
sample employed by ABJ is that more current time periods have been characterized by volatile demand in
many industries. In particular, unexpected national events of 2001 interjected large, primarily downward
exogenous shifts of demand that should enhance our ability to detect sticky cost behavior.
Following the description of how the sample was developed in ABJ we refine our sample with
these steps:
1. We delete firms with duplicate issues (same CNUM, different CICs).
2. Next, we delete observations that are missing data on either sales revenue or SG&A costs for
the current year or the previous year.
3. Next, we drop observations where SG&A costs are greater than sales revenue in the current
2
year.
4. Our next step, not explicitly mentioned by ABJ, but required for the log transformation, is to
delete observations with non-positive amounts for either sales revenue or SG&A.
5. The final step, completed after the descriptive statistics are computed, but before the
regression is estimated, is to trim the top and bottom 0.5% of the estimation sample.
An alternative filter would be to delete observations where SG&A costs are greater than sales in both the current
and the preceding year. We choose to follow ABJ in their sample development. Our primary concern is not with the
empirical tests of ABJ; it is with the interpretation of the results of those tests. A second issue, which we discuss
below, does concern the appropriate observations to include in the sample, but not how to treat unusual
observations such as these (where SG&A costs are greater than sales revenue). Although researchers might disagree
about the appropriate treatment of unusual observations, this issue is not germane to our concerns with ABJ.
Specifically, after completing steps 1 5 above, the 0.5% of the observations with the highest
and lowest values of each of the three variables in the regression are identified and removed
3
(concurrently) from the sample remaining after step 5.
Table 1 summarizes these sequential data screening steps and documents their impact on the estimation
sample size.
Descriptive statistics for our sample are shown in Table 2. Panel A of Table 2 provides
comparative data on the sizes of the current sample and the ABJ sample. Our sample is slightly more than
twice the size of the ABJ sample both in terms of firm-years and individual firms covered. Selected
descriptive statistics from the original ABJ article (ABJ, Table 1) are reproduced in Table 2 to facilitate
comparison of our results with theirs. Comparing the descriptive statistics of the current sample in Panel
B with those of the ABJ sample, our sample is characterized by larger (non-inflation adjusted) firms
(median sales of $91 million compared to $88 million in ABJ) that have higher SG&A costs (median
costs of $19 million compared to $17 million in ABJ). The distribution of the ratio of SG&A costs to
sales revenue is very similar in the two samples with the current sample showing a slightly higher ratio.
The two samples are also similar in terms of the number of years in which sales revenue and/or
SG&A decline as can be seen by comparing the results shown in Panel C of Table 2. Our sample shows
slightly higher frequency of sales declines (29% versus 27% in ABJ) and SG&A declines (27% versus
25% in ABJ). Given a sales decline, however, the median amount of the decline is essentially the same in
both samples (10.7% versus 11% in ABJ). These comparisons suggest that the characteristics of the two
samples are broadly similar. This is reasonable, because although there is a significant difference in the
size of the samples, there is considerable overlap in the years covered.
Table 3 reports the results of estimating the basic ABJ regression model with our estimation
sample. Comparing our results with those of ABJ, we find a negative and significant 2 indicating that our
An alternative is trimming the sample one variable at a time (sequentially). We note this because when reviewing
Table 2, the trimming operation reduces the sample by more than 1% in total, but not by 1% for each of the three
variables.
sample is also characterized by sticky SG&A costs as defined by ABJ. Our coefficient estimate for 0, the
intercept, is slightly lower (0.0419 versus 0.0481 in ABJ) and 1, the estimate of the change in SG&A
cost that follows increased sales revenue, is somewhat larger than that reported in ABJ (0.5733 compared
to .5459 in ABJ). Our coefficient estimate for 2, the estimated change in SG&A costs that follows a
decline in sales revenue, is much lower in absolute value than that of ABJ (-0.0395 compared to -0.1914
in ABJ). Thus, in our sample, costs are much less sticky and the (absolute) response of costs to a sales
decline, measured by (1 + 2), is greater.
2.3 What Constitutes Sticky Costs?

Before continuing with our evaluation of the robustness of the sticky cost phenomenon, we pause
to consider what cost behavior (not estimated results) reflects stickiness. ABJ define sticky costs as
follows:
costs are sticky if the magnitude of the increase in costs associated with an
increase in volume is greater than the magnitude of the decrease in costs
associated with an equivalent decrease in volume. (p. 48, emphasis added)
They argue that in rejecting the null hypothesis of cost changes being invariant to the direction of change
in activity, their results require us to reject the traditional model of cost behavior in favor of an
alternative model in which sticky costs occur because managers deliberately adjust the resources
committed to activities (p. 47, emphasis added). For example, managers may purposely delay
reductions to committed resources until they are more certain about the permanence of a decline in
demand (p. 48). On page 49, ABJ note, Sticky costs occur because there are asymmetric frictions in
making resource adjustmentsforces acting to restrain or slow the downward adjustment process more
than the upward adjustment process. Thus, their basic argument is that uncertainty and adjustment costs
interject a lag in the relationship of activity to costs. However, even with such delays, the expectation
4
Restricting the sample to the same period studied by ABJ (1979 1998) does not change our results qualitatively
from those for the full sample (1978 2004). Our results for the ABJ period differ because the firms included in
COMPUSTAT have changed (our sample includes more firms for the 1979-1998 period. Again, we are not trying to
replicate the ABJ results; we take their empirical results as a starting point for evaluating their analysis of cost
management.
remains that cost should move in the same direction as activity. ABJ suggest something similar when they
note both on page 53 and page 56 that the sticky cost hypothesis is conditional on 1 > 0. In other words,
the issue is not whether costs and revenues move in the same direction, but whether costs move as quickly
for sales declines as they do for sales increases. This raises the issue of sample selection.
Both the ABJ sample and our sample include observations where SG&A costs have increased
following a sales revenue decline (or have decreased following a sales revenue increase). Empirically,
firm-year observations where revenue decreases (hence Decrease_Dummy = 1) and SG&A costs increase
will have the effect of decreasing (making less positive or more negative) the estimated coefficient 2,
although the related managerial decisions cannot be attributable to the sticky cost hypothesis. To
determine the effect on the results from including these observations, we develop a second sample that
includes only firm-year observations where SG&A costs and sales revenue move in the same direction;
thus we exclude firm-year observations in which SG&A costs increase following sales revenue reductions
or SG&A costs decline following sales revenue increases. In what follows, we refer to this as the
restricted sample and the initial sample (used in developing the results reported in Tables 1-3) as the
full sample.
As shown in Table 4, Panel A, the sample size of the restricted sample is about 21% smaller than
the original (full) estimation sample. To obtain the restricted sample, we remove 16,259 observations in
which SG&A costs increased following a sales revenue decline and 13,646 observations in which SG&A
costs decreased following a sales revenue improvement.5 In Table 4, Panel B, we find that the restricted
sample is not characterized by cost stickiness; indeed, the coefficient on 2 is positive and significant.6
The results in this section indicate that the sticky SG&A costs found by ABJ remains even when
a longer time series is employed. However, there is also evidence that the phenomenon is sensitive to how
These numbers do not match the difference in sample sizes between Tables 3 and 4 because the observations were
removed before the trimming of the extreme observations for the regression analysis.
6
Taking a somewhat different approach, Banker, et al. (2006) find that conditioning the model by the direction of
the sales change in the previous year also reverses the results found by ABJ.
stickiness is defined. When samples are restricted to the direction of changes in costs being consistent
with sales changes, the results are much less strong. Thus, conclusions about cost stickiness are sensitive
to the precise meaning of the sticky cost hypothesis. Specifically, the researcher must determine if
stickiness is conditional only on 1 > 0 (ABJ, p. 53). If so, tests of the sticky cost hypothesis should
restrict the sample to only those observations where changes in SG&A costs are in the same direction as
the changes in sales revenue.
2.4 Investigating the Sticky Cost Hypothesis in Industry and Year Subsamples
Like ABJ, we next investigate whether the sticky cost phenomenon is a general one or is
restricted to specific industries or time periods. We do this for two reasons. First, we want to understand if
evidence of cost stickiness is sensitive to sample selection. Second, we want to understand how cost
stickiness varies with selected firm characteristics, such as industry, to help guide future research designs
by identifying potentially important covariates. In each of the sub-sample tests described, we follow ABJ
in dropping observations from the sample for the reasons given in steps 1 5 described above in
developing the full sample. In addition, we trim each sub-sample of the top and bottom 0.5% of
observations before estimating the regression.
ABJ estimate the basic regression year-by-year and report a significant z-statistic for the sticky
cost coefficient (p. 56). The results of year-by-year regressions for our data set are shown in Table 5.
Panel A contains the results for the full sample and Panel B presents the results for the restricted sample.
As shown in Table 5, the hypothesis of sticky costs is not supported in the full sample; the mean
coefficient is negative, but insignificantly different from zero. In the restricted sample, the mean
coefficient is positive and significant. In the full sample, the estimated coefficient 2 is negative and
significant in eleven of the 26 years. In the restricted sample, it is negative and significant in no year.
However, it is also positive and significant in 25 of the 26 years.
ABJ consider the possibility that unspecified firm or industry factors influence the degree of
stickiness in SG&A costs. ABJ test this proposition using a fixed effects model to allow for separate
10
intercepts for major industries (p. 54) and a random coefficients model in which all the model coefficients
are assumed to be normally distributed across firms (p. 56-7). In the next set of tests, we estimate ABJ
Model (1) for each of the 2-digit SIC codes.7 Thus, ours is a less restricted model because we allow all
coefficients of Model (1) to differ by industry and do not constrain industries to be related to one another
through a common underlying distribution for any coefficient. The results reported in Table 6, Panel A,
where the full sample is used, indicate that the sticky cost hypothesis is supported in 22 of 65 industries,
is positive and significant in 13 industries, and is not significant in 32. When the restricted sample is used
(Table 6, Panel B), the sticky cost hypothesis is supported in only 7 of the industries, the coefficient is
positive and significant in 32, and it is insignificant in 28.
The results in this section suggest that stickiness in SG&A costs is much less widespread and less
stable over time than might have been inferred from ABJ. Subramaniam and Weidenmier (2003) also
provide evidence that cost stickiness is more prevalent in certain industries. In sum, the results suggest
that sticky SG&A costs are an industry and time period-specific phenomenon. At a minimum, these
results suggest that researchers interested in cost management should consider research designs that
exploit differences among industries and time periods.
2.5 The Stickiness of Other Cost Categories
If managerial discretion is the cause of SG&A cost stickiness, we might expect to see other cost
categories, for example those most likely to be subject to control by managers, exhibit sticky behavior.
Indeed, Noreen and Soderstrom (1997) find weak evidence of asymmetric cost response in 12 of the 16
hospital overhead cost categories that they study. If, as ABJ argue, cost stickiness is evidence of overt
cost management, the costs that are most likely to be sticky are those that are subject to managerial
discretion. Obvious categories to investigate are advertising expenses (Compustat Item # 45), which is a
We disregard industries with SIC Codes greater than 81, which tend to be public organizations (e.g., schools and
public agencies).
11
component of SG&A, research and development (R&D) costs (Compustat Item #46), labor costs
(Compustat Item # 42), and property plant and equipment costs (Compustat Item # 30).
We estimate the basic model substituting advertising costs for SG&A costs. We use the full
sample period (1978 2004) and follow the same procedures for screening and trimming the data. The
results of estimating regression of Model (1) with advertising costs indicate that advertising costs do not
exhibit stickiness (Table 7) for either the full or the restricted sample. In both cases, the coefficient on the
sticky cost variable is positive and significant. This differs from the results ABJ reported (page 58).
Although we find a similar estimate for 1 in the full sample (0.5525 versus 0.6298 in ABJ), the estimate
for 2 is not only positive, but also relatively large (0.2446).
Repeating the analysis using R&D costs, the results are again not consistent with the sticky cost
hypothesis for the unrestricted sample. The results are also not consistent with the hypothesis when we
use the restricted sample; the estimate for 2 is positive, significant, and large (0.5753). In the case of both
advertising and R&D, the results suggest that in both the full and restricted samples, the reaction of
managers in the face of a decline in sales revenue is to cut spending by an even greater proportion.
ABJ consider the moderating effect of labor intensity on the stickiness of SG&A costs. However,
we consider a more direct test of their proposition that, Dismissing employees is costly because
employers must pay severance costs lose investments made in firm-specific training and experience
productivity losses because morale declines when employees are laid off, and they may experience more
turnover because employee loyalty is eroded (p. 52). We estimate Model (1) for two measures of
employee costslabor costs (Compustat Item #42) and the number of employees (Compustat Item #29).
The results are reported in Table 7.
Labor costs exhibit stickiness in the unrestricted sample, but not in the restricted sample; the
coefficient is positive and significant. The number of employees does not exhibit stickiness in either
sample. It is positive and significant in both samples. Stickiness in cost but not the number of employees
would be consistent with managers retaining higher cost-per-hour employees (among those included in
12

8
the labor cost calculation). Whether this is because of managerial discretion or because of labor
agreements with respect to seniority, is impossible to determine in this analysis. As we discuss in Section
3, one concern with the ABJ tests is the use of accounting data rather than physical measures of input and
activity volume. One explanation for the different results we obtain using labor costs and the number of
employees is that the cost of labor is changing at a rate different from that of revenue.
The last cost category we consider is expenditures on property, plant, and equipment (PP&E)
Compustat Item #30. Again, ABJ tested the impact of asset intensity on stickiness of SG&A costs.
However, we perform a more direct test of their proposition that, Unless long-term contracts exist, it is
relatively easy to scale down purchased resources when demand drops, but disposing of assets is costly
because the company must pay selling costs and lose firm-specific investments (installation and
customization costs (p. 51). We estimate whether PP&E expenditures exhibit stickiness. The results,
reported in Table 7 indicate that PP&E expenditures are not sticky in either the full or restricted samples;
the estimated coefficient, 2, in both samples is positive and significant.
2.6 A Summary on the Empirical Analysis of Sticky Costs

Reviewing the empirical results in this section, we find support for the sticky cost hypothesis for
SG&A costs in our full sample. We find mixed evidence of SG&A cost stickiness for different industries
and time periods when we use the full sample. However, the results are very different when we restrict the
sample to those observations characterized by changes in costs moving in the same direction as changes
in sales revenue. In the restricted sample we find fairly consistent cost behavior that is anti-sticky; costs
decrease more rapidly in the face of sales declines than they increase with sales increases. Moreover, we
find no evidence of stickiness, and indeed observe anti-sticky behavior in other categories of cost that
Hamermesh and Pfann (1996, p. 1279) summarize empirical economic studies that examine adjustments in labor
costs as follows: It is unclear from this growing literature whether there is one particular nonconvex or asymmetric
specification that describes adjustment of employment demand best. It is more likely that firms distinguished by the
skills of their workers and the nature of shocks to demand have underlying adjustment costs with different
structures. This conclusion is consistent with both our findings about cost stickiness in different industries and in
different cost categories. Hinting at issues that we discuss in section 3.2, the authors go on to say, The literature
also tells us nothing about the nature of the costs of adjusting worker-hours as opposed to the demand for workers.
13
seem as likely to be subject to managerial discretion as SG&A costs. This is true whether we use the full
or the restricted samples. Taken together, these results suggest that before we conclude something about
managerial behavior from empirical tests of the sticky cost hypothesis, we should consider more closely
exactly what these tests tell us.
3.
Technology or Managerial Behavior? Can We Tell?

The results reported in Section 2 suggest that sticky costs, although present in some industries, in
some time periods and for some cost categories, are not universal. The question then arises whether it is
sticky cost behavior that varies from one subsample to another or whether the test design is flawed (or
both). We start by looking at the economic theory of cost. This allows us to identify how sticky cost
behavior might, in principle, manifest itself in the data. We next consider the conjectures of ABJ and
identify reasons why their tests might not be able to distinguish the reasons for the existence of sticky cost
behavior (e.g., whether it arises from cost management in the ABJ sense or is mechanical in the sense that
it is technologically determined). In the course of this discussion we highlight ambiguities in the research
literature in the meanings of various widely used terms; ambiguities that we argue have further confused
the discussion about cost behavior and cost management.
There are four major issues that complicate the empirical design of tests of cost management.
First, the assumptions required to move from the theory of cost behavior to the empirical model are
unlikely to be accurate descriptions of the world. Second, because managerial discretion generally
extends beyond resource utilization to pricing, we must consider the dangers of testing theory about real
relations using nominal data. Third, costs are typically classified according to accounting conventions.
However, economic theory is concerned with total costs. Analysis of an arbitrary subset of (accounting)
costs using theory that describes the relation between activity and total costs is fraught with difficulties,
difficulties that are compounded when managerial discretion over cost accounting is acknowledged and
when changing accounting rules may cause cost categories to be defined differently in different time
periods or industries. Finally, and perhaps most important, it is difficult to identify any characteristic of
14
the firm that is not, in some sense, the result of managerial discretion or, in this case cost management.
These complications, especially the first two, are not unusual in empirical tests of accounting theories, but
they should be considered explicitly so that the results of the empirical analysis are interpreted in light of
these limitations. In the sections below we discuss these concerns and their implications.
We acknowledge that all empirical work requires that assumptions be made and that imperfect
measures must be specified for theoretical constructs. We raise these issues here, not because we believe
that ABJ might have found better measures or employed better assumptions, but because it is useful to
make these issues explicit when we use the findings of sticky costs to infer something about cost
management.
3.1 The Cost Function

The cost function specifies how total cost, C, is related to output quantity, q, and factor prices
(wage rates, cost of capital, etc.), pj. The cost function is based on the production function and is derived
by minimizing total costs (the sum of payments for each of the factors of production) given that the firm
produces a specified quantity of output. In competitive output markets, quantity and factor prices can be
considered exogenous, so, in theory, regression analysis can be used to estimate the cost function once a
functional form is specified.
A common production function is the Cobb-Douglas production function, which for two factors
of production, capital and labor for example, is:
q = AK L .
The sum of + represents the returns to scale of the production process. If + > 1 , the process
exhibits increasing returns to scale and if + < 1 the process exhibits decreasing returns to scale. The
Cobb-Douglas function is a particularly convenient assumption because, as is well-known, the
corresponding cost function is of the same form as the production function. Specifically, the cost function
can be written as:
15
C = Bq1/( + ) ( pK pL ) ,
(1)
where B is a function of the parameters, A, , and . If we want to study how costs change over time in
response to changes in quantity, or activity, we can add time subscripts to the cost function (1) assuming
the underlying scale parameters, , and , are time-invariant. (We will allow the constant, A, and hence,
B, to vary over time.) Thus, at time t, total costs are equal to,
Ct = Bt qt1/( + ) ( pKt pLt ) .
(2)
It is clear from equation (2) that if we want to estimate the cost function over time (or how it
grows over time), we need either to include data on factor prices or to assume that factor prices are
constant over time. If we do not, the estimate will be biased because of the omitted variables. For this
discussion, we assume that factor prices remain constant over time. We make this assumption not because
we necessarily view it as descriptive, but because it is implicit in the ABJ empirical model, which does
not include factor prices.
Given this assumption on factor prices, we can write the equation describing the one period
growth in cost (Ct/Ct-1), as,
Ct
Ct 1 =
Bt
Bt 1
( )
qt
1/( + )
(3)
qt 1
Taking the logs of both sides of equation (3), we derive the basic estimating equation,9
ln
( )
Ct
Ct 1 =
+ 1 ln
( )
qt
qt 1
(4)
where
( )
1
and 1 =
.
0 = ln Bt B
t 1
+
For notational convenience, we omit the firm subscript and the error term.
16
ABJ conjecture that, because of managerial decisions, the decline in costs when activity falls is
not as great as the increase in costs when activity increases. To test this, they add an indicator variable to
the regression equation (4) that takes on the value one when activity falls and is zero otherwise. Thus, as
discussed in Section 2, ABJ estimate a version of the following equation:
ln
( )
Ct
Ct 1 =
+ 1 ln
( )
qt
qt 1 +
Decline * ln
( )
qt
qt 1
(5)
where Decline is the indicator variable. In the ABJ tests, costs, C, are represented by SG&A costs and
quantity, q, is measured by sales revenue.
Successful empirical work almost always requires that researchers make compromises with the
underlying theory when developing models for testing theories. We want to emphasize that the discussion
in this subsection is not intended as a pedantic objection to the empirical work of ABJ or others in this
area. Rather, we want to make explicit the compromises that have been taken in translating theory into an
empirical test so that readers may judge the applicability of the results for testing the sticky cost
hypothesis. We have identified six issues here:
1. The theory is a theory of total costs while the empirical tests are based on one component
of cost (SG&A);
2. The empirical model assumes factor prices are constant;
3. The theory relates cost to the quantity of activity; the empirical tests are based on the
nominal value of the activity;
4. The theory assumes the product portfolio is constant over time;
5. The theory assumes competitive output markets;
6. The theory assumes a specific functional form for the production function of all
industries.
Taken together these issues suggest that it will be difficult to rely on archival data on SG&A costs over a
twenty-year period to infer whether the sticky cost phenomenon holds.
17
3.2 Logical (and Terminological) Issues When Testing for Stickiness

In their paper, ABJ contrast two models describing cost behavior. In the first, which they term,
the traditional model of cost behavior that pervades the accounting literature (p. 47), costs are classified
as either fixed or variable and variable costs change proportionately with changes in the activity driver
(pp. 47-8). The second model is one, in which managers deliberately adjust resources in response to
changes in volume (p. 48). They go on to say, This model [Model 1] distinguishes between costs that
move mechanically with changes in volume and costs that are determined by the resources committed by
managers (p. 48).
It is useful to identify the factors associated with costs and to understand the cost management
process. There are, however, several problems with the empirical model ABJ use to distinguish between
their two alternatives. Some of these are problems of terminology, terms that are ambiguous or illdefined. Some of the problems concern the data used in the tests. We discuss three general problems
below. Although they are not completely independent, we discuss separately: nominal versus real activity;
accounting classification of costs; and, the meaning of managerial discretion.
Real versus Nominal Activity

As shown in equation (5), the driver of cost is the quantity produced. This is a physical volume
measure. ABJ recognize the physical nature of the factor that causes costs to be incurred, using the terms
activity and volume throughout. Thus (ABJ, p. 48),:
The behavior of SG&A costs can be meaningfully studied in relation to revenue
activity because sales volume drives many of the components of SG&A (Cooper
and Kaplan [1998, p.341]). In its annual SG&A survey, CFO Magazine performs
extensive analyses of SG&A costs in relation to sales revenue (Mintz [1999]).
In footnote 2 to this passage, ABJ note,
We use sales revenue as an imperfect proxy for sales volume because sales
volume is not directly observable.
The problem is that growth in activity, when activity is measured by sales revenue, can occur
because of changes in prices, changes in volume, or both. The cost function, which is the basis of the
18
regression equation for the tests of stickiness, assumes activity (quantity) is exogenous. If the activity
measure is not exogenous, then the regression is misspecified.10 The problem is especially serious in this
case because the purpose of the regression is to identify if managers are exercising discretion. One of the
discretionary choices managers have is to set sales prices, which in turn influence sales volume. Thus,
inferences about managerial decisions on cost management are tied to managerial decisions about pricing.
In deriving the regression (5) above, we noted that the constant term would be zero if the
parameters of the production function remain constant. However, these parameters explain total cost,
given constant input prices, as a function of the (physical) volume of output. When sales price changes
are allowed, but not explicitly included as variables in the regression, we could observe differences in the
relative prices of SG&A resources and output in the constant term. In other words, the drift referred to
above could simply be price changes that differ between input resources and output volume. Most
important, tests for sticky costs using nominal data for quantity prevent researchers from making any
conclusions about cost management and the steps firms have taken in response to demand changes.
Accounting Classification of Costs
The regression equation in (5) is based on total costs. ABJ estimate the effect on one component
of cost, SG&A costs, using accounting data, such as that compiled in Compustat. In our sample, SG&A
costs represent about 25% of sales (see Table 2, Panel B) and about 30% of total operating costs. It is
important to consider the implications of using accounting data, which report costs using a classification
scheme designed for financial reporting purposes on tests that are meant to distinguish between
mechanical and managerial theories of cost behavior. As ABJ describe it (p. 49),
When volume falls, managers must decide whether to maintain committed
resources and bear the costs of operating with unutilized capacity or reduce
committed resources and incur the adjustment costs of retrenching and, if volume
is restored, replacing committed resources at a later date.
10
If the activity measure is not exogenous, then it is possible to evaluate the properties of the production function by
estimating the associated profit function, which expresses profits as a function of input and output prices only. Of
course, there are also measurement problems with constructs such as profit.
19
A problem with using one cost component from the accounting system for analyzing cost
behavior is that the classification of cost is itself open to managerial discretion. By substituting a single
component of cost for total costs in equation (5), we implicitly assume that resources are employed in a
specific (e.g., SG&A) task or are unemployed (unutilized). The following (simplistic) example highlights
the problem with this assumption.
Suppose a firm employs a single resource (one unit of labor) in addition to the manager, to
produce and sell a product. The cost of the labor ($1) is completely fixed by contract. Demand for the
resource is either one unit, which can be produced by one unit of labor, or zero. The manager assigns
labor to production when demand is one unit and to sales when demand is zero. Costs, in this example,
are technologically fixed. However, in the accounting system, SG&A is reported as $1 when demand is
zero and $0 (manufacturing costs are $1) when demand is one unit. Thus, when activity falls, SG&A
increases. Over time, based on the results of regression (5), it would appear as if costs are sticky, when,
in fact, they are contractually (mechanically) determined and invariant to activity.
It is difficult to think about any component of cost, especially one reported in the financial
accounting system that is not subject to managerial discretion, i.e., one that can be said to be related
mechanically to volume. Even costs (resources) that have an engineering (physical) relation to volume
(e.g., chemicals or raw materials) might not be completely determined mechanically because the costs
reported will include costs associated with waste and other inefficient practices. Accounting discretion is
an important factor to consider in assessing how well these tests capture cost management behavior. A
question at a more general level, however, is whether managers manage accounting costs or
economic costs. Estimating these effects for economic (opportunity) costs would be much more
difficult, but current tests assume, for example, that accounting measures (including, for example,
depreciation and amortization) are the focus of management cost reduction efforts.
20
Managerial Discretion
If the motivation for testing the sticky cost hypothesis is to distinguish between cost behavior that
results from a mechanistic relation between activity and cost from that which is the result of managerial
decision-making, it is important to clarify what cost behavior is uniquely associated with each theory. To
that end it is useful to consider two forms of discretion: that which is evident in current
(contemporaneous) decisions and that which is evident in past decisions.
In ABJ and related papers, cost behavior is evaluated by correlating current growth in SG&A (or
other cost) with current revenue growth. The hypothesized pattern of sticky costs is argued to be
inconsistent with the traditional fixed and variable cost model. However, it is one thing to reject the
traditional model and quite another to conclude that deviations in empirical cost behavior from
(symmetric) proportionality is evidence of managerial discretion. Indeed, the cost function that represents
the mechanical alternative to managerial discretion is itself the result of a managers (prior) decision.
ABJ use forward looking behavior by managers to bolster their conclusionnoting that managers
consider the permanence of activity changes as well as adjustment costs in deciding whether to alter
resource levels. However, they fail to note that in prior periods managers also considered future demand,
input prices and adjustment costs in selecting an appropriate production technology. In the current period
manager makes decisions about resources (and costs) given the constraints (or adjustment costs)
associated with prior technology choices.
The technology choice is not limited to the physical machinery for production. Labor contracts or
take-or-pay contracts for materials, for example, are entered into based on the costs of adjustments, the
likelihood that demand will fall, and so on. Thus, the (current) operating manager might be faced with
constraints on what can be done when demand falls. Whether this leads empirically to sticky costs
depends on the nature of the contracts and the technology as well as the magnitudes of the revenue growth
(or decline). The point is that a manager faced with these constraints might consider this to be
21
mechanical. Alternatively, because the constraints are the result of a previous managerial decision, it
might be appropriate to consider it managerial discretion.
Earlier we discussed the problem of nominal and real changes. An important decision the
manager makes is pricing. Faced with a decline in demand, for example, the manager might choose
between reducing SG&A resources or lowering the price to maintain demand. However, the empirical
tests of ABJ cannot distinguish between lower activity (physical volume) and lower revenue with the
same (or greater) physical volume. If SG&A resources are driven by (physical) activity, then attempts by
the manager to maintain market share by lowering price will show up in the data as stickiness of SG&A
costs when, in fact, SG&A costs are mechanically following (physical) sales volume. This might not be
symmetric with changes in demand conditions if the firm faces competitive pressures that make it
difficult to profitably raise price.
The purpose of making these points is to emphasize the difficulties of disentangling cost drivers
when managers have discretion over both resource utilization and the accounting for the costs of these
resources. This does not diminish the importance of understanding why and how firms manage costs.
What it suggests is that the distinction between mechanical changes in costs and changes in costs
resulting from managerial discretion is artificial and unlikely to be convincingly determined. More
directly, what does the sign of 2 tell us about cost management? Suppose that in all tests, 2 is
insignificantly different from zero. Does this imply there is no cost management or does it mean that
managers actively adjust costs to maintain a constant ratio with sales revenue? The empirical results on 2
cannot be used to infer anything about how firms manage costs without more structure on what we mean
by the term cost management.
4.
Developing Data to Study Cost Behavior and Cost Management

We conclude from our investigation that no single model of cost behavior prevails. In some
settings and time periods cost stickiness is the norm while in others it is not. (In fact, in some cases,
results that might be termed consistent with anti-stickiness, costs declining faster than sales declines, is
22
observed.) We further conclude that even when cost stickiness is observed, we can infer very little about
management decision-making or cost management from accounting data and measures of activity that
confound price and volume effects. In sum, we are in a position not unlike that described by Alan Blinder
(1991, p. 89), who surveyed research on sticky prices and concluded11:
[economists] attach great importance to the phenomena of wage and/or price
stickiness Yet, more than a half-century after Keynes published The General
Theory, the phenomenon itself [wage and price stickiness] remains poorly
understood. Just why are wages or prices sticky? It is not that economists have
ignored these questions. One could literally fill many volumes with good
empirical studies of wage and price stickiness, and many more with clever
theories purporting to explain these phenomena. Yet, despite all this work, the
range of admissible theories is wider than ever, and new theories continue to crop
up faster than old ones are rejected.
Like economists, accounting researchers understand the importance of cost behavior and of
theories of cost management, and have a range of plausible theories for why managers might take
decisions that would lead to incomplete or asymmetric cost reactions to activity changes.
Nonetheless, the answer to the question, why costs are (or are not) sticky, has eluded us. Blinder
(1991, pp. 89-90) argues that the main reason for the failure to make progress on the question of
why wages and prices are sticky:
is that most of the theories are empty in the following specific sense: Either
they involve unobservable variables in an essential way, or they carry no real
implications other than that prices are sluggish in some unmeasureable sense,
or both All of the theories share exactly the same prediction: that prices are
sticky So how are we to discriminate among them? (emphasis in original)
As we have done in this paper, Blinder identifies econometric and theoretical issues that are
unlikely to make prior research approaches fruitful before concluding:
This lack of scientific progress makes one wonder about the basic strategy that
economists have been pursuing. Is there a better way? (p. 89)
If it really matters which theory is correct, but conventional modes of economic
inquiry cannot adjudicate the dispute, economic science would appear to be in
deep trouble. Fortunately, one other common characteristic of the theories
11
Recall from our earlier note that economists use this term to refer to the delayed reaction of wages and prices to
changes in activity, not per se to asymmetric reactions to increases and decreases in activity.
23
suggests a way out: Virtually every theory of prices rigidity describes a chain of
reasoning which allegedly leads the firm to conclude that a change in price is
inadvisable (emphasis in original, p. 90)
Blinder argues separately that it does matter which theory is correct, both for the value of economics as a
descriptive science and as an important determinant of optimal macroeconomic policy.
The parallels to accounting research on sticky costs are striking. ABJ, Cooper and Kaplan (1998),
and Noreen and Soderstrom (1997) are but a few examples of papers that describe in some detail the
chain of reasoning that might cause managers to adjust (or not to adjust) the level and mix of resources.
For example, ABJ (p. 49) posit that:
When volume falls, managers must decide whether to maintain committed
resources and bear the costs of operating with unutilized capacity or reduce
committed resources and incur the adjustment costs of retrenching and, if volume
is restored, replacing committed resources at a later dateWhen demand
increases, managers increase committed resources to the extent necessary to
accommodate additional sales. When volume falls, however, some committed
resources will not be utilized unless managers make the deliberate decision to
remove them managers must evaluate the likelihood that a drop in demand is
temporary when deciding whether to adjust committed resources downward.
Manager decisions to maintain unutilized resources may also be caused by

personal considerations and result in a form of agency costs Managers may
retain unutilized resources to avoid personal consequences of retrenchment, such
as loss of status when a division is downsized or the anguish of dismissing
familiar employees
In this discussion, several factors are hypothesized to influence managers response to changes in
demand. First, managers consider the costs of adjusting the level and mix of resources to bring about a
new equilibrium relation between inputs and demand as compared to maintaining existing resources and
incurring costs of excess resources (decreased demand), or of peak utilization of resources or foregone
demand (increased demand).12 Adjustment costs take two forms: costs that are borne by the firm and costs
12
The economics literature on input factor adjustment costs is extensive. Lucas (1967) provides one of the earliest
discussions of capital adjustment processes. Hamermesh (1995) and Hamermesh and Pfann (1996) provide
contemporary reviews of the literature and discuss sources of adjustment costs. On the issue of asymmetric response
to changes in activity, Hamermesh and Pfann (1996) conclude that, The vast literature on dynamic factor demand
has been organized around the concept of costs of adjustment. The standard assumption has been that these costs are
convex and symmetricThe assumption is not supported by microeconomic data: on a variety of data sets a rapidly
growing body of economic research has demonstrated that other functional forms describe the technology of
24
that are borne by the manager. ABJ point out that in some settings, an agency problem may arise if
managers are motivated to avoid personal adjustment costs, accepting instead a suboptimal relation
between inputs and demand that costs the firm more than firm adjustment costs alone. After evaluating
the costs of adjusting resources to match the new level of demand, managers consider the transitory or
permanent nature of the demand shift. The permanence of the demand shift affects the adjustment
decision through its influence on the expected value of future savings associated with rebalancing inputs
to match demand and on the likelihood of incurring future adjustment costs. Although ABJ consider cost
management only in relation to exogenous shifts in demand, the above arguments also extend to other
exogenous shifts that are likely to precipitate managers reconsidering the level and mix of resources, in
particular, exogenous shifts in input prices or availability.
Although not raised in ABJ, economics provides additional candidate theories about why
managers might respond differently to changes in demand. We have already mentioned the need to
consider prior discretionary technology choices in crafting an appropriate response to demand changes.
Firms that select flexible technologies in prior periods have in effect already incorporated the real option
value of being capable to produce efficiently a wide range of output and, perhaps, a broad mix of products
(Marschak and Nelson [1962], Moel and Tufano [2002], Kallapur and Eldenburg [2005]). Although these
technology decisions are sunk with respect to current period demand shifts, a given level of demand
change would be associated with different adjustment costs for different technology choices. Another
avenue of theorizing that draws on the economics of industrial organization might consider the source of
the demand change as a factor in optimal cost management behavior. For example, we might expect
adjustment of individual inputs into production better. No doubt some firms behavior may be described as
symmetric quadratic costs; but on every one of the sets of microeconomic data in which it has been examined this
standard assumption is dominated by some alternative(emphasis in original, pp. 1287-8). Like Blinder (1991)
Hamermesh and Pfann argue that research is needed to identify characteristics of firms, industries, and input factor
markets that are associated with different patterns of adjustment cost. They recommend field-based research in this
endeavor: By expanding our direct observation of what businesses do and how managers thought processes
condition those actions, we should be able to gain additional industry insight to the nature of adjustment costs. This
approach means combining the accounting studies with the powerful organizing ability of economic theory to
provide information on the size of adjustment costs and their implications for economic behavior. (p. 1289)
25
managers to respond differently if demand decreases are the result of a competitors increased market
share rather than an overall decline in total demand. Indeed, a vigorous counter attack in the former case
might be predicted to lead to anti-sticky costs. There is precedent for considering how competition
affects cost structure (e.g., Banker and Johnston [1993]); thus, it seems reasonable to hypothesize that
variables that influence cost structure might also influence cost adjustments.
ABJ discuss (but do not test) the theory that managerial incentives might diverge from the
interests of the firm. In this respect, they continue to use economic theory to explain managerial decision
making. Although one must consider carefully whether the individual manager is the appropriate unit of
analysis in relation to firm-level costs, another potentially fruitful line of theorizing about cost
management decisions derives from the behavioral literature that examines cognitive factors affecting
managerial judgment and decision-making. For example, cognitive theory has proved useful in explaining
why managers incorporate irrelevant sunk costs, disregard relevant opportunity costs, and respond
differently to a given economic event under alternative accounting treatments (e.g., Lipe [1993], VeraMunoz [1998], Luft and Shields [2001]).
In sum, paralleling Blinders (1991) discussion of research on price rigidity, there are a number of
plausible theories, both economic and behavioral, for why managers might take different decisions in
response to exogenous shocks that create an imbalance in the relation between input resources and output.
The challenge is designing research programs to explore the contextual settings in which these theories
apply and the relative power of these theories to explain cost behavior.
In considering alternative approaches for testing theories of price rigidity, Blinder focuses on the
chain of reasoning which allegedly leads the firm to conclude that a change in price is inadvisable,
concluding that If actual decision makers really think the way one of these theories says, they ought to
know that they do (p. 90). To that end, he proposed a new (particularly for economists) mode of inquiry
in which managers are interviewed to shed light on the relative descriptive validity of twelve major
26
competing theories for explaining the speed of price adjustment in that managers firm.13 We believe that
a similar approach would be fruitful for creating a more systematic understanding of cost management
and the resultant cost behavior.
Similar to the Blinder study, the purpose of structured interviews would be to understand what
events precipitate managers considering a change to the level or mix of resources (e.g., demand changes,
input price or availability changes, competitive changes). Then, given that such a change is under
consideration, the interviewer would explore the relative descriptive validity of alternative theories on the
nature and extent of adjustment that is undertaken. The latter question includes consideration of whether
adjustment processes are symmetric with respect to increases and decreases in demand as well as the
factors associated with any asymmetric response. Important questions in the area of judgement and
decision making related to cost management that could be explored in interviews include: 1) how does the
managers assessment of adjustment costs compare with those of the firm, 2) which costs does the
manager consider to be eligible candidates adjustment, and 3) how are alternatives to adjustment
identified and valued? Finally, taking an information systems perspective, it would be useful to
investigate how management accounting information is implicated in the analysis of alternatives and
whether important management control systems (e.g., decision rights, reward systems) moderate
managers response to exogenous changes in demand. As Blinder notes, because it is likely that managers
will have a large (and complex) set of reasons for their actions, it is important that the interviewers be
able to deviate from the interview protocol to record the thought process.
Although, as in the Blinder study, we believe that interview data will provide important insight
into the cost management process, we would also expect that combining interviews with other research
methods of both a qualitative and quantitative natures, (e.g., field-based observation, analysis of archival
cost data (Anderson and Widener [2006])) would provide important corroborating evidence and a
stronger basis for testing theory (e.g., Hamermesh and Pfann, 1996, p. 1288). Moreover, although like
13
Blinder (1991) provides preliminary evidence from this study. Complete research results as well as interview
protocols and a discussion of the research design and data collection are found in Blinder et al. (1998).
27
Blinder, we believe that structured interviews and field-based research are more appropriate at this early
stage of understanding managerial behavior, later inquiries may employ more cost-effective means of
standardized data collection (e.g., mail, telephone, or internet-based surveys).
5.
Conclusion
ABJ sought to provide large sample evidence about whether managers take overt actions to align
costs with production activity and whether differences in adjustment costs or in managers personal
incentives produce differing responses to increased as compared to decreased activity.14 In this paper we
reconsider this question, first by revisiting the estimated models of prior research with additional data and
tests and then by considering anew the foundational models of production economics that define the
parameters of the estimated models.
We conclude from our empirical tests that although we find evidence of sticky SG&A costs in a
larger, 27-year sample, the results are not robust for many important sub-samples. We also find that the
results are considerably weaker when we limit our estimation sample to firm years in which cost changes
move in the same direction as activity changes. Although ABJ theorized about sticky costs in a manner
that implies this condition, they do not condition tests of cost stickiness on data meeting this criterion.
When we continue with this more circumscribed sample to re-examine the effects of industry and year,
we again find little evidence of widespread stickiness of SG&A costs. Finally, when we explore the
premise of sticky cost behavior for other types of costs that are also subject to managerial discretion,
including: advertising, labor wages, headcount, research and development, and property, plant and
equipment; we find no consistent mode of cost behavior; cost changes are almost equally likely to exhibit
14
Throughout the paper, we, like ABJ, focus on a rather narrow range of cost management. That is, we analyze
how managers react to a change in sales activity (changes in the demand side of the market). We generally are silent
about how managers react to changes in the supply side. For example, we ignore through much of the discussion
how managers cope with changes in input prices or supply disruptions. If researchers are to understand cost
management better, however, it is clear that we need to consider how managers might respond differently to
exogenous shocks that arise in the input versus the output markets.
28
asymmetries of both the sticky and anti-sticky (rising faster when revenues increase) variety as they are
to vary linearly with sales revenue changes.
These mixed results cause us to return to first principles; the economic theory of production that
is the foundation of cost equations. From this vantage point we evaluate critically the data and tests used
by ABJ in relation to the research question. We conclude that the tests and the data (SG&A costs) that
ABJ are unlikely to provide convincing tests of whether costs adjust asymmetrically with respect to
activity increases as compared to decreases. Moreover, we argue that although a perfect execution of the
test implied by economic theory would allow researchers to evaluate the sticky hypothesis, it would not
allow a researcher to discriminate among theories of the source cost behavior (e.g., overt cost
management versus a mechanistic movement of costs with activity). In particular, these data and tests
cannot distinguish between technological and managerial factors associated with stickiness or the absence
thereof. The problem with the approach is rooted in ambiguity about what defines managerial discretion
(cost management), how managerial discretion about redeploying versus releasing resources interacts
with recording costs in the accounting system, and data problems associated with the use of revenues to
proxy for activity when managerial responses to demand changes include both price and quantity of
output.
Although we critique the methods of prior studies, we believe that the research questions that
have been raised, namely, what explains cost behavior and the role of the manager in controlling costs,
are absolutely central to the field of management accounting. Consequently, we consider in the latter part
of the paper how future research might more meaningfully address these very important questions. We
find meaningful parallels in economic studies of price and input adjustment processes, and we argue that
the conclusion of these studies that the research must get closer to the data generating process to be able
to distinguish among a variety of theories also holds for management accounting research on cost
management.
Cost management practices of firms are a legitimate, yet long-neglected area of study for
management accountants. Although there are many studies about cost accounting systems and their
29
potential for improved decision-making, there are few studies about how managers use these or other
information systems to decide what steps to take to control costs. We believe that field-based evidence,
collected both from interviews with key managers who have authority and responsibility for cost
management and from firms archival accounting records, should be used to probe more deeply into cost
management practices. A few well-designed, field-based inquiries could then be used to inform the
construction of instruments to collect data on a wider scale, such as through mail or web-based surveys.
In sum, the conjectures of ABJ and others about the role of managerial discretion in actions that
lead to stickiness in cost were based on anecdotal evidence and observations about practice (e.g., Cooper
and Kaplan, 1998; Noreen and Soderstrom, 1997). These conjectures are reasonable and we are not
arguing that adjustment costs are not asymmetric or that managerial actions are unlikely to generate
asymmetric changes in cost in association with changes in production activity. Rather, we argue that the
definition of cost management and the theory that predicts asymmetric response to changes in demand is
not sufficiently advanced to produce meaningful tests using public accounting data on costs. In particular,
we are concerned that the contest between mechanistic cost behavior and managerial discretion as
explanations for observed sticky cost behavior, if it exists, is neither meaningful nor productive. We agree
that a better understanding of costs and cost management practices by firms is an area of research that has
been neglected by management accounting researchers and is a worthwhile endeavor. By returning to the
field to develop more precise theory about conditions that are likely to generate different cost
management behavior we will be in a better position at a later date to construct better large sample tests of
the theory.
30
Table 1
Steps Taken to Arrive at the Sample
and the Number of Observations Dropped at Each Step
Step
Observations (FirmYears) Deleted
Beginning raw sample (1978 2004)

1. Delete duplicate issues
2. Delete observations that are missing
data on either sales revenue or SG&A
costs for the current or preceding year
3. Delete observations with SG&A >
Sales for the current year
4. Delete observations where Sales or
SG&A are non positive for the current
or preceding year
5. Delete observations that exhibit
extreme values of the regression
variables (i.e., in the top and bottom
0.5% of the distribution)
Observations (FirmYears) Remaining

533,928
7,098
526,830
371,609
155,221
13,468
141,753
781
140,972*
2,272
138,700**
* Used for the descriptive statistics reported in Table 2.

** Used for the regression analysis reported in Table 3.
31
Table 2
Comparison of Descriptive Statistics from the Current 27-Year Sample
and from the ABJ 20-Year Sample
Panel A Comparative Sample Sizes of the Current Sample and the ABJ Sample:
Sample Size
Number of observations:
Number of firms:
Current Sample
140,972
17,338
ABJ Sample
64,663
7,629
Panel B Distribution of Annual Revenue and SG&A Costs:

Observations from the Full Sample
Sales revenue (millions)

SG&A costs (millions)
SG&A costs as a % of sales
Mean
$1,222.85
$223.12
27.23%
Standard
Deviation
$6,351.04
$1,095.70
18.50%
Median
$91.20
$18.81
23.06%
Lower
Quartile
$20.92
$4.86
14.09%
Upper
Quartile
$432.57
$79.51
35.29%
Standard
Deviation
$5,983.43
$1,042.49
17.79%
Median
$87.53
$17.49
22.62%
Lower
Quartile
$17.51
$4.56
13.66%
Upper
Quartile
$447.75
$79.12
34.31%
Descriptive Statistics Reported by ABJ*
Sales revenue (millions)

SG&A costs (millions)
SG&A costs as a % of sales
Mean
$1,277.09
$229.45
26.41%
* Source: ABJ Table 1, Panel A
32
Table 2 (continued)
Comparison of Descriptive Statistics from the Current 27-Year Sample
and from the ABJ 20-Year Sample
Panel C Periodic Fluctuations in Revenue and SG&A Costs:
Observations from the Full Sample
Sales
revenue
SG&A
costs
Percentage
of FirmYears with
Negative
Change
from
Previous
Period
28.89%
Mean
Percentage
Decrease
Across
Periods
27.03%
Standard
Deviation of
Percentage
Decreases
Across
Periods
Median
Percentage
Decreases
Across
Periods
Upper
Quartile of
Percentage
Decreases
Across
Periods
Lower
Quartile of
Percentage
Decreases
Across
Periods
16.17%
16.50%
10.67%
22.47%
4.34%
16.06%
16.45%
10.50%
22.28%
4.21%
18.64%
Median
Percentage
Decreases
Across
Periods
10.99%
Upper
Quartile of
Percentage
Decreases
Across
Periods
23.76%
Lower
Quartile of
Percentage
Decreases
Across
Periods
4.38%
16.40%
10.07%
21.63%
3.94%
Descriptive Statistics Reported by ABJ*
Sales
revenue
SG&A
costs
Percentage
of FirmYears with
Negative
Change
from
Previous
Period
27.01%
Mean
Percentage
Decrease
Across
Periods
17.45%
24.98%
15.67%
Standard
Deviation of
Percentage
Decreases
Across
Periods
* Source: ABJ Table 1, Panel B
33
Table 3
Comparison of the Basic Regression Model (Model (I)) Results
from the Current Sample and from the ABJ Sample
Regression specification is for ABJs Model (1):
SG & Ai,t
Revenuei,t
Revenuei,t
log
= 0 + 1 log
+ i,t
SG & Ai,t 1
Revenuei,t 1
Revenuei,t 1
Observations from the Current Sample (1978 2004):
Coefficient estimate
t-statistic
(Intercept)
(Direct
Effect)
(Sticky
Measure)
0.0419
(54.73)
0.5733
(239.73)
-0.0395
(-7.10)
Adj. R2
0.4042
Number of
Observations
138,700
Results Reported by ABJ (1979 1998)*:
(t-statistic)
(Intercept)
(Direct
Effect)
(Sticky
Measure)
0.0481
(39.88)
0.5459
(164.11)
-0.1914
(-26.14)
Adj. R2
0.3663
Number of
Observations
63,958
* Source: ABJ Table 2
34
Table 4
Relative Size of the Restricted and Full Sample and
Results of the Basic Regression Model (Model (I)) Results
When the Sample Excludes Positive (Negative) Changes in SG&A
for Negative (Positive) Sales Changes
(The Restricted Sample)
Panel A Sample Sizes, Full and Restricted

Observations Percentage
140,972
100.0%
16,259
11.5
13,646
9.7
111,067
78.8%
Full sample size (before trimming)

less SG&A increases following Sales decline
less SG&A decreases following Sales increase
Restricted sample size (before trimming)
Panel B Regression Results for the Restricted Sample:

SG & Ai,t
Revenuei,t
Revenuei,t
log
= 0 + 1 log
+ i,t
SG & Ai,t 1
Revenuei,t 1
Revenuei,t 1
t-statistic
(Intercept)
(Direct
Effect)
(Sticky
Measure)
0.0495
(68.17)
0.6590
(314.97)
0.1393
(27.97)
Adj. R2
0.6283
Number of
Observations
109,303
35
Table 5
Basic Regression Model (Model (I)) Results
For Individual Years
Panel A Full Sample (Includes positive (negative) changes in SG&A for negative (positive) Sales
changes):
1
0
(Intercept)
Year
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
Mean
z-statistic
0.0709
0.0709
0.0704
0.0728
0.0486
0.0433
0.0592
0.0687
0.0389
0.0408
0.0337
0.0450
0.0418
0.0251
0.0270
0.0305
0.0267
0.0446
0.0449
0.0455
0.0402
0.0324
0.0343
0.0235
0.0370
0.0344
0.0432
14.99
t-statistic
19.07
18.52
17.74
17.99
11.30
9.48
13.81
15.34
8.43
9.49
8.14
11.56
10.67
6.63
6.98
8.75
7.35
12.97
13.25
12.42
10.59
8.42
9.11
6.21
10.85
9.78
(Direct
Effect)
0.4874
0.5274
0.5988
0.5263
0.5253
0.5932
0.5293
0.5042
0.5749
0.5515
0.5331
0.5257
0.4948
0.5859
0.6156
0.6471
0.6105
0.6345
0.6384
0.6601
0.6318
0.5928
0.4619
0.4068
0.4106
0.5262
2
t-statistic
(Sticky
Measure)
36.15
39.70
46.30
35.77
39.35
45.07
38.95
38.12
43.43
43.44
39.22
39.29
34.59
44.07
49.38
60.54
57.11
64.22
69.03
65.26
58.19
57.96
34.60
27.56
33.46
43.05
0.0710
-0.0090
-0.1623
0.0078
0.0162
-0.1294
0.0188
0.0564
-0.0588
-0.0264
0.0351
0.0821
0.0788
-0.0332
-0.0897
-0.1659
-0.1414
-0.0570
-0.0601
-0.2094
-0.0639
-0.0658
0.0129
0.2077
0.2504
-0.0431
0.5563
41.06
Number of Negative and Significant Coefficients

Number of Positive and Significant Coefficients
Number of Insignificant Coefficients
t-statistic
2.41
-0.31
-5.40
0.27
0.58
-3.80
0.64
2.12
-1.86
-0.88
1.14
2.85
2.66
-1.15
-2.95
-5.89
-5.14
-2.07
-2.31
-7.93
-2.39
-2.30
0.50
7.70
8.34
-1.33
-0.0220
-1.04
11
6
9
36
Table 5 (continued)
For Individual Years
Panel B Restricted Sample (Excludes positive (negative) changes in SG&A for negative (positive)
Sales changes):
1
0
(Intercept)
Year
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
Mean
z-statistic
0.0611
0.0745
0.0733
0.0713
0.0462
0.0584
0.0628
0.0743
0.0537
0.0552
0.0406
0.0496
0.0415
0.0294
0.0417
0.0446
0.0413
0.0588
0.0499
0.0514
0.0530
0.0486
0.0359
0.0200
0.0471
0.0441
0.0507
18.97
t-statistic
18.13
21.79
19.55
17.41
11.32
13.83
15.20
17.37
12.54
14.05
10.52
13.42
10.70
7.90
10.94
13.34
12.67
17.62
15.78
15.41
14.73
13.43
9.78
5.01
14.31
13.39
(Direct
Effect)
0.6124
0.5962
0.6455
0.6059
0.6517
0.6412
0.6445
0.5989
0.6671
0.6225
0.6570
0.6528
0.6042
0.6916
0.6607
0.6959
0.6694
0.6761
0.7087
0.7281
0.7034
0.6460
0.6216
0.6263
0.5515
0.6054
2
t-statistic
51.63
53.35
56.54
45.39
54.80
56.69
53.48
51.04
58.45
57.38
55.25
54.35
47.88
57.63
59.34
74.28
75.74
76.59
88.64
85.06
74.17
73.91
51.38
43.83
50.85
55.84
0.6469
79.78

(Sticky
Measure)
0.1174
0.2087
0.1353
0.1461
0.0840
0.1997
0.1430
0.2083
0.1862
0.1661
0.1845
0.2029
0.1957
0.0447
0.2077
0.0882
0.1329
0.2189
0.1055
0.0718
0.0833
0.2002
0.0612
0.1820
0.3752
0.1568
t-statistic
4.53
8.25
4.82
5.54
3.38
6.49
5.46
8.66
6.57
6.28
6.82
7.87
7.41
1.71
7.36
3.39
5.55
8.50
4.62
3.05
3.54
7.88
2.64
7.00
14.06
5.28
0.1596
11.64
0
25
1
37
Table 6
For Industries at the 2-Digit SIC Level
Summary of Coefficient Estimates of 2
Panel A Full Sample (Includes positive (negative) changes in SG&A for negative (positive) Sales
changes):
SIC Code
01
02
07
08
09
10
12
13
14
15
16
17
20
21
22
23
24
25
26
27
28
29
30
Estimated 2
- 0.1312
0.0948
0.0807
0.1415
- 0.0669
0.0147
0.4655*
0.0922*
- 0.4615*
0.2356
- 0.1143
- 0.1585
- 0.0017
- 0.1399*
- 0.1569*
- 0.1415*
- 0.1127*
- 0.0448
0.1281*
- 0.0145
0.1372*
0.0955
- 0.1608*
SIC Code Estimated 2

31
- 0.0950
32
- 0.0559
33
- 0.1964*
34
- 0.1255*
35
- 0.1534*
36
- 0.1218*
37
- 0.0964*
38
- 0.0553*
39
- 0.0404
40
- 0.1978
41
0.5476
42
- 0.0738
44
- 0.4074*
45
0.1572
46
- 0.2470*
47
0.2740*
48
0.1589*
49
0.2152*
50
- 0.1009*
51
0.0904*
52
- 0.3923*
53
- 0.2032*
SIC Code
54
55
56
57
58
59
60
61
62
63
64
65
67
70
72
73
75
76
78
79
80
81
Estimated 2
0.0998*
0.0903
- 0.1268*
- 0.2180*
0.1244*
- 0.0801*
- 0.3187*
0.1533
- 0.1025
- 0.0123
- 0.0317
0.2265*
0.2395*
0.0697
0.1303
0.0062
- 0.2864*
0.4783*
0.0669
- 0.0435
0.0661
0.7826
* Significant at the 5% level (two-sided test)

22
13
32
38
Table 6 (continued)
For Industries at the 2-Digit SIC Level
Summary of Coefficient Estimates of 2
Panel B Restricted Sample (Excludes positive (negative) changes in SG&A for negative (positive)
Sales changes):
SIC Code
01
02
07
08
09
10
12
13
14
15
16
17
20
21
22
23
24
25
26
27
28
29
30
Estimated 2
0.0671
0.2480
0.1158
- 0.1379
- 0.0631
0.4349*
0.7931*
0.3120*
- 0.5661*
0.1689*
- 0.0977
0.1064
0.1244*
- 0.1251
- 0.1098*
0.0383
- 0.0457
- 0.0352
0.2003*
0.1067*
0.2849*
0.6366*
- 0.0475
SIC Code Estimated 2

31
- 0.0565
32
0.0580
33
0.0186
34
- 0.0328
35
0.0511*
36
0.0829*
37
- 0.0782*
38
0.1457*
39
0.1306*
40
0.0720
41
1.2038*
42
- 0.0123
44
- 0.0508
45
0.1916*
46
0.4938
47
0.4825*
48
0.4992*
49
0.4054*
50
0.0208
51
0.3375*
52
- 0.3624*
53
- 0.1374*
SIC Code
54
55
56
57
58
59
60
61
62
63
64
65
67
70
72
73
75
76
78
79
80
81
Estimated 2
0.0834*
0.0576
- 0.0728*
- 0.1148*
0.3239*
0.0642
0.1189*
0.1265
- 0.0242
0.5655*
0.0033
0.2522*
0.2591*
0.2050*
0.2441*
0.2392*
0.2426
0.6968*
0.3057*
0.1897*
0.3108*
0.8030
* Significant at the 5% level (two-sided test)

7
32
28
39
Table 7
Results of the Basic Regression Model (Model (I)) Results
for Various Cost Categories
Costi,t
Revenuei,t
Revenuei,t
log
= 0 + 1 log
+ i,t
Costi,t 1
Revenuei,t 1
Revenuei,t 1
Advertising Costs
Full
Restricted
Sample
Sample
0.0213
0.0742
(6.67)
(24.09)
R&D Costs
Full
Restricted
Sample
Sample
0.0534
0.1015
(21.42)
(42.22)
Labor Costs
Full
Restricted
Sample
Sample
0.0405
0.0394
(34.77)
(35.89)
Employees
Full
Restricted
Sample
Sample
-0.0008
0.0251
(-0.92)
(26.44)
PP&E Costs
Full
Restricted
Sample
Sample
0.0414
0.0758
(32.75)
(54.39)
0.5525
(61.08)
0.8683
(107.77)
0.3532
(54.45)
0.6175
(103.32)
0.6219
(130.75)
0.7053
(169.38)
0.3722
(151.80)
0.4813
(199.06)
0.4721
(135.50)
0.6719
(189.47)
0.2446
(11.76)
0.6626
(35.75)
0.1356
(8.45)
0.5753
(38.20)
-0.0828
(-6.73)
0.1535
(13.57)
0.0289
(5.91)
0.2958
(57.12)
0.1535
(19.02)
0.3384
(42.22)
No. of
Observations
Adjusted R2
47,901
33,142
59,407
40,273
32,420
26,041
159,321
113,094
149,179
104,499
0.1398
0.4667
0.0889
0.4000
0.4309
0.6370
0.2170
0.4700
0.1930
0.4350
40
References
Anderson, M. C., R. D. Banker, and S. N. Janakiraman (2003). Are selling, general, and administrative
costs "sticky"? Journal of Accounting Research 41(1), 47 63.
Anderson, S.W. and S.K. Widener. (2006) Doing Quantitative Field Studies in Management Accounting.
In C.S. Chapman, C.; A.G. Hopwood, and M.D. Shields (Eds.). Handbook of Management Accounting
Research. (Vol. 1) Oxford: Elsevier.
Banker, R. D. and L. Chen (2006). Predicting earnings using a model based on cost variability and cost
stickiness.The Accounting Review. 81 (2), 285-308.
Banker, R. D. and H. H. Johnston (1993, July). An empirical study of cost drivers in the U.S. airline
industry. The Accounting Review 68 (3), 576601.
Banker, R. D., M. Ciftci, and R. Mashruwala (May 2006). The Effect of Prior-Period Changes on Cost
Behavior, Unpublished Working Paper.
Blinder, A. S. (1991, May). Why are prices sticky? Preliminary results from an interview study. The
American Economic Review 81(2), 8996.
Blinder, A. S., Canetti, E.R.D, Lebow, D.E. and J.B. Rudd (1998). Asking About Prices: A New Approach
to Understanding Price Stickiness. Russell Sage Foundation Publications.
Cooper, R. and R. Kaplan (1998). The design of Cost Management Systems: Text, Cases, and Readings.
Upper Saddle River, NJ: Prentice-Hall.
Hamermesh, D.S. (1995) Labour demand and the source of adjustment costs. The Economic Journal. 105
(430): 620-634.
Hamermesh, D.S. and G.A. Pfann. (1996) Adjustment costs in factor demand. Journal of Economic
Literature. 34 (3): 1264-1292.
Kallapur, S. & Eldenburg, L. (2005). Uncertainty, Real Options, and Cost Behavior: Evidence from
Washington State Hospitals. Journal of Accounting Research. (forthcoming).
Lipe, M.G. (1993) Analyzing the variance investigation decision: The effects of outcomes, mental
accounting, and framing. Accounting Review. 68 (4): 748-764.
Lucas, R.E. (1967) Optimal investment policy and the flexible accelerator. International Economic
Review. 8 (1): 78-85.
Luft, J. and M. Shields. (2001) Why does fixation persist? Experimental evidence on the judgment
performance effects of expensing intangibles. Accounting Review. 76 (4): 561-588.
Marschak, T. & Nelson, R. (1962). Flexibility, uncertainty and economic theory. Metroeconomica 14: 4258.
Mintz, S. Penny Pinchers: Our Sixth Annual SG&A Survey. CFO Magazine, December 1999.
41
Moel, A. & Tufano, P. (2002). When are Real Options Exercised? An Empirical Study of Mine Closings.
Review of Financial Studies. 15 (1): 35-64.
Noreen, E. and N. Soderstrom (1994). Are overhead costs strictly proportional to activity?: Evidence from
hospital departments. Journal of Accounting and Economics 17 (1-2), 255278.
Noreen, E. and N. Soderstrom. (1997). The accuracy of proportional cost models: Evidence from Hospital
Service Departments. Review of Accounting Studies. 2: 89-114.
Subramaniam, C. and M. L. Weidenmier (2003). Additional evidence on the sticky behavior of costs.
Working paper, Texas Christian University.
Vera-Munoz, S.C. (1998) The effects of accounting knowledge and context on the omission of
opportunity costs in resource allocation decisions. Accounting Review. 73 (1): 47-73.
42

SSRN Id975135

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

SSRN Id975135

Uploaded by

Copyright:

Available Formats

Understanding Cost Management:

What Can We Learn from the Evidence on Sticky Costs?

Understanding Cost Management

Understanding Cost Management

Although we reject the mechanical model as a meaningful benchmark, we believe that it is

Understanding Cost Management

Understanding Cost Management

ABJ RevisitedTesting for Cost Stickiness

2.1. The Basic Sticky Cost Model

Understanding Cost Management

2.2 The Sample

Understanding Cost Management

(concurrently) from the sample remaining after step 5.

Understanding Cost Management

2.3 What Constitutes Sticky Costs?

Understanding Cost Management

Understanding Cost Management

Understanding Cost Management

Understanding Cost Management

Understanding Cost Management

2.6 A Summary on the Empirical Analysis of Sticky Costs

Understanding Cost Management

Technology or Managerial Behavior? Can We Tell?

Understanding Cost Management

3.1 The Cost Function

Understanding Cost Management

Ct = Bt qt1/( + ) ( pKt pLt ) .

Understanding Cost Management

Understanding Cost Management

3.2 Logical (and Terminological) Issues When Testing for Stickiness

Real versus Nominal Activity

Understanding Cost Management

Understanding Cost Management

Understanding Cost Management

Understanding Cost Management

Developing Data to Study Cost Behavior and Cost Management

Understanding Cost Management

Understanding Cost Management

Manager decisions to maintain unutilized resources may also be caused by

Understanding Cost Management

Understanding Cost Management

Understanding Cost Management

Understanding Cost Management

Understanding Cost Management

Understanding Cost Management

Understanding Cost Management

Observations (FirmYears) Deleted

Beginning raw sample (1978 2004)

Observations (FirmYears) Remaining

* Used for the descriptive statistics reported in Table 2.

Understanding Cost Management

Panel B Distribution of Annual Revenue and SG&A Costs:

Sales revenue (millions)

Descriptive Statistics Reported by ABJ*

Sales revenue (millions)

* Source: ABJ Table 1, Panel A

Understanding Cost Management

Descriptive Statistics Reported by ABJ*

* Source: ABJ Table 1, Panel B

Understanding Cost Management

Results Reported by ABJ (1979 1998)*:

* Source: ABJ Table 2

Understanding Cost Management