Professional Documents
Culture Documents
Abstract
The use of activity-based costing (ABC) has been steadily, if not rapidly, spreading
on an international level. This fact notwithstanding, the economic bene®t associated
with adopting ABC is suspect, at best. In an eort to shed additional light on this
apparent dilemma, this paper empirically investigates the stock market eects of an-
nouncing the adoption of an ABC system. The research methodology includes both
parametric and non-parametric tests for excess market returns from a seemingly unre-
lated regressions model with a matched pairs sample of ®rms. The analysis indicates that
the installation of an ABC system in the United States is not associated with a signi-
®cant (either positive or negative) stock market reaction. Ó 1999 Elsevier Science Inc.
All rights reserved.
1. Introduction
Ever since the 1980s, it has been widely argued that cost management sys-
tems are in need of major change (e.g., Johnson and Kaplan, 1987, pp. 1±18).
q
The authors wish to thank Wolfgang Bessler, Kwok Leung, Eric Noreen, Albert Paulson, Steve
Loeb, Kimberly Smith, Krishnamoorthy Surysekar, Jerold Zimmerman, and the participants at the
research workshops of The University of Essex, London Business School, University of
Manchester, The University of Toronto, and Rensselaer Polytechnic Institute for their comments
on earlier drafts of this paper.
*
Corresponding author. Tel.: +1-301-405-2255; fax: +1-301-405-0359; e-mail: lgor-
don@rhsmith.umd.edu
1
Ernst & Young Alumni Professor of Managerial Accounting.
0278-4254/99/$ ± see front matter Ó 1999 Elsevier Science Inc. All rights reserved.
PII: S 0 2 7 8 - 4 2 5 4 ( 9 9 ) 0 0 0 0 9 - 5
230 L.A. Gordon, K.J. Silvester/Journal of Accounting and Public Policy 18 (1999) 229±251
One change, which has received widespread attention, concerns the allocation
of indirect costs via activity-based costing (ABC). In fact, ABC has become an
issue of increasing and fundamental concern to numerous researchers and
practitioners around the world. In response to this concern, numerous ®rms
have either implemented, or are considering implementing, ABC.
For example, a 1997 survey by the Cost Management Group (1998, p. 1) of
the Institute of Management Accountants notes that 39% of its member or-
ganizations have ``at least approved ABC implementation.'' A survey of UK
®rms (Innes and Mitchell, 1995, p. 141) shows that almost 19.5% of the re-
spondents were using ABC and that 27.1% were considering its adoption. A
survey of 134 Finnish manufacturing units (Lukka and Granlund, 1996, p. 17)
found that 5% were in the process of implementing ABC and another 24% were
considering its use. Although ABC has only been sparsely used by Japanese
®rms in the past, there are many ®rms in Japan that are seriously considering
introducing ABC systems, with a particular emphasis on activity-based man-
agement concepts (Sakurai, 1995, p. 26). Hence, while US companies may have
taken the initial lead, there is clearly a signi®cant international movement to-
ward adopting ABC.
Despite this movement toward adopting ABC, its bene®ts have been as-
serted largely through anecdotal, self-reported survey data, and case study
evidence (see the next section). Accordingly, it is not surprising that many
researchers question the inherent value of an ABC system. For example,
Bromwich and Bhimani (1989, p. 3) note that the evidence does not suggest
that the adoption of ABC will improve pro®ts, while Dopuch (1993, p. 617)
observes that the payo from an ABC system might not justify its imple-
mentation costs. Shields (1995, pp. 159±161) shows that the bene®ts derived
from implementing and using ABC vary greatly among ®rms and depend upon
several behavioral and organizational dimensions. Innes and Mitchell (1995,
p. 150) argue that it may be dicult to separate the real bene®ts of ABC from
the claimed bene®ts, since those claiming the bene®ts are often the individuals
responsible for the adoption and development of the ABC systems within their
own companies (i.e., a vested interest issue is present). Some have even gone so
far as to challenge the basic logic underlying the use of ABC (e.g., Piper and
Walley, 1991, p. 42; Piper and Walley, 1990, p. 54) and ABC's relevance to
operational decisions (Johnson, 1992, p. 141). In a recent survey-based em-
pirical study by Foster and Swenson (1997, p. 136), it was concluded that the
way ABC's success is measured has an eect on studies regarding the derived
bene®ts of ABC.
As indicated by the above, we seem to be on the horns of the following
dilemma. On the one hand, the number of ®rms adopting ABC seems to be
spreading rapidly into a multitude of countries. On the other hand, there is
good reason to doubt whether the adoption of an ABC system results in an
economic bene®t to a ®rm. In fact, mounting evidence suggests that many of
L.A. Gordon, K.J. Silvester/Journal of Accounting and Public Policy 18 (1999) 229±251 231
the ABC systems implemented in the late 1980s and early 1990s were failures
(e.g., Roberts and Silvester, 1996, p. 23; Malmi, 1997, p. 460). Accordingly, the
purpose of our study is to empirically investigate the performance eects as-
sociated with the adoption of an ABC system. Our study diers signi®cantly,
however, from other work published in the area in that the stock market eect
of announcing the installation of an ABC system is assessed. Thus, our mea-
sure of success is based on publicly available stock market data. Further, the
identi®cation of when a ®rm introduced ABC is also based on publicly avail-
able information.
The results of our study indicate that, in general, the announcement of the
adoption of an ABC system among US-based ®rms did not have a signi®cant
(either positive or negative) stock market eect. These ®ndings are particularly
relevant to organizations that are considering, but have not yet implemented,
an ABC system.
The remainder of our paper will proceed as follows. In Section 2, the main
hypothesis of the empirical study is developed. Section 3 describes the empirical
study conducted to test this hypothesis, including methodological concerns.
The results and implications of the study are discussed in Section 4 of the
paper. Section 5 oers some concluding comments.
2. Hypothesis development
2
Conceptually, an ABC system could aid in shifting indirect costs to direct cost categories (where
the cost object is a product's cost by uncovering previously unidenti®ed relationships between the
factors of production and the products being produced).
232 L.A. Gordon, K.J. Silvester/Journal of Accounting and Public Policy 18 (1999) 229±251
ABC systems have many potential bene®ts, including: (1) helping to identify
non-value added activities, (2) improving the ability of managers to make
pricing, production, and investment decisions through the provision of more
accurate product and process costs, and (3) improving the integration of the
strategic and operating/production processes of the organization (Gordon,
1998, p. 142). Case studies supporting these claimed bene®ts are plentiful (e.g.,
Bhimani and Pigott, 1992; Cooper et al., 1992a,b; Cooper and Turney, 1989;
Kaplan, 1990; Innes and Mitchell, 1990, 1991; Turney and Anderson, 1989).
Despite the above, empirical evidence on the value of ABC is limited. Foster
and Gupta (1990, p. 327), for example, found that non-volume cost drivers add
little explanatory power to the behavior of indirect costs, 3 while Banker and
Johnston (1993, p. 587) show both volume-based and operations-based cost
drivers to be statistically signi®cant within the US airline industry. Spicer
(1990, p. 143) notes that ``we do not as yet have any real systematic evidence
that relate the use of ABC systems to improved internal and external perfor-
mance measures.'' Bromwich and Bhimani (1989, p. 3), while noting that ``. . .
activity costing changes product costing substantially'', also argue that there
``is little to suggest that it enhances pro®ts.'' Noreen (1991, p. 165) also points
out that ``. . . the widespread and rapid adoption of ABC systems is an inter-
esting phenomenon in and of itself ± particularly since it is not obvious that on
balance ABC systems as implemented provide greater bene®ts relative to costs
than any other possible costing systems.'' Shields (1995, p. 159), in a survey of
143 ®rms, found a wide variation across ®rms in terms of the bene®ts derived
from ABC. Innes and Mitchell (1995, p. 150), while noting that ``those
adopting ABC considered its application had been successful,'' pointed out
that the ``survey respondents had a vested interest in their ®rm's ABC appli-
cation.''
Innes and Mitchell (1995, p. 151) argue that there is a need for more em-
pirical (and objective) research on the topic. Young and Selto (1991, p. 296),
Anderson (1995, p. 48), and Shields (1995, p. 154) also call for further empirical
research on the performance eects of ABC. Shank (1989, p. 47) points out
that there is nothing conceptually new in activity-based accounting. Hence, the
question regarding the net value of ABC remains unresolved, in large part due
to the limited systematic and objective empirical evidence regarding its per-
formance eects. Nevertheless, as noted earlier, the use of ABC systems is
steadily, if not rapidly, spreading on an international level.
In constructing an objective empirical analysis of the net bene®ts of an ABC
implementation, the methodology should be based upon a conceptual frame-
3
Although only volume-related cost drivers were signi®cant in their study, Foster and Gupta
(1990, pp. 335,336) warn against generalizing from their results due to data limitations (i.e., all of
their data were from 37 facilities of a single electronics company).
L.A. Gordon, K.J. Silvester/Journal of Accounting and Public Policy 18 (1999) 229±251 233
4
In a related study, Nair (1979, pp. 239±241) has empirically shown that accounting techniques
can have an eect on the relative ranking of economic investment models.
234 L.A. Gordon, K.J. Silvester/Journal of Accounting and Public Policy 18 (1999) 229±251
3. Empirical study
5
One of the primary alleged bene®ts of ABC is its explicit dierentiation of process/activity costs.
Such recognition of the cost of production processes can play a crucial role in the elimination of
non-value added activities or ineciencies (Kelly, 1991, p. 43).
L.A. Gordon, K.J. Silvester/Journal of Accounting and Public Policy 18 (1999) 229±251 235
technology. The Dos Santos et al. (1993) study isolates a positive and signi®-
cant market reaction to announcements concerning investments in innovative
technologies, by utilizing both traditional sources (The Wall Street Journal and
PR Newswire and non-traditional sources (American Banker, American Metal
Markets, Minneapolis Star, and Greenville News).
Interestingly, the Securities and Exchange Commission (SEC), itself, deems
it necessary to closely monitor the timing and eect of the release of such
periodicals because of their impact upon the market. Under Section 10(b) of
the US Securities Exchange Act of 1934 (US Congress, 1934 ), the SEC has
historically sought to (among other goals) ``guarantee that no single group of
traders enjoys unfair informational advantages over the rest of the market''
(Harpaz, 1985, p. 1036). Although originally targeted at scalping by journal-
ists, 6 the SEC has been fairly aggressive in attempting to extend the scope of
Section 10(b) to include a general duty by ®nancial reporters to the readership
(Harpaz, 1985, pp. 1035±1038).
Business Week has a policy that precludes selling advance copies of their
periodical before the ocial release date for each issue (see footnote 9). This
policy regarding no early releases, combined with SEC's timing concerns, prior
impact of similar periodical articles, and our veri®cation that the ABC adop-
tions were not previously revealed through another news medium, provide
substantiation that the Business Week article released new and potentially
valuable information to the market on the day of the announcement.
Analysis of reactions in the information market. Intuitively, it seems rea-
sonable to argue that movements in security markets should be a re¯ection of
the activity in the information markets. Indeed, as noted in Foster (1979, pp.
273±274), the existence of ``superior insight'' in analyzing public information
does not preclude an ecient market hypothesis. Therefore, one would expect
that the Business Week article (if informative to ®nancial markets) would also
have generated an immediate and corresponding amount of interest in the
general manufacturing and business community. The novelty of this article and
its impact have also been noted by Campi (1992, p. 5).
Furthermore, as noted by Beaver (1981, p. 32), ``...empirical studies of
change in accounting methods are viewed as testing market eciency with
respect to more information than merely the knowledge that a change took
place.'' Therefore, to the extent that the market did not already fully under-
stand the implications of the adoption of ABC, the extensive Business Week
article would have conveyed valuable information to the market. Accordingly,
it is crucial to emphasize that the Business Week article by Port et al. (1988) not
6
Scalping is generally considered to occur when a journalist uses private advance information
about the content of forthcoming articles to personally take advantageous positions in the stock
market.
L.A. Gordon, K.J. Silvester/Journal of Accounting and Public Policy 18 (1999) 229±251 237
only revealed the ABC adopting ®rms but also provided substantial discussion
regarding the initiation, utilization, and purported bene®ts of such costing
innovations. Ceteris paribus, it seems reasonable to argue that the market
impact of this extensive and substantive Business Week article concerning ABC
may actually be stronger than the announcements found in the traditionally
utilized release formats (e.g., The Wall Street Journal, corporate proxy state-
ments, etc.). In addition, the Business Week article reports on the adoption of
ABC by ®rms during the period when the costing methodology was receiving a
great deal of attention. Hence, if the market was ever to be sensitive to the
adoption of ABC by ®rms, the mid-to-late 1980s would have surely been the
time period for this to happen.
to control for size. 7 The matching procedure consisted of choosing the control
®rm within the same 3-digit or 4-digit SIC code as the experimental ®rm with
the closest match on Total Net Assets. There were actually 16 ®rms named in
the Port et al. (1988, pp. 108, 112) article. However, four of these ®rms were not
contained in the 1988 and 1989 CRSP tapes for the time period around the
event under consideration. Of the 12 usable experimental ®rms, matches were
obtainable for 10 of the ®rms. 8 A list of both the experimental ®rms and
matched control ®rms is provided in Table 1. Also included in Table 1 is a
statistical assessment of the quality of the matches. Based upon a Wilcoxon
Signed Rank Test, the experimental and control groups of ®rms do not dier
signi®cantly (at a 0:10, two-tailed) when matched on Total Net Assets as a
control for size.
The release date of the Port et al. (1988) article, for purposes of our study,
was Monday, 30 May 1988 (i.e., the day the 6 June 1988 issue of Business Week
was released to the general public via placement on the news stands). 9 The 31st
of May is the ®rst trading day following the release of Port et al. (1988) an-
nouncement on the 30 May holiday. Hence, we refer to 30 May as the release
date and 31 May as the announcement date. Since Business Week was released
to the general news stands while the market was closed on 30 May, the market
actually had two days (both 30 and 31 May) to assess the informational
content before impounding the value of the information in stock prices on 31
May.
This Business Week issue may also have been released to a limited number
of selected news stands on Friday, 27 May 1988 (see footnote 9). To control for
this possibility, Friday, 27 May 1988 is also included in the event period.
Therefore, given the nature of the Business Week release, the event period
was constructed to contain four separate and consecutive trading days (i.e., one
trading day before, the trading day of the announcement, and two trading days
7
As an additional control, a review of the 1988 Wall Street Journal Index was made to check for
announcements concerning the ®rms. The review indicated that no sales or earnings announce-
ments were made by any of the experimental or control ®rms during the event period (27 May
1988±2 June 1988)
8
Of the 12 experimental ®rms, nine were matched on a 4-digit SIC code, one was matched on a 3-
digit SIC code, and two ®rms were unable to be matched. One unmatched ®rm did not have a
match at the 3-digit SIC code level, and the 2-digit SIC code was deemed inadequate as an industry
match. The second unmatched ®rm did not have a match even at a 2-digit level. In general, the size
of these two unmatched ®rms (as measured by Total Net Assets) far over-shadowed the size of any
potential industry-matched control ®rm.
9
This information was obtained from Business Week.
L.A. Gordon, K.J. Silvester/Journal of Accounting and Public Policy 18 (1999) 229±251 239
Table 1
Experimental and control ®rms
Experimental ®rms Control ®rms
Firm number SIC Total net Firm number SIC Total net
and name code assetsa and name code assets
($000) ($000)
E1: Westing- 3600 16,937,305 C1: Philips 3600 26,398,008
house Corp. N.V.
E2: Eaton Corp. 3714 3,033,800 C2: Dana 3714 4,786,379
Corp.
E3: Northern 3661 5,878,199 C3: Harris 3663 1,643,719
Telecom Corp.
E4: Northrop 3721 3,139,200 C4: Grumman 3721 2,565,984
Corp. Corp.
E5: Unisys 3570 11,534,602 C5: Digital 3570 10,111,500
Corp. Equipment
Corp.
E6: General 3711 164,063,000 C6: Ford 3711 143,366,000
Motors Corp. Motor Corp.
E7: United 3724 12,748,301 C7: Allied 3724 10,005,000
Technologies Signal Corp.
E8: Parker- 3728 1,741,802 C8: Sunstrand 3728 1,567,030
Hanni®n Corp. Corp.
E9: General 3721 6,118,098 C9: McDonnell 3721 11,885,000
Dynamics Corp. Douglas Corp.
E10: Honeywell 3882 5,089,098 C10: Johnson 3822 2,013,099
Controls
a
Mean dierence (the mean dierences were computed by subtracting the total net assets of the
control ®rm from the total net assets of the experimental ®rm; the ®gures for the total net assets
were obtained from the 1988 CRSP tapes and represent the end-of-year numbers for that year).
Mean dierence in the total net assets: $ 1,594,149, median dierence in the total net assets:
$ 2,909,550, Wilcoxon Signed Rank Statistic: T 34, p-value of the two-tailed Wilcoxon statistic:
0.5560.
after the announcement): 27 May, 31 May, 1 June, and 2 June 1988 (28 and 29
May were weekend days, and 30 May was Memorial Day). The estimation
period used in the study contains 271 daily returns and covers the period 2 May
1988±31 May 1989. 10
10
Our study utilizes a forward-looking estimation period with the event period placed at the
beginning of the estimation period. Such a design allows us to abstract from the chaotic eects of
the 19 October 1987 stock market crash when the Dow dropped 508 points (22.6%) in a single day.
This forward-looking approach is not new, and it has previously been utilized in the SUR-based
event study literature (Binder, 1985b, p. 178).
240 L.A. Gordon, K.J. Silvester/Journal of Accounting and Public Policy 18 (1999) 229±251
One of the key issues in evaluating the robustness of a study of this type is
validation of the actual dates of information release and ensuring that no early
pre-release of information took place. Therefore, we purposefully undertook
an extremely careful approach to these matters. First, we performed an ex-
haustive search of the existing literature (including The Wall Street Journal,
stock analysts reports, SEC reports, etc.) to ensure that once of the information
regarding ABC adoption or planned adoption was previously publicly released.
Second, we veri®ed that no confounding events occurred during the event
window (see footnote 7). Third, we veri®ed the actual date of the Business
Week release to the general public and obtained written con®rmation of the
release date from Business Week (see footnote 9). Fourth, we personally spoke
with representatives of ®rms active in ABC practice and in professionally
oriented ABC research and conferences to see if they knew of any previous
releases of the information. Our research indicated that the news released in
Port et al. (1988) article was, indeed, new to the market. Finally, we also have
made every attempt possible to statistically control for any possible event-date
confounding by de®ning the event period to consist of four separate and
consecutive trading dates.
3.7. Methodology
in Business Week. This severe form of calendar clustering lessens the proba-
bility that the residuals will be independent because of their contemporaneous
cross-sectional correlation. In addition, independence of the residuals is also
thought to be lessened by industry eects within the sample. That is, many of
the sample ®rms operate in similar industries and, therefore, it is reasonable to
assume that their calendar-based residuals might be related to one another.
Finally, it is also possible that ABC may have dierential impacts across ®rms
depending upon certain ®rm-speci®c factors. Therefore, the standard OLS
assumption of homoscedasticity may also be inappropriate.
In light of the foregoing and due to the methodological need for joint hy-
pothesis testing, a seemingly unrelated regressions (SUR) model will be used in
our study (Binder, 1985a, pp. 370±372; Binder, 1985b, pp. 170±171; Smirlock
and Kaufold, 1987, pp. 352±354). 11 The SUR model allows for contempora-
neous correlation of the error terms across ®rms, as well as non-constant
variance (heteroscedasticity) of the disturbance terms across ®rms. Utilizing
this approach, one can separate the basic market model into a system of
seemingly unrelated sets of ®rm regressions, which are (in reality) related via
contemporaneous correlation in their error terms.
The SUR model which is used to test the hypothesis underlying our study is
presented below.
R~1t a10 a11 DSt b10 R~mt b11 R~mt DSt
c11 D1 c12 D2 c13 D3 c14 D4 e~1t ;
~
R2t a20 a21 DSt b20 R~mt b21 R~mt DSt
c21 D1 c22 D2 c23 D3 c24 D4 e~2t ;
..
.
R~Nt aN 0 aN 1 DSt bN 0 R~mt bN 1 R~mt DSt
cN 1 D1 cN 2 D2 cN 3 D3 cN 4 D4 ~eNt ;
where R~it is the return for a security for ®rm i, in period t, net of the risk-free
rate, i 1; . . . ; N ; t 1; . . . ; T ; R~mt is the return for an equally weighted market
portfolio in period t, net of the risk-free rate; e~it is the random error term; DSt is
the announcement shift dummy variable; contains a 1 for every observation
11
Other SUR studies have utilized diering approaches to the control issue in calendar based
event studies. Binder (1985a,b), Rose (1985), and Schipper and Thompson (1983) used the existence
of multiple legislative announcements across all ®rms for their control mechanism. In a study of the
eect of the Mexican debt crises, Smirlock and Kaufold (1987) examined two separate and
dierently sized portfolios of exposed and non-exposed banks. To the best of our knowledge, our
study is the ®rst matched pairs analysis performed within a SUR methodology. As indicated
previously, our methodology is driven by the ®nite number of ABC adopting ®rms, the single
adoption announcement, and the need to perform joint hypothesis testing.
242 L.A. Gordon, K.J. Silvester/Journal of Accounting and Public Policy 18 (1999) 229±251
between the ®rst day of the event period and the last day in the estimation
period inclusive; zero otherwise; cia is the abnormal returns (event) coecient
for ®rm i on day a of the event period, a 1; . . . ; 4 corresponding to 5/27, 5/31,
6/1, 6/2; Da is the abnormal returns (event) dummy variable; contains a single
1 during day a of the event period, a 1; . . . ; 4 corresponding to 5/27, 5/31, 6/1,
6/2; zero otherwise; ai0 is the intercept for ®rm i; ai1 is the intercept shift coef-
®cient for ®rm i; bi0 is the slope coecient on the equally weighted market
portfolio return, Rmt , for ®rm i; bi1 is the slope shift coecient for ®rm i.
Daily ®rm returns were gathered from the University of Chicago Center for
Research in Security Prices (CRSP) tape. The equally weighted daily market
index from the CRSP tape is used for R~mt , as recommended by Brown and
Warner (1980, p. 239). Risk-free daily rates were obtained from the monthly
Federal Reserve Statistical Publication G.13 (1987±1988) for 3-month Trea-
sury Bills. In essence, the approach described above will test whether or not the
ABC announcement was associated with a statistically signi®cant stock market
reaction on any one (or more) of the four trading days in the announcement
period under consideration. Out estimation and testing of daily event coe-
cients follows the example of Smirlock and Kaufold (1987, pp. 352±354) in
their study of the Mexican debt crisis. This type of approach is appropriate
when the information release date can be exactly identi®ed; accordingly, it
allows for a more exact identi®cation and detailed analysis of the impact of
information on market returns than an analysis of cumulative abnormal re-
turns averaged over multiple event days.
In the case where ®rms show stock market return reactions (i.e., have cia
coecients on a given day a) of the opposite sign, a joint hypothesis test should
be more powerful than an average hypothesis test in detecting market reac-
tions. However, when all or most of the ®rms show reactions of the same sign
(on a given day a), an average hypothesis test may be more powerful than the
joint test. Accordingly, since it is not known ex ante whether the market would
impound the value of ABC adoptions to all ®rms in a similar manner, both
average and joint hypothesis testing are undertaken. Therefore, the main hy-
pothesis underlying the empirical study will be tested via the two operational
hypotheses noted below. In each case, the alternate hypothesis is the logical
alternative.
Null joint Hypothesis 1A. All event coecient
cia 0 for day a 27 and 31
May, 1 and 2 June; for all i, i 1; . . . ; N .
P
Null average Hypothesis 1B. The average event coecient
1=N i cia 0 for
day a, a 27 and 31 May, 1 and 2 June; i 1; . . . ; N .
L.A. Gordon, K.J. Silvester/Journal of Accounting and Public Policy 18 (1999) 229±251 243
The test statistic initially used in testing the joint and average hypotheses 1A
and 1B will be Theil's F statistic. 12 This test statistic allows for contempora-
neous correlation in the disturbance terms and has been used extensively in
prior calendar-based SUR event studies for testing of similar joint and average
hypotheses (Binder, 1985a, pp. 370±372; Binder, 1985b, pp. 171,172).
The results of estimating the four separate daily event parameters for each of
the control and experimental ®rms are presented in Table 2. 13 14 15 None of
12
Theil's (Theil, 1971, pp. 314,402) F statistic can be stated as follows.
PN
NT ÿ ^ ÿ1
IX ÿ1 C 0 ÿ1
c ÿ C B
^ 0
CX 0
R ^
i1 Ki
c ÿ C B
;
q
R ÿ X B^ 0
R^ ÿ1
I
R ÿ X B ^
where c ÿ C B^ the vector of linear constraintsPbeing tested, c the vector of dimension q 1,
C of full row rank and is of dimension q
i Ki xi,
P
B the estimated coefficient vector of dimension i Ki xi,
Ki the number of parameters estimated in equation i,
N the total number of equations in the regression system,
T the number of observations in time,
q the number of restrictions tested,
P
X the matrix of independent variables of dimension NT
i Ki ,
R the vector of dependent variables of dimension NT 1,
^ the sample covariance matrix of disturbances of dimension N N ,
R
I the identity matrix of dimension T T : Theil's statistic is exactly distributed, when the null
hypothesis is stated in terms of average market abnormal P returns (Binder, 1985b, p. 173). In
general, Theil's F is asymptotically distributed F
q; NT ÿ i Ki .
13
Although not separately reported, the intercept, intercept shift, slope, and slope shift
coecients were also analyzed for signi®cance at the a 0:10 level (two-tailed). For the
experimental ®rms, none of the intercepts, intercept shifts, or slope shift coecients were
signi®cant. For the control ®rms, none of the intercepts or intercept shift coecients
were signi®cant, but two of the slope shift coecients were signi®cant. Eight of ten slope
coecients were signi®cant for the experimental ®rms, and eight of ten slope coecients were
signi®cant for the control ®rms. These results on the signi®cance of the coecients resemble those
previously reported under a similar methodology (Smirlock and Kaufold, 1987, pp. 352±360)
14
The general non-signi®cance of the shift variables led us to re-estimate the models without the
intercept shifts and slope shifts. Doing so resulted in no material changes to any of the event
coecients, their signi®cance levels, or the hypothesis testing. However, it did cause swings in the
signi®cance levels of some of the basic slope and intercept estimates. Therefore, for control
purposes, the shift variables were retained in the regression system.
15
As is apparent from the SUR model structure, the independent variables are the same on the
right side of each regression equation. Under these circumstances, the parameter estimates and
standard deviations obtained under this system of equations are essentially identical to those
obtained via OLS. However, the use of SURs is still advantageous in that it allows for joint
hypothesis testing, which OLS would not allow.
244 L.A. Gordon, K.J. Silvester/Journal of Accounting and Public Policy 18 (1999) 229±251
Table 2
Estimated event parameters (cia )a
Firms 27/5/88 31/5/88 1/6/88 2/6/88
Experimental ®rm
E1: Westinghouse Corp. 0.0045 0.0241 ÿ0.0176 ÿ0.0082
E2: Eaton Corp. ÿ0.0020 ÿ0.0063 0.0003 0.0070
E3: Northern Telecom 0.0030 0.0205 ÿ0.0167 0.0047
E4: Northrop ÿ0.0064 0.0110 ÿ0.0191 ÿ0.0095
E5: Unisys 0.0001 0.0014 0.0125 ÿ0.0054
E6: General Motors ÿ0.0002 0.0169 0.0078 ÿ0.0056
E7: United Technologies ÿ0.0162 0.0226 ÿ0.0055 0.0021
E8: Parker-Hanni®n ÿ0.0149 0.0141 ÿ0.0152 0.0046
E9: General Dynamics 0.0012 0.0087 ÿ0.0285 ÿ0.0050
E10: Honeywell ÿ0.0024 0.0040 ÿ0.0033 0.0058
the estimated event parameters on 27 May are signi®cant for either the ex-
perimental or the control ®rms. Similarly, only 1 out of the 20 estimated event
parameters on 2 June is signi®cant (a 0:10, two-tailed). However, on the day
of the general announcement (31 May) and the day after the announcement (1
June), we see that a number of the estimated event parameters are signi®cant
for both the experimental and the control ®rms. Of particular note is the fact
that 9 of the 10 event coecients for the experimental ®rms are positive on the
day of the general announcement (31 May); this pattern is not repeated for the
control ®rms on 31 May. However, perhaps of more import is the fact that all
of the signi®cant coecients for both the control and the experimental ®rms
are positive on 31 May, and all of the signi®cant coecients for both the
control and the experimental ®rms are negative on 1 June. This sign and sig-
ni®cance pattern naturally raises the question of whether any real dierences
L.A. Gordon, K.J. Silvester/Journal of Accounting and Public Policy 18 (1999) 229±251 245
Table 3
Tests of joint and average Hypotheses 1A and 1Ba
27/5/88 31/5/88 1/6/88 2/6/88
Null hypothesis F value F value F value F value
Pr > F Pr > F Pr > F Pr > F
D.F.b D.F. D.F. D.F.
Experimental group
Hypothesis 1A 0:4390 1:2335 1:9021 0:3754
All cia 0 0:9280 0:2636 0:0403 0:9577
for day given 10 10 10 10
5260 5260 5260 5260
Hypothesis 1B 0:4317 5:1952 2:7550 0:0351
Average cia 0 0:5112 0:227 0:0970 0:8515
for day given 1 1 1 1
5260 5260 5260 5260
Control group
Hypothesis 1A 0:6856 3:0732 0:8176 0:6179
All cia 0 0:7389 0:0007 0:6117 0:7999
for day given 10 10 10 10
5260 5260 5260 5260
Hypothesis 1B 0:1148 3:8278 2:4902 0:4976
Average cia 0 0:7347 0:0505 0:1146 0:4806
for day given 1 1 1 1
5260 5260 5260 5260
a
Signi®cant at (two-tailed): *0.10 level; **0.05 level; ***0.01 level.
b
Degrees of freedom are presented with the numerator degrees of freedom followed vertically by
the denominator degrees of freedom.
exist between the experimental and control ®rms on the two days under dis-
cussion.
The results of testing the average and joint hypotheses on both the
experimental and the control groups are contained in Table 3. For
the experimental group, the Average Hypothesis 1B is rejected on 31 May at
the a 0:05 (two-tailed) and on 1 June at the a 0:10 (two-tailed); the
Joint Hypothesis 1A is rejected on 1 June at the a 0:05 (two-tailed). For
the control group, the Average Hypothesis 1B is rejected on 31 May
at the a 0:10 (two-tailed) and is not rejected on 1 June at the a 0:10
(two-tailed); the Joint Hypothesis 1A is rejected on 31 May at the a 0:01
(two-tailed). Therefore, both the experimental and control groups
experienced signi®cant abnormal returns on the day of and the day
following the Business Week announcement. In particular, both the
control and experimental groups experienced positive and signi®cant
(at a 0:10, two-tailed) average excess returns on the day of the
246 L.A. Gordon, K.J. Silvester/Journal of Accounting and Public Policy 18 (1999) 229±251
Table 4 recaps the average daily estimated event parameter for each group
and reports the dierence between the average estimated event parameters
(experimental average ± control average) for each day of the four day event
period. The table indicates that on the day before the Business Week an-
nouncement and the two days after the announcement, the control group av-
erage event coecient exceeds the experimental group average event coecient.
In other words, the control group appears to outperform the experimental
group (on average) on the day before and the two days Business Week an-
nouncement. However, on the day of the announcement (31 May), the ex-
perimental group average coecient value exceeds the control group average
coecient ± i.e., the experimental group appears to outperform the control
group. Although the signs are consistent with an instantaneous and positive
eect from the 31 May announcement, the dierences are not statistically
signi®cant at conventional levels for the 10 pairs. 18
In order to further investigate whether or not the abnormal excess returns for
the experimental ABC ®rms exceed the abnormal excess returns for non-ABC
matched control ®rms on a pair-by-pair rather than on an average basis, a non-
parametric one-tailed Wilcoxon Signed Rank Test was performed. The
16
In order to test for the potential and diering impact of information leakage eects versus
announcement eects, the model was also estimated with a dierent speci®cation of the event
periods. Instead of four separate daily parameters, two event periods were structured for the
Business Week announcement. The ®rst period was 2-day information leakage period that included
26 and 27 May (the two trading days before the 31 May announcement). The second period was a
3-day announcement period that included 31 May, 1 and 2 June. None of the F statistics associated
with the hypothesis testing regarding the information leakage event parameter or the announce-
ment period event parameter were signi®cant at the a 0:10 level (two-tailed) under this
speci®cation. This is not surprising given the opposite signs of the average reactions of the two
groups on 31 May and 1 June.
17
The potential for the market to completely impound the value of the ABC announcement on a
single trading day is strengthened by the fact that the information was physically on the market for
an additional non-trading day before the 31 May announcement date in our study.
18
Sign tests were not performed to assess the signi®cance of the trend because such tests are
reliant upon the assumption of statistical independence. This assumption is violated by the
contemporaneous correlation created by the calendar and industry clustering extant in the data set.
L.A. Gordon, K.J. Silvester/Journal of Accounting and Public Policy 18 (1999) 229±251 247
Table 4
Test of the dierences between the experimental and control group average estimated parametersa
Firm 27/5/88 31/5/88 1/6/88 2/6/88
Average experimental coecient ÿ0.0033 0.0117 ÿ0.0085 ÿ0.0010
Average control coecient ÿ0.0015 0.0096 ÿ0.0077 0.0034
Wilcoxon Signed Rank Test Statistics 19 and p-values for the dierences in the
estimated parameter values between the pairs of ®rms for each of the four days of
the event period are presented in Table 5. Although the p-value of the Wilcoxon
statistic on 31 May is substantially lower than on any of the other days of the
event period, the pairs comparison does not indicate a statistically signi®cant
dierence between the experimental and control groups on any of the 4 days. 20
Although our ®ndings indicate no signi®cant market eect for the group of
®rms as a whole, that does not preclude the possibility that individual ®rms
may have gained from the new ABC system. Indeed, given the recent ®ndings
by Shields (1995, p. 159), there is strong reason to believe that the bene®ts of
ABC are contingent upon various behavioral and organizational factors.
Further, anecdotal and case study evidence indicates that a great variety of
ABC systems are in existence and that ®rms are experiencing signi®cant
19
The Wilcoxon Signed Rank Test is superior in this situation to a general sign test, because the
Wilcoxon utilizes both the signs of the dierences between the pairs as well as size of the dierences.
A general sign test (such as the Binomial Test) utilizes only the sign of the dierences. While the
Wilcoxon is a distribution free test, it does assume that the true underlying distribution of
dierences is symmetric about its median (see, e.g., Van Matre and Gilbreath, 1983, pp. 514±518).
20
Sensitivity analysis was also performed on the sample by conducting the Wilcoxon Signed
Rank Test on all 9-pair subsets chosen from the 10 pairs. None of these combinations indicated
signi®cantly higher experimental excess returns (as compared to the control excess returns) on any
of the four event days.
248 L.A. Gordon, K.J. Silvester/Journal of Accounting and Public Policy 18 (1999) 229±251
Table 5
Wilcoxon Signed Rank Test for dierences in the event coecients between experimental and
control groups on a pair-by-pair basis
27/5 31/5 1/6 1l2/6
T value T value T value T value
p-value p-value p-value p-value
10 pairs 23 32 28 12
5. Concluding comments
References
Anderson, S.W., 1995. A framework for assessing cost management system changes: the case of
activity based costing implementation at General Motors, 1986±1993. Journal of Management
Accounting Research 7, 1±51.
L.A. Gordon, K.J. Silvester/Journal of Accounting and Public Policy 18 (1999) 229±251 249
Babad, Y.M., Balachandran, B.V., 1993. Cost driver optimization in activity-based costing. The
Accounting Review 68 (3), 563±575.
Ball, R., Brown, P., 1968. An empirical evaluation of accounting numbers. Journals of Accounting
Research 6 (2), 103±126.
Banker, R.D., Johnston, H.H., 1993. An empirical study of cost drivers in the US airline industry.
The Accounting Review 68 (3), 576±601.
Beaver, W.H., 1981. Market eciency. The Accounting Review 56 (1), 23±37.
Bhimani, A., Pigott, D., 1992. Implementing ABC: a case study of organizational behavioural
consequences. Management Accounting Research 3 (2), 119±130.
Binder, J.J., 1985a. On the use of the multivariate regression model in event studies. Journal of
Accounting Research 23 (1), 370±383.
Binder, J.J., 1985b. Measuring the eects of regulation with stock price data. Rand Journal of
Economics 16 (2), 167±183.
Bromwich, M., Bhimani, A., 1989. Management accounting: evolution not revolution. Chartered
Institute of Management Accountants, London.
Brown, S.J., Warner, J.B., 1980. Measuring security price performance. Journal of Financial
Economics 8 (3), 205±258.
Campi, J.P., 1992. It's not as easy as ABC. Journal of Cost Management 6 (2), 5±11.
Cooper, R., Turney, P.B.B., 1989. Hewlett-Packard: Roseville Networks Division. Harvard
Business School, Case 9-187-117.
Cooper, R., Kaplan, R.S., Maisel, L.S., Morrissey, E., Oehm, R.M., 1992a. From ABC to ABM.
Management Accounting 70 (10), 54±57.
Cooper, R., Kaplan, R.S., Maisel, L.S., Morrissey, E., Oehm, R.M., 1992b. Implementing Activity-
Based Cost Management: Moving from Analysis to Action. Institute of Management
Accountants, Montvale, NJ.
Cost Management Group of the Institute of Management Accountants, 1998. Cost Management
Update (81).
Dopuch, N., 1993. A perspective on cost drivers. The Accounting Review 68 (3), 615±620.
Dos Santos, L.B., Peers, K., Mauer, D.C., 1993. The impact of information technology
investment announcements on the market value of the ®rm. Information Systems Research 4
(1), 1±23.
Fama, E., Fisher, L., Jensen, M., Roll, R., 1969. The adjustment of stock prices to new
information. International Economic Review 10 (1), 1±21.
Federal Reserve Statistical Publication G.13, 1987±1988. Open market money dates and bond
prices ± selected interest dates. Monthly release dates of 1 September 1987±9 September
1988.
Foster, G., 1979. Brilo and the capital market. Journal of Accounting Research 17 (1), 262±274.
Foster, G., Gupta, M., 1990. Manufacturing overhead cost driver analysis. Journal of Accounting
and Economics 12 (1±3), 309±337.
Foster, G., Swenson, D., 1997. Measuring the success of activity based cost management and its
determinants. Journal of Management Accounting Research 9, 109±141.
Gonedes, N.J., Dopuch, N., 1979. Economic analyses and accounting techniques: perspective and
proposals. Journal of Accounting Research 17 (2), 384±410.
Gordon, L.A., 1998. Managerial Accounting: Concepts and Empirical Evidence, 3rd edu.
McGraw-Hill, New York.
Haka, S.F., Gordon, L.A., Pinches, G.E., 1985. Sophisticated capital budgeting selection
techniques and ®rm performance. The Accounting Review 60 (4), 651±669.
Harpaz, T., 1985. Financial reporters, the securities laws and the ®rst amendment: where to draw
the line. Fordham Law Review 8 (3), 1035±1062.
Innes, J., Mitchell, F., 1995. A survey of activity-based costing in the UK's largest companies.
Management Accounting Research 6 (2), 137±153.
250 L.A. Gordon, K.J. Silvester/Journal of Accounting and Public Policy 18 (1999) 229±251
Innes, J., Mitchell, F., 1990. Activity Based Costing: A Review with Case Studies. Chartered
Institute of Management Accountants, London.
Innes, J., Mitchell, F., 1991. Activity Based Costing: A Case Study of Development and
Implementation. Chartered Institute of Management Accountants, London.
Johnson, H.T., 1992. Relevance Regained. The Free Press, New York.
Johnson, H.T., Kaplan, R.S., 1987. Relevance Lost: The Rise and Fall of Management
Accounting. Harvard Business School Press, Boston, MA.
Kaplan, R.S., 1992. In defense of activity-based cost management. Management Accounting 74 (5),
58±63.
Kaplan, R.S., Shank, J.K., Horngren, C.T., Boer, G., Ferrara, W.L., Robinson, M.A., 1990.
Contribution margin analysis: no longer relevant/strategic cost management: the new
paradigm. Journal of Management Accounting Research 2, 1±32.
Kelly, K., 1991. A bean-counter's best friend. In: Business Week Special 1991 Bonus Issue: The
Quality Imperative, pp. 42, 43.
Larcker, D.F., 1983. The association between performance plan adoption and corporate capital
investment. Journal of Accounting and Economics 5 (1), 3±30.
Lukka, K., Granlund, M., 1996. Cost accounting in Finland: current practice and trends of
development. The European Accounting Review 5 (1), 1±28.
Malmi, T., 1997. Towards explaining activity-based costing failure: accounting and control in a
decentralized organization. Management Accounting Research 8 (4), 459±480.
Mitchell, F., 1994. A commentary on the applications of activity based costing. Management
Accounting Research 5 (3), 261±277.
Nair, R., 1979. Economic analyses and accounting techniques: an empirical study. Journal of
Accounting Research 17 (1), 225±242.
Noreen, E., 1991. Conditions under which activity-based cost systems provide relevant costs.
Journal of Management Accounting Research 3, 159±168.
Piper, J.A., Walley, P., 1990. Testing ABC logic. Management Accounting (British) 68 (8), 37, 42.
Piper, J.A., Walley, P., 1991. ABC relevance not found. Management Accounting (British) 69 (3),
42, 44, 54.
Port, O., King, R., Hampton, W.J., 1988. The productivity paradox. Business Week, 100±114.
Roberts, M.W., Silvester, K.J., 1996. Why ABC failed and how it may yet succeed. Journal of Cost
Management 10 (4), 23±35.
Rose, N.L., 1985. The incidence of regulatory rents in the motor carrier industry. Rand Journal of
Economics 16 (3), 299±318.
SAS Institute Inc., 1982. SAS/ETS User's Guide: 1982 Edition and SAS User's Guide: Statistics,
version 5 edition. SAS Institute, Cary, NC.
Sakurai, M., 1995. The past and future of Japanese management accounting. Journal of Cost
Management 9 (3), 21±30.
Schipper, K., Thompson, R., 1983. The impact of merger-related regulations using exact
distributions of test statistics. Journal of Accounting Research 23 (1), 408±415.
Shank, J.K., 1989. Strategic cost management: new wine, or just new bottles?. Journal of
Management Accounting Research 1, 47±65.
Shields, M.D., 1995. An empirical analysis of ®rms' implementation experiences with activity-based
costing. Journal of Management Accounting Research 7, 148±166.
Smirlock, M., Kaufold, H., 1987. Bank foreign lending, mandatory disclosure rules, and the
reaction of bank stock prices to the Mexican debt crisis. Journal of Business 60 (3), 347±364.
Spicer, B.H., 1990. New directions in management accounting practice and research. Management
Accounting Research 2, 139±146.
Theil, H., 1971. Principles of Econometrics. Wiley, New York.
Turney, P., Anderson, B., 1989. Accounting for continuous improvement. Sloan Management
Review 30 (2), 37±47.
L.A. Gordon, K.J. Silvester/Journal of Accounting and Public Policy 18 (1999) 229±251 251