APPLIED STOCHASTIC MODELS IN BUSINESS AND INDUSTRY

Appl. Stochastic Models Bus. Ind. 2000; 16:159}177
Cointegration analysis of brand and category sales: Stationarity
and long-run equilibrium in market shares
Shuba Srinivasan¹*¹¹ and Frank M. Bass`
A
¹The A. Gary Anderson School of Management, University of California, Riverside, CA 92521, U.S.A.
`The University of Texas System Eugene C. McDermott Professor of Management at the University of Texas at Dallas,
Richardson, TX 75083-0688, U.S.A.
SUMMARY
The present study uses modern time series methodology to understand long-run equilibrium in markets and
provides additional evidence of the frequent existence of stationary market shares for frequently purchased
consumer products. Dekimpe and Hanssens, Marketing Science 1995; 14(2):G109}121 using a database of
over 400 prior studies, found that 78 per cent of the market share series they studied were stationary, but that
68 per cent of the sales series were evolving. Our "ndings reconcile these results. A major contribution of this
paper is its demonstration that the prior empirical evidence that a majority of sales series is in evolution is
consistent with stationary market shares, if brand sales and category sales are cointegrated. To the extent
that competitive activities have an e!ect on market share, an implication of our "ndings is that these
activities may, in general, only have a temporary e!ect on market share. Finally, we distinguish, from
a strategic perspective, between sales and share response at the primary-demand level (category sales),
selective-demand level (brand sales) and relative-position level (market share) and identify strategic scenarios
depending upon their stable/evolving nature. Copyright 2000 John Wiley & Sons, Ltd.
KEY WORDS: econometric models; competition; stationarity; evolution; cointegration analysis
1. INTRODUCTION
1.1. Overview
Interest in the potential stationarity of market shares of brands arose initially because of the
development and "tting of zero-order stochastic brand choice models of the type proposed by
Ehrenberg [1] or of the equivalent model shown by Bass et al. [2]. These models assume that
choice probabilities of heterogeneous consumers are constant over purchase occasions. The
*Correspondence to: S. Srinivasan, The A. Gary Anderson School of Management, University of California, Riverside,
CA 92521, U.S.A.
¹Assistant Professor
¹E-mail: shuba.srinivasan@ucr.edu
A
E-mail: mzjb@utdallas.edu
Received 3 February 1999
Copyright 2000 John Wiley & Sons, Ltd. Revised April 2000
*We will return to a detailed description of the data shortly in Section 3.
averages of the choice probabilities over consumers at a given time are the market shares (share of
choices) for that time period. The assumed constancy of underlying choice probabilities would
imply unchanging market shares over time. The stochastic choice models have been shown to
provide reasonably good descriptions of aggregate purchase incidence data for brands as well as
of aggregate brand switching data. Inasmuch as market shares are known to be changing over
time because of promotions and other marketing activities, the question naturally arises as to
how "ts could be good when they are based on an assumption that implies the existence of
unchanging market shares.
Ehrenberg [3] examines markets from a stochastic brand choice framework and has concluded
that &...most markets are, however, more or less mature and approximately stationary*at least in
the medium term (that is, when measured in periods of a week or more and up to a year or two).
They may have some promotional ups and downs or the like superimposed'. This seems to
suggest the existence of unchanging individual choice probabilities that are temporarily pertur-
bed by marketing activity. In Figure 1(a)}1(c) we provide a graphical illustration of the phenom-
enon of stationarity in the market share of Heinz, over a 72-month period from 1960 to 1965, for
the period from 1986 to 1988, and over the period from 1992 to 1994, respectively.* Casual
inspection of these graphs suggests stationary share behaviour.
These market share patterns are important, particularly in light of "ndings in the literature.
Research has found that for frequently purchased consumer goods, market shares are predomi-
nantly stationary and sales series are largely evolving (e.g., References [3}6]). Notably, Dekimpe
and Hanssens [5], using a database of over 400 prior studies, found that 78 per cent of the market
share series they studied were stationary, but that 68 per cent of the sales series were evolving. Our
"ndings reconcile these results. A major contribution of this paper is its demonstration that the
prior empirical evidence that a majority of sales series is in evolution is consistent with stationary
market shares, if brand sales and category sales are cointegrated. We obtain this unique result as
an outcome of the study. To the extent that competitive activities have an e!ect on market share
[7], an implication of our "ndings is that these activities may, in general, only have a temporary
e!ect on market share. We shall discuss and utilize formal tests of stationarity and we shall
introduce and use cointegration analysis in an application of this methodology with respect to
long-run equilibrium in markets.
2. BACKGROUND
2.1. Emerging empirical generalizations concerning stationarity and evolution of
market share and sales
As pointed out in Section 1, some recent studies have examined the stationarity vs evolution in
market share and sales. Research has found that for frequently purchased consumer goods,
market shares are predominantly stationary (e.g., References [3}6]) and sales are mostly in
evolution. Therefore, marketing mix variables only appear to have a temporary e!ect on share,
while there is a potential for long-term e!ects on sales.
160 S. SRINIVASAN AND FRANK M. BASS
Copyright 2000 John Wiley & Sons, Ltd. Appl. Stochastic Models Bus. Ind. 2000; 16:159}177
Figure 1. (a) Market share series for Heinz Brand in ketchup category, 1960}1965; (b) Market share series
for Heinz Brand in ketchup category, 1986}1988; (c) Market share series for Heinz Brand in ketchup
category, 1992}1994.
Dekimpe et al. [8], used time-series analysis to examine the long-run e!ects on sales of price
promotions for ketchup, liquid detergent, soup and yogurt. Using weekly scanner data for
a period of 113 weeks, they estimated vector}autoregressive models (VAR) and vector}error
correction models (VECM). They estimated a system of equations where brand sales, price and
BRAND SALES AND CATEGORY SALES 161
Copyright 2000 John Wiley & Sons, Ltd. Appl. Stochastic Models Bus. Ind. 2000; 16:159}177
competitor's prices (all endogenous) are a function of lagged endogenous variables. Using
impulse}response analysis, they traced the over time impact of a one-standard error shock in
price. They concluded that price promotions have a signi"cantly di!erent impact on sales of
national brands versus private labels, but these e!ects are only temporary. One exception is the
soup market, where they observed a small long-term e!ect (a positive for private label and
a negative e!ect for two national brands).
Using a similar model, Dekimpe and Hanssens [9] examined the persistence e!ects on sales of
changes in marketing variables when sales series are evolving. The interested reader in referred to
Campbell and Mankiw [10], Evans [11] and Lutkepohl and Reimers [12] for other applications
of persistence analysis. Using persistence analysis. Dekimpe and Hanssens [13] measured the
long-run e!ects of advertising on sales, and concluded that advertising does have a long-run
impact on sales. They show that when sales are evolving it is possible for temporary variations in
marketing variables to have a permanent e!ect on sales (hysteresis). Therefore, it is possible when
competitive variables in#uence an evolving series, that temporary changes in a competitive
variable will have long-run (permanent) e!ects. On the other hand, in stationary markets,
Dekimpe et al. [8] have shown (when both market shares and competitive variables are
stationary) it is very unlikely that one will observe permanent o!setting competitive e!ects due to
temporary changes in marketing e!ort. This would suggest that these marketing expenditures
only have a temporary e!ect.
Recent studies have investigated the evolution of market shares. Bronnenberg et al. [14]
studied the ready-to-drink tea category and observed market shares are strongly in#uenced by
retailer distribution decisions through the early growth stage, but this e!ect diminishes over time.
The authors use a logically consistent model to quantify how a brand's coupled market share and
distribution evolve and to answer questions about the relative e$ciency of in#uencing share (pull)
vs seeking distribution (push). Franses et al. [15] utilize cointegration techniques to quantify the
long-run e!ects of marketing e!ort. However, they are primarily concerned with aberrant
observations that hamper the quality of data. In their illustrative example, they "nd di!erent
results for an outlier robust cointegration model as compared to non-robust methods.
Jedidi et al. [16] studied the long-run impact of advertising and promotions on brand choice
and purchase quantity. They propose a utility framework, di!erent from the models above,
utilizing individual level scanner panel data rather than aggregate weekly data. Parameters in the
utility function (intercept, price and promotion) are a function of long-termadvertising, long-term
promotions and loyalty where long-run advertising and promotions are speci"ed as a geometric
series of past advertising and promotional activities. They found a negative long-run e!ect on
brand equity, measured by the intrinsic brand preferences of the utility function, and increasing
price and price promotion sensitivity due to long-run promotions. Advertising only had a signi"-
cant positive e!ect on brand equity. They also conducted a simulation analysis to determine the
impact of changes in marketing strategies.
2.2. Modelling stationarity and long-run equilibrium using unit-root econometrics
Modern time-series methodology such as unit-root econometrics and cointegration analysis
are well suited to study the issue of stationarity and long-run equilibrium in markets. First, unit
root tests are used to test for the presence or absence of a unit root in the data. Once a long-run
component is identi"ed as being present in a series of interest, cointegration analysis is used to
identify whether trends in variables are related to each other in the long run. The idea of using
162 S. SRINIVASAN AND FRANK M. BASS
Copyright 2000 John Wiley & Sons, Ltd. Appl. Stochastic Models Bus. Ind. 2000; 16:159}177
*See Reference [23] for a detailed discussion of the econometric analysis of non-stationary data.
cointegration analysis in the study of non-stationary time series comes from the work of Engle
and Granger [17]. Cointegration analysis has been applied in a number of di!erent "elds, for
example, to study the e!ects of public policy on the long-run relationship between narcotics usage
and property crime [18], to measure the long-run relationship between advertising and sales
[19, 20], to forecast new automobile sales [21], to forecast freight rates [22], to determine the
long-run e!ect of marketing mix variables [9, 15]. For a detailed discussion of cointegration
analysis we refer the reader to Banerjee et al. [23], Enders [24] and Harris [25].
A long-run equilibrium exists between brand sales, S
R
, and category sales, C
R
, if the amount e
R
by which the actual observations S
R
deviate from the equilibrium level C
R
, given the observed
level C
R
, is a stationary time series. The equilibrium error e
R
given by +S
R
!C
R
, contains useful
information since the brand sales will move in a direction that satis"es the equilibrium.* In other
words, if +S

!C

, is low, then brand sales are low relative to category sales and we might
expect to see an increase in brand sales in future periods. On the other hand, if the term
+S

!C

, is high, we might expect a decrease in brand sales in future periods. An advantage
of cointegration analysis is that the extent of adjustment in a given period to deviations from the
long-run equilibrium is given in the estimated equation. Several di!erent testing procedures are
available to determine the existence of long-run cointegrating relationships such as the Engle and
Granger method and the multivariate extension, the Johansen method [26, 27]. Table I presents
a summary of the above-mentioned literature on long-run e!ects of marketing mix variables on
sales and share.
2.3. Objectives and contributions
The major objectives of this paper are: (1) to provide additional evidence of the frequent existence
of stationary market shares for frequently purchased consumer products; (2) to demonstrate that
the prior empirical evidence that a majority of sales series is in evolution is consistent with
stationary market shares, if brand sales and category sales are cointegrated; and (3) to distinguish,
from a strategic perspective, between market response at the primary-demand level (category
sales), selective-demand level (brand sales) and relative-positive level (market share) and identify
strategic scenarios depending upon their stable/evolving nature.
In dealing with the "rst objective, our analysis adds to the growing support for the empirical
generalization that for many markets, market shares are stationary (see for example, References
[5, 6]). Using data from 28 brands across eight product categories, we "nd that market shares
either for all brands or for major brands are stationary.
As observed earlier, previous research has found that shares tend to be stationary while sales
tend to be in evolution. What does this mean? This study contributes to an understanding and
reconciliation of this empirical observation, analytically and empirically. Given that brand and
category sales are evolving, we investigate whether these trends are related to each other. For
example, is an upward (downward) trend in brand sales related to an upward (downward) trend in
category sales? The trends would be related if there exists a long-run equilibrium or cointegrating
relationship between brand and category sales. We demonstrate that the presence of a cointegrat-
ing relationship between brand and category sales also implies that market shares are mean
stationary and hence is consistent with the above-mentioned empirical observation.
BRAND SALES AND CATEGORY SALES 163
Copyright 2000 John Wiley & Sons, Ltd. Appl. Stochastic Models Bus. Ind. 2000; 16:159}177
Table I. Studies looking at the long-run sales/share e!ects of marketing mix variables.
Study Data Model Findings
Dekimpe and
Hanssens [9]
1. Monthly pharmaceutical
data for 5 yr, 2. BRANDAID
data
VECM 1. Sales calls, advertising and
price di!erential have long-run
impact on sales;
2. Advertising promotion and
prices have no long-run
impact on sales
Dekimpe et al. [8] 113 Weeks scanner data
four-categories
VAR Impulse
response
analysis
Promotions only have
temporary e!ect in stationary
market. In non-stationary
markets, small lasting e!ect,
positive for private label,
and negative for two national
brands
Franses et al. [15] Weekly scanner data for
grocery product
Cointegration
analysis
Distribution, advertising and
promotions have long-run
impact on sales
Jedidi et al. [16] Scanner panel data, 691
households. 8 yr (1984}1992).
One product class with
four brands
Individual level
choice model,
and quantity
model
Advertising has long-run
positive e!ect on sales,
promotions have a negative
long-run e!ect
Bronnenberg et al.
[14]
Weekly grocery data
1991}1996 (257 weeks) of
ready-to-drink tea
VAR Distribution in#uences
long-run share through early
growth, and diminishes during
later stages of product life cycle
Dekimpe and
Hanssens [5]
Database of 400 prior
analyses
Unit root tests Sales series are mostly
evolving while shares are
stationary
Dekimpe and
Hanssens [13]
Monthly advertising and
sales data
VAR
Persistence
analysis
Advertising has a long-term
e!ect on sales
Lal and
Padmanabhan [6]
Yearly data from IRI
Factbook
Regression
analysis
Relative promotional
expenditures have no impact
on long-run shares
Baghestani [19] 54 Yearly advertising and
sales observations
VECM Advertising and sales are
cointegrated
We deal with the third objective, in a deviation from previous research by distinguishing, from
a strategic perspective, between market response at the primary-demand level (category sales),
selective-demand level (brand sales) and relative-position level (market share) and identifying
potential scenarios depending upon their stable/evolving nature. In doing so, we extend the work
of Dekimpe et al. [8], who focus only on brand and category sales. From a strategic perspective, it
is useful to make a distinction between sales and market share. To illustrate, consider two
scenarios, one where market shares are stationary in a category that has stationary sales and the
other where market shares are stationary in a category that has evolving sales. Under the "rst
scenario, "rms are unable to improve the long-run relative and absolute positions, whereas under
the second scenario the long-run absolute positions may improve even though "rms are unable to
164 S. SRINIVASAN AND FRANK M. BASS
Copyright 2000 John Wiley & Sons, Ltd. Appl. Stochastic Models Bus. Ind. 2000; 16:159}177
*Of these eight scenarios, the following three cannot occur for logical reasons: stationary market shares and evolving
brand sales of all brands in a stationary category, stationary market shares and stationary brand sales of all brands in an
evolving category, evolving market shares and stationary brand sales of all brands in a stationary category.
¹Another scenario is evolving market shares and evolving brand sales in an evolving category. While this scenario is
potentially interesting, we do not consider it in the paper since it is not present in the data used and leave further analysis
to future research.
improve their relative positions. From a managerial perspective, this distinction is important, for
instance, because the long-term pro"tability implications of the two scenarios could be quite
di!erent. Further, category sales may be driven by exogenous factors but market shares are not
and they would depend on the relative marketing activities of the "rms. We illustrate the strategic
scenarios using data from eight consumer product categories (ketchup, soup, toilet tissue, sugar,
peanut butter, dry detergent and tuna) for which scanner data are available and provide the
strategic implications of each of these scenarios. These implications underscore the importance of
distinguishing between market response at the primary-demand level (category sales), selective-
demand level (brand sales) and relative-position level (market share), which has not been done in
prior research.
The structure of the remainder of the paper is as follows. In Section 3, we provide an overview
of the strategic scenarios with respect to market shares, brand and category sales and discuss the
estimation approach. In Section 4, we describe the scanner data used and provide the empirical
results. Section 5 presents conclusions and o!ers directions for future research.
3. LONG-TERM BEHAVIOUR OF MARKET SHARES, BRAND
AND CATEGORY SALES
3.1. Testing univariate equilibrium share models
When a series may be appropriately modelled as depending on a constant plus a coe$cient times
a lag of the series plus a random term, testing whether a series is stationary or evolving may
be accomplished by means of the well-known test proposed by Dickey and Fuller [28]. When
more than one lag is involved the appropriate test is the augmented Dickey}Fuller test (ADF)
[23]. We use these tests in examining the stationarity of market shares, sales and category sales.
3.2. Overview of scenarios
Stationary vs evolving conditions in market shares, brand and category sales give rise to eight
possible scenarios.* We consider the following scenarios: stationary market shares and stationary
brand sales for all brands in a stationary category (SSS), stationary market shares and stationary
brand sales for a sub-set of major brands in a stationary category (SSS), stationary market shares
and evolving sales for all brands in an evolving category (SEE), stationary market shares and
evolving sales for a sub-set of major brands in an evolving category (SEE), and, evolving market
shares and evolving brand sales in a stationary category (EES).¹ We discuss these in turn:
(i) Stationary market shares and stationary brand sales for all brands in a stationary category
(SSS). This refers to the case where all market shares are stationary as are brand and category
sales. What does this mean from a strategic perspective? Stationary market shares imply that all
BRAND SALES AND CATEGORY SALES 165
Copyright 2000 John Wiley & Sons, Ltd. Appl. Stochastic Models Bus. Ind. 2000; 16:159}177
gains and losses are short-lived. To the extent that competitive activities have an e!ect on market
share [7], an implication of our "ndings is that these activities may, in general, only have
a temporary e!ect on market share.
(ii) Stationary market shares and stationary brand sales for a sub-set of brands in a stationary
category (SSS). Under this scenario, while a sub-set of major brands has stationary market shares,
to the extent that marketing activity has an e!ect on shares, this scenario also implies that at least
two brands are engaged in a share war with one brand gaining at the expense of another. This
suggests that while marketing activity may cause temporary perturbations in market shares for
the sub-set of brands with stationary shares, this may not be case for the sub-set with evolving
shares, consistent with Srinivasan et al. [29].
(iii) Stationary market shares and evolving brand sales in an evolving category (SEE). This refers
to a scenario where all brands have stable market share in an expanding or declining market.
From a strategic perspective, this kind of behaviour can arise due to two conditions. First, the
category expansion may result from macro-economic or demographic factors. Second, marketing
e!orts of "rms may drive the observed evolution in category sales. In the former case, it is
conceivable that brands do not change their marketing mix but merely ride the long-run demand
waves in an industry. But what is the meaning of stationary market shares and evolving sales? If
brand sales and category sales are cointegrated this also implies that the market is in long-run
equilibrium. Cointegration analysis is useful in "nding stationary linear combinations of category
and brand sales. If such regressions exist, then this is consistent with market shares being
stationary or stable. From a strategic perspective, this implies that "rms are unable to improve
their relative position in spite of improving their absolute long-run performance with respect to
sales.
(iv) Stationary market shares and evolving brand sales for a sub-set of major brands in an evolving
category (SEE). This scenario has marketing strategy implications that are similar to those for
scenario (iii) for the sub-set of brands that has stationary market shares. This scenario also implies
that at least two brands have evolving market shares, and to the extent that market shares are
in#uenced by marketing variables, the two "rms may be engaged in a long-run market share
battle.
(v) Evolving market shares and evolving brand sales in a stationary category (EES). Under this
scenario, given that the category sales are stationary, to the extent that marketing activity has an
e!ect on shares, at least two brands are engaged in a market share battle with one brand gaining
at the expense of another in the long run. This is consistent with Ehrenberg [30], &. . . in near-
stationary markets, everybody has to run hard to stand still.' A strategic implication is that
competitive marketing activity may have a long-term impact on market shares for the evolving
sub-set.
3.3. Assessment of market equilibrium using cointegration analysis
If the sales series and category sales series are evolving, we investigate whether trends in variables
are related to each other. The trends are related to each other if there exists a long-run
equilibrium or cointegrating relationship between the variables. The concept of cointegration is
a powerful one since it describes the existence of an equilibrium or stationary relationship among
two or more time series, each of which is individually non-stationary [23]. In other words, an
equilibrium relationship would imply that, even if they diverge from each other in the short run,
such deviations are stochastically bounded or diminishing over time. Marketing variables may
166 S. SRINIVASAN AND FRANK M. BASS
Copyright 2000 John Wiley & Sons, Ltd. Appl. Stochastic Models Bus. Ind. 2000; 16:159}177
*We are grateful to Clive W. Granger for pointing this out to us and motivating the following discussion.
¹For the three- and four-brand case, one can show that a similar condition requiring the market shares across brands to
sum to 1 is obtained.
¹Although y
G R
"e
G R
/C
R
will have E( y
G R
)"0 and E( y
G R
, y
G RI
)"0 for all kO0, E(y`
G R
) is not time constant. Thus S
R
/C
R
does not have time-constant second moments but is &mean-stationary'. We are thankful to Clive W. Granger and an
anonymous referee for pointing this out to us.
show trending patterns in the typical two-year span of available scanner data. Cointegration
models may therefore prove useful, as such models can allow for the presence of stochastic
and deterministic trends in the data [15]. If a long-run equilibrium relationship holds
between brand sales S
R
and category sales C
R
, then they are related by an equilibrium relationship
such as:
S
R
"C
R
#e
R
(1)
The existence of a long-term relationship implies that equilibrium error e
R
represents a stationary
process. From an empirical perspective, under scenario (iii), brand sales are cointegrated with
category sales and market shares are stationary. We demonstrate this scenario analytically using
a two-brand case.* Let S
¹
and S
`
be the sales of the two brands that are evolving or I(1), and their
sum, the category sales, is also I(1). If S
¹
is cointegrated with C, then there exists a constant
¹
such that
S
¹ R
"
¹
C
R
#e
¹ R
(2)
Similarly, if S
`
is cointegrated with C, then there exists a constant
`
such that
S
` R
"
`
C
R
#e
`R
(3)
As before, the existence of a long-termrelationship implies that the equilibrium errors e
¹ R
and e
` R
represent stationary processes. Equations (2) and (3) can be written as
(1!
¹
)S
¹ R
!
¹
S
` R
"e
¹ R
(4)
!
`
S
¹ R
#(1!
`
) S
` R
"e
` R
(5)
For Equations (4) and (5) to hold, it follows that e
¹ R
"!e
` R
and
¹
#
`
"1. The latter
condition is logical consistency constraint that the market shares across brands sum to 1. This is
consistent with the interpretation of
¹
and
`
as the long-run market shares of the brands.¹
Equations (2) and (3) can be written as
S
¹ R
/C
R
"
¹
#e
¹ R
/C
R
(6)
S
` R
/C
R
"
`
#e
` R
/C
R
(7)
Hence if e
G R
(where i"1, 2) is iid, zero mean and independent of C
R
, then y
G R
"e
G R
/C
R
will have
E(y
G R
, y
G RI
)"0 for all kO0 and so will look like white noise in terms of its mean properties.¹
This implies that the market shares de"ned by S
¹ R
/C
R
and S
` R
/C
R
are mean-stationary and have
BRAND SALES AND CATEGORY SALES 167
Copyright 2000 John Wiley & Sons, Ltd. Appl. Stochastic Models Bus. Ind. 2000; 16:159}177
*Though the likelihood-ratio test and AIC/SIC information criterion approach are equivalent, the real di!erence is in the
signi"cance levels, which are usually "xed at 5 per cent in hypothesis testing, and are selected automatically (typically
starting at more than 10 per cent in small samples but converging to 0 for larger ones) in the information criterion
approach. Loose signi"cance levels are not necessarily the best choice in cointegration analysis and the discussion on this
issue cannot be regarded as closed with a clear guideline. Model choice by residual correlation statistics (such as
Ljung}Box Q) has often been reported to yield the poorest results.
long-run equilibrium values given by
¹
and
`
. Thus we demonstrate, analytically and
empirically (in Section 4), that the prior empirical evidence that a majority of sales series is in
evolution is consistent with stationary market shares, if brand sales and category sales are
cointegrated.
Several di!erent testing procedures are available to determine the existence of such long-run
cointegrating relationships such as the Engle and Granger method [17] and the multivariate
extension, the Johansen method [26, 27]. While the former uses an ordinary least-squares
approach, the latter uses a maximum-likelihood method that tests for the number of cointegrat-
ing regressions. In this paper we will apply the Johansen maximum-likelihood estimator, which is
the most widely used approach. The interested reader is referred to Maddala and Kim [31, pp.
155}197], who present an excellent review on estimation of cointegrated systems.
3.4. Vector error correction models (VECM)
Engle and Granger [17] show that the presence of a cointegrating relationship implies that the
data are generated according to a partial adjustment or error-correction mechanism. The
equilibrium error ensures that, after short-run deviations from the equilibrium, the system will
return to its long-run equilibrium. In other words, the short-run dynamics are in#uenced by the
deviation from the long-run relationship. These short-run dynamics are captured in the VECM.
The error-correction model is obtained by adding the terms containing the lagged residuals as
shown below:
S
¹ R
S
` R
C
R
"
a
¹"
a
`"
a
`"
#


0 0
0
1`
0
0 0
!
e
1¹ R¹
e
1` R¹
e
! R¹
#
,

G¯¹
b
¹¹
b
¹`
b
¹`
b

b
``
b
``
b

b
``
b
``
S
¹ RG
S
` RG
C
RG
#
u
¹ R
u
` R
u
` R
(8)
where N is the order to the model determined using the log-likelihood criterion.* The
error correction term is obtained from the cointegrating regression. The coe$cients

,
1`
and
!
measure the speed of adjustment of the dependent variables towards the long-run
equilibrium.
To summarize, our framework consists of three major steps. First, we assess the presence of
evolution versus stability in sales (or market share) by observing the behaviour of the series over
time. The absence of a unit root indicates mean-reverting behaviour or stationarity. The presence
of a unit root indicates that there is an evolutionary component in sales or market share. Next, if
the data indicate the presence of unit roots, we investigate whether these non-stationary compo-
nents are related to each other. Finally, if cointegration has been established, we test for the
existence of a partial adjustment equilibrium or error-correcting behaviour. Further, we illustrate
the strategic scenarios described in Section 3.2 using data from the eight consumer product
categories.
168 S. SRINIVASAN AND FRANK M. BASS
Copyright 2000 John Wiley & Sons, Ltd. Appl. Stochastic Models Bus. Ind. 2000; 16:159}177
*The use of di!erent samples for the two databases on ketchup causes us to classify the two into di!erent scenarios and we
shall return to this issue shortly.
¹Retailers typically assign each product a SKU (similar to the universal product code (UPC)) and use it as a unique
identi"er for their products to help track sales, etc.
¹We start by including 12 lags, determine the greatest lag with a signi"cant t-value and re-test using this number of lags.
A constant is included and we test for trends and seasonality in the data.
A
Seasonality can cause non-stationarity and persistent e!ects (see for example, Reference [35]). Controlling for seasonal-
ity, we "nd that the series is stationary.
4. DATA DESCRIPTION AND EMPIRICAL RESULTS
A.C. Nielsen household scanner panel data from two test markets in the U.S. on the purchases of
ketchup, peanut butter, tuna, toilet tissue, dry detergent, soup and sugar were used to construct
the time series of weekly shares, brand and category sales. In addition, we used another database
for the ketchup category, that of store movement data supplied by A.C. Nielsen for the period
from 1992 to 1994. The latter has a sample of 690 stores across ten regions in the US.* Of these
categories, two databases, the ketchup scanner panel data and the soup scanner panel data, have
also been examined by Dekimpe et al. [8]. We compare our "ndings with theirs later in this
section.
Our database aggregates across varieties to the brand level, consistent with our focus on the
maintenance of market share equilibrium at the brand level rather than at the Stock Keeping
Unit (SKU) level.¹ However, this causes some loss of information due to possible underlying
heterogeniety among varieties. Thus, the competitive interaction e!ects may be subjected to
aggregation bias [32, 33]. To control for this bias, we performed pooling tests to determine
whether we can pool the di!erent varieties for a brand. Over 90 per cent of the varieties could be
pooled consistent with Link [34], who recommends using data within homogeneous sub-sets to
overcome the bias issue. Market share of a brand is calculated as the share of the pooled unit sales
of the brand. In all categories, we consider the major brands and aggregated all brands other than
the major brands into one "gure referred to as Rest in all the categories. This resulted in a total of
28 brands in the eight databases that we analyse. Each time series consists of 104 weekly
observations. We provide a brief description of the brands considered in each category in
Table II. We discuss the results for each one of the "ve scenarios in turn.
(i) Stationary market shares and stationary brand sales in a stationary category (SSS). We "nd
this case holds for two categories*ketchup and soup. We describe the results for each of these
categories in turn. We conducted the unit root test; the results in Table II showthat market shares
of all brands are stationary while the brand sales of all brands as well as category sales are
stationary.¹ Therefore, the ketchup market is consistent with stationarity or with the existence of
a long-run equilibrium.
Next, we examine the soup category. We control for seasonal e!ects in sales by using the
seasonal equivalent of Equation (5) for the ADF test (see Reference [5]).
A
Our results summarized
in Table II show that market shares of Campbell brand and Rest are stationary as are brand sales
and category sales. Thus the soup category is consistent with stationarity. Therefore, we "nd that
in the long run the relative position of the players is una!ected by the marketing mix activity and
all gains and losses are temporary, consistent with Lal and Padmanabhan [6].
We should point out that these ketchup category results are consistent with those obtained by
Dekimpe et al. [8] using the same data source. We extend their analysis by examining market
BRAND SALES AND CATEGORY SALES 169
Copyright 2000 John Wiley & Sons, Ltd. Appl. Stochastic Models Bus. Ind. 2000; 16:159}177
T
a
b
l
e
I
I
.
C
l
a
s
s
i
"
c
a
t
i
o
n
o
f
S
c
e
n
a
r
i
o
s
º
.
S
c
e
n
a
r
i
o
s
M
a
r
k
e
t
B
r
a
n
d
C
a
t
e
g
o
r
y
M
a
r
k
e
t
s
h
a
r
e
E
m
p
i
r
i
c
a
l
o
b
s
e
r
v
a
t
i
o
n
:
S
t
r
a
t
e
g
i
c
s
h
a
r
e
s
a
l
e
s
s
a
l
e
s
A
D
F
r
a
n
g
e
a
n
d
c
a
t
e
g
o
r
y
,
(
y
e
a
r
)
,
b
r
a
n
d
s
i
m
p
l
i
c
a
t
i
o
n
s
s
i
g
n
i
"
c
a
n
c
e
l
e
v
e
l
I
:
S
S
S
S
t
a
t
i
o
n
a
r
y
(
a
l
l
)
S
t
a
t
i
o
n
a
r
y
(
a
l
l
)
S
t
a
t
i
o
n
a
r
y
(
a
l
l
)
!
3
.
5
6
t
o
!
7
.
1
0
'
!
3
.
0
3
*
1
.
K
e
t
c
h
u
p
(
1
9
8
6
}
1
9
8
8
)
H
e
i
n
z
,
H
u
n
t
s
,
D
e
l
M
o
n
t
e
,
R
e
s
t
2
.
S
o
u
p
(
1
9
8
6
}
1
9
8
8
)
C
a
m
p
b
e
l
l
a
n
d
R
e
s
t
A
l
l
g
a
i
n
s
a
n
d
l
o
s
s
e
s
a
r
e
t
e
m
p
o
r
a
r
y
I
I
:
S
S
S
S
t
a
t
i
o
n
a
r
y
(
m
a
j
o
r
b
r
a
n
d
s
u
b
-
s
e
t
)
S
t
a
t
i
o
n
a
r
y
(
m
a
j
o
r
b
r
a
n
d
s
u
b
-
s
e
t
)
S
t
a
t
i
o
n
a
r
y
!
4
.
1
9
t
o
!
6
.
6
7
'
3
.
¹
o
i
l
e
t
t
i
s
s
u
e
(
1
9
8
5
}
1
9
8
7
)
N
o
r
t
h
e
r
n
a
n
d
R
e
s
t
A
l
l
g
a
i
n
s
a
n
d
l
o
s
s
e
s
a
r
e
t
e
m
p
o
r
a
r
y
f
o
r
t
h
e
m
a
j
o
r
b
r
a
n
d
s
u
b
-
s
e
t
I
I
I
:
S
S
E
S
t
a
t
i
o
n
a
r
y
(
a
l
l
)
E
v
o
l
v
i
n
g
(
a
l
l
)
E
v
o
l
v
i
n
g
!
4
.
4
0
t
o
!
7
.
3
6
'
!
3
.
0
2
°
!
5
.
6
5
t
o
!
7
.
1
0
'
4
.
K
e
t
c
h
u
p
(
1
9
9
2
}
1
9
9
4
)
H
e
i
n
z
,
H
u
n
t
s
,
D
M
,
R
e
s
t
5
.
S
u
g
a
r
(
1
9
8
5
}
1
9
8
7
)
C
&
H
a
n
d
R
e
s
t
6
.
P
e
a
n
u
t
b
u
t
t
e
r
(
1
9
8
5
}
1
9
8
7
)
J
i
!
,
P
e
t
e
r
P
a
n
,
S
k
i
p
p
y
a
n
d
R
e
s
t
L
o
n
g
-
r
u
n
a
b
s
o
l
u
t
e
p
e
r
f
o
r
-
m
a
n
c
e
i
s
e
v
o
l
v
i
n
g
b
u
t
"
r
m
s
a
r
e
u
n
a
b
l
e
t
o
i
m
-
p
r
o
v
e
r
e
l
a
t
i
v
e
p
o
s
i
t
i
o
n
I
V
:
S
E
E
S
t
a
t
i
o
n
a
r
y
(
m
a
j
o
r
b
r
a
n
d
s
u
b
-
s
e
t
)
E
v
o
l
v
i
n
g
(
m
a
j
o
r
b
r
a
n
d
s
u
b
-
s
e
t
)
E
v
o
l
v
i
n
g
!
3
.
3
2
t
o
!
5
.
6
7
'
7
.
D
r
y
d
e
t
e
r
g
e
n
t
(
1
9
8
6
}
1
9
8
8
)
C
h
e
e
r
a
n
d
O
x
y
d
o
l
S
a
m
e
a
s
c
a
s
e
(
i
i
i
)
f
o
r
t
h
e
s
t
a
t
i
o
n
a
r
y
s
u
b
-
s
e
t
V
:
E
E
S
E
v
o
l
v
i
n
g
(
a
l
l
)
E
v
o
l
v
i
n
g
(
a
l
l
)
S
t
a
t
i
o
n
a
r
y
N
o
n
e
a
r
e
s
i
g
n
i
"
c
a
n
t
8
.
¹
u
n
a
(
1
9
8
5
}
1
9
8
7
)
C
h
i
c
k
e
n
o
f
S
e
a
,
S
t
a
r
K
i
s
t
,
R
e
s
t
M
a
r
k
e
t
i
n
g
a
c
t
i
v
i
t
y
m
a
y
i
m
p
r
o
v
e
r
e
l
a
t
i
v
e
a
s
w
e
l
l
a
s
a
b
s
o
l
u
t
e
p
o
s
i
t
i
o
n
o
f
s
o
m
e
o
f
t
h
e
b
r
a
n
d
s
º
A
.
C
.
N
i
e
l
s
e
n
:
S
i
n
g
l
e
-
s
o
u
r
c
e
s
c
a
n
n
e
r
d
a
t
a
(
1
9
8
6
}
1
9
8
8
)
,
E
R
I
M
s
c
a
n
n
e
r
d
a
t
a
(
1
9
8
5
}
1
9
8
7
)
a
n
d
s
t
o
r
e
m
o
v
e
m
e
n
t
d
a
t
a
(
1
9
9
2
}
1
9
9
4
)
.
'
d
e
n
o
t
e
s
s
i
g
n
i
"
c
a
n
c
e
a
t
t
h
e
1
p
e
r
c
e
n
t
l
e
v
e
l
.
*
S
i
g
n
i
"
c
a
n
c
e
a
t
t
h
e
5
p
e
r
c
e
n
t
l
e
v
e
l
.
170 S. SRINIVASAN AND FRANK M. BASS
Copyright 2000 John Wiley & Sons, Ltd. Appl. Stochastic Models Bus. Ind. 2000; 16:159}177
Table III. Johansen's FIML test of cointegrating rank.
Likelihood 5 Per cent
''º°°
'
°''' ""`
Hypothesized
ratio critical value No. of CE(s)

''º°°

°''' ""`
87.34 39.89 Yes Noneº
47.51 24.31 Yes At most 1º
21.67 12.53 Yes At most 2º
0.91 3.84 No At most 3
ºdenotes rejection of the hypothesis at 5 per cent signi"cance level.
shares, in addition to brand and category sales. In a divergence from their results, we "nd that the
soup market is stationary, whereas they found evidence of evolution in sales. One possible
explanation is that we control for seasonality. As they point out, more work is needed to enhance
the power of unit root tests as it is sometimes di$cult to classify a series as being stationary or
evolving with a low persistence level.
(ii) Stationary market shares and stationary brand sales for a sub-set of major brands in
a stationary category (SSS). We "nd this to be valid in one category*toilet tissue. The unit root
test results summarized in Table II show that the market shares of Northern and Rest are
stationary, while the market shares of Charmin and Cottonelle are evolving. Further, the sales of
Northern and Rest brands are stationary, while those of Charmin and Cottonelle are evolving.
Category sales are stationary. Therefore, a sub-market consisting of two major brands has
stationary market shares. From a strategic perspective, to the extent that marketing activity has
an e!ect on shares, this scenario implies that at least two brands may be engaged in a share war
with one brand gaining at the expense of another. These brands are Charmin and Cottonelle in
the toilet tissue category. Thus, while marketing activity causes temporary perturbations in
market shares for the sub-set of brands with stationary shares, this may not be the case for the
sub-set with evolving shares [29].
(iii) Stationary market shares and evolving brand sales in an evolving category (SEE). This case
refers to a situation where market shares are stationary, or in long-run equilibrium, even though
brand and category sales are evolving. We "nd this condition to hold in three catego-
ries*ketchup, sugar and peanut butter. Unit root test results, which are summarized in Table II,
show that market shares of all brands are stationary. The sales of the brands*Heinz, Hunts, Del
Monte and Rest*are evolving. Next, we examine the category sales. Our test results show that
category sales are also evolving. In this case, we proceed to examine whether there is a cointegrat-
ing relationship between each of the brand and category sales.
Cointegration analysis is performed on four variables: sales of Heinz, Hunts, Del Monte and
category sales. First, we need to determine the number of lags to include in the cointegration
relation. We use the Johansen and Juselius estimation method [27]. The PCFiml software is used
for estimation purposes [36]. This method determines the cointegration rank, the number of
signi"cant cointegrating vectors and the associated vectors with the long-run parameters. Several
di!erent test procedures exist to determine the cointegration rank, with the most widely used
being the trace test and the eigenvalue test (see References [26, 27]).
The Johansen FIML test indicated the presence of three cointegrating relationships among the
variables. The results of this analysis are provided in Table III.
BRAND SALES AND CATEGORY SALES 171
Copyright 2000 John Wiley & Sons, Ltd. Appl. Stochastic Models Bus. Ind. 2000; 16:159}177
Table IV. Long-run equilibrium relationship between brand sales and category
salesº.
Variable S
'°'"´
S
'""''
S
'°' `º"'°
C 0.607' 0.126' 0.056'
(0.005) (0.0028) (0.0014)
ADF statistic !7.15 !4.88 !6.59
Critical value !2.76 !2.76 !2.76
Cointegration Yes Yes Yes
ºThe standard error of the estimate is included in parenthesis. The critical value (at 5 per cent
level) for unit root tests obtained from Philips and Oularis [38].
'Signi"cant at p(0.01.
We "nd that there are three signi"cant cointegrating vectors. The "rst cointegrating vector
shows the long-run relationship between Heinz sales and category sales, the second one the
long-run relationship between Hunts sales and category sales and the third the relationship
between Del Monte sales and category sales. The results reported are the so-called restricted
long-run relationships. Following Doornik and Hendry [37], we "rst estimated the unrestricted
relationship and next tested for restriction using log-likelihood ratio tests. Based on this analysis,
we concluded that sales of Heinz is only signi"cant in the "rst long-run relationship and was
constrained to zero in the other long-run relationships. Similarly, we concluded that sales of
Hunts (Del Monte) is only signi"cant in the second (third) long-run relationship and was
constrained to zero in the other long-run relationships. The cointegrating vectors are reported in
Table IV.
As described in the previous section, the concept of cointegration allows us to describe the
existence of an equilibrium relationship between two series, each of which is non-stationary. The
intuition is that non-stationary movements in category sales remove non-stationary movements
in brand sales, so that only transitory components are left in the residuals.
We test for the condition that e
G R
, the residuals from each of the cointegrating regressions, as in
Equations (2) and (3), are iid, zero mean and independent of C
R
. As pointed out in Section 3.2, this
will then imply that y
G R
"e
G R
/C
R
will have E( y
R
, y
RI
)"0 for all kO0 and so will look like white
noise in terms of its mean properties. Johansen's FIML test has provided the cointegrating
vectors and has established the stationarity of the residuals of the cointegrating regressions.
Nevertheless, we run one more test proposed by Philips and Oularis [37] to con"rm that the
residuals from the cointegrating regression are white noise (in Table IV). Further, we test for
independence between e
G R
and C
R
. In all three cases, we "nd that the residuals are white-noise
processes and independent of C
R
. Therefore, the interpretation of the coe$cients is that the
long-run market share of Heinz is 60.7 percent, the long-run market share of Hunts is 12.6
percent and the long-run market share of Del Monte is 5.5 percent; all are signi"cant at the 1
percent level, and the long-run market shares are mean-stationary. While each of the brand and
category sales may move up or down, a long-run equilibrium exists, towards which the market
adjusts.
In Johansen's method, the cointegrating regression is estimated jointly with the error correc-
tion model. As pointed out earlier, the idea is that a portion of the disequilibriumfrom one period
to the next is corrected in the next period. The output from the estimation of the error-correction
model is shown in Table V.
172 S. SRINIVASAN AND FRANK M. BASS
Copyright 2000 John Wiley & Sons, Ltd. Appl. Stochastic Models Bus. Ind. 2000; 16:159}177
Table V. Results of the VECM for the ketchup categoryº.
Variable S
'°'"´
S
'""''
S
'°' `º"'°
C (!1) !0.646'
(0.208)
S
'°'"´
(!1) 0.584° !0.013°
(0.264) (0.007)
S
'""''
(!1)
S
'°' `º"'°
(!1) 0.138 !0.199'
(0.124) (0.085)
ES
'°'"´
(!1) !1.257' * *
(0.283)
ES
'""''
(!1) * !0.188' *
(0.070)
ES
'°' `º"'°
(!1) * * !0.395'
(0.097)
ºThe standard error of the estimate is included in parenthesis.
'Signi"cant at p(0.01, °Signi"cant at p(0.05 and °Signi"cant at p(0.10.
The sign * indicates the row variables is assumed to have no e!ect on the dependent
variable in the corresponding column. (!1) denotes the one-period lag. The number of lags is
determined using the log-likelihood criterion. ES
'°'"´ R
"S
'°'"´ R
!0.607C
R
; ES
'""'' R
"
S
'""'' R
!0.126C
R
and ES
'°' `º"'° R
"S
'""'' R
!0.056C
R
.
We estimated the system of equations speci"ed in (8) using full information maximum
likelihood. Following Dekimpe and Hanssens [9], we only include parameter estimates, which
have a t-value greater than one. Insigni"cant parameters are deleted step-wise, "rst deleting the
parameter with the lowest t-value and next re-estimating the equation, deleting the parameter
with the next lowest t-value and so on.
In the case of three major brands, Heinz, Hunts and Del Monte, the coe$cients of the
error-correction terms are negative as expected and signi"cant at the 1 percent level with
magnitudes of !1.25, !0.18 and !0.395, respectively. As mentioned earlier, an interpretation
of these coe$cients is that they re#ect the speed of adjustment of the corresponding dependent
variable towards the equilibrium. Thus, it is worth noting that Heinz, the market leader in terms
of market share, has the fastest speed of adjustment towards equilibrium while Hunts, the market
follower, is the slowest in terms of speed of adjustment. Del Monte, the market challenger, adjusts
faster than Hunts to equilibrium, suggesting that Del Monte is aggressively defending its share.
The results are consistent with the two goals of marketing as stated by Bass et al. [39], &One
purpose of marketing activity is to make behaviour non-stationary in a direction that is
favourable to a brand. Another purpose is to prevent behaviour from becoming non-stationary in
a direction which is not favourable to a brand'.
One di!erence we observe in the ketchup category results across the two di!erent data sets is
that, for the period from 1986 to 1988, market shares are stationary, and sales of all major brands
and category sales are stationary, while in the data set for the period from 1992 to 1994, market
shares are stationary but sales of all major brands and category sales are evolving. One possible
explanation for this is the di!erence between the two samples. The sample from 1986 to 1988 uses
weekly household scanner data on two test markets in the U.S., whereas the sample from 1992 to
1994 uses store movement data from 690 stores across 10 regions in the U.S. Yet, there is
BRAND SALES AND CATEGORY SALES 173
Copyright 2000 John Wiley & Sons, Ltd. Appl. Stochastic Models Bus. Ind. 2000; 16:159}177
*The empirical estimates from cointegration analysis and error correction models for the categories, sugar, peanut butter
and detergent are available with the authors and are not reported here in the interest of space.
consistency in the results with respect to the variables that are the focus of this paper: market
shares of major brands are stationary in both samples. This is an important point since category
sales may be driven by exogenous factors, hence regional di!erences matter, but market shares
are not and may depend only on the relative marketing activities of the "rms.
For the sugar category, unit root tests indicate that the market shares of C&H brand and Rest
are stationary while the sales of C&H as well as the brand sales of Rest are evolving as are
category sales. For the peanut butter category, these tests indicate that market shares of Ji!,
Skippy, Peter Pan and Rest are stationary, while brand and category sales are evolving.
Cointegration analysis and error-correction modelling for the sugar and peanut butter categories
reveal that these markets are also consistent with the existence of long-run equilibrium.*
Dekimpe and Hanssens [5] found that 78 per cent of the market share series they studied were
stationary, but that 68 per cent of the sales series were evolving. Our "ndings reconcile these
results.
(iv) Stationary market shares and evolving brand sales for a sub-set of major brands in an evolving
category (SEE). The dry detergent category is an exemplar for this scenario. We consider "ve
major brands*Surf, Tide, Cheer, Oxydol and Rest. Unit root tests summarized in Table II show
that market shares of Cheer and Oxydol are stationary, while those of Surf, Tide and Rest are
evolving. The sales of the brands Surf, Cheer, Oxydol and Rest are evolving while Tide sales
are stationary. The test indicates that category sales are also evolving. Next, we examine the
presence of a cointegrating relationship between the brand sales of Cheer and Oxydol and
category sales by performing the regressions of brand sales with category sales. We "nd that
a sub-set of the market consisting of the major brands Cheer and Oxydol is in long-run
equilibrium, while the sub-set consisting of Tide, Surf and Rest is evolving. In contrast, the Tide
brand has stationary sales and a non-stationary market share in a category that is evolving
downward and this suggests that Tide is gaining ground relative to competitors. These cases are
of interest. The question of the long-run relationship between the shares and marketing activity of
the brands is beyond the scope of this paper and we postpone this analysis to further research (see
for example, Reference [29]).
(v) Evolving market shares and evolving brand sales in a stationary category (EES). The brands
considered in the tuna category are Chicken-of-Sea, Starkist and Rest. The test results sum-
marized in Table II indicate that brand market shares and brand sales are evolving, while
category sales are stationary. Since all the market shares are evolving and sum to 1, this implies
that the cointegrating rank using the Johansen method is 1 with the long-run equilibrium
relationship between the brands having the coe$cients (1, 1, 1). Therefore, from a strategic
perspective, marketing activity may improve the relative position (share) as well as the absolute
position of some brands under this scenario.
5. CONCLUSIONS AND FUTURE RESEARCH
The major focus and contribution of this paper has been to study the long-run equilibrium in
market shares. We accomplish three objectives in this paper. First, our analysis adds to the
growing support for the empirical generalization that for many markets, market shares are
174 S. SRINIVASAN AND FRANK M. BASS
Copyright 2000 John Wiley & Sons, Ltd. Appl. Stochastic Models Bus. Ind. 2000; 16:159}177
stationary and in long-run equilibrium. Using data from 28 brands across eight product
categories, we "nd that market shares either for all brands or for major brands are stationary.
Dekimpe and Hanssens [5], using a database of over 400 prior studies, found that 78 per cent of
the market share series they studied were stationary, but that 68 per cent of the sales series were
evolving. Our "ndings reconcile these results. While each of the brand and category sales may
move up or down, there exists a long-run equilibrium towards which the market adjusts. A major
contribution of this paper is its demonstration, empirically and analytically, that the prior
empirical evidence that a majority of sales is in evolution is consistent with stationary market
shares, if brand and category sales are cointegrated. To the extent that competitive activities have
an e!ect on market share [7], an implication of our "ndings is that these activities may, in general,
only have a temporary e!ect on market share.
From a strategic perspective, we distinguish between market response at the primary demand
level (category sales), selective demand level (brand sales) and relative position level (market
share) and identify strategic scenarios depending upon their stable/evolving nature. In doing so,
we extend the work of Dekimpe et al. [8], who focus only on brand and category sales. The study
of market shares is important for several reasons. Among them, category and brand sales may be
driven by exogenous factors, while market shares depend mostly on the relative marketing
activity of "rms. Further, the managerial implications are di!erent; for example, consider the
implications of scenarios (i) and (iii) outlined in Section 3.2. Under the former, "rms are unable to
improve the long-run relative and absolute positions whereas under the latter scenario the
long-run absolute positions may improve even though "rms are unable to improve their relative
positions. From a managerial perspective, this distinction is important; for instance, the long-
term pro"tability implications of the two scenarios could be quite di!erent.
Our results indicate that promotions and other marketing activities are unlikely to increase
share in the long run. Recently, Mela et al. [40] examined the long-term e!ects of promotion and
advertising on consumer sensitivities to price and promotion and found that although these
variables do in#uence consumer sensitivities, they do not in#uence long-term market share,
a result that is entirely consistent with the existence of long-run equilibrium.
There exists the potential to disturb equilibrium through new products and product improve-
ments and, from a managerial perspective, the most fruitful approach for obtaining share gains
would appear to be an emphasis on such activities. A recent study reported by James Findlay of
Information Resources Inc. [44] indicates that of 240 product categories with thousands of brands
scanned in 1992 and 1993, only 164 brands increased share by more than three share points; of
those 164 brands, 80 per cent were associated with new product initiatives of one sort or another.
An additional area of managerial exploration associated with stationary markets is that of
possible reduced promotional spending. Ehrenberg et al. [4] have studied the after-e!ects of
price-related consumer promotions across many di!erent brands and product categories. They
assessed four aspects of possible after-e!ects of promotional sales peaks, including before-to-after
sales and before-to-after repeat buying. They concluded that consumer promotions for estab-
lished brands have no noticeable e!ect on subsequent sales or brand loyalty. Further, they
suggest that management implications point towards reduced rather than increased spending on
price-related promotions. However, not very much is known about equilibrium spending levels
that are required to maintain equilibrium shares. Unilateral reductions would seem to be risky
but worth exploring. Procter and Gamble recently experimented with reductions in consumer
and trade promotions [42, 43] with mixed results. A fruitful area of future research is to examine
the role of competitive marketing variables in maintaining market share equilibrium.
BRAND SALES AND CATEGORY SALES 175
Copyright 2000 John Wiley & Sons, Ltd. Appl. Stochastic Models Bus. Ind. 2000; 16:159}177
ACKNOWLEDGEMENTS
We thank the Editor, Jef L. Teugels and the referees for their very helpful comments that have greatly
improved the paper. We also thank Peter Popkowski Leszczyc, Philip Hans Franses, Clive W. Granger,
Dominique M. Hanssens and the Marketing Seminar participants at the University of Texas at Dallas, the
University of California, Irvine and the University of California, Riverside for their comments on an earlier
version of the paper.
REFERENCES
1. Ehrenberg ASC. Repeat Buying: Facts, ¹heory and Data. Oxford University Press: New York, 1972 (1988 2nd edn).
2. Bass FM, Jeuland AP, Wright GP. Equilibrium stochastic choice and market penetration theories. Management
Science 1976; 22(10):1051}1063.
3. Ehrenberg ASC. Theory or well-based results: which comes "rst? In Research ¹raditions in Marketing, Laurent G,
Lillien GL, Pras B (eds). Kluwer Academic Publisher: Boston, 1994, 79}105.
4. Bass FM, Pilon TL. A stochastic brand choice framework for econometric modeling of time series market share
behavior. Journal of Marketing Research 1980; 17:486}497.
5. Dekimpe MG, Hanssens DM. Empirical generalizations about market evolution and stationarity. Marketing Science
1995; 14(2):G109}121.
6. Lal R, Padmanabhan V. Competitive response and equilibria. Marketing Science 1995; 14(2):G101}108.
7. Blattberg RC, Briesch R, Fox EJ. How promotions work. Marketing Science 1995; 14(2):G122}132.
8. Dekimpe MG, Hanssens DM, Silva-Risso JM. Long-run e!ects of price promotions in scanner markets. Journal of
Econometrics 1999; 89:269}291.
9. Dekimpe MG, Hanssens DM. Sustained spending and persistent response: a new look at long-term marketing
pro"tability. Journal of Marketing Research 1999; forthcoming.
10. Campbell JY, Mankiw NG. Are output #uctuations transitory? Quarterly Journal of Economics 1987; 102(4):857}880.
11. Evans G. Output and unemployment dynamics in the United States. Journal of Applied Econometrics 1989; 4:213}237.
12. Lutkepohl H, Reimers H. Impulse-response analysis of cointegrated systems. Journal of Economic Dynamics and
Control 1992; 16(1):53}78.
13. Dekimpe MG, Hanssens DM. The persistence of marketing e!ects on sales. Marketing Science 1995; 14(1):1}21.
14. Bronnenberg J, Mahajan V, Vanhonacker WR. The emergence of market structure in new repeat-purchase categories:
a dynamic approach and an empirical application. Journal of Marketing Research 1999, forthcoming.
15. Franses PH, Kloek T, Lucas A. Outlier robust analysis of long-run marketing e!ects for weekly scanning data.
Journal of Econometrics 1999; 89:293}315.
16. Jedidi K, Mela CF, Gupta S. Managing advertising and promotion for long-run pro"tability. Marketing Science 1999;
18(1):1}22.
17. Engle RF, Granger CWJ. Cointegration and error correction: representation, estimation and testing. Econometrica
1987; 55(2):251}276.
18. Powers K, Hanssens DM, Hser Y, Anglin MD. Measuring the long-term e!ects of public policy: the case of narcotics
use and property crime. Management Science 1991; 37(6):627}644.
19. Baghestani H. Cointegration analysis of the advertising-sales relationship. Journal of Industrial Economics 1991;
34:671}681.
20. Zanias GP. The long run, causality, and forecasting in the advertising-sales relationship. Journal of Forecasting 1994;
13(7):601}610.
21. Franses PH. Modeling new product sales: an application of cointegration analysis. International Journal of Research
in Marketing 1994;11:491}502.
22. Veenstra AW, Franses PH. A cointegration approach to forecasting freight rates in the dry bulk shipping sector.
¹ransportation Research - A 1997; 31(6):447}458.
23. Banerjee A, Dolado J, Galbraith JW, Hendry DF. Cointegration, Error-Correction, and the Econometric Analysis of
Non-Stationary Data. Oxford University Press: Oxford, 1993.
24. Enders W. Applied Econometric ¹ime series. Wiley: New York, 1995.
25. Harris R. Cointegration Analysis in Econometric Modeling. Prentice-Hall: Englewood Cli!s, NJ, 1995.
26. Johansen S. Statistical analysis of cointegration vectors. Journal of Economic Dynamics and control, 1988; 12:231}254.
27. Johansen S, Juselius K. Maximum likelihood estimation and inference on cointegration - with applications to the
demand for money. Oxford Bulletin of Economics and Statistics 1990; 52(2):169}210.
28. Dickey DA, Fuller WA. Distribution of the estimators for autoregressive time series with a unit root. Journal of the
American Statistical Association 1979; 74:427}431.
29. Srinivasan S, Popkowski Leszczyc PTL, Bass FM. Market share response and competitive interaction: the role of
temporary, gradual and structural changes in prices. =orking paper, 2000.
176 S. SRINIVASAN AND FRANK M. BASS
Copyright 2000 John Wiley & Sons, Ltd. Appl. Stochastic Models Bus. Ind. 2000; 16:159}177
30. Ehrenberg ASC. Description and prescription. Journal of Advertising Research 1997; 37(6):17}22.
31. Maddala GS, Kim I. ;nit Roots, Cointegration and Structural Change. Cambridge University Press: Cambridge, UK,
1998.
32. Christen M, Gupta S, Porter JC, Staelin R, Wittink DR. Using market-level data to understand promotion e!ects in
a nonlinear model. Journal of Marketing Research 1997; 34:322}334.
33. Pesaran MH, Smith R. Estimating long-run relationships from dynamic heterogeneous panels. Journal of
Econometrics 1995; 68:79}113.
34. Link R. Are aggregate scanner data models biased? Journal of Advertising Research 1995; RC8}12.
35. Franses PH. Periodicity and Stochastic ¹rends in Economic ¹ime Series. Oxford University Press: Oxford, 1996.
36. Hendry DF, Doornik JA. Empirical Econometric Modelling ;sing PcGive 9.0 for =indows. International Thomson
Business Press, 1996.
37. Doornik JA, Hendry DF. Modelling Dynamic Systems: ;sing PcFiml 9.0 for =indows. International Thompson
Business Press: Boston, MA, 1996.
38. Philips PCB, Ouliaris S, Asymptotic properties of residual based tests for cointegration. Econometrica 1990;
58:165}193.
39. Bass FM, Givon MM, Kalwani MU, Reibstein D, Wright GP. An investigation into the order of the brand choice
process. Marketing Science 1984; 3(4):267}287.
40. Mela CF, Gupta S, Lehmann DR. The long-term impact of promotions and advertising on consumer brand choice.
Journal of Marketing Research 1997; 34:248}261.
41. Ehrenberg ASC, Hammond K, Goodhardt GJ. The after-e!ects of price-related consumer promotions. Journal of
Advertising Research 1994; 11}21.
42. Wall Street Journal. P&G ad chief plots demise of the coupon. 1996; B1:6.
43. Wall Street Journal. Down the drain: move to drop coupons puts Proctor & Gamble in sticky PR situation. 1997; A1.
44. Findlay JF. How to beat the odds in the new products game. Paper presented for Information Resources Inc. at the
Marketing Science Conference, University of Florida, Gainesville, March 1996.
45. Granger CWJ. Cointegrated variables and error-correcting models. Unpublished ;CSD Discussion Paper 83-13,
1983.
BRAND SALES AND CATEGORY SALES 177
Copyright 2000 John Wiley & Sons, Ltd. Appl. Stochastic Models Bus. Ind. 2000; 16:159}177

only have a temporary e!ect on market share. Our "ndings reconcile these results. market shares are predominantly stationary and sales series are largely evolving (e. some recent studies have examined the stationarity vs evolution in market share and sales. 2. however. A major contribution of this paper is its demonstration that the prior empirical evidence that a majority of sales series is in evolution is consistent with stationary market shares. more or less mature and approximately stationary*at least in the medium term (that is. References [3}6]) and sales are mostly in evolution. BASS averages of the choice probabilities over consumers at a given time are the market shares (share of choices) for that time period.. while there is a potential for long-term e!ects on sales. Ind. Research has found that for frequently purchased consumer goods. if brand sales and category sales are cointegrated. They may have some promotional ups and downs or the like superimposed'.160 S.. particularly in light of "ndings in the literature. In Figure 1(a)}1(c) we provide a graphical illustration of the phenomenon of stationarity in the market share of Heinz. Research has found that for frequently purchased consumer goods. the question naturally arises as to how "ts could be good when they are based on an assumption that implies the existence of unchanging market shares. Dekimpe and Hanssens [5]. respectively. 2000.most markets are.* Casual inspection of these graphs suggests stationary share behaviour. for the period from 1986 to 1988. The assumed constancy of underlying choice probabilities would imply unchanging market shares over time.g. found that 78 per cent of the market share series they studied were stationary.. The stochastic choice models have been shown to provide reasonably good descriptions of aggregate purchase incidence data for brands as well as of aggregate brand switching data. Stochastic Models Bus. in general. marketing mix variables only appear to have a temporary e!ect on share. SRINIVASAN AND FRANK M. BACKGROUND 2.1. Appl. over a 72-month period from 1960 to 1965. when measured in periods of a week or more and up to a year or two). Therefore. This seems to suggest the existence of unchanging individual choice probabilities that are temporarily perturbed by marketing activity. Copyright 2000 John Wiley & Sons. 16:159}177 . *We will return to a detailed description of the data shortly in Section 3. These market share patterns are important.g. Notably. We obtain this unique result as an outcome of the study. Inasmuch as market shares are known to be changing over time because of promotions and other marketing activities. Emerging empirical generalizations concerning stationarity and evolution of market share and sales As pointed out in Section 1. but that 68 per cent of the sales series were evolving. Ltd. using a database of over 400 prior studies. References [3}6]).. and over the period from 1992 to 1994. Ehrenberg [3] examines markets from a stochastic brand choice framework and has concluded that &. an implication of our "ndings is that these activities may. To the extent that competitive activities have an e!ect on market share [7]. We shall discuss and utilize formal tests of stationarity and we shall introduce and use cointegration analysis in an application of this methodology with respect to long-run equilibrium in markets. market shares are predominantly stationary (e.

Ind. Ltd. They estimated a system of equations where brand sales. 1992}1994. used time-series analysis to examine the long-run e!ects on sales of price promotions for ketchup. 16:159}177 . 1986}1988.BRAND SALES AND CATEGORY SALES 161 Figure 1. [8]. (b) Market share series for Heinz Brand in ketchup category. Appl. Dekimpe et al. (a) Market share series for Heinz Brand in ketchup category. soup and yogurt. they estimated vector}autoregressive models (VAR) and vector}error correction models (VECM). Using weekly scanner data for a period of 113 weeks. Stochastic Models Bus. 1960}1965. 2000. (c) Market share series for Heinz Brand in ketchup category. price and Copyright 2000 John Wiley & Sons. liquid detergent.

they are primarily concerned with aberrant observations that hamper the quality of data. cointegration analysis is used to identify whether trends in variables are related to each other in the long run. 2. Once a long-run component is identi"ed as being present in a series of interest. that temporary changes in a competitive variable will have long-run (permanent) e!ects. Recent studies have investigated the evolution of market shares. Evans [11] and Lutkepohl and Reimers [12] for other applications of persistence analysis. but this e!ect diminishes over time. 2000. [14] studied the ready-to-drink tea category and observed market shares are strongly in#uenced by retailer distribution decisions through the early growth stage. they "nd di!erent results for an outlier robust cointegration model as compared to non-robust methods.162 S. Using persistence analysis. Dekimpe and Hanssens [9] examined the persistence e!ects on sales of changes in marketing variables when sales series are evolving. Ind. and concluded that advertising does have a long-run impact on sales. Dekimpe et al. Using impulse}response analysis. Franses et al. BASS competitor's prices (all endogenous) are a function of lagged endogenous variables. However. On the other hand. where they observed a small long-term e!ect (a positive for private label and a negative e!ect for two national brands). First. but these e!ects are only temporary. [16] studied the long-run impact of advertising and promotions on brand choice and purchase quantity. Bronnenberg et al. They propose a utility framework. they traced the over time impact of a one-standard error shock in price. 16:159}177 . [15] utilize cointegration techniques to quantify the long-run e!ects of marketing e!ort.2. Modelling stationarity and long-run equilibrium using unit-root econometrics Modern time-series methodology such as unit-root econometrics and cointegration analysis are well suited to study the issue of stationarity and long-run equilibrium in markets. it is possible when competitive variables in#uence an evolving series. One exception is the soup market. Using a similar model. price and promotion) are a function of long-term advertising. The interested reader in referred to Campbell and Mankiw [10]. The authors use a logically consistent model to quantify how a brand's coupled market share and distribution evolve and to answer questions about the relative e$ciency of in#uencing share (pull) vs seeking distribution (push). This would suggest that these marketing expenditures only have a temporary e!ect. Dekimpe and Hanssens [13] measured the long-run e!ects of advertising on sales. utilizing individual level scanner panel data rather than aggregate weekly data. Jedidi et al. In their illustrative example. measured by the intrinsic brand preferences of the utility function. Appl. Ltd. [8] have shown (when both market shares and competitive variables are stationary) it is very unlikely that one will observe permanent o!setting competitive e!ects due to temporary changes in marketing e!ort. in stationary markets. long-term promotions and loyalty where long-run advertising and promotions are speci"ed as a geometric series of past advertising and promotional activities. unit root tests are used to test for the presence or absence of a unit root in the data. The idea of using Copyright 2000 John Wiley & Sons. di!erent from the models above. Therefore. They show that when sales are evolving it is possible for temporary variations in marketing variables to have a permanent e!ect on sales (hysteresis). Advertising only had a signi"cant positive e!ect on brand equity. and increasing price and price promotion sensitivity due to long-run promotions. Parameters in the utility function (intercept. They found a negative long-run e!ect on brand equity. Stochastic Models Bus. They concluded that price promotions have a signi"cantly di!erent impact on sales of national brands versus private labels. SRINIVASAN AND FRANK M. They also conducted a simulation analysis to determine the impact of changes in marketing strategies.

and (3) to distinguish. As observed earlier. (2) to demonstrate that the prior empirical evidence that a majority of sales series is in evolution is consistent with stationary market shares. from a strategic perspective. is low. Ltd. selective-demand level (brand sales) and relative-positive level (market share) and identify strategic scenarios depending upon their stable/evolving nature. S . Appl. *See Reference [23] for a detailed discussion of the econometric analysis of non-stationary data. if the amount e R R R by which the actual observations S deviate from the equilibrium level C . to forecast new automobile sales [21]. 27].* In other words. we might expect a decrease in brand sales in future periods. Cointegration analysis has been applied in a number of di!erent "elds.3. between market response at the primary-demand level (category sales). Given that brand and category sales are evolving. the Johansen method [26. Stochastic Models Bus. is high. given the observed R R level C . Table I presents a summary of the above-mentioned literature on long-run e!ects of marketing mix variables on sales and share. In dealing with the "rst objective. The equilibrium error e given by +S ! C . then brand sales are low relative to category sales and we might R\ R\ expect to see an increase in brand sales in future periods. 2000. A long-run equilibrium exists between brand sales. we "nd that market shares either for all brands or for major brands are stationary. to study the e!ects of public policy on the long-run relationship between narcotics usage and property crime [18]. 20]. is an upward (downward) trend in brand sales related to an upward (downward) trend in category sales? The trends would be related if there exists a long-run equilibrium or cointegrating relationship between brand and category sales. C . Using data from 28 brands across eight product categories.BRAND SALES AND CATEGORY SALES 163 cointegration analysis in the study of non-stationary time series comes from the work of Engle and Granger [17]. An advantage R\ R\ of cointegration analysis is that the extent of adjustment in a given period to deviations from the long-run equilibrium is given in the estimated equation. to forecast freight rates [22]. 15]. for example. Copyright 2000 John Wiley & Sons. contains useful R R R R information since the brand sales will move in a direction that satis"es the equilibrium. Ind. 16:159}177 . Objectives and contributions The major objectives of this paper are: (1) to provide additional evidence of the frequent existence of stationary market shares for frequently purchased consumer products. For example. 2. We demonstrate that the presence of a cointegrating relationship between brand and category sales also implies that market shares are mean stationary and hence is consistent with the above-mentioned empirical observation. References [5. What does this mean? This study contributes to an understanding and reconciliation of this empirical observation. 6]). if +S ! C . [23]. if brand sales and category sales are cointegrated. Several di!erent testing procedures are available to determine the existence of long-run cointegrating relationships such as the Engle and Granger method and the multivariate extension. to determine the long-run e!ect of marketing mix variables [9. For a detailed discussion of cointegration analysis we refer the reader to Banerjee et al. if the term +S ! C . analytically and empirically. our analysis adds to the growing support for the empirical generalization that for many markets. market shares are stationary (see for example. we investigate whether these trends are related to each other. Enders [24] and Harris [25]. previous research has found that shares tend to be stationary while sales tend to be in evolution. to measure the long-run relationship between advertising and sales [19. On the other hand. and category sales. is a stationary time series.

and diminishes during later stages of product life cycle Sales series are mostly evolving while shares are stationary Advertising has a long-term e!ect on sales Relative promotional expenditures have no impact on long-run shares Advertising and sales are cointegrated 1. Monthly pharmaceutical VECM data for 5 yr. we extend the work of Dekimpe et al. consider two scenarios. SRINIVASAN AND FRANK M. in a deviation from previous research by distinguishing. [15] Weekly scanner data for grocery product Scanner panel data. [16] Bronnenberg et al. Under the "rst scenario. "rms are unable to improve the long-run relative and absolute positions. Appl. From a strategic perspective. advertising and promotions have long-run impact on sales Advertising has long-run positive e!ect on sales. Ind. [8] 113 Weeks scanner data four-categories VAR Impulse response analysis Franses et al. In doing so. between market response at the primary-demand level (category sales). Sales calls. one where market shares are stationary in a category that has stationary sales and the other where market shares are stationary in a category that has evolving sales. and quantity model VAR Jedidi et al. from a strategic perspective. who focus only on brand and category sales. In non-stationary markets. advertising and price di!erential have long-run impact on sales. To illustrate. BRANDAID data Dekimpe et al. whereas under the second scenario the long-run absolute positions may improve even though "rms are unable to Copyright 2000 John Wiley & Sons. 16:159}177 . [14] Dekimpe and Hanssens [5] Dekimpe and Hanssens [13] Lal and Padmanabhan [6] Baghestani [19] Unit root tests VAR Persistence analysis Regression analysis VECM We deal with the third objective. One product class with four brands Weekly grocery data 1991}1996 (257 weeks) of ready-to-drink tea Database of 400 prior analyses Monthly advertising and sales data Yearly data from IRI Factbook 54 Yearly advertising and sales observations Cointegration analysis Individual level choice model. BASS Table I. small lasting e!ect. 2000. 2. selective-demand level (brand sales) and relative-position level (market share) and identifying potential scenarios depending upon their stable/evolving nature. Stochastic Models Bus. 691 households. [8]. 2. Studies looking at the long-run sales/share e!ects of marketing mix variables. Ltd. 8 yr (1984}1992). positive for private label. Study Dekimpe and Hanssens [9] Data Model Findings 1. promotions have a negative long-run e!ect Distribution in#uences long-run share through early growth.164 S. it is useful to make a distinction between sales and market share. and negative for two national brands Distribution. Advertising promotion and prices have no long-run impact on sales Promotions only have temporary e!ect in stationary market.

we describe the scanner data used and provide the empirical results. we do not consider it in the paper since it is not present in the data used and leave further analysis to future research. Section 5 presents conclusions and o!ers directions for future research. BRAND AND CATEGORY SALES 3. Overview of scenarios Stationary vs evolving conditions in market shares. soup. Stochastic Models Bus. We illustrate the strategic scenarios using data from eight consumer product categories (ketchup. Testing univariate equilibrium share models When a series may be appropriately modelled as depending on a constant plus a coe$cient times a lag of the series plus a random term. and. we provide an overview of the strategic scenarios with respect to market shares. stationary market shares and stationary brand sales of all brands in an evolving category. While this scenario is potentially interesting.* We consider the following scenarios: stationary market shares and stationary brand sales for all brands in a stationary category (SSS). Ltd.BRAND SALES AND CATEGORY SALES 165 improve their relative positions. When more than one lag is involved the appropriate test is the augmented Dickey}Fuller test (ADF) [23]. stationary market shares and evolving sales for all brands in an evolving category (SEE). brand and category sales give rise to eight possible scenarios. evolving market shares and stationary brand sales of all brands in a stationary category. evolving market shares and evolving brand sales in a stationary category (EES). stationary market shares and stationary brand sales for a sub-set of major brands in a stationary category (SSS). 2000. stationary market shares and evolving sales for a sub-set of major brands in an evolving category (SEE). We use these tests in examining the stationarity of market shares. The structure of the remainder of the paper is as follows. In Section 4.R We discuss these in turn: (i) Stationary market shares and stationary brand sales for all brands in a stationary category (SSS). peanut butter. dry detergent and tuna) for which scanner data are available and provide the strategic implications of each of these scenarios. Further. What does this mean from a strategic perspective? Stationary market shares imply that all *Of these eight scenarios. testing whether a series is stationary or evolving may be accomplished by means of the well-known test proposed by Dickey and Fuller [28]. which has not been done in prior research. for instance. 3.1. brand and category sales and discuss the estimation approach. RAnother scenario is evolving market shares and evolving brand sales in an evolving category. Ind. 3. LONG-TERM BEHAVIOUR OF MARKET SHARES. category sales may be driven by exogenous factors but market shares are not and they would depend on the relative marketing activities of the "rms. toilet tissue. Copyright 2000 John Wiley & Sons. These implications underscore the importance of distinguishing between market response at the primary-demand level (category sales). because the long-term pro"tability implications of the two scenarios could be quite di!erent. sugar. 16:159}177 .2. From a managerial perspective. Appl. sales and category sales. In Section 3. This refers to the case where all market shares are stationary as are brand and category sales. selectivedemand level (brand sales) and relative-position level (market share). this distinction is important. the following three cannot occur for logical reasons: stationary market shares and evolving brand sales of all brands in a stationary category.

only have a temporary e!ect on market share. Under this scenario. This is consistent with Ehrenberg [30].166 S. Cointegration analysis is useful in "nding stationary linear combinations of category and brand sales. Under this scenario. SRINIVASAN AND FRANK M. then this is consistent with market shares being stationary or stable. . to the extent that marketing activity has an e!ect on shares. This scenario has marketing strategy implications that are similar to those for scenario (iii) for the sub-set of brands that has stationary market shares. the two "rms may be engaged in a long-run market share battle. we investigate whether trends in variables are related to each other. 2000. this scenario also implies that at least two brands are engaged in a share war with one brand gaining at the expense of another. 16:159}177 . in general. Ind. this may not be case for the sub-set with evolving shares. To the extent that competitive activities have an e!ect on market share [7]. In the former case. (ii) Stationary market shares and stationary brand sales for a sub-set of brands in a stationary category (SSS). such deviations are stochastically bounded or diminishing over time. this implies that "rms are unable to improve their relative position in spite of improving their absolute long-run performance with respect to sales.3. Stochastic Models Bus. to the extent that marketing activity has an e!ect on shares. The concept of cointegration is a powerful one since it describes the existence of an equilibrium or stationary relationship among two or more time series. this kind of behaviour can arise due to two conditions. (iv) Stationary market shares and evolving brand sales for a sub-set of major brands in an evolving category (SEE). This refers to a scenario where all brands have stable market share in an expanding or declining market. (iii) Stationary market shares and evolving brand sales in an evolving category (SEE). . (v) Evolving market shares and evolving brand sales in a stationary category (EES). and to the extent that market shares are in#uenced by marketing variables. Assessment of market equilibrium using cointegration analysis If the sales series and category sales series are evolving. But what is the meaning of stationary market shares and evolving sales? If brand sales and category sales are cointegrated this also implies that the market is in long-run equilibrium. This scenario also implies that at least two brands have evolving market shares. an implication of our "ndings is that these activities may. Marketing variables may Copyright 2000 John Wiley & Sons. each of which is individually non-stationary [23]. even if they diverge from each other in the short run. in nearstationary markets. given that the category sales are stationary. In other words. everybody has to run hard to stand still. consistent with Srinivasan et al. the category expansion may result from macro-economic or demographic factors. First. marketing e!orts of "rms may drive the observed evolution in category sales. while a sub-set of major brands has stationary market shares. an equilibrium relationship would imply that. it is conceivable that brands do not change their marketing mix but merely ride the long-run demand waves in an industry. This suggests that while marketing activity may cause temporary perturbations in market shares for the sub-set of brands with stationary shares. Ltd. [29]. Appl. Second. BASS gains and losses are short-lived. From a strategic perspective. The trends are related to each other if there exists a long-run equilibrium or cointegrating relationship between the variables. From a strategic perspective.' A strategic implication is that competitive marketing activity may have a long-term impact on market shares for the evolving sub-set. at least two brands are engaged in a market share battle with one brand gaining at the expense of another in the long run. 3. If such regressions exist. &.

E(y ) is not time constant. one can show that a similar condition requiring the market shares across brands to sum to 1 is obtained.BRAND SALES AND CATEGORY SALES 167 show trending patterns in the typical two-year span of available scanner data. Appl. brand sales are cointegrated with category sales and market shares are stationary. We are thankful to Clive W. 2000. and their   sum. the existence of a long-term relationship implies that the equilibrium errors e and e  R  R represent stationary processes. The latter  R  R   condition is logical consistency constraint that the market shares across brands sum to 1. RFor the three. under scenario (iii). This is consistent with the interpretation of and as the long-run market shares of the brands.R   Equations (2) and (3) can be written as S /C "  R R S /C "  R R #e /C   R R #e /C   R R (6) (7) Hence if e (where i"1. 16:159}177 . If a long-run equilibrium relationship holds between brand sales S and category sales C . From an empirical perspective. y )"0 for all kO0.and four-brand case. Ltd.S G R G R\I This implies that the market shares de"ned by S /C and S /C are mean-stationary and have  R R  R R *We are grateful to Clive W. then they are related by an equilibrium relationship R R such as: S " C #e R R R (1) The existence of a long-term relationship implies that equilibrium error e represents a stationary R process. Cointegration models may therefore prove useful. then there exists a constant such that   S " C #e  R  R  R (3) (2) As before. zero mean and independent of C . then there exists a constant   such that S " C #e  R  R  R Similarly. Stochastic Models Bus. it follows that e "!e and # "1. is also I(1). as such models can allow for the presence of stochastic and deterministic trends in the data [15]. y )"0 for all kO0 and so will look like white noise in terms of its mean properties. If S is cointegrated with C.* Let S and S be the sales of the two brands that are evolving or I(1). Thus S /C G R G R R G R G R G R\I G R R R does not have time-constant second moments but is &mean-stationary'. Granger and an anonymous referee for pointing this out to us. then y "e /C will have G R R G R G R R E(y . if S is cointegrated with C. SAlthough y "e /C will have E(y )"0 and E(y . Copyright 2000 John Wiley & Sons. the category sales. We demonstrate this scenario analytically using a two-brand case. Equations (2) and (3) can be written as (1! )S ! S "e   R   R  R ! S #(1! )S "e   R   R  R (4) (5) For Equations (4) and (5) to hold. Granger for pointing this out to us and motivating the following discussion. 2) is iid. Ind.

155}197]. First. Loose signi"cance levels are not necessarily the best choice in cointegration analysis and the discussion on this issue cannot be regarded as closed with a clear guideline. we illustrate the strategic scenarios described in Section 3. Model choice by residual correlation statistics (such as Ljung}Box Q) has often been reported to yield the poorest results. Appl. The error-correction model is obtained by adding the terms containing the lagged residuals as shown below: S a  R  1 S " a # 0  R  C a 0 R  0 1 0 0 0 ! e b b   e # b b b 1 R\    G b e b b ! R\    1 R\ . that the prior empirical evidence that a majority of sales series is in evolution is consistent with stationary market shares. the latter uses a maximum-likelihood method that tests for the number of cointegrating regressions. Finally. 2000. Ltd. Several di!erent testing procedures are available to determine the existence of such long-run cointegrating relationships such as the Engle and Granger method [17] and the multivariate extension. Copyright 2000 John Wiley & Sons. the system will return to its long-run equilibrium. 1 1 and measure the speed of adjustment of the dependent variables towards the long-run ! equilibrium. if the data indicate the presence of unit roots. The coe$cients . Ind. In other words. Thus we demonstrate. *Though the likelihood-ratio test and AIC/SIC information criterion approach are equivalent. we investigate whether these non-stationary components are related to each other. which are usually "xed at 5 per cent in hypothesis testing. To summarize. While the former uses an ordinary least-squares approach. who present an excellent review on estimation of cointegrated systems. In this paper we will apply the Johansen maximum-likelihood estimator. we assess the presence of evolution versus stability in sales (or market share) by observing the behaviour of the series over time.  b S u  R\G  R S # u (8)  R\G  R C u R\G  R where N is the order to the model determined using the log-likelihood criterion. and are selected automatically (typically starting at more than 10 per cent in small samples but converging to 0 for larger ones) in the information criterion approach. Vector error correction models (VECM) Engle and Granger [17] show that the presence of a cointegrating relationship implies that the data are generated according to a partial adjustment or error-correction mechanism. 3. if cointegration has been established. The presence of a unit root indicates that there is an evolutionary component in sales or market share. Next.4. our framework consists of three major steps. the Johansen method [26.* The error correction term is obtained from the cointegrating regression. after short-run deviations from the equilibrium. Stochastic Models Bus. which is the most widely used approach. BASS long-run equilibrium values given by and . the real di!erence is in the signi"cance levels. SRINIVASAN AND FRANK M. the short-run dynamics are in#uenced by the deviation from the long-run relationship. Further.2 using data from the eight consumer product categories. The interested reader is referred to Maddala and Kim [31. if brand sales and category sales are cointegrated. pp. we test for the existence of a partial adjustment equilibrium or error-correcting behaviour.168 S. These short-run dynamics are captured in the VECM. The equilibrium error ensures that. 27]. The absence of a unit root indicates mean-reverting behaviour or stationarity. 16:159}177 . analytically and   empirically (in Section 4).

Thus the soup category is consistent with stationarity. determine the greatest lag with a signi"cant t-value and re-test using this number of lags. Each time series consists of 104 weekly observations. Nielsen for the period from 1992 to 1994. Market share of a brand is calculated as the share of the pooled unit sales of the brand.* Of these categories. We conducted the unit root test.R However. the competitive interaction e!ects may be subjected to aggregation bias [32. peanut butter.C. In addition. (i) Stationary market shares and stationary brand sales in a stationary category (SSS). We extend their analysis by examining market *The use of di!erent samples for the two databases on ketchup causes us to classify the two into di!erent scenarios and we shall return to this issue shortly. [8].C. etc. brand and category sales. A constant is included and we test for trends and seasonality in the data. Our database aggregates across varieties to the brand level. we performed pooling tests to determine whether we can pool the di!erent varieties for a brand. Nielsen household scanner panel data from two test markets in the U. RRetailers typically assign each product a SKU (similar to the universal product code (UPC)) and use it as a unique identi"er for their products to help track sales. who recommends using data within homogeneous sub-sets to overcome the bias issue. that of store movement data supplied by A. [8] using the same data source. Controlling for seasonality. this causes some loss of information due to possible underlying heterogeniety among varieties. we "nd that in the long run the relative position of the players is una!ected by the marketing mix activity and all gains and losses are temporary. We discuss the results for each one of the "ve scenarios in turn. the ketchup market is consistent with stationarity or with the existence of a long-run equilibrium.S Therefore. Ind. To control for this bias. Stochastic Models Bus. have also been examined by Dekimpe et al. Next. 2000.BRAND SALES AND CATEGORY SALES 169 4. dry detergent. tuna. SWe start by including 12 lags.A Our results summarized in Table II show that market shares of Campbell brand and Rest are stationary as are brand sales and category sales. We compare our "ndings with theirs later in this section. DATA DESCRIPTION AND EMPIRICAL RESULTS A. Therefore. Thus. A Seasonality can cause non-stationarity and persistent e!ects (see for example. on the purchases of ketchup. two databases. Reference [35]). the ketchup scanner panel data and the soup scanner panel data. We provide a brief description of the brands considered in each category in Table II. 33]. Copyright 2000 John Wiley & Sons. We should point out that these ketchup category results are consistent with those obtained by Dekimpe et al. We "nd this case holds for two categories*ketchup and soup. the results in Table II show that market shares of all brands are stationary while the brand sales of all brands as well as category sales are stationary. toilet tissue. consistent with our focus on the maintenance of market share equilibrium at the brand level rather than at the Stock Keeping Unit (SKU) level. We control for seasonal e!ects in sales by using the seasonal equivalent of Equation (5) for the ADF test (see Reference [5]). we used another database for the ketchup category. we examine the soup category. Ltd. We describe the results for each of these categories in turn. Over 90 per cent of the varieties could be pooled consistent with Link [34]. The latter has a sample of 690 stores across ten regions in the US. Appl. soup and sugar were used to construct the time series of weekly shares.S. This resulted in a total of 28 brands in the eight databases that we analyse. we "nd that the series is stationary. we consider the major brands and aggregated all brands other than the major brands into one "gure referred to as Rest in all the categories. 16:159}177 . In all categories. consistent with Lal and Padmanabhan [6].

Rest 5. Hunts.03* Category sales Market share ADF range and signi"cance level Empirical observation: category.36 !3. ¹oilet tissue (1985}1987) Northern and Rest 4.19 to !6.32 to !5. Brand sales Stationary (all) Stationary (all) !3. BASS III: SSE Stationary (major brand sub-set) Stationary (all) Evolving !4. Ltd. SRINIVASAN AND FRANK M.65 to !7.10 !3. ERIM scanner data (1985}1987) and store movement data (1992}1994). *Signi"cance at the 5 per cent level. Stochastic Models Bus. 2000. Classi"cation of Scenarios . Rest 2. (year). . ¹una (1985}1987) Chicken of Sea. Peter Pan. Dry detergent (1986}1988) Cheer and Oxydol 8. Ketchup (1992}1994) Heinz. Hunts. brands Strategic implications All gains and losses are temporary Scenarios Market share I: SSS Stationary (all) 2000 John Wiley & Sons.02 !5. 16:159}177 A.170 Copyright Table II. denotes signi"cance at the 1 per cent level. Rest Appl. Ketchup (1986}1988) Heinz. DM.67 Same as case (iii) for the stationary sub-set Marketing activity may improve relative as well as absolute position of some of the brands V: EES Stationary (major brand sub-set) Evolving (all) Stationary Evolving (major brand sub-set) Evolving (all) None are signi"cant 1.56 to !7. Sugar (1985}1987) C&H and Rest 6.10 Stationary (major brand sub-set) Evolving (all) All gains and losses are temporary for the major brand sub-set Long-run absolute performance is evolving but "rms are unable to improve relative position IV: SEE Evolving !3. Skippy and Rest 7. Ind. Peanut butter (1985}1987) Ji!. Nielsen: Single-source scanner data (1986}1988).67 S.40 to !7. Soup (1986}1988) Campbell and Rest 3.C. II: SSS Stationary !4. Star Kist. Del Monte.

31 12. One possible explanation is that we control for seasonality. Therefore. The unit root test results summarized in Table II show that the market shares of Northern and Rest are stationary. to the extent that marketing activity has an e!ect on shares.84 Yes Yes Yes No None At most 1 At most 2 At most 3  '     Hypothesized No. while marketing activity causes temporary perturbations in market shares for the sub-set of brands with stationary shares. This case refers to a situation where market shares are stationary. (iii) Stationary market shares and evolving brand sales in an evolving category (SEE). the number of signi"cant cointegrating vectors and the associated vectors with the long-run parameters. These brands are Charmin and Cottonelle in the toilet tissue category. we examine the category sales. Stochastic Models Bus.89 24. sugar and peanut butter.BRAND SALES AND CATEGORY SALES 171 shares. The Johansen FIML test indicated the presence of three cointegrating relationships among the variables. Johansen's FIML test of cointegrating rank. In a divergence from their results. 2000. the sales of Northern and Rest brands are stationary. in addition to brand and category sales. Next. we proceed to examine whether there is a cointegrating relationship between each of the brand and category sales. Thus. we "nd that the soup market is stationary. this scenario implies that at least two brands may be engaged in a share war with one brand gaining at the expense of another. First. Copyright 2000 John Wiley & Sons. Category sales are stationary. Del Monte and Rest*are evolving. of CE(s) denotes rejection of the hypothesis at 5 per cent signi"cance level. Several di!erent test procedures exist to determine the cointegration rank. a sub-market consisting of two major brands has stationary market shares. We "nd this condition to hold in three categories*ketchup.91 5 Per cent critical value     39. with the most widely used being the trace test and the eigenvalue test (see References [26. while the market shares of Charmin and Cottonelle are evolving. 16:159}177 . We use the Johansen and Juselius estimation method [27]. (ii) Stationary market shares and stationary brand sales for a sub-set of major brands in a stationary category (SSS). Hunts.51 21. show that market shares of all brands are stationary. Cointegration analysis is performed on four variables: sales of Heinz. Further. Our test results show that category sales are also evolving. The sales of the brands*Heinz. 27]). From a strategic perspective. Table III. more work is needed to enhance the power of unit root tests as it is sometimes di$cult to classify a series as being stationary or evolving with a low persistence level. this may not be the case for the sub-set with evolving shares [29]. whereas they found evidence of evolution in sales. This method determines the cointegration rank. Appl. Likelihood ratio  87. Ind. we need to determine the number of lags to include in the cointegration relation. Hunts. even though brand and category sales are evolving. or in long-run equilibrium.53 3. The PCFiml software is used for estimation purposes [36].67 0. Unit root test results. which are summarized in Table II. Del Monte and category sales. The results of this analysis are provided in Table III.34 47. while those of Charmin and Cottonelle are evolving. As they point out. We "nd this to be valid in one category*toilet tissue. In this case. Ltd.

172 S. Variable C ADF statistic Critical value Cointegration S &  S & 0.0028) !4. Long-run equilibrium relationship between brand sales and category sales .88 !2. SRINIVASAN AND FRANK M.76 Yes S " . BASS Table IV.126 (0.

056 (0. Based on this analysis. The results reported are the so-called restricted long-run relationships. y )"0 for all kO0 and so will look like white G R G R R R R\I noise in terms of its mean properties. As pointed out earlier. In Johansen's method. In all three cases. The intuition is that non-stationary movements in category sales remove non-stationary movements in brand sales. towards which the market adjusts.76 Yes The standard error of the estimate is included in parenthesis. the long-run market share of Hunts is 12. the cointegrating regression is estimated jointly with the error correction model.0014) !6.5 percent.005) !7. we "rst estimated the unrestricted relationship and next tested for restriction using log-likelihood ratio tests.607 (0. 16:159}177 . Similarly. and the long-run market shares are mean-stationary.2. Therefore. all are signi"cant at the 1 percent level.59 !2. The cointegrating vectors are reported in Table IV.6 percent and the long-run market share of Del Monte is 5. we test for independence between e and C . Stochastic Models Bus. The critical value (at 5 per cent level) for unit root tests obtained from Philips and Oularis [38]. As pointed out in Section 3. The "rst cointegrating vector shows the long-run relationship between Heinz sales and category sales. Johansen's FIML test has provided the cointegrating vectors and has established the stationarity of the residuals of the cointegrating regressions. are iid. we "nd that the residuals are white-noise G R R processes and independent of C . the second one the long-run relationship between Hunts sales and category sales and the third the relationship between Del Monte sales and category sales. we concluded that sales of Hunts (Del Monte) is only signi"cant in the second (third) long-run relationship and was constrained to zero in the other long-run relationships. the idea is that a portion of the disequilibrium from one period to the next is corrected in the next period. We "nd that there are three signi"cant cointegrating vectors.76 Yes 0. the residuals from each of the cointegrating regressions.7 percent. each of which is non-stationary. the interpretation of the coe$cients is that the R long-run market share of Heinz is 60.01.15 !2. Copyright 2000 John Wiley & Sons. a long-run equilibrium exists. we concluded that sales of Heinz is only signi"cant in the "rst long-run relationship and was constrained to zero in the other long-run relationships. Appl. Signi"cant at p(0. as in G R Equations (2) and (3). Further. this R will then imply that y "e /C will have E(y . zero mean and independent of C . we run one more test proposed by Philips and Oularis [37] to con"rm that the residuals from the cointegrating regression are white noise (in Table IV). Ltd. As described in the previous section. 2000. The output from the estimation of the error-correction model is shown in Table V. Following Doornik and Hendry [37]. Nevertheless. Ind. We test for the condition that e . + 0. so that only transitory components are left in the residuals. the concept of cointegration allows us to describe the existence of an equilibrium relationship between two series. While each of the brand and category sales may move up or down.

Variable C (!1) S (!1) &  S (!1) & S (!1) " .BRAND SALES AND CATEGORY SALES 173 Table V. Results of the VECM for the ketchup category .

+ ES (! 1) &  ES (!1) & ES (!1) " .

+ !1.138 (0.208) 0.264) 0.188 (0.124) * !0.584 (0.070) * S & S " .257 (0.283) * * S &  !0.646 (0.

097) The standard error of the estimate is included in parenthesis.013 (0. + !0. ES " & &  R R & R S !0.10.05 and Signi"cant at p(0.007) !0. The sign * indicates the row variables is assumed to have no e!ect on the dependent variable in the corresponding column. Signi"cant at p(0.199 (0. The number of lags is determined using the log-likelihood criterion.126C and ES "S !0.607C .056C .01.395 (0. ES "S !0.085) * * !0. Signi"cant at p(0. (!1) denotes the one-period lag.  R & R R " .

which have a t-value greater than one. As mentioned earlier. !0. the market challenger.S. we only include parameter estimates. Ind. 2000. the coe$cients of the error-correction terms are negative as expected and signi"cant at the 1 percent level with magnitudes of !1. the market leader in terms of market share.18 and !0. Another purpose is to prevent behaviour from becoming non-stationary in a direction which is not favourable to a brand'. [39]. Stochastic Models Bus. Appl. Hunts and Del Monte. Heinz. 16:159}177 . is the slowest in terms of speed of adjustment. One di!erence we observe in the ketchup category results across the two di!erent data sets is that.S. respectively. Del Monte.. Yet. it is worth noting that Heinz.25. One possible explanation for this is the di!erence between the two samples. whereas the sample from 1992 to 1994 uses store movement data from 690 stores across 10 regions in the U. Thus. suggesting that Del Monte is aggressively defending its share. + R & R R We estimated the system of equations speci"ed in (8) using full information maximum likelihood. &One purpose of marketing activity is to make behaviour non-stationary in a direction that is favourable to a brand. Ltd. adjusts faster than Hunts to equilibrium. for the period from 1986 to 1988. the market follower. an interpretation of these coe$cients is that they re#ect the speed of adjustment of the corresponding dependent variable towards the equilibrium. Insigni"cant parameters are deleted step-wise. deleting the parameter with the next lowest t-value and so on.395. The results are consistent with the two goals of marketing as stated by Bass et al. Following Dekimpe and Hanssens [9]. has the fastest speed of adjustment towards equilibrium while Hunts. market shares are stationary but sales of all major brands and category sales are evolving. In the case of three major brands. while in the data set for the period from 1992 to 1994. "rst deleting the parameter with the lowest t-value and next re-estimating the equation. market shares are stationary. The sample from 1986 to 1988 uses weekly household scanner data on two test markets in the U. there is Copyright 2000 John Wiley & Sons. and sales of all major brands and category sales are stationary.

Therefore.* Dekimpe and Hanssens [5] found that 78 per cent of the market share series they studied were stationary. 1. market shares are *The empirical estimates from cointegration analysis and error correction models for the categories. SRINIVASAN AND FRANK M. BASS consistency in the results with respect to the variables that are the focus of this paper: market shares of major brands are stationary in both samples. CONCLUSIONS AND FUTURE RESEARCH The major focus and contribution of this paper has been to study the long-run equilibrium in market shares. Cheer. First. hence regional di!erences matter. while category sales are stationary. These cases are of interest. while the sub-set consisting of Tide. The dry detergent category is an exemplar for this scenario. but that 68 per cent of the sales series were evolving. we examine the presence of a cointegrating relationship between the brand sales of Cheer and Oxydol and category sales by performing the regressions of brand sales with category sales. Tide and Rest are evolving. unit root tests indicate that the market shares of C&H brand and Rest are stationary while the sales of C&H as well as the brand sales of Rest are evolving as are category sales. this implies that the cointegrating rank using the Johansen method is 1 with the long-run equilibrium relationship between the brands having the coe$cients (1. Unit root tests summarized in Table II show that market shares of Cheer and Oxydol are stationary. peanut butter and detergent are available with the authors and are not reported here in the interest of space. Appl. our analysis adds to the growing support for the empirical generalization that for many markets. Starkist and Rest. but market shares are not and may depend only on the relative marketing activities of the "rms.174 S. sugar. while brand and category sales are evolving. from a strategic perspective. Oxydol and Rest. Stochastic Models Bus. We "nd that a sub-set of the market consisting of the major brands Cheer and Oxydol is in long-run equilibrium. We consider "ve major brands*Surf. Cointegration analysis and error-correction modelling for the sugar and peanut butter categories reveal that these markets are also consistent with the existence of long-run equilibrium. while those of Surf. Reference [29]). The test results summarized in Table II indicate that brand market shares and brand sales are evolving. In contrast. 16:159}177 . Peter Pan and Rest are stationary. Cheer. (v) Evolving market shares and evolving brand sales in a stationary category (EES). The question of the long-run relationship between the shares and marketing activity of the brands is beyond the scope of this paper and we postpone this analysis to further research (see for example. marketing activity may improve the relative position (share) as well as the absolute position of some brands under this scenario. For the sugar category. Next. The brands considered in the tuna category are Chicken-of-Sea. 5. Tide. Skippy. 2000. Oxydol and Rest are evolving while Tide sales are stationary. We accomplish three objectives in this paper. (iv) Stationary market shares and evolving brand sales for a sub-set of major brands in an evolving category (SEE). The sales of the brands Surf. The test indicates that category sales are also evolving. For the peanut butter category. Surf and Rest is evolving. This is an important point since category sales may be driven by exogenous factors. the Tide brand has stationary sales and a non-stationary market share in a category that is evolving downward and this suggests that Tide is gaining ground relative to competitors. Copyright 2000 John Wiley & Sons. these tests indicate that market shares of Ji!. 1). Since all the market shares are evolving and sum to 1. Ind. Our "ndings reconcile these results. Ltd.

Our "ndings reconcile these results. Our results indicate that promotions and other marketing activities are unlikely to increase share in the long run. selective demand level (brand sales) and relative position level (market share) and identify strategic scenarios depending upon their stable/evolving nature. not very much is known about equilibrium spending levels that are required to maintain equilibrium shares. From a managerial perspective. Unilateral reductions would seem to be risky but worth exploring. Recently. an implication of our "ndings is that these activities may. if brand and category sales are cointegrated. that the prior empirical evidence that a majority of sales is in evolution is consistent with stationary market shares. in general. From a strategic perspective. the most fruitful approach for obtaining share gains would appear to be an emphasis on such activities. of those 164 brands. category and brand sales may be driven by exogenous factors. They assessed four aspects of possible after-e!ects of promotional sales peaks. A recent study reported by James Findlay of Information Resources Inc. [8]. only 164 brands increased share by more than three share points. Further. Among them. 80 per cent were associated with new product initiatives of one sort or another. this distinction is important. for example. they do not in#uence long-term market share. [40] examined the long-term e!ects of promotion and advertising on consumer sensitivities to price and promotion and found that although these variables do in#uence consumer sensitivities. Appl. "rms are unable to improve the long-run relative and absolute positions whereas under the latter scenario the long-run absolute positions may improve even though "rms are unable to improve their relative positions. Ehrenberg et al. a result that is entirely consistent with the existence of long-run equilibrium. [4] have studied the after-e!ects of price-related consumer promotions across many di!erent brands and product categories. Copyright 2000 John Wiley & Sons. In doing so. for instance. the longterm pro"tability implications of the two scenarios could be quite di!erent. A major contribution of this paper is its demonstration. Further. The study of market shares is important for several reasons. While each of the brand and category sales may move up or down. There exists the potential to disturb equilibrium through new products and product improvements and. Using data from 28 brands across eight product categories. there exists a long-run equilibrium towards which the market adjusts. They concluded that consumer promotions for established brands have no noticeable e!ect on subsequent sales or brand loyalty. from a managerial perspective.BRAND SALES AND CATEGORY SALES 175 stationary and in long-run equilibrium. while market shares depend mostly on the relative marketing activity of "rms.2. we extend the work of Dekimpe et al. An additional area of managerial exploration associated with stationary markets is that of possible reduced promotional spending. 43] with mixed results. consider the implications of scenarios (i) and (iii) outlined in Section 3. 16:159}177 . but that 68 per cent of the sales series were evolving. empirically and analytically. Ind. However. they suggest that management implications point towards reduced rather than increased spending on price-related promotions. we "nd that market shares either for all brands or for major brands are stationary. who focus only on brand and category sales. A fruitful area of future research is to examine the role of competitive marketing variables in maintaining market share equilibrium. found that 78 per cent of the market share series they studied were stationary. [44] indicates that of 240 product categories with thousands of brands scanned in 1992 and 1993. Mela et al. the managerial implications are di!erent. Dekimpe and Hanssens [5]. To the extent that competitive activities have an e!ect on market share [7]. Procter and Gamble recently experimented with reductions in consumer and trade promotions [42. Ltd. including before-to-after sales and before-to-after repeat buying. we distinguish between market response at the primary demand level (category sales). Stochastic Models Bus. Under the former. using a database of over 400 prior studies. only have a temporary e!ect on market share. 2000.

Johansen S. Journal of the American Statistical Association 1979. Bass FM. Anglin MD. Jeuland AP. Long-run e!ects of price promotions in scanner markets. Ind. Dekimpe MG. Maximum likelihood estimation and inference on cointegration . gradual and structural changes in prices. 13(7):601}610. 89:293}315. Franses PH. Ehrenberg ASC. Management Science 1976. Dolado J. Veenstra AW. 1995. Vanhonacker WR. Journal of Economic Dynamics and Control 1992. Dickey DA. 2. 2000. The long run. 12. Oxford University Press: New York. 31(6):447}458. Wright GP. 16(1):53}78. Mela CF.with applications to the demand for money. =orking paper. Reimers H. Lal R.176 S. 1994. Copyright 2000 John Wiley & Sons. 22. estimation and testing. Pras B (eds). 7. 37(6):627}644. Dekimpe MG. 16:159}177 . Dekimpe MG. 1995. and forecasting in the advertising-sales relationship. 13. NJ. Outlier robust analysis of long-run marketing e!ects for weekly scanning data. Oxford University Press: Oxford. Journal of Applied Econometrics 1989. 1972 (1988 2nd edn). Silva-Risso JM. Marketing Science 1995. Lillien GL. Bass FM. Stochastic Models Bus. 52(2):169}210. Hanssens DM. Market share response and competitive interaction: the role of temporary. 3. and the Econometric Analysis of Non-Stationary Data. Jedidi K. Journal of Econometrics 1999. We also thank Peter Popkowski Leszczyc. 17:486}497. Harris R. Bronnenberg J. causality. A cointegration approach to forecasting freight rates in the dry bulk shipping sector. Theory or well-based results: which comes "rst? In Research ¹raditions in Marketing. Marketing Science 1995.11:491}502. 14.A 1997. the University of California. Prentice-Hall: Englewood Cli!s. Pilon TL. SRINIVASAN AND FRANK M. 18. Cointegration and error correction: representation. Applied Econometric ¹ime series. 20. Banerjee A. forthcoming. Ltd. Measuring the long-term e!ects of public policy: the case of narcotics use and property crime. Journal of Marketing Research 1999. 9. 14(2):G101}108. Hanssens DM. 55(2):251}276. Gupta S. Teugels and the referees for their very helpful comments that have greatly improved the paper. 89:269}291. Journal of Forecasting 1994. Powers K. Wiley: New York. 27. Hendry DF. Journal of Econometrics 1999. 29. Baghestani H. 2000. Modeling new product sales: an application of cointegration analysis. 74:427}431. Statistical analysis of cointegration vectors. Riverside for their comments on an earlier version of the paper. Journal of Economic Dynamics and control. Econometrica 1987. Dekimpe MG. Mankiw NG. Distribution of the estimators for autoregressive time series with a unit root. Briesch R. Enders W. 19. Appl. Franses PH. Equilibrium stochastic choice and market penetration theories. Johansen S. Hanssens DM. 14(2):G122}132. Competitive response and equilibria. 16. Granger CWJ. BASS ACKNOWLEDGEMENTS We thank the Editor. The emergence of market structure in new repeat-purchase categories: a dynamic approach and an empirical application. Galbraith JW. 10. Jef L. Management Science 1991. ¹heory and Data. 28. REFERENCES 1. Laurent G. 12:231}254. 4:213}237. Kloek T. Franses PH. Are output #uctuations transitory? Quarterly Journal of Economics 1987. 17. International Journal of Research in Marketing 1994. 79}105. 11. 14(2):G109}121. Journal of Industrial Economics 1991. The persistence of marketing e!ects on sales. Granger. Journal of Marketing Research 1980. Hanssens DM. Marketing Science 1999. 6. Clive W. Zanias GP. 26. 18(1):1}22. Ehrenberg ASC. Managing advertising and promotion for long-run pro"tability. Repeat Buying: Facts. 5. Marketing Science 1995. Srinivasan S. Hser Y. 15. 23. Irvine and the University of California. 4. Lutkepohl H. Dominique M. 1988. 24. Campbell JY. Engle RF. Lucas A. Kluwer Academic Publisher: Boston. Marketing Science 1995. Sustained spending and persistent response: a new look at long-term marketing pro"tability. Empirical generalizations about market evolution and stationarity. Cointegration. forthcoming. ¹ransportation Research . Fuller WA. Philip Hans Franses. 25. A stochastic brand choice framework for econometric modeling of time series market share behavior. Padmanabhan V. Juselius K. Impulse-response analysis of cointegrated systems. 8. Mahajan V. Fox EJ. Journal of Marketing Research 1999. 14(1):1}21. 1993. Evans G. Bass FM. 21. Cointegration Analysis in Econometric Modeling. Output and unemployment dynamics in the United States. 102(4):857}880. Oxford Bulletin of Economics and Statistics 1990. Hanssens and the Marketing Seminar participants at the University of Texas at Dallas. How promotions work. Popkowski Leszczyc PTL. 22(10):1051}1063. Error-Correction. Cointegration analysis of the advertising-sales relationship. Hanssens DM. 34:671}681. Blattberg RC.

Hendry DF. 42. Cointegrated variables and error-correcting models. International Thomson Business Press. B1:6. 2000. Are aggregate scanner data models biased? Journal of Advertising Research 1995. Mela CF.BRAND SALES AND CATEGORY SALES 177 30. How to beat the odds in the new products game. 34:322}334. Ehrenberg ASC. Stochastic Models Bus. March 1996. 1998. 39.0 for =indows. Gainesville. Lehmann DR. Gupta S. 1996. at the Marketing Science Conference. Ind. 32. Reibstein D. Journal of Advertising Research 1994. Ouliaris S. MA. Findlay JF.0 for =indows. Staelin R. The after-e!ects of price-related consumer promotions. Estimating long-run relationships from dynamic heterogeneous panels. Wittink DR. 1996. 3(4):267}287. 36. Doornik JA. Doornik JA. Ltd. Givon MM. Modelling Dynamic Systems: . 1996. 40. Christen M. Journal of Econometrics 1995. 31. 34:248}261. Journal of Advertising Research 1997. Gupta S. 38. Hammond K. 16:159}177 . Ehrenberg ASC. Kalwani MU. 43. 37(6):17}22. Econometrica 1990. Wright GP. Appl. Porter JC. Hendry DF. Copyright 2000 John Wiley & Sons. Philips PCB. International Thompson Business Press: Boston. 11}21. Periodicity and Stochastic ¹rends in Economic ¹ime Series. Smith R. P&G ad chief plots demise of the coupon. Maddala GS. Marketing Science 1984.sing PcGive 9. Asymptotic properties of residual based tests for cointegration. UK. Wall Street Journal. 34. 1983. Wall Street Journal. Granger CWJ. Goodhardt GJ. 58:165}193. . Paper presented for Information Resources Inc. 35. 37. 1996. 33. Using market-level data to understand promotion e!ects in a nonlinear model. Down the drain: move to drop coupons puts Proctor & Gamble in sticky PR situation. A1. Journal of Marketing Research 1997. RC8}12. Bass FM. Journal of Marketing Research 1997. 1997. Empirical Econometric Modelling . An investigation into the order of the brand choice process. 41. 44. Oxford University Press: Oxford. Link R. 45. Description and prescription.sing PcFiml 9. Cambridge University Press: Cambridge. Kim I. Cointegration and Structural Change. The long-term impact of promotions and advertising on consumer brand choice. Franses PH. Unpublished .nit Roots. 68:79}113. University of Florida. Pesaran MH.CSD Discussion Paper 83-13.