price war: what is it good for? store incidence and basket...
TRANSCRIPT
1
Price War: What is It Good for?
Store Incidence and Basket Size Response to the Price War in Dutch Grocery Retailing
Harald J. van Heerde1
Els Gijsbrechts
Koen Pauwels
OCTOBER 13, 2005
Acknowledgements: The authors gratefully acknowledge the support of Aimark and Publi-info, which provided the data for this study. The assistance of Alfred Dijs, Dick Valstar, Peter Gouw, Ton Luyten and Vincent van Witteloostuyn (Europanel) is especially acknowledged. For their constructive comments and feedback, we thank Marnik Dekimpe, Scott Neslin, Laurens Sloot and seminar participants at the Marketing Dynamics Conference, the Marketing Science Conference and North-East Marketing Consortium. Research funding was provided by the Marketing Science Institute and by the Netherlands Organization for Scientific Research (for the first author).
1 Harald van Heerde is Professor of Marketing at Tilburg University, PO Box 901513, 5000 LE Tilburg, The Netherlands ([email protected]).. Els Gijsbrechts is Professor of Marketing at Tilburg University, PO Box 901513, 5000 LE Tilburg, The Netherlands ([email protected]). Koen Pauwels is Associate Professor of Business Administration, Tuck School of Business at Dartmouth, Hanover, New Hampshire 03755 ([email protected]).
2
ABSTRACT
Price War: What is It Good for?
Store Incidence and Basket Size Response to the Price War in Dutch Grocery Retailing
While retail price wars have received much business press and some research attention, it
is unclear how they affect the major components of retailer performance: store incidence and
basket size. Based on consumer hand scan and perceptual data, we analyze the impact of a recent
price war with a multivariate Tobit II model. We find that increased consumer price sensitivity
offsets the improved price image for the pioneer, while it benefits competitors with high value-
for-money.
Keywords: price war, Tobit II model, grocery retailing, store incidence, basket size, price image.
3
INTRODUCTION
Retail price competition has become increasingly vivid in recent years, reducing retailer
profitability (Ailawadi 2001). Discounters such as Walmart, Aldi and Lidl are challenging
traditional retail formats on both sides of the Atlantic (Business Week October 6, 2003). In the
United States, Walmart now controls a large part of the retail market and is driving down prices
at other retailers. In the Netherlands, over 52% of households frequently shopped at hard
discounters Aldi or Lidl in the Fall of 2003, up from 30% in 2001 (GfK 2003). The reaction of
traditional retailers has varied from focusing on quality and service to engaging the challengers
with lower prices. However, such lower prices may trigger price wars, which can last for a long
time and strongly affect all market players (Rao, Bergen and Davis 2000).
Price wars are generally believed to hurt revenues and long-term prospects
(Brandenburger and Nalebuff 1996). Unless the initiating company has a cost advantage, price
wars are good for “absolutely nothing” (Henderson 1997). While the antecedents of price wars
have been well documented (see the next section), there is hardly any empirical research on their
consequences. Indeed, a recent review concludes: “It is unclear what the overall effects of price
wars are. Price wars are often assumed to lead to losses for the firms involved in the battle … It
is, therefore, important to research how price wars affect firms in the industry, whether these
effects are uniformly distributed, and how such effects persist in the long run through lower
reference prices” (Heil and Helsen 2001, p.96).
In our grocery retailer setting, we study the consequences of a price war on consumer
perceptions and consumer behavior. In particular, we investigate how a price war affects the
major components of retailer performance: store incidence and basket size (spending per
4
shopper).2 Moreover, does a price war make consumers more responsive to prices? Do its
consequences differ by retailer positioning (on service and value-for-money dimensions), and
does the price war pioneer have an advantage over followers? The current retailer price war in
the Netherlands presents an ideal opportunity to analyze these issues. Based on a data set
consisting of consumer hand scan panel and store perception data from before and after the start
of this national price war, we estimate the impact of pricing policies on incidence, basket size,
price sensitivity and image for the six major Dutch grocery retailers.
The remainder of this paper is organized as follows. The next section discusses the price
war literature, focusing on the gaps this paper aims to address. Next, we discuss the multivariate
Tobit II model to quantify price war effects on store incidence and basket sizes. The subsequent
section describes the empirical setting, and details our data sets. The estimation outcomes are
presented next, and we end by providing a discussion and limitations.
RESEARCH BACKGROUND
Price War: Definition and importance.
Price wars are characterized by competing firms struggling to undercut each other’s
prices (Assael 1990). Likewise, Urbany and Dickson (1991) refer to a “price-cutting
momentum,” the downward price pressure that drives other competitors to follow the initial
move. Unlike typical intense price competition, price wars lead to prices that are not sustainable
in the long run (Schunk 1999). A review of business press articles and academic literature
generated the following definitional conditions of a price war (Heil and Helsen 2001): (1) a
strong focus on competitors instead of on consumers, (2) the pricing interaction as a whole is
undesirable to the competitors, (3) the competitors did neither intend nor expect to ignite a price
2 We use “basket size” and “spending” as synonyms throughout this paper.
5
war, (4) the competitive interaction violates industry norms, (5) the pricing interaction occurs at
much faster rate than normal, (6) the direction of pricing is downward, and (7) the pricing
interplay is not sustainable.
Price wars have become a “fact of life” in a wide range of industries (Rao, Bergen, and
Davis 2000). Business press and academic papers report on price wars in industries ranging from
electricity (Fabra and Toro 2005), oil (Slade 1992), telecom (Young 2004, Rao et al. 2000), cars
(Breshnahan 1987), airlines (Busse 2002, Traça 2004), and fast food (Gayatri 2004) to groceries
(e.g., the Asda-Tesco battle in the UK, Barnes 2004). Price wars erupt at various levels in the
distribution channel, and with growing frequency and intensity (Heil and Helsen 2001). This
leads Rao et al. (2000, p. 116) to conclude that “If you’re not in a battle currently, you probably
will be fairly soon”. Moreover, Heil and Helsen (2001, p.85) observe that the overall effects of
price wars appear to be “seemingly unsurpassed by any other form of competitive exchange”.
Price War Antecedents.
Given the importance of price wars, it is surprising that the academic marketing literature
on price wars is limited to a few publications only. Most price war papers belong to an
economic, often game-theoretic, stream of research. Taken together, the literature does provide a
fairly comprehensive picture of the antecedents of price wars. However, and this is the main
contribution of this paper, there is very little systematic research into the consequences of price
wars (see next subsection).
The antecedents are referred to by Heil and Helsen (2001) as “early warning signals”.
Five types of factors may trigger price wars, or at least facilitate their ignition and affect their
intensity: market, competitive entry, firm, product, and consumer characteristics. Concerning
6
market characteristics, price wars often occur in concentrated or oligopolistic settings, where
firms have tacit collusive agreements (Busse 2002). Outside events may lead one firm to
“undercut” the prevailing price and trigger a chain of reactions away from the tacit agreement.
Outside events triggering a price war may be external shocks in (i) industry demand (e.g.
Rotemberg and Saloner 1986, Slade 1990), (ii) capacity (Fabra and Toro 2005), or (iii) costs
(Rao et al. 2000, Eckert 2004). Demand increases may make it more profitable for firms to (be
the first to) deviate from tacit collusion (Rotemberg and Saloner 1986, Haltiwanger and
Harrington 1991). At the same time, demand decreases or market saturation also cause firms to
abandon collusions (Green and Porter 1984; Tirole 2001, p. 252), leading to price cuts and strong
retaliation (Slade 1990, Eilon 1993).
Competitive entry constitutes another price war trigger. In some cases, the threat of a
price war by the incumbent suffices to deter entry (Milgrom and Roberts 1982, Fudenberg and
Tirole 1986, Elzinga and Mills 1999). In other instances, the price war ignites after newcomers
have actually entered the market, especially if they charge lower prices and gain share. In such a
case, the price war escalation can be explained by Organski (1968)’s power shift paradigm. This
theory states that changes in the power division underlie wars: in order to block a rising
challenger, the market leader launches a preemptive war while it is still powerful. Price is the
likely instrument of choice, as it can be changed fast and easily, and leads to immediate,
measurable results (Kalra, Raju and Srinivasan 1998). This is especially the case if the challenger
charges lower prices, leading firms to over-compete (Griffith and Rust 1997, Leeflang and
Wittink 1996).
Firm characteristics may further contribute to price war eruption. Price wars are more
likely to occur in markets with one renowned “tough” firm but no single price leader (Heil and
7
Helsen 2001) or where price decisions are decentralized (Cyert, Kumar and Williams 1995).
Moreover, companies with high exit barriers (Heil and Helsen 2001), high fixed costs or idle
capacity (Scherer 1980), and high stakes in the market or a worsened financial situation (Busse
2002) are more inclined to initiate a price war or enter an ongoing battle. The nature of the
product, also, signals the likelihood of a price war. Industries with low-differentiation products,
strong assortment overlap between market players, and head-to-head new product competition
constitute more fertile ground for a price battle (Busse 2002, Eilon 1993). Finally, customer
characteristics inductive to price wars are new customer segments, low (and/or declining) brand
loyalty and high (and/or increasing) price sensitivity (Heil and Helsen 2001, Rao et al. 2000,
Sairamesh and Kephart 2000, Klemperer 1989, Eilon 1993).
Price War Consequences
While the antecedents of price wars are well documented, much less is known about their
consequences. The “conventional wisdom” is that price wars affect a diversity of stakeholders,
including suppliers, shareholders, and society as a whole (Rao et al. 2000, Busse 2002) and that,
in the absence of a strong and sustainable cost advantage for the pioneer, price wars are good for
“absolutely nothing” (Henderson 1997).
Companies engaging in a price war do so for a variety of anticipated positive
consequences. Firms may initiate a price drop because they prefer short-term over long-term
profits, hoping to benefit from demand acceleration or from a lag in competitive response
(Bhattacharya 1997, Busse 2002). Conversely, by charging drastically lower prices, firms may
bring about a market clear-out and profit from reduced competition in the long run (Fudenberg
and Tirole 1986, Klemperer 1989, Rao et al. 2000). In many instances, however, the price cut is
8
a necessary defensive move against a competitive threat or a new entrant. Through a sizeable
price reduction, the firm seeks to halt the loss of customers or even to re-attract clientele (Elzinga
and Mills 1999, Klemperer 1989). In addition, the price drop may act as a signal. By indicating
its determination to pursue the battle till the end, the firm may prevent competitors from
engaging in (further) detrimental actions (Klemperer 1989, Besanko and Shanley 1996). At the
same time, a widely advertised price cut may signal to consumers that the firm is not expensive
and establish a more favorable price image (Rao et al. 2000, Busse 2002, Heil and Helsen 2001).
Among the negative consequences of price wars, apart from a reduction in margins, are
undesirable changes in purchase behavior. Consumers may develop unrealistically low reference
prices, shop around more, and come to consider price as the key purchasing criterion (Heil and
Helsen 2001). Note that such increased price sensitivity is especially troublesome for high-end
market players, while it may actually help low-end competitors (Boulding et al. 1994). Indeed,
consumer experiments by Wathieu et al. (2004) show strong evidence for this “price salience”
effect in a brand setting: offering and retracting discounts decreases the subsequent choice share
for high-priced brands, while it increases the choice share of low-priced brands. A similar
argument may apply to retailing. Interestingly, the implied increase in consumers' price
sensitivity may be either temporary (dying out after the “shock” of the initial price moves and
supporting communication), or permanent (in case the price war sets “new price standards,”
causing decreased willingness to pay).
While there is much speculation about these possible consequences, empirical backup is
lacking in the current marketing literature. To the best of our knowledge, not a single empirical
paper has systematically distinguished the impact of a price war on consumers’ company price
image, price sensitivity, firm selection and spending. This is an important gap, since the net
9
outcome for firms involved in a price war hinges upon these (possibly countervailing) effects.
Unfortunately, researchers used to lack the necessary data on consumer perceptions and behavior
before and during the price war. Our data set on the recent Dutch retailing price war allows us to
overcome this hurdle.
Moreover, when studying price war effects, it is instructive to consider differences
between market players. For one, the relative price position of the firm, and the extent to which it
changes by its involvement in the price war, influences whether increased price sensitivity will
help or hurt firm performance. For another, the firm’s role in the price war may be an important
factor (Rao et al. 2000, Elzinga and Mills 1999). As indicated by Busse (2002), price war
initiators possibly enjoy a first mover advantage over followers in regaining a favorable price
image. Our analysis will, therefore, study the price war’s consequences for each of the major
market players, in particular the price war impact on their consumers’ (a) store incidence, (b)
basket size, (c) price sensitivity and (d) store price image.
MODEL
To study the consequences of the price war for national retail chains, we model purchase
behavior of a national panel of Dutch households before and after the start of the price war (more
details on the data are given in the next section). A household faces choices along two
dimensions: which of the stores will be visited (possibly more than one in a given week) and how
much to spend at each store. We develop a model for the incidence decision and spending level
of every household h (h=1, …, H), for every chain i (i=1, ..., S) in every week t (t= 1, …, T).
Given that a household may visit multiple stores in one week, and given the left-censored nature
of household expenditures, we adopt a multivariate Tobit-II model (Fox, Montgomery, and
10
Lodish 2004). The incidence of a store visit of household h for store i in week t ( hitz ) is
described by a probit model:
(1) ��� >
=otherwise 0
0 if 1 *hit
hit
zz ,
In a given week t, a household h may visit multiple stores. Hence hitz equals one for those stores.
The latent variable, *hitz , is modeled through a linear model:
(2) hithithihit uxz ++= ζι '* .
Second, the observational equation for expenditures, hity , is:
(3) ��� =
= otherwise 0
1 if *hithit
hit
zyy ,
where the model for the logarithm of the latent variable, *hity , is
(4) hithithihit vy εωα ++= '* )ln(
The independent variables in the store incidence equation ( hitx ) and spending equation ( hitv )
need not be the same, and their impact is allowed to be different as well. We specify the
independent variables after we have given more details about the data.
The intercepts in equations (2) and (4) capture individual-specific store preferences. We
assume that these intercepts are randomly distributed around store means:
(5) hiihi τψι +=
and
(6) hiihi ξδα += .
Clearly, the stores visited and the amounts spent depend on time and budget constraints,
and are interdependent between stores. Our model allows for this by embedding equations (2)
and (4), across the stores, in a multivariate framework. More specifically, like Fox et al. (2004),
we assume that the each of the error vectors '1 ),...,( hStthht uuu = and '
1 ),...,( hStthht εεε = follow a
11
multivariate normal distribution, with full variance-covariance matrices:
),0(~ ΛMVNuht and ),0(~ ΣMVNhtε . Intrinsic store preferences for patronage and spending
may also be correlated across stores, leading to multivariate normal distributions for the error
terms in (5) and (6) as well: ),0(~ ιτ VMVNh and ),0(~ αξ VMVNh .
Estimation of this hierarchical, multivariate Tobit-II model is possible using MCMC
procedures. Technical details of our approach are given in the Appendix.
THE DUTCH PRICE WAR IN GROCERY RETAILING: SETTING AND DATA
Empirical setting
The igniting spark of the Dutch price war in grocery retailing was hard to miss for Dutch
residents. On October 20th, 2003 the leading grocery retail chain Albert Heijn decided to slash its
prices for more than 1000 products and communicated this in double-page color advertisements
in all national and local newspapers. The ad's headline said “From now on, your daily groceries
are much more inexpensive,” and the ad made clear the chain was committed to decrease its
prices systematically and permanently. The price decreases applied to many national brands from
multiple categories. For example, the price for a 1.5 liter bottle of Coca Cola went down from
�1.23 to �1.12 (−9%).
While Albert' Heijn's operation to decrease prices had been undertaken in complete
secrecy, all major competitors reacted within 2 days without waiting for analysis of consumer
responses. One of these major competitors, Super de Boer, decreased prices for 1000 products,
e.g., it now offered a bottle of Coca Cola for �1.09, down from �1.23. The next week, Albert
Heijn decreased prices for 1500 products. For example, Coca Cola’s price became �1.02. To
further illustrate competitive reactions, Figure 1 shows the average prices charged by the six
12
major chains for non-alcoholic beverages (e.g., sodas and fruit juices). When Albert Heijn cut
prices in week 43 of 2003, competition followed rapidly, leading to a downward trend in prices.
[Insert Figure 1 about here]
The antecedents of this national price war fit Organski’s (1968) “power transition
paradigm”. In Fall 2003, the major Dutch retailer Albert Heijn faced just such a situation: its
leadership and price image position had been declining for years, as hard discounters Aldi and
Lidl3 now reached the majority of Dutch retail consumers. Compared to other West-European
countries, the Dutch retailing industry has low average price levels (an index of 92), but a high
price spread across retailers (from a 75 to a 109 index for national brands). Moreover, the
economic recession enhanced consumer price sensitivity; a major threat for a high price-high
service retailer such as Albert Heijn. On top of that, Albert Heijn had allowed the price gap with
the least expensive supermarket chains to increase from 3% to 16% in the 1993-2003 period. As
a result, discount retailers managed to increase their market share from 13% to 22% between
1996 and 2003, while Albert Heijn’s market share shrunk from 31% to 26% (Van Aalst et al.
2005).
The price war is unprecedented in Dutch retailing. Within one year, seven more price
cutting rounds occurred; involving different brands (national versus private label) and categories,
resulting in negative retail margins for hundreds of products (Van Aalst et al. 2005). As for scope
and breadth, this national price war dwarfs both documented incidents in the grocery industry
mentioned by Heil and Helsen (2001). First, UK retailers Tesco and Asda crossed swords with
price cuts on private labels, leading to “nimbling share prices” (ibid). The second price war
involved the price war between an incumbent in Houston and two market entrants (H.E.B. and
3 Lidl opened many new outlets in the past decade. To illustrate: the average distance from the panelists in our sample to their closest Lidl store decreased from more than 8 to 5.5 km in the period July 2002-May 2004.
13
Food Lions), yielding a 2% drop in food prices (ibid). In our case, the price war is nation-wide,
average prices for national grocery brands drop by 11% (Consumer report 2004), resulting in a
3.9% drop in the consumer price index and the lowest inflation level in 15 years. The scope and
breadth of the price war is also evident from the large number of parties substantially affected.
Smaller suppliers (such as farmers) are going out of business, while large national brand
manufacturers face being dropped from retailer assortments if they refuse to decrease wholesale
prices further (as happened with Venco licorice and Sunil laundry detergent at Edah and Super
De Boer). About 30,000 employees in the grocery industry lost their jobs because of this price
war, while small specialized grocery stores face bankruptcy as they lost over 10% of their
customers, and 33% of their profits. The loss in added value for the Dutch retailing industry is
estimated to be 900M Euros in 1 year, which is supposedly split half-way between the retailers
and their suppliers (van Aalst et al. 2005). As a result, the majority of food executives and
industry analysts argue that these low prices are not sustainable in the long run (ibid) – an
observation typical for price war settings (Heil and Helsen 2001). Other “definitional conditions”
(ibid) that clearly apply to the 2003-2004 Dutch retail market include the strong competitive
focus and the fast downward spiral in prices.
Data Sources
Our dataset combines several sources. First, we use purchase records from the Dutch GfK
consumer hand scan panel across a period of almost three years (July 1, 2001-May 31, 2004).
Panel members scan at home all their purchases at all Dutch grocery retailers, and the data are
sent electronically to GfK Benelux. This GfK panel consists of 4400 households, which
represents a stratified national sample. We use this source to operationalize our dependent
14
variables (store visits and store spending) as well as weekly household- and store specific basket
prices. A unique advantage of consumer hand scan data (over in-store scanned data via
household ID cards) is that the market research agency does not need the permission for data
collection from the retail chains. Such permission is increasingly problematic both in Europe
(typically for the hard discounters) and the US (Walmart).
GfK also provided household perceptions of grocery retailing chains. Every six months,
part of the panelists are surveyed on their perception of price image, produce quality, and waiting
time. Based on these surveys, GfK prepares Christmas and Summer reports for the Dutch
grocery industry. On top of these biyearly reports, GfK conducted an additional survey a few
weeks after the start of the price war. We obtained the store image data at the individual
household level for the same three-year time period, and for each week t we assigned the
perceptions from the measurement moment that is closest to week t. For the households that were
not surveyed for a specific Christmas- or Summer report, we imputed image data using the two-
way linear model - a typical and commonly used best-fit imputation approach (see Little and
Rubin 1987, Chapter 2).
We obtained data from IRI and Publi Info (both in the Netherlands) on weekly feature
and display for all items sold in Dutch grocery retailing chains across the same four-year period.
We used these variables to operationalize household- and store-specific feature- and display
variables. Finally, Reed Business provided us with the sizes (in square meters) and the locations
(zip codes) for all Dutch grocery stores and each year in our data set. The former data is a useful
proxy for assortment size of each chain's store nearest to the household. The latter data,
combined with the zipcodes of all household panelists from GfK, allows us to compute the
Euclidean distance between a household and the closest store from each chain.
15
Data selection
We model incidence and expenditures at the six largest chains with national coverage.
Figure 2 summarizes the store perception data in two main dimensions (according to GfK);
service and value for money, to illustrate the positioning of these chains prior to the price war.
Albert Heijn is the market leader who initiated the price war. It is a high-price, high-service
chain, which is illustrated by its scores on price image, produce quality and waiting time (see
Table 1). Super de Boer is located in the same high-price, high-service segment. In the middle
segment there are two chains as well: C1000, with good scores for service and value, and Edah,
with low scores on both dimensions. The two hard discounters (low price, low service) are Aldi
and Lidl.
[Insert Table 1 and Figure 2 about here]
Since the panel composition changes over time, we decided to select the 1885 households
that remained in the panel across the three-year period. We use the first 26 weeks of this period
(Week 27 of 2001 till week 52 of 2001) as the initialization period for determining households’
expenditures across categories. The final 126 weeks (Week 1 of 2002 – Week 22 of 2004) are
used for model calibration. The price war started in week 43 of 2003, and hence we have 94
weeks before the price war start and 32 weeks afterwards. The full data set consists of 1,425,060
observations (purchases of 1885 households at 6 retail chains over 126 weeks).
Independent variables
Store selection and spending depend on a trade-off between various shopping benefits
and costs (see, e.g., Bell, Ho, and Tang 1998, Tang, Bell, and Ho 2001). We show in Table 2 the
independent variables representing benefits and costs included in our models. Key drivers of
store benefits include a favorable store price image, a high level of product quality (often
16
signaled by top-quality produce), a large assortment, and appealing promotional offers (Sirohi et
al. 1998, Bell et al. 1998, Tang et al. 2001, Fox et al. 2004). Store familiarity or spending habits
have been found to affect store patronage and basket size as well (Bell et al. 1998, Rhee and Bell
2002). Such state dependence in households’ shopping behavior can be captured using lagged
purchase indicators (Seetharaman 2003), such as, for instance, lagged expenditures in the store.
[Insert Table 2 about here]
Shopping costs encompass (i) transaction or fixed shopping costs, which increase with
store distance and waiting time at the checkout, as well as (ii) variable costs, i.e., prices paid to
acquire the products (see, e.g., Bawa and Ghosh 1999, Bell et al. 1998, Popkowski-Leszczyc,
Sinha, and Sahgal 2004). We operationalize prices (and feature and display) relative to the
market average (see Table 2) in order to capture both own- and cross-effects. Cross-store effects
are very likely in the Netherlands, a highly urbanized country with a very high supermarket
density. In addition, consumers’ shopping behavior may be shaped by seasonal factors, including
special events or holiday periods. Last but not least, in line with the focal interest of this paper,
we include several price war variables. Based on the previous sections, we allow for a separate
impact of the price war on each of the chains’ intercepts, and this for store incidence as well as
basket size. We also test whether the price war affects consumer’s price sensitivity for both
decisions, either permanently (using the interaction between log price and the price war step
dummy) and/or temporarily (by multiplying this interaction with an inverted trend (1/ �) where �
starts to count at one at the first week of the price war, see Table 2).
We use the same independent variables in the store incidence equation ( hitx ) and
spending equation ( hitv ), except for two variables. We drop “display” from the store incidence
equation, because this marketing instrument is only observed by shoppers in the store. Likewise,
17
we drop feature promotions from the spending equation, since feature promotions represent out-
of-store communication intended to enhance store incidence. We verified both restrictions, and
found that display does not affect store incidence significantly, and feature does not affect
spending significantly. E
RESULTS
Store incidence
We present the store incidence results in the left-hand part of Table 3. Except for
perceived Produce Quality - the impact of which is not significant - the “benefit” variables
(StoreSurface, Feature and LagExpend) have positive and significant effects on store incidence.
The positive impact of StoreSurface (.480) is consistent with store size being a proxy for
assortment size, while the coefficient of lagged expenditures (.051) indicates positive state
dependence. As for “costs”, we find that a higher Distance between a household and a store
(which means: more travel time and costs) has the expected negative effect on store incidence
(−.104). The impact of perceived Waiting Time on store incidence is not significant, while the
effect of price is found to be significantly positive (.123), which may seem counterintuitive.
Similar positive relationships between price and store incidence were also observed by Fox et al.
(2004). One possible explanation is that higher-priced stores are frequently visited for small
basket trips and fill-in trips (see, e.g, Lal and Rao 1997, Bell and Lattin 1998, Popkowski-
Leszczyc et al. 2004), which is consistent with reported consumer behavior in the Netherlands
(GfK 2003). Another explanation may be that high-end stores have a loyal consumer base, which
visits the store whether or not prices are high in any given week. The seasonal effect estimates
indicate a decreased propensity to visit grocery stores in the Christmas week (Week52: −.188)
and in the first week of the year (Week1: −.789), possibly because stores limit their opening
hours (grocery stores are closed on December 25 and 26, and January 1), and consumers prefer
18
to stay at home with family and friends. In the Easter week there is an increase in store visit
propensity (.079), plausibly because consumers want to shop for holiday meals, and the longer
opening hours (relative to Christmas) allow them to do so.
[Insert Table 3 about here]
Focusing on the impact of the price war variables, several interesting findings emerge.
First, Albert Heijn did not manage to increase its store incidence propensity, as the coefficient of
the PW*AlbertHeijn variable is insignificant. However, the price war did lead to significantly
negative intercept effects for two competitors, Aldi (−.118) and Edah (−.083). Interestingly, one
competitor (C1000) experienced a significant increase in store incidence propensity after the start
of the price war (.044). Note in Figure 2 that C1000 is perceived to offer good service and value-
for-money. Therefore, once price-sensitive consumers are induced to check out traditional
grocery chains again, C1000 is a logical choice. In Keller et al.’s (2002) conceptual framework,
C1000’s good service provides a point of parity with high-service competitors, while its good
value-for-money provides a desirable and deliverable point of difference. Second, as of the start
of the price war, the price coefficient becomes significantly lower, part of which is permanent
−.355), and part of which is temporary (−.121). This empirical finding corroborates the
prediction of Busse (2002) and Heil and Helsen (2001) that price wars increase the price
sensitivity of consumers.
Expenditures
The estimates for the log (ln) of expenditures equation are given in right-hand part of
Table 3. All the “benefit” variables have the expected positive and significant effects. Spending
increases with PriceImage (.015), ProduceQuality (.005) and Display (.005). Moreover, it
increases with StoreSurface (.340); consistent with the notion that larger assortments allow the
19
fulfillment of more consumer needs, and lagged expenditures (.005); consistent with positive
state dependence. On the “cost” side, a longer distance to the store leads to significantly lower
expenditures (−.007). This may be true either because transportation from the store to home by
foot or bike (which are not uncommon in the Netherlands) becomes increasingly difficult for
larger basket sizes, or because consumers visit these “far away” stores for fill-in trips on their
way home from work. Surprisingly, we find that a worse perceived WaitingTime is associated
with higher expenditures (−.004). Our tentative explanation is that some consumers use hard
discounters to stockpile large quantities. They are willing to wait longer in line at discounters
because they economize a lot by buying here. As a result, we observe a negative correlation
between perceived waiting time and expenditures. The elasticity of expenditures to price is
virtually zero and insignificant (−.051). This implies that the elasticity of quantity to price is
minus one, and hence demand for groceries are neither price elastic nor price inelastic before the
price war. As for seasonalities, the effects of the pre-Christmas-week (Week51: .109), the
Chrismas week (Week52: .024) and the Easter week (.139) on ln expenditures are positive,
whereas the effect of the year's first week on expenditures is negative (Week1: −.359), possibly
due to consumer economizing or dieting.
Again, the price war variables reveal some striking results. Ceteris paribus, the price war
causes significant decreases in ln expenditures for both the price war initiator (PW*Albert Heijn:
−.049) and three followers (PW*C1000: −.038; PW*Edah: −.095; PW*Super de Boer: −.058). In
other words, the price war does not increase sales volume enough to compensate for the large
price drops, consistent with the notion that the grocery market is not that price elastic in the
Netherlands. The effects of the price war on expenditures at the hard discounters (PW*Aldi;
PW*Lidl) are insignificant. Interestingly, the coefficient of PW*lnPrice*(1/�) indicates that, after
20
the start of the price war, the elasticity of expenditures to price temporarily becomes more
negative (−.085). This finding is again consistent with Busse (2002) and Heil and Helsen (2001).
However, although the permanent effect (−.107) is negative as expected, it is not significant.
Hence, while the price war causes an enduring change in sensitivity to price changes for store
incidence, no such lasting effect occurs for basket size. After a dust settling period, consumers’
expenditures – once they are in the store - are not systematically more dependent on price than
before the price war started.
Decomposing the net impact of price war on incidence and expenditures
It is important to note that the price war affects multiple aspects in the models for store
incidence and expenditures. First, there is a direct impact of the price war on the store intercepts,
captured by terms such as PW*AlbertHeijn, PW*Aldi, etc. Second, the price war may affect the
price coefficient via the value of the relative price variable (lnPrice). To illustrate that this
variable remains relatively stable, we show in the top panel of Figure 3 the over-time pattern of
the relative price (i.e. exp(lnPrice), averaged across households) for each of the six chains. Since
the chains have followed each other’s price moves rapidly, there is no clearly visible change in
relative prices after the start of the price war (week 43 of 2003).
Third, the price war affects the price coefficients in the models for store incidence and
expenditures permanently via the term PW*lnPrice. Since we are interested in long-term changes
only, we exclude the temporary change captured by PW*lnPrice*(1/�), since this term
approaches zero as � → �. Fourth, the price war affects the price image of some chains as is
evident from the lower panel of Figure 3. In particular, the price war initiator, Albert Heijn,
experiences a major improvement of its price image after the start of the price war, although its
21
actual relative price position has not really changed (see top of Figure 3). Since price image is
included as a separate variable in the model, we need to incorporate the impact of its change as a
fourth component in our assessment of the net price war effect.
[Insert Figure 3 about here]
To quantify the net effect across these four components, we proceed as follows. We
consider the quarter before the price war start as the pre-war period. For the post-start period, we
take the quarter that begins in the first new measurement of the price image (week 49 of 2003).
For both periods and for each store, we compute the average ln Price and PriceImage variables,
and substitute those in the part of the store incidence and ln expenditures model that we label as
the “Price Part”(where, like before, the subscript i is a store indicator):
(7) }Image Price{}lnPrice*PW{}lnPrice{PW}{Part Price 43210 iiiiii βββββ ++++= .
Comparing these Price Parts in the pre-war and post-war period then allows us to assess the net
price war effect.
Figure 4 displays the pre- and post price war-start Price Parts for each of the six chains.
Table 4 decomposes these Price Parts changes into the three underlying components (bracketed
terms in equation (7): (i) the baseline shift, (ii) the shift due to change in relative price levels, (iii)
the shift due to change in price coefficient times relative price level, and (iv) the shift due to
price image changes. For the price war initiator Albert Heijn, we observe that the price war has a
negative effect on the Price Part of the incidence model. What happens is that Albert Heijn’s
relative price position (lnPrice) changes only marginally after the start of the price war, and
hence there is no gain from the term lnPrice2β (−1% contribution to the net change). However,
the increased price sensitivity makes the term lnPrice*PW3β very negative (−167%
contribution), which is not compensated by the baseline change PWi1β (+53%) or by the change
22
in the price image component Image Price4β (+15%). Albert Heijn also experiences a net
decrease in the price part for the expenditures model, which is composed of an intercept decrease
(−82%) and a decrease due to increased price sensitivity (−28%).
[Insert Figure 4 and Table 4 about here]
The two hard discounters, Aldi and Lidl, are the clear winners of the price war. The price
war enhances the Price Parts of both their incidence specification and their ln expenditures
model. Table 4 shows that for both firms and both models, the increased price sensitivity more
than compensates the loss in intercepts. Aldi and Lidl benefit from the increased price sensitivity,
since they have been able to maintain their very low price position after the price war started to
escalate (see Figure 3).4
Besides Albert Heijn, also Edah and Super de Boer experience net losses in incidence and
expenditures due to the price war. The decomposition in Table 4 shows that Edah primarily loses
business due to intercept reductions; its average price position prevents losses due to increased
price sensitivity. Super de Boer, on the other hand, remains a high-priced chain (see Figure 3),
and as a result, significant proportions of its net reductions in the price parts for both components
are attributable to the increased price sensitivity. Finally, C1000 experiences a net price war
benefit in the price part of the store incidence equation yet a decrease in the expenditure
equation. C1000’s baseline increase in incidence (+70%) is further enhanced by a gain due to
increased price sensitivity that favors its relatively low price level (+31%). However, for the
expenditure equation, C1000’s benefit from the increased price sensitivity (+49%) does not
compensate the change attributable to the baseline loss (−164%).
4 In the calculation of the Price Parts, we use all parameters irrespective of significance levels. If we were to set the insignificant permanent price sensitivity change in the expenditure model equal to zero, only one substantial conclusion changes: the expenditures at Aldi and Lidl stay the same.
23
DISCUSSION
In conclusion, price wars do not appear to be good “for absolutely nothing” (Henderson,
1997). Our analysis reveals that price wars trigger several - possibly countervailing - effects, and
that their net impact varies by market player. By decomposing the net price war effect into
several underlying components, this study improves our understanding of what might motivate
firms to engage in a price war, as well as why they often fail.
We show that the price war initiator Albert Heijn succeeds in improving its price image
without hurting its quality and service image5. This is consistent with the perspective that Albert
Heijn’s move was initially meant as a market correction. The retailer simply allowed the price
gap to get too large in the period 2001 till Fall 2003, exactly during the time that consumers
became more price conscious due to the deteriorating economic conditions. Despite their
continued belief in the retailer’s quality and service, fewer and fewer shoppers could justify
paying such higher prices. In such a situation, Albert Heijn’s “correcting” price move may have
been the only approach left to re-engage customers (Chen and McMillan 1992, p. 563). The
“market correction” view also appears consistent with the perceptual data: the price war boosts
Albert Heijn’s price image - after a 2-year slide. Therefore, rich perceptual data can help
companies (and researchers) to monitor changes in consumer associations driving retailer choice,
and thus complement actual purchase information.
Our result that the high-service pioneer may benefit from a price war by improving its
price image, is also consistent with the power transition paradigm (Organski 1968). Instead of
simply reminding shoppers about its great quality and service (which apparently no longer
5 A separate analysis showed that these components did not decrease significantly after the start of the price war.
24
suffices), the market leader launches a pre-emptive strike while it is still powerful, i.e., before all
shoppers have lost interest. Price is the logical weapon of choice: it is easy to change fast and
directly addresses the pioneer’s perceived weakness compared to the hard discounters (Kalra et
al. 1998). In other words, quality and service may be valuable points of difference for high-
service retailers in competition with hard discounters, but low prices increasingly become a point
of parity in times of demanding consumers who want “more for less” (Keller, Sternthal and
Tybout 2002): In general, “points of differentiation are no longer enough to sustain a brand
against competitors” (p. 82). Two criteria for these points are desirability (which is clearly the
case for low price in retailing) and deliverability. There is the rub for high-service retailers:
convincing shoppers they can consistently deliver low prices while maintaining and building
desirable differentiation on service and quality.
In our empirical setting, deliverability turned out to be a major problem for the pioneer
Albert Heijn. Their aim to “become less expensive than the market average,” which was
massively announced in the press, entailed dramatic consequences since the competition reacted
rapidly and fiercely. Heil and Helsen (2001) observe that many price wars ignite unintended,
since firms often seem to lack an understanding of the market and industry conditions leading to
a price war. Despite that “even the smallest amount of look-ahead in agents’ pricing algorithms
can significantly reduce or eliminate the occurrence of price wars” (Tesauro and Kephart 2000),
Albert Heijn may have overlooked that its aim to improve its relative price position would
strongly depend on the extent to which the competition would allow them to do so. And, as it
appeared, the competition did not allow much leeway.
Our results strongly suggest, however, that it is not just incorrect anticipation of
competitive moves that turned the price war initiated by Albert Heijn into a “Pyrrhic Victory” at
25
best. Rather, it is the lack of improvement in relative price position combined with increased
consumer price sensitivity that ultimately lead to lower store incidence and decreased basket
sizes. Our study provides factual support for the observation that price wars may dramatically
increase consumers’ focus on price (Busse 2002, Heil and Helsen 2001), both in their firm
patronage and spending decisions. We show that insufficient perspective of these changes in
consumer response may entail disastrous consequences for the pioneer. Our results therefore
extend the experimental findings by Wathieu et al. (2004) from brand to retail competition. Our
study provides external validity for cautioning high-end market players not to focus on price as a
competitive weapon: any short-term gain may be more than offset by increased price sensitivity.
However, for the price war followers, fortunes are mixed as well. The high-service
follower, most similar to the pioneer, also gets hit with lower incidence and basket sizes. Its
lower prices appear simply to subsidize existing customers, while it hardly enjoys an improved
price image, like the pioneer does. The same holds for the low service “middle-of-the-market”
competitor. In contrast, the higher service-higher value for money competitor succeeds in
increasing store incidence, but at the expense of lower basket sizes. In other words, the company
“subsidizes” its broadening of customer base. As this firm is perceived to offer good service and
value-for-money, it does not suffer from higher consumer price sensitivity. Finally, both hard
discounters strongly benefit from the increased sensitivity to price, and they manage to continue
their growth in incidence and basket size.
Our findings corroborate Rao et al.’s claim that it is unwise for a player without an
obvious cost advantage to engage in a price war (Rao et al. 2000). Similarly, “Companies think
they can compete on price even though they do not have a true cost advantage. … Cutting prices
to gain market share -as opposed to doing it because of a cost advantage- can often permanently
26
hurt profits and revenues” (Henderson 1997, p. 156). Although we do not have data on the cost
structure of the price-war initiator, its high-service profile seems to conflict with cost leadership.
High-service companies should focus on the use of non-price instruments (Rao et al. 2000) such
as loyalty cards (Busse 2002) to reduce competition among retailers, and create “win-win
situations” (Brandenburger and Nalebuff 1996).
This study has several limitations, providing leads for future research. First, our data
come from one price war in one country, and further studies are needed to establish whether our
findings generalize to other price war situations. In particular, we cannot ascertain to what extent
Albert Heijn’s service leadership drives the pioneering benefits (in price image) we observe.
Would second-tier retailers obtain similar benefits from initiating a price war? Moreover, other -
better - options may be open to high-service retailers to counter the hard-discount challenge; we
simply did not observe them in our data. Second, we do not have retailer (or manufacturer)
margin data that would allow us to directly assess the profitability consequences of the price war.
Third, while our models account for preference heterogeneity, they do no allow for response
heterogeneity since we are interested in overall effects. Previous studies suggest that accounting
for response heterogeneity does not affect overall response parameters significantly (see, e.g.,
Bijmolt, Van Heerde, and Pieters 2005’s recent meta-analysis of price elasticity), while it only
marginally improves overall fit and predictive validity, and may render estimation results
unstable (Ailawadi, Neslin and Gedenk, 1999). For this reason, Ailawadi et al (1999, p 195)
recommend that “response heterogeneity be included only if it is of interest to the modeler per
se”. In order to accommodate response heterogeneity, we would have to expand the current
model considerably, and model estimation already takes 17 days on a special computer server.
Fourth, we model competitive reactions only through their effect on relative price (the publicized
27
focus of the price initiator and followers), which implicitly assumes that each competitor has a
similar effect on the performance of the focal chain. Again, while future research may analyze
how each competitor’s price has a different impact, incorporating explicit cross-price effects
would greatly complicate the already strenuous model estimation. Finally, we model prices as
exogenous vis-à-vis household decisions on incidence and expenditures, as it seems much more
likely that chains base their prices on competition than on unobserved individual-level demand
shocks (see Erdem, Imai, and Keane 2003 for a similar argument). Incorporating endogenous
prices and complicated feedback loops, for instance in vector-autoregressive models of aggregate
data (e.g. Pauwels et al. 2002), would allow quantitative analysis of the antecedents and
momentum of a price war, a fascinating topic for future research.
Despite these limitations, our analysis of the Dutch retailing price war generates at least
one strong recommendation: for Albert Heijn to call a cease fire. Indeed, the pioneer’s goal of
restoring its price image was achieved within a few weeks, while the stated objective of
“becoming less expensive than the market average” remains elusive even after a year of
bloodshed. It is highly likely that a reconciliatory move would receive warm competitor
reception: during the price war, competitors appear to only match, not to amplify the pioneer’s
price reductions, which signals their commitment to defend their position without escalating the
situation (Chen and McMillan 1992). As business press reports estimate the one-year added-
value loss at �900M for retailers and suppliers, many employees lost their jobs and smaller
companies are on the verge of bankruptcy, we believe a cease fire would be welcomed by
retailers and manufacturers alike.
28
References
Ailawadi, Kusum L. (2001), “The Retailer Power-Performance Conundrum: What Have We Learned?,” Journal of Retailing, 77 (3), 299-318.
Ailawadi, Kusum L., Karen Gedenk and Scott Neslin (1999), “Heterogeneity and Purchase Event Feedback in Choice Models: an Empirical Analysis with Implications for Model Building,” International Journal of Research in Marketing, 16 (3), 177-198.
Assael, Henry (1990). Marketing, Englewood cliffs, NJ: Prentice Hall. Barnes, Rachel (2004), “Tesco Complains to ASA over Asda Low-Price Ads,” Marketing, August 11, 5. Bawa, Kapil and Avijit Ghosh (1999), “A Model of Household Grocery Shopping Behavior,” Marketing
Letters, 10 (2), 149-160. Bell, David and James Lattin (1998), “Shopping Behavior and Consumer Preference for Store Price
Format: Why “Large Basket” Shoppers Prefer EDLP,” Marketing Science, 17 (1), 66-88. Bell, David, Teck-Hua Ho and Christopher S. Tang (1998), “Determining Where to Shop: Fixed and
Variable Costs of Shopping,” Journal of Marketing Research, 35 (August), 352-369. Besanko, David, David Dranove, and Mark Shanley (1996), Economics of Strategy, New York: Wiley. Bhattacharya, R. (1996), “Bankruptcy and Price Wars,” Working Paper, University of Melbourne. Bijmolt, Tammo H.A., Harald J. Van Heerde, and Rik G.M. Pieters (2005), “New Empirical
Generalizations on the Determinants of Price Elasticity,” Journal of Marketing Research, 42 (May), 141-156.
Boulding, William, Eunkyu Lee and Richard Staelin (1994), “Mastering the Mix: Do Advertising, Promotion, and Sales Force Activities Lead to Differentiation?,” Journal of Marketing Research, 31 (May), 159-172.
Brandenburger, Adam M. and Barry J. Nalebuff 1996, “Co-opetition”, New York: Doubleday. Breshnahan, Timothy F. (1987), “Competition and Collusion in the American Automobile Industry: the
1955 Price War,” The Journal of Industrial Economics, 35 (4), 457-482. Business Week (2003), “Wal-Mart: Too Powerful?,” October 6, 46-55. Busse, Meghan R. (2002), “Firm Financial Conditions and Airline Price Wars,” RAND Journal of
Economics, 33 (2), 298-318. Chen, Ming-Jer and Ian C. MacMillan (1992), “Nonresponse and Delayed Response to Competitive
Moves: the Roles of Competitor Dependence and Action Irreversibility,” Academy of Management Journal, 35 (3), 539-570.
Consumer Report (Consumentengids) (2004), “Consument Wint Eerste Slag,”March, p. 25. Cyert, Richard, Praveen Kumar, and Jeff Williams. (1995), “Impact of Organizational Structure on
Oligopolistic Pricing,” Journal of Economic Behavior and Organization, 26 (1), 1-15. Eckert, Andrew (2004), “An Alternating-Move Price-Setting Duopoly Model with Stochastic Costs,”
International Journal of Industrial Organization, 22 (7), 997-1015. Eilon, Samuel (1993), “The Effect of a Price War in a Duopoly,” International Journal of Management
Science, 21 (6), 619-627.
29
Elzinga, Kenneth G. and David E. Mills (1999), “Price Wars Triggered by Entry,” International Journal of Industrial Organization, 17 (2), 179-198.
Erdem, Tülin, Susumu Imai, and Michael P. Keane (2003), “Brand and Quantity Choice Dynamic Under Price Uncertainty,” Quantitative Marketing and Economics, 1 (1), 5-64.
Fabra, Natalia and Juan Toro (2005), “Price Wars and Collusion in the Spanish Electricity Market,” International Journal of Industrial Organization, 23 (3-4), 155-181.
Fox, Edward J., Alan L. Montgomery, and Leonard M. Lodish (2004), “Consumer Shopping and Spending across Retail Formats,” Journal of Business, 77 (2), S25-S60.
Fudenberg, Drew and Jean Tirole (1986), “A Theory of Exit in Duopoly,” Econometrica, 54 (4), 943-960. Gayatri, D. (2004), “Burger King: Revitalizing the Brand,” ECCH Collection, ICFAI Business School
Case Development Centre, Hyderabad, India. Gfk (2003), Jaargids 2003, Growth from Knowledge, The Netherlands. Green, Edward J. and Robert H. Porter (1984), “Noncooperative Collusion under Imperfect Price
Information,” Econometrica 52 (1), 87-100. Griffith, David E. and Roland Rust (1997), “The Price of Competitiveness in Competitive Pricing,”
Journal of the Academy of Marketing Science, 25 (2), 109-116. Haltiwanger, John and Joseph E. Harrington (1991), “The Impact of Cyclical Demand Movements on
Collusive Behavior,” RAND Journal of Economics, 22 (1), 89-106. Heil, Oliver P. and Kristiaan Helsen (2001), “Toward an Understanding of Price Wars: Their Nature and
How They Erupt,” International Journal of Research in Marketing, 18 (1-2), 83-98. Henderson, David R. (1997), “What Are Price Wars Good For? Absolutely Nothing,” Fortune, 135 (9),
156. Kalra, Ajay, Surendra Raju, and Kannan Srinivasan (1998), “Response to Competitive Entry: A Rationale
for Delayed Defensive Reaction,” Marketing Science, 17 (4), 380-405. Keller, Kevin Lane, Brian Sternthal and Alice M. Tybout (2002), “Three Questions You Need to Ask
About Your Brand,” Harvard Business Review, 80 (9), 80-86. Klemperer, Paul (1989), “Price Wars Caused by Switching Costs,” Review of Economic Studies, 56 (3),
405–420.
Lal Rajiv and Ram Rao (1997), “Supermarket Competition: the Case of Every Day Low Pricing,” Marketing Science, 16 (1), 60-80.
Leeflang, Peter S.H. and Dick R. Wittink (1996), “Competitive Reaction versus Consumer Response: Do Managers Overreact?,” International Journal of Research in Marketing, 13 (2), 103-121.
Little, Roderick J. A and Donald B. Rubin (1987), Statistical Analysis with Missing Data, New York: John Wiley.
Milgrom, Paul and John Roberts (1982), “Predation, Reputation, and Entry Deterrence,” Journal of Economic Theory, 27 (2), 280-312.
Organski, A.F.K. (1968). World Politics, New York: Knopf.
30
Pauwels, Koen, Dominique Hanssens, and S. Siddarth (2002), “The Long-Term Effects of Price Promotions on Category Incidence, Brand Choice and Purchase Quantity,” Journal of Marketing
Research, 39 (November), 421-439. Popkowski-Lesczyc, Peter T.L, Ashish Sinha and Anna Sahgal (2004), “The Effect of Multi-Purpose
Shopping on Pricing and Location Strategy for Grocery Stores,” Journal of Retailing, 80 (2), 85-99.
Rao, Akshay R., Mark E. Bergen and Scott Davis (2000), “How to Fight a Price War,” Harvard Business Review, 78 (2), 107-117.
Rhee, Hongjai and David R. Bell (2002), “The Inter-Store Mobility of Supermarket Shoppers,” Journal of Retailing, 78 (4), 225-237.
Rotemberg, Julio J. and Garth Saloner (1986), “A Supergame-Theoretic Model of Business Cycles and Price Wars During Booms,” American Economic Review, 76 (3), 390–407.
Sairamesh, Jakka and Jeffrey O. Kephart (2000), “Price Dynamics and Quality in Information Markets,” Decision Support Systems, 28 (1-2), 35-47.
Scherer, F.M. (1980). Industrial Market Structure and Economic Performance, 2nd Ed., Chicago: Rand McNally College Publishing Co.
Schunk, H. (1999), “The Evolution of Competitive Interaction,” Working Paper, University of Mainz. Seetharaman, P.B. (2003), “Modeling Multiple Sources of State Dependence in Random Utility Models:
A Distributed Lag Approach,” Marketing Science, 23 (2), 263-271. Sirohi, Niren, Edward W. McLaughlin, and Dick R. Wittink (1998), “A Model of Consumer Perceptions
and Store Loyalty Intentions for a Supermarket Retailer,” Journal of Retailing, 74 (2), 223-245. Slade, Margaret E. (1990), “Strategic Pricing Models and Interpretation of Price-War Data,” European
Economic Review, 34 (2-3), 524-537. Slade, Margaret E. (1992), “Vancouver’s Gasoline-Price Wars: An Empirical Exercise in Uncovering
Supergame Strategies,” Review of Economic Studies, 59 (April), 257-276. Tang Christopher S., David Bell and Teck-Hua Ho (2001), “Store Choice and Shopping Behavior: How
Price Format Works, ” California Management Review, 43 (2), 56-74. Tesauro, Gerald J. and Heffrey O. Kephart (2000), “Foresight-Based Pricing Algorithms in Agent
Economies,” Decision Support Systems, 28 (1-2), 49-60. Tirole, Jean (2001). The Theory of Industrial Organization. Cambridge, MA: MIT Press. Traça, Daniel (2004), “Virgin Blue: Fighting with National Champions,” INSEAD Case 204-032-1. Urbany, Joel E., and Peter R. Dickson (1991), “Competitive Price-Cutting Momentum and Pricing
Reactions,” Marketing Letters, 2 (4), 393-402. Van Aalst, Marcel, Laurens Sloot, Leo van der Blom, and Leo Kivits (2005), Het ‘Grote Voordeel’ van
één Jaar Prijsoorlog, Erasmus Food Management Instituut 2005-01, January, ISBN 90-77015-18-3. Wathieu, Luc, A.V. Muthukrshnan and Bart J. Bronnenberg (2004), “The Asymmetric Effect of Discount
Retraction on Subsequent Choice”, Journal of Consumer Research, 31 (4), 652-657. Young, Shawn (2004), “MCI Offer Shows Price War Persists In Long Distance,” Wall Street Journal,
May 14.
31
Appendix: MCMC Estimation of a Hierarchical Multivariate Type-2 Tobit Model6
We stack (1) the dependent variables of equations (2) and (4) for all households h and time
periods t so that ],,,[ **21
*11
* ′= HiTiii yyyy � and ],,,[ **21
*11
* ′= HiTiii zzzz � , (2) the error terms of these
equations for all households h and time periods t so that εi = [ ε1i1,ε1i2,…,εHiT ]′ and ui = [ u1i1,u1i2,…,uHiT ]′,
(3) the predictor variables for the store incidence equation, Vi = [ v1i1,v1i2,…,vHiT ]′, and the predictor
variables for the ln expenditures equation, Xi = [ x1i1,x1i2,…,xHiT ]′, We allow for contemporaneous
correlation of the error terms in equations (2) and (4) by adopting the SUR forms shown below.
����
�
�
����
�
+
����
�
�
����
�
=
�����
�
�
�����
�
����
�
�
����
�
+
����
�
�
����
�
=
�����
�
�
�����
�
S
2
1
2
1
2
1
S
2
1
2
1
2
1
,
u
u
u
X
X
X
z
z
z
V
V
V
y
y
y
S*S
*
*
S*S
*
*
����ζ
ε
εε
ω
where, for {i = 1, …, S}:
εi is an HT vector of disturbances such that E(εi) = 0 and E(εiε′j) = σijIHT, and
ui is an HT vector of disturbances such that E(ui) = 0 and E(uiu′j) = λijIHT,
with Σ =
����
�
�
����
�
=Λ
����
�
�
����
�
SSSS
S
S
SSSS
S
S
���
���
���
�
����
�
�
�
����
�
�
21
22221
11211
21
22221
11211
,
σσσ
σσσσσσ
Finally, for clarity we rewrite the SUR equations above as: y* =Vω + ε and z* = Xζ + u.
Next we specify the hierarchies associated with the two shopping decisions. We stack (i) the
intercept coefficients for all households h so that ααααi = [α1i,α2i,…,αHi ]′ and ιιιιi = [ι1i,ι2i,…,ιHi ]′ and (ii) the
error terms of the hierarchical equations for all households h, ξξξξi = [ ξ1i ,ξ2i ,…, ξHi ]′, and ττττi = [τ1i,τ2i,…,
τHi]′. We allow for contemporaneous correlation of unexplained preferences across store chains in
equations (5) and (6) by using the SUR forms below.
6 This appendix is largely based on Fox et al. (2004).
32
����
�
�
����
�
+
����
�
�
����
�
����
�
�
����
�
=
�����
�
�
�����
�
����
�
�
����
�
+
����
�
�
����
�
����
�
�
����
�
=
�����
�
�
�����
�
SSH
H
H
SSSH
H
H
S �
�
�
�
�
�
�
�
�
�
�
�
�
�
�
�
�
�
��������
2
1
2
1
2
1
2
1
2
1
2
1
,
1
11
1
11
where 1H is a H by 1 vector of ones, ξi is an H by 1 vector of disturbances such that E(ξi) = 0 and E(ξiξ′j)
= Hij Iαν and τi is an H by 1 vector of disturbances such that E(τi) = 0 and E(τiτ′j) = Hij Iιν ,
with Vα =
�����
�
�
�����
�
=
�����
�
�
�����
�
���
���
���
�
���
���
���
SSSS
S
S
SSSS
S
S
vvv
vvv
vvv
V,
vvv
vvv
vvv
�
����
�
�
�
����
�
�
21
22221
11211
21
22221
11211
We summarize the SUR equations above as α = Wδ + ξ and ι = Wψ + τ, where W HSI 1⊗= .
We use an MCMC approach to estimate the marginal distributions of the latent dependent
variables, parameters and covariances. The MCMC algorithm involves sampling sequentially from the
relevant conditional distributions over a large number of iterations. These draws can be shown to
converge to the marginal posterior distributions. Our implementation of the MCMC algorithm has twelve
steps that are described below.
Conditional distributions
The first implementation step requires that we specify conditional distributions of the relevant
variables. The solutions of these distributions follow from the normality assumption of the disturbances
terms. We employ natural conjugate priors. Specifications of the conditional distributions are as follows:
1. *hity is yhit if yhit > 0, otherwise *
hity is drawn from a normal distribution.
( )�
��
�−−�++′
>�
−≠≠
≠otherwiseE(
0~,,,
11hi)��,�)(yy��vN
yyyy
jijjijii*
i,th,j*
i,th,j-jjijhihit
hithit*i,th,j
*hit αω
where���
�
�
���
�
�
−−+−−=�
���
�
�
���
�
−−=
≠ jjji
ijii
*i,th,j
*hit
*ht
�
��
y
y
y and
33
As the notation suggests, the *hty vector and Σ matrix are partitioned between the store chain of
interest, i, and all other store chains, j ≠ i. Without loss of generality, we have assumed the store chain
of interest to be the first. Each chain is then drawn in succession for household h, conditioning on
*j,th,iy ≠ , a vector of latent dependent variables for all j ≠ i, and �.
2. We next draw the latent dependent variable values for the probit component of the model. If the
indicator variable zhit = 1, then *hitz is drawn from a normal distribution, truncated below at 0.
Otherwise, *hitz is drawn from a normal distribution, truncated above at 0.
( )( )jijjijii*
i,th,j*
i,th,jjjijhihitThi*
i,th,j*hit ��,�)(zz���xN,,�,zz 11 E~ −
≠≠−
≠ −−++ζ
where: ���
�
�
���
�
−−+−−=���
�
�
���
�
−−=
≠ jjji
ijii
*i,th,j
*hit
*ht
�
��
z
z
z and
The latent probit dependent variables are drawn using the inverse cdf method.
3. The parameters in ω are drawn from a SUR model with variance/covariance matrix of disturbances �.
( )( )( ),OyIVO~N,y *(t)HT
)(t)(t*(t)(t) ⊗�′� −−− 111 , where ( )( ) 111 −−− ⊗�′= VIVO HT)(t
4. The parameters in ζ are drawn from a SUR model with variance/covariance matrix of disturbances Λ.
)111 ),P)zI(X~N(P(,z *(t)HT
)(t)(t*(t)(t) ⊗′ −−−ζ , where 111 −−− ⊗= )X)I((X'P HT)(t
5. The vector of household intercepts αh is drawn from a SUR model with variance/covariance matrix of
disturbances Σ: ),Q)�V)rI(~N(Q(U'�,V,,y� )(t)(t-a
�
hT)(t)(t)(t
�
)(t(t)*(t)(t)h
11111111 −−−−−−− +⊗�� , where
( ) 11111 −−−−− +⊗�= )(t�T
)(t V)UIU'(Q , TSIU 1⊗= , and
(t)
hS
h
h
*(t)hS
*(t)h
*(t)h
�
h
V
V
V
y
y
y
r
����
�
�
����
�
−
�����
�
�
�����
�
= 2
1
2
1
�.
34
6. The vector of household intercepts ιh is also drawn from a SUR model with variance/covariance
matrix of disturbances Λ.
)11111111 ),R�V)rI~N(R(U'(,�,V,,�z� )(t)(t-�
�
hT)(t)(t)(t
t)(t(t)*(t)(t)
h−−−−−−− +⊗
where ( ) 11111 −−−−− +⊗= )(t�T
)(t V)UIU'(R , and
(t)
hS
h
h
*(t)hS
*(t)h
*(t)h
�
h
X
X
X
z
z
z
r ���
�
�
����
�
−
�����
�
�
�����
�
= 2
1
2
1
�.
7. The vector of hyper-parameters, δ, is drawn from a SUR model with variance/covariance matrix of
disturbances Vα: ),F)�V)�I~N(S(W'(V�,,V,V|�� �
(t)H
)(t��
)(t�
(t)(t) 1111 −−−− +⊗ , where
( ) 1111 −−−− +⊗= �H)(t V)WIW'(VF � .
8. The vector of hyper-parameters, ψ, is drawn from a SUR model with variance/covariance matrix of
disturbances Vι: )111 ),G�V)�I~N(T(W'(V�,V,�� �
(t)H
)(t��
(t)(t) −−− +⊗ , where
( ) 1111 −−−− +⊗= �H)(t V)WIW'(VG � .
9. Σ is drawn from an inverted Wishart distribution with HT+νΣ degrees of freedom.
( )( )11 −����
− ++�� ��'V, HT~Wish , ,V,V,,�,y (t)�
(t)(t)*(t)(t)(t)
10. Λ is also drawn from an inverted Wishart distribution with HT +νΛ degrees of freedom.
( )( )11 −− ++ uu'V, HT~Wish , ,V,V,,�,z�
(t)�
(t)(t)*(t)(t)(t) . Next, Λ is transformed into a
correlation matrix for identification purposes.
11. Vα is drawn from an inverted Wishart distribution with H+να degrees of freedom.
( )( )11 −− ++ ��'V, H~Wish , V,,V,�,yV ����
(t)�
(t)*(t)(t)(t)�
12. Vι is drawn from an inverted Wishart distribution with H+νι degrees of freedom.
35
( )( )11 −− ++ ��'V, H~Wish , V,,V,�,zV ����
(t)�
(t)*(t)(t)(t)� ζ . Next, Vι is transformed into a correlation
matrix for identification purposes.
Prior distributions
The second implementation step is to specify prior distributions for the parameters of interest.
Note that the priors are set to be non-informative so that inferences are driven by the data.
The prior distribution of δ is MVN(δ ,Vδ), where 0=δ and Vδ = diag(103).
The prior distribution of ψ is MVN(ψ ,Vψ), where 0=ψ and Vψ = diag(103).
The prior distribution of Σ-1 is Wishart: W(νΣ,VΣ), where νΣ = S+2 and VΣ = diag(10-3).
The prior distribution of Λ-1 is Wishart: W(νΛ,VΛ), where νΛ = S+2 and VΛ = diag(10-3).
The prior distribution of 1-�
V is Wishart: W(να, αV ), where να = S+2 and αV = diag(10-3).
The prior distribution of 1-�
V is Wishart: W(νι, ιV ), where νι = S+2 and ιV = diag(10-3).
Initial values
The third implementation step is to set initial values for the parameters of the marginal
distributions. The starting values for ω and δ are computed by OLS, using ln(ybit) as the dependent
variable of the regression. The covariance matrix, Σ, is initiated by computing the sample covariances of
this regression’s residuals. In a similar fashion, the starting values for the patronage equation parameters,
�i, are computed by OLS, using zhit as the dependent variable, and the residuals from this regression, ubit,
are used to compute the sample correlations, which serve as the initial value for Λ.
The final step is to generate N1+N2 random draws from the conditional distributions. We use a
“burn in” of N1=10,000 iterations. To reduce autocorrelation in the MCMC draws, we “thin the line,”
using every 50th draw in the final N2 = 10,000 draws for our estimation. In this way, 200 draws are used
to estimate marginal posterior distributions of the parameters of interest.
36
Table 1 Descriptive statistics of the six chains
Albert Heijn Super de Boer
C1000 Edah Aldi Lidl
Positioning (GfK source)
Service Service Middle Middle Discount Discount
Market share in 2002
24% 11% 15% 8% 11% 1%
Weekly store incidence
.35 .16 .27 .13 .25 .09
Weekly expenditures given expenditures > 0
28.52 27.78 27.78 23.71 20.97 16.71
Price image (1=lowest, 7 = highest)
5.2 5.5 6.0 5.8 6.8 6.7
Produce quality (1=lowest, 7 = highest)
6.4 6.2 6.2 5.5 4.4 4.8
Waiting time (1=longest, 7 = shortest)
5.9 6.1 6.1 5.7 4.6 5.5
Distance to panelists (km)
2.3 4.0 3.1 5.2 3.2 6.6
Store surface (m2)
1344 895 860 1013 422 615
Relative Price Index
1.17 1.09 .95 .99 .58 .58
Relative Feature (i.e., minus market average)
.6 1.0 -1.2 1.3 -1.8 -1.0
Relative Display (i.e., minus market average)
.3 .8 -.7 1.0 -.6 -.6
37
Table 2 Overview of independent variables in the incidence (I) and expenditure (E) models
Variable
Operationalization
Equation
Store benefits
PriceImagehit Price image of store i for household h in week t measured on a seven-point scale (1=worst, 7=best)b
I, E
ProduceQualityhit An important indicator of perceived chain quality, produce quality image of store i for household h in week t measured on a seven-point scale (1=worst, 7=best)b
I, E
StoreSurfacehit Floor surface (in 10,000 m2) of closest store of chain i to household h in week tb. This variable is an important indicator for assortment size.
I, E
Featurehit Feature activity of store i in week t for household h, i.e., a weighted average of store i’s feature activities in category c in week t with household h’s category shares as weightsa
I
Displayhit Display activity of store i in week t for household h, i.e., a weighted average of store i’s display activities in category c in week t with household h’s category shares as weightsa
E
LagExpendhit ln expenditures for household h in store i in week t-1 I, E
Store costs
Distancehit Distance (km) between household h and store i in week tb I, E
WaitingTimehit Checkout waiting time image of store i for household h in week t measured on a seven-point scale (1=longest, 7=shortest)b. This variable is an important indicator for perceived service.
I, E
lnPricehit ln relative basket price of household h for store i in week t, i.e., a weighted average of store i’s price in category c in week t with household h’s category shares as weights. It is operationalized relative to other stores by dividing it by the before-log weighted average across stores for household h in each week t.
I, E
Seasonalities
Week1t Dummy that is 1 in week 1, 0 else I, E
Week51t Dummy that is 1 in week 51, 0 else I, E
Week52t Dummy that is 1 in week 52, 0 else I, E
Eastert Dummy that is 1 if Easter is in week t I, E
Price war variables
PWt*Storei Price war step dummy * store i’s dummy. PWt = 0 before start price war, 1 after, and Storei equals 1 for store i and 0 else
I, E
PWt*lnPricehit Price war step dummy * ln Pricehit I, E
PWt*lnPricehit*(1/�) Previous interaction * (1/�), where � (=1,2,...) counts from first week of price war I, E
a: Feature and Display variables capture own- and cross-effects since they are operationalized relative to other stores by subtracting the mean across stores for household h in each week t b: obtained from the measurement moment that is closest to week t
38
Table 3 Posterior distributions of response parameters
Model for store incidence Model for ln expenditures
percentiles 2.5 50 97.5 2.5 50 97.5
Intercept Albert Heijn -0.857 -0.748* -0.677 6.952 7.008 * 7.063
Intercept Aldi -1.135 -1.076* -1.010 6.981 7.042 * 7.096
Intercept C1000 -1.216 -1.125* -1.065 6.871 6.930 * 6.982
Intercept Edah -1.819 -1.722* -1.652 6.495 6.566 * 6.662
Intercept Lidl -1.812 -1.752* -1.670 6.721 6.772 * 6.838
Intercept Super de Boer -1.842 -1.744* -1.683 6.875 6.941 * 6.998
PriceImage 0.004 0.012* 0.020 0.012 0.015 * 0.019
ProduceQuality -0.005 0.003 0.010 0.001 0.005 * 0.009
StoreSurface 0.220 0.480* 0.750 0.160 0.340 * 0.540
Feature 0.005 0.007* 0.009
Display 0.001 0.005 * 0.010
LagExpend 0.050 0.051* 0.052 0.004 0.005 * 0.006
Distance -0.107 -0.104* -0.100 -0.010 -0.007 * -0.005
WaitingTime -0.010 -0.002 0.006 -0.008 -0.004 * -0.001
lnPrice 0.016 0.123* 0.238 -0.142 -0.051 0.051
Week1 -0.821 -0.789* -0.756 -0.384 -0.359 * -0.339
Week51 -0.022 0.012 0.039 0.124 0.109 * 0.143
Week52 -0.222 -0.188* -0.159 0.003 0.024 * 0.041
Easter 0.057 0.079* 0.100 0.125 0.139 * 0.156
PW*AlbertHeijn 0.017 -0.024 0.050 -0.077 -0.049 * -0.017
PW*Aldi -0.242 -0.118* -0.006 -0.141 -0.025 0.055
PW*C1000 0.020 0.044* 0.068 -0.021 -0.038 * -0.008
PW*Edah -0.108 -0.083* -0.060 -0.113 -0.095 * -0.079
PW*Lidl -0.135 -0.013 0.104 -0.137 -0.021 0.064
PW*Super de Boer -0.046 -0.015 0.019 -0.079 -0.058 * -0.039
PW*lnPrice -0.563 -0.355* -0.120 -0.323 -0.107 0.035
PW*lnPrice*(1/�) -0.240 -0.121* -0.015 -0.149 -0.085 * -0.007
* the 95% posterior density excludes 0.
39
Table 4 Decomposition of the net effect of the price war on store incidence and expenditures
Store incidence (Price Part) ln Expenditures (Price Part)
Net
change∆
Intercept∆
lnPrice∆ Price
coefficient∆ Price image
Net change
∆Intercept
∆lnPrice
∆ Pricecoefficient
∆ Price image
Albert Heijn -0.03 53% -1% -167% 15% -0.06 -82% 0% -28% 10% Aldi 0.07 -181% 1% 282% -2% 0.03 -78% -1% 184% -5% C1000 0.06 70% -2% 31% 2% -0.01 -164% 5% 49% 10% Edah -0.08 -104% 0% 3% 0% -0.09 -102% 0% 1% 0% Lidl 0.17 -7% 2% 107% -1% 0.03 -63% -3% 173% -7% Super de Boer -0.04 -33% 2% -71% 2% -0.07 -86% -2% -15% 2%
Note: each percentage represents the part of the total change that is attributable to a change in a component.
40
(Source: GfK)
200320
200322
200324
200326
200328
200330
200332
200334
200336
200338
200340
200342
200344
200346
200348
200350
200352
200402
200404
200406
200408
200410
200412
week
0.04
0.06
0.08
0.10
0.12
Pric
e No
n-Al
coho
lic B
ever
ages
(pric
e pe
r ml)
chainALBERT HEIJNALDIC1000EDAHLIDLSUPER DE BOER
Figure 1 Price (in Euro) for non-alcoholic beverages in the six chains
over time
41
Figure 2 Positioning of the six major Dutch retail chains in Summer 2002 (Source: GfK)
LowerValue
MoreService
LessService
Higher Value
C1000
Superde Boer
Edah
Aldi
Lidl
��������
��� ������
Albert Heijn
42
200205200209200213200217200221200225200229200233200237200241200245200249200301200305200309200313200317200321200325200329200333200337200341200345200349200401200405200409200413200417200421200425
week
0.50
0.60
0.70
0.80
0.90
1.00
1.10
1.20
Price
chainALBERT HEIJNSUPER DE BOEREDAHC1000LIDLALDI
200205200208200211200214200217200220200223200226200229200232200235200238200241200244200247200250200301200304200307200310200313200316200319200322200325200328200331200334200337200340200343200346200349200352200403200406200409200412200415200418200421200424
Week
5.0
5.5
6.0
6.5
7.0
Pri
ce Im
age
chainALDILIDLC1000EDAHSUPER DE BOERALBERT HEIJN
Figure 3 Relative Price and Price image for the six chains over time