See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/228667260

Market Structure and Competition in the Retail

Discount Industry
Article in Journal of Marketing Research September 2007
Impact Factor: 2.52 DOI: 10.1509/jmkr.46.4.453

CITATIONS

READS

27

763

3 authors, including:
Ting Zhu
University of British Columbia - Vancouver
13 PUBLICATIONS 144 CITATIONS
SEE PROFILE

Available from: Ting Zhu

Retrieved on: 12 April 2016

Market Structure and Competition in the Retail

Discount Industry
Ting Zhu, Vishal Singh, Mark Manuszak1
August 15, 2007

1 Ting

Zhu is Assistant Professor of Marketing at the Graduate School of Business, University

of Chicago, Vishal Singh is Associate Professor of Marketing at the Tepper School of Business,
Carnegie Mellon University, and Mark Manuszak is an economist at the Federal Reserve Board
of Governors. The authors can be contacted via email at: tingzhu@chicagogsb.edu (Zhu), vsingh@andrew.cmu.edu (Singh), and mark.d.manuszak@frb.gov (Manuszak). Mailing address:
5000 Forbes Ave, Carnegie Mellon University, Pittsburgh, PA 15213. Phone: 412-268-7392.
The views expressed here are those of the authors and do not necessarily reflect the views of
the Board of Governors of the Federal Reserve System or its staff.

Abstract
We examine competition between the three major firms in the retail discount industry, WalMart, Kmart and Target, using two distinct, but related, approaches. We first consider a
model that examines the determinants of store presence for the three chains in local markets.
For each local market, we assume that each chain chooses whether to operate a store and,
if so, which store format. These decisions are based on the evaluation of underlying payoffs
which depend on observable and unobservable market characteristics including the presence of
competing firms. Notably, we allow for substantial asymmetries across firms in terms of their
preferred characteristics and the impact of competition. Since each firms decision depends on
the decisions of its rivals, the model involves a discrete game between the chains, the outcome
of which determines the realized market structure. We use numerical techniques to characterize
the equilibrium in each market and estimate the model by simulated maximum likelihood using
data on market characteristics and structures for a cross-section of local markets. Hence, even
without detailed store-level data, our analysis provides insight into competitive interactions
between these chains. We then augment our analysis to evaluate the determinants of one
important store-level variable that we do have available, namely store-level revenues. We posit
a regression model relating revenues of each store to various market characteristics including the
presence of competing firms. However, since the presence of competing stores is not randomly
determined, we use the results of our intial market structure model to generate correction terms
for the endogeneity of market structure.
Our results from both exercises illustrate a number of important asymmetries across the
chains, many of which corroborate the conventional wisdom about positioning in the industry. From the market structure model, we find that Wal-Mart prefers lower income, low cost,
family-oriented markets whereas Target requires substantially different market characteristics,
such as high income levels, to operate a store. Kmart, on the other hand, exhibits fewer systematic patterns in the market characteristics that it finds attractive, perhaps reflecting a lack
of strategic focus on the part of that firm. Notably, we find that Kmart and Wal-Mart do not
inherently differ in their ability to operate in certain markets as markets with similar sizes could
support either chain as the sole operator. The difference between the chains, however, arises
in their vulnerability to competition. Kmart is much more adversely affected by the presence
of Wal-Mart than the converse. Together, these findings suggest that Wal-Marts crucial competitive advantage does not lie in its ability to operate where its rivals could not, but rather
is its ability to better weather competition with its rivals. Our analysis of store-level revenues
reinforces many of these findings. Once we correct for the endogeneity of market structure,
we find that Wal-Mart has a large negative effect on the store-level revenues of its rivals while
Target has a somewhat lesser effect with Kmart trailing behind. In total, these various results
confirm the standard view of the industry as one in which Wal-Mart is dominant with Target
serving more of a niche role and Kmart struggling to find its footing.

Keywords: Endogenous market structure, discrete games, retail competition, discount

stores

Introduction

The year 1962 marked a turning point in the history of retailing with the emergence of
three retail stores that would change the course of business in years to come. Having
observed the success of the supermarket format in grocery retailing, three companies
anticipated a similar opportunity for the sale of low-priced general merchandise items. In
1962, S.S. Kresge Company opened its first discount store in Garden City, Michigan and
called it Kmart. Shortly thereafter, Dayton Hudson Corporation, which also operated
stores in the upper Midwest, opened its first discount store named Target in Roseville,
Minnesota. In the same year, an independent operator named Sam Walton opened
another discount store in Rogers, Arkansas: Wal-Mart. Over the next few decades, these
firms expanded their operations both domestically and internationally to reach dominant
positions in the retailing world.
Despite the sheer size and economic importance of discount retailers1 along with the
intensive coverage that they receive in the popular press, academic researchers have only
recently begun to rigorously examine their strategic decision-making and the effect of
competition between them.2 This relative dearth of formal empirical analysis does not
stem from a lack of interest, rather it largely reflects data constraints. In contrast to
other segments of the retail market such as grocery stores where detailed scanner data
have allowed careful examination of demand features including substitution patterns
across retailers (e.g., Bell and Lattin 1998) or retailers interaction with manufacturers
(Sudhir 2001), data on prices and sales volumes for discount retailers are generally limited.3 Moreover, discount retailers tend to carry a much wider assortment of products
such as groceries, electronics, jewelry, and clothing which complicates the collection and
comparison of product information, such as prices, across firms.
In this paper, we study competition between the three dominant discount retailers
using two distinct, but interrelated, techniques. First, we use an indirect approach to
examine competition between the three chains in which we examine the determinants of
only one, albeit important, decision in retailing: whether to operate a store in a local
1

Wal-Mart alone has annual revenues of over $300 billion making it the worlds largest company in
addition to being the largest retailer.
2
See, e.g., Holmes (2005), Zhu and Singh (2006), and Jia (2007.) A large literature does exist that
relates various variables, such as local market wages (Basker 2005) or sales at local retailers (Singh et
al. 2006), to the presence of discount retailers like Wal-Mart. This latter literature does not examine
competitive issues across chains.
3
For example, Wal-Mart does not share its point of sales data even with specialized market research
firms such as IRI and ACNielsen.

market. This decision is important since, unlike other marketing mix elements such
as pricing, product assortment or promotions, market presence is less adjustable in the
short-run without incurring significant costs. Our initial approach evaluates the effect
of various factors, including local market conditions and competitive considerations, on
realized market structures in local markets. Hence, even without comprehensive storelevel data, our analysis broadly provides information about competition between the
chains based on their store operation patterns across markets.
The basic intuition behind our initial analysis is based on revealed preference arguments: the fact that we observe some combination of firms in a market reveals something
about the underlying economic profits in those markets since firms would be present only
if they expect positive profits, but not otherwise. We formalize this portion of our analysis by assuming that, for a cross-section of local markets, each firm chooses from one of
three alternatives: no store, operate a standard discount store, or operate a supercenter.
We do not directly observe profits, but assume that each firms underlying payoff depends on market characteristics, unobservable factors, and firm-specific factors including
its format choice. In addition, the store operation and format decisions of its competitors affect each firms profits. This last consideration implies strategic interdependencies
in the firms decisions since Target, for example, must consider the possible decision of
Wal-Mart when choosing whether to operate in a particular market. The requirement
that the profit-maximizing decisions of all firms must be simultaneously consistent implies conditions that reflect a Nash equilibrium of the game between the chains. Due to
the discrete nature of the store operation decision, these equilibrium conditions involve
inequality restrictions on latent profits which, as initially suggested by Bresnahan and
Reiss (1990, 1991a, 1991b - hereafter BR) and Berry (1992), can be used to estimate
features of latent profits through an estimation approach such as maximum likelihood.
Intuitively, this approach parallels the standard single-agent discrete choice models
(McFadden 1982) that have been used extensively in the marketing literature. In that
case, the researcher makes inferences about unobserved latent utility based on the inequality restrictions that observed behavior implies about utilities. In the case of a
discrete game such as the one that we consider, however, payoffs and resulting behavior
reflect the interaction of decisions of multiple agents. Thus, estimation is based on an
oligopolistic equilibrium concept rather than on individual utility maximization.
In addition to data on store presence, we also have some information about store-level
outcomes, specifically revenues for individual stores of each chain. Using these data,
our second approach to examine competition between the chains involves an evaluation
2

of the impact of different combinations of competitors on the revenues of individual

stores. Notably, this second exercise relies on our initial analysis of store operation
patterns since Wal-Mart, for example, does not randomly choose the markets in which
it is present. As a result, when relating Targets store-level revenues to Wal-Marts
presence, we must account for the endogeneity of the observed market structure. We
use our initial model that characterizes market structures to provide correction terms to
account for this endogeneity in store-level revenue regressions.
Our evaluation of store-level revenues not only provides information about competition between the firms, but can also serve a practical purpose for analysts of this industry.
An analyst who wishes to perform formal statistical analysis to generate a forecast of
store-level sales revenue or to undertake an evaluation of individual store performance
would presumably consider how sales depend on the presence of competing stores. Doing
so, however, requires an appropriate treatment of the endogeneity of market structure
such as the one that we employ. More generally, one could envision evaluating the impact
of market structure on a wide variety of market outcomes such as some measure of prices
(Mazzeo 2002b, Singh and Zhu 2007, Manuszak and Moul 2007), profits (Buzzell and
Gale 1987) or product lines (Cohen and Mazzeo 2004, Watson 2004). Any such exercise
relating a market outcome to a market structure variable must, in some way, reflect the
non-random determination of market structure. Our application to store-level revenues
in the discount retail industry provides such an example.
An important component of our analysis involves firm and format heterogeneity. In
both our examination of store operation patterns and our revenue regressions, we allow for differences in competitive effects across both firms and formats. Allowing such
heterogeneity is likely crucial in our application since, given its positioning and image,
we may expect Wal-Mart to have a different impact on the profitability of Kmart than
Target. Moreover, the effect of a Wal-Mart on Kmart may depend on whether Wal-Mart
operates a conventional discount store or the relatively novel, but widely touted, superstore format. Similarly, the impact of market conditions may vary across firms as well as
store formats leading us to allow for heterogeneity in that dimension as well. Whether
any such differences across firms actually exist is an empirical issue. Whereas previous
work has typically allowed limited heterogeneity across firms (e.g., Berry 1992, Mazzeo
2002a), the presence of a small number (three) of identifiable firms (Wal-Mart, Kmart,
Target) allows us to consider more flexible patterns of various effects related to specific
firm identities.
To estimate the market structure model and the revenue regressions, we assemble
3

a unique dataset that includes locations for all Wal-Mart, Kmart and Target stores in
operation during 2003 in the United States. While we do not generally observe detailed
profits or product-level sales or price information for each store, we do observe store
format information as well as store-level revenues. We partition the United States into
a series of smaller local markets for which we collect an extensive set of measures to
reflect market-specific demand and cost conditions. Hence, for a cross-section of local
markets, our dataset contains the presence and store formats for each retailer along with
revenues for each store and a series of variables related to other market conditions. We
use these data to estimate the market structure model by simulated maximum likelihood.
We then estimate the store-level revenue regression model including the correction term
for endogenous market structure using a two-stage estimation procedure similar to that
in Mazzeo (2002b.)4
Our results reveal a number of important features of firm behavior in this industry,
many of which lend support to the conventional wisdom about the strategic positioning of the chains. In particular, substantial heterogeneity exists across firms regarding
the market characteristics that they find attractive. From the market structure model,
we find that Wal-Mart strongly prefers to enter markets that have lower retail wages
and more households with vehicles and children. At the same time, Wal-Mart is less
concerned with incomes in a market. Conversely, Target tends to prefer markets with
higher incomes, but is less concerned with the vehicle holdings or number of children
for households in a market. Somewhat surprisingly, the importance of higher customer
incomes for Target even extends to markets with higher wage rates as Target prefers
markets with higher wages despite the increased costs in such markets. These findings
corroborate the conventional view of Wal-Mart as a firm that is highly concerned with
costs and locates at the outskirts of population centers in less affluent, family-oriented
markets. In contrast, Target prefers to locate in more affluent, urban communities.
Our most notable results from the market structure model can be seen in the market
populations required to support the presence of different firms and formats. Given all
other market characteristics evaluated at their typical levels, we find that both Wal-Mart
formats would require similar populations to enter as monopolists. More interestingly,
we find that the breakeven population size for a monopoly Kmart discount store is only
slightly larger than that required for a monopoly Wal-Mart store. In other words, neither
format is clearly a dominant format for Wal-Mart in a typical market where it can
4

This approach to deal with the endogeneity of market structure builds upon the work of Heckman
(1979.)

operate as a monopolist, and that same market could just as easily support a Kmart
monopoly. The crucial difference between these firms, however, arises due to the effect
of competition. While Wal-Mart naturally prefers markets in which its competitors are
not present, it is able to operate in markets as a duopolist at much smaller population
sizes than Kmart can. Put differently, our results suggest that Wal-Marts advantage
over Kmart is not necessarily its superior ability to operate in markets which Kmart
would find unattractive anyway. Instead, Wal-Marts crucial strength lies in its ability
to weather competition to a greater extent than its rivals.
Target, on the other hand, requires a substantially larger population to operate a store
even as a monopolist. An alternative interpretation of this finding is that other market
conditions must be sufficiently attractive for Target to offset its apparent need for a large
customer base. This finding again reflects strategic positioning by Target as a firm that
targets certain types of markets which may be quite different from the typical market.
However, as with Kmart, Target again finds markets in which Wal-Mart is present to be
very unattractive especially compared to analogous markets which instead have a Kmart.
And again, the effect that Target experiences due to Wal-Marts presence is much larger
than the converse. In fact, for both Target and Kmart, the implied effect of a Wal-Mart
is more than twice the reverse effect that those firms have on Wal-Mart. In short, these
types of findings confirm the conventional view of Wal-Mart as the dominant player in
this industry.
Our findings from the revenue regressions reinforce many of these results. When we
account for the endogeneity of market structure, we find that all firms tend to experience
declines in store revenues when faced with competitors. However, the effect on the
revenue of Kmart or Target stores is again largest when Wal-Mart is present. Conversely,
Kmart tends to have a relatively small impact on the revenues of either of its rivals
which mirrors our findings in the market structure model and suggests a relatively fragile
position for that chain in the industry. Target, while not affecting Kmarts revenues
as much as Wal-Mart, does have a substantial impact on the revenues of Wal-Mart
discount stores, possibly resulting from careful selection by Target of markets with certain
characteristics that are especially amenable to its operations.
Throughout our estimates of the revenue regressions, we find that correcting for the
endogeneity of market structure is crucial. When we fail to perform such a correction,
we tend to understate the impact of competition on revenues. For Wal-Mart, many of
our results suggest that revenues rise in the presence of competing firms, results that are
generally reversed when we include the correction term. For the other firms, our estimates
5

of the effect of competition on revenues become substantially more negative when we

include our correction term. These results demonstrate the importance of accounting for
the non-random determination of market structure since an analyst performing such an
exercise could draw very misleading conclusions and produced flawed forecasts from a
simple regression of revenues on market structure variables.
The rest of the paper is organized as follows. In the next section we provide a brief
overview of the discount store industry and discuss the data. Section 3 presents the
market structure model and discusses estimation of that model. Section 4 presents a
simple regression model for store-level revenues and discusses the procedure to correct
for the endogeneity of market structure. Section 5 presents the estimation results while
Section 6 concludes with suggestions for future research.

Industry and Data Description

The year 1962 marked a turning point in the history of retailing due to the emergence
of three retail stores who would come to dominate the retail industry in years to come.
By 1962, the supermarket format had become well-established in the grocery industry.
The adoption of this format replaced small, high margin neighborhood stores with large,
self-service stores that tended to locate in low-rent areas and engaged in vigorous price
competition. A similar opportunity to transform another segment of the retail market
appeared to exist in the sale of general merchandise items. This apparent opportunity
led to the creation of three discount retail chains, Kmart, Target and Wal-Mart, all of
whom opened their first stores in different parts of the country in 1962. While several
other players have existed in the discount general merchandise industry, these three
chains would come to dominate the retailing world. Over time, they each expanded
their operations across the country and internationally in the process developing several
other retail formats, most notably the supercenter. This format combines the general
merchandise aspect of the standard discount stores with a full-service grocery store. In
addition, these stores include several ancillary services such as pharmacies, dry cleaning,
vision centers, automotive care centers, hair salons, and income tax preparation resources
(in season), providing consumers with a true one-stop shopping environment.5
In what follows, we provide a brief history of the three chains. Since these companies
have received tremendous coverage in the business press, our discussion is relatively brief.6
5

Singh et al. (2006) provide a detailed discussion of the supercenter format.

Interested readers are referred to various Harvard Business Review case studies (Ghemawat 1986,
Ghemawat and Friedman 1999, Noe 1998, Miller and Noe 2000) and business press books such as Sam
6

2.1

Company Overview

The first Kmart opened as a retail outlet for the S.S. Kresge Company which was one of
largest general merchandise retailers in the country. By 1977, the Kmart discount format
accounted for almost 95% of the companys revenues leading S.S. Kresge Corporation to
change its name to Kmart Corporation. Being part of a well-financed retailer, Kmart
experienced tremendous initial expansion with the opening of over 1000 stores during its
first decade. By early 1980s, the company had over 2000 discount stores and had ventured into Canada and Puerto Rico. Kmart developed a reputation for low prices and was
particularly known for its blue light special promotions. It surpassed Sears, Roebuck
and Co. in 1990 to become the worlds largest retailer. Mimicking its rival Wal-Mart, the
company opened its first Super Kmart Center in Medina, Ohio in July, 1991. Between
1985 and 1992, Kmart also diversified into other businesses by acquiring companies such
as Builders Square, Walden Books, Borders Bookstore, PACE Membership Warehouse,
Office Max, Sports Authority, and Payless Drug Stores. However, the company encountered serious financial troubles with a reported loss of $974 million in 1993. By the late
1990s, the company had divested equity interest in most of its acquisitions to concentrate on its core discounts stores and supercenters. With ever-increasing competition
from rivals Wal-Mart and Target, the company introduced several new brands such as
the Martha Stewart Everyday line and even re-introduced its blue light special in 2001.
Despite these initiatives, Kmart filed for Chapter 11 bankruptcy protection on January
22, 2002 and announced the closure of over 600 stores soon after. Kmart emerged from
Chapter 11 bankruptcy in 2003 with a new chairman, Edward Lampert, and has recently
acquired Sears Roebuck & Co.
Target originated as a part of Dayton Hudson Corporation which, during its history, has also operated several other department stores including Marshall Fields and
Mervyns. In 2000, the company changed its name to Target Corporation and subsequently sold off most of its other divisions. Target traditionally has been the smallest
of the Big 3 discount stores before overtaking Kmart in 2002 to take the number two
spot.7 The company saw dramatic growth during the mid-1990s with the total square
footage of its stores doubling between 1996 and 2001. Following in the footsteps of
Walton : Made In America, (Sam Walton and John Huey), Bantam (1993); The Wal-Mart Decade:
How a New Generation of Leaders Turned Sam Waltons Legacy into the Worlds #1 Company (Robert
Slater) Portfolio Hardcover (2003); On Target: How the Worlds Hottest Retailer Hit a Bullseye
( Laura Rowley) Wiley (2003); Kmarts Ten Deadly Sins: How Incompetence Tainted an American
Icon, (Marcia Layton Turner) John Wiley & Sons (2003).
7
Kmarts recent acquisition of Sears has put it back ahead of Target.

its rivals, it opened its first supercenter, Super Target, in Omaha, Nebraska in 1995.
Historically, Target has attempted to differentiate itself from Wal-Mart and Kmart by
introducing higher-end products in order to position itself as an upscale discounter.8
It has partnered with several designers to offer contemporary merchandise, runs its own
Target Guest credit card, and offers nationwide bridal and baby shower gift registries. A
recent survey by Retail Forward, Inc. finds the demographic profile of shoppers at Target stores to be more affluent and quite different in other dimensions than Wal-Mart and
Kmart (Retail Forward Channel Report: Discount Stores/Supercenters, August 2004).
The company currently operates approximately 1,250 stores with over $40 billion in revenues.
Wal-Marts initial expansion was quite different from its rivals. As is well known,
whereas its competitors focused on urban locations, Wal-Mart primarily expanded by
opening outlets in small rural towns. In addition, Wal-Marts initial growth was quite
subdued compared to Kmart. For example, in 1980, Wal-Mart operated less than 300
stores while Kmart had nearly 2000 stores. The company grew dramatically over the
next decade, however, with 1,250 stores and sales of over $20 billion by 1988 when Sam
Walton stepped down as the CEO of the company. His successor, David Glass, who
had a background in the grocery industry, transformed Wal-Mart over the next several
years with the opening of the first Wal-Mart Supercenter.9 This new format fueled the
companys growth making it the worlds largest retailer in 1991 and the worlds largest
company by 2002. With its EDLP strategy and the Always Low Prices, Always slogan,
Wal-Mart has firmly established itself as the undisputed low price leader in the industry.
Analysts attribute their low-price leadership primarily to their hard negotiating skills
with the suppliers and their excellence in supply chain management. According to its
2004 annual report, the company currently operates 538 Sams Clubs, 2,949 domestic
stores (discount and supercenters) and 1,355 international stores, generating over $250
billion in revenues.

2.2

Data Description

Our data includes locations (street addresses and zip codes) of all stores in operation as of
2003 for Wal-Mart, Kmart, and Target. Besides the location data, we also observe some
other information about each store. In particular, we know each stores format (discount
8

Targets panache is reflected in the pop-culture use of a French-influenced pronunciation for its name.
In the early 80s, the company also experimented with several other formats including Sams Warehouse Club, Deep Discount Drugstore, and Helens Arts and Crafts store. Only Sams Club has survived
to the current day.
9

store versus supercenter) and opening date. The store opening date information was
collected using the historical volumes of the Directory of Discount Retailers as well as by
calling individual stores. Finally, for each store, we observe the average weekly revenue
for 2003.10
The total number of stores belonging to each chain is presented in Table 1.11 WalMart is the largest player with almost 3000 total stores and an equal distribution of across
the two store formats. By contrast, Kmart and Target are relatively minor players in the
supercenter format. In Figure 1, we show the distribution of stores for these three chains
across the country. Not surprisingly, the highest number of stores are found in the two
largest states in the country, California and Texas. The three chains have the highest
share (in terms of number of stores) in their states of origin: Kmart in Michigan, Wal-Mart
in Arkansas, and Target in Minnesota. The chains also tend to have a disproportionate
share of stores in states that neighbor their headquarters state. However, Wal-Mart
has more stores than its competitors in all but 6 states in the country. In addition to
Michigan and Minneapolis, these states include California, Pennsylvania, Hawaii, and
New Jersey.12
An ideal model of endogenous market structure would consider dynamic entry patterns by the three chains accounting for the loction of distribution centers as well as the
actual street-level locations of stores within local markets. As we discuss in the conclusion, some progress has been made in the formulation and estimation of such models,
but this literature is still in development. Characterizing equilibria in such models is
complex, and problems with multiple equilibria are severe. Both of these issues arise in
the context of our model, but our framework involves simplifications that alleviate them
in order to yield a model that is reasonably tractable and yet still rich enough to provide
insight into competition between the chains. In particular, we partition the contiguous
United States into smaller local markets. For each of these local markets, we gather a
number of variables reflecting demand and cost conditions, and we aggregate our store
data up to the level of these markets. We then look to explain differences in outcomes
such as market structures across the local markets.
The practice of breaking large geographical regions into smaller local markets has
been widely used in the literature that considers the determinants of market structure.
For example, BR (1990, 1991a) consider isolated rural towns, Berry (1992) examines
10

The revenue data were acquired from Trade Dimension.

We do not consider Sams Club as its main competitors are other membership clubs such as Costco
and BJs.
12
Wal-Mart and Kmart are tied with 10 stores each in South Dakota.
11

city-pair airports, Mazzeo (2002a) studies highway exits, and Seim (2006) uses mediumsized cities. In such an application, any definition of a market is going to be arbitrary
on some level. A common concern is that the local markets must be small enough that
we expect firms in a market to compete with one another, but should be big enough to
ensure that consumers in one market do not travel extensively to another location for
their purchases.
In our empirical application, we considered three alternative market definitions: (i)
Only non-MSA counties; (ii) All counties in the US; and (iii) Census tracts which share
the first 3 digits of the census tract number.13 However, the use of data in only non-MSA
markets is problematic in our study as we end up discarding over 60% of the stores.
This problem is particularly severe in the case of Target which primarily enters urban
markets. Next, using all counties as markets also seems problematic due to extreme
heterogeneity in their characteristics. In particular, market size measured in terms of
population exhibits large variation across MSA and non-MSA counties with average MSA
markets more than ten times larger than non-MSA markets. While large variation across
markets is not a problem in general, this variation is associated with unusually heavy
store presence in some cases. For example, Los Angeles county alone has over 40 Target
stores. Our third market definition eliminates both of these problems to a large extent.
The first three digits of the tract number creates sub-markets in large urban counties
while leaving rural counties largely unaffected. For example, the roughly 800 large MSA
counties are split into approximately 5,000 sub-markets, while there is only a marginal
increase in the number of markets in the non-MSA counties. In addition, we find very
few markets (less than 3%) with multiple stores from the same firm, allowing us to focus
on only store operation decisions and not the decision on how many stores to open in
a market. As a result, we choose to focus on this market definition in our empirical
application yielding 7,698 markets.14
For each of these smaller markets, we gathered data on a variety of cost and demand
variables from the 2000 U.S. Census and the Bureau of Labor Statistics. In addition to
population, these variables include: median income (Income); the percentage of population with no vehicles (Novehicle); the percentage of population with children under the
age of 18 (Childratio); the average retail wage in a market (Retail Wage); the state sales
13

A census tract is identified by a unique code, SS-CCC-TTTTTT, where SS represents the State
FIPS code, CCC represents the County FIPS Code, and TTTTTT is the census tract number. Our
third definition defines a market as the tracts within a county with the same first three digits in their
tract numbers.
14
We exclude a small number of markets with population less than 5000.

10

tax rate (Tax Rate); and the price of electricity for commercial users (Electricity Price.)
We include these variables to account for some of the demand and cost factors that may
affect market outcomes. In addition, we also obtained information about the number of
supermarkets (Num Supermarket) from the Economic Census. Finally, for each market
we compute the distance from a centroid point of that market to the headquarters and
the nearest distribution center of each chain. Table 2 provides summary statistics of
these variables for the 7,698 markets in our data.
We link each of the stores in our data to these local markets. Table 3 presents the
market configurations that we observe in these markets along with the average values
for the population and income variables.15 The letters W, K, and T respectively represent the presence of Wal-Mart, Kmart and Target in the markets. For example, (0,0,0)
represents markets with no firms while (W,K,0) are markets with both Wal-Mart and
Kmart. Table 3 reveals a few patterns in the data. First, markets with no firms tend
to have significantly smaller populations. Second, Wal-Mart operates as a monopolist in
far more markets than its competitors. However, this table does not identify the forces
underlying the disproportionate number of Wal-Mart monopolies. On the one hand, the
typical population and income in Wal-Mart monopoly markets tend to be lower than in
markets for the other two chains, particularly when compared to Target. This pattern
suggests that Wal-Mart may be able to systematically enter markets that are unattractive to its competitors. On the other hand, an alternative explanation could be that
the prospect of competition from Wal-Mart dissuades Target or Kmart from opening
stores in markets with a Wal-Mart. Simultaneously, competitive effects could decrease
the number of monopoly markets for Target and Kmart since Wal-Mart may be able to
operate in many markets requiring the other two chains to be very careful about choosing
the markets in which they subsequently are monopolists. Our full model will shed light
on these different possibilities.
In summary, our data consists of a cross-section of local markets. For each market,
we observe the presence and format choice of each chain. In addition, we observe demand
and cost conditions in each market. In what follows, we develop a model of endogenous
market structure reflecting the store operation decisions of these three firms in each local
market. Application of this model to our cross-section of markets will allow us to draw
inferences about the factors that govern the store operation and format choices of the
companies including the importance of competition between them. We then present a
store-level revenue regression model and discuss a correction approach to deal with non15

To conserve space, we ignore store formats in Table 3.

11

random determination of market structures using the results of our market structure
model.

A Model of Retail Market Configurations

In this section, we begin by formulating a model to characterize the store operation decisions of the three chains in local markets. Formally, this model will involve a discrete
game between the firms. Beginning with BR (1990, 1991a, 1991b), the last decade has
seen a large increase in empirical research devoted to the study of discrete games. The
general goal of inference in discrete games is analogous to that of single-agent decision
problems, namely using restrictions implied for certain choices to be optimal to estimate
features of underlying payoffs. However, whereas the single-agent problem involves a
decision made in isolation, the interdependencies in agents payoffs in a game theoretic
setting necessitates the use of some equilibrium concept, most commonly Nash equilibrium or some refinement of Nash equilibrium. Moreover, unlike structural models of
supply and demand that provide marginal conditions for outcomes, the discrete nature
of decisions in such settings implies threshold conditions for latent profits.
A common feature of these models is the presence of multiple equilibria in which certain values of the underlying latent payoffs are simultaneously consistent with more than
one equilibrium outcome. Researchers have generally adopted one of the two approaches
to deal with the problem. First, models sometimes yield a prediction that is unique across
all equilibria for some outcome such as the total number of active firms (BR 1990, Berry
1992.) In our application, however, we are explicitly interested in asymmetries related
to the identity of the firms. An approach based on an aggregate outcome such as the
total number of firms would provide limited information about such asymmetries. As a
result, we adopt the second main approach to deal with multiple equilibria, namely the
imposition of additional structure on the model to guarantee equilibrium uniqueness. In
particular, following Mazzeo (2002a), we assume that firms sequentially make their decisions about whether to operate in a particular market, an assumption that, along with
other mild restrictions on payoffs, implies a unique subgame perfect Nash equilibrium
(SPNE) for the game.16 As we discuss in more detail below, we assume that the sequence
of decisions follows the observed timing at which the chains entered a market, if at all.
16

We also follow the literature by assuming that the data is generated by a pure strategy equilibrium
and that the model satisfies restrictions sufficient to guarantee the existence of at least one such equilibrium. Exceptions to the focus on pure strategy equilibria include Bajari, Hong and Ryan (2004) and
Sweeting (2004).

12

The focus on asymmetries across firms, which dictates many of our assumptions, is
crucial in our application. Because of data limitations or tractability issues, most previous studies have either assumed no heterogeneity in payoffs across firms (Bresnahan
and Reiss 1991a) or have limited firm-level heterogeneity to differences in fixed costs of
operation (Berry 1992) or discrete differences in competitive effects across broad classes
of firms (Mazzeo 2002a, Cohen and Mazzeo 2004, Seim 2006.) In contrast, heterogeneity of competitive effects across firms is likely crucial in our application since, given its
positioning and image, we may expect Wal-Mart to have a different impact on the profitability of Kmart compared to Target. However, an important aspect of the discount
retail industry is the existence of three dominant players. Hence, we know the identity
of all actual and potential entrants in a market, and the number of potentially active
firms is small. These feature of our application allows us to avoid strong restrictions on
heterogeneity across firms in order to examine whether important asymmetries are, in
fact, present.
An additional complication with discrete models of strategic interaction is the complexity of the equilibrium conditions required to yield a certain outcome. In terms of the
stochastic structure of the model, these equilibrium conditions imply inequalities that
involve complex regions of integration for the models unobservables. As suggested by
Berry (1992), we use simulation methods to compute the required integrals. The game
is solved numerically by backwards induction, and we use simulated maximum likelihood
procedures for estimation.

3.1

Payoff Functions for Retailers

We model a firms decision about whether to operate in a market and which format to
choose simultaneously. Each firm, Wal-Mart (W ), Kmart (K), and Target (T ), has
three options in a local market: operate a discount store (d), operate a supercenter (s) ,
or do not operate in the market (0). The expected present discounted value of profit for
firm i {W, T, K} with store format f {d, s} in market m is specified by the payoff
function
if m (Dim , Xim , if m ; if ) =
if m (Dim , Xim ; if ) + if m

(1)

where
if m (Dim , Xim ; if ) reflects profits known, to the analyst, up to some parameters
if , and if m is a profit shock for firm i with store format f in market m. In
if m (.),
the term Dim represents the store operation and format decisions of all other firms in
market m. In the current context, Dim is an indicator vector with 4 (2 2) elements,
13

since each competitor has a choice of two store formats. Xim are observed characteristics
of market m that impact the profitability of retailer i and can include variables that are
common across all firms in a market, such as market population, as well as characteristics
that are specific to firm i in market m. The vector if includes parameters reflecting the
impact of demand and cost conditions, as well as competitive interactions. As emphasized
earlier, we allow the parameters if to be firm- and format-specific. Thus, we allow for
the fact that different firms and formats may (i) impact competitors differently and (ii)
put different weight on the market conditions. In terms of the competitive effects, we
allow both the identity and format of the competing firms as well as the identity and
format of the firm of interest to be reflected in the payoff parameters. In other words,
the impact of a Wal-Mart supercenter can be different for Target discount stores and
supercenters. Moreover, both effects can differ if a Wal-Mart discount store is present
rather than a supercenter. In the empirical section, we assume that the average profits,

if m (.), are linear in the market characteristics. In addition, we assume that the presence
of different competing types yields discrete shifts in profitability.17 Finally, if a firm does
not operate in a market, its payoff in that market is normalized to zero.
The profit function in (1) should be viewed as an approximation to the reduced form
outcome of second stage competition of a two-stage game in which firms initially make
decisions about whether to operate in a market followed by price or quantity competition
given those operation decisions. The last term in the payoffs, if m , is a component of
profits for firm i with store format f in market m which is observed by the players, but
not by analyst.18 This term reflects the fact that it is virtually impossible to collect data
on all relevant factors that impact the store operation decisions of firms. We assume that
the unobservables if m are independent of the market characteristics Xim . Moreover, to
account for the variety of unobserved heterogeneity that could be present in our application, we specify a very rich error structure to allow for correlations in unobservables
across firms in a market. In particular, we decompose if m as
if m = um + uim + uif m

(2)

where um is a market-specific component that is common across all firms and formats,
uim captures firm-specific unobserved components specific to market m, and uif m is
17

In other words, our specification involves multiple indicator variables corresponding to the presence of
different firm-format combinations for competing firms. The competitive effects are then the coefficients
on these indicator variables.
18
We assume that all actual and potential firms observe all components of payoffs, both for themselves
and for other firms. Hence, we consider a complete information game between the firms. Seim (2006)
and Zhu and Singh (2006) consider a similar framework with incomplete information.

14

an idiosyncratic component for each choice alternative in market m. An example of a

market-level shock is zoning ordinances that are used to prevent entry by big box retailers.
Such zoning restrictions vary across markets and are difficult to collect and quantify,
but could yield negative market-level shocks that impact all retailers evaluation of the
market. Market-level shocks could also impact each firm differently. For example, some
local communities oppose entry by Wal-Mart in their neighborhoods, but such an attitude
may not apply to Target. Such market-level shocks along with other unobserved demand,
cost, or logistic features that are specific to each firm are captured by uim. In the empirical
application, we assume that uif m N (0, 1)19 and that the other random components
are normally distributed with mean zero and different variances to be estimated.

3.2

Conditions for Nash Equilibrium

The model above yields a system of simultaneous equations for a discrete outcome variable, namely the store operation decision of each firm, with discrete endogenous variables
reflecting the actions of other firms. We could envision characterizing a firms decision
with the following conditions: operate the most profitable format when that choice yields
a payoff that exceeds the payoff from not having a store and do not operate a store otherwise. In other words, given the actions of other firms in the market summarized by
Dim , firm i operates format f in market m if
if m (Dim , Xim , if m ; if ) > 0
and
if m (Dim , Xim , if m ; if ) > if 0m (Dim , Xim , if 0m ; if 0 )

(3)

and does not operate in the market if

if m (Dim , Xim , if m ; if ) < 0
and
if 0m (Dim , Xim , if 0m ; if 0 ) < 0

(4)

Since all firms are behaving optimally in response to one another in equilibrium, these
conditions imply a system of inequalities that must be simultaneously satisfied.20
These threshold conditions imply restrictions on the underlying observables. As such,
one might be tempted to use a discrete choice model for each firms decision using the
19

The normalization of the variance of uif m reflects the standard need to normalize the scale of
utility in discrete choice models. We also assume that the components of if m are uncorrelated across
i, f and m.
20
These conditions would be those associated with a pure-strategy Nash equilibrium of a simultaneous
move game.

15

threshold conditions to derive probabilities for each outcome and then selecting the parameter values that maximize the likelihood of the observed outcomes across markets.
However, such an approach is inappropriate since the decisions of competitors enter a
firms payoffs as explanatory variables, and the competitors decisions in turn depend
on the firms decision. Since a players actions depend on all the common knowledge
variables including the errors, a naive model that ignores this endogeneity would yield
inconsistent estimates.21
Suppose, however, that one looked to construct a joint probability that correctly
reflects the restrictions that are simultaneously imposed on all firms. The difficulty with
directly implementing this approach is that conditions (3) and (4) are sufficiently weak
that certain values of the unobservables could be consistent with multiple outcomes. In
other words, certain values of the unobservables may satisfy these conditions for a Target
supercenter or a Wal-Mart supercenter, but not both. The presence of these types of
multiple equilibria causes difficulties for estimation since the threshold conditions (3)
and (4) do not provide exact probabilities for the different outcomes, thereby precluding
application of maximum likelihood estimation.22
As discussed above, we resolve this difficulty by making an assumption on the sequence
of moves that guarantees a unique SPNE. Intuitively, such an assumption removes the
problem of multiple equilibria by allowing the first mover to gain the market in any
regions of multiplicity. Continuing the example from above, if Wal-Mart were the first
mover in a market that could support either a Target supercenter or a Wal-Mart supercenter, then a Wal-Mart supercenter monopoly would be the unique SPNE of the
sequential move game. We considered two alternative for the sequence of moves, both of
which implicitly reflect likely entry patterns over time for the chains. First, we allowed
the sequence of moves to be based on the distance of each firm from its headquarters
with the firm nearest to its headquarters moving first. Intuitively, this sequence allows
the order of moves to reflect the historic pattern of expansion for each firm outward from
its home territory. Second, we consider a sequence based on the actual store opening
dates. In this case, any firms that operate in a market made their decisions in an order
based on their actual entry dates with earlier entrants moving first. In markets where
two or more firms are not present so that entry dates for the chains without stores are
not defined, we assume that any existing firms moved first while the firms not present
21

We illustrate such inconsistencies in the results section.

Ciliberto and Tamer (2004) note that conditions (3) and (4) can be used to derive bounds on the
probabilities of different outcomes. They use these bounds as the basis of an estimation routine.
22

16

made decisions in an order based on their distance from their headquarters. In other
words, for a monopoly market, the monopolist moved first followed by the firm closer
to its headquarters among the two remaining firms. This second approach involves the
same ideas behind the first, but also exploits the fact that we do have information about
the actual timing of entry decisions. However, from a practical point of view, both of
these alternatives yielded similar results. As a result, we only present results associated
with the second move sequence.
The SPNE refinement implies additional equilibrium restrictions on payoffs. In general, the presence of multiple players each of whom have three actions implies that the
extensive form of the game is sufficiently complex that succinct characterization of all
equilibrium conditions is difficult. As a result, we solve the game numerically using
simulation methods as we discuss in the next section.

3.3

Numerical Analysis of Equilibrium and Estimation

We solve the game numerically by backwards induction. Given a set of parameter values
and a vector of simulated draws for the unobserved components of firms profits, we
calculate the payoffs of all firms under all market configurations. We can then find the
last movers best response to any of the nine combinations of actions by the other players.
For example, when the first mover chooses d and the second mover chooses s, we can
determine the third movers optimal response to the actions (d, s) as
R3 (d, s) = arg max a3 {0,d,s} 3 (d, s, a3 ) .

(5)

where we have suppressed the dependence of payoffs on explanatory variables, parameters

and unobservables to conserve space. We can repeat this process for any combination of
moves (a1 , a2 ) by the first two players to obtain R3 (a1 , a2 ). Turning to the second mover,
that firm anticipates the best response of the subsequent mover as required in a SPNE.
Hence, we evaluate the second movers best responses to possible actions by the first
mover given the implications for the optimal action of the last mover to obtain R2 (a1 ).
Again, if the first mover chooses d, the second movers best response will be defined as
R2 (d) = arg max a2 {0,d,s} 2 [d, a2 , R3 (d, a2 )] .

(6)

Finally, the first mover chooses its action to maximize its profits recognizing the optimal
responses of both later movers:
R1 = arg max a1 {0,d,s} 1 [a1 , R2 (a1 ) , R3 (a1 , R2 (a1 ))] .
17

(7)

The SPNE given the parameter values and random draws is thus (R1 , R2 (R1 ) , R3 (R1 , R2 (R1 ))).
As noted earlier, we use this algorithm, which implicitly checks the various inequalities
that must be satisfied in a SPNE, due to the complex inequalities that must be satisfied
to yield a particular outcome as a SPNE. Even if we derived analytical inequalities, we
would still face extremely complex regions of integration which would be exacerbated
by the flexible error structure that we employ. Our use of simulation and numerical
methods allows us to sidestep these problems. However, our approach does introduce a
computational burden as we must compute the SPNE separately for each set of random
draws.
We employ the previous algorithm in a maximum likelihood estimation routine. Given

a set a parameter values and a set of random draws (urm , urW m , urKm , urT m , urW md , urKmd , urT md , urW ms , urKms , u
r
we solve for the SPNE in market m. Letting Dm
denote the predicted SPNE decisions

for the rth random draws, we can repeat this procedure for R sets of random draws to
obtain a simulated probability of the observed outcome
R

1X
r
Pb (Dm |Xm ; ) =
1 (Dm
= Dm |Xm ; )
R r=1

(8)

that is consistent as the number of simulated draws grows large, where Dm represents the
observed decisions of the three firms and 1(.) equals one if the condition in parentheses is
true and zero otherwise. In other words, we use the fraction of times that our simulations
yield a SPNE that coincides with the actual outcomes to approximate the probabilities
of those events. Given these simulated probabilities, we can then form the loglikelihood
function
L (D, X; ) =

M
X

ln Pb (Dm |Xm ; )

m=1

and can estimate the parameters of the model using simulated maximum likelihood:
b = arg max L (D, X; ) .

(9)

Store Revenues and Market Structure

The estimates of the latent payoff functions from the model in the previous section
provide information about the factors that govern the long-run profits of the different
firms. As such, these estimates are interesting on their own. However, such analysis
can also serve as a useful input into an application in which one wishes to relate some
18

other market outcome to the observed market structure. Such applications have a long
history in economics and marketing as researchers have examined relationships between
market structure measures and prices, mark-ups, revenues, or profits (e.g., Weiss 1989,
Buzzard and Gale 1987, Manuszak and Moul 2007, Singh and Zhu 2007.) Similarly, one
could consider how product offerings (Cohen and Mazzeo 2004, Watson 2004), promotion
levels, product introduction, or product innovation vary with the extent of competition in
a market. Typically, this type of application involves a regression of the outcome variable
of interest on a market structure variable such as concentration ratios or the number of
firms.
Such studies embody what is termed the structure-conduct-performance paradigm in
economics. However, regressions of this type have been subject to substantial criticism.
Market structures are not randomly assigned, a necessary condition for a standard regression model to yield consistent estimates of the relationship between market structure
and an outcome variable. Of particular concern is the possibility that unobserved demand and cost conditions influence both the outcome of interest and the realized market
structure. Put briefly, market structure is endogenous, and any regression using market
structure as an explanatory variable must recognize that fact.23
One solution to this problem builds on the work of Heckman (1979) by constructing
correction terms for the regression model that explicitly account for the non-random determination of market structure. As proposed by Mazzeo (2002b), with a suitable model
to characterize the observed market structures, one could employ a two-stage estimation
approach in which the market structure model is estimated followed by estimation of a
regression model including correction terms derived from the first-stage estimation. Intuitively, the correction terms reflect the correlation between unobservable factors that
influence both the outcome variable and the realized market structure.
We employ this technique to estimate a regression model in which we examine how
store-level revenues for each chain depend on competition in a market. Such a regression
could be useful for a number of reasons. First, an analyst could simply wish to learn how
competition affects store revenues for these chains. Alternatively, such a model could
be used to generate forecasts for store-level revenues. Such forecasts would presumably
account for differences in competition across markets, and our approach provides a technique to correctly capture such effects. Finally, one could evaluate store-level performance
23

The composite nature of the error terms and the non-structural form of the regression model can
make it difficult to determine what biases might arise. See Singh and Zhu (2007) for a formal derivation
of some potential biases in a price-market structure application.

19

based on realized store revenues relative to some expected level of revenues. Again, an
appropriate characterization of expected revenues would require correct measurement of
the importance of competition in a market.

4.1

A Corrected Revenue Regression Model

Formally, we consider the following regression model for the revenues of firm i with format
f in market m:


q
ln qif m = Zm if + f Dm , if
+ qif m

(10)

In this model, qif m represents store revenues while Zm are exogenous market conditions


q
and qif m are unobservables that influence revenue. The function f Dm , if
represents
the impact of the underlying market structure described in Dm on revenues. As in our
market structure model, an important consideration is the possibility that the effect of
competition or the effect of market characteristics varies depending on firm identities and
q
formats. Consequently, we consider a flexible parameterization reflected in if and if
that allows such differential effects.
To reflect the endogeneity of market structure, we permit the unobserved components
in the revenues of firm i with store format f in market m to be correlated with the error
terms underlying firm is operation decisions. We assume that the error terms for the
sales volumes regression and the payoff function follow24
q

if m
0
2 if m im m
uif m

1
0
0

N 0 , if m
2
uim
0
0
im
0
im
2
m
0
0 m
0
um

(11)

where m , for example, is the covariance between qif m and um . Given this assumption,
qif m |uif m , uim , um is normally distributed with mean


m
im
E qif m |uif m , uim , um = if m uif m + 2 uim + 2 um .
im
m

(12)

Given some observed market structure D, we can express qif m as

qif m |D = if m E [uif m |D ] +

im
m
q
E [uim |D ] + 2 E [um |D ] + vif
m,
2
im
m

(13)

Note that we assume that qif m is uncorrelated with ujgm and ujm for all j 6= i and g 6= f . However,
the correlation between qif m and the market-wide unobservable, um , implies that this error terms is also
correlated with unobservable factors that influence the operation decisions of the other firms.
24

20

q
where vif
m is now a pure idiosyncratic mean zero error term. The key intuition is that
the expectation of qif m is different under different market structures which would imply

biased estimates of standard regression estimation of (10).

The condition that market structure D obtains implicitly involves restrictions on all of
the unobservables from the market structure model. These restrictions, in turn, depend
on parameters from that model. With estimates of the parameters from the market
structure model, we can compute implied values for the conditional expectations and can
introduce a correction term into our initial regression. Denoting the computed value of
E [uif m |D ], for example, with E [uif m |D ], we can consider


im
m
q
q
ln qif m |Dm = Zim if +f Dm , if
+if m E [uif m |Dm ]+ 2 E [uim |Dm ]+ 2 E [um |Dm ]+
vif
m
im
m
(14)
q
25
where vif
m is now mean zero conditional on Dm .

Computation of the correction terms is complex. Consider E [uif m |D ]. Letting ~m be

a vector stacking all of the unobservables if m = um +uim +uif m from the market structure
model and denoting the set of those unobservables consistent with market structure D
as (D), we can, after some manipulation, obtain
E [uif m |D ]

= E
m (D)]
if m |~
Z [u

uif m P [~m (D) |uif m ] f (uif m ) duif m

=
P [~m (D)]

(15)

Computation of this term is difficult due to both the irregular regions of integration
reflected in (D) and the integral required for the numerator of this expression. For
these reasons, we compute these correction terms through simulation methods.26 In our
actual implementation, we scale E [uim |D ] by im and E [um |D ] by m . Hence, the
correction procedure yields three additional parameters to be estimated:

m im
,
m im

and

if m .

Estimation Results

In this section, we present results from our model of endogenous market structure and
the corrected revenue regressions. Before doing so, we present two alternative market
q
q
Technically, the new error term in this regression, vif
m , is comprised of vif m as well as approximation
[uif m |D ]. However, both
error that arises due to the use of first-stage estimates to obtain values of E
of these components are mean zero conditional on D.
26
The details about the simulation procedure are available upon request from the authors.
25

21

structure models. The first involves a binary probit model in which each firm takes the
store operation decisions of its competitors as given. The second formulation takes the
sequential nature of operation decisions into account, but ignores any strategic considerations. In particular, early movers ignore the potential effect of their decisions on later
movers who, in turn, take previous decisions as exogenous so that
P (D1m , D2m , D3m |Xm , ) = P (D1m |Xm, ) P (D2m |D1m , Xm , ) P (D3m |D1m , D2m , Xm , )
where D1m , D2m and D3m represent the decisions of the first, second and third firms
respectively. We consider these two alternatives to our full model in order to illustrate the
importance of accounting for strategic interaction between the firms. Computationally,
such models are much simpler than our full model, and one could envision using a variant
of such a model to naively examine how the operation decisions of Kmart, for example,
vary with presence of Wal-Mart treating Wal-Marts presence as exogenous.
The estimates presented in Table 4 suggest that such a naive approach is problematic. While we omit estimates of the cost and demand variables to conserve space, we
generally find that most have reasonable signs. The problem arises in the estimates
of the parameters reflecting competitive effects. In both models, we find estimates of
competitive effects that are positive suggesting that the presence of competitors makes
a market more attractive. Such a finding is troublesome and reflects the treatment of
competitors actions as exogenous. Since the actions of all firms depend on all of the
common knowledge variables including the errors, the competition variables that appear
on the right hand-side and the error term have a non-zero correlation. In this case, rather
than recognizing that the presence of both Wal-Mart and Target, for example, can be
driven by attractive observed and unobserved conditions, the simple models attribute
their joint presence to a positive effect of one on the the other.
Table 5 presents the results from our full market structure model. The coefficients
from the model are somewhat difficult to interpret due to the reduced form nature of the
payoff function and the normalization of the variance of uif m . Nevertheless, the signs
and relative magnitudes of the coefficients are informative. For example, we find that
the population coefficients are positive and significant for all stores and formats indicating, not surprisingly, that all firms prefer markets with higher populations. Moreover,
the importance of population far outweighs considerations associated with other market
conditions as a one standard deviation increase in population has a much larger affect
on the attractiveness of a market than a similar change for any of the other variables.27
27

All non-indicator explanatory variables have been standardized to have a mean of zero and standard

22

The coefficients for many of the other variables suggest sharp differences between the
chains, often in ways that corroborate popular views of the industry. For example, neither
Wal-Mart nor Kmart are very concerned with median income in a market as the coefficient
estimates on income for both chains are negative, but often statistically insignificant.
Wal-Mart does, however, find markets with higher wages to be significantly less attractive.
In contrast, Target clearly finds more affluent markets to be more profitable even to
the point where it prefers markets with higher wages despite the higher costs. Further
contrasts arise in the coefficients for variables related to household composition and
automobile holdings. Wal-Mart prefers markets in which more households have children
and cars, while Target is less concerned with those variables. These findings support
the conventional wisdom that Wal-Mart prefers to locate on the outskirts of city limits
in family-oriented middle and lower income markets while placing a large emphasis on
controlling costs. In contrast, our results confirm that Target has attempted to position
itself as an upscale discounter that caters to a higher income clientele and, consequently,
finds relatively urban markets to be more attractive. Kmart appears to be less systematic
than its rivals in terms of the characteristics that it favors, perhaps reflecting a lack of
strategic focus on the part of that chain.28
The most notable asymmetry across payoffs can be seen in the competitive effects,
all of which are negative in Table 5 indicating the superiority of this model to the simple
approaches in Table 4. These results suggest that Wal-Mart has a disproportionate effect
on its two rivals. For both Kmart and Target, the largest competitive effects are those
associated with Wal-Mart, and the implied effect of a Wal-Mart is more than twice the
reverse effect that those firms have on Wal-Mart. Even for the smaller effects that WalMart experiences, however, the magnitude of these competitive effects is quite large,
requiring substantial improvements in other market characteristics to offset competition
between firms. No clear patterns appear across store formats, but the generally low
values of the effects associated with Kmart for the other two firms combined with larger
effects that they have on Kmart would suggest that Kmart is in a particularly tenuous
position.
One way to concretely interpret the implications of our findings is to consider the
population sizes required to support different configurations of firms. While we could
deviation of one.
28
Naturally, an interesting issue which is beyond the scope of the current paper would be to examine
whether Kmarts more recent operation patterns exhibit greater dependence on certain market characteristics. In addition, one could consider whether Wal-Mart and Target have remained dedicated to the
apparent strategies that we document.

23

perform similar calculations for other variables such as income, population is attractive
because it is clearly important and can take a wide range of values in actual markets.
With other market characteristics fixed at their average levels, we can compute threshold
population sizes at which latent payoffs under different market structures are zero.29
Following BR (1991a), variation in these breakeven market sizes across market structures
then provides concrete evidence about the nature of competition between these firms.
When we perform such calculations, we obtain a number of interesting findings. First,
in a typical non-MSA market, we find that both Wal-Mart formats would require essentially the same population, approximately 29,000 for a discount store and 30,000 for a
supercenter, to operate in the absence of any competitors. This result suggests that neither Wal-Mart format is clearly dominant for Wal-Mart in a typical market where it can
operate as a monopolist despite the great attention paid to the supercenter format in the
popular press.30 Interestingly, we find that a similar market size, 31,000, could support
a single Kmart discount store. The similarity between this value and those of the WalMart formats is notable since it suggests that Wal-Marts competitive advantage does not
necessarily lie in its ability to operate in markets that would otherwise be unattractive
to its rivals. In other words, the large number of Wal-Mart monopolies in our data does
not entirely reflect Wal-Marts superior ability to place stores in certain markets that
its rivals would find inherently unprofitable. Target, on the other hand, would require a
substantially larger population size, approximately 78,000 people, to operate as the sole
firm in a typical market. This last finding reinforces the fact that the typical market
nationwide is not the typical market for Target which instead focuses on markets with
particular characteristics.
However, while Wal-Marts ability to operate as a single firm does not differ drastically from that of Kmart, the crucial difference arises in settings where multiple firms
operate. As noted above, the competitive effects exhibit substantial asymmetries with
the largest effects running from Wal-Mart to its competitors and not vice versa. These
asymmetries have striking effects on the population at which the different firms would
be willing to operate in various market structures. For a typical non-MSA market, by
a population of 65,000, Wal-Mart would be willing to operate a discount store in the
presence of either a Kmart or Target discount store. However, Kmart would be willing
29

Of course, the realized market structure will depend on the values of the unobservables which this
calculation necessarily ignores.
30
This could reflect that the distinction between the two formats for Wal-Mart is not a first-order
concern since the supercenter format has become so ubiquitous while the discount store format has
evolved in ways that make it similar to the supercenter format.

24

to face Wal-Mart only with a population in excess of approximately 108,000 while Target
would require a population of roughly 140,000 which again emphasizes Targets strategic
decision to operate in atypical markets. Thus, for a typical market with a population between roughly 65,000 and 108,000, Wal-Mart would operate alone absent any substantial
unobserved shocks. Although Kmart could operate profitably on its own in many markets and may do so in particularly small markets that cannot also support a Wal-Mart, it
would not operate in those moderately sized markets which would also attract Wal-Mart.
Put differently, our results suggest that the large number of Wal-Mart monopolies in the
data reflects Wal-Marts crucial ability to weather competition to a greater extent than
its rivals, thereby detering its rivals from operating in many markets in the first place.
To assess whether our market structure model captures the main features of the data,
we used our model to conduct out-of-sample predictions for firms operation decisions.
For this exercise, we drew a random sample of 1,000 markets and re-estimated the model
with the remaining 6,698 markets. For the holdout sample, our overall hit rate for
this exercise was 67% which is quite reasonable considering that our predicted market
structure is based on identities of individual firms and not just the total number of active
firms. For example, in monopoly markets, the hit rates are not based on observed and
predicted monopolies, but are based on the observed and predicted number of monopoly
Wal-Mart, Kmart and Target markets. The hit rates are even higher (78%) if we ignore
the format choice of firms and focus just on identities.
Turning now to the store-level revenue regressions, we present results from that exercise in Table 6. To illustrate the importance of the correction term, we estimate the
model with and without that term; the top panel shows the competitive interaction
parameters from a basic linear regression that does not correct for the endogeneity of
market structure while the lower panel shows the results of the corrected model. Both
models also include demographic variables, but we omit those results to conserve space.
The results of this exercise yield a number of notable findings. First, the correction
procedure clearly matters. When we do not employ the correction term, we find that
Wal-Marts revenues for both formats are significantly higher when it competes with the
other firms. While the positive coefficients are not necessarily inconsistent as these are
revenues not profits, the large magnitude is a bit alarming.31 However, when we account
for the endogeneity of market structure, this problem is resolved to a large extent. The
corrected estimates suggest that Wal-Marts revenue decreases with the presence of other
firms. Moreover, the estimates for the two other firms fall (i.e. become more negative)
31

Recall that our outcome variable is the natural logarithm of revenues.

25

when the correction term is included. Such patterns likely reflect the fact that markets
which yield high revenue for unobserved reasons also tend to attract other firms. Overall,
the differences across the corrected and uncorrected regressions suggest that one could
draw very misleading conclusions about the determinants of store-level revenue without
correctly accounting for the endogeneity of market structure.
The parameter estimates from the corrected regressions mirror the competitive effects
from the market structure model. For Kmart, the effect of Wal-Mart on revenues is
substantially larger than the effect of Target. A similar pattern emerges for Targets
revenues with Wal-Mart again having the largest effect. These findings reinforce the
results of the market structure model of Wal-Mart as the most fearsome competitor
among these three firms. Moreover, the results again suggest the relatively vulnerable
position of Kmart. Whereas the estimates suggest that Target has a substantial effect
on the revenues of Wal-Mart stores, the effect of Kmart on Wal-Mart is much lower.
On their own, these parameter estimates provide information about competition in
the industry, again painting a picture in which Wal-Mart is dominant followed by Target
with Kmart struggling to find its way. However, the parameter estimates of revenues
accounting for the endogeneity of market structure not only reveal a more sensible pattern
of competition among retailers, but can also be used for the purposes of forecasting or
evaluation of store performance. In either case, an analyst would presumably account for
the impact of competition on an outcome variable such as revenue. Our results suggest
that correctly doing so would require a correction method such as the one which we
employ.
For example, suppose one wished to link managerial compensation to store performance. Typical measures of store performance that could be used for such a purpose are
net income, total store sales, and comparable store sales. The last of these involves an
assessment of a stores relative performance within a retail chain. In order to evaluate
relative performance, however, one must know the appropriate baseline level of store
sales which could be complicated by variation in local characteristics such as the number of competitors. A regression model relating store performance to market conditions,
including the presence of competitors, could provide such a baseline. An analysis of the
residuals for a particular store could then reveal whether it performed usually well or
poorly during the past year.
To determine baseline sales, our previous discussion of the importance of the correction term suggests that inclusion of that term could be important. And, in fact, we find
that it is. For example, out of the 313 Wal-Mart discount stores that face competition
26

from Kmart and Target, we find that a standard regression would incorrectly characterize
18 of these stores as poor performers relative to peer stores when they were, in fact, not
based on the implications of the corrected model. Similarly, a standard regression would
fail to identify 20 stores that performed relatively poorly, instead suggesting that their
performance was adequate. Hence, when performing a yardstick comparison of revenues
across stores, a standard regression model which ignores the endogeneity of market structure would misclassify the performance of approximately 10% of these Wal-Mart stores.
Subsequent managerial compensation and incentives could correspondingly be distorted.

Conclusion and Future Research

In this paper, we investigate competition between the three big box discount retailers:
Wal-Mart, Kmart and Target. Due to data limitations, we initially analyze how store
operation patterns for the chains vary in a cross-section of local markets. Using a unique
dataset, we estimate a model in which the firms play a strategic game and choose whether
to operate in a market and, if so, which format of store. By relating the observed outcomes to the equilibrium restrictions required for a particular market structure to obtain,
we estimate how the payoffs of the firms depend upon local market conditions and competitive considerations. These estimates provide insights into the trade-offs that firms
make in balancing favorable market conditions against higher competition. Notably, we
allow for substantial observed and unobserved heterogeneity across the firms. In particular, unlike previous empirical applications of models of endogenous market structure, the
small number of known firms allows us to permit the competitive effects to depend on
firm identities. We solve the model by backwards induction using numerical techniques
and estimate the model by simulated maximum likelihood.
We then use the results of this model as inputs into a more standard application
in which we relate one observed market-specific outcome for the firms, store-level revenues, to market characteristics. One important market characteristic that can influence
store-level revenues is the presence of competing firms. However, as our previous model
illustrates, market structures are not randomly assigned leading to concern that standard regression techniques could yield inconsistent estimates when applied to a model
that relate store-level revenues to market structure variables. To deal with this problem,
we present and implement a two-stage procedure in which we use estimates from the
market structure model to construct correction terms to account for the endogeneity of
market structure in the revenue regression.
27

Our estimates of both models reveal a number of asymmetries across the chains, some
of which are consistent with conventional views of the industry. For example, while all
firms prefer markets with larger population, higher income is more important to the
profitability of Target than it is for Wal-Mart or Kmart. Conversely, Wal-Mart places
more emphasis on household composition and costs reflected in retail wages than do
its rivals. Yet, any attractive market characteristics, such as larger population, must be
balanced against the concern that those features could encourage operation of rival firms.
We find that these competitive considerations are quite substantial for all of the chains.
However, Wal-Mart is clearly in a dominant position as it is able to face competition from
its rivals in much less attractive markets than its rivals are able to face it. In contrast,
Kmart appears to be in a more tenuous position as it has less of an effect on its rivals
than the reverse. Finally, Target appears to have effectively identified particular types of
markets which serve its brand positioning. Notably, however, we find that Kmart, were
it the sole store, would be able to operate in many of the same markets that Wal-Mart
finds attractive. In other words, Wal-Marts advantage is not necessarily its ability to
operate as a single firm in markets that would be unattractive to its rivals. Instead, the
possibility of competition from Wal-Mart deters its rivals from operating stores, but the
reverse effect is not as substantial.
Our revenue regressions illustrate the importance of the correction procedure as omission of the correction terms appears to understate the importance of competitors on storelevel revenues for each chain. Our corrected estimates again reveal important asymmetries across the chains. While Wal-Marts has a strong negative influence on the revenues
of its rivals and Target has a smaller, but still substantial negative effect on revenues
of Wal-Mart and Kmart, Kmart has a lesser effect on its rivals. These findings again
reinforce the conventional view of the industry in which Wal-Mart is dominant followed
by Target while Kmart is struggling to find its market niche.
There are, of course, several caveats to our study that suggest directions for future
research. While our model is rich enough to provide insight into the competitive landscape facing these firms, we make a number of assumptions to obtain a tractable, albeit
still complex model. First, we consider our local markets to be independent. Holmes
(2005) and Jia (2007) have relaxed this assumption to investigate issues associated with
cannibalization across stores and economies of density associated with the location of distribution centers. However, Holmes (2005) does not consider competitive effects across
chains while Jias (2007) model can yield a large number of equilibria. Techniques for
dealing with multiplicity to that extent are not well-developed although promising tech28

niques include those of Ciliberto and Tamer (2004) and Andrews et al. (2004) which are
based on bounds for the outcome probabilities. Second, we do not consider the location
of stores within individual markets. Characterizing the equilibria of such location games
is notoriously difficult, and multiple equilibria are again endemic to such models. Nevertheless, some recent work has begun to investigate such models (e.g. Seim 2006, Zhu
and Singh 2006), albeit in models of imperfect information. Finally, like most other papers in the literature, our model is static. It therefore misses the rich dynamics inherent
in strategic decisions like market entry and exit. As discussed in Section 2 above, the
three chains in the discount store industry started in the same year. Four decades since
its inception, the industry has evolved with one its the player Wal-Mart becoming
the worlds largest company, while another (Kmart) filed for Chapter 11 bankruptcy. It
would be interesting to study the competitive dynamics in this industry using some of
the recently proposed methodologies on multi-agent dynamic games (Aguirregabiria and
Mira 2004, Pakes et al. 2004, Pesendorfer and Schmidt-Dengler 2003). Although applications of these methodologies pose significant challenges in terms of data requirements
and computational difficulties, with advancements in computing power these hurdles can
be overcome in the future.

29

References
Aguirregabiria, V. and P. Mira (2004), Sequential Estimation of Dynamic Discrete
Games, Working Paper, Boston University.
Andrews, D. W. K., S. Berry, and P. Jia (2004), Confidence Regions for Parameters
in Discrete Games with Multiple Equilibria with an Application to Discount Chain Store
Location, Working Paper, Yale University.
Bajari, P., Hong, H., and Ryan, S., (2004), Identification and Estimation of Discrete
Games of Complete Information, Working Paper, Duke University.
Basker, Emek (2005), Job Creation or Destruction? Labor-Market Effects of WalMart Expansion, Review of Economics and Statistics, 87 (1), 174-83.
Berry, S. T. (1992), Estimation of a Model of Entry in the Airline Industry, Econometrica, 60 (4), 889-917.
Bell, D. and J. Lattin (1998), Shopping Behavior and Consumer Preference for Store
Price Format: Why Large Basket Shoppers Prefer EDLP, Marketing Science, 17 (1),
66-88
Bresnahan, T. F. and P. C. Reiss (1990), Entry in Monopoly Markets, Review of
Economic Studies, 57 (4), 531-53.
(1991a), Entry and Competition in Concentrated Markets, Journal of Political
Economy, 99 (5), 977-1009.
(1991b), Empirical-Models of Discrete Games, Journal of Econometrics, 48 (1-2),
57-81.
Buzzell, R.D.and B.T. Gale (1987), The PIMS PrinciplesLinking Strategy to Performance. The Free Press, New York.
Ciliberto, F. and E. Tamer (2004), Market Structure and Multiple Equilibria in
Airline Markets, Working Paper, Princeton University.
Cohen, A. and M. Mazzeo (2004), Market Structure and Competition among Retail
Depository Institutions , Working Paper, Northwestern University.
Ghemawat, P. (1986), Wal-Mart Stores Discount Operations, Harvard Business
Case No. 9-387-018.
Ghemawat, P. and G. Friedman (1999), Wal-Mart in 1999, Harvard Business Case
No. 9-799-118.
Heckman, J. J. (1979), Sample Selection Bias as a Specification Error, Econometrica, 47 (1), 153-61.
Holmes, Thomas (2005), The Diffusion of Wal-Mart and Economies of Density,
Working Paper, University of Minnesota.
Jia, P. (2007), What Happens When Wal-Mart Comes to Town: An Empirical
Analysis of the Discount Retailing Industry, Working Paper, MIT.
Manuszak, M. D. and C. C. Moul (2007), Prices and Endogenous Market Structure
in Office Supply Superstores, Working Paper, Washington University.
Mazzeo, M. J. (2002a), Product Choice and Oligopoly Market Structure, Rand
Journal of Economics, 33 (2), 221-42.
30

 (2002b), Competitive Outcomes in Product-differentiated Oligopoly, Review of

Economics and Statistics, 84 (4), 716-28.
McFadden, D. (1982), Qualitative Response Models, in Hildenbrand, W. (ed.),
Advances in Econometrics, Cambridge University Press, Cambridge.
Miller, G. S. and C. F. Noe (2000), Sears, Roebuck and Co. vs. Wal-Mart Stores,
Inc., Harvard Business Case No. 9-387-018.
Noe, C. F. (1998), Kmart Corp., Harvard Business Case No. 9-199-017.
Pakes, A., M. Ostrovsky, and S. Berry (2004), Simple Estimators for the Parameters of Discrete Dynamic Games with Entry/Exit Examples, Working Paper, Harvard
University.
Pesendorfer, M. and P. Schmidt-Dengler (2003), Identification and Estimation of
Dynamic Games, Working Paper, London School of Economics.
Seim, K. (2006), An Empirical Model of Firm Entry with Endogenous Product-Type
Choices , Rand Journal of Economics, 37 (3), 619-42.
Singh, V., K. Hansen, R. Blattberg (2006), Market Entry and Consumer Behavior:
An Investigation of a Wal-Mart Supercenter, Marketing Science, 25 (5), 457-476
Singh, V. and T. Zhu (2007), Pricing and Market Concentration in Oligopoly Markets, Working Paper, Carnegie Mellon University.
Sweeting, A. (2004), Coordination Games, Multiple Equilibria and the Timing of
Radio Commercials, Working Paper, Northwestern University.
Sudhir, K. (2001), Structural Analysis of Manufacturer Pricing in the Presence of a
Strategic Retailer, Marketing Science, 20 (3), 244-64.
Watson, R. (2004), Product Variety and Competition in the Retail Market for Eyeglasses, Working Paper, University of Texas at Austin.
Weiss, L. (1989), Concentration and Price. The MIT Press, Cambridge, MA.
Zhu, T. and V. Singh (2006), Spatial Competition with Endogenous Location Choices:
An Application to Discount Retailing, Working Paper, Carnegie Mellon University.

31

Table 1: Number of Stores in the United States

Store Names

Number of Stores

Kmart Discount Store

Super Kmart Center

1425
59

Target Discount Store

SuperTarget Center

1133
127

Wal Mart Store

Wal Mart Supercenter

1490
1442

Table 2: Summary Statistics of the Market Characteristics

Varaibles

MEAN

Population
Income
Novehicle
Childratio
Retail Wage
Num of Supermarket
Electricity Price
Tax Rate
Dummy of MSA
Distance to Wal-Mart Dist Ctr
Distance to Kmart Dist Ctr
Distance to Target Dist Ctr
Dist to Wal-Mart HQ
Dist to Kmart HQ
Dist to Target HQ

36,722
41,967
0.10
0.23
22,130
302
8.12
5.42
0.67
95
138
137
816
740
898

STD

MIN

39,526 5,004 607,599

16,753 4,127 166,369
0.11 0.00
0.79
0.04 0.00
0.54
4,431 8,998 44,513
447
6
1935
2.13 5.29
12.80
1.33
0
7.25
0.47
0
1
61
0
546
109
0
706
82
0
652
380 1.94
1,712
561 1.81
2,201
382 0.54
1,708

N = 7698

Table 3: Market Characteristics by Market Structure

Market Structure Population Income

(0,0,0)

23,534

41,124 4,379

(W,0,0)
(0,K,0)
(0,0,T)

35,883
41,227
54,340

37,429 1,420
43,636
414
52,003
260

(W,K,0)
(W,0,T)
(0,K,T)

51,016
87,265
76,857

39,201
49,535
50,462

467
307
120

(W,K,T)

113,951

47,844

331

Total

MAX

7698

of
of
of
of
of
of

Wal-Mart Store
Wal-Mart Sup
Kmart Store
Kmart Sup
Target Store
Target Sup

0.346
0.248
0.405
-0.224

0.047
0.187
0.057
0.153

0.303
0.318
0.217
0.362

0.051
0.205
0.065
0.151

0.486
0.427

0.440
-0.139

0.056
0.220
0.073
0.274

0.156
0.261
-0.008
-0.268

0.063
0.233
0.089
0.268

0.450
0.314
0.487
0.385

Presence
Presence
Presence
Presence
Presence
Presence

Wal-Mart Store
Wal-Mart Sup
Kmart Store
Kmart Sup
Target Store
Target Sup

Parameter

of
of
of
of
of
of

Wal-Mart
Store
Supercenter
Est
Std Err
Est
Std Err

0.280
0.364

0.315
-0.042

0.147
0.151

0.137
0.325

0.106
0.055

0.328
-0.717

0.070
0.061

0.077
0.193

0.141
0.000

0.230
-0.077

0.232
0.197

0.223
0.378

Kmart
Store
Supercenter
Est
Std Err
Est
Std Err

0.050
0.052

0.055
0.145

Kmart
Store
Supercenter
Est Std Err
Est Std Err

Table 4-2: Estimated Parameters Probit Model with Sequence of Move

Presence
Presence
Presence
Presence
Presence
Presence

Parameter

Wal-Mart
Store
Supercenter
Est Std Err
Est Std Err

Table 4-1: Estimated Parameters Probit Model

0.058
0.063
0.051
0.190

0.452
0.844
0.061
-0.361

0.150
0.144
0.113
0.395

0.360
0.067
0.319
0.530

0.072
0.067
0.058
0.212

0.391
0.636
0.016
-0.776

0.163
0.135
0.128
0.622

Target
Store
Supercenter
Est
Std Err
Est
Std Err

0.555
0.403
0.489
0.485

Target
Store
Supercenter
Est Std Err
Est Std Err

-13629.1

2.391

Std of Market Component

Log Likelihood

0.731

Std of Firm Random Component

Wal-Mart Store
Wal-Mart Supercenter
Kmart Store
Super K
Target Store
Super Target

-0.641
-0.924
-0.468
-0.368

of
of
of
of
of
of

Presence
Presence
Presence
Presence
Presence
Presence

0.024

0.032

0.046
0.071
0.052
0.120

-0.572
-0.570
-0.769
-0.395

0.048
0.051
0.052
0.136

0.045
0.031
0.043
0.032
0.048
0.043
0.028
0.057
0.047
0.056
0.031
0.032

Intercept
Population
Income
Childratio
Novehicle
Retail Wage
Tax Rate
Num SuperMarket
Electricity Price
MSA
Distance to HQ
Distance to Dist Ctr

0.167
0.994
-0.223
0.181
-0.653
-0.232
-0.006
-0.316
0.057
-0.100
-0.373
-0.253

0.133
0.705
-0.042
0.142
-0.227
-0.424
-0.025
-0.213
0.040
-0.417
-0.231
-0.068

Parameters
0.052
0.025
0.039
0.032
0.045
0.041
0.028
0.044
0.040
0.060
0.036
0.028

Wal-Mart
Store
Supercenter
Est
Std Err
Est
Std Err

Table 5: Estimated Parameters Market Structure Model

1.411

-1.511
-1.326

-1.329
-0.562

0.111
0.771
-0.017
0.031
-0.153
-0.017
-0.040
-0.089
0.024
-0.175
-0.078
-0.100

0.036

0.046
0.046

0.038
0.051

0.044
0.023
0.028
0.026
0.030
0.035
0.023
0.028
0.024
0.041
0.023
0.027
-1.664
-1.420

-0.662
-0.838

-0.989
0.754
-0.007
0.106
-0.214
0.046
-0.023
0.008
-0.050
-0.430
-0.264
-0.202
0.093
0.093

0.078
0.158

0.072
0.020
0.033
0.034
0.031
0.042
0.046
0.011
0.041
0.086
0.029
0.053

Kmart
Store
Supercenter
Est
Std Err
Est
Std Err

0.888

-1.304
-1.282
-0.679
-0.328

-0.860
0.828
0.294
-0.014
-0.015
0.125
0.009
0.051
-0.026
-0.001
-0.183
-0.033

0.021

0.048
0.057
0.041
0.138

0.063
0.010
0.016
0.010
0.021
0.010
0.018
0.020
0.016
0.060
0.012
0.030

-1.742
-1.578
-1.672
-0.379

-0.899
0.988
0.242
0.078
-0.078
0.059
-0.126
-0.150
-0.026
-0.515
-0.028
-0.157

0.103
0.096
0.101
0.181

0.104
0.033
0.021
0.011
0.046
0.053
0.040
0.054
0.037
0.108
0.034
0.119

Target
Store
Supercenter
Est
Std Err
Est
Std Err

0.050
0.093
0.051
0.069

-0.088
0.052
-0.204
-0.196
0.086
-0.053
0.349

Market Firm Format Error

Firm Error
Market Error

0.028
0.037
0.046

0.025
0.087
0.030
0.064

0.215
0.336
0.106
0.015

Without Error Correction

Presence of Wal Store
Presence of Wal Sup
Presence of Kmart Store
Presence of Kmart Sup
Presence of Target Store
Presence of Target Sup
With Error Correction
Presence of Wal Store
Presence of Wal Sup
Presence of Kmart Store
Presence of Kmart Sup
Presence of Target Store
Entry by Target Sup
-0.084
-0.047
0.183

-0.060
-0.057
-0.068
-0.145

0.114
0.110
0.100
0.008

0.049
0.042
0.051

0.052
0.095
0.054
0.065

0.023
0.084
0.028
0.051

Wal-Mart
Store
Supercenter
Est
Std Err
Est
Std Err

Table 6: Estimated Parameters Revenue Regression

0.028
0.056
0.110

-0.181
-0.168

-0.112
-0.095

-0.094
-0.081

-0.032
-0.034

0.030
0.029
0.036

0.041
0.039

0.035
0.049

0.018
0.019

0.017
0.040

-0.004
-0.036
-0.065

0.088
0.002

0.191
0.297

0.015
-0.059

0.141
0.239

0.111
0.091
0.100

0.153
0.143

0.114
0.199

0.089
0.099

0.075
0.166

Kmart
Store
Supercenter
Est
Std Err
Est
Std Err

0.017
0.003
0.073

-0.109
-0.148
-0.060
-0.040

-0.022
-0.071
0.005
0.017

0.012
0.011
0.015

0.022
0.021
0.018
0.032

0.012
0.012
0.010
0.027

0.049
0.041
-0.015

0.139
0.036
0.022
-0.441

0.114
0.002
-0.015
-0.456

0.078
0.056
0.070

0.128
0.122
0.095
0.191

0.092
0.089
0.055
0.179

Target
Store
Supercenter
Est
Std Err
Est
Std Err

6
0
0

1142
130
186

27
14
16

35
21
29

20
11
14

16
8
5

HI
6
7
0

43
23
37

23
18
9

11
10
7

9
9
2

26
15
8

50
20
31

22
8
11

10
10
4

8
7
4

284
20
35

53
13
14
82
10
10

46
33
66

53
25
22

82
12
10

79
6
4

114
29
26

71
39
29

75
31
12

85
29
10

092/ 39/ 23

87
45
4
37

105
88
44

Wal-Mart Stores
Kmart Stores
Target Stores

64
9
2

116
70
64

68
92
52

103
40
40

29
17
3

158
113
79

60
29
15

78
62
37

101
47
3
32

77
45
31

104
106
31

Figure 1: Distribution of Stores in the United States.

Distribution of Stores in the United States

34
43
29

4
0
0

MD
39
24
24

NH
26
9
5

MA
42
20
20

7
6
2

RI
8
CT
1
30
2
9
11 DE

22
6
1