Measuring competition in spatial retail
| Author | Oleksii Khvastunov,Paul L.E. Grieco,Paul B. Ellickson |
| DOI | http://doi.org/10.1111/1756-2171.12310 |
| Date | 01 March 2020 |
| Published date | 01 March 2020 |
RAND Journal of Economics
Vol.51, No. 1, Spring 2020
pp. 189–232
Measuring competition in spatial retail
Paul B. Ellickson∗
Paul L.E. Grieco∗∗
and
Oleksii Khvastunov∗∗
Wepropose and estimate a spatially aggregateddiscrete-choice model with overlapping consumer
choice sets and demographic-driven heterogeneity that varies by chain. Our approachavoids the
need to define markets ex ante and capturesrich substitution patterns, even in the absence of price
data. An application to the US grocery industry illustrates the importance of location, format,
and the spatial distribution of consumers in shaping the competitive environment. Contrary to
conventional wisdom, we find substantial cross-format competition between supercenters, clubs,
and traditional grocers. Finally, we evaluate two representative mergers between supermarket
chains to demonstrate how our estimates inform antitrust policy.
1. Introduction
Measuring competition requires specifying the relevant consumer choice set, accounting
for overlap in ownership across products, and identifying the degree of substitution between the
relevant options. In the case of spatial retail competition, both the choice set and the extent of
differentiation depend on how consumers value the trade-off between travel distance and store
features (assortment, amenities, and prices). How far are consumers willing to travel to obtain
better or cheaper products? Also, how are these choices shaped and constrained by the retail
environment that surrounds them? To answer these questions, we propose and estimate a model
of spatial demand that flexibly captures these trade-offs and provides concrete measures of both
substitution patterns and competitive pressures. We illustrate our approach by evaluating the
antitrust implications of two high-profile grocery mergers.
Although one could, in principle, tackle the richness of spatial retail demand (e.g., firms
offering distinct bundles of heterogeneous products to multihoming shoppers choosing howoften
to shop and how much to buy) with a fully articulated structural model of supply and demand,
∗University of Rochester; paul.ellickson@simon.rochester.edu.
∗∗The Pennsylvania State University; paul.grieco@psu.edu,oleksii.khvastunov@gmail.com.
We are grateful for insightful comments on earlier drafts from Chris Conlon, Jessie Handbury, Charles Murry, Mark
Roberts, Marc Rysman, Steven Salop, and two anonymous referees. We also wish to acknowledge seminar participants
at Chicago Booth, the Haas School of Business, the University of Rochester, the University of Hong Kong,Penn State,
the 2016 NYU Stern IO day,and the 2016 Souther n Economic Association meetings in WashingtonDC.
C2020, The RAND Corporation. 189
190 / THE RAND JOURNAL OF ECONOMICS
we take a simpler approach that abstracts away from much of this complexity while retaining
the ability to flexibly estimate substitution patterns between stores. To do so, we exploit two key
institutional features of modern retail chain competition, namely, that prices and assortments
are primarily firm-rather than store-level decisions (Della Vigna and Gentzkow, 2017; Hitsch,
Hortac¸su, and Lin, 2020). This regularity allows us to capture several of the key aspects of
the underlying consumer choice problem via chain-level fixed effects, interacted with income,
that reflect the overall pricing, assortment, and service levels of rival outlets, without requiring
information on their actual values. Although the significant reduction in complexity (and data
requirements) this provides is not without cost, it permits a relatively simple framework for
analyzing spatial competition that still captures the rich manner in which demand is allocated
across space. Absent information on prices, we exploit the rich covariationbetween store locations
and local consumer demographics to pin down substitution patterns and quantify the extent of
market power at varying levels of geographic aggregation.1
This article fits between two existing approaches for estimating retailer demand, each of
which have important limitations in this context. Traditional aggregate-demand models, which
divide the spatial landscape into independent markets, are forced to ignore distance effects, as
measuring these would require specifying overlapping choice sets. Although micro-level choice
models can include direct information on travel distance, choice-based sampling often leads
individual-level analyses to omit several components of the choice set—namely, the stores that
were not visited—thereby obscuring the role of competition. Estimating retail demand is further
complicated by the large number of products and complex set of prices that comprise a given
shopping basket, which often requires the construction of ad hoc price indices and/or the omission
of relevant components of the product line.
To address these limitations, we develop a model of spatial demand that links store-level
aggregate revenues to consumer choices by exploiting the relationship betweenthe exact location
of every store and the distribution of consumer demographics in the residential geography that
surrounds it. Our empirical model builds upon and extends the approach proposed by Holmes
(2011) for sales volumes at Wal-Mart outlets, by incorporating competition from rival firms and
multiple store formats.2This approach is similar in spirit to Davis (2006), who aggregates over
geographic areas to estimate spatial demand for movie theaters, a specialized retail environment
where firms sell a relatively small number of products at fixed prices. The spatially aggregated
discrete-choice framework we propose extends the simple nested-logit model of store choice,
in which stores are grouped into nests according to retail format (e.g., club store, supermarket,
or supercenter), by aggregating over spatially heterogeneous consumer types. Shoppers differ
by their location (census tract), income, family size, and vehicle ownership status, allowing
us to capture flexible substitution patterns across stores by leveraging geographic variation in
these observed preference shifters. Consumers in each location allocate grocery expenditures
across a group of nearby retail outlets that are distinguished by their in-store amenities, distance
from the consumer’s location, and chain affiliation. Critically, consumers’ utility for all store
characteristics—including chain affiliation—are allowed to vary with observed demographics,
such as income and vehicle ownership. This flexibility ensures the ability to compute highly
localized measures of competitive overlap, including the store-and chain-level diversion ratios
that are key inputs to competition policy.
1A primary motivation for pursuing this alternative source of variation is the challenge of obtaining price data for
all outlets. However,the widespread practice of uniform grocery pricing (e.g., Thomassen et al., 2017) suggests that price
effects may not be well identified from cross-sectional variation alone, evenif prices were observed. In this sense, our
approach complements Conlon and Mortimer (2013), who use variation in the availability of products in lieu of price
changes to measure diversion
2Holmes’ analysis of Wal-Martabstracted from competition to focus on how the dynamics of Wal-Mart’sexpansion
decisions were influenced bythe scale economies associated with operating a dense network of stores. A similar approach
is used by Seim and Waldfogel(2013) to model the Pennsylvania state liquor monopoly.
C
The RAND Corporation 2020.
ELLICKSON, GRIECO AND KHVASTUNOV / 191
As location is a primary driver of store choice, a consumer’s utility for shopping at a given
store depends on their distance from the store and whether they own a car. Income impacts
consumer spending in two ways, first through the overall budget allocated to food and second,
through the stores they choose to shop at. The model captures the well-known tendency for a
consumer’s share of income spent on food at home to fall with income (e.g., via either basic
“Engel’s law”-type effects or the fact that wealthyconsumers tend to spend propor tionatelymore
on food outside the home), and the mix of stores they frequent to change as well (e.g., high-income
consumers are more likely to shop at Whole Foods than at Aldi). By leveraging the rich spatial
distribution of stores and consumer demographics, the model delivers a clear characterization of
which consumers prefer which stores, the competitive impact that stores exert on one another,
and the degree of substitution with between stores and the outside option.
Using a store-level census of the grocery industry, we demonstrate that this model can
identify the key determinants of store choice and competitive overlap in this complex setting.
Our results highlight the importance of distance, format, and consumer heterogeneity in shaping
the competitive landscape of retail trade. Consistent with earlier studies, we find that consumers
prefer to travel relatively short distances for groceries. This disutility of distance increases
quickly with income but is moderated by vehicle ownership (Smith, 2006; Eizenberg, Lach,
and Yiftach, 2016). Although markets are geographically localized, store type matters as well.
Consumers, particularly less affluent ones and almost exclusively those with cars, are willing to
travel significantlyf arther to shop at club stores rather than traditional supermarkets. This accords
with the empirical analysis of Lagakos(2016), who finds that car ownership alone explains roughly
two thirds of the cross-country differences in the relative use of modern (and far more efficient)
big-box retail technologies, and provides further evidence that the benefits of these innovations
may be regressive (Lagakos, 2016; Atkin, Faber, and Gonzalez-Navarro, 2018; Eizenberg, Lach,
and Yiftach, 2016), perhaps even leaving some transportation-constrained consumers un-served.
The greater spatial reach enjoyed by club stores also has direct implications for merger policy.
Due to their more limited product selection, club stores have previously been excluded from the
competitive set (Hosken, Olson, and Smith, 2012); our analysis suggests that this is a mistake, as
club stores now compete on relatively equal footing with conventional supermarkets.
We illustrate how our approach can inform antitrust policy in practice by examining two
high-profile mergers. First, werevisit the contentious Whole Foods and Wild Oats case, which was
challenged (but eventuallyapproved) in 2007. We then consider another merger,between Delhaize
and Ahold, that was recently approved in 2016. Our analysis reveals that the grocery industry
contains many submarkets with localized competition: chains of the same format compete more
strongly,b ut chains that targetspecific income segments are able to compete in relatively distinct
niches. However, there remains substantial overlap, with traditional grocery stores attracting a
diverse set of customers, and competing intensively with most stores in their local catchment
area. Consistent with this latter point, we find evidence supporting the court’s opinion that “when
Whole Foods does enter a new market where Wild Oats operates, Whole Foods takes most of
its business from other retailers, not from Wild Oats” (Varner and Cooper, 2007). This finding
is driven by the high degree of substitution between premium organic firms and conventional
supermarkets for most consumers. Due to this substitution, we find that only a tiny fraction of the
tracts in which Whole Foods and Wild Oats overlapped should have raised antitrust concerns.
On the other hand, our store-choice model also revealsstrong cross-format competition from
supercenters and club stores, which has important implications for the second merger weconsider.
We find that club stores represent significant competitors to traditional grocers, due in large part
to consumers’ greater willingness to travel to them. We illustrate the importance of including
clubs in the market by comparing the combined position of the two merging chains (Ahold and
Delhaize) both with and without including club stores in the analysis. Wefind that the number of
tracts where the thresholds provided in merger guidelines would raise antitrust concerns is over
20% higher when clubs are ignored. This finding is intuitive,as the ex ante exclusion of club stores
results in a model that overlooks the presence of significant substitute outlets and consequently
C
The RAND Corporation 2020.
To continue reading
Request your trial