Productivity Dynamics and the Role
of “Big-Box” Entrants in Retailing∗
Florin Maican†and Matilda Orth‡
December 10, 2014
Abstract
Entry of large (“big-box”) stores along with a drastic fall in the number of stores are striking trends in retail. We use a dynamic model to
measure the impact of large entrants on productivity, allowing for a controlled productivity process and accounting for prices, local markets, and
the endogeneity of entry. Using data on all retail food stores in Sweden, we
find that incumbents’ productivity increase after large entry and that the
magnitude of the increase declines with the productivity of incumbents.
Our findings highlight that large entrants play a crucial role in driving
productivity growth.
Keywords: Retail markets; imperfect competition; industry dynamics; productivity; dynamic structural model.
JEL Classification: C24, L11, L50, L81, O3.
∗
We would like to thank Daniel Ackerberg, Victor Aguirregabiria, Mats Bergman,
Jan De Loecker, Pierre Dubois, Martin Dufwenberg, Lennart Hjalmarsson, Randi Hjalmarsson, Jordi Jaumandreu, Amil Petrin, Vincent Réquillart, Rune Stenbacka, Johan
Stennek, Måns Söderbom, and seminar participants at Toulouse School of Economics
and the University of Gothenburg for valuable comments and discussions. In addition, we thank participants at EEA 2008 (Milano), EARIE 2007 (Valencia), the Nordic
Workshop in Industrial Organization 2007 (Stockholm), the Conference of the Research
Network on Innovation and Competition Policy 2007 (Mannheim), and the Swedish
Workshop on Competition Research 2007 (Stockholm) for helpful comments and suggestions. Special thanks to the Trade Union Institute for Economic Research (FIEF)
and the Swedish Retail Institute (HUI) for providing the data. We gratefully acknowledge financial support from the Swedish Competition Authority and the Jan Wallander
and Tom Hedelius Foundation.
†
Research Institute of Industrial Economics (IFN) and University of Gothenburg,
Box 640, SE-405 30, Göteborg, Sweden, Phone +46-31-786 4866, Fax +46-31-786 4154,
E-mail: [email protected]
‡
Research Institute of Industrial Economics (IFN), Box 55665, SE-102 15, Stockholm,
Sweden, Phone +46-8-665 4531, Fax +46-8-665 4599, E-mail: [email protected]
1
Introduction
Entry and competition are widely viewed as important to productivity
growth.1 Although entry and exit have been found to play a more crucial role for labor productivity growth in retail than in manufacturing,
there have been few attempts to estimate multi-factor productivity in retail
markets (Foster et al., 2006). Recently developed methods of estimating
production functions have in fact been applied almost exclusively to manufacturing industries.2 The major structural change in retail markets during
the last few decades is the entry of large (“big-box”) stores along with a
drastic fall in the number of stores. The most striking example of this is the
expansion of Walmart, which has been found to greatly lower retail prices
and increase the exit of retail stores in the U.S., the “Walmart effect.”3
For instance, the number of single-store retailers in the U.S. declined by 55
percent from 1963 to 2002 (Basker, 2007). Retail markets in Europe also
follow the “big-box” trend, although on a smaller scale, for example, with
Carrefour, Metro, Schwartz, and Tesco. Despite this significant structural
change, its impact on productivity has received little attention. Our goal
is to estimate multi-factor productivity in retail markets and measure the
impact of increased competition from the entry of large stores on incumbents’ productivity while controlling for prices and local markets.
A central contribution of this paper is that we provide a modeling framework to estimate multi-factor productivity in retail markets and to identify
heterogeneous responses of large entrants on productivity. A key advantage
of our study is that we characterize the full distribution of changes in store
productivity from large entrants in local markets. Specifically, we examine
the influence of large entrants on changes in the distribution of local market
productivity, aggregate weighted productivity in local markets, and exit.
1
Aghion and Griffith (2005) and Syverson (2011) provide excellent surveys of recent
literature.
2
Olley and Pakes (1996), Levinsohn and Petrin (2003), Ackerberg et al. (2006),
Ackerberg et al. (2007), and Doraszelski and Jaumandreu (2013).
3
Basker (2005), Basker (2007), Jia (2008), Basker and Noel (2009), and Holmes
(2011). Fishman (2006) and Hicks (2007) provide a general discussion on the Walmart
effect.
1
Detailed data on all retail food stores in Sweden from 1996 to 2002 provide
us with a unique opportunity to investigate the questions at hand.
Retail food is important to analyze because it accounts for 15 percent of
consumers’ budgets (Statistics Sweden, 2005) and thus constitutes a large
share of retailing. The Swedish market follows two crucial trends common
among nearly all OECD countries. One of these is a structural change toward larger but fewer stores; in fact, the total number of stores in Sweden
declined from 36,000 in the 1950s to less than 6,000 in 2003 (Swedish National Board of Housing, Building, and Planning, 2005). The second trend
is the presence of entry regulations, giving local authorities the power to
determine whether a store can enter the local market. Our results are important for policymakers who implement such entry regulations in local
markets. In fact, the consequences of retail regulations (e.g., supermarket
dominance) are frequently debated among policymakers in Europe (European Parliament, 2008; European Competition Network, 2011; European
Commission, 2012).
The paper relates to three strands of literature. First, it links to the
growing literature on retail productivity (Foster et al., 2006; Schivardi and
Viviano, 2011; Basker, 2012; Basker, 2014). In detail, we add to the scarce
literature on how to measure and understand heterogeneity in multi-factor
productivity in retail markets. The second focuses on dynamic models
of productivity heterogeneity within industries (Ericson and Pakes, 1995).
Such studies have found that increased competition from high productivity
entrants forces low productivity firms to exit, increasing the market shares
of more productive firms (Syverson, 2011). The productivity distribution is
thus truncated from below, increasing the mean and decreasing dispersion
(Syverson, 2004; Asplund and Nocke, 2006). Third, the paper relates to
the vast literature on how competition affects productivity,4 to which we
propose a complementary approach to quantify effects of entry on produc4
Previous theoretical work has emphasized both positive and negative effects, while
empirical work has often emphasized positive effects. Examples of recent contributions
are MacDonald (1994), Nickell (1996), Aghion and Griffith (2005), and De Loecker
(2011). Examples of papers that examine the effect of liberalization and competition
policy on productivity growth include Bertrand and Kramarz (2002), Djankov et al.
(2002), Buccirossi et al. (2013), Maican and Orth (2014), and Sadun (2014).
2
tivity.
The model incorporates the following key features of retail markets.
First, stores operate in local markets. Second, large entrants causally influence store productivity, i.e., increased competition from large entrants
forces stores to improve their productivity and induces exit. Our measure of large entrants is the number of large stores that enter local markets in each period. Third, lack of data on prices and quantities at the
firm/establishment level is common in many industries, especially in retail, due to difficulties in measuring output (Griffith and Harmgart, 2005;
Reynolds et al., 2005). We augment the production function with a simple
horizontal product differentiation demand system (CES), where exogenous
demand shifters and large entrants affect prices, and obtain an industry
markup (Klette and Griliches, 1996). We analyze the identification of the
effect of large entrants on both technical productivity (shocks to productivity separate from demand) and quality-adjusted productivity (the sum of
technical productivity and remaining shocks to demand).5 Fourth, to proxy
for store productivity, we use the labor demand function from stores’ shortrun optimization problem together with high-quality data on store-specific
wages.6
The role of large entrants is directly linked to competition policy because of entry regulations in the OECD, although such regulations are much
5
Most studies of imperfectly competitive industries that use sales or value-added as
a measure of output do not control for unobserved prices, although some examples exist
(Katayama et al., 2009, Levinsohn and Melitz, 2006, De Loecker, 2011, Doraszelski and
Jaumandreu, 2013). Product-level data on prices and quantities have recently been used
together with establishment level data to estimate production functions and demand in
manufacturing. Similar data for services are, to the best of our knowledge, not available.
Retail is complex due to its multi-product, multi-store and multi-market nature. In
addition, stores offer different product assortments, and our approach requires data
on all stores in local markets. Identifying the impact of large entrants on technical
productivity therefore requires additional assumptions (e.g., timing).
6
A common characteristic of retail markets is lumpy investment and a lack of data
on intermediate inputs such as the stock of products (materials). In retail, a static
labor assumption is less restrictive than in many other industries, as part-time work
is common, the share of skilled labor is low, and stores frequently adjust labor due to
variations in customer flows.
3
more restrictive in Europe than in the U.S.7 The main rationale for such
regulations is that new entrants generate both positive and negative externalities that require careful evaluation by local authorities. Advantages,
such as productivity gains, lower prices, and wider product assortment, contrast with disadvantages, notably, fewer stores and environmental effects.
Because large entrants are expected to impact market structure extensively,
they are carefully evaluated in the planning process.
Our empirical results show that large entrants force low productivity
stores to exit and surviving stores to increase productivity. There is significant heterogeneity in the impact of large entrants on store productivity within and across local markets. The median increase in incumbents’
productivity due to a large entrant is 4-6 percent. A key result is that
productivity increases most among incumbents in the bottom part of the
productivity distribution and declines with the productivity of incumbents.
A large entrant increases productivity 1-2 percentage points more for a
store in the 25th local market productivity percentile than for a store in
the 75th percentile. The average aggregate weighted impact of a large entrant on productivity at the local market level is 3-5 percent, using output
market shares as weights. Our results show that it is important to allow
large entrants to affect the distribution of store productivity and control
for omitted prices and local market demand. The finding that a more liberal design and application of entry regulations would support productivity
growth is robust to various identification strategies, including controlling
for possible endogeneity issues concerning large entrants. From a policy
perspective, our results are informative to policymakers who decide over
entry of new stores in local markets.
The next section describes the retail food market and the data. Section
3 presents the modeling framework, identification and estimation. Section
4 reports the empirical results. Section 5 summarizes and draws conclusions. In several places, we refer to an online appendix containing various
analyses that are not discussed in detail in the paper.
7
Hoj et al. (1995); Pilat (1997); Boylaud and Nicoletti (2001); and Griffith and
Harmgart (2005).
4
2
The retail food market and data
The Swedish retail food market consists of a mix of different firm organizations with a clear tendency toward independent and franchise stores and
with firms functioning as wholesale providers. Decisions regarding pricing,
inputs, and exit are thus traditionally made by individual store owners.
For our purposes, we focus on the recent increase in centralized decisions
of firms to enter large stores into local markets (henceforth referred to as
large entry) together with the historical network of independent and franchise stores.
Stores belong to four main firms. ICA consists of a group of independent store owners and began through collaboration on wholesale provision.
Axfood contains a mix of independent and franchise stores.8 Bergendahls
has a mix of franchises and centrally owned stores and operates mainly in
the south and southwest of Sweden. Coop, by contrast, consists of centralized cooperatives, with decisions made at the local or national level.
Despite its cooperative structure, independent store owners in Coop have
the power to decide on, e.g., pricing and labor. Stores affiliated with these
four firms together constitute approximately 92 percent of market shares
in 2002: ICA (44 percent), Coop (22 percent), Axfood (23 percent), and
Bergendahls (3 percent). Various independent owners make up the remaining 8 percent of market share.9 All four firms include both small and large
stores.
The Swedish regulation, the Plan and Building Act (PBA), empowers
the 290 municipalities to make decisions regarding the applications of new
entrants. PBA is viewed as a major barrier to entry, resulting in diverse
outcomes, e.g., in price levels, across municipalities (Swedish Competi8
In 2000, Axel Johnson and the D-group (D&D) merged to form Axfood, initiating
more centralized decision-making and more uniformly designed store concepts from 2001
and onwards.
9
International firms with hard discount formats entered the Swedish market after
the study period: Netto in 2002 and Lidl in 2003.
5
tion Authority, 2001:4). Several reports stress the need to better analyze
how regulation affects market outcomes (Swedish Competition Authority,
2001:4, 2004:2). Large entrants are often newly built stores in external
locations, making regulations highly important. The online appendix (Section A) describes the PBA in greater detail.
Data. The data consist of two micro-data sets, DELFI and FS-RAMS.
Both contain yearly information on all retail food stores in Sweden from
1996 to 2002. DELFI contains information about store type, chain/firm,
and sales space (in square meters). A store is assumed to enter if it is
observed in the data in year t but not t − 1, and a store is assumed to exit
if it is observed in year t but not t + 1. DELFI defines a unit of observation
as a store based only on its geographical location (i.e., only its physical
address), and it is used to define large entrants.
The most disaggregated level for which more accurate input and output
measures exist is organization number (Statistics Sweden, SCB).10 SCB
provides data at this level based on tax reporting. Financial Statistics
(FS) provides input and output measures, and Regional Labor Statistics
(RAMS) provides data on wages for all organization numbers, from 1996 to
2002, belonging to SNI code 52.1, “Retail sales in non-specialized stores,”
which covers the four dominant firms (ICA, Coop, Axfood, and Bergendahls).11 In few cases, an organization number can consist of more than
one store (“multi-store”) in the same municipality for which we observe
total, not average, inputs and outputs. In FS-RAMS, entry and exit are
defined only on the basis of organization numbers.12
For our purposes, we estimate productivity for each organization num10
A so-called organization number specifies the identity of a corporate body. The
Swedish Tax Authority (Skatteverket) has a register of all organization numbers used
for tax reporting. The numbers are permanent and unique, i.e., one number follows
the corporate body throughout its whole existence, and two identical organization numbers do not exist. The register contains the date of registration of the organization
number and information regarding any exit/bankruptcy (Swedish Tax Authority, 2011).
Anonymous codes in FS-RAMS entail that we do not know the exact identity of an
organization number.
11
SNI (Swedish National Industry) classification codes build on the EU standard
NACE.
12
In FS-RAMS, we observe the municipality in which each organization number is
physically located. Exit in FS-RAMS may thus be due to re-organizations, for example.
6
ber and year using FS-RAMS and define physical entry of big-box stores
based on DELFI. We remove large entrants from FS-RAMS when estimating productivity. Finally, we collect local market demographic information
(population, population density, average income, and political preferences)
from SCB. Online appendix A provides more information about the data.
Local markets. Food products fulfill daily needs and are often of relatively short durability. Thus, stores are located close to consumers. Travel
distance when buying food is relatively short (except if prices are sufficiently low), and proximity to home and work are thus key considerations
for consumers in choosing where to shop. The size of the local market
for each store depends on its type and the distance between stores. We
assume that retail markets are isolated geographic units.13 The 21 counties in Sweden are clearly too large to be considered local markets for our
purposes, and the 1,534 postal areas are probably too small, especially for
large stores (on which we focus). The 88 local labor markets take into
account commuting patterns, which are important for hypermarkets and
department stores, while the 290 municipalities appear to be more suitable
for large supermarkets. As noted, municipalities are also the location of
local government decisions regarding new entrants. We therefore use municipalities to define local markets.
Large entrants. We define the five largest types of stores (hypermarkets,
department stores, large supermarkets, large grocery stores, and other14 )
as “large” and four other types of stores (small supermarkets, small grocery
stores, convenience stores, and mini markets) as “small.” This classification
accords with the Swedish Competition Authority (see, e.g., Swedish Competition Authority, 2002:6). In terms of the Swedish market, we believe
that these types are representative of being “large.”15 In light of the entry
13
A complete definition of local markets requires information about the exact distances between stores. Without this information we must rely on already existing measures.
14
Stores classified as other stores are large and externally located.
15
Because the store type classification in DELFI is extremely detailed, grouping stores
into two size classes is not highly restrictive. Sales, sales space, and other store characteristics suggest that it is reasonable to group, e.g., hypermarkets and large grocery
stores together, and to separate large and small supermarkets (online appendix A). Alternatively, we define observations in FS-RAMS with sales above the 5th percentile of
7
regulation, we only consider the physical entry of large stores (defined only
based on address). A store that is re-classified into one of the large store
types during the period is thus not counted as a large entry. Gas station
stores, seasonal stores, and stores under construction are excluded, as they
do not belong to the SNI-code 52.1 in FS-RAMS.
A concern, when analyzing the link between large entrants and productivity growth, is the endogeneity of large entry.16 Local authorities make
decisions regarding such entry. We use political preferences in municipalities and the number of large entrants in neighboring markets as instruments
for large entrants in a setting that distinguishes between the impact of large
entrants on demand and productivity shocks (see Section 3.1).17 We use
variation in political preferences across local markets throughout the election periods 1994-1998 and 1999-2002 to add exogenous variation to the
numbers of large entrants.18 We expect non-socialist local governments to
have more liberal views regarding large entrants. Section 3.1 and online
appendix B discuss in detail concerns regarding the endogeneity of large
entrants in our model setting.
Descriptive statistics. Table 1 presents descriptive statistics of the
Swedish retail food industry from the two data sets DELFI and FS-RAMS
for 1996-2002. In FS-RAMS, the number of observations decreases by about
17 percent (from 3,714 to 3,067). The share of large stores increases from
large stores’ sales in DELFI as large; otherwise as small. The empirical results (available
from the authors upon request) are consistent with those reported here.
16
See Bertrand and Kramarz (2002), Schivardi and Viviano (2011), and Sadun (2014).
Studies based on U.K. data have used major policy reforms to handle the endogeneity
of entry (Aghion et al., 2009; Sadun, 2014).
17
Data on the number of applications and rejections for each municipality are not
available in Sweden. Even if this information would have been available, it is not completely exogenous because the number of applications is easily influenced by current
local government policies. No major policy reforms changing the conditions for large
entrants occurred in Sweden during the study period.
18
The Social Democratic Party is the largest party nationally, with 40.6 percent of
seats on average. It collaborates with the Left Party (8 percent) and the Green Party
(4.2 percent). The non-socialist group consists of the Moderate Party (18 percent), most
often aligned with the Center Party (13.2 percent), the Christian Democratic Party (5.9
percent), and the Liberal Party (5.6 percent). Twenty-two percent of municipalities had
a non-socialist majority during 1996-1998, increasing to 32 percent during 1999-2002.
The non-socialists had 8.6-85 percent, averaging 40.7 percent in 1996-1998 and 44.1
percent in 1999-2002.
8
19 percent to nearly 26 percent during the sample period. While total sales
space remains virtually constant, mean sales space increases by 33 percent.
Thus, there has been a major structural change toward larger but fewer
stores. The wage-bill increases by over 22 percent (in real terms), while the
number of employees increases by only 9 percent. Total sales increase by
approximately 26 percent. Aggregate value-added per employee increases
from SEK 247.22 thousand to SEK 277.69 thousand during the period (12
percent). The corresponding increase in value-added per sales space is from
SEK 7.29 thousand to SEK 8.72 thousand (19 percent).
Table 2 shows the distribution of stores and firms across all local markets
and years. The average number of stores is 23, with a standard deviation
of 35. A majority of markets consist of stores that belong to three firms,
whereas almost no markets consist of stores that belong to a single firm.
Most stores belong to ICA, about twice as many as belong to Coop and
Axfood in the upper part of the distribution. On average, as many as 7.25
stores belong to ICA, and slightly less than 4 each belong to Coop and
Axfood.19
Table 3 shows the median characteristics of local markets with and
without large entrants during 1997-2002. Based on all stores, average valueadded per employee increases from SEK 249.33 thousand to SEK 266.28
thousand (7 percent) during the study period, whereas average value-added
per sales space (m2 ) increases from SEK 4.85 thousand to SEK 5.55 thousand (14 percent). The median number of stores varies from 22 to 54 in
large entry markets, compared with 13 to 15 in non-entry markets. The
number of markets with at least one large entrant varies from 6 to 23.
Among these, up to three large entrants become established in the same
market in the same year. As expected, median entry and exit are higher in
large entry than in non-entry markets, and so are median population and
population density.
19
ICA stores operate in almost all 290 markets. Coop decreases from 236 to 227
markets and Axfood from 276 to 266 during the study period. Bergendahls stores are in
21 markets at the beginning and 42 markets at the end. ICA, Axfood, and Coop have
similar store size distributions throughout the whole distribution. Median (mean) store
size is 316 (540) square meters for ICA, 350 (620) for Axfood, 400 (620) for Coop, and
448 (1,297) for Bergendahls. Hence, most stores are small.
9
3
Modeling the impact of large entrants on
productivity
This paper uses a general strategy to measure the effect of entry of large
stores on stores’ productivity shocks, while controlling for local market
characteristics and unobserved prices. Our framework describes a store by
a vector of state variables consisting of productivity ω, capital stock k, the
number of large stores eL that enters a local market in period t, and other
local market demand shifters x.20 We assume that capital is a dynamic
input that accumulates according to Kt+1 = (1−δ)Kt +exp(it ), where δ is
the depreciation rate. Productivity follows a controlled first-order Markov
process with P (dωjt |ωjt , eLmt−1 ), i.e., large entrants affect productivity.21
The fact that it takes time for stores to adjust their productivity in response
to increased competition justifies the assumption of a lagged effect of large
entrants on productivity. We assume that firms decide on entry of large
stores and that individual stores cannot influence this decision. Large entry
is an exogenous state variable and we assume that stores do not form
expectations about future values of large entrants.
Service generating function and imperfect competition. Stores sell
products and services following a Cobb-Douglas technology:
p
qjt = βl ljt + βk kjt + ωjt
+ upjt ,
(1)
where qjt is the log of service output by store j at time t; ljt is the log of
labor input; and kjt is the log of capital input. The service output does not
20
We follow the common notation of capital letters for levels and small letters for
logs for all variables except eL , which is in levels.
21
The investment measures the difference between real gross expenditures on capital
and real gross retirement of capital. Online appendix A provides details regarding the
construction of the capital stock in our empirical application. The correlation between
capital stock (equipment) and sales space (m2 ) is 0.63. The model can be extended to
allow for spillover effects by adding the number of large entrants in neighboring markets
L
L
(−mn ) eL
−mn ,t−1 to the productivity process, i.e., P (dωjt |ωjt−1 , em,t−1 , e−mn ,t−1 ).
10
include the items that are purchased at the wholesaler and sold in the store,
p
i.e., intermediate inputs. The unobserved ωjt
is technical productivity, and
p
ujt is a shock to service output that is not predictable during the period
in which inputs can be adjusted and stores make exit decisions.22 In other
p
words, all endogeneity problems regarding inputs are concentrated in ωjt
.
Because service output is difficult to measure in retail markets and is
therefore unobserved in many data sets, we use deflated value-added yjt as a
proxy, which includes store prices. The log of deflated value added is given
by yjt = qjt + pjt − pIt , where pjt and pIt are logs of service output prices
at store and industry level. When a store has some market power and the
output includes prices, as in retail food, its price influences productivity
measure (Foster et al., 2008). To account for this, we consider a standard
horizontal product differentiation demand system (CES)
1
1
1
1
1
pjt = pIt + qjt − qmt − βe eLmt − x0 mt β x − udjt ,
η
η
η
η
η
(2)
where qmt are log of aggregated service output in local market m, and udjt
represents remaining store level shocks to demand (Klette and Griliches,
1996; Levinsohn and Melitz, 2006; De Loecker, 2011).
The unobserved prices pjt are explained by variations in inputs and
aggregate demand. We use the current number of large entrants eLmt and
observed local market demand shifters x0 mt to control for local market
demand. The parameter η (< −1 and finite) captures the elasticity of
substitution among stores.23 The demand system implies a single elasticity
of substitution for all stores. Thus, there are no differences in cross-price
η
), and
elasticities, i.e., we have a constant markup over marginal cost ( 1+η
1 24
).
the Lerner index is ( |η|
In the case of value-added, upjt can be associated with measurement error when
there is the same measurement error in intermediate inputs and output (Gandhi et al.,
2011).
23
The vertical dimension is to some extent also captured because deflated output
measures both quantity and quality, which is correlated with store type (size).
24
Empirical studies on our Swedish retail food stores find that large stores offer
slightly lower prices (about 3 percent) and have only a modest impact on prices in
surrounding stores (less than 1 percent) (Asplund and Friberg, 2002). However, it may
22
11
Combining unobserved store price pjt in (2) and the service production
function (1), we have the service generating function
yjt =
1 + η1 [βl ljt + βk kjt ] − η1 qmt − η1 βe eLmt − η1 x0 mt β x
p
− η1 udjt + 1 + η1 upjt .
+ 1 + η1 ωjt
(3)
The impact of large entrants on the productivity process. The
nature of the remaining demand shocks udjt are crucial for the identification
strategy and whether it is possible to separate the impact of large entrants
p
on technical productivity ωjt
from their impact on demand.25 If udjt are
correlated over time, we can only identify the impact of large entrants on
quality-adjusted productivity, i.e., the sum of technical productivity and
p
p
1
demand shocks ωjt = ωjt
− 1+η
and udjt follow
udjt .26 We then assume that ωjt
independent Markov processes, i.e., expected productivity at time t conditional on the information set Fjt−1 does not depend on udjt . The i.i.d. shocks
are denoted by jt = (1+ η1 )upjt . If udjt are i.i.d. shocks, jt = (1+ η1 )upjt − η1 udjt ,
we can use a timing assumption on large entrants to separate the impact
of large entrants on technical productivity from their impact on demand
(De Loecker, 2011). The key difference between quality-adjusted producp
is the interpretation of the results,
tivity ωjt and technical productivity ωjt
that is, on whether productivity can be measured with or without demand
shocks. For expositional simplicity and to avoid carrying forward different
definitions in what follows, we will use ωjt , referring to it as productivity
yjt =
1+
[βl ljt + βk kjt ] − η1 qmt − η1 βe eLmt − η1 x0 mt β x
+ 1 + η1 ωjt + jt .
1
η
(4)
be that markups vary across stores, depending on size and concept.
25
Because of difficulties in defining a “price” measure, as a consequence of stores
offering different product baskets and services, our analysis will always require additional
assumptions (e.g., a timing assumption) to separately identify the effect of large stores
on demand and supply. Section 3.1, and online appendix B provide more details.
26
This implies that there are lagged effects of large entrants on prices even after
controlling for a wide range of local market characteristics and current large entrants.
12
We assume that productivity follows a first-order nonlinear Markov process: ωjt = E[ωjt |Fjt−1 ] + ξjt , i.e.,
ωjt = h(ωjt−1 , eLmt−1 ) + ξjt ,
(5)
where h(·) is an approximation of the conditional expectation, and ξjt are
shocks to productivity and are mean independent of all information known
at t−1. Large entrants immediately affect stores’ residual demand and thus
local market equilibrium prices but affect store productivity with a one-year
lag. We provide reduced form evidence that this assumption is not rejected
by our data (online appendix B) and argue that it is relatively unrestrictive
in application to the Swedish retail food market (Section 4.1). Specifically,
when regressing store sales or value-added on both eLmt and eLmt−1 , the
coefficient on lagged large entrants is not significant.27 Stores determine
their own prices and adjust prices quickly, and consumers can easily switch
stores. Because we use yearly data, and entry is regulated, consumers are
aware of and have time to adjust to the new market structure, i.e., the
previous number of large entrants does not affect current prices.28
Our framework involves explicit modeling of unobservables that drive
store output and their relationship to observables (Haavelmo, 1943). We
also discuss the identification and empirical findings, omitting the effect of
eLmt on demand. In this case, we measure the impact of the previous number
of large entrants (eLmt−1 ) on both technical productivity and demand shocks
27
Online appendix B provides a detailed discussion of the identification of the effect
of large entrants on productivity and demand, and Tables B.2-3 present reduced-form
results.
28
The productivity measure can be affected by capital mismeasurement, e.g., stores
vary their capital services by changing the utilization of the capital stock, or capital
service flows for retail space might be higher in urban/dense areas than in typical big-box
locations. Because these data are not available, we assume that the flow of equipment
services is proportional to the stock of equipment (ideally, we would like to know how
each machinery (technology) and square meter of sales space is used in the store).
We minimize capital mismeasurements by controlling for the common characteristic of
capital service flows for retail space at the local market level.
13
associated with current large entrants (eLmt ).29
3.1
Identification and estimation
We use the labor demand function from stores’ static profit maximization
problem to recover productivity together with a good measure of storespecific wages (Doraszelski and Jaumandreu, 2013).30 Labor is a static
and variable input chosen based on current productivity. This assumption
has the advantages that we can include many stores with zero investment
and abstract from assumptions regarding stores’ dynamic programming
problem. For several reasons, this assumption is less restrictive in retail
than in many other industries. Part-time workers are common. As many
as 40 percent of employees in retail food work part time, compared with
20 percent in the Swedish economy as a whole (Statistics Sweden). The
share of skilled labor is low in retail. Only 15 percent of all retail employees
had a university education in 2002, compared with 32 percent in the total
Swedish labor force (Statistics Sweden). Moreover, we find no systematic
differences in hiring educated workers between small and large stores in our
data. Stores have long and similar opening hours and adjust their labor
due to variations in customer flows over the day, week, month and year.
The training process might also be shorter than in many other industries.31
We relax the static labor assumption in the robustness section.
The general labor demand function that arises from stores’ optimization problem is ljt = ˜lt (ωjt , kjt , wjt , qmt , eLmt , xmt ) where ˜lt (·) is an unknown
29
Complexity of the retail food industry makes it difficult to model all channels that
improve productivity. Changes in store productivity provide information about the
response to a large entrant, apart from physical entry and exit.
30
Intermediate inputs would be an excellent choice of proxy for productivity in retail
markets (Levinsohn and Petrin, 2003). Ideally we would like to have data on the stock
of products (materials), but such data are unfortunately not available in many data sets
on service industries. The complexity of food products and the fact that stores have
different product assortments make it difficult to collect data on the stock of products
for all stores. The investment policy function is restrictive to use because retail stores
make lumpy investments, and we can only use stores with positive investment (Olley
and Pakes, 1996).
31
We assume that the labor market is efficient, so that there are no training, hiring
or firing costs, no labor supply constraints for stores (they can hire when they want),
and no labor market rigidities.
14
function, and wjt is the log of wage rate at the store level. To back out
productivity, the following key assumptions must hold. First, the labor
demand function must be strictly monotonic in productivity, which holds
when labor is a static input and more productive stores do not have disproportionately higher markups than less productive stores (Levinsohn and
Melitz, 2006). Second, ωjt is the only unobservable entering the labor demand function. This rules out, e.g., measurement error, optimization error
in labor, and a model in which exogenous productivity is not single dimensional. We assume that the observed variation in store wages is due to
differences in exogenous market conditions (Ackerberg et al., 2007). Our
detailed register data of wages for all employees in Swedish retail are less
subject to measurement errors due to reporting. The number of full-time
adjusted employees is our measure of labor.
Third, we require helpful variation in store-specific wages.32 Even if
store wages change over time, we need additional variation at the store
level if we are to control for time effects in the estimation.33 High-quality
data on store-specific wages, the fact that stores set wages, and the prevalence of temporary job contracts and part-time work ensure the existence
of wage variation across stores. The coefficient of variation for wages is
about 18 percent across stores and 53 percent across municipalities. Storespecific wages regressed on market- and year-fixed effects, store size and
32
Our measure of wages is a good reflection of exogenous changes in the price of
labor because the 22 percent growth in the retail wage-bill during the period (Table 1)
is consistent with the 24 percent growth in aggregate real wages in Sweden (Statistics
Sweden). The average wage contains both the price of labor and its composition, e.g.,
age, gender, and skill groups. In Sweden, we do not expect compositional effects due
to some employees working overtime or differences in opening hours across stores. A
one-sided t-test shows that we cannot reject the null of equal means of the share of educated employees (0.064) for both small and large stores. However, wages might pick up
unobserved worker quality. Because worker quality is unobserved by the econometrician
but observed by stores, we have two unobservables to control for, which complicates
estimation. Instead, the unobserved quality will enter into our productivity measure.
However, this is not a large concern in the retail food market, where the quality of
workers is expected to be fairly homogenous (online appendix C).
33
In the absence of store level wages, however, it may be difficult to estimate the
coefficients of static inputs in the Cobb-Douglas case (Bond and Söderbom, 2005). The
proposed estimation strategy assumes that the first-order condition for labor does not
include the derivative of the wage rate with respect to labor.
15
observed labor quality (education) show that there are other unobserved
factors at the store level, for example, bargaining negotiations, experience,
etc., that explain variations in store wages (online appendix C). Using data
for the year 2000, market dummy variables alone explain only 9.7 percent
of the variation in wages. By adding capital and a dummy for large stores
and local market controls, we explain 14.2 percent of the wage variation.
By including the number of employees as an additional measure of store
size, the variation in wages explained by the covariates increases to 15.7
percent. Fourth, we form moment conditions in the estimation, using information about when in the productivity process stores choose inputs and
firms make decisions regarding large entry (discussed below).
Estimation. By inverting the labor demand function ˜lt (·) to obtain productivity ωjt and substituting the result into (4), the service generating
L
) + jt ,where φt (·) =
function
becomes yjt = φt (ljt , wjt , kjt , qmt , emt , xmt
0
1
1
1
1
L
1 + η [βl ljt + βk kjt ] − η qmt − η βe emt − η xmt βx + 1 + η1 ωjt . Estimation is performed in two steps. The aim of the first step is to remove output
and demand shocks from productivity. The first step yields an estimate of
φt (·), φ̂t , and helps to recover productivity ωjt without i.i.d. shocks jt as
follows:
h
η
1
ωjt (β) = (1+η) φ̂t − 1 + η [βl ljt + βk kjt ] + η1 qmt + η1 βe eLmt
i
(6)
0
+ η1 xmt β x ,
where β = (βl , βk , η, βe , β x ). To obtain φ̂t using the OLS estimator, we
use the moment conditions:34 E[jt |f (ljt , wjt , kjt , qmt , eLmt , xmt )] = 0, t =
1, · · · , T, where f is vector valued instrument functions (Wooldridge,
2009). In the second step, we nonparametrically regress ωjt (β) on a polynomial expansion of order three in ωjt−1 (β) and eLmt−1 . The labor coefficient βl
is identified from the moment E[ξjt |ljt−1 ] = 0. Capital is a dynamic input,
so that the coefficient for capital βk is identified from E[ξjt |kjt , kjt−1 ] = 0.
Large entrants influence productivity with a one-year lag. In the case of
technical productivity, only current large entrants influence prices. The
34
A polynomial expansion of third-order is used.
16
moment E[ξjt |eLmt ] = 0 is used to identify the coefficient for large entrants
βe . The moment E[ξjt |qmt−1 ] = 0 is used to estimate η, and E[ξjt |xjt−1 ] = 0
is used to estimate β x . The parameters β are estimated by minimizing the
GMM objective function
0 1 0
1 0
W ξ(β) A
W ξ(β) ,
min QN =
β
N
N
(7)
−1
0
0
where A is the weighting matrix defined as A = N1 W ξ(β)ξ (β)W , and
W is the matrix of instruments.35 Estimation is performed at the industry
level, controlling for local market conditions. Standard errors are computed
using Ackerberg et al. (2012).36 We denote by Mlm the specification that
uses the above moment conditions. This is our main specification that allows for imperfect competition and treats the number of large entrants as
exogenous.
Endogeneity of large entrants and aggregate service output. As
noted, firms can decide to enter large stores in markets with favorable characteristics such as short distance to a distribution center or good logistics.
The assumption E[jt |eLmt ] = 0 does not hold when jt includes shocks due
to advertising, sales promotion activities related to large entrants, and distribution (transportation). These shocks affect stores differently and might
also impact the aggregate service output, i.e., E[jt |qmt ] = 0 does not hold.
To account for the endogeneity of large entrants, we use the share of nonsocialist seats in local governments (Bertrand and Kramarz, 2002; Schivardi
and Viviano, 2011; Sadun, 2014), the number of large entrants in other
markets (Hausman type of instruments), and the previous number of large
entrants as instruments. In case of the endogeneity of the aggregate service output, we can use lagged values and aggregate service output in other
35
Wooldridge (2009) and ACF (equation (27)) suggest a one-step estimator using
GMM based on moment conditions E[jt |Fjt ] = 0 and E[(1+ η1 )ξjt +jt |Fjt−1 ] = 0. Even
if this estimator is more efficient than the two-step estimator, it is very computationally
demanding in our case due to a large number of parameters to be estimated.
36
Bootstrapping might not be the best choice when the underlying model is more
complicated. It requires additional computation time, optimization errors may appear,
and the choice of stores in different samples yields different effects of competition from
large entrants, implying that a large number of bootstraps may be required.
17
markets as instruments.
We obtain φ̂ in the first step, using the GMM estimator and the moment
P
P
conditions E[jt |f (ljt , wjt , kjt , qmt−1 , o6=m qot , eLmt−1 , o6=m eLot , polmt , xmt )] =
0, t = 1, · · · , T, where f is vector valued instrument functions (Wooldridge,
2009).37 We denote by Mlme the specification that controls for the endogeneity of the number of large entrants in the first-step of the estimation.
In the empirical implementation, we find no significant changes in the elasticity η when controlling for the endogeneity of aggregate quantity qmt . For
this reason, we mainly discuss the specifications Mlm and Mlme .
When regressing the current number of large entrants on political preferences, we find that an increase in the share of non-socialist seats (level) in
a municipality positively affects the number of large entrants (Table B.2 in
the online appendix). This result is robust to observed characteristics and
year- or market-fixed effects, indicating the relevance of our instrument.
Large entrants might enter in regions with good distribution, suggesting
that the number of large entrants in neighboring markets can be used as
an instrument (Hausman, 1997; Petrin and Train, 2010). The results in
Table B.2 (online appendix) show that the number of large entrants in
neighboring markets is an important determinant of the current number of
entrants in a market.
To be an effective instrument for large entrants, political preferences
(i.e., the share of non-socialist seats) should not be related to local market
demand or reflect characteristics of the population that favour shopping at
big-box stores but can boost productivity at other stores. This raises the
following concerns. First, the outcomes of elections might be influenced
by economic conditions. Political business cycles can only affect our results if there is substantial ability to predict future demand shocks when
politicians are elected. We also investigate median local market characteristics for socialist markets with large entrants and non-socialist markets
without large entrants. There are between 1-6 socialist markets with large
entrants and 82 to 147 non-socialist markets without large entrants during
37
Similarly, E[jt |wjt−1 ] = 0 can be used to control for endogeneity of wages. In the
second step, there is no endogeneity problem of large entry because of our assumption
regarding the productivity process, i.e., E[ξjt |eL
mt−1 ] = 0.
18
the study period. Socialist markets with large entrants are larger markets
(population) and have lower population density than non-socialist markets
without large entrants. In addition, these two groups of markets do not
significantly differ in income per capita. Importantly, we control for local market characteristics (income, population, population density) when
estimating productivity. The second concern is that political preferences
might capture local policies other than entry regulations. In Sweden, PBA
is rather exceptional, enabling local politicians to play a key role. Furthermore, in our context, the number of large entrants in other markets is an
appropriate instrument if the number of large entrants in other markets
reflects common trends or demand shocks only specific to large entrants,
e.g., unobserved advertising.38 To check the validity of the instruments, we
also report the partial F-test suggested by Staiger and Stock (1997) (Table
4). The test validates the proposed instruments showing that they are not
weakly correlated with the number of large entrants. Although the proposed instruments are not perfect when there are correlated unobservables
across markets, we believe they are the best instruments, given the available data and modeling framework, and they have been used extensively
in the empirical literature.
3.2
Some remarks on the empirical implementation
Large entrants. We analyze the impact of the number of large stores eL
that enters a local market in period t on the productivity of incumbent
stores, which are defined as all stores other than entrants. The choice to
model entry of new large stores instead of the total number of large incumbents is due to the entry regulation focusing on entrants’ impact on
consumers and market structure. Because all stores determine their own
prices in Sweden, and a majority of stores operate as independent or franchise units, we model each store as a separate unit that decides on prices,
38
Hausman type instruments are widely used, as they are always available, but they
are controversial (Hausman, 1997).
19
inputs, and exit.39
Entry by a large store might have different impacts depending on the
size of the local market. For a fixed definition of large stores, the number
of large entrants is a good proxy for scale of entry relative to market size
in industries characterized by spatial differentiation. To explicitly incorporate sales space, one could replace large entrants by their sales space and
evaluate the impact of an additional m2 on incumbents’ productivity. Alternatively, one could model large entrants’ optimal entry size in a dynamic
game framework. To deal with scale of entry, we use a detailed definition of
large entrants, control for store size and market size, and calculate marginal
effects of large entrants on incumbents’ productivity, where such effects are
robust to market heterogeneity.
Demand system. Although our CES demand model is restrictive because
of data constraints, several features make it less restrictive in the Swedish
retail food market than in many other industries. First, stores determine
their own prices, and we do not expect a single store to influence the market price because local markets contain many stores. On average, there are
30 stores in markets with large entrants and 15 in markets without large
entrants (Table 3). Second, all stores offer a wide range of products, i.e.,
we assume that stores serve the same basic function for consumers – to
provide food. In Sweden, price (and quality) differences for a homogenous
product basket are found to be small between firms and stores (Asplund
and Friberg, 2002). Given our data constraints, we focus on the key dimension of differentiation in location.
Alternative approaches: Dynamic games. To model large entry and
exit in a dynamic game framework would require additional assumptions,
e.g., on functional forms of payoff and cost functions and aggregation re39
If we aggregate and analyze decisions of, e.g., pricing at the firm level (instead of
the store level), we lose much of the dynamics crucial to our analysis of the Swedish
retail food market. National pricing with market power of firms rather than stores is
more common in other countries (e.g., the U.K.). To analyze the relationship between
firms and stores in more detail, we would require data on the identity of (multi-) stores
for which we observe inputs and outputs. The decision to exit or continue is made at
the store level, although firms can influence the decision of each store through possible
chain effects. Section 2 provides details regarding the organization of firms.
20
strictions, and raise concerns about multiple equilibria, equilibrium selection mechanisms and computational complexity (Pakes et al., 2007). The
benefit of a dynamic game setting is that we can endogenize entry. Given
the main goal of the current paper, we believe that the advantages of our
detailed analysis of store productivity dynamics outweighs the potential
limitations of the single agent framework.
4
Results
Table 4 presents estimates of the service generating function, using labor
as a proxy for productivity and previous large entrants in the productivity
process and controlling for prices using current large entrants and local
market characteristics (population, population density, income) (Mlm ). In
addition, we control for the endogeneity of large entrants in the first-step
of the estimation, using political preferences, number of large entrants in
other markets, and the previous number of large entrants as instruments
(Mlme ). We also present results under perfect competition, using the basic
implementation of Ackerberg et al. (2006), with labor demand as a proxy
(Ml ), and OLS. A major advantage of the specifications Mlm and Mlme is
that they use a controlled productivity process and control for unobserved
prices, which otherwise might downwardly bias the scale estimator (omitted price bias) (Klette and Griliches, 1996). Another advantage is that they
yield an estimate of market output, which makes it possible to compute the
implied demand elasticity (η) and an average industry markup controlling
for local market competition.
As theory suggests, the estimate of returns to scale (βl + βk ) in the
Mlm estimator is greater (1.505) than in the OLS (1.121) and Ml (1.005)
estimators. The point estimate for labor is 0.674, and that for capital is
0.304.40 Few studies that use a production function framework emphasize
40
Omitting to control for unobserved demand shocks, we expect the coefficients for labor and capital to be upwardly biased, owing to the positive correlation between inputs
and demand shocks. Maican and Orth (2012) discuss returns to scale under imperfect competition. After controlling for local market competition, the capital coefficient
21
the returns to scale in service industries. Increasing returns to scale are
expected in industries with high consumer participation, geographic dispersion, and multi-market contacts (economies of density). The scale is
likely to increase with the degree of self-service and is found to be higher
in retail food than in other retail sectors (Ofer, 1973).41
The estimate of the implied elasticity of demand is -2.858 in Mlm . Thus,
the implicit assumption that η=−∞, often used in empirical studies, does
not hold. The markup, defined as price over marginal cost, is 1.530. Our
estimates are consistent with previous findings based on retail data (Hall,
1988; Roeger, 1995; Maican and Orth, 2014).42 The coefficient for population is positive and statistically significant, and the coefficient for population density is close to zero.43 When controlling for the possible endogeneity
of large entry in the first-step in Mlm , the results show similar coefficients
for labor, capital and large entrants (Mlme ). The demand elasticity (|η|)
increases slightly from 2.85 to 3.27, and the scale decreases somewhat from
1.505 to 1.420.
increases, which is in the direction of controlling for selection bias (Olley and Pakes,
1996).
41
For food retailing in Israel, Ofer (1973) estimates returns to scale at 1.42 and at
1.31 when controlling for supermarkets. Bairam (1994) estimates returns to scale at
approximately 1.30 for fruit and vegetables, based on Australian data. These estimates
rely on Cobb-Douglas technology and value-added but do not control for simultaneity,
selection or omitted price bias.
42
The aggregate mark-up (η/(1 + η)) depends on the estimated elasticity of demand
η at the industry level, i.e., a larger |η| implies a lower mark-up. Hall (1988) uses
aggregate sector time series U.S. data and finds a markup about 1.42 for retail trade
and 1.53 for services. Using the same data, Roeger (1995) finds a mark-up about 1.50
for food and kindred products. Using a nested logit demand model with store level
prices for a product basket in Swedish retail food (2001-2008) and assuming a Nash
equilibrium, Maican and Orth (2014) find an estimated average price elasticity about -3
for large stores and -3.8 for small stores. Based on their product basket, their estimated
average mark-up across all stores is about 1.20.
43
The impact of a large entrant on residual demand and hence prices is, on average,
roughly 2 percent. This small positive effect might be due to the fact that our simple
demand system, owing to data constraints, only allows us to estimate average effects
and does not consider distributional effects. Large entrants may, e.g., reduce prices in
nearby stores. Our finding that large stores have a modest impact on prices is consistent
with previous studies of the Swedish retail food market (Asplund and Friberg, 2002).
22
4.1
Large entrants and productivity
The next step is to investigate whether large entrants influence the productivity of stores. We analyze whether large entrants have a greater impact
on one part of the local market productivity distribution than another
(Section 4.1.1) and use individual store’s marginal effects to evaluate the
impact of large entrants on aggregate weighted productivity in the local
market (Section 4.1.2).
Productivity measure. To obtain a measure that is comparable across
different estimators, we recover productivity from the service generating
function
η
[yjt − (1 + η1 )[βl ljt + βk kjt ] + η1 qmt
ωjt = 1+η
(8)
+ η1 βe eLmt + η1 x0mt β x ].
The productivity measure in (8) contains the i.i.d. jt shocks. To recover
productivity without i.i.d. shocks, one can use the inverse labor demand
function given in equation (6). Our results are robust to using this alternative measure. The average productivities obtained from both measures
(output and proxy) are close, but there are distributional differences and,
as expected, higher variance when using the service generating function.
For Mlm , the ratio of the interquantile range to the median is about 0.07
and 0.09 for productivity recovered from labor demand and output, respectively.
The reduced-form evidence suggests that previous large entrants do not
significantly impact demand (Section 3 and online appendix B), i.e., there
are no remaining persistent demand shocks related to large entrants. Given
the complexity of the industry, no persistence in the remaining demand
shocks upjt (i.e., technical productivity) may be debated. In what follows,
we focus on productivity recovered from output and estimated using our
main specifications, Mlm and Mlme .
Transitions in the productivity distribution. To explore changes in
productivity distribution in local markets, we classify incumbents into six
percentile bins (p10, p10-25, p25-50, p50-75, p75-90, p90) for each year,
based on productivity. We then follow movements between percentile bins
or exit over time, with productivity estimated by Mlm .
23
Low productive incumbents in markets without large entry decrease
their productivity or stay low productive without being forced to exit (Table 5). The share of incumbents that remain in p10 is 5 percentage points
higher in markets without large entry. More than 20 percent and 13 percent of stores in the two lowest percentile bins exit in entry markets, but
only 16 percent and 11 percent of such stores exit in non-entry markets.
In both market groups, exit also occurs among stores in p90. This might
be due to re-structuring and re-organization of incumbent stores because,
although large entrants are “physical entry”, the data only allow us to link
estimated productivity and exit based on organization number. To avoid
possible selection problems due to this characteristic of our data, we control for survival probabilities when estimating productivity and find that
the results are robust to this selection problem (Olley and Pakes, 1996).44
4.1.1
Store-level heterogeneity
Using our semiparametric model to control for standard problems (simultaneity and endogeneity), and following a specification entirely consistent
with our model, we approximate h(ωjt−1 , eLmt−1 ) using a third-order polynomial expansion in its arguments.45 We only consider incumbent stores
and exclude stores that enter (see next subsection for exit).
Table 6 (Panel A) shows the results for the impact of large entrants on
store productivity under our main specification, Mlm and Mlme , recovering
productivity from output. To account for the heterogeneity of stores’ productivity levels and in the impact of large entrants across local markets,
we evaluate the marginal effects of large entrants for different productivity
percentiles at the local market level (10th, 25th, 50th, 75th, and 90th).
First, we calculate the marginal effects of large entrants for each productivity percentile measure in each local market. Second, we compute the
44
Selection is discussed in detail in online appendix D.
The static entry process implies no endogeneity problem of large entrants because
eL
mt−1 is uncorrelated with current innovation in productivity ξjt (Section 3). We model
L
eL
mt−1 as a continuous variable in the Markov process because emt−1 is larger than one
in some local markets.
45
24
mean and standard deviation of the marginal effects for each productivity
percentile measure across local markets. This gives us the average productivity change across local markets following a large entrant for each
percentile.
The median impact of an additional large entrant on store productivity
across local markets is 6.5 percent in Mlm and 4.5 percent after controlling for the endogeneity of large entrants in Mlme . There is, however, high
dispersion in the impact across markets, as indicated by the standard deviations of 0.022 for Mlm and 0.018 for Mlme , respectively. We also find high
dispersion in store labor productivity (value-added per full-time adjusted
employee) across entry markets, where average labor productivity growth
is approximately 7-8 percent.
A key result in Table 6 is that the impact on productivity decreases
toward the upper parts of the productivity distribution. Large entrants
force low productive incumbents to improve their productivity more than
high productive incumbents.46 A large entrant increases productivity 3-4
percentage points more for a store in the 10th local market productivity
percentile than for a store in the 90th percentile (Mlm and Mlme ). The
corresponding difference is 1-2 percentage points for a store in the 25th
productivity percentile compared to a store in the 75th percentile. These
findings are in line with recent empirical literature on productivity (Syverson, 2004; Asplund and Nocke, 2006; Collard-Wexler, 2011).
For comparison, we use semiparametric estimates but without controlling for imperfect competition in local markets (Ml ) and simple OLS. A
new large store decreases median productivity by about 2 percent under
Ml . The unexpected negative impact of a large entrant suggests that we
must control for demand in local markets. The adjusted R2 for the productivity process regression is, moreover, 2-3 times lower under Ml than
under Mlm . Considering a simple parametric specification that explains
productivity by the number of large entrants, ωjt = βe eLmt + uejt , where uejt
are i.i.d., we can interpret βe as the effect on productivity when estimating
46
Using a simple linear specification, the results also suggest that large entrants
increase productivity, but the impact decreases with the productivity of incumbents.
25
the service generating function, yjt = β0 +βl ljt +βk kjt +βe eLmt +uejt +υjt , by
OLS. The coefficient for large entrants is positive but small (0.0003) and
not statistically significant. In addition to the standard problems of production function estimation and the use of strong assumptions to identify
βe (i.e., E[uejt + υjt |emt ] = 0), this specification does not address the effect
of large entrants on prices. Omitting to control for the impact of current
large entrants on prices in Mlm results in a 2 percentage point lower median impact of large entrants on store productivity (results not reported).
Hence, part of the productivity increase caused by large entrants is in fact
an effect on prices, which is important to control for (De Loecker, 2011).
An advantage of our approach (Mlm and Mlme ) is that it yields marginal
effects for every store. In other words, the impact of large entrants on productivity varies across stores, depending on their values of ωt−1 and eLmt−1 ,
and we can recover the full distribution in each local market. Table 6
(Panel B) shows that the support of individual effects of large entrants on
an incumbent’s productivity increases in the range of 0.017-0.117 (Mlm )
and 0.008-0.09 (Mlme ).47 Hence, there is substantial heterogeneity in the
impact of large entrants on productivity across incumbents. Variations in
geographic distance between large entrants and incumbents (spatial differentiation) and the possible existence of unobserved persistent demand
shocks, which yield quality-adjusted productivity instead of technical productivity, may partly explain this finding.
4.1.2
Productivity in local markets
We use effects on individual incumbents to evaluate changes in aggregate
local market productivity that result from a large entrant, which is an issue
of particular interest to policymakers who decide over entry of new stores
P m
in local markets. Aggregate productivity in market m is ωmt = nj=1
sjt ωjt ,
where sjt is the market share of store j in period t, and nm is the number of
47
Because of exit and product differentiation this does not imply that large entrants
will continuously increase productivity among incumbents. Without controlling for local
market competition and prices (Ml ), the lower bound of the support is negative.
26
stores. We compute the change in aggregate local market productivity due
to a large entrant as a weighted sum of individual stores’ marginal effects,
P m
using store market shares as weights, i.e., nj=1
sjt ∂e∂h(·)
. Note that eLmt−1
L
mt−1
measures the number of large entrants. As before, we only focus on changes
in the productivity of incumbent stores.
Table 6 (Panel C) shows the distribution of weighted aggregate local
market productivity growth of incumbents following a large entrant, i.e.,
Pnm
∂h(·)
. There is variation across local markets, with aggregate proj=1 sjt ∂eL
mt−1
ductivity increases ranging from 0.002 to 10.6 percent under Mlm (0.0027.8 percent under Mlme ). The median contribution of a large entrant to
local market productivity growth of incumbents is 5.7 percent under Mlm
(4 percent under Mlme ).48 These figures are 0.7 (0.5) percentage points
lower under Mlm (Mlme ) than the median increase in productivity computed using distribution measures of local market productivity reported as
averages across markets (Panel A in Table 6). This indicates that there
are more stores with a relatively low productivity increase from a large
entrant (marginal effect) and/or that stores with relatively low productivity increases as a result of a large entrant have larger market shares and
therefore receive larger weights.
Decomposition. To understand the contribution of the large entrants
to local market productivity growth of incumbents, we can use a simple
decomposition. The change in aggregate productivity in market m can be
written as
Pnm
Pnm
Pnm
∂h(·)
∆ωmt ≡
j=1 sjt−1 ωjt−1 '
j=1 sjt ∂eL
j=1 sjt ωjt −
mt−1
Pnm
L
+ j=1 ∆sjt h(ωjt−1 , emt−1 − 1)
P m
+ nj=1
sjt−1 [h(ωjt−1 , eLmt−1 − 1) − ωjt−1 ],
where the first term defines the weighted contribution of a large entrant
to productivity growth among incumbents in the local market;49 the second term is the contribution of the stores with increasing market shares,
regardless of a large entrant; and the third term is aggregate productivity
48
The marginal effects are 0.5-1 percent greater when the full sample of stores is used
and we evaluate the marginal effect for one large entrant in each market.
49
L
Note that ∂e∂h(·)
' [h(ωjt−1 , eL
L
mt−1 ) − h(ωjt−1 , emt−1 − 1)].
mt−1
27
growth, regardless of a large entrant in the local market, using previous
market shares as weights.50
Using local market aggregation of incumbents’ productivity, the findings
show a large dispersion in yearly productivity growth across local markets
with large entrants, where the median growth is about 9 percent. Because
our focus is on the first term, we discuss only the median value of each term
in the decomposition. As noted, a large entrant results in a 5.7 percent median increase under Mlm (4 percent under Mlme ) in productivity growth
among incumbents (first term). Regardless of a large entrant, the median
contribution to local productivity growth by the incumbent stores that increase their market share is approximately 2 percent (second term), but
the standard deviation is high (0.29). These findings suggest that large entrants might enter growing markets or that they create additional demand.
Without a large entry, aggregate productivity growth contributed by incumbents that increase productivity at initial sales levels is also dispersed
across local markets, with a median of -0.7 percent (third term).
Exit. While exit mainly occurs from the bottom part of the distribution,
entrants are found across the whole distribution (not reported), consistent
with previous findings in retail markets (Foster et al., 2006). According to
our model, stores decide whether to exit or continue at the beginning of period t, based on their information set consisting of the previous or current
state variables, that is, productivity, capital, large entrants, and demand
shifters (Section 3). When we identify technical productivity and control
for demand with observable demand shifters (eLmt , xmt ), so that shocks to
demand udjt are i.i.d., we assume that these shocks are not predictable by
50
There are a few remarks about this decomposition. First, we only use information
about incumbent stores in local markets exposed to large entry. Second, because most
markets have only one large entrant in a given year, we focus on the impact of one
rather than all large entrants (the decomposition can be extended to accommodate a
large number of entrants). Third, the sum of the three terms approximates the aggregate
productivity growth ∆ωmt because of the numerical approximations when computing
the derivative of the nonparametric function h(·) with respect to eL
mt−1 . Fourth, we use
store information and show how much of the productivity growth among incumbents at
the local market level is the effect of a large entrant.
28
stores when exit decisions are made.51 If stores can observe or predict the
demand shocks udjt after we control for observable demand shifters, it is not
possible to estimate the exit regression as below without including them in
the productivity process (quality-adjusted productivity).
Table 7 shows regression results for the probability of exit.52 The first
specification (column 1) relies on the pure stopping rule and does not consider stores’ positions in the local market productivity distribution. In line
with both theory and previous empirical studies (Olley and Pakes, 1996),
exit is less likely if productivity and the capital stock are high but more
likely if the market size is large. The coefficient of large entry has the expected positive sign but is not significant at conventional significance levels.
The expanded specification (column 2) includes interaction terms of large
entrants with the six local market productivity dummies, using the middle
group (p50-75) as a reference. The coefficient for the interaction term is
positive and jointly significant with the coefficient of large entry for p10
and p25. The probability of exit is about 0.02 higher after large entry for
stores in the bottom part of the productivity distribution than for those in
the middle.
To summarize, our results in Tables 5-7 show that large entrants increase store’s productivity, local market productivity growth, and exit.
These findings provide information about the observed trend toward larger
but fewer stores in the retail industry. That large entrants have the largest
effect on incumbents with low productivity is an indirect effect of competition. Large entrants induce exit, bring new demand but also capture
demand from incumbent stores, where the net effect is higher competitive
pressure on stores with low productivity.53
51
Exit decisions include physical exit and re-structuring/re-organization of stores,
which cause changes in stores’ organization numbers.
52
The exit regressions in Table 7 represents reduced-form exit policies, i.e., exit is a
function of the state variables. Note that by replacing the productivity with the inverse
of the optimal labor policy function, we obtain an exit policy function similar to the
selection equation presented in online appendix D.
53
The dynamic effect of competition is included in the measured effect of a new
entrant on productivity, and the positive effect on productivity should be put in balance
with a decrease in the number of stores, i.e., lower product differentiation available to
consumers. Maican and Orth (2013) use a dynamic entry-exit game with store type
29
4.2
Robustness and specification tests
This section discusses the main robustness and specification tests. The
online appendix presents additional robustness results.
Relaxing the timing assumption on labor. If there are hiring and
firing costs of employees, labor is a static and fixed input. We can then use
current labor ljt as an instrument in our main specification. The results
are directly comparable with those when labor is static and variable, i.e.,
Ml(m) in Tables 4 and 6. Under perfect competition (Ml ), the coefficient for
labor decreases from 0.843 to 0.647, and the coefficient for capital increases
from 0.162 to 0.240. Controlling for imperfect competition (Mlm ), the labor coefficient decreases to 0.491, the capital coefficient increases to 0.412,
and demand elasticity is -2.88 (similar to the findings in Table 4). Using
the moment condition based on current labor gives similar support of the
marginal effect of large entrants when productivity is recovered from the
service generating function, i.e., [0.017, 0.10] compared with [0.017, 0.117]
(Table 6).
Alternative production technology. For our main specification Mlm ,
we relax the Cobb-Douglas technology in equation (1) and consider a
2
2
+ βlk ljt kjt +
+ βkk kjt
translog production function qjt = βl ljt + βk kjt + βll ljt
p
p
ωjt + ujt , which requires the estimation of three additional parameters:
labor squared (βll ), capital squared (βkk ), and the interaction between labor and capital (βlk ). The results, not reported but available from the
authors upon request, are consistent with our previous findings. Large entrants have a greater impact on low productive incumbents than on high
productive incumbents. An additional large entrant increases productivity
by about 4 percent for a 10th percentile productivity store, by about 2
percent for a median store, and by about 0.1 percent for a 90th percentile
store.
Other robustness. Our identification strategy and empirical findings are
differentiation to study large entrants and industry dynamics in Swedish retail food
2001-2008 in detail.
30
robust to the choice of the labor demand function, e.g., a parametric labor demand function (Section E in the online appendix). By controlling
for possible endogeneity of wages in the first step, the coefficient of labor
decreases sightly, from 0.674 to 0.671.
We also estimate the contribution of all entrants to aggregate productivity growth during 1997-2002, using various productivity decompositions
(Griliches and Regev, 1995 and Foster et al., 2001). Incumbent stores
that increase their productivity at the initial sales level contribute approximately 8 percent (within) and net entry contributes 2-4 percent (Section
F in the online appendix). These findings suggest the importance of understanding the factors that drive within productivity growth.
5
Conclusions
The present study provides new insights into competition and productivity differences among retail stores. Net entry is found to foster almost all
labor productivity growth in the U.S. retail sector (Foster et al., 2006).
However, multi-factor productivity in retail markets has rarely been studied, in contrast to manufacturing. We provide a first attempt to use recent
advances in semiparametric estimation of production functions to estimate
productivity in retail markets and investigate how entry of large (“bigbox”) stores influences stores’ efficiency shocks and demand shocks. On
both sides of the Atlantic, the pros and cons of the big-box format have
been widely debated (the “Walmart effect”). We provide a dynamic model
that takes key features of retail markets into account. Apart from large
entrants, we emphasize the importance of local markets, imperfect competition, lumpy investments, and limited access to quantity data on products
purchased and sold by each store. We analyze whether large entrants force
low productivity stores out of the market and increase productivity among
surviving stores with different positions in the productivity distribution.
Our empirical application relies on detailed data on all retail food stores in
Sweden, a sector that is representative of many OECD markets in terms of
31
market structure and regulation.
The results show substantial heterogeneity in the positive effects of large
entrants on future productivity. A key finding is that productivity increases
decline toward the upper part of the productivity distribution, implying
that productivity increases relatively more among low productivity incumbents than among high productivity incumbents. The median increase in
incumbents’ productivity due to a large entrant is 4-6 percent. In addition, productivity increases by 1-2 percentage points more for a store in
the 25th local market productivity percentile than for a store in the 75th
percentile. The average aggregate weighted increase in productivity at the
local market level is 3-5 percent, using output market shares as weights.
The findings are informative to policymakers who decide over entry of new
stores in local markets and are robust to various identification strategies
It is important to consider that stores compete in local markets. Furthermore, not controlling for the contemporaneous effect of large entrants on
prices leads to underestimation of their impact on productivity. We conclude that entry of big-box stores catalyzes retail productivity growth.
Our findings contribute knowledge relevant to competition policy, as
entry regulation issues are a significant concern to policymakers in Europe,
where such regulations are generally much more restrictive than in the U.S.
As an example, the European Commission recently highlighted an investigation of the food sector (European Commission, 2012). We argue that
a more restrictive design and application of entry regulations can hinder
aggregate productivity growth in local markets. In addition to productivity, entry regulations compound a wide range of other aspects. How to
balance potential productivity growth against increased traffic and broader
environmental effects and to consider dynamic game frameworks are interesting topics for future research (Pakes et al., 2007; Holmes, 2011; Dunne
et al., 2013; Maican and Orth, 2013; and Sweeting, 2013).
32
References
Ackerberg, D., L. Benkard, S. Berry, and A. Pakes (2007):
“Econometric Tools for Analyzing Market Outcomes,” Handbook of
Econometrics, 6, 4171–4276.
Ackerberg, D., K. Caves, and G. Fraser (2006): “Structural Identification of Production Functions,” Mimeo, UCLA.
Ackerberg, D., X. Chen, and J. Hahn (2012): “A Practical Asymptotic Variance Estimator for Two-Step Semiparametric Estimators,” Review of Economics and Statistics, 94(2), 481–498.
Aghion, P., R. Blundell, R. Griffith, P. Howitt, and S. Prantl
(2009): “The Effects of Entry on Incumbent Innovation and Productivity,” Review of Economics and Statistics, 91(1), 20–32.
Aghion, P., and R. Griffith (2005): Competition and Growth: Reconciling Theory and Evidence. MIT Press, MA.
Asplund, M., and R. Friberg (2002): “Food Prices and Market Structure in Sweden,” Scandinavian Journal of Economics, 104(4), 547–567.
Asplund, M., and V. Nocke (2006): “Firm Turnover in Imperfectly
Competitive Markets,” Review of Economic Studies, 73(2), 295–327.
Bairam, E. (1994): Homogeneous and Nonhomeogeneous Production
Functions. Avebury, Ashate Publishing Ltd.
Basker, E. (2005): “Job Creation or Destruction? Labor-Market Effects
of Wal-Mart Expansion,” Review of Economics and Statistics, 87(1),
174–183.
(2007): “The Causes and Consequences of Wal-Mart’s Growth,”
Journal of Economic Perspectives, 21(3), 177–198.
(2012): “Raising the Barcode Scanner: Technology and Productivity in the Retail Sector,” American Economic Journal: Applied
Economics, 4(3), 1–27.
(2014): “Change at the Checkout: Tracing the Impact of a Process
Innovation,” Journal of Industrial Economics ( forthcoming).
Basker, E., and M. Noel (2009): “The Evolving Food Chain: Competitive Effects of Wal-Mart’s Entry into the Supermarket Industry,” Journal
of Economics and Management Strategy, 18(4), 177–198.
33
Bertrand, M., and F. Kramarz (2002): “Does Entry Regulation Hinder Job Creation? Evidence from the French retail industry,” Quarterly
Journal of Economics, 117(4), 1369–1413.
Bond, S., and M. Söderbom (2005): “Adjustment Costs and the Identification of Cobb-Douglas Production Functions,” Working Paper 05/04,
Institute of Fiscal Studies.
Boylaud, O., and G. Nicoletti (2001): “Regulatory reform in retail
distribution,” OECD Working Paper 32.
Buccirossi, P., L. Ciari, T. Duso, G. Spagnolo, and C. Vitale
(2013): “Competition Policy and Productivity Growth: An Empirical
Assessment,” Review of Economics and Statistics ( forthcoming), 95(4),
1324–1336.
Collard-Wexler, A. (2011): “Productivity Dispersion and Plant Selection in Ready-Mix Concrete Industry,” Mimeo, New York University.
De Loecker, J. (2011): “Product Differentiation, Multi-Product Firms
and Estimating the Impact of Trade Liberalization on Productivity,”
Econometrica, 79(5), 1407–1451.
Djankov, S., R. La Porta, F. Lopez-de Silanes, and A. Shleifer
(2002): “The Regulation of Entry,” Quarterly Journal of Economics,
CXVII(1), 1–37.
Doraszelski, U., and J. Jaumandreu (2013): “R&D and Productivity: Estimating Endogenous Productivity,” Review of Economic Studies,
80(4), 1338–1383.
Dunne, T., S. Klimek, M. Roberts, and Y. Xu (2013): “Entry,
Exit and the Determinants of Market Structure,” The RAND Journal of
Economics, 44(3), 462–487.
Ericson, R., and A. Pakes (1995): “Markov-Perfect Industry Dynamics: A Framework for Empirical Work,” Review of Economic Studies,
62(1), 53–83.
European Commission (2012): “The Economic Impact of Modern Retail
on Choice and Innovation in the EU Food Sector,” Written Declaration.
European Competition Network (2011): “ECN Brief Extended Issue,” ECN Brief 05/2011, European Competition Network.
34
European Parliament (2008): “On Investigating and Remedying the
Abuse of Power by Large Supermarkets Operating in the European
Union,” Written Declaration 0088, European Parliament.
Fishman, C. (2006): The Wal-Mart Effect. The Penguin Group.
Foster, L., J. Haltiwanger, and C. Krizan (2001): “The Link Between Aggregate and Micro Productivity Growth: Evidence from Retail
Trade,” NBER Working Paper 13783.
(2006): “Market Selection, Reallocation, and Restructuring in
the U.S. Retail Trade Sector in the 1990s,” Review of Economics and
Statistics, 88(4), 748–758.
Foster, L., J. Haltiwanger, and C. Syverson (2008): “Reallocation,
Firm Turnover, and Efficiency: Selection on Productivity or Profitability?,” American Economic Review, 98(1), 394–425.
Gandhi, A., S. Navarro, and D. Rivers (2011): “On the Identification
of Production Functions: How Heterogenous is Productivity?,” mimeo,
University of Wisconsin-Madison.
Griffith, R., and H. Harmgart (2005): “Retail Productivity,” International Review of Retail, Distribution and Consumer Services, 15(3),
281–290.
Griliches, Z., and H. Regev (1995): “Productivity and Firm Turnover
in Israeli Industry: 1979-1988,” Journal of Econometrics, 65, 175–203.
Haavelmo, T. (1943): “The Statistical Implications of a System of Simultaneous Equations,” Econometrica, 1, 1–12.
Hall, R. (1988): “The Relation between Price and Marginal Cost in U.S.
Industry,” Journal of Political Economy, 96(5), 921–947.
Hausman, J. (1997): The Economics of New Goodschap. Valuation of New
Goods Under Perfect and Imperfect Competition, p. 207Â248. University
of Chicago Press.
Hicks, M. J. (2007): The Local Economic Impact of Wal-Mart. Cambria
Press, Youngstown, New York.
Hoj, J., T. Kato, and D. Pilat (1995): “Deregulation and Privatisation
in the Service Sector,” OECD Working Paper 25.
35
Holmes, T. (2011): “The Diffusion of Wal-Mart and Economies of Density,” Econometrica, 79(1), 253–302.
Jia, P. (2008): “What Happens When Wal-Mart Comes to Town: An
Empirical Analysis of the Discount Retailing Industry,” Econometrica,
76(6), 1263–1316.
Katayama, H., S. Lu, and J. Tybout (2009): “Firm-level Productivity
Studies: Illusions and a Solution,” International Journal of Industrial
Organization, 27(3), 403–413.
Klette, T., and Z. Griliches (1996): “The Inconsistency of Common
Scale Estimators when Output Prices are Unobserved and Endogenous,”
Journal of Applied Econometrics, 11(4), 343–361.
Levinsohn, J., and M. Melitz (2006): “Productivity in a Differentiated
Products Market Equilibrium,” Mimeo, Harvard University.
Levinsohn, J., and A. Petrin (2003): “Estimating Production Functions Using Inputs to Control for Unobservables,” Review of Economic
Studies, 70(2), 317–341.
MacDonald, J. (1994): “Does Import Competition Force Efficient Production?,” Review of Economics and Statistics, 76(4), 721–727.
Maican, F., and M. Orth (2012): “On Estimation of Returns to Scale
under Imperfect Competition,” Mimeo, Department of Economics, University of Gothenburg.
(2013): “Entry Regulations, Product Differentiation and Determinants of Market Structure,” Mimeo, Reserach Institute of Industrial
Economics (IFN) and University of Gothenburg.
(2014): “A Dynamic Analysis of Retail Productivity,” Mimeo, Reserach Institute of Industrial Economics (IFN) and University of Gothenburg.
Nickell, S. (1996): “Competition and Corporate Performance,” Journal
of Political Economy, 104(4), 724–746.
Ofer, G. (1973): “Returns to Scale in Retail Trade,” Review of Income
and Wealth, 19(4), 363–384.
Olley, S., and A. Pakes (1996): “The Dynamics of Productivity in the
Telecommunications Equipment Industry,” Econometrica, 64(6), 1263–
1297.
36
Pakes, A., S. Berry, and M. Ostrovsky (2007): “Simple Estimators for the Parameters of Dyscrete Dynamic Games, with Entry/Exit
Examples,” RAND Journal of Economics, 38(2), 373–399.
Petrin, A., and K. Train (2010): “A Control Function Approach to Endogeneity in Consumer Choice Models,” Journal of Marketing Research,
XLVII, 3–13.
Pilat, D. (1997): “Regulation and Performance in the Distribution Sector,” OECD Working Papers 180.
Reynolds, J., E. Howard, D. Dragun, B. Rosewell, and
P. Ormerod (2005): “Assessing the Productivity of the UK Retail
Sector,” International Review of Retail, Distribution and Consumer Research, 15(3), 237–280.
Roeger, W. (1995): “Can Imperfect Competition Explain the Difference
between Primal and Dual Productivity Measures? Estimates for U.S.
Manufacturing,” Journal of Political Economy, 103(2), 316–330.
Sadun, R. (2014): “Does Planning Regulation Protect Independent Retailers?,” Review of Economics and Statistics ( forthcoming).
Schivardi, F., and E. Viviano (2011): “Entry Barriers in Retail Trade,”
Economic Journal, 121(155), 145–170.
Staiger, D., and J. Stock (1997): “Instrumental Variables Regression
with Weak Instruments,” Econometrica, 65(3), 557–586.
Statistics Sweden (2005): “Hushållens ekonomi - vem lånar, vem konsumerar? (The Economy of Households: Who lends, who comsumes?),”
Mimeo, Stockholm.
Swedish Competition Authority (2001:4): “Kan Kommunerna
Pressa Matpriserna? (Can the Municipalities Put Pressure on Prices?),”
Technical Report 4, Stockholm.
(2002:6): “Dagligvaruhandeln: Struktur, Ägarform och relation
till leverantörer? (Retail Food: Market structure, ownership and suppliers),” Technical Report 6, Stockholm.
Swedish National Board of Housing, Building, and Planning
(2005): “Shopping Matters! Retailing and Sustainable Urban Development,” Technical Report.
37
Sweeting, A. (2013): “Dynamic Product Positioning in Differentiated
Product Markets: The Effect of Fees for Musical Performance Rights on
the Commercial Radio Industry,” Econometrica, 81(5), 1763–1803.
Syverson, C. (2004): “Market Structure and Productivity: A Concrete
Example,” Journal of Political Economy, 112(2), 1181–1222.
(2011): “What Determines Productivity?,” Journal of Economic
Literature, 49(2), 326–365.
Wooldridge, J. M. (2009): “On Estimating Firm-Level Production
Functions Using Proxy Variables to Control for Unobservables,” Economics Letters, 104(3), 112–114.
38
Table 1: Characteristics of the Swedish Retail Food Market
A. FS-RAMS
Year
No. of
stores
1996
1997
1998
1999
2000
2001
2002
B. DELFI
Year
3,714
3,592
3,482
3,398
3,287
3,094
3,067
No. of
employees
74,100
73,636
74,696
74,758
77,180
76,905
80,931
Wage
bill
Value
added
Total
sales
9,882,234
10,322,136
10,766,043
11,110,785
11,536,063
11,522,482
12,081,931
18,319,407
18,838,130
19,185,120
19,570,472
20,389,492
20,748,902
22,473,696
141,743,876
142,840,611
147,726,647
152,160,949
154,106,865
158,512,132
179,335,162
Value added
per employee
247.22
255.83
256.84
261.78
264.18
269.79
277.69
Large
Large
Mean sales
Total sales
Total
Value added per
stores
entry
space (m2 )
space (m2 )
sales
sales space
1996
905
21
538
2,510,028
129,326,000
7.29
1997
925
8
550
2,483,248
126,732,397
7.58
1998
926
9
587
2,552,794
130,109,604
7.52
1999
936
18
604
2,514,367
133,156,023
7.72
2000
948
23
654
2,587,952
138,314,044
7.80
2001
942
28
689
2,471,510
139,352,920
8.23
2002
932
5
718
2,525,084
142,532,944
8.72
NOTE: FS-RAMS is provided by Statistics Sweden and consists of all organization numbers in SNI code 52.1, i.e.,
“multi-store” units that contain one store or several (e.g., due to the same owner). Sales (incl. 12% VAT), valueadded, wages, value-added per employee and sales space are measured in thousands of 1996 SEK (1USD=6.71SEK,
1EUR=8.63 SEK). DELFI is provided by Delfi Marknadspartner AB. Sales in DELFI are collected by surveys and
reported in classes, while sales are based on tax reporting in FS-RAMS. Therefore, total sales are lower in DELFI than
in FS-RAMS. Value-added per employee is defined using the number of full-time adjusted employees in FS-RAMS.
Value-added per sales space (m2 ) is defined using value-added from FS-RAMS and sales space from DELFI. From
1996 to 2002, the total population in Sweden increased from 8,844,499 to 8,940,788.
39
Table 2: Distribution of stores and firms across local markets and years
Total
No. of
Share of pop
no. of
firms
with nearest
stores
store < 2km
Minimum
0
0
0
0
0
3
1
0.45
10th percentile
2
0
1
0
2
7
2
0.59
25th percentile
3
1
1
0
3
9
3
0.66
50th percentile
5
2
2
0
5
15
3
0.75
75th percentile
9
4
5
0
8
25
3
0.82
90th percentile
15
8
8
1
16
44
3
0.91
Maximum
86
93
88
12
218
460
4
1.00
Mean
7.25
3.66
3.91
0.22
8.25
23.29
2.86
0.74
Std. deviation
7.74
6.76
5.81
0.89
16.87
35.34
0.55
0.12
NOTE: This table shows the distribution of the number of stores and firms across local markets as well as the share
of population with less than 2 kilometers to the nearest store. ICA, Axfood, Coop and Bergendahls are defined as
firms. Municipalities, considered as local markets, increase from 288 to 290 due to three municipality break-ups
during the period, which gives a total of 2,021 market-year observations. Distance to the nearest store is calculated
based on 800x800 meter grids and is only available for 2002 (290 observations).
ICA
Axfood
No. of stores
Coop
Bergendahls
40
Others
Table 3: Local market characteristics
Year
1997
1998
1999
2000
2001
2002
A. Productivity measures for all markets: mean (std. dev.)
Value-added per employee
249.33
252.11
271.66
256.93
258.93
266.28
(70.04)
(49.95)
(149.64)
(54.98)
(64.79)
(57.62)
Value-added per sales space (m2 )
4.85
5.01
5.11
4.95
5.16
5.55
(4.77)
(5.16)
(5.29)
(5.72)
(5.71)
(5.97)
Total no. of markets
288
288
289
289
289
290
B. Markets with large entrants: median
No. of stores
37.00
54.00
29.00
32.00
33.00
22.00
No. of all entrants
2.00
2.00
3.00
2.00
1.00
2.00
No. of all exits
3.00
2.00
2.00
3.00
1.00
-.Population
57,441.00
60,429.00
37,195.00
48,250.00
58,361.00
22,907.00
Population density
80.88
57,92.00
68.03
79.38
77.29
52.77
Per capita income
149.10
157.60
161.60
170.30
179.10
177.60
Total no. of markets
10
9
20
20
23
6
C. Markets without large entrants: median
No. of stores
15.00
15.00
15.00
14.00
13.00
14.00
No. of all entrants
0.00
0.00
1.00
0.00
0.00
0.00
No. of all exits
0.00
1.00
1.00
1.00
0.00
-.Population
14,827.00
15,133.00
14,322.00
14,154.00
14,068.00
15,207.00
Population density
25.80
25.78
25.22
25.60
24.75
26.20
Per capita income
143.30
149.10
155.90
162.50
168.40
175.90
Total no. of markets
278
279
269
269
266
284
NOTE: 1996 is left out because entrants are not observed. Municipalities, considered as local markets, increase
from 288 to 290 due to three municipality break-ups during the period. Value-added per employee is defined
using the number of full-time adjusted employees in FS-RAMS. Value-added per employee and sales space are
in thousands of 1996 SEK (1USD=6.71SEK, 1EUR=8.63 SEK). Sales space, stores, entrants and exits come
from DELFI. Population density is defined as total population per square kilometer in the municipality.
41
Table 4: Service generating function estimates
Nonparametric
Ml
Mlm
(2)
(3)
0.843
0.674
(0.006)
(0.005)
0.162
0.304
(0.004)
(0.004)
Mlme
(4)
0.679
(0.005)
0.307
(0.004)
”
“
Market output − η1
0.350
(0.013)
0.304
(0.013)
Log of population
0.018
(0.003)
-0.003
(0.004)
0.027
(0.003)
0.021
(0.004)
1.505
-2.858
1.420
-3.279
1.530
1.438
Log no. of labor
Log of capital
OLS
(1)
0.948
(0.006)
0.167
(0.004)
Log of population density
Scale (βl + βk )
Demand “elasticity
” (η)
Markup
1.121
1.005
η
1+η
Partial-F test (first-step)
194.11
(p-value)
(0.000)
No. of obs.
23,521
17,747
17,747
17,747
NOTE: The dependent variable is the log of deflated value added. Labor
is measured as the number of full-time adjusted employees. All regressions
include year dummies. In all specifications that control for imperfect competition, reported parameters include elasticity, specifically, (1 + η1 )βl for labor,
(1 + η1 )βk for capital, − η1 βx for exogenous demand shifters, and − η1 βe for
large entry (equation (3)). OLS refers to ordinary least square regression. All
M specifications include previous large entrants in the productivity process.
Ml is Ackerberg, Caves, and Fraser’s (2006) two-step estimation method, using labor demand as a proxy for productivity; Mlm is a two-step estimation
method that uses a nonparametric labor demand function as a proxy for productivity and controls for imperfect competition but assumes that wages and
large entrants are exogenous; Mlme is a two-step estimation method that
uses a nonparametric labor demand function and controls for imperfect competition and the endogeneity of large entrants in the first step. The share
of non-socialist seats in the local government and the number of large entrants in other local markets (Hausman type) are used as instruments for
current large entry. Reported standard errors (in parentheses) are robust to
heteroscedasticity. All M specifications use previous the capital stock and labor as instruments, and standard errors are computed using Ackerberg et al.
(2012). Market output is measured as the market share weighted output in
the municipality. Markup is defined as price over marginal cost.
42
Table 5: Transition matrix from t-1 (column) to t (row) in percentage
Percentile
<p10
p10-p25
p25-p50
p50-p75
p75-p90
>p90
Markets with large entrants in t-1
<p10
22.09
12.14
8.75
5.65
3.83
2.50
p10-p25
22.09
26.43
10.83
9.68
7.10
5.83
p25-p50
18.60
22.86
34.58
26.21
13.11
4.17
p50-p75
6.98
15.71
24.58
29.84
30.60
7.50
p75-p90
5.81
7.14
7.08
13.71
20.22
23.33
>p90
3.49
2.14
2.92
3.63
12.02
30.00
Exit
20.93
13.57
11.25
11.29
13.11
26.67
Markets without large entrants in t-1
<p10
27.29
14.91
8.16
3.93
3.93
4.31
p10-p25
22.66
24.02
15.33
8.25
8.25
3.56
p25-p50
18.23
30.61
32.24
22.39
22.39
6.26
p50-p75
8.68
11.51
22.46
32.21
32.21
13.59
p75-p90
3.09
4.70
8.07
15.50
23.98
23.30
>p90
3.76
2.97
3.76
7.68
17.19
27.29
Exit
16.30
11.29
9.99
10.04
11.16
21.68
NOTE: Productivity is estimated using Mlm described in Section 3. Productivity is backed out from
the value-added generating function. Municipalities are considered as local markets. Large entrants
in period t-1 are defined as the five largest store types in the DELFI data (hypermarkets, department
stores, large supermarkets, large grocery stores, and other stores).
43
Table 6: Summary statistics: Large entrants and future productivity
Panel A: Marginal effects of large entrants for different productivity percentiles in local markets
10th
25th
50th
75th
90th
percentile
percentile
percentile
percentile
percentile
productivity
productivity
productivity
productivity
productivity
Mean
0.082
0.074
0.065
0.053
0.045
Mlm
Std.Dev.
0.025
0.022
0.022
0.023
0.024
Mean
0.060
0.054
0.045
0.036
0.029
Mlme
Std.Dev.
0.020
0.018
0.018
0.019
0.019
Adjusted R2
0.988
0.988
No. of obs.
12,760
12,760
Panel B: Distribution of marginal effect of large entrants on individual store’s future productivity
Support
Mlm
Mlme
[0.017 0.117]
[0.008 0.090]
Panel C: Weighted change in aggregate local market productivity after large entry
Mlm
Mlme
Minimum
0.0002
0.0002
25th percentile
0.040
0.028
50th percentile
0.057
0.040
Mean
0.050
0.036
75th percentile
0.063
0.047
Maximum
0.106
0.078
NOTE: The figures are computed using the estimated controlled Markov process of productivity. Panel A shows means and standard deviations of the marginal effects across local
entry markets and years. For each market and year, we computed the effects for various
local productivity percentiles. Panel B shows the support for the impact of large entrants
on productivity. The support is computed using 1,000 simulation draws from the estimated
distribution of productivity. Panel C shows the distribution of changes in aggregate local
market productivity after large entry. The figures are computed asPweighted averages of innm
L
dividual stores’ productivity, using market shares as weights, i.e.,
j=1 sjt (∂h(·)/∂emt−1 )).
Mlm is a nonparametric two-step approach that controls for imperfect competition, where
wages and large entrants are exogenous. Mlme is a nonparametric two-step approach that
controls for imperfect competition, where large entrants are endogenous. The share of nonsocialist seats in the local government, the previous number of large entrants, and the number of large entrants in other local markets (Hausman type) are used as instruments for
large entry.ˆ Productivity is recovered from the value-added generating function:
ωjt =
˜
(η/(1 + η)) yjt − (1 + 1/η)[βl ljt + βk kjt ] + (1/η)qmt + (1/η)x0mt β x + (1/η)βe eL
mt . We drop
extreme values by removing 3 percent of the observations from each tail of the marginal effect
distribution. Large entrants in period t − 1 are defined as the five largest store types in the
DELFI data (hypermarkets, department stores, large supermarkets, large grocery stores, and
other stores).
44
Table 7: Regression results: Exit
Mlm
Log of productivityt
Large entrantst-1
p10*Large entrantst-1
p10-p25*Large entrantst-1
p25-p50*Large entrantst-1
p75-p90*Large entrantst-1
p90*Large entrantst-1
Log of capitalt
(1)
-0.124
(0.027)
0.043
(0.043)
(2)
-0.115
(0.092)
0.336
(0.132)
0.263
(0.131)
0.193
(0.118)
0.080
(0.143)
0.189
(0.145)
-0.082
(0.013)
0.066
(0.022)
0.001
(0.017)
-0.196
(0.221)
-0.090
(0.014)
Log of populationt
0.054
(0.023)
Log of population densityt
-0.004
(0.017)
Log of incomet
-0.054
(0.224)
Year dummies
Yes
Yes
No. of obs.
11,132
11,132
NOTE: This table shows probit regressions on exit. Productivity
is estimated using Mlm described in Section 3. Reported standard
errors (in parentheses) are robust to heteroscedasticity. Large entrants in period t-1 are defined as the five largest store types in the
DELFI data (hypermarkets, department stores, large supermarkets, large grocery stores, and other stores). We use six percentile
bins for productivity in each local market and year, with p50-75
used as reference group.
45