Upload
others
View
9
Download
0
Embed Size (px)
Citation preview
Entry into Local Retail Food Markets in Sweden:
A Real Options Approach∗
Sven-Olov Daunfeldt†, Matilda Orth‡, and Niklas Rudholm§
Abstract
In this paper a real options approach is used, incorporating uncertainty
and irreversibility of investments, to study the number of stores enter-
ing the Swedish retail food market between 1994 and 2002. The results
presented in this paper supports the claim that uncertainty affects the
entry decision. According to the results, entry is less frequent in highly
concentrated local retail food markets characterized by a high degree
of uncertainty; whereas higher profit opportunities seem to increase
the probability of entry.
Key words: Real options, uncertainty, retail food, entry, negative
binomial regression.
JEL classification: L13, L81.
∗Financial support from FORMAS and the Jan Wallander and Tom Hedelius founda-tion is gratefully acknowledged.
†The Swedish Research Institute of Trade (HUI), S-103 29, Stockholm, Sweden; and De-partment of Economics, University of Gävle, S-804 26, Gävle, Sweden. E-mail: [email protected].
‡Corresponding Author. Department of Economics, School of Economics andCommercial Law, Gothenburg University, S-405 30 Gothenburg, Sweden. E-mail:[email protected]
§Department of Economics, Umeå University, S-901 87 Umeå, Sweden. E-mail:[email protected].
1
1 Introduction
The question of firm entry has received considerable attention in the litera-
ture (see e.g., Dunne et al., 1988; Audretsch and Fritsch, 1994; Keeble and
Walker, 1994; Love, 1996) and it is well known that entry rates vary strongly
across industries (see e.g., Dunne et al, 1988; Berglund and Brännäs, 2001).
However, entry within the retail food industry is, to our knowledge, only
analyzed in a very small number of economic studies. Coterill and Haller
(1992), who analyzed the entry behavior of leading supermarket chains in
the United States, provides one rare exception. Their results indicated that
the decision to enter a market was influenced both by the market structure
and the capabilities of potential competitors.
Over the last decade the retail food industry has been characterized by
great changes (see e.g., Clarke et al., 2002). Large firms, e.g., Wal-Mart,
Aldi, Carrefour and Tesco, have expanded into new markets and become
even larger, the store concepts have become more visualized and chain store
operators have chosen to invest more in store brands. The retail food mar-
ket in Sweden provides no exception. However, in most other countries the
share of total retail food sales accounted for by the four largest chain store
operators varies between 30 and 80 percent (see Table 6.5 in Clarke et al.,
2002); while, in Sweden, three chain store operators (ICA, COOP and Ax-
food) have a total market share exceeding 89 percent in 2002. It is still
an open question if the market structure in Sweden only supports a limited
number of chain store operators or if ICA, COOP and Axfood’s dominant
2
market position is a result of low competitiveness and entry barriers.1
The purpose of this paper is to analyze entry within the Swedish retail
food industry. One shortcoming with most previous entry studies, irrespec-
tive of the market analyzed, is that they ignore both uncertainty regarding
future market conditions and the effects of potential investment irreversibil-
ity, i.e., sunk investment costs. In this paper, the possible irreversibility of
the investment and the uncertainty regarding future profits are included in
both the theoretical and the empirical part of the paper. More specifically,
the economic value of being able to defer the decision to enter a market is
studied using the theory of real options, see e.g., Dixit and Pindyck (1994).
The predictions of the model is then tested using a unique data set that
cover all retail food stores in Sweden during the years 1994 to 2002. The
paper contributes to the existing literature in the following ways; First, as far
as we know, this paper is one of exceptionally few studies that analyzes entry
within the retail food industry. Second, in contrast to most previous entry
studies, we incorporate uncertainty and investment irreversibility in both
the theoretical and the empirical analysis. Third, in comparison to previous
entry studies, we are also able to control for several additional confounding
factors such as industry, region - and store type specific determinants of
entry.
1Recently, the Swedish competition authority (see Swedish competetion authority, re-port 2002:5) released a report that indicated that prices in Sweden were around 11 per-cent higher, on average, than food retail prices in the European Union. According tothe Swedish competition authority, the price difference was primarly explained by lowcompetetivness. On the other hand, Asplund and Friberg (2002) presented results thatindicated that the degree of market concentration only had a modest effect on food retailprices in Sweden and Bergman (2004) argues that the relatively high food retail prices inSweden primarly can be explained by high direct and indirect taxes.
3
The empirical results presented in this paper supports the claim that
uncertainty affects the decision to establish a new retail food store in a spe-
cific region. According to the results, a high degree of uncertainty regarding
future profits reduces entry into the Swedish retail food market; while higher
potential profits seem to attract entry. However, according to the calculated
marginal effects, the impact on entry of both profits and uncertainty is small
in size.
Moreover, entry is found to be less common in highly concentrated mar-
kets, and regions characterized by relatively low purchasing power. The
calculated marginal effects indicate that the degree of market concentration
and the purchasing power have a greater impact on entry compared to the
effects of potential profits and investment uncertainty. Finally, regions char-
acterized by a non-socialist local government seem to attract more entry,
suggesting that the decision to establish a new store is influenced by the
political preferences of the local government as well.
The paper is organized as follows. In the next section, the theoretical
framework of our study is presented. The data and the empirical method are
described in section 3; while the results are presented in section 4. Finally,
section 5 concludes the paper.
2 Theoretical Model
The theoretical model underlying the empirical study is based on the theory
of real options, see e.g., Dixit and Pindyck (1994). Note that there are three
important characteristics of the investment problem that must be fulfilled
4
for this approach to be appropriate. First, there must exist some degree
of uncertainty about the future ”state of the world”, i.e. some uncertainty
about future market conditions. Second, the decision to enter a market must
entail some irreversible commitment of resources, i.e. some of the investment
cost must be sunk. Finally, the potential entrant must have some discretion
as to the timing of entry into the market.
It is reasonable to assume that the decision to establish a new store is
influenced both by the degree of uncertainty regarding future profits and
the irreversibility of the investment. When a chain store operator wants to
invest in a new store in a specific region, they do not know with certainty how
the demand conditions and prices will develop in the future. In addition,
they know that there are other agents facing similar decisions and that their
decisions will affect future revenues, conditional on entry. Opening a new
store also involves large sunk costs that cannot be retrieved if the investment
fails. As building costs are approximately equal across regions in Sweden,
while real estate values are much lower in less densely populated areas, the
level of sunk costs are much greater in less densely populated areas.
2.1 Investment under uncertainty
The decision process, when a chain store operator plans to invest in a new
store in a specific region, can be described as follows:2 First, the profitabil-
ity of the project is evaluated. The purchasing power of the region is an
important decision variable when future revenues conditional on entry are
2The description of the entry decision process is based on interviews with FredrikBergström, president at the Swedish Research Institute of Trade (HUI) and Peder Larsson,chief operating officer, at ICA Sweden AB.
5
calculated, where purchasing power often is approximated using population
size and measures regarding the private consumption of convenience goods.
In the next step, the supply side of the market is analyzed to determine if a
new store will prevail given the degree of local competition. Finally, possible
entry barriers are investigated.
Now consider firm j which is planning to invest in a new store and define
the firm’s appropriately discounted expected revenues, Djt, of entry as
Djt = Et
∞Zt
ujτe−ρτdτ, (1)
where ujτ is the revenue, and the firm specific discount rate is given by
ρ. In the theoretical model, the potential entrant is assumed to know the
current level of revenues ujt, while future values are uncertain due to possible
changes in market conditions. In addition, the potential entrant is assumed
to have discretion over the timing of investment.3 The revenues generated
by an investment in a food retail store is assumed to evolve according to the
geometric Brownian motion
dDjt = αDjtdt+ σDjtdzt. (2)
where dzt is the increment of a Wiener process, σ the volatility coefficient
and α the drift rate. Equation (2) implies that the current expected revenues
of the decision to enter the market is known to the potential entrant. Fu-
3 In practice, some municipalities in Sweden have used the Planning and Building Act(PBA) to prevent or delay entry. Although these municipalities have not in general beenable to prevent entry, it might have been delayed for some entrants.
6
ture values are, however, uncertain due to the uncertainty concerning future
market conditions.
A chain store operator who plans to build a new food retail store usually
knows the size of the investment costs, Ij , through estimates from contrac-
tors, as well as the costs of operating a certain type of store concept, cjτ ,
from prior experience. The discounted expected costs, Ujt, of entry can then
be written
Ujt = Et
∞Zt
f(cjτ , Ij)e−ρτdτ (3)
where cjτ is the operating costs of the food retail store, and Ij is the size
of the initial investment to set up operations, including the costs for land,
buildings etc. The initial investment, Ij , is assumed to be fully sunk if the
investment fails.4 Sunk costs in combination with uncertainty about future
revenues creates an option value of waiting, as long as there is a possibility
that the chain store management will revise their estimate of future revenues
downward a period from now and choose not to enter the market (see e.g.,
McDonald and Siegel, 1986; Dixit and Pindyck, 1994).
As shown by Dixit and Pindyck (1994), the solution to an investment
problem of this type is given by a second order differential equation which
can be written;
1
2σ2D2
jtF00(Djt) + αDjtF
0(Djt)− ρF (Djt) = 0. (4)
4One could include some form of ”salvage value” if the investment fails in the model.This would, however, not alter any of the testable predictions from our model, whilemaking the theoretical model more complex. The possible ”salvage value” in case of failureis thus excluded from the theoretical model. In the empirical section, the salvage value ofthe investment is included in our measurement of the irreversibility of the investment.
7
where F (Djt) denotes the value of the firm’s option to the market at time
t.5 The value of the option thus satisfies this equation along with boundary
conditions
F (0) = 0
F (D∗j ) = D∗j − Uj
F 0(D∗j ) = 1
The first boundary condition arises from the observation that if the value
D goes to zero, it will stay at zero (i.e. zero is an absorbing state for the
geometric Brownian motion). The second condition is the ”value-matching
condition” expressing the fact that at the optimal trigger level (D∗j ), the pay-
off is simply the net profit, i.e. revenues minus costs. The third condition is
the ”smooth pasting condition”, see e.g., Dixit (1993).
The general solution to equation (4) is given by
F (Dj) = A1Dβ1j +A2D
β2j (5)
where A1 and A2 are constants to be determined, and where β1 and β2 are
the positive and the negative root, respectively, of the fundamental equation
Q =1
2σ2β(β − 1) + αβ − ρ = 0. (6)
As shown by Dixit and Pindyck (1994), the solution to this problem is
5A thorough description of the process used to arrive at equation (4) is given in Dixitand Pindyck (1994), chapter 5.
8
given by the following equations6:
A1 =(β1 − 1)(β1−1)
ββ11 U
(β1−1)j
(7)
and
D∗j =β1
(β1 − 1)Uj (8)
where β1 is the positive root of the fundamental equation presented above.
The implication of this is that uncertainty concerning future profit opportu-
nities creates a value of waiting for more information, since β1/(β1−1) > 1 .
For comparison the optimal behavior without uncertainty would be to enter
the market whenever Djt ≥ Ujt.
In order to apply the model empirically using our data, two additional
problems must be solved. First, a potential problem is that equation (8)
gives an expression of the optimal level D∗j at which to enter, while what is
observed in our dataset is the actual time at which entry occurs. Second,
we observe the total number of entrants; while the model considers the
decision problem for each potential entrant. We must, therefore, also derive
an expression for the expected number of firms entering the market during
the discrete time period t to t+ 1.
Oksendal (1995) has shown that the optimal time to enter the market
under the conditions set out above will be given by T ∗j , where
T ∗j = inf©t > 0 : Djt /∈ (0,D∗j )
ª. (9)
6This follows since the first boundary condition implies that A2 = 0.
9
Hence, T ∗j is the first time at which the firm’s revenues exceed the interval
(0,D∗j ). If Dj0 ≥ D∗j , then E(T ∗j ) = 0, and immediate entry is optimal.
Now assume that expected switching times, T , among firms which has
not entered the market at time t, has a distribution given byR∞t g(T ) dT ,
which is assumed to be constant over time. The expected number of firms
entering the market, Nt, during the discrete time period t to t + 1 is then
given by
Nt =
Z t+1
tg(T )dT = h(u, c, I, σ, α) (10)
Equation (10) gives the number of firms entering the market between time
t and t+1. All variables affecting the revenues of the entry decision as pre-
sented by equation (1), and the costs as presented by equation (2), will enter
the analysis by affecting the distribution of T through D∗j = Ujβ2/(β2− 1).
Note that increased uncertainty affects equation (10) by shifting the mean
of the switching time distribution to the right. Fewer firms will then reach
their critical value during (t, t+ 1) and enter the market. This means that
entry will be less frequent in markets where the uncertainty concerning rev-
enues is large. Variables increasing the revenues of entering a market will
increase entry, while the opposite is true for variables which increase the
costs of the investment. Finally, if the size of the sunk cost (i.e. the level of
irreversibility) increases, entry will be less frequent.
10
3 Data and Econometric specification
3.1 Data
The empirical analysis is based on data obtained from DelfiMarknadsparter
AB (DELFI), Statistics Sweden (SCB) and the Swedish Research Institute
of Trade (HUI). The data obtained from DELFI cover all stores operating in
the Swedish retail food industry from 1993 to 2002 and include store specific
information such as revenues, location, sales area, store type and chain store
affiliation. Our dependent variable in the empirical analysis is the number of
stores in a specific category j (j = 1, ..., 12) entering the Swedish retail food
industry in municipality m (m = 1, ..., 290) in period t (t = 1994, ..., 2002).
The independent variables are grouped into three different categories:
Industry-specific factors, regional characteristics and store type indicators.
Industry-specific factors include measurements of average revenues per firm
per unit of sales area (REV AREA), uncertainty (UNCERT ), the level of
irreversibility of the investments (IRREV ERS) and the concentration of
chain stores at the local market (CCR).
We use municipality-specific data that are provided by Statistics Sweden
(SCB) to control for region specific determinants of entry. These data in-
clude measures of the average per capita income (INCOME), the presence
of a university or a university college (DUNIV ) and the political prefer-
ences in the municipality (DCONS). The relative purchasing power of the
region (PPOWER) is approximated using a sales index developed by the
Swedish Research Institute of Trade (HUI). An index above 100 indicates
11
that the region is characterized by more sales than expected from the size
of the population and vice versa. Descriptive statistics and definitions for
our variables used in the empirical analysis are presented in Table 1.
- Table 1 About Here -
The retail food industry consists of a number of different type of stores,
e.g., gas-station and convenience stores, grocery stores, supermarkets and
hypermarkets. The service level, the product assortment and the location
differ dramatically between these trading concepts. For instance, hypermar-
kets are classified as self-service stores with at least 2 500 square meters of
sales area, a broad range of food and non-food products, external location
and supported with no less than 300 parking spaces; while gas-station stores
are small stores in connection to gas stations with a very limited product
mix.7 It can be argued that a hypermarket not necessarily competes with
a gas-station store because the service level, product mix, prices and loca-
tion differ considerably between these type of store concepts. This means
that entry within the retail food market may be influenced by store specific
factors. In Table 2, descriptive statistics concerning entry are disaggregated
into different type of stores.
- Table 2 About Here -
In the beginning of the study period the Swedish retail food market was
dominated by four groupings of chain stores (ICA, COOP, Axel Johnson and
7Classification according to DELFI.
12
the D-group). In 1998, Axel Johnson and the D-group merged under the
name of Axfood. As can be seen from Table 3, these three companies control
89 percent of the market in 2002. This means that the Swedish food retail
trade market is highly concentrated; which may prevent entry of newcomers
and indicate low competitiveness. On the other hand, in a European context,
the Swedish retail food market is relatively small, located in the periphery
and characterized by high establishment costs. Hence, it can be argued that
the Swedish retail food market is a "natural" oligopoly supporting only a
limited number of chain store operators.
- Table 3 About Here -
As can be seen from Table 3, the number of stores operating in the
Swedish retail food market has decreased during the study period. ICA is
clearly the dominant agent in the Swedish retail food market, with a market
share that approximately equals 44 percent in 2002. In some municipali-
ties, ICA has a market share that exceeds 70 percent. Traditionally, ICA
has been a cooperation of independent stores that collaborates regarding
purchasing, transport and marketing. However, in recent years, ICA has
adopted a more centralized decision-making regarding, for instance, prod-
uct assortment. COOP has traditionally been the second largest agent in
the Swedish retail food market and consists of a group of regional coopera-
tives that are centrally coordinated. As can be seen from Table 3, COOP’s
share of the market has decreased during the study period. The third large
agent in the Swedish retail food market, Axfood, was founded in 1998 and
13
did initially consist of a very heterogenous group of stores.8 In recent years,
however, Axfood has increased the amount of centralized decision-making,
limited the number of chain store concepts, and changed the strategy in
favor of low-price segment shops. Axfood has today a market share that
marginally exceeds COOP’s share of the market. The Bergendahl-group is
often considered to be the fourth actor in the Swedish retail food indus-
try. However, the Bergendahl-group is mainly established in the southwest
of Sweden and its share of the total market is only around three percent
(although increasing rapidly). The residual group Others in Table 3 con-
sists of a number of different stores and concepts, e..g., gas-station stores,
convenience stores (e.g., 7-eleven), and foreign establishments. Note that a
number of foreign firms have entered the Swedish food retail trade market
recently, e.g., Danish Netto in 2002 and German Lidl in 2003. Their share
of the Swedish retail food market is increasing but still below 2 percent (see
the Swedish Competition Authority, report 2004:2).
3.2 Econometric specification
As the number of firms entering a market is a positive integer, a count data
model is used. The common starting point for most count data analysis is
the Poisson regression model. However, a restrictive feature of the Poisson
regression model is the moment restriction E(Nmt|Xmt) = var(Nmt|Xmt) =
µmt, where Nmt denotes the number of entrants in municipality m at time
t, Xmt is a vector of independent variables and µmt is the mean. Since the
conditional variance in most cases exceed the conditional mean (overdisper-
8Note that the name Axfood was not introduced until 2000.
14
sion) it is useful to consider the negative binomial (NB) regression model
instead. The NB model has density
Pr(Nmt|Xmt) =Γ(Nmt + φ−1)
Γ(Nmt + 1)Γ(φ−1)
µφ−1
φ−1 + µmt
¶φ−1 µµmt
φ−1 + µmt
¶Nmt
(11)
φ ≥ 0, Nmt = 0, 1, 2, 3...,
where Γ(.) is the gamma function and φ is a dispersion parameter. The
variance function for the NB model used in this study is given by µmt+φµ2mt.
Hence, this model allows for a flexible relationship between the conditional
mean and the conditional variance in the model.
Since the number of entrants in different markets is observed over time
(longitudinal/panel data) it is possible to control for store, industry and
municipality specific heterogeneity. In the present study, where the num-
ber of store types and municipalities is fairly small, fixed effects NB models
are easily estimated with maximum likelihood by specifying the exponential
mean function as exp(α+PJ
j=1 γjSjt +PM
m=1 ηmRmt + β0sZmt + δ
0zYmt +
θ1T + θ2T2).9 The store type specific indicator variables, Sjt, are equal to
one if the observation is for store type j and zero otherwise, while the munic-
ipality specific indicator variables, Rmt, are equal to one if the observation is
9 In addition, random effects NB models have also been estimated. In these models, therandom effects are assumed beta distributed which give a closed form analytic expressionfor the unconditional density forming the basis for maximum likelihood estimation (se e.g.Hausman et al 1984). However, if the individual effects are correlated with the regressors,the random effects specification of the model will suffer from inconsistency due to omittedvariables. The null hypothesis of orthogonality of the random effects and the regressorshave been tested using a Hausman test (Greene, 1993, p.479). A value of the test statisticχ2 = 114.61 makes it possible to reject the null hypothesis at all reasonable significancelevels. As such, all results presented in the paper are from the fixed effects specificationsof the model.
15
for municipality m. Characteristics of the incumbents and the local market
are captured by the vector Zmt, Ymt is a vector of regional determinants of
entry, T is the time trend10, α is a constant, and γj , ηm, β0s (s = 1, ..., 5), δ
0z
(z = 1, ..., 4), θ1and θ2 are parameters to be estimated.
Industry-specific factors, Zmt, include profit opportunities (REV AREA),
investment uncertainty (UNCERT ), investment irreversibility (IRREV ERS)
and the chain store concentration rate (CCR). The potential return of in-
vesting in a new store in municipality m is assumed to be captured by the
average revenues per firm per unit of sales area at time t−1.11 Lagging this
variable corresponds directly to the potential entrant’s decision problem,
since entrants only have access to other firms’ annual reports with a one
year time lag. Second, this setup makes it possible to alleviate a possible
endogeneity problem, since previous years’ values concerning sales revenues
are, by definition, predetermined. Uncertainty concerning the future state of
the market is measured by the variance in the average revenue per firm per
unit of sales area in municipality m, at time t−1. This variance is measured
by the conditional variance from an autoregressive conditional heteroskedas-
ticity (ARCH) model using the five first lags of the squared error terms in
the regression.12 The sales area multiplied by the negative of the popula-
10We have also used time specific fixed effects instead of the time trend and the timetrend squared. However, the qualitative results do not differ from the results presented inthis paper.11 Ideally, one would like to have data regarding the actual profit per krona invested
for each firm. However, data availability only makes it possible to calculate the averagerevenue for each firm per unit of sales area. Thus, sales area is used as a proxy for thesize of the investment of each firm, while revenue is used as a proxy for the profits of thefirm.12Several specifications for the UNCERT variable have been tried. The final specifica-
tion used in this paper was choosen using the consistent Akaike information criteria.
16
tion density is used as a measure of the irreversibility of the investments.
This is based on the fact that while building costs are approximately equal
across regions in Sweden, real estate values are much lower in less densely
populated areas. For a given level of initial investment in sales area, the
(possible) salvage value is thus much less in less densely populated areas,
and the level of sunk costs are much greater.
The effect of uncertainty on entry should be more pronounced when
the irreversibility of the investment is large. Thus, an interaction term
(UNC ∗IRR) between the level of uncertainty and the level of irreversibility
is also included in the model. Market concentration is often measured by a
Herfindahl index that equals the sum of squares of the market shares of all
firms located in a specific market. In our case, this measure is not a good
proxy for competitiveness because it does not give any indication if it is a
single actor (e.g., ICA) that totally dominates the local market. CCR is,
therefore, computed as the sum of squares of the market shares of the group
of chain stores (see Table 3 above), i.e., CCRkm = s21m + s22m + ...+ s2km is
the k-firm chain store concentration rate; where s21m is the sum of square
of ICA’s market share in municipality m, and so on. This means that if all
chain store groupings have the same size, then the chain store concentration
rate is equal to 1/k; whereas it is equal to one if the entire local market is
supplied by one chain store operator.
There exist a number of models that have been used in the empirical
entry literature to explain patterns of entry. Virtually all of these models
predict that entry should occur until expected profits in each period are
driven to zero and we, therefore, expect entry to be positively related to
17
REV AREA. In accordance with the predictions of our theoretical model
discussed in Section 2, we expect UNCERT and IRREV ERS to be nega-
tively related to entry. Highly concentrated markets are supposed to indicate
low competitiveness, which means that entry should be negatively related
to the value of the CCR variable.
Entry is also assumed to be determined by region-specific factors, Ymt.
Previous studies (see e.g., Daunfeldt et al., 2002; Fritsh and Falck, 2003)
have shown that regional factors as well as firm specific factors seem to be
important determinants of firm entry. The regional characteristics are aver-
age income in the region (INCOME), the relative purchasing power of the
region (PPOWER), the presence of higher education (DUNIV ), and the
political preferences of the local government (DCONS). A number of pre-
vious studies (see e.g., Audretsh and Fritsch, 1994; Davidsson et al., 1994;
and Guesnier, 1994) have indicated that more entry occurs in regions where
demand is high. This implies that high average income in the municipality
and relatively high purchasing power should be associated with more entry.
Audretsh and Fritsch (1994) among others have also found that entry is pos-
itively influenced by the level of education in the region, possibly indicating
that firms demand highly skilled labor. More entry is, therefore, expected in
municipalities with established universities and/or university colleges. Fi-
nally, the type of political leadership, socialist or non-socialist, might have
an effect on entry. It can, for instance, be argued that firms within the retail
food industry prefer a non-socialist local government because this type of
leadership implement more beneficial policies.
18
4 Empirical results
The estimation results are presented in Table 4 and Table 5. Two different
versions of the empirical model have been estimated; one with no region
specific fixed effects and one model where municipality specific indicator
variables were included in the analysis. The hypothesis of equal intercept
terms across municipalities is tested using a likelihood ratio test.13 A value of
the test statistic λ (289) = 843.82 makes it possible to reject the hypothesis
of equal intercept terms across municipalities at all reasonable significance
levels. We interpret this result as evidence in favor of the model that in-
cludes region-specific effects and only these estimation results are henceforth
presented.14
Table 4 About Here -
According to the results presented in Table 4, REV AREA has a positive
and significant effect on entry into the Swedish retail food industry. This
indicates that higher expected profits attract more entry into the Swedish
retail food industry. Note, however, that this result has not been widely
reported in previous entry studies (see Geroski, 1995). In accordance with
our expectations, UNCERT has a significantly negative impact on entry,
suggesting that potential entrants take uncertainty into account when they
decide whether to invest in a new store. IRREV ERS and UNC ∗ IRR are13The municipality specific fixed effects in model 2 (289 parameter estimates) are ex-
cluded in order to save space.14The first five autocorrelation coefficients of the residual are 0.018 (0.0044), -0.00089
(0.0024), -0.0030 (0.0019), 0.00063 (0.0020), and -0.0016 (0.0013); where the associatedstandard errors are reported in parantheses. Since the autocorrelation coefficients arefairly small, autocorrelation seem not to be an important problem.
19
both negative, but not statistically significant at the conventional 5% level.
CCR has a negative and statistically significant influence on entry into the
Swedish retail food industry, indicating that entry is less likely in regions
where one chain store operator dominates a large share of the local market.
Turning to the variables reflecting regional differences, INCOME is not
statistically significant at the 5% level; while PPOWER has a positive and
significant effect on entry. Hence, the results indicate that entry is not influ-
enced by the average income in the municipality, but occur more frequently
in regions characterized by relatively higher sales for convenience goods.
The presence of a university or university college in the region (DUNIV )
seem to decrease the number of entrants; while the results indicate that en-
try is more common in municipalities characterized by a non-socialist local
government (DCONS).
In order to get a sense of the economic impact of the explanatory vari-
ables on entry, marginal effects have been calculated at the mean value of
the variables included in the analysis (see Table 4). The expected num-
ber of entrants was then calculated for a hypothetical state that has the
mean value of the explanatory variables. Then the expected number of
entrants has been recalculated, increasing each statistically significant con-
tinuous variable by one standard deviation. Finally, the resulting change in
the expected number of entrants, measured in percentage points, has been
calculated. According to the results, if REV AREA increases with one stan-
dard deviation, this leads to an increase in mean entry by 2.16 percent, while
an increase in UNCERT with one standard deviation leads to an decrease
in entry by 2.40 percent. As such, the effects of REV AREA and UNCERT
20
on entry are quite small in size. The impact of relatively higher sales for
convenience goods in the region show that an increase in PPOWER by one
standard deviation leads to an 7.22 percent increase in entry. Using the
marginal effects, and evaluating the effect of an increase in CCR equaling
one standard deviation from it’s mean, results in a decrease in entry by
24.83 percent. Hence, the chain store concentration rate has a rather large
negative impact on entry.
The estimated parameters concerning the store specific indicator vari-
ables are presented in Table 5 and the results indicate that patterns of entry
differ significantly across different types of stores. As can be seen from
Table 5, department stores and supermarkets tend to enter less frequently
compared to hypermarkets; while entry is relatively more common for small
firms such as gas-station stores and mini markets. This suggests that the
Swedish food retail market is being polarized into two growing groups; hy-
permarkets and different types of small convenience stores, while the group
consisting of department stores and ordinary supermarkets is declining.
-Table 5 About Here -
5 Conclusions
In this paper we have studied entry into the Swedish retail food industry
during the years 1994 to 2002 using the theory of real options. The theoret-
ical part of this paper is based on the theory of real options (see e.g., Dixit
and Pindyck, 1994). There are three important characteristics of the prob-
lem which must be fulfilled for this approach to be appropriate. First, there
21
must exist some degree of uncertainty about the future ”state of the world”,
i.e. some uncertainty about future market conditions.¨Second, the decision
to enter a market must entail some irreversible commitment of resources,
i.e. some of the investment cost must be sunk. Finally, the potential entrant
must have some discretion as to the timing of entry into the market. In the
paper we have argued that it is reasonable to assume that these conditions
are fulfilled when a chain store operator plans to open a new store in a
specific region.
Under these conditions, Dixit and Pindyck (1994) has shown that entry
will be positively correlated to profits; while the level of uncertainty regard-
ing the profits and the level of irreversibility of the investment, i.e., the size
of the sunk costs, is negatively correlated to entry. In the empirical part of
the paper, these predictions are tested using a novel data set that cover all
stores at the Swedish retail food market between 1994 and 2002. In addi-
tion, we have also been able to control for additional confounding factors
such as the market concentration, the purchasing power in the region, the
presence of higher education in the region, the ideological inclination of the
local government as well as municipality and store type specific fixed effects.
As the number of firms entering a market is a positive integer, a negative
binomial fixed effect model have been used to study the number of stores
entering the Swedish retail food market. The results presented in the paper
supports the theoretical predictions of the model. First, entry is found to be
higher in municipalities characterized by high profit opportunities. Entry is,
moreover, found to be less common in local markets characterized by a high
degree of uncertainty regarding future revenues. This suggests that chain
22
store operators within the Swedish retail food industry take uncertainty into
account when they decide whether to invest in a new store. Note, however,
that the impact of potential profits and uncertainty was shown to be rather
small in size. The level of the irreversibility of the investment and an in-
teraction term between irreversibility and uncertainty had no statistically
significant effect on entry. In addition, it is found that a high degree of mar-
ket concentration prevents entry of newcomers. According to the marginal
effects, the impact of higher concentration on entry is large compared to the
effects of profit opportunities and investment uncertainty. This result might
be explained by possible entry barriers at the Swedish food retail market;
where incumbents, for example, engage in strategic behavior to prevent the
entry of new competitors.
Turning to the regional specific determinants of entry, the relative pur-
chasing power of the region is positively correlated to entry and the effect is
approximately three times as large as the effect of increased profit opportu-
nities and uncertainty. It is also found that more entry occur in regions with
non-socialist local governments. The latter result suggests that institutional
factors might have an affect on entry. It is, for instance, possible that non-
socialist local governments more frequently use the Plan and Building Act
(PBA) to prevent entry of hypermarkets and out-of-town shopping centres.
This is an interesting question for further research.
23
References
Asplund, M., and R. Friberg. (2002), "Food Prices and Market Structure
in Sweden", Scandinavian Journal of Economics, 104, 547-567.
Audretsch, D.B., and M. Fritsch. (1994), ”The Geography of Firm Births
in Germany”, Regional Studies, 28, 359-365.
Berglund, E., and Brännäs, K. (2001), ”Plants’ Entry and Exit in Swedish
Municipalities”, The Annals of Regional Studies, 35, 431-448.
Bergman, M. (2004), "Anpassas svenska priser till eurpeisk nivå?" (Do
Prices in Sweden Adjust to a European Level?), Ekonomisk Debatt,
32(7), 21-36.
Clarke, R., S. Davies., P. Dobson., and M. Waterson. (2002), Buyer Power
and Competetion in European Food Retailing, Edward Elgar, Cheltenham,
UK.
Coterill, R., and L. Haller. (1992), ”Barrier and Queue Effects: A Study of
Leading U.S. Supermarket Chain Entry Patterns”, Journal of Industrial
Economics, 40, 427-440.
Daunfeldt, S-O., N. Rudholm., and F. Bergström. (2002), ”Entry into Retail
- and Wholesale Trade Markets”, Umeå Economic Studies 599, Depart-
ment of Economics, University of Umeå.
Dixit, A. (1993), The Art of Smooth Pasting. Harwood Academic Publish-
ers.
Dixit, A., and R. Pindyck. (1994), Investment Under Uncertainty. Prince-
ton University Press, Princeton.
24
Dunne, T., M.J. Roberts., and L. Samuelson. (1988), ”Patterns of Firm
Entry and Exit in U.S. Manufacturing Industries”, RAND Journal of
Economics, 19, 495-515.
Fritsch, M., and Falck, O. (2003), ”New Firm Formation by Industry over
Space and Time”, DIW Discussion Papers 322, German Institute for
Economic Research, Berlin.
Greene, W.H. (1993), Econometric analysis, Prentice-Hall, New Jersey.
Hausman, J., B. Hall., and Z. Griliches. (1984), "Economic Models for
Count Data with an Application to the Patents-R&D Relationships",
Econometrica, 52, 909-938.
Keeble, D., and S. Walker. (1994), ”New Firms, Small Firms and Dead
Firms: Spatial Patterns and Determinants in the United Kingdom”, Re-
gional Studies, 28, 411-427.
Love, J.H. (1996), ”Entry and Exit: A County-level Analysis”, Applied Eco-
nomics, 28, 441-451.
McDonald, R., and D. Siegel. (1986),”The Value of Waiting to Invest”,
Quarterly Journal of Economics, 101, 707-728.
Oksendal, B. (1995), Stochastic Differential Equations: An Introduction with
Applications. Fourth edition, Springer-Verlag, Berlin.
Swedish Competition Authority (2002), Priserna kan pressas! (Prices can
be forced down!), report 2002:5, Stockholm.
Swedish Competition Authority (2004), Konsumenterna, matpriserna och
konkurrensen (Consumers, retail food prices and competition), report
25
2004:2, Stockholm.
26
Table 1. Descriptive statistics (standard deviations in parenthesis)
Variable Mean Definition and source
ENTRY 0.095 Number of entrants for each store type j (j=1,...,12) in municipality
(0.542) m (m=1,...,290) at time period t (t=1994,...,2002). Source: DELFI
REVENUES 92 836 Revenues in 1000 SEK grouped in 19 classes; where the average
(193 025) revenue in the first 18 classes are 750, 1500, 2500, 3500, 4500, 5500,
7000, 9000, 12 500, 17 500, 22 500, 27 500, 35 000, 45 000, 55 000,
67 500, 87 500, 100 000. Revenues greater than 100 000 is recorded
as the true value. Source: DELFI
SALES AREA 1794.38 Internal floor area (in square meters) used for selling and displaying
(2974.33) goods and services. Source: DELFI
REVAREA 30.31 REVENUES /SALES AREA. Source: DELFI(132.24)
UNCERT 48597 The variance of REVENUES / SALES AREA in municipality m at time
(3412922) t -1, measured by the conditional variance from an autoregressive
conditional heteroskedasticity (ARCH) model using the five first lags
of the squared error terms in the regression Source: DELFI
IRREVERS -726 144 -POPDENS*SALES AREA. Source: DELFI and Statistics Sweden.
(6 921 672)
POPDENS 135.92 Population density, measured by the ratio of the population to square
(435.59) kilometers, in municipality m. Source: Statistics Sweden.
UNC*IRR -5.83E+09 Interaction term of UNCERTAINTY and IRREVERSIBILITY.
(1.9E+11) Source: DELFI
CCR 0.390 Herfindahl index (the sum of squared market shares) of chain store
(0.098) concentration, calculated from revenues, in municipality m. Source: DELFI
INCOME 155.97 Per capita income in the municipality measured in 100 000 SEK.
(24.24) Source: Statistics Sweden
PPOWER 100.30 (Cm/Popm)/(C swe/PopSwe), where Cm and C swe is the consumption of
(43.76) convenience goods in municipality m and Sweden, respectively; while
Popm and PopSwe measures the size of the population.
Source: The Swedish Research Institute of Trade.
DUNIV 0.18 Dummy variable taking the value one if there exist a university in the
(0.38) municipality. Source: Statistics Sweden
DCONS 0.27 Dummy variable taking the value one if there is a conservative local
(0.44) government in the municipality. Source: Statistics Sweden
STORE TYPE 234.79 Stores are categorized into different types; where 100=stores under
(41.68) construction, 110=hypermarkets, 120=department stores,
200=supermarkets, 210=grocery stores, 220=small supermarkets,
230=small grocery stores, 240=convenience stores, 270=gas-station stores,
280=mini markets, 290=seasonally opened stores, 299=other stores.
Source: DELFI
No. obs. 14 209
27
Table 2: Descriptive statistics, store types (mean with standard deviations in parenthesis)
Store type # entrants Revenues Sales area No. obs.
Under construction 0.8 (0.45) 80 700 (114 510) 3 604 (3 506) 5
Hypermarket 0.04 (0.23) 250 000 (166 508) 3 878 (2 650) 794
Department store 0.09 (0.10) 65 425 (42 193) 1 229 (784) 408
Supermarket 0.01 (0.11) 94 294 (91 065) 2 192 (1 621) 2 113
Grocery store 0.07 (0.31) 243 606 (348 360) 4 215 (5 230) 5 953
Small supermarket 0.006 (0.08) 36 582 (25 386) 836 (484) 1 483
Small grocery store 0.03 (0.26) 111 871 (231 492) 2 081 (3 193) 8 792
Convenience store 0.14 (0.75) 28 689 (80147) 665 (1 582) 12 191
Gas-station store 0.23 (0.65) 22 790 (29 082) 498 (819) 11 907
Mini market 0.10 (0.79) 80 294 (166 313) 1 747 (2 618) 25 647
Season store 0.02 (0.15) 4 102 (4 637) 214 (269) 379
Other store 0.33 (0.48) 102 205 (146 131) 5 162 (3823) 39
28
Table 3: Market shares in the Swedish retail food market 1994-2002
(number of stores are reported in parenthesis)
owner/year 1994 1995 1996 1997 1998 1999 2000 2001 2002
ICA 44.2 43.9 43.6 44.0 44.7 44.1 43.4 43.6 43.9(2653) (2546) (2378) (2258) (2179) (2090) (2000) (1870) (1861)
COOP 24.9 25.2 25.2 23.9 23.5 22.5 22.3 22.4 22.4(1447) (1436) (1353) (1321) (1337) (1057) (1009) (924) (903)
Axel Johnson 5.5 5.2 5.1 4.6(522) (488) (437) (270)
D-group 17.0 17.2 17.3 18.0(856) (815) (785) (884)
Axfood 22.9 22.69 23.3 23.0 22.7(1110) (1069) (1027) (917) (903)
Bergendahls 1.9 1.8 1.9 2.1 2.0 2.1 2.3 2.8 2.9(87) (67) (60) (56) (60) (55) (55) (78) (83)
Others 6.6 6.7 6.8 7.4 7.1 8.4 8.7 8.2 8.1(2044) (2056) (2159) (2146) (2209) (2651) (2618) (2433) (2448)
Total # of stores 7589 7408 7172 6935 6895 6922 6709 6222 6198
29
Table 4. Estimation results
Variable (Parameter) Coefficients Marginal effects
REVAREA (β1) 1.6E-04∗∗∗ 1.05E-06
(2.47)
UNCERT (β2) -7.13E-09∗∗ -4.64E-11
(-2.24)
IRREVERS (β3) -2.32E-09 -1.51E-11
(-0.71)
UNC*IRR (β4) -1.41E-13∗ -9.19E-16
(-1.43)
CCR (β5) -2.88∗∗∗ -0.0187
(-2.83)
INCOME (δ1) 9.1E-03 5.9E-05
(0.97)
DUNIV (δ2) -0.95∗∗∗ 4.74E-03
(-2.73)
DCONS (δ3) 1.11∗∗∗ 9.81E-03
(7.19)
PPOWER (δ4) 1.6E-03∗∗ 1.04E-05
(1.87)
T 0.14∗∗ 9.29E-04
(1.79)
T 2 -0.022∗∗∗ 1.42E-04
(-3.40)
Log Likelihood -3195.76
Note: t-values (Robust -White) in parenthesis.∗∗∗significant at the 1 percent level,∗∗at the 5 percent level and∗at the 10 percent level.
30
Table 5. Estimation results store type indicators
Variable (Parameter) Coefficients Marginal effects
Intercept (α) -4.34∗∗∗ -
(-2.79)
Under construction (γ1) 2.92∗∗∗ 0.11
(6.16)
Department store (γ2) -2.22∗∗∗ -0.0061
(-2.40)
Supermarket (γ3) -0.74∗∗ -0.0036
(-2.06)
Grocery store (γ4) 1.29∗∗∗ 0.015
(5.05)
Small supermarket (γ5) -1.21∗∗∗ -0.0050
(2.62)
Small grocery store (γ6) 0.22 0.0015
(0.75)
Convenience store (γ7) 1.79∗∗∗ 0.025
(7.29)
Gas-station store (γ8) 2.72∗∗∗ 0.060
(11.14)
Mini market (γ9) 1.56∗∗∗ 0.018
(6.18)
Season store (γ10) -0.081 -0.00051
(-0.15)
Other store (γ11) 1.46∗∗∗ 0.021
(3.31)
Note: t-values (Robust -White) in parenthesis.∗∗∗significant at the 1 percent level,∗∗at the 5 percent level and∗at the 10 percent level.