Comparison and one-stop shopping after big-box retail entry: A spatial difference-in-difference analysis

Journal of Retailing and Consumer Services 40 (2018) 175–187

Contents lists available at ScienceDirect

Journal of Retailing and Consumer Services journal homepage: www.elsevier.com/locate/jretconser

Comparison and one-stop shopping after big-box retail entry: A spatial difference-in-difference analysis

MARK

⁎

Mengjie Hana, Oana Mihaescub, , Yujiao Lia, Niklas Rudholmc a b c

Dalarna University, Sweden HUI Research, Sweden Dalarna University and HUI Research, Sweden

A R T I C L E I N F O

A B S T R A C T

JEL classifications: D22 L11 L25 L26 P25

This paper empirically measures the potential spillover effects of big-box retail entry on the productivity of incumbent retailers in the entry regions, and investigates whether the effects differ depending on 1) if the entry is in a rural or urban area, and 2) if the incumbent retailers are within retail industries selling substitute or complement goods to those found in IKEA. To identify the IKEA-entry effect, a difference-in-difference model is suitable, but traditionally such estimators neglect the possibility that firms’ sales are determined by a process with spatially interactive responses. If ignored, these responses may cause biased estimates of the IKEA entry effect due to spatial heterogeneity of the treatment effect. One objective of this paper is thus to propose a spatial difference-in-difference estimator accounting for possible spatial spillover effects of IKEA entry. Particular emphasis is placed on the development of a suitable weight matrix accounting for the spatial links between firms, where we allow for local spatial interactions such that the outcome of observed units depends both on their own treatment as well as on the treatment of their neighbors. Our results show that for complementary goods retailers (or one-stop shopping retailers) in Haparanda and Kalmar, productivity increased by 35% and 18%, respectively, due to IKEA entry. No statistically significant effects were found for the entries in Karlstad and Gothenburg, indicating that it is mainly incumbents in smaller entry regions that benefit from IKEA entry. Also, for incumbent retailers selling substitute (or comparison shopping) goods no significant effects were found in any of the entry regions, indicating that it is mainly retailers selling complementary goods that benefit from IKEA entry. Finally, our results also show that ignoring the possibility of spatially correlated treatment effects in the regression models reduces the estimated impact of the IKEA entries in Haparanda and Kalmar on productivity in one-stop shopping retail firms with 3% and 0.1% points, respectively.

Keywords: Big-box entry Production functions Retail productivity Retail entry subsidies Spatial difference-in-difference

1. Introduction The question of how big-box retail entry affects the productivity of existing retail firms in the entry regions is paramount to local policymakers. Local governments are often ready to subsidize big-box retail entry under the justification that it will have significant positive spillover effects on the existing businesses in the region (Nilsson, 2015). There has been, however, little interest in empirically testing this argument in spite of the fact that if positive externalities on productivity are absent, the use of taxpayer money for such subsidies cannot be justified on economic efficiency grounds (Greenstone et al., 2010). Previous studies on the effects of big-box retail entry on surrounding businesses have mainly investigated the impact on retail revenues or retail employment, and have been mainly based on the entry of WalMart stores in the USA. The results of these studies diverge, with some

⁎

finding positive (Davidson and Rummel, 2000; Artz and Stone, 2012) and others negative (e.g., Merriman et al., 2012) impacts on retail revenues. Furthermore, while Basker (2005) and Hicks (2007) both found that Wal-Mart entry increase retail employment by approximately 100 jobs in the entry regions in the year of entry, others have found that big-box entry negatively affects retail employment (Jones and Doucet, 2000; Hicks, 2008; Neumark et al., 2008). Outside Wal-Mart and the US market, Jones and Doucet (2000) and Hernandez (2003) studied the Canadian market and big-box entry in general, while Daunfeldt et al. (2016, 2017) investigated the impact of IKEA entry on revenues and employment in Swedish municipalities. Daunfeldt et al. reported an increase in durable goods retail revenues by an average of 20% and in durable goods retail employment by an average of 17% in Swedish municipalities where IKEA chose to enter during the 2004–2007 period (Daunfeldt et al., 2017). They also

Corresponding author. E-mail addresses: [email protected] (M. Han), [email protected] (O. Mihaescu), [email protected] (Y. Li), [email protected] (N. Rudholm).

http://dx.doi.org/10.1016/j.jretconser.2017.10.003 Received in revised form 1 September 2017; Accepted 10 October 2017 0969-6989/ © 2017 Elsevier Ltd. All rights reserved.

Journal of Retailing and Consumer Services 40 (2018) 175–187

M. Han et al.

outcome of one incumbent retailer depends both on their own treatment as well as on the treatment of their neighboring incumbent retailers. Our results show that for complementary and non-related goods retailers (or one-stop shopping retailers) in Haparanda and Kalmar, productivity increased by 35% and 18%, respectively, due to IKEA entry. No statistically significant effects were found for the entries in Karlstad and Gothenburg, indicating that it is mainly incumbents in smaller entry regions that benefit from IKEA entry. Also, for incumbent retailers selling substitute goods no significant effects were found in any of the entry regions, indicating that it is mainly retailers selling complementary goods that benefit from IKEA entry. Finally, our results also show that ignoring the possibility of spatially correlated treatment effects in the regression models reduces the estimated impact of the IKEA entries in Haparanda and Kalmar on productivity with 3% and 0.1% points, respectively. The article is organized as follows: Section 2 presents the theoretical foundations for why big-box entry should affect incumbent retailer productivity and the importance of considering industry differences in retailing; Section 3 presents our identification strategy and empirical model; Section 4 presents the data, descriptive statistics and estimation results; and Section 5 summarizes and discusses the findings of the study.

reported that these effects decrease with distance from IKEA (Daunfeldt et al., 2016). Most studies on the effects of big-box retail entry on productivity have investigated the effect on regional level productivity, while those focusing on firm-level productivity are scarce. In 2007, Basker reported that much of the productivity growth in the US general merchandise sector was driven by the growth of Wal-Mart, while Maican and Orth (2012a) found that big-box entry in the Swedish retail food sector increased the productivity of incumbent firms in the entry regions. Additionally, Maican and Orth (2012b) also showed that more liberal entry regulations increased productivity in the Swedish retail sector, the effect being larger in smaller markets in Sweden. The inverse relationship between the size of the local retail market and the effect of big-box entry on productivity was also emphasized in Håkansson et al. (2016). Their results indicated that big-box entry increased the productivity of incumbent firms in rural entry regions where the IKEA entry was large relative to the local retail market, while no productivity spillover effects could be found in the case of the urban IKEA entry included in the analysis. From theory, we know that the impact of big-box entry on incumbent retailer productivity might differ depending on what type of products the incumbents are selling. Increases in the productivity of incumbent firms due to big-box retail entry is a result of both supply- and demand-side spillovers, and while it has been argued that there is no difference in how supply-side spillovers impact existing firms within the same industry (McCann, 2001), demand-side spillovers may have different effects on productivity depending on whether the products sold by the incumbents are substitutes or complements to those sold by the big-box retailer. Co-location of retailers selling substitutes establishes the basis for comparison shopping, while co-location of retailers selling complements establishes the basis for one-stop shopping (Håkansson et al., 2016). But while both comparison and one-stop shopping have the power to increase the customer base and thus sales for the firms in the retail cluster, comparison shopping also increases competition for the incumbent retailers in the cluster (McCann, 2001). The purpose of this study is to investigate a question overlooked in previous studies: do the effects of big-box retail entry differ depending on if the incumbent retailers are within retail industries selling substitutes or complements to goods found in IKEA? Additionally, we will also research whether the impact differs depending on if the entry is in a rural or urban area. Methodologically, the estimation of how IKEA entry might affect the productivity of incumbent retailers in the entry municipalities is not an easy task. Previous studies of how IKEA affects incumbent retailers in local economies (Daunfeldt et al., 2016; Håkansson et al., 2016) use traditional difference-in-difference estimations, after first having tried to select control group municipalities with similar characteristics to the entry municipalities under study. We follow these studies in that we first select control group municipalities we believe to be similar to the entry regions in terms of the determinants of incumbent retailers’ productivity development in the absence of entry. As in Håkansson et al. (2016), for the rural IKEA entries in the period 2006–2007, we use the municipalities deemed suitable for entry by IKEA in the period 2013–2016 as controls, while for the urban entry in Gothenburg, Stockholm is used as the control region. As in Daunfeldt et al. (2016) and Håkansson et al. (2016), we use a difference-in-difference model to identify the IKEA-entry effect. The estimators used in these papers neglect however the possibility that firms’ sales are determined by a process with spatially interactive responses, which, if ignored, may cause biased estimates of the IKEA entry effect due to spatial heterogeneity of the treatment effect. One additional purpose of this paper is thus to propose a spatial differencein-difference estimator that accounts for possible spatial spillover effects of IKEA entry. Particular emphasis is placed on the development of a suitable spatial weight matrix accounting for the spatial links between firms, where we allow for local spatial interactions such that the

2. Theoretical framework and previous studies Big-box retail entry can have both direct and indirect effects on productivity in the entry regions. IKEA entry will have a direct effect on average productivity if IKEA itself is more productive than the average for the already existing retailers in the entry area, and it will have indirect effects if there are also productivity spillovers to incumbent retailers in the entry area. Such spillovers can affect the supply and demand of the incumbent retailers. As mentioned in the introduction, most studies on the effects of bigbox retail entry have investigated the effect on the regional level, also including the output of the big-box retailer in the data. As such, these studies focus on the aggregate of both direct and indirect effects of bigbox entry on productivity (and other outcome variables such as sales or employment). For the case of IKEA entry in Sweden, the studies of Daunfeldt et al. (2017) and Rudholm et al. (2017), take this approach while using different empirical methods to measure the impact of IKEA entry on sales, employment and productivity, respectively. In this paper, we instead study the impact of IKEA entry on incumbent retailers in the entry areas. As such, our focus is on the indirect, or spillover, effects of IKEA entry on the productivity of incumbent retailers in the IKEA entry areas. We start by discussing supply side spillovers and then turn to demand side spillovers of big-box entry. Ever since the early contributions of Marshall (1890), Hotelling (1929), and Weber (1929), economists and economic geographers alike have analyzed supply-side spillovers due to firm co-location. According to these theories, firm co-location decreases input costs, facilitates labor matching, and creates knowledge spillovers (McCann, 2001; O’Sullivan, 2003). More recent literature has further associated knowledge spillovers to increases in productivity (Lucas, 1988; Grossman and Helpman, 1991; Glaeser, 1999) and pointed out that inter-firm learning might be affected not only by geographical, but by other types of proximity as well (e.g., cognitive, organizational, social, and institutional) (Boschma, 2005). Moreover, these theoretical contributions also find that there must be some optimal level of specialization in the type of firms that co-locate for knowledge spillovers to occur. An increase in retail density through co-location by very similar retailers would yield a low level of knowledge spillovers (as firms are very similar one firms’ knowledge will already be known by other, similar firms), while also creating a high level of competition for the co-located firms. If co-located firms are instead too dissimilar, for example from totally different industries, the knowledge of one firm would contain little or no value 176

Journal of Retailing and Consumer Services 40 (2018) 175–187

M. Han et al.

that a retail firm located in a region that receives a big-box retail establishment cannot also be observed in the counterfactual state of not receiving that establishment. Also, as Greenstone et al. (2010) points out, firms choose locations in an effort to maximize profits and the chosen entry regions will in all likelihood differ substantially from randomly selected regions. If we are to correctly measure how big-box entry affects incumbent retailers in the entry regions, we need to identify control regions that are identical to the entry regions in terms of the determinants of incumbent retailers’ productivity in the absence of entry. In two studies of how entry by IKEA affects revenues and employment in the entry municipalities in Sweden (Daunfeldt et al., 2016, 2017), control municipalities were identified using logit estimations to detect municipalities where IKEA had not entered, but which had a similar probability of entry based on observables. However, as pointed out by Greenstone et al. (2010), such a strategy implies that the entry decision can be correctly modeled by the observable characteristics of the municipalities, while in most cases many important characteristics are generally unknown and unobserved by the researcher. One obvious example is the subsidies given to IKEA for entering certain municipalities, which in some cases become known for the entry municipalities after entry, but which remain unknown for the alternative entry sites (i.e., potential control municipalities). Following Håkansson et al. (2016), we choose the next round of IKEA entries, completed during the 2013–2016 period, as control municipalities for three of the four IKEA entries in Sweden during the 2006–2007 period, i.e., Kalmar (2006), Haparanda (2006), and Karlstad (2007). The municipalities in the 2013–2016 round of entries, i.e., Borlänge, Uddevalla, and Umeå, are similar to the entry municipalities in several observables.1 Most importantly, however, is that both the entry- and control-group municipalities were deemed suitable entry locations by IKEA itself within a 10-year period, indicating that these municipalities should be similar in terms of both observable and unobservable determinants of IKEA entry. Daunfeldt et al., (2016, 2017) have also demonstrated that for the Swedish entries under study, the regional spillover effects regarding revenues and employment of IKEA entry were limited to the entry municipalities, so we focus our study on the potential spillover effects within the above-mentioned entry municipalities. The fourth IKEA entry under study took place in the urban metropolitan area of Gothenburg in 2004, and is the second IKEA store in the region. Gothenburg is Sweden's second largest city; its metropolitan area consists of 13 municipalities and has approximately 1 million inhabitants. As explained in Håkansson et al. (2016), only two other Swedish metropolitan areas can reasonably be considered potential controls for Gothenburg: Stockholm and Malmö.2 However, Malmö is a special case with its close ties to another metropolitan area, Copenhagen. The Stockholm metropolitan area consists of 26 municipalities, with somewhat more than 2 million inhabitants, and as shown by Håkansson et al. (2016) is more suited as a control for Gothenburg than Malmö. Most importantly, Stockholm has two IKEA stores, one in the northern and another in the southern part of the metropolitan core, meaning that both Stockholm and Gothenburg have been deemed suitable as two-store metropolitan areas by IKEA. In Fig. 1, the entries studied here (i.e., Gothenburg metropolitan area and Haparanda, Kalmar, and Karlstad municipalities) are marked in stripes while the control regions (i.e., Stockholm metropolitan area and Borlänge, Uddevalla, and Umeå municipalities) are marked in light yellow. The other previous IKEA-entry municipalities are marked in dark green.

for the others. For the retailing industry, demand-side spillovers are also important. While supply-side spillovers have been argued to have the same effects for retailers within the same location and industry (McCann, 2001), demand-side spillovers clearly impact the co-located retailers differently depending on whether the products sold are substitutes or complements. Co-location of retailers selling substitutes establishes the basis for comparison shopping, while co-location of retailers selling complements establishes the basis for one-stop shopping (Håkansson et al., 2016). In the case of comparison shopping, co-location helps firms to increase their productivity by taking advantage of the customer base attracted by the marketing and reputation of their rivals and by the possibility of minimizing consumer uncertainty. In the case of one-stop shopping, productivity premiums are obtained through an increase in the customer base attracted to a retail trade area by the prospect of minimizing shopping costs for the basket of desired products (Brown, 1989; Chung and Kalnins, 2001). While one-stop shopping is associated with increases in the productivity of the clustered firms, comparison shopping may create both benefits and costs for the affected firms. As already discussed, co-location of firms selling similar products is likely to increase the customer base for each of the firms in the cluster. However, co-location of similar firms also increases the level of competition for these firms. This may provoke a variety of reactions from the incumbent firms: some of them will find a way to respond by increasing their productivity and thus survive; but other firms, who lack an appropriate reaction, will eventually be displaced by more productive retailers. Big-box retailers are of special importance in this context since local governments (at least in Sweden) often subsidize big-box entry in the belief that it will positively affect the productivity of existing retail firms in the entry region (Nilsson, 2015). However, studies researching the relationship between big-box entry and incumbents’ productivity are scarce, and mainly investigating effects of Wal-Mart entry in the US market. Many of them are also focused on studying effects on aggregate productivity at the regional level. These studies show that entry by the highly productive big-box retailers is enough, by itself, to increase productivity in the retail sector in the entry regions. According to Basker (2007), Wal-Mart's real value added per worker is 40% higher than that of other general merchandise retailers, and its productivity increased by 55% over the 20-year 1982–2002 period. Entry by big-box retailers may also create significant competitive pressure in the entry regions, because these large retail establishments are highly productive due to economies of scale and due to innovations in logistics, distribution, inventory control, computerization and use of specialized software, and communications (Foster et al., 2006; Basker, 2007). Foster et al. (2006) reported that labor productivity growth in the US retail sector in the 1990s was mainly due to entry by national chains in local markets displacing less productive local retailers, while Jia (2008) reported that Wal-Mart entry caused 50–70% of the net exit of small discount retailers in the US market, and that the exiting establishments were 25% less productive than the surviving incumbents. For the Swedish retail market, Maican and Orth (2012) found that bigbox entry forced low-productivity stores to exit, and that surviving stores experienced productivity increases of approximately 3%. Finally, there seems to also be a relationship between the size of the new entry and market size in the entry area. Maican and Orth (2012b) showed that more liberal entry regulations had larger effects on retail in smaller markets in Sweden. The inverse relationship between the size of the local retail market and the effect on productivity was also emphasized in Håkansson et al. (2016). 3. Method

1

See Håkansson et al. (2016) for examples. As noted by Håkansson et al. (2016), municipalities in Swedish metropolitan areas are geographically very small and not individually representative of the market of new IKEA stores. As such, we use metropolitan areas for comparison when studying the urban IKEA entry in Gothenburg. 2

3.1. Identification strategy The fundamental identification problem that we need to solve is 177

Journal of Retailing and Consumer Services 40 (2018) 175–187

M. Han et al.

Fig. 1. IKEA-entry, control, and previous IKEA-entry municipalities in Sweden.

3.2. A difference-in-difference estimation model with spatially correlated treatments

(2010) and Håkansson et al. (2016), using a production-function difference-in-difference regression model to measure the effect of a new IKEA store on the productivity of incumbent one-stop or comparison shopping retail firms located in the entry municipalities. In our most

After selecting suitable control regions, we follow Greenstone et al. 178

Journal of Retailing and Consumer Services 40 (2018) 175–187

M. Han et al.

difference-in-difference regression if the outcome of one treatment unit is correlated to the neighboring units, i.e., if IKEA entry does not only have a direct effect on the incumbent firms in the entry regions, but also indirect effects. From the demand side, as pointed in Section 2, retailers selling complementary products that are closely located to each other promotes one-stop shopping, making the sales of one retailer dependent on the sales of closely located retailers as well. Similarly, if selling substitute goods, retailer co-location will increase competition, which also might facilitate increased productivity. From the supply side, retailers can also benefit from co-located retailers in terms of knowledge spillovers reducing costs and increasing productivity. For these reasons, the possibility of spatial interaction effects needs to be considered in the difference-in-difference model. In this paper, the effect of spatial correlation is not assumed to be constant between two retailers in the entry municipalities. Instead, firms located closer to each other are assumed to affect each other more than those located within some distance from each other. To model this relationship we use a distance decay function represented by the exponential correlation function:

general specification, the retailers are assumed to use a production technology that can be described by the transcendental logarithmic (translog) production function developed by Christensen et al. (1971). This functional form can be seen as a second-order Taylor series approximation with a remainder term of an arbitrary production function, and can be written as follows:

lnQi, t = β1 ln Li, t − 1 + β2 ln Ki, t − 1 + β3 ln2 Li, t − 1 + β4 ln2 Ki, t − 1 + β5 ln Li, t − 1 ln Ki, t − 1 + Ri, t ,

(1)

where lnQi,t is a measure of output in retailing and Li,t –1 and Ki,t –1 are measures of the labor and capital inputs in retailing, respectively, both lagged one period to alleviate a potential endogeneity problem. As discussed in Håkansson et al. (2016), the revenues of the studied retail firms must be discounted using a price index to arrive at a relevant measure of retail output. As such, following a suggestion of the OECD regarding how to measure output in multiple product firms (OECD, 2001), output (Qi,t), is measured for each firm and year and is defined as revenues of firm i in year t discounted by the Swedish consumer price index (CPI). The log transformation of output in retailing (Qi,t) has the additional benefit of making the parameter estimate related to the effect of IKEA entry on incumbent-firm productivity from our differencein-difference model interpretable in percentage terms after using the formula 100 × [exp(treatment effect) – 1]. Labor (Li,t–1) is measured as the total wages paid by firm i at time t – 1, while capital (Ki,t –1) is measured as the sum of payments to capital owners, i.e., the sum of interest payments, dividends to owners, and other financial expenses for firm i at time t – 1. Both variables have been log transformed. Finally, Ri,t is the remainder term of the Taylor series approximation, which in most empirical work is assumed to contain a constant and a random error term, making Eq. (1) a standard ordinary least squares regression model to be estimated. However, as we are interested in measuring how the entry of an IKEA store in a region affects the productivity of incumbent firms, i.e., whether IKEA entry causes a shift in the production function of the affected firms, holding the levels of the capital and labor inputs constant, our remainder term needs to take this into account. As such, we propose the following remainder term:

Ri, t = β0 + β6 TRi + β7 TPt + β8 (TPt *TRi ) + εi, t ,

d corr (d ) = exp ⎧ − ⎫ , ⎨ ⎬ φ ⎩ ⎭ where corr (d ) is the correlation function of the distance, d, and where φ is the range parameter specified by the empirical data-driven semivariogram approach. Suppose that we have two observations of retail output lnQi1, t and lnQi2, t from firms having different locations, and that the distance between them is d . Our approach then assumes that the variance between each spatially correlated observation pair, γ (d) = Var (lnQi1, t − lnQi2, t ) , is a function of the distance such that γ (d ) increases when d increases. However, when d for any pair of observations is larger than φ , γ (d ) never increases. Thus, fitting a theoretical line to the empirical γ (d ) is the most efficient way of finding φ .3The empirical findings for the four municipalities are shown in Fig. 2. We can easily specify φ for each municipality: Gothenburg 3.3 km, Haparanda 1.4 km, Kalmar 1.9 km and Karlstad 4.3 km, thus confirming the findings of Daunfeldt et al., (2016, 2017) that spillover effects of IKEA entry in Sweden are confined to the entry municipalities, or in the case of Gothenburg, to the metropolitan area. Furthermore, corr (d ) is decreasing to 0 when the distance goes to infinity and takes the value 1 when d = 0 . Following Delgado and Florax (2015), the correlation function forms the spatial lag operator Ws (i.e., Ws is the spatial weight matrix), and it is a block-diagonal rowstandard matrix. There are t = 10 (2001–2010) non-zero blocks in which the correlation between each retail firm and IKEA is a function of their Euclidean distance. Therefore, the spatial interaction is defined as TRL*TP = (I + ρWs ) TR*TP , where index L indicates that TR now has a spatial lag defined by Ws . ρ ≠ 0 is an unknown parameter which specifies how strong the correlation between co-located retailers is, and it is estimated by an iterative approach to the maximum likelihood function (Bailey and Gatrell, 1995). Thus, the spatial translog production function difference-in-difference model4 becomes:

(2)

where β0 is a constant, TRi is an indicator variable equal to one for firms located in the IKEA-entry region, and zero otherwise, and where TPt is an indicator variable equal to one for the treatment period (i.e., years after IKEA entry), and zero otherwise. Our key variable of interest is the interaction between TPt and TRi, as this will provide an estimate of the treatment effect, i.e., of how the output of incumbent retailers in the entry region after IKEA entry compares with their own output before entry, and with the output of control group retailers throughout the study period, holding the levels of inputs (i.e., labor and capital) constant. This type of difference-in-difference estimator is one of the tools most frequently used in applied economics research to evaluate the effects of public interventions and other treatments of interest on relevant outcome variables (Abadie, 2005). Combining Eqs. (1) and (2) gives:

lnQi, t = β0 + β1 ln Li, t − 1 + β2 ln Ki, t − 1 + β3 ln2 Li, t − 1 + β4 ln2 Ki, t − 1

ln Qi, t = β0 + β1 ln Li, t − 1 + β2 ln Ki, t − 1 + β3 ln2 Li, t − 1 + β4 ln2 Ki, t − 1

+ β5 ln Li, t − 1 ln Ki, t − 1 + β6 TRi + β7 TPt + β8 (I + ρ Ws )(TPt *TRi) + εi, t .

+ β5 ln Li, t − 1 ln Ki, t − 1 + β6 TRi + β7 TPt + β8 (TPt *TRi ) + εi, t .

(4) (3)

We expect that firms in the IKEA-entry regions selling complementary products become more productive in that they increase their output for given levels of labor and capital after IKEA entry. If agglomeration economies caused by the new IKEA create, for example,

Eq. (3) is a translog production function difference-in-difference model in which β8 estimates the conditional average treatment effect. The ordinary least square estimation for β8 indicates a significant IKEA entry effect on incumbent retailer productivity when we reject the hypothesis that β8 = 0 (Delgado and Florax, 2015). One assumption for Eq. (3) to measure the IKEA entry effect correctly is the stable unit treatment value assumption (SUTVA) (Rubin, 1978, 1990). Under SUTVA, the treatment of one unit is unrelated to the treatment of other units. However, a violation of SUTVA occurs in the conventional

3 See Cressie and Wikle (2011) and Han et al. (2016) for details regarding the estimation method. 4 For comparison, we will also present results from two other specifications, a CobbDouglas production function difference-in-difference model and a traditional, non-production function difference-in-difference model. Se section 4 and Appendix 1.

179

Journal of Retailing and Consumer Services 40 (2018) 175–187

M. Han et al.

Fig. 2. Range parameter, φ , in meters.

therefore be interpreted as a general equilibrium reduced form effect that combines the impact of the IKEA store itself, and all other associated changes in the retail environment due to IKEA entry.

knowledge spillovers that make labor more productive, or if there are spillovers from the marketing and reputations of other firms in the entry region, we would expect a positive outcome, especially since these firms do not directly compete with IKEA. For the comparison shopping retailers, i.e., the ones selling substitute goods to IKEA, the above would still hold, but for these firms co-location with IKEA also intensifies competition, and could do so to the extent that the positive agglomeration effects are completely counteracted. This means that, if the impact of positive spillovers and increased competition are of approximately equal size, we could observe a zero productivity impact for comparison shopping retailers due to IKEA entry. As pointed out by Greenstone et al. (2010), who use an identification strategy and empirical method similar to ours, it should be noted that it is not only the spillover effects of the IKEA store itself that are measured; rather, our measure also captures the impact of all other changes associated with IKEA entry that occur in the retail environment of the entry municipality. One such change is that more retailers will likely want to establish stores near the new IKEA, increasing competition for the incumbent retailers located in the same area, making some retailers leave the market and others enter.5 The results should

4. Data, descriptive statistics and estimation results To investigate the effect of a new IKEA store on the productivity of incumbent retailers in IKEA-entry municipalities, we use data on all limited liability companies in the retail trade industry that were active in the entry- and control municipalities at some point between 2001 and 2010. The data were collected from PAR, a Swedish consulting firm that compiles this information from the Swedish Patent and Registration Office (PRV), and include all variables registered in the annual reports. The determination of if a retail firm belongs to the comparison- or one-stop shopping category is made based on the firms reported main activity within the Swedish SNI-2002 code system.6 Descriptive statistics for the variables included in the empirical model are presented in Table 1 (one-stop shopping) and Table 2 (comparison shopping), and the means and standard deviations of output, labor, and capital indicate that the data from the entry and control municipalities are quite similar, not only for retail output but also for our measures of capital and labor. The results from estimating Eq. (4) and the two other, less general,

5 In order to give the reader an idea about retail dynamics in the entry municipalities before and after IKEA entry we present descriptive statistics of entry, exit and the number of retail firms in Appendix 1. These statistics do not show any dramatic changes in entry, exit or the number of firms at the time of IKEA entry.

6

180

Similar to the European NACE-code system.

Journal of Retailing and Consumer Services 40 (2018) 175–187

M. Han et al.

Table 1 Means and standard deviations, variable description and data source, one-stop shopping retailers. Variable

Haparanda

Kalmar

Karlstad

Gothenburg

Urban control

Rural control

Variable description

Data source

lnQi,t

15.90 (0.91)

15.44 (1.09)

15.47 (1.17)

15.37 (1.16)

15.38 (1.28)

15.39 (1.28)

Revenues of firm i in year t discounted by CPI

PAR/Statistics Sweden/ own calculations

lnKi,t

– 1

4.08 (1.62)

4.00 (1.38)

4.07 (1.57)

3.58 (1.66)

3.63 (1.80)

3.84 (1.70)

Sum of interest payments, dividends to owners, and other financial expenses of firm i in year t

PAR/own calculations

lnLi,t

– 1

6.51 (0.75)

6.50 (1.01)

6.49 (1.13)

6.41 (1.13)

6.38 (1.24)

6.41 (1.17)

Total wages paid by firm i in year t

PAR/own calculations

ln2Ki,t

−1

19.25 (14.64)

17.93 (11.03)

18.98 (13.09)

15.53 (12.75)

16.46 (14.79)

17.61 (14.08)

lnKi,t

– 1

squared

PAR/own calculations

ln2Li,t

−1

42.95 (9.92)

43.31 (12.55)

43.46 (14.24)

42.33 (14.45)

42.30 (15.90)

42.41 (14.68)

lnLi,t

– 1

squared.

PAR/own calculations

26.82 (11.69)

26.53 (11.46)

27.22 (13.43)

23.74 (13.73)

24.20 (15.31)

25.56 (14.76)

lnLi,t

– 1

multiplied by lnKi,t

TPt

0.46 (0.50)

0.51 (0.50)

0.50 (0.50)

0.50 (0.50)

0.50 (0.50)

0.48 (0.50)

Indicator variable equal to one for treatment period years

Own calculations

TRi

1 (0)

1 (0)

1 (0)

1 (0)

0 (0)

0 (0)

Indicator variable equal to one for retailers located in treatment municipalities

Own calculations

TPt × TRi

0.46 (0.50)

0.51 (0.50)

0.50 (0.50)

0.50 (0.50)

0 (0)

0 (0)

Indicator variable equal to one for retailers located in treatment municipalities in treatment years

Own calculations

lnLi,t-1lnKi,t −1

PAR/own calculations

– 1

Table 2 Means and standard deviations, variable description and data source, comparison shopping retailers. Variable

Haparanda

Kalmar

Karlstad

lnQi,t

N.A. (N.A.) N.A. (N.A.) N.A. (N.A.) N.A. (N.A.) N.A. (N.A.) N.A. (N.A.) N.A. (N.A.) N.A. (N.A.) N.A. (N.A.)

15.45 (1.25) 4.00 (1.35) 6.38 (1.31) 17.83 (10.44) 42.38 (14.88) 26.42 (12.28) 0.41 (0.49) 1 (0) 0.41 (0.49)

15.18 (1.12) 4.20 (1.72) 6.18 (1.18) 20.60 (19.49) 39.57 (14.43) 27.08 (16.30) 0.45 (0.50) 1 (0) 0.45 (0.50)

lnKi,t

– 1

lnLi,t

– 1

2

ln Ki,t

− 1

ln2Li,t

− 1

lnLi,t TPt

− 1lnKi,t

− 1

TRi TPt × TRi

Gothenburg

15.25 (1.04) 3.81 (1.51) 6.28 (1.11) 16.80 (11.54) 40.73 (13.04) 24.69 (12.31) 0.47 (0.50) 1 (0) 0.47 (0.50)

Urban control

Rural control

Variable description

Data source

15.22 (1.07) 3.58 (1.48) 6.29 (1.08) 14.99 (11.12) 40.69 (13.21) 23.05 (11.80) 0.47 (0.50) 0 (0) 0 (0)

15.50 (0.95) 4.31 (1.29) 6.54 (1.01) 20.27 (10.19) 43.85 (11.14) 28.74 (10.61) 0.45 (0.50) 0 (0) 0 (0)

Revenues of firm i in year t discounted by CPI

PAR/Statistics Sweden/ own calculations PAR/own calculations

Sum of interest payments, dividends to owners, and other financial expenses of firm i in year t Total wages paid by firm i in year t

PAR/own calculations

lnKi,t

– 1

squared

PAR/own calculations

lnLi,t

– 1

squared.

PAR/own calculations

lnLi,t

– 1

multiplied by lnKi,t

PAR/own calculations

– 1

Indicator variable equal to one for treatment period years Indicator variable equal to one for retailers located in treatment municipalities Indicator variable equal to one for retailers located in treatment municipalities in treatment years

Own calculations Own calculations Own calculations

Table 3 IKEA entry effect on productivity in one-stop shopping incumbent retailers. Haparanda

Translog model Cobb–Douglas model Difference-in-difference model

Kalmar

Karlstad

Gothenburg

Spatial

Standard

Spatial

Standard

Spatial

Standard

Spatial

Standard

34.99%* 27.89% 27.00%

32.05%* 25.48% 24.60%

17.70%** 23.99%*** 37.30%**

17.59%*** 18.29%** 14.34%

1.21% 2.02% −9.06%

1.21% 2.22% −6.39%

−1.39% −1.39% −1.09%

−3.25% −2.18% 2.74%

*** significant at the 1% level, ** significant at the 5% level, * significant at the 10% level. Estimation results of TPt × TRi, dependent variable lnQi,t; results in percentage terms, i.e., calculated using the formula 100 × [exp(β8) – 1].

The results show that for Haparanda, IKEA entry increased productivity in incumbent retailers with 35% (significant at the 10% level) in the translog model, while the corresponding results when estimating the Cobb-Douglas production function was 28% and no statistically

models can be found in Appendix 1 for both one-stop and comparison shopping. The results regarding the estimate of the IKEA entry effect on productivity, for all estimated models, for one-stop shopping incumbent retailers, are presented in Table 3. 181

Journal of Retailing and Consumer Services 40 (2018) 175–187

M. Han et al.

Table 4 IKEA entry effect on productivity in comparison shopping incumbent retailers. Haparanda

Translog model Cobb–Douglas model Difference-in-difference model

Kalmar

Karlstad

Gothenburg

Spatial

Standard

Spatial

Standard

Spatial

Standard

Spatial

Standard

n.a. n.a. n.a.

n.a. n.a. n.a.

−19.10% 6.93% 3.46%

−10.33% 5.34% 13.31%

−11.40% −7.23% 39.65%

−8.42% −4.97% 32.98%

9.63% 13.66%* −6.20%

−5.07% −4.50% 11.40%

*** significant at the 1% level, ** significant at the 5% level, * significant at the 10% level. Estimation results of TPt × TRi, dependent variable lnQi,t; results in percentage terms, i.e. calculated using the formula 100 × [exp(β8) – 1].

papers have also ignored the possibility that the impact of big-box entry on incumbent retailers might be spatially correlated. The purpose of this paper has thus been twofold. First, we divide the available data into firms selling complementary products and those selling substitute goods to IKEA, the big-box entry under study in this paper, to investigate if and how the impact on productivity in incumbent firms differ. Secondly, we allow for the possibility of spatially correlated treatment effects. The results show that big-box retail entry has positive effects on productivity, but only if the entry occurs in a rural area, and if the incumbent retailers are selling complementary goods rather than substitute goods. The rural entry of IKEA in Haparanda increased productivity in incumbent one-stop shopping retailers by 35%, while the entry in Kalmar increased productivity by 18%. For the larger entry municipalities of Karlstad and Gothenburg, no significant effects were found for one-stop shopping retailers, and for comparison shopping retailers, no statistically significant effects were found in any of the estimated models. The finding that an IKEA entry has a larger impact on incumbent retailers in smaller markets is not that surprising bearing in mind how large an IKEA entry is compared to existing retail in the smaller entry municipalities. Although IKEA never releases any local statistics regarding sales, an IKEA store has been estimated to have a yearly turnover of between 700 and 1000 million SEK. This can then be related to the retail market in Haparanda, where total yearly retail turnover was approximately 440 million SEK before IKEA entered the market. For Gothenburg, total retail turnover was instead approximately 24 500 million SEK in the year before IKEA entry.8 As such, one might expect a larger impact on existing retailers in Haparanda than in Gothenburg. The findings in this paper are of importance for incumbent retailers in potential future big-box entry municipalities in Sweden, as how these incumbents will be affected depends largely on the type of product they sell. While productivity levels of incumbents that are in direct competition with the potential big-box retail entrant are not affected by its entry, retailers selling complementary products will, at least on average, be positively affected. The results could also be of importance for local governments when deciding whether to support IKEA entry. The local government may have less reason to subsidize IKEA entry in regions with many retailers selling similar products with those found in IKEA, since no significant spillovers are expected. It may make sense, however, to support entry in regions where the retailing cluster consists mostly of retailers selling complementary products to those found in the IKEA stores.

significant effect was found for the traditional difference-in-difference estimation. For Kalmar, the increase in productivity was 18% (significant at the 5% level), 24% and 37% (significant at the 1% and 5% levels) in the translog, Cobb-Douglas and difference-in-difference models, respectively. For the larger entry municipalities of Karlstad and Gothenburg, no statistically significant effects were found in any of the estimated models. The estimation results for how IKEA entry affects productivity for comparison (substitute) product incumbent retailers is presented in Table 4. For Haparanda, there are no comparison shopping retailers in the annual report data7 and our main model shows no statistically significant effects of IKEA entry on productivity for the other three entry regions. These results can be compared to those of Håkansson et al. (2016), who also report that IKEA entry had a statistically significant impact on the productivity of incumbent retailers only in the smaller entry municipalities of Haparanda and Kalmar. However, our results also show the importance of taking into account what type of products that the incumbent retailers are selling when studying potential spillover effects of big-box retail entry as we find positive spillover effects only for incumbent retailers selling complements goods to the products sold by the big-box retailer. A final result is that ignoring the possibility of spatially correlated treatment effects reduces both the estimated effect and, in some cases, also the statistical significance of the point estimate. For Haparanda, the reduction in the point estimate is approximately 3% points in the translog model and the estimated effect is no longer statistically significant at conventional levels. For Kalmar, the estimated effect decreases by some 0.1%, while remaining statistically significant at the 5% level as in the estimations presented above. This shows the importance of also considering that when IKEA enters, the effect of such entry on the incumbent retailers might be spatially correlated. Otherwise, estimates of the entry effect might be biased. 5. Summary and discussion Traditionally, research investigating how big-box retail entry affects incumbent retailers have in most cases ignored that the effects might be different for retailers selling different types of products. For example, while the impact of big-box retail entry on incumbent retailers selling complementary products should be positive, the impact on retailers selling substitute products in direct competition with the big-box retailer might not. From a methodological point of view, most previous 7 There are a few other stores selling furniture and other substitute goods to IKEA in Haparanda. However, either these belong to some national chain and are therefore not in the dataset, or only a minor part of total sales within these firms is from retailing, thus making their annual reports belong to other industries within the SNI 2002 code system. It should be also noted that all firms not belonging to a national chain, selling substitute goods to IKEA and based in Haparanda (Table A1.1 in the Appendix) are located outside the 1.4 km radius mentioned in Fig. 2 and thus solely included in the control group.

8 Data over total retail turnover for Haparanda and Gothenburg was collected from Trade in Sweden (Handeln i Sverige), which is a free database that includes information on a variety of retail related variables, e.g., revenues, employment, a sales index, and their development over time. The database is maintained by HUI Research and can be accessed at http://www.handelnisverige.se/ (in Swedish).

182

Journal of Retailing and Consumer Services 40 (2018) 175–187

M. Han et al.

Acknowledgments

is gratefully acknowledged. The authors would also like to thank participants in seminars held at Mälardalen University, September 12, 2016 and Dalarna University, April 7, 2017 for valuable comments and suggestions.

Research funding from the Swedish Retail and Wholesale Development Council (Handelns Utvecklingsråd, grant number 2015:4) Appendix 1. Entry and exit of firms

Tables A1.1–A1.3 below present descriptive statistics on the entry, exit and number of retail firms for comparison and one-stop shopping in the entry municipalities for the period under study. Numbers for the years following IKEA entry in each municipality is marked in bold. It should be noted that the dataset at hand is not ideally suited for identifying entry and exit. The only way of identifying entry and exit is by recording when a specific organization number for a retail firm appears in or leaves the dataset. There can, however, be a number of reasons for this, other than a formal entry or exit of a retail firm. If, for example, the firm is sold, it is often the case that the firm is registered with a new organization number as there are certain benefits of being a “new firm” in Sweden. This would then create one exit and one entry as we measure it, but in reality this has only been an administrative change. Note also that since we have quite large numbers of entries and exits, while the total number of firms do not change that much, this might be quite common in our dataset. As such, the numbers regarding entry and exit should be interpreted with caution. Looking at the total number of firms instead, we see that these are quite stable during the period under study, but with a drop in the number of firms in 2009 and 2010. This drop is likely due to the onset of the financial crisis which hit Sweden during 2009, resulting in firm exits the following years. Table A1.1 Number of active retailers in comparison and one-stop shopping in the IKEA entry municipalities. Haparanda

2001 2002 2003 2004 2005 2006 2007 2008 2009 2010

Kalmar

Karlstad

Gothenburg

Comparison

One-stop

Comparison

One-stop

Comparison

One-stop

Comparison

One-stop

25 25 29 24 25 23 21 22 21 18

179 198 196 182 196 186 182 174 173 162

36 36 40 34 36 33 29 31 26 23

225 239 238 229 255 244 235 227 221 201

36 37 39 34 33 30 30 30 30 27

238 258 264 251 266 255 248 246 241 220

130 126 108 106 103 103 101 102 103 93

738 792 786 801 797 789 793 808 783 714

Table A1.2 Number of entries and exits each year in comparison shopping in the IKEA entry municipalities. Haparanda

2001 2002 2003 2004 2005 2006 2007 2008 2009 2010

Kalmar

Karlstad

Gothenburg

Entry

Exit

Entry

Exit

Entry

Exit

Entry

Exit

– 7 8 1 4 2 5 1 2 4

– 7 4 6 3 4 7 0 3 7

– 9 10 3 6 3 5 2 2 6

– 9 6 9 4 6 9 0 7 9

– 9 10 4 5 4 7 2 3 4

– 8 8 9 6 7 7 2 3 7

– 33 17 23 22 22 20 21 18 15

– 37 35 25 25 22 22 20 17 25

Table A1.3 Number of entries and exits each year in one-stop shopping in the IKEA entry municipalities. Haparanda

2001 2002 2003 2004 2005 2006 2007 2008 2009 2010

Kalmar

Karlstad

Gothenburg

Entry

Exit

Entry

Exit

Entry

Exit

Entry

Exit

– 50 45 31 40 26 27 27 25 24

– 31 47 45 26 36 31 35 26 35

– 60 58 46 56 37 34 34 31 28

– 46 59 55 30 48 43 42 37 48

– 62 62 42 51 35 37 39 35 31

– 42 56 55 36 46 44 41 40 52

– 214 180 183 172 136 157 171 137 105

– 160 186 168 176 144 153 156 162 174

183

Journal of Retailing and Consumer Services 40 (2018) 175–187

M. Han et al.

Appendix 2. Estimation results Our main estimation results are presented in Table A1 below, in which a translog production function is used in the analysis. In this appendix, we also present the results of a Cobb–Douglas production-function model (Table A2) and of an ordinary difference-in-difference regression model (Table A3). For Gothenburg, Stockholm is the control region and for the other three entry regions, Umeå, Uddevalla, and Borlänge are used as control regions in our estimations. The results indicate only small differences between the translog and the Cobb–Douglas models, while the differences are larger when a non-production-function difference-in-difference model is estimated. See appendix Tables A2.1–A2.12

Table A2.1 Estimation results, dependent variable lnQi,t, translog model, spatially correlated treatment effects (one-stop shopping). Haparanda

lnKi,t – 1 lnLi,t – 1 ln2Ki,t – 1 ln2Li,t−1 lnLi,t−1lnKi,t TPt TRi TPt × TRi n R2

−1

Kalmar

Karlstad

Gothenburg

coef.

std. err.

coef.

std. err.

coef.

std. err.

coef.

std. err.

0.059 −0.635*** 0.117*** −0.003 0.003 −0.085*** 0.259** 0.300* 1828 75.54%

0.058 0.091 0.008 0.005 0.010 0.032 0.103 0.156

0.086 −0.677*** 0.121*** −0.006 0.000 −0.069** −0.023 0.163** 2314 74.54%

0.053 0.080 0.007 0.005 0.009 0.028 0.037 0.076

0.097** −0.543*** 0.110*** 0.001 −0.005 −0.081*** −0.026 0.012 2487 76.73%

0.049 0.072 0.007 0.004 0.008 0.030 0.040 0.067

0.226*** −0.554*** 0.108*** 0.003 −0.024*** −0.048*** 0.006 −0.014 7801 70.49%

0.025 0.042 0.004 0.002 0.004 0.016 0.017 0.023

*** significant at the 1% level, ** significant at the 5% level, and * significant at the 10% level.

Table A2.2 Estimation results, dependent variable lnQi,t, translog model (one-stop shopping). Haparanda

lnKi,t – 1 lnLi,t – 1 ln2Ki,t −1 ln2Li,t −1 lnLi,t −1lnKi,t TPt TRi TPt × TRi n R2

−1

Kalmar

Karlstad

Gothenburg

coef.

std. err.

coef.

std. err.

coef.

std. err.

coef.

std. err.

0.059 −0.634*** 0.117*** −0.003 0.003 −0.085*** 0.261** 0.278* 1828 75.54%

0.058 0.091 0.008 0.005 0.010 0.032 0.103 0.148

0.091* −0.681*** 0.122*** −0.007 −0.000 −0.091*** −0.066 0.162*** 2314 74.57%

0.053 0.080 0.007 0.005 0.009 0.031 0.045 0.062

0.097** −0.543*** 0.110*** 0.001 −0.005 −0.082*** −0.027 0.012 2487 76.73%

0.049 0.072 0.007 0.004 0.008 0.030 0.040 0.054

0.226*** −0.554*** 0.108*** 0.003 −0.024*** −0.035* 0.024 −0.033 7801 70.49%

0.025 0.042 0.004 0.002 0.004 0.019 0.023 0.032

*** significant at the 1% level, ** significant at the 5% level, and * significant at the 10% level.

Table A2.3 Estimation results, dependent variable lnQi,t, Cobb–Douglas model, spatially correlated treatment effects (one-stop shopping). Haparanda

lnKi,t – 1 lnLi,t – 1 TPt TRi TPt × TRi n R2

Kalmar

Karlstad

Gothenburg

coef.

std. err.

coef.

std. err.

coef.

std. err.

coef.

std. err.

0.090*** 0.752*** −0.056* 0.174 0.246 1828 72.15%

0.012 0.018 0.033 0.109 0.166

0.084*** 0.746*** −0.039 −0.053 0.215*** 2314 70.79%

0.011 0.016 0.030 0.040 0.081

0.104*** 0.747*** −0.052* −0.047 0.020 2487 73.48%

0.010 0.015 0.031 0.041 0.072

0.117*** 0.697*** −0.029* −0.013 −0.014 7801 66.79%

0.005 0.008 0.016 0.018 0.024

*** significant at the 1% level, ** significant at the 5% level, and * significant at the 10% level.

184

Journal of Retailing and Consumer Services 40 (2018) 175–187

M. Han et al.

Table A2.4 Estimation results, dependent variable lnQi,t, Cobb–Douglas model (one-stop shopping). Haparanda

lnKi,t – 1 lnLi,t – 1 TPt TRi TPt × TRi n R2

Kalmar

Karlstad

Gothenburg

coef.

std. err.

coef.

std. err.

coef.

std. err.

coef.

std. err.

0.090*** 0.752*** −0.056* 0.177 0.227 1828 72.15%

0.012 0.018 0.033 0.109 0.157

0.083*** 0.748*** −0.056* −0.088* 0.168** 2314 70.78%

0.011 0.016 0.033 0.048 0.066

0.104*** 0.747*** −0.054* −0.050 0.022 2487 73.48%

0.010 0.015 0.032 0.042 0.058

0.117*** 0.697*** −0.021 −0.002 −0.022 7801 66.79%

0.005 0.008 0.020 0.024 0.034

*** significant at the 1% level, ** significant at the 5% level, and * significant at the 10% level.

Table A2.5 Estimation results, dependent variable lnQi,t, difference-in-difference model, spatially correlated treatment effects (one-stop shopping). Haparanda

TPt TRi TPt × TRi n R2

Kalmar

Karlstad

Gothenburg

coef.

std. err.

coef.

std. err.

coef.

std. err.

coef.

std. err.

0.204*** 0.175 0.239 1828 32.49%

0.051 0.170 0.259

0.198*** 0.078 0.317** 2314 28.51%

0.047 0.062 0.127

0.024*** 0.076 −0.095 2487 30.08%

0.051 0.067 0.116

0.100*** −0.036 −0.011 7801 14.83

0.026 0.028 0.038

*** significant at the 1% level, ** significant at the 5% level, and * significant at the 10% level.

Table A2.6 Estimation results, dependent variable lnQi,t, difference-in-difference model (one-stop shopping). Haparanda

TPt TRi TPt × TRi n R2

Kalmar

Karlstad

Gothenburg

coef.

std. err.

coef.

std. err.

coef.

std. err.

coef.

std. err.

0.204*** 0.178 0.220 1828 32.49%

0.051 0.170 0.244

0.201*** 0.084 0.134 2314 28.37%

0.051 0.075 0.103

0.202*** 0.073 −0.066 2487 30.07%

0.051 0.069 0.094

0.093*** −0.044 0.027 7801 14.74%

0.033 0.039 0.055

*** significant at the 1% level, ** significant at the 5% level, and * significant at the 10% level.

Table A2.7 Estimation results, dependent variable lnQi,t, translog model, spatially correlated treatment effects (comparison shopping). Haparanda

lnKi,t – 1 lnLi,t – 1 ln2Ki,t −1 ln2Li,t −1 lnLi,t −1lnKi,t TPt TRi TPt × TRi n R2

−1

Kalmar

Karlstad

Gothenburg

coef.

std. err.

coef.

std. err.

coef.

std. err.

coef.

std. err.

n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a.

n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a.

0.141 −0.307** 0.091*** −0.009 −0.000 0.013 0.150 −0.212 324 78.02%

0.183 0.129 0.014 0.012 0.027 0.064 0.081 0.169

−0.028 −0.364*** 0.090*** −0.012 0.025 −0.013 0.009 −0.121 326 77.97%

0.147 0.129 0.014 0.009 0.024 0.065 0.084 0.156

0.311*** −0.753*** 0.135*** −0.004 −0.035*** −0.057 0.027 0.092 1075 69.62%

0.073 0.110 0.010 0.007 0.012 0.036 0.039 0.070

*** significant at the 1% level, ** significant at the 5% level, and * significant at the 10% level.

185

Journal of Retailing and Consumer Services 40 (2018) 175–187

M. Han et al.

Table A2.8 Estimation results, dependent variable lnQi,t, translog model (comparison shopping). Haparanda

lnKi,t – 1 lnLi,t – 1 ln2Ki,t −1 ln2Li,t −1 lnLi,t −1lnKi,t TPt TRi TPt × TRi n R2

−1

Kalmar

Karlstad

Gothenburg

coef.

std. err.

coef.

std. err.

coef.

std. err.

coef.

std. err.

n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a.

n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a.

0.139 −0.290** 0.089*** −0.009 −0.000 0.005 0.136 −0.109 324 77.96%

0.184 0.128 0.014 0.116 0.027 0.067 0.085 0.129

−0.027 −0.364*** 0.090*** −0.012 0.025 −0.015 0.005 −0.088 326 77.96%

0.147 0.129 0.014 0.009 0.024 0.065 0.085 0.125

0.318*** −0.758*** 0.136*** −0.004 −0.036*** −0.037 0.053 −0.052 1075 69.58%

0.073 0.110 0.010 0.007 0.012 0.046 0.052 0.008

*** significant at the 1% level, ** significant at the 5% level, and * significant at the 10% level.

Table A2.9 Estimation results, dependent variable lnQi,t, Cobb–Douglas model, spatially correlated treatment effects (comparison shopping). Haparanda

lnKi,t – 1 lnLi,t – 1 TPt TRi TPt × TRi n R2

Kalmar

Karlstad

Gothenburg

coef.

std. err.

coef.

std. err.

coef.

std. err.

coef.

std. err.

n.a. n.a. n.a. n.a. n.a. n.a. n.a.

n.a. n.a. n.a. n.a. n.a.

0.096*** 0.646*** −0.003 0.112 0.067 324 73.88%

0.027 0.033 0.069 0.087 0.178

0.091*** 0.677*** −0.018 −0.049 −0.075 326 72.58%

0.025 0.035 0.072 0.093 0.172

0.091*** 0.071*** −0.044 0.010 0.128* 1075 64.20%

0.014 0.020 0.039 0.042 0.076

*** significant at the 1% level, ** significant at the 5% level, and * significant at the 10% level.

Table A2.10 Estimation results, dependent variable lnQi,t, Cobb–Douglas model (comparison shopping). Haparanda

lnKi,t – 1 lnLi,t – 1 TPt TRi TPt × TRi n R2

Kalmar

Karlstad

Gothenburg

coef.

std. err.

coef.

std. err.

coef.

std. err.

coef.

std. err.

n.a. n.a. n.a. n.a. n.a. n.a. n.a.

n.a. n.a. n.a. n.a. n.a.

0.096*** 0.646*** −0.005 0.110 0.052 324 73.88%

0.028 0.033 0.072 0.091 0.137

0.091*** 0.678*** −0.020 −0.053 −0.051 326 72.57%

0.025 0.035 0.072 0.093 0.139

0.090*** 0.706*** −0.027 0.034 −0.046 1075 64.12%

0.014 0.020 0.050 0.057 0.082

*** significant at the 1% level, ** significant at the 5% level, and * significant at the 10% level.

Table A2.11 Estimation results, dependent variable lnQi,t, difference-in-difference model, spatially correlated treatment effects (comparison shopping). Haparanda

TPt TRi TPt × TRi n R2

Kalmar

Karlstad

Gothenburg

coef.

std. err.

coef.

std. err.

coef.

std. err.

coef.

std. err.

n.a. n.a. n.a. n.a. n.a.

n.a. n.a. n.a.

0.016 0.098 0.034 324 33.23%

0.110 0.139 0.283

0.010 −0.443 0.334 326 25.01%

0.118 0.150 0.282

−0.048 0.026 −0.064 1075 5.08%

0.064 0.069 0.124

*** significant at the 1% level, ** significant at the 5% level, and * significant at the 10% level.

186

Journal of Retailing and Consumer Services 40 (2018) 175–187

M. Han et al.

Table A2.12 Estimation results, dependent variable lnQi,t, difference-in-difference model (comparison shopping). Haparanda

TPt TRi TPt × TRi n R2

Kalmar

Karlstad

Gothenburg

coef.

std. err.

coef.

std. err.

coef.

std. err.

coef.

std. err.

n.a. n.a. n.a. n.a. n.a.

n.a. n.a. n.a.

−0.012 0.056 0.125 324 33.30%

0.115 0.145 0.217

0.004 −0.451*** 0.285 326 25.05%

0.119 0.151 0.227

−0.088 −0.025 0.108 1075 5.11%

0.081 0.092 0.133

*** significant at the 1% level, ** significant at the 5% level, and * significant at the 10% level. J. Priv. Enterp. 22, 137–160. Hicks, M.J., 2008. Estimating Wal-Mart's impact in Maryland: a test of identification strategies and endogeneity tests. East. Econ. J. 34, 56–73. Håkansson, J., Li, Y., Mihaescu, O., Rudholm, N., 2016. Big-box retail entry in urban and rural areas: Are there productivity spillovers to incumbent retailers? HUI Working Papers No. 118. Jia, P., 2008. What happens when Wal-Mart comes to town: an empirical analysis of the discount retailing industry. Econometrica 76, 1263–1316. Jones, K., Doucet, M., 2000. Big-box retailing and the urban retail structure: the case of the Toronto area. J. Retail. Consum. Serv. 7, 233–247. Maican, F., Orth, M., 2012a. Productivity dynamics and the role of big-box entrants in retailing. Research Institute of Industrial Economics Working Paper No. 898. Research Institute of Industrial Economics, Stockholm. Maican, F., Orth, M., 2012b. A dynamic analysis of entry regulations and productivity in retail trade. Research Institute of Industrial Economics Working Paper No. 939. Research Institute of Industrial Economics, Stockholm. Marshall, A., 1890. Principles of economics. McMillan and Co., London. McCann, P., 2001. Urban and regional economics. Oxford University Press, Oxford. Merriman, D., Persky, J., Davis, J., Baiman, R., 2012. The impact of an urban Wal-Mart store on area businesses: the Chicago case. Econ. Dev. Q. 26, 321–333. Neumark, D., Zhang, J., Cicarella, S., 2008. The effect of Wal-Mart on local labor markets. J. Urban Econ. 63, 405–430. Nilsson, H., 2015. The impact of big-boxes on local retail: What happens when IKEA comes to town? Dalarna University Licentiate Theses, No. 4. Dalarna University, Dalarna. OECD, 2001. Measuring Productivity: Measurement of Aggregate and Firm Level Productivity Growth, a Manual. Organization for economic co-operation and development, Paris. O’Sullivan, A., 2003. Urban Economics, 5th ed. McGraw-Hill, Irwin, New York. Rubin, D.B., 1978. Bayesian inference for causal effects: the role of randomization. Ann. Stat. 6, 34–58. Rubin, D.B., 1990. Formal odes of statistical inference for causal effects. J. Stat. Plann. Inference 25, 279–292. Rudholm, N., Li, Y., Carling, K., 2017. How does IKEA Entry Affect Labor Productivity in Durable Goods Retailing? A Synthetic Control Approach. Manuscript, Dalarna University.

References Abadie, A., 2005. Semiparametric difference-in-difference estimator. Rev. Econ. Stud. 79, 1–19. Artz, G.M., Stone, K.E., 2012. Revisiting Wal-Mart's impact on Iowa small-town retail: 25 years later. Econ. Dev. Q. 26, 298–310. Bailey, T.C., Gatrell, A.C., 1995. Interactive spatial data analysis. Basker, E., 2007. The causes and consequences of Wal-Mart's growth. J. Econ. Perspec. 21, 177–198. Basker, E., 2005. Job creation or destruction? Labor market effects of Wal-Mart expansion. Rev. Econ. Stat. 87, 174–183. Brown, S., 1989. Retail location theory: the legacy of Harold Hotelling. J. Retail. 65, 450–470. Christensen, L.R., Jorgenson, D.W., Lau, L.J., 1971. Conjugate duality and the transcendental logarithmic production function. Econometrica 39, 255–256. Chung, W., Kalnins, A., 2001. Agglomeration effects and performance: a test of the Texas lodging industry. Strateg. Manag. J. 22, 969–988. Cressie, N., Wikle, C.K., 2011. Statistics for spatio-temporal data. John Wiley & Sons, Inc. Daunfeldt, S.-O., Mihaescu, O., Nilsson, H., Rudholm, N., 2017. What happens when IKEA comes to town? Reg. Stud. 51, 313–323. Daunfeldt, S.-O., Mihaescu, O., Nilsson, H., Rudholm, N., 2016. When IKEA Enters: Do local retailers win or lose? HUI Working Paper No. 109. HUI Research, Stockholm. Davidson, S.M., Rummel, A., 2000. Retail changes associated with Wal-Mart's entry into Maine. Int. J. Retail Distrib. Manag. 28, 162–169. Delgado, M.S., Florax, R.J.G.M., 2015. Difference-in-differences techniques for spatial data: local autocorrelation and spatial interaction. Econ. Lett. 137, 124–126. Foster, L., Haltiwanger, J., Krizan, C.J., 2006. Market selection, reallocation, and restructuring in the U.S. retail trade sector in the 1990s. Rev. Econ. Stat. 88, 748–758. Greenstone, M., Hornbeck, R., Moretti, E., 2010. Identifying agglomeration spillovers: evidence from winners and losers of large plant openings. J. Polit. Econ. 118, 536–598. Han, M., Håkansson, J., Rönnegård, L., 2016. To what extent do neighbouring populations affect local population growth over time. Popul., Space Place. 22, 68–83. Hernandez, T., 2003. The impact of big-box internationalization on a national market: a case study of Home Depot Inc. in Canada. Int. Rev. Retail Distrib. Consum. Res. 13, 77–98. Hicks, M.J., 2007. Job turnover and wages in the retail sector: the influence of Wal-Mart.

187