The Effect of Mergers on Retail Prices: Evidence from Germany

Transcription

1 The Effect of Mergers on Retail Prices: Evidence from Germany Dennis Rickert Jan Phillip Schain Joel Stiebale June 6, 2017 Preliminary and incomplete. Please do not cite without authors permisison. Abstract This paper estimates the effects of a merger between a German supermarket chain and a discounter on consumer prices. We exploit the fact that retailers set prices at a local level and use a difference-difference estimator to compare regional markets with pre-merger overlap of acquirer and target to a control group of products in unaffected markets. Our results indicate a price increase due to the merger, where this price increase is solely driven by supermarkets, but not by discounters. We also provide evidence that retail prices were more likely to increase in local markets with high expected changes in concentration and that remedies imposed where not sufficient to offset anti-competitive effects. Finally, our novel identification strategy allows to isolate the upward pricing pressure due to increased concentration from global efficiency gains. JEL codes: D22, K21, L11, L66, L81, L69 Keywords: Mergers and acquisitions, ex-post merger evaluation, retail markets, retail prices, competition. Duesseldorf Institute for Competition Economics (DICE), Heinrich-Heine University Duesseldorf. 1

2 1 Introduction Mergers and acquisitions (M&As) play an important role in the world economy. The combined value of worldwide M&As exceeded $4 trillion in 2015 and major deals often make news headlines. 1 In particular, the retail sector has experienced a number of significant mergers within the last two decades, which caused the evolution to highly concentrated market structures all over Europe. In Germany, France, and the UK, the combined market share of the five biggest retailers exceeds 70% (Inderst, 2013), whereas in Northern European countries the three largest retailers hold combined market shares of around 90% (Allain et al., 2013). Since retail purchases constitute a high share of consumption expenditures, increasing concentration is a central topic both in competition policy and within the industrial organization literature. Economic theory offers opposing predictions regarding the impact of M&As. On the one hand, mergers can be used by firms to eliminate (intra-brand) competition in downstream markets which increases market power and prices at the expense of consumers (von Ungern-Sternberg, 1996; Dobson and Waterson, 1997). On the other hand, they may induce efficiency gains which benefit consumers in the form of lower prices or new and improved products. The net effects are particularly difficult to predict in the presence of buyer buyer power in such vertical market structures. First, merging firms might be able to negotiate better terms due to size related discounts (e.g. Katz, 1987; Scheffman and Spiller, 1992) and a reduction in the number of suppliers alternatives (e.g. Chipty and Snyder, 1999). In this context, retailers increase their countervailing bargaining power vis-a-vis the suppliers (Galbraith, 1954) and benefit from efficiency gains in the distribution channel. 2 However, it is by no means clear that efficiency gains are passed on to consumers since the pass-through depends on demand curvature (Bulow and Pfleiderer, 1983; Weyl and Fabinger, 2013), the intensity of downstream competition (Dobson and Waterson, 2005), and the type of vertical agreement firms contract on (Gaudin, 2016). Since theoretical predictions are quite ambiguous, a growing empirical literature has estimated the effects of M&As on prices and has produced mixed results. 3 For instance, Hosken et al. (2012) analyze 14 retail mergers and find that prices were raised after some mergers while they decreased or remained unchanged in other cases. Allain et al. (2013) report a significant price increase after a merger among French retailers, while Argentesi et al. (2016) find no significant price changes after mergers in the Dutch retail market. The ambiguity is further confirmed by the heterogene- 1 See, for instance, accessed Feb 11, This, in turn, could lead to higher prices for consumers since suppliers have to recuperate their losses at the expense of less powerful retailers. This mechanism is referred to as waterbed effect (Inderst and Valletti, 2011). 3 See the overview of related literature in Ashenfelter et al. (2014). 2

3 ity in national antitrust authority decisions regarding mergers. While there is a consensus among antitrust authorities on the existence of high levels of buyer power (e.g. Rodrigues, 2006; Competition Commission, 2008; Bundeskartellamt, 2014b), the implications for competition policy vary substantially. For instance, the European Commission and the Federal Trade Commission expect rather welfare-enhancing effects from buyer power increases. Consequently, both recently approved 88% respectively 89% of all proposed mergers without constraints (Allain et al., 2013). In contrast, Germany s federal cartel office (Bundeskartellamt) either denied or imposed strict remedies on all recently proposed mergers, which reflects their concerns about consumer surplus and producer profits. 4 The aim of this paper is to investigate the price effects of a merger between two German retail chains, which was challenged by competition authorities but eventually cleared subject to remedies. We first follow the existing literature on ex-pots merger evaluations to estimate the causal net price effect due to the merger. However, given the ambiguous results in the theoretical and empirical literature, we develop a novel and unique identification strategy to identify the upward pressure pricing (UPP) that is due to a higher market concentration and the downward pressure pricing (DPP) that stems from lower costs. The latter occurs from economies of scale and scope, for example by increases in global bargaining power vis-à-vis the manufacturers or efficiency gains from sharing warehouse capacities. However, we aim to further contribute to the literature by analyzing heterogeneous effects which can potentially further explain the inconclusiveness of recent empirical studies. In particular, we investigate varying price responses of private label products and national brands, between discounters and supermarkets, and how these effects vary with predicted changes in regional retail market concentration. We exploit a rich consumer-level panel data set from which we extract information about prices, products and quantities at the regional level. Our empirical analysis focuses on 4 product categories: dairy products, coffee, diapers, and toilet paper and thus includes both differentiated and rather homogeneous products. We use a difference-in-difference estimator which exploits the fact that retailers set prices at the local market level. Our identification strategy for the net price effect follows the empirical literature on retrospective merger analysis. It is based on a comparison of local markets in which both acquirer and target operated in the pre-merger period to control markets that did not experience a change in market structure. In a subsequent setup, we explore concentration effects and efficiency gains separately. The UPP is identified by the comparison of local markets in which both firms were active pre-merger to a comparison group of markets in which only one of the merging parties was operating. The former group was affected both by a change in market concentration and 4 See, for instance, Bundeskartellamt (2014a) on denied mergers or Bundeskartellamt (2005, 2010a) on conditionally excepted mergers. To the latter we could also add Rewe/Wasgau

4 by potential efficiency gains, while the latter markets were only affected by potential national-wide cost savings in one of the retail chains. The DPP is identified by comparing markets that experienced potential cost efficiencies but not an increase in concentration since there was only one of the two merging retail chain operating with markets that remained unchanged post merger where non of the merging parties had positive pre-merger market shares. Our results indicate that merging retailers and their competitors raised consumer prices in affected markets, where this price increase is of the same magnitude for private label products and national brands. While, on average, estimated price changes are small, prices increased substantially in regional markets with high predicted changes in market concentration. Price adjustments are concentrated in products sold by supermarkets, but there is little evidence for changes in pricing strategies among products sold by discounters. We investigate the effects of remedies imposed by the German cartel office which involved the sale of the target firm s retail stores to a competitor in regional markets with high pre-merger market shares of acquirer and target. The results indicate that prices were less likely to rise in these markets, which implies that imposed remedies might not have been sufficient to prevent anti-competitive effects of the merger. When we separate the effects into UPP and DPP, we find as expected a significant downward pressure on prices due to potential efficiency gains. The negative effect on prices is higher in more competitive markets which implies imperfect pass-through rates of cost savings. Consistent with economic intuition, results indicate that cost savings could not be immediately realized and it took about one year for prices to adjust to higher efficiency. Moreover, firms raised prices in markets with overlap immediately after the merger was approved, where this price increase remains relatively stable over the whole post-merger period. The rest of this paper is organized as follows. Section 2 discusses the German retail market and the merger case. In section 3, we provide a description of our consumer-level panel data set, section 4 details our identification strategy. Results of the empirical analysis are presented in section 5, and section 6 concludes. 2 The merger in the German retail market In this section, we first describe the German grocery retail market structure in terms of buyer power concerns in section 2.1, before providing detailed background information on the merger in section 2.2, describing pre- and post-merger market structure (section 2.3), and characterizing the local component of price competition among retailers (section 2.4) 4

5 2.1 German retailing market The German retail sector already has developed to a highly concentrated market structure. Induced by an expansive merger and acquisition strategy, the five largest retailers in Germany have increased their market shares in the two preceeding decades from 50% to over 80% in 2014 (Inderst, 2013), which is above the average of 70% in other Western European countries and well above the US average of 33% (Allain et al., 2013). Prominent examples for recent mergers in Germany are the cases of Edeka/Trinkgut (Bundeskartellamt, 2010a), (Bundeskartellamt, 2014a), and Wasgau/Rewe (Bundeskartellamt 2013). The ongoing concentration process in Germany has raised concerns about consumer surplus and producer profits. As a consequence, the grocery sector has been under stringent scrutiny and steady investigations by the German national antitrust authority. Finding that high market shares are indicative of high bargaining power, it has initiated a total of 14 cases between 2004 to 2011 (Haucap et al., 2014), which often included unannounced raids at retail headquarters (Bundeskartellamt, 2010b). Other major concern were vertical price fixing agreements and presumptions of below cost pricing practices (Bundeskartellamt, 2014b). These concerns culminated in a sectoral analysis of the market, which according to the Act against Restraints on Competition ( 32e Gesetz gegen Wettbewerbsbeschränkungen, GWB) can be conducted if competition in the market appears to be seriously distorted. The main purpose of the investigation was the assessment of retail buyer power and the degree of buyer power abuse (Bundeskartellamt, 2014b). Consistent with results of other antitrust authorities, the German cartel office found evidence of high levels of buyer power. However, the conclusions for merger decisions were diverging from the viewpoint of the European Commission and the FTC who expect rather welfare-enhancing effects from buyer power increases. Whereas the Bundeskartellamt either denied the proposed mergers (e.g. Bundeskartellamt, 2014a) or imposed strict remedies (e.g. Bundeskartellamt, 2005, 2010a), EC and FTC recently approved 88% respectively 89% of all proposed mergers without constraints (Allain et al., 2013). 2.2 The merger We consider a merger of the two retailers R1 and R2 with pre-merger market shares of 25% and 5% which was proposed at the end of 2007 and approved in midyear There have been two other mergers in 2005 and 2010, but no other major retail mergers outside the Safe-Harbour region between 2006 and 2009, which allows a clean analysis of the merger effects. The Outsiders O1-O3 split the remainder rather equally (20%,15%,15%) The acquirer R1 is a multi-line retailer with two different retailing formats, supermarkets and discounters, which we label 5

6 R1 S and R1 D henceforth. The target R2 is a discounter. Since aggregated post-merger market shares exceeded the safe-harbor threshold of 22% (Competition Commission, 2008). Thus, the merger was in the focus of the national cartel office which identified regions where the firms had large market share and competition would be severely distorted. The merger was approved under the condition R1 divests and sells 378 out of 2700 stores in question to outsider O1. R1 then converted (relabeled) 1800 stores R1 D and the remaining stores kept their former label R2, but were effectively under the control of R1. The outsider O1 pursued the same strategy in the remedy regions, where it was allowed to obtain the stores of the target. It relabeled the acquired stores into its own discount retail chain O1 D. Basically, the imposed remedies then affect regions with overlap of O1 and R2 since there was another merger. Consequently, we will continue to refer to the merger of R1 and R2, but treat remedy regions as part of the treatment group in our basline specification. However, having this in mind, it seems worth studying also heterogeneous effects for remedy and non-remedy regions. 2.3 Pre- and post-merger market structure 34 different retailers are active on the German grocery retailing market which can be grouped into three formats: Discounter, drugstore, and supermarket with market shares of 51.76%, 3.02%, and 44.56%. Furthermore, there are some specialized retailers, such as, for instance, cash-and-carry stores, pharmacies and also online retailers, which distribute some of the products. However, neither these specialized retailers, nor Internet purchasing are reasonable alternative distribution channels for grocery purchases in the sample period. Thus, we exclude all products from these channels from our analysis. Consequently, insiders are defined as the two merging firms, which are a supermarket and a discounter. Outsiders are all remaining firms. Table 1 presents average market shares per category distinguishing between national brands and private labels, and retailing format pre and post merger. The market share analysis provides evidence that private labels are an important driver on the market. Particularly, discounters assortment consist of a large share of private labels, but we see that other retail formats also offer a high proportion of private label products. Discounters have the highest market shares in the toilet paper market, the yoghurt market and the milk market selling their private label products. Other retail formats dominate in the coffee and diaper market by selling national brands. Table 2 shows average prices per category distinguishing between treatment and control group, and retailing format pre and post merger. We see that pre and post merger, in treatment and control group, discounters have lower prices in all categories but milk. Post-merger, supermarkets raise their prices for toilet paper, coffee and diapers in treatment markets relatively to control markets, while 6

7 Table 1: Market shares in percent Supermarkets Discounters Pre Post Pre Post Toiletpaper NB PL Coffee NB PL Yoghurt NB PL Diaper NB PL Milk NB PL Table 2: Average prices by product category Supermarkets Discounters Pre Post Pre Post Toilet paper Control Treat Coffee Control Treat Yoghurt Control Treat Diaper Control Treat Milk Control Treat Respectively, pricing units are given the following: cents per unit, cents per gram, cents per gram, cents per unit, cents per Liter/1000 the average prices of milk and yogurt change only little around the time of the merger. 2.4 Local Market definition and national bargaining In order to identify the causal effect of the merger on retail prices, an accurate market definition is essential. Unlike in the UK (Dobson and Waterson 2005), but as in the US (Dafny et al. 2012), in the Netherlands (Argentesi et al. 2016) or in France (Allain et al. 2015), German retailers adopt a local pricing strategy. Retailers local pricing policy stems from the fact that the two largest retailers, Edeka and Rewe, evolved from former buying cooperatives of local merchants, which were subsequently transformed into national retail chains with centralized headquarters. Due to this historical development, the national retailers give many degrees of freedom to the local merchants regarding individual price and variety decisions while bundling purchasing activities of local retailers. 7

8 Figure 1: Analysis of Variance in Local and National Component These historical background is backed up by additional evidence from the of the merging firm s homepage. R1 owns 11,400 stores operated by roughly 4,500 independent merchants who adapt the day-to-day business activities to local needs and market specifics. 5 The retail brand R1 has derived from regional cooperative built to join forces in purchasing activities. To date of the merger, there have been seven regional wholesale cooperatives derived from former regional buying groups, who are the operational entities delivering items to the stores of the independent merchants and coordinating central issues regarding distributions and sales. These wholesale cooperatives may also own retail outlet stores and production facilities. The distribution is effected from 38 distribution warehouses managed by the seven wholesale cooperatives. Furthermore, there is a central headquarter coordinating commodities transactions ( Nationales Warengeschäft ) at the national level. The headquarter employs a national purchasing strategy for several product categories, such as fruits, vegetables, and flowers as well as all non-food product categories. We interpret these stylized facts as strong evidence that local merchants receive their stock from central distribution warehouses, which is coordinated by the national headquarters. While retail prices are set at the regional level, there is evidence that wholesale prices are determined nationally. As we discuss in the next section, these facts play an important role in our identification strategy. The assumption that wholesale prices are thus considered to be determined at the national level seems reasonable if there is no price discrimination among local merchants. The hypothesis is backed up by practices of the German antitrust authorities who define national procurement markets (Bundeskartellamt, 2014b, p.132). Another justification might be found in 5 Due to confidentiality agreements, we are neither allowed to display names nor links to homepages. 8

9 the recent discussion in the media on so-called wedding rebates, where retailer R1 demanded better purchasing conditions for all its stores after the merger. However, at the same time, local retailers have the a degree of freedom to local prices. This could be implemented by price discrimination with respect to regional demographic characteristics or region-specific discounts to consumers respectively promotions such as buy-one-get-one-free offers. Consequently, local retail prices could vary across regions within retailers even for products whose wholesale list price are bargained at the national level. Since the local market definition (and the nature of retail price setting) is crucial to the merger evaluation, we use our dataset to provide descriptive insights on the local dimension of retail price setting practices. Figure 1 supports the hypothesis of local pricing by showing the decomposition of price variance into national and local variance. To this extent, we regress for each product category considered that are toilet paper, coffee, yogurt, diapers, and milk prices on retailer-brand fixed effects. This simple reduced-form regression yields an R 2 indicating the explanatory power of the national pricing component at the retail-brand level. By construction, the leftover of the regression yields the variance which cannot be explained by national pricing strategies. Thus, the figure plots 1 R 2 over time and product categories. For all product categories, we find a high degree of variance which can be explained by local components, ranging from roughly 40% for milk and yogurt to around 90% for diapers. In additional estimations, we regress prices on regional characteristics and find additional support for local pricing. In particular, table 12 shows that regional prices vary with local market conditions and that retailers set higher prices for regions with high percentage of children, high population density, and high market concentration. The Bundeskartellamt ackknowledges these local components of retail competition and defines 345 local catchment areas, which are the basis for decisions on mergers or abuse of dominance (Bundeskartellamt, 2007). In contrast to this rather broad market definition, we define local markets at a more disaggregate level, which corresponds to the classification of municipalities. According to this classification, there are roughly 12,000 local markets and the set of competitors therefore contains all stores located in this municipality. Municipalities have an average size of km 2 (median of km 2 ) and a standard deviation of km 2 (Destatis 2017). However, our assessment of the average size of a local market could be too small, when following the local market definition of European competition commission (European Commission, 1999) who propose to center given retail stores and draw circles around these stores with corresponding radii of 20 minutes driving time by car (roughly kilometers). Our market definition could be, on average, too narrow, while being to wide for some outliers such as Hamburg or Berlin. That is why we conduct a number of robust check, where we aggregate small municipalities and disaggregate big municipalities. 9

10 Based on these stylized facts, we choose an empirical strategy, in particular regarding the definition of treatment and control group, which fits to the observed market characteristics and retailer strategies. 3 Data To assess merger effects, we use household panel enriched with regional information from two other sources. We present and describe this dataset in section 3.1 before reporting details on construction of products and prices in section Data description The primary dataset is a representative survey of households distributed across all regions of Germany obtained by GfK Panel Services. The GfK Panel Services collects information on all actual transactions of up to 20,000 households who are selected with respect to geographical, social, and economic characteristics to be representative of the German population. This dataset entails two distinct features which makes it well-suited for the purpose of our analysis. First, all panel members track their entire purchase history using home-scanning devices. Thus, it contains information on the name of the brand, the label type (national brand or private label), the retailer (e.g., supermarket, discounter, drugstore, or specialized shop), and type of product (including package size among other descriptive variables) as well as the actual transaction price (including any discounts and promotions). Thus, it gives a more detailed picture on household shopping behavior compared to checkout scanner data which can only track purchases within a particular store or datasets from other marketing agencies in Germany, which do not provide information on all discounters (see e.g., Draganska et al. 2010). Second, the data encompass detailed information on panel member characteristics, including the postal code of their accommodation and their yearly income. Thus, information is reported in detail for the individuals and can be aggregated to the retailer-productregional level such that the analysis is representative for the German population. We enrich our analysis with data obtained from INKAR providing regional information on counties ( Kreise ) and municipalities ( Gemeinden ). For this purpose, we match postal code with a municipality identifier for which we use the Deutsche Post matching key. In the considered time period, there has been a number of reforms in municipalities, where postal codes were reallocated to other municipalities or where new municipalities have been created by consolidating old ones. To take adequate account, we retrospectively allocate postal codes to the definition of the year of Furthermore, there are some cases where the local postal code may belong to two (or more) 10

11 municipalities. By assuming that there is no systematic assignment to either control or treatment group, we feel confident to drop these cases, which leaves us with 78% of the total observations. Having matched the postal code to the municipality level also allows to identify all purchases and prices within a regional market. While we lack precise information about the location of stores, we can assign consumer level purchases to regions and retail chains. We assume that a retail chain is located in a certain region if we observe purchases at the retail chain in the region. Finally, we use additional data from the German cartel office available at the 2 digit postal code level to identify regions with remedies. 3.2 Construction of products and prices Our empirical analysis focuses on 4 product categories: dairy products (including milk and yogurt), coffee, diapers, and toilet paper, which vary in the degree of product differentiation. The selected products resemble a reasonable shopping basket with homogeneous base products, such as milk and toilet paper, more differentiated products, such as yogurt and coffee and a technologically more sophisticated product, such as diapers. Even though our database entails a rich set of products attributes, it lacks the universal product code (UPC). To emulate the code, we build a unique product identifier from information on the brand name, package size, and product type. The latter varies over product categories. It defines the number of layers for toilet paper, the per diaper size for the diaper category, the coffee type (ground coffee, espresso, crema, or pads/capsules) for the coffee category, and the fat content for dairy products. Two examples according to our product definition are Brand A s 250 grams espresso coffee and R1 s private label diaper bigpack for kg juniors. Furthermore, the same product sold by two different retailers is treated as two different alternatives because prices vary not only across products but also for a given product across retail chains and regions. Observing a though representative subsample of the German population has the disadvantage that for some regions and some products we observe low (or no) frequencies per day. Consequently, we aggregate the data to quarterly periods to ensure a sufficient number of observations. Between 2005 and 2010, we observe 4,109,022 purchases. 8% these observations are purchases of toilet paper, 8.84% coffee-, 44.23% yogurt-, 3.51% diaper- and 34.58% milk purchases. On average, we observe 59 purchases per period and local market ranging between 1 and 9425 with a standard deviation of 208. We then build mean prices per product, retailer and region in Eurocents per unit of size. This unit of size depends on the product category. De facto, it is either grams or milliliter for food products, i.e., coffee and dairy products. For diapers and toilet paper, it is a per-unit price. Our 11

12 primary price definition is the transaction price, which is the effective price paid at the checkout counter. However, the detailed dataset provides information on whether or not a product was purchased within a promotional activity. This allows us to distinguish between regular prices and promotional prices. In our baseline setting we build seperate mean prices for target and acquirer. To avoid composition effects (e.g., the merger target had fewer national brands in its portfolio), as a robustness check we treat the merging parties as a single firm in markets with pre-merger overlap. 4 Empirical Strategy The aim of this study is (i) the ex-post evaluation of price effects due to the merger between retailers R1 and R2 and (ii) the decomposition of the net effect into a competition- and an efficiency effect. To this extent, we develop a novel and unique identification strategy, whose underlying assumptions we describe in section 4.1 before presenting the empirical specification in section Identification A simple before-after analysis is not sufficient to estimate the effects of mergers on prices. Observed price changes might also be attributed to demand shifts or cost constraints. Rather we aim to compare price changes around the merger to a counterfactual scenario in which no merger took place. For this purpose, we exploit the fact that neither the target retailer nor the acquirer owns retail stores in each local market in Germany. Therefore, we compare regions which experienced a change in market concentration, i.e., markets in which both R1 and R2 were active before the merger, to markets without pre-merger overlap. Our identification strategy thus relies on the assumption that firms use a regional pricing strategy such that changes in product prices are independent across regional markets after controlling for a range of control variables and fixed effects. The stylized facts in section 2.4 and the market definition exercises of the European competition commission (European Commission, 1999) as well as the German cartel office (Bundeskartellamt, 2014b) provide strong evidence for local pricing decisions, such that we can rule out national pricing strategies. The causal price effect due to the merger is identified by the implementation of a simple differencein-difference (DiD) estimator. The DiD approach compares pre- and post-merger prices of treated regions (i.e., regions affected by the merger) to pre- and post-merger prices in a control group. Taking double differences isolates the merger effect from other factors that might impact prices, such as demand shocks or liquidity constraints. However, the estimator relies on a parallel trend assumption, which states that absent the merger prices in treatment and control group would have moved identically. For this reason, we describe how we operationalize our local market definition 12

13 and the assignment of local markets to treatment or control group in the following subsections Treatment- and control groups in baseline scenario Local markets are defined by the boarders of municipalities. The dataset albeit containing very rich information about product and consumer characteristics lacks information on the retail location. However, the it provides the location of every consumer at the postal code level, which allows to infer store locations by assuming that households shop in the local market they inhabit. We match the postal code to regional data at municipality level via a matching key. Subsequently, we are able to identify all purchases and prices within a local market and for each quarter. From purchases within the municipality we infer the location of insider firms R1 and R2. The set of competitors therefore contains all stores located in this municipality. We follow the literature on retail merger evaluation (e.g. Houde, 2012) to define the treatment group as local markets affected by the merger. More precisely, the baseline specification of our treatment group contains regions which experienced a change in market concentration, i.e. markets in which both R1 and R2 were active before the merger. As the merger was approved solely under the condition to sell roughly 300 stores to competitors in some local markets, our definition of treatment groups includes local markets where the competitor acquired the target. Consequently, we treat the remedies as another merger in other markets. Treatment groups are compared to the control group of markets without pre-merger overlap, where we assume parallel price trends for treatment and control group absent the merger. The broadest definition of a control group would contain all local markets without a pre-merger market overlap of acquirer and target. However, we follow Allain et al. (2013) and exclude regional markets that are located geographically close to treatment markets, which could be indirectly affected by the merger and thus contaminate results. To this extent, we assume that each store is located at the municipality center and define so-called catchment areas around these stores. 6 We then construct an algorithm which (i) identifies the center of each municipality, (ii) calculates the distance to all other municipalities, and (iii) eliminates all municipalities within a distance of less than 15 kilometers. Figure 2(a) illustrates how we define treatment and control groups for our baseline specification. Treatment regions, which consist of local markets where both R1 and R2 were active, are colored in red. Control groups colored in green are defined as all firms in a market without pre-merger overlap and outside a circle of 15 kilometers around treatment regions. According to this definition, control groups are defined as local markets with (i) outsiders solely, (ii) outsiders and R1 or R2, or 6 We use the user-written command opencagegeo (Zeigermann, 2016) to obtain longitude and latitude of each municipality. 13

14 Figure 2: Definition of Treatment and Control Groups Treatment Group Control Group R2 Treatment Group Control Group R2 Excluded Group Excluded Group R1 R2 R1 R2 15 km R 1 15 km R1 R1 R2 R1 R2 R1 R2 R1 R2 (a) Specification 1: Baseline (b) Specification 2: Net Price Effect Treatment Group Control Group R2 Treatment Group Control Group R2 Excluded Group Excluded Group R1 R2 R1 R2 15 km R1 15 km R1 R1 R2 R1 / R2 R1 R2 R1 / R2 (c) Specification 3: Upward Pricing Pressure (d) Specification 4: Downward Pricing Pressure (iii) R1 or R2, but without outsiders. Grey-colored local markets are excluded from the definition of the control group if they are situated within a 15km circle around a treatment group. This above specification of treatment and control group identifies local market power and if there are any regional efficiency gains. The control group may be ill-suited for the purpose of analysis if retailers engage in national pricing strategies, where the internalization of competition effects is adapted globally. In this case, the control groups are confounded and it cannot be argued that they are unaffected by the merger. However, stylized facts and reduced-form regression show strong evidence against the national pricing hypothesis. Moreover, this specification serves as a simple test for this hypothesis. The reason being, in case of national pricing we expect to find no significant price differences between treatment and control group Disentanglement of concentration effects and efficiency gains The baseline scenario merely sums up efficiency gains and concentration effects, where the sign of the overall effect indicates which channel dominates. However, it does not allow disentangling the magnitude of price changes due to efficiency gains from those that stem from a change in market power. The subsequent paragraph describes three modifications of specifying treatment and control group to explore the effect of (the pass-through of) efficiency gains and concentration effects. First, insiders may have profited from global efficiency gains, which potentially are passed- 14

15 Table 3: Insider and Outsider per Region Municipality County Frequency Percent Frequency Percent None R1 or R2 3, R1 and R2 1, Total 6, through to consumers. In order to learn more on the magnitude of pass-through of global efficiency gains, we exclude all local markets from the control group, which contain either R1 or R2 (see Figure 1(b)). Intuitively, we now compare markets with pre-merger overlap to a control group of unaffected markets without insiders. In this setup, the price effect is only identified via the price reactions of competitors since we estimate price effects on the market-retailer-brand level and the insiders are not present in the control group. Thus, we compare prices of outsiders in treatment and control group, where we expect price effects if outsiders act strategically in response to the merger. Eliminating insiders from the control group abstracts from regional efficiency gains and allows to focus on efficiency gains at the national level. Recall that abstracting from regional efficiency gains such as increase in local bargaining power or cost savings from local transportation are not quite relevant for the considered merger as we have outlined in section 2.4. The reason being, R1 pursues a national purchasing strategy and distribution cost savings are internalized at the national level. Even though the specification excludes both insiders, the control group remains a good comparison for the treatment since (i) we have a sufficient number of cases (see table 3) and (ii) outsider O1 a close competitor of the same format with a similar business strategy is active in many of those regions. Moreover, graph shows that the parallel trend assumption holds. In this specification, price effects should to be lower compared to the baseline scenario since insiders potentially benefit from nation-wide efficiency gains and these might be (partly) passed on to consumers. Such global cost savings cancel out in the previous specification and now affect the treatment group solely. Second, we further adapt the definition of the control group to identify the upward-pricing pressure due to merger. In this specification, we change the control group in a way that it contains all local markets where at least one of the insiders exists (see Figure 1(c)), which is the opposite of the afore-mentioned case, in which control groups are defined as local markets with outsiders only. The specification identifies the pricing pressure due to an increase in market power since insiders are in both the treatment and the control group, global efficiency gains cancel out on both sides. This assumption hinges on the assumptions (i) that retailers employ a national bargaining strategy on the supply side instead of leaving negotiations to local merchants and (ii) there is equal pass-through in 15

16 treatment and control group. While we feel confident to assume the former, the latter is difficult to justify. However, if there are indeed local efficiency gains, then prices of the control group decrease after the merger, and our estimates are a lower bound. Finally, we investigate the pass-through of efficiency gains. In this specification, our control group consists of markets with outsiders only, while the treatment group contain markets in which one but not both of the merging parties were active before the merger (Figure 1(d)). Since, in this specification, we compare markets without overlap to a control group of markets without insiders, changes in local competition do not affect the estimates, and price changes stem from changes in efficiency only. 4.2 Empirical implementation For our baseline specification, we use a simple difference-in-differences (DiD) estimator to analyze the effects of the merger on regional consumer prices: ln(p igjt ) = α igj + θ post t MA g + δ t + [x gtβ + η it + ω kt ] + ε igjt (1) where ln(p igjt ) denotes the logarithmic product price set by retail chain i in regional market g (defined at the county level, e.g. Cologne), for brand j at quarterly time period t. We treat branded products of the merging retail chains like a product of one retailer in markets with pre-merger overlap of R1 and R2. Private labels sold by the target firm in re-branded stores will not enter the estimation since they disappear in the post-merger period. α igj is a retailer-market-product fixed effect, post t takes on a value of one in all post-merger periods, MA g is a dummy variable indicating regions affected by the merger and δ t denotes time dummies. In some specifications, we add the terms in brackets: x gt controls for time-varying demand heterogeneity at the regional level through changes in average income, population density and unemployment. η it denotes retail chain-time fixed effects which control for national-wide price changes across retailers. Note that these also capture any concentration and efficiency effects of the merger which do not vary across regions. ω kt controls for overall price changes in product categories (k) across time. 7 Finally, ε igjt is an error term. Our main coefficient of interest is the DiD-parameter θ which indicates differences between within-product changes in consumer prices across regions. Since our dependent variable is retailer-product-region specific, while our treatment indicator only varies across regions within years, we compute standard errors that are clustered at the region level. However, we also experiment with alternatives such as clustering at a lower level and boot- 7 To capture product-specific trends, we had to aggregate products j to product categories k since we do not observe purchases of all products in all regions. 16

17 strapping. As discussed in the previous subsection, we exclude counties which have no pre-merger overlap between R1 and R2 but share a border with a merger county or are located less than 15 kilometers away from a market directly affected by the merger, to ensure that our control group is not contaminated by indirect merger effects. We extend our baseline specification in several dimensions. First, we analyze heterogeneous effects across private label products and national brands by estimating the following specification: ln(p igjt ) = α igj + θ 1 post t MA g + θ 2 post t MA g PL j + δ t + [x gtγ + η it + ω kt ] + ε igjt (2) where PL j takes on a value of one if product j is sold under a private label. In this specification, θ 1 measures the treatment effect on products sold under national brand names, while θ 2 indicates differences between price effects on private label products relative to those for national brands. Next, we extend (1) to allow for heterogeneous retail chain-specific treatment effects: ln(p igjt ) = α igj + ϕ 1 post t MA g + ϕ 2 post t MA g DC i + δ t + [x gtπ + η it + ω kt ] + ε igjt (3) where DC i takes on a value of one for discounters and ϕ 2 measures differences in price changes between supermarkets and discounters. We also use a similar specification in which we replace DC by a dummy variable for insiders to distinguish between effects on merging parties and third nonmerging competitors. In another extension of our baseline model, we investigate whether price increases are more likely to occur in markets with high expected changes in concentration. For this purpose, we follow Dafny et al. (2012) and rely on the predicted change in the Herfindahl-Index (HHI) induced by the merger: simhhi gk = 2 Acqshare gk Tarshare gk (4) where Acqshare gk and Tarshare gk denote the pre-merger market shares for region g and product category k of acquirer and target, respectively. For instance, if acquirer and target have pre-merger market shares of 10% each, HHI, the sum of squared market shares, would be expected to change by We use this predicted change in concentration to test the hypothesis that price increases are more likely to occur in regional markets with substantial changes in retail concentration in the following equation: ln(p igjt ) = α igj + τ 1 post t MA g + τ 2 post t MA g simhhi gk + δ t + [x gtκ + η it + ω kt ] + ε igjt (5) 17

18 In this specification, τ 1 estimates the effect of the merger on prices that is independent of initial market shares of acquirer and target. τ 2 captures heterogeneity of treatment effects with respect to variation in expected regional retail market concentration. Finally, we analyse whether remedies imposed by the German cartel office, which required the sale of several retail stores to a competitor in regional markets with high pre-merger market shares of acquirer and target, had the desired effects. For this purpose, we estimate: ln(p igjt ) = α igj + λ 1 post t MA g + λ 2 post t MA g Rem g + δ t + [x gtν + η it + ω kt ] + ε igjt (6) where Rem g takes on a value of one if remedies were imposed in market g and MA g indicates regions affected by the merger which includes regions with and without remedies. Thus, λ 1 measures the estimated impact of the merger in non-remedy regions and λ 2 indicates whether there are different effects in regional markets in which remedies have been imposed. 5 Results This section presents the empirical findings. We first discuss the baseline results in section 5.1 where we analyze average and heterogeneous treatment effects by comparing relative price changes of the treatment group with the entire control group. In section 5.2 we present the results where we separately identify upward pressure pricing, downward pressure pricing and the net effect of both forces. 5.1 Baseline Results Table 4 presents the baseline results. Models 1-3 show results for the average treatment effect with different subset of controls. The estimated treatment effect of the merger ranges between 0.37% and 0.47% and is significant in all models. In addition to region-retailer-brand fixed effects and categorytime fixed effects, model 3 controls for retailer-time fixed effects and regional controls, such as median income, share of the households with at least one child, median age and population density. This will be the main specification from here on. Model 4 adds an interaction term with the predicted change in the HHI. The baseline effect of this specification remains positive and significant, but the predicted effect of the merger increases with the change in HHI, where this increase is statistically significant. To demonstrate the economic significance of this effect, we evaluate the effect at certain parameter values. The change in HHI is 0.5 at the maximum. This means the treatment effect ranges from 0.38% to 3.81% showing the importance of market concentration for the expected changes in prices. 18

19 Table 4: Baseline Results Model 1 Model 2 Model 3 Model 4 Model 5 Treat * *** *** *** * (0.0020) (0.0020) (0.0020) (0.0020) (0.0032) Treat HHI * (0.0407) Retailer-Time FE no yes yes yes yes Regional Controls no no yes yes yes Region all all all all remedy FEs for Region-Retailer-Brand, Category-Time, Time N R Standard errors in parentheses. Clustered at regional level. Significant at 1% ***, Significant at 5% **, Significant at 10% * Model 5 shows the results for remedy regions only. The estimated effect in remedy regions is larger than the average effect in specifications where remedy and non-remedy regions are included. This is consistent with the selection criteria of the remedy regions by the German cartel office. As the remedy regions consists of the more concentrated local markets, we expect that the merger would produce higher effects in those regions. Table 5 shows heterogeneous treatment effects. Model 6 interacts the treatment effect with a dummy variable for discount retailers (DC). We see that the positive effect on prices is fully driven by multi-line retailing formats and the interaction term with discounter formats exactly offsets the positive effect of multi-line retailers. The average treatment effect for multi-line retailers is 0.9% and highly significant. Model 9 extends the specification of Model 6 by an interaction term with the predicted change in HHI. Again we see that the baseline effect is positive and significant for multi-line retailers and is exactly offset by the effect for discounters. The interaction of multi-line retailers with the change in HHI is positive and significant. The treatment effect ranges between 0.78% and 6.38% at the maximum. The interaction of the discounter treatment effect with the change in HHI is negative and offsets the positive interaction term of the multi-line retailers. This is consistent with the findings that the average treatment effect is driven by the multi-line retailer as shown in model 6. Model 7 shows the average treatment effect on prices of national brands versus private label. Controlling for separate effects on PL, we estimate a higher coefficient for Treat compared to model 3. The interaction term between treat and PL is insignificant indicating that prices of PL change in 19

20 Table 5: Heterogeneous Treatment Effects: Private Labels (PL) & Discounters (DC) Model 6 Model 7 Model 8 Model 9 Model 10 Treat *** ** *** *** * (0.0025) (0.0030) (0.0033) (0.0026) (0.0020) Treat DC *** ** ** (0.0033) (0.0056) (0.0034) Treat PL (0.0037) (0.0046) Treat PL DC (0.0066) Treat HHI ** ** (0.0512) (0.0611) Treat DC HHI (0.0696) Treat PL HHI (0.0856) FEs for Region-Retailer-Brand, Retailer-Time, Category-Time, Time + Regional Controls N R Standard errors in parentheses. Clustered at regional level. Significant at 1% ***, Significant at 5% **, Significant at 10% * a similar way as NB. The negative interaction term with private labels in model 7 might be driven by the fact that they are mostly sold by discounters, for which we find no statistically significant effect. To disentangle private label price effects from discounter price effects, we use a triple interaction of private labels and discounters. Indeed, model 8 shows that within multi-line retailers, prices change similarly for NB and PL while the difference between discounters and supermarkets does not differ by much across the two product categories. This indicates that the negative interaction term in model 7 is indeed driven by the fact that most private labels are sold by discounters, which do not change their prices significantly in response to the merger. Model 10 shows results from a specification that is similar to model 7 but includes an interaction term between treat and DeltaHHI. In this model, the variable treat measures the joint average baseline treatment effect for private labels and national brands. However, we allow for different effects for private labels and national brands with a change in HHI. As before, we see that national brand prices increase with the expected change in HHI, and this effect is statistically significant. However, the interaction term of change in HHI and private labels is negative and offsets the positive effect for NB almost completely. Even though not being significant, it indicates that private label prices might not react as sensitive as national brands to changes in concentration. 20

21 5.2 UPP, DPP, and Net Effect Model 11 in Table 6 identifies price effects net of national-wide efficiency gains in merging retailers by restricting the control group to regions in which exactly one of the merging retailers was active. Intuitively, insiders in both treatment and control group can benefit from lower input costs due to the merger. If these cost savings are passed-on to consumers, outsiders in both treatment and control group are likely to adjust their prices as well. However, market concentration changes in the treatment group only. As expected, there a larger treatment effect in this specification than in our baseline model. Prices increase, on average, by 0.526% due to the increase in market concentration. Model 12 identifies the net effect of upward pressure pricing and downward pressure pricing. In this specification, the treatment group consist of outsiders in markets with pre-merger overlap of acquirer and target while the control group includes outsiders in markets without any presence of acquirer or target. Differences between treatment and control therefore potentially stem from both changes in market concentration and from outsiders reaction to efficiency gains of insiders. We find that there is no significant change in prices indicating that outsiders response to insiders efficiency gains (which do not affect estimates in model 11) are an important determinant of post-merger prices. In model 13, we use the same control group as in model 12, but the treatment group consists of markets in which only one of the merging retailers was active prior to the merger. Here the retailers in the treatment group operate in a market where the parties potentially experience a reduction in costs which might induce price changes from both insiders and outsiders, but there is no concentration effect. This allows us to identify downward pressure on prices due to nationalwide efficiency gains. The treatment effect is negative and slightly non-significant on average. We would expect that efficiency gains are not realized immediately, but price effects due to a higher concentration should should occur right after the merger and remain relatively constant in postmerger periods. Furthermore, we would expect that the pass through of lower costs to the consumer in form of lower prices is higher in competitive markets (add source, e.g., Boone Indicator). Model 14 and 15 in Table 6 correspond to model 11 and 12, respectively, taking into account time specific treatment effects. As expected, the price effects that are due to concentration are immediately realized, while remaining relatively stable afterwards (model 14). In the specification of model 15 the statistical and economical significance of the treatment effect is increasing with time passed after the merger. The effect is non-significant in the first half of 2009 while being three times as large and significant in the second half of Model 14 tests that the price increases are not statistically significant from each other, but in model 15 the effect is significantly higher in absolute terms in the 21

22 Table 6: UPP, DPP and Net Effect Model 11 Model 12 Model 13 Model 14 Model 15 Effect UPP Net DPP UPP DPP Treat *** ( ) ( ) ( ) Treat * ( ) ( ) Treat * ( ) ( ) Treat ** ( ) ( ) Treat ** ** ( ) ( ) FEs for Region-Retailer-Brand, Category-Time,Time + Regional Controls N Standard errors in parentheses * p 0.10, ** p 0.05, *** p 0.01 second half of 2010 compared to the first half in Table 7 replicates Table 6 but includes interactions interactions between HHI and the baseline effect. We estimate this model to test if the pass-through of efficiency gains is higher in more competitive markets. Even though in model 16 the coefficient of the interaction term is non-significant, the combined marginal effect for any value of HHI is significantly different from zero. An alternative specification (not shown here) where the treatment effect is interacted with a dummy variable, that takes on the value of 1 if the HHI is larger than the median, shows that there is a significantly larger effect in less concentrated markets. Model 17 shows that the net effect is significantly positive for highly concentrated markets. In model 18 we see that the baseline effect of the downward pressure on prices is significantly negative, and the pass-through of the efficiency gains is larger in more competitive markets, which is consistent with economic intuition. Model 19 and 20 include time specific treatment dummies for upward pressure and downward pressure on prices interacted with HHI, which are not shown in the table Robustness Checks To be done. 8 Note even though model 19 shows different base effects for the time specific treatment effect this does not imply different treatment effects. We saw in Table 6 model 14 that the average effect is similar in each time period 22

23 Table 7: UPP, DPP, Net Effect and HHI Model 16 Model 17 Model 18 Model 19 Model 20 Effect UPP Net DPP UPP DPP Treat * ** ( ) ( ) ( ) Treat HHI * *** ( ) ( ) ( ) Treat ( ) ( ) Treat * ( ) ( ) Treat * *** ( ) ( ) Treat * *** ( ) ( ) FEs for Region-Retailer-Brand, Category-Time,Time + Regional Controls N Standard errors in parentheses * p 0.10, ** p 0.05, *** p 0.01 All models include the baseline effect of HHI without interaction Model 4 and 5 include interactions with HHI for each time specific treatment effect 6 Conclusion In this paper we analyze the effect of a retail merger on consumer prices in Germany. Exploiting regional variation of prices and the fact that some regions are affected by the merger and some are not, which then serve as the control group, we are able to identify a causal effect on prices due to the retail merger. We find a positive effect of the merger on prices which increased on average by 0.45% relative to the control group. The estimates effect increases with predicted changes in market concentration and amount to 3.8% in markets with the highest pre-merger market share of acquirer and target. We show that the treatment effect is driven by multi-line retailers and that the average price reactions are larger in remedy regions where the acquirer had to sell the target stores to an outsider. We also provide evidence that efficiency gains partly offset price increases due to market power in some regions. First, prices do not rise significantly in markets with no change in market concentration. Second, price increases are most significant when we use a control group that captures responses to changes in efficiency of the merging parties that are common across regions. References Allain, Marie-Laure, Claire Chambolle, Stephane Turolla, and Sofia Villas-Boas, The 23

24 impact of retail mergers on food prices: evidence from France, Argentesi, Elena, Paolo Buccirossi, Roberto Cervone, Tomaso Duso, and Alessia Marrazzo, The effect of retail mergers on variety: An ex-post evaluation, Technical Report, University of Bologna Ashenfelter, Orley C, Daniel Hosken, and Matthew C Weinberg, Did Robert Bork Understate the Competitive Impact of Mergers? Evidence from Consummated Mergers, Technical Report, National Bureau of Economic Research Bulow, Jeremy I and Paul Pfleiderer, A note on the effect of cost changes on prices, Journal of political Economy, 1983, 91 (1), Bundeskartellamt, Beschluss: B FA-27/05, Fusion: EDEKA/Spar, Bonn: Bundeskartellamt, 2005., Beschluss B 2333/07, Fusion: Edeka/Tengelmann, Bonn: Bundeskartellamt, 2007., Beschluss B247250Fa52/10, Fusion: Edeka/Trinkgut, Bonn: Bundeskartellamt, 2010., Pressemeldung des Bundeskartellamtes vom : Bundeskartellamt bestätigt Durchsuchung von Einzelhandelsunternehmen und Markenartikelherstellern wegen des Verdachts abgestimmter Endverbraucherpreise, Bonn: Bundeskartellamt, 2010., Beschluss B 2-96/14, Fusion: Edeka/Tengelmann, Bonn: Bundeskartellamt, 2014., Sektoruntersuchung Lebensmitteleinzelhandel. Darstellung und Analyse der Strukturen und des Beschaffungsverhaltens auf den Märkten des Lebensmitteleinzelhandels in Deutschland, Bonn: Bundeskartellamt, Chipty, Tasneem and Christopher M Snyder, The role of firm size in bilateral bargaining: A study of the cable television industry, Review of Economics and Statistics, 1999, 81 (2), Competition Commission, The Supply of Groceries in the UK Market Investigation, London: Stationery Office, Dafny, Leemore, Mark Duggan, and Subramaniam Ramanarayanan, Paying a premium on your premium? Consolidation in the US health insurance industry, American Economic Review, 2012, 102 (2), Dobson, P. W. and M. Waterson, Countervailing Power and Consumer Prices, Economic Journal, 1997, 107 (441),

25 Dobson, Paul W and Michael Waterson, Chain-Store Pricing Across Local Markets, Journal of Economics & Management Strategy, 2005, 14 (1), European Commission, Decision: IV/M.1221, Rewe/Meinl, Brussels: European Commission, Galbraith, J. K., Countervailing Power, American Economic Review, 1954, 44 (2), 1 6. Gaudin, Germain, Pass-through, vertical contracts, and bargains, Economics Letters, 2016, 139, 1 4. Haucap, Justus, Ulrich Heimeshoff, Gordon J. Klein, Dennis Rickert, and Christian Wey, Die Bestimmung von Nachfragemacht im Lebensmitteleinzelhandel: Theoretische Grundlagen und empirischer Nachweis, WUW: Wirtschaft und Wettbewerb, 2014, 64 (10), Hosken, Daniel S, Luke Olson, and Loren Smith, Do retail mergers affect competition? Evidence from grocery retailing, Federal Trade Commission, Bureau of Economics, Working Paper, 2012, (313). Houde, Jean-Francois, Spatial Differentiation and Vertical Mergers in Retail Markets for Gasoline, American Economic Review, 2012, 102 (5), Inderst, R., Implications of Buyer Power and Private Labels on Vertical Competitionand Innovation, Inderst, Roman and Tommaso M. Valletti, Buyer Power and the Waterbed Effect*, The Journal of Industrial Economics, 2011, 59 (1), Katz, M. L., The Welfare Effects of Third-Degree Price Discrimination in Intermediate Good Markets, American Economic Review, 1987, pp Rodrigues, J., Buyer Power and Pass-Through of Large Retailing Groups in the Portuguese Food Sector, Lisbon: Autoridade da Concorrencia, Scheffman, David T and Pablo T Spiller, Buyers Strategies, Entry Barriers, and Competition, Economic Inquiry, 1992, 30 (3), von Ungern-Sternberg, Thomas, Countervailing power revisited, International Journal of Industrial Organization, 1996, 14 (4), Weyl, E Glen and Michal Fabinger, Pass-through as an economic tool: Principles of incidence under imperfect competition, Journal of Political Economy, 2013, 121 (3),

26 7 Appendix 7.1 Graphs Baseline 26

27 7.1.2 NET 27

28 7.1.3 UPP 28

29 7.1.4 DPP 29