Supermarket pricing-strategies4995 (1)
Upcoming SlideShare
Loading in...5

Supermarket pricing-strategies4995 (1)







Total Views
Views on SlideShare
Embed Views



0 Embeds 0

No embeds



Upload Details

Uploaded via as Adobe PDF

Usage Rights

© All Rights Reserved

Report content

Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

  • Full Name Full Name Comment goes here.
    Are you sure you want to
    Your message goes here
Post Comment
Edit your comment

Supermarket pricing-strategies4995 (1) Supermarket pricing-strategies4995 (1) Document Transcript

  • informs ®Vol. 27, No. 5, September–October 2008, pp. 811–828issn 0732-2399 eissn 1526-548X 08 2705 0811 doi 10.1287/mksc.1080.0398 © 2008 INFORMS Supermarket Pricing Strategies Paul B. Ellickson Department of Economics, Duke University, Durham, North Carolina 27708, Sanjog Misra William E. Simon School of Business Administration, University of Rochester, Rochester, New York 14627, M ost supermarket firms choose to position themselves by offering either everyday low prices (EDLP) across several items or offering temporary price reductions (promotions) on a limited range of items. While this choice has been addressed from a theoretical perspective in both the marketing and economic literature, relatively little is known about how these decisions are made in practice, especially within a competitive envi- ronment. This paper exploits a unique store level data set consisting of every supermarket operating in the United States in 1998. For each of these stores, we observe the pricing strategy the firm has chosen to follow, as reported by the firm itself. Using a system of simultaneous discrete choice models, we estimate each store’s choice of pricing strategy as a static discrete game of incomplete information. In contrast to the predictions of the theoretical literature, we find strong evidence that firms cluster by strategy by choosing actions that agree with those of its rivals. We also find a significant impact of various demographic and store/chain characteristics, providing some qualified support for several specific predictions from marketing theory. Key words: EDLP; promotional pricing; positioning strategies; supermarkets; discrete games History: Received: March 22, 2006; accepted: February 27, 2008; processed by David Bell.1. Introduction is likely to shop, we know relatively little about howWhile firms compete along many dimensions, pricing pricing strategies are chosen by retailers. There arestrategy is clearly one of the most important. In many two primary reasons for this. First, these decisionsretail industries, pricing strategy can be characterized are quite complex: managers must balance the pref-as a choice between offering relatively stable prices erences of their customers and their firm’s own capa-across a wide range of products (often called every- bilities against the expected actions of their low pricing) or emphasizing deep and frequent Empirically modeling these actions (and reactions)discounts on a smaller set of goods (referred to as requires formulating and then estimating a complexpromotional or PROMO pricing). Although Wal-Mart discrete game, an exercise which has only recentlydid not invent the concept of everyday low pricing, become computationally feasible. The second is thethe successful use of everyday low pricing (EDLP) lack of appropriate data. While scanner data setswas a primary factor in their rapid rise to the top have proven useful for analyzing consumer behavior,of the Fortune 500, spawning a legion of followers they typically lack the breadth necessary for tack-selling everything from toys (Toys R Us) to building ling the complex mechanics of inter-store competi-supplies (Home Depot). In the 1980s, it appeared that tion.1 The goal of this paper is to combine newlythe success and rapid diffusion of the EDLP strategy developed methods for estimating static games withcould spell the end of promotions throughout much a rich, national data set on store level pricing poli-of retail. However, by the late 1990s, the penetration cies to identify the primary factors that drive pricingof EDLP had slowed, leaving a healthy mix of firms behavior in the supermarket industry.following both strategies, and several others employ- Exploiting the game theoretic structure of ouring a mixture of the two. approach, we aim to answer three questions that Not surprisingly, pricing strategy has proven to be have not been fully addressed in the existing liter-a fruitful area of research for marketers. Marketing ature. First, to what extent do supermarket chainsscientists have provided both theoretical predictions tailor their pricing strategies to local market condi-and empirical evidence concerning the types of con- tions? Second, do certain types of chains or storessumers that different pricing policies are likely toattract (e.g. Lal and Rao 1997, Bell and Lattin 1998). 1 Typical scanner data usually reflect decisions made by only a fewWhile we now know quite a bit about where a person stores in a limited number of markets. 811
  • Ellickson and Misra: Supermarket Pricing Strategies812 Marketing Science 27(5), pp. 811–828, © 2008 INFORMShave advantages when it comes to particular pricing from their rivals, stores choose strategies that match.strategies? Finally, how do firms react to the expected This finding is in direct contrast to existing theoreticalactions of their rivals? We address each of these ques- models that view pricing strategy as a form of dif-tions in detail. ferentiation, providing a clear comparative static that The first question naturally invites a market pull future pricing models must address.driven explanation in which consumer demographics Our paper makes both substantive and method-play a key role in determining which pricing strategy ological contributions to the marketing literature. Onfirms choose. In answering this question, we also the substantive front, our results offer an in-depthaim to provide additional empirical evidence that will look at the supermarket industry’s pricing practices,inform the growing theoretical literature on pricing delineating the role of three key factors (demand,related games. Since we are able to assess the impact supply, and competition) on the choice of pricingof local demographics at a much broader level than strategy. We provide novel, producer-side empiri-previous studies, our results provide more conclusive cal evidence that complements various consumer-sideevidence regarding their empirical relevance. models of pricing strategy. In particular, we find qual- The second question concerns the match between ified support for several claims from the literaturea firm’s strategy and its chain-specific capabilities. on pricing demographics, including Bell and Lattin’sIn particular, we examine whether particular pricing (1998) model of basket size and Lal and Rao’s (1997)strategies (e.g., EDLP) are more profitable when firms positioning framework, while at the same time high-make complementary investments (e.g. larger stores lighting the advantages of chain level investment.and more sophisticated distribution systems). The Our focus on competition also provides a structuralempirical evidence on this matter is scant—this is the complement to Shankar and Bolton’s (2004) descrip-first paper to address this issue on a broad scale. Fur- tive study of price variation in supermarket scannerthermore, because our data set includes all existing data, which emphasized the role of rival actions. Oursupermarkets, we are able to exploit variation both most significant contribution, however, is demonstrat-within and across chains to assess the impact of store ing that stores in a particular market do not use pric-and chain level differences on the choice of pricing ing strategy as a differentiation device but insteadstrategy. coordinate their actions. This result provides a direct Finally, we address the role of competition posed challenge to the conventional view of retail compe-in our third question by analyzing firms’ reactions tition, opening up new and intriguing avenues forto the expected choices of their rivals. In particular, future theoretical research. Our econometric imple-we ask whether firms face incentives to distinguish mentation also contributes to the growing literature inthemselves from their competitors (as in most models marketing and economics on the estimation of staticof product differentiation) or instead face pressures discrete games, as well as the growing literature onto conform (as in network or switching cost mod- social interactions.2 In particular, our incorporation ofels)? This question is the primary focus of our paper multiple sources of private information and our con-and the feature that most distinguishes it from earlier struction of competitive beliefs are novel additions towork. these emerging literatures. Our results shed light on all three questions. First, The rest of the paper is organized as follows. Sec-we find that consumer demographics play a signifi- tion 2 provides an overview of the pricing landscape,cant role in the choice of local pricing strategies: firms explicitly defining each strategy and illustrating thechoose the policy that their consumers demand. Fur- importance of local factors in determining store levelthermore, the impact of these demographic factors decisions. Section 3 introduces our formal model ofis consistent with both the existing marketing liter- pricing strategy and briefly outlines our estimationature and conventional wisdom. For example, EDLP approach. Section 4 describes the data set. Section 5is favored in low income, racially diverse markets, provides the details of how we implement the model,while PROMO clearly targets the rich. However, a key including the construction of distinct geographic mar-implication of our analysis is that these demographic kets, the selection of covariates, our two-step estima-factors act as a coordinating device for rival firms, tion method, and our identification strategy. Section 6helping shape the pricing landscape by defining anequilibrium correspondence. Second, we find that 2 Recent applications of static games include technology adop-complementary investments are key: larger stores tion by internet service providers (Augereau et al. 2006), prod-and vertically integrated chains are significantly more uct variety in retail eyewear (Watson 2005), location of ATMlikely to adopt EDLP. Finally, and most surprisingly, branches (Gowrisankaran and Krainer 2004), and spatial differenti- ation among supermarkets (Orhun 2005), discount stores (Zhu et al.we find that stores competing in a given market have 2005), and video stores (Seim 2006). Structural estimation of socialincentives to coordinate their actions. Rather than interactions is the focus of papers by Brock and Durlauf (2002),choosing a pricing strategy that distinguishes them Bayer and Timmins (2006), and Bajari et al. (2005).
  • Ellickson and Misra: Supermarket Pricing StrategiesMarketing Science 27(5), pp. 811–828, © 2008 INFORMS 813provides our main empirical results and discusses Table 1 Descriptive Statisticstheir implications. Section 7 concludes with directions Variable Obs Mean Std. dev. Min. Maxfor future research. Strategy EDLP 17 388 0 28 0 45 0 12. The Supermarket Pricing HYBRID PROMO 17 388 17 388 0 38 0 34 0 48 0 47 0 0 1 1 Landscape MSA characteristics Size (sq. miles) 333 1 868 31 1 943 99 46 40 11 229 62.1. Pricing Strategy Choices Density (pop ’000 333 10 42 9 62 0 91 49 06Competition in the supermarket industry is a complex per sq. mile)phenomenon. Firms compete across the entire retail Avg. food expenditure 333 663 64 1 201 37 16 04 9 582 09and marketing mix, enticing customers with an attrac- ($ ’000)tive set of products, competitive prices, convenient Market variables Median household size 8 000 2 66 0 35 1 32 5 69locations, and a host of other services, features, and Median HH income 8 000 35 255 59 9 753 95 18 109 60 81 954 60promotional activities. In equilibrium, firms choose Proportion Black 8 000 0 08 0 14 0 00 0 97the bundle of services and features that maximize Proportion Hispanic 8 000 0 06 0 13 0 00 0 98 Median vehicles in HH 8 000 2 12 0 33 0 56 3 37profits, conditional on the types of consumers theyexpect to serve and their beliefs about the actions of Chain/store characteristics Vertically integrated 17 388 0 51 0 50 0 00 1 00their rivals. A supermarket’s pricing strategy is a key Store size (sqft ’000) 17 388 28 99 16 34 2 00 250 00element in this multidimensional bundle. Independent store 17 388 0 23 0 42 0 00 1 00 The majority of both marketers and practitioners Number of stores 804 390 15 478 45 1 00 1 399 00 in chainframe a store’s pricing decision as a choice betweenoffering everyday low prices or deep but tempo-rary discounts, labeling the first strategy EDLP and which asked individual store managers to choosethe second PROMO (Table 1).3 4 Not surprisingly, which of the following categories best described theirthe simple EDLP-PROMO dichotomy is too narrow store’s pricing policy:to adequately capture the full range of firm behav- • Everyday Low Price (EDLP): Little reliance onior. In practice, firms can choose a mixture of EDLP promotional pricing strategies such as temporaryand PROMO, varying either the number of categories price cuts. Prices are consistently low across thethey put on sale or changing the frequency of sales board, throughout all packaged food departments.across some or all categories of products. Practitioners • Promotional (Hi-Lo) Pricing: Heavy use of spe-have coined a term for these practices—hybrid pric- cials, usually through manufacturer price breaks oring. What constitutes HYBRID pricing is necessarily special deals.subjective, depending on an individual’s own beliefs • Hybrid EDLP/Hi-Lo: Combination of EDLP andregarding how much price variation constitutes a Hi-Lo pricing strategies.departure from pure EDLP. Both the data and defini- According to Trade Dimensions, the survey wastions used in this paper are based on a specific store designed to allow for a broad interpretation of thelevel survey conducted by Trade Dimensions in 1998, HYBRID strategy, as they wanted it to capture devia- tions along either the temporal (i.e., number of sales3 This is clearly a simplification—a supermarket’s pricing policy per year) or category based dimensions (i.e., numberis closely tied to its overall positioning strategy. Pricing strategies of categories on deal). We believe that pricing strat-are typically chosen to leverage particular operational advantages egy is best viewed as a continuum, with pure EDLPand often have implications for other aspects of the retail mix. Forexample, successful implementation of EDLP may involve offering (i.e., constant margins across all categories) on onea deeper and narrower product line than PROMO, allowing firms end and pure PROMO (i.e. frequent sales on all cate-to exploit scale economies (in particular categories), reduce their gories) at the other. This data set represents a coarseinventory carrying costs, and lower their advertising expenses. On discretization of that continuum.the other hand, PROMO pricing gives firms greater flexibility inclearing overstock, allows them to quickly capitalize on deep man-ufacturer discounts, and facilitates the use of consumer loyalty pro- 2.2. Supermarket Pricing: A Closer Lookgrams (e.g. frequent shopper cards). In other words, the choice of Without observing data on individual stores, it mightpricing strategy is more than just how prices are set: it reflects the be tempting to conclude that all pricing strategies areoverall positioning of the store. This paper focuses on the pricing determined at the level of the chain. While there aredimension alone, taking the other aspects of the retail mix as given. certainly incentives to choose a consistent policy, theWhile this is limiting, modeling the entire retail mix is beyond thescope of this paper. data reveals a remarkable degree of local heterogene-4 Note that we focus on the choice of pricing strategy and abstract ity. To examine the issue more closely, we focus in onaway from issues related to more tactical decisions about how prices a single chain in a single market: the Pathmark chainare (or should be) set (see e.g., Kumar and Rao 2006). in New Jersey. Figure 1 shows the spatial locations of
  • Ellickson and Misra: Supermarket Pricing Strategies814 Marketing Science 27(5), pp. 811–828, © 2008 INFORMSFigure 1 Pathmark Stores in New Jersey Table 3 Pricing Strategy by Firm Type % EDLP % HYBRID % PROMO “Large” firms: 41.0 Chain 33 37 30 Vertically integrated 35 36 29 Large store size 32 38 30 Many checkouts 31 39 30 40.5 “Small” firms: Independent 22 28 50 Not vertically integrated 21 32 47 Small store size 23 26 52 40.0 Few checkouts 22 26 52 small chains, using four alternative definitions of 39.5 “large” and small.5 While large chains seem evenly EDLP distributed across the strategies and “small” chains HYBRID seem to favor PROMO, firm size is not the primary PROMO 39.0 determinant of pricing strategy. –75.5 –75.0 –74.5 –74.0 The second noteworthy feature of the Pathmark data is that even geographically proximate stores adopt quite different pricing strategies. While there isevery Pathmark store in New Jersey, along with its some clustering at the broader spatial level (e.g. northpricing strategy. Two features of the data are worth versus south New Jersey), the extent to which theseemphasizing. We address them in sequence. strategies are interlaced is striking. Again, looking First, Pathmark does not follow a single strategy beyond Pathmark and New Jersey confirms that thisacross its stores: 42% of the stores use PROMO pric- within-chain spatial heterogeneity is not unique toing, 33% follow EDLP, and the remaining 25% use this particular example: while some chains clearlyHYBRID. The heterogeneity in pricing strategy favor a consistent strategy, others appear quiteobserved in the Pathmark case is not specific to this responsive to local factors. Broadly speaking, theparticular chain. Table 2 shows the store level strate- data reveal only a weak relationship between geog-gies chosen by the top 15 U.S. supermarkets (by raphy and pricing strategy. While southern chainstotal volume) along with their total store counts. Aswith Pathmark, the major chains are also surprisingly such as Food Lion are widely perceived to favorheterogeneous. While some firms do have a clear EDLP and Northeastern chains like Stop & Shop arefocus (e.g. Wal-Mart, H.E. Butt, Stop & Shop), oth- thought to prefer PROMO, regional variation doesers are more evenly split (e.g. Lucky, Cub Foods). not capture the full story. Table 4 shows the per-This pattern extends to the full set of firms. Table 3 cent of stores that choose either EDLP, HYBRID, orshows the pricing strategies chosen by large and PROMO pricing in eight geographic regions of the United States. While PROMO pricing is most popular in the Northeast, Great Lakes, and central Southern Table 2 Pricing Strategies of the Top 15 Supermarkets regions, it is far from dominant, as both the EDLP and Firm Stores % PROMO % HYBRID % EDLP HYBRID strategies enjoy healthy shares there as well. EDLP is certainly favored in the South and Southeast, Kroger 1 399 47 40 13 Safeway 1 165 52 43 5 but PROMO still draws double digit shares in both Albertson’s 922 11 41 48 regions. This heterogeneity in pricing strategy can Winn-Dixie 1 174 3 30 67 be illustrated using the spatial structure of our data Lucky 813 35 38 27 set. Figure 2 plots the geographic location of every Giant 711 29 60 11 store in the United States, along with their pricing Fred Meyer 821 22 60 18 Wal-Mart 487 1 26 73 Publix 581 13 71 16 5 The four definitions of firm size are: chain/independent, vertically Food Lion 1 186 2 12 86 integrated and not, large/small store, and many/few checkouts. A&P 698 55 30 15 A chain is defined as having 11 or more stores, while an indepen- H.E. Butt 250 1 3 96 dent has 10 of fewer. Vertically integrated means the firm operates Stop & Shop 189 50 43 7 its own distribution centers. Large versus small store size and many Cub foods 375 26 34 40 versus few checkouts are defined by the upper and lower quartiles Pathmark 135 42 25 33 of the full store level census.
  • Ellickson and Misra: Supermarket Pricing StrategiesMarketing Science 27(5), pp. 811–828, © 2008 INFORMS 815 Table 4 Pricing Strategies by Region Table 5 Local Factors Region % PROMO % HYBRID % EDLP EDLP HYBRID PROMO West Coast 39 39 22 Local demographics Northwest 32 51 17 Median household 2 84 (0.331) 2 81 (0.337) 2 80 (0.329) South West 20 48 32 size South 32 25 43 Median household 34,247 (14,121) 36,194 (15,121) 36,560 (16,401) income Southern Central 45 27 28 Median vehicles 2 12 (0.302) 2 13 (0.303) 2 09 (0.373) Great Lakes 54 29 17 in HH North East 40 37 23 Median age 35 4 (4.59) 35 8 (4.98) 35 7 (4.25) South East 23 37 40 Proportion Black 0 128 (0.182) 0 092 (0.158) 0 110 (0.185) Proportion Hispanic 0 078 (0.159) 0 073 (0.137) 0 070 (0.135) Strategies of rivalsstrategy. As is clear from the three panels correspond- Percent of rivals using 49 (31) 49 (25) 52 (23)ing to each pricing strategy, there is no obvious pat- same strategytern: all three strategies exhibit quite uniform cover- Note. The main numbers in each cell are means, standard deviations are inage. Taken together, these observations suggest look- elsewhere for the primary determinants of pricing strategy. We turn next to the role of market demo- graphics and then to the nature and degree of com-Figure 2 Spatial Distribution of Store Pricing Strategy petition. Table 5 contains the average demographic char- EDLP stores acteristics of the local market served by stores of each type.6 PROMO pricing is associated with smaller households, higher income, fewer automobiles per capita, and less racial diversity, providing some ini- tial support for Bell and Lattin’s (1998) influen- tial model of basket size.7 However, the differences in demography, while intuitive, are not especially strong. This does not mean that demographics are irrelevant, but rather that the aggregate level patterns linking pricing strategy and demographics are not overwhelming. Isolating the pure impact of demo- graphic factors will require a formal model, which we HYBRID stores provide below. The final row of Table 5 contains the share of rival stores in the competing market that employ the same strategy as the store being analyzed. Here we find a striking result: 50% of a store’s rivals in a given loca- tion employ the same pricing strategy as the focal store. Competitor factors also played a lead role in the work of Shankar and Bolton (2004), which ana- lyzed pricing variability in supermarket scanner data. In particular, they note that “what is most striking, however, is that the competitor factors are the most dominant determinants of retailer pricing in a broad PROMO stores framework that included several other factors” (p. 43). Even at this rather coarse level of analysis, the data 6 Roughly corresponding to areas the size of a ZipCode, these “local markets” are defined explicitly in §5.2. 7 Bell and Lattin (1998) find that the most important features of shopping behavior can be captured by two interrelated choices: basket size (how much you buy) and shopping frequency (how often you go). They suggest that large or fixed basket shoppers (i.e. those who buy more and shop less) will more sensitive to the overall basket price than those who shop frequently and will therefore prefer EDLP pricing to PROMO. They present empirical evidence that is consistent with this prediction.
  • Ellickson and Misra: Supermarket Pricing Strategies816 Marketing Science 27(5), pp. 811–828, © 2008 INFORMSreveal that most stores choose similar pricing strate- MSA). Before proceeding further, we must introducegies to their rivals. This pattern clearly warrants a some additional notation. Stores belonging to a givenmore detailed investigation and is the focus of our chain c = 1 C, that are located in a local mar-structural model. ket lm = 1 Lm , in an MSA m = 1 M, will be l l Stepping back, three key findings emerge. First, su- indexed using icm = 1 Nc m . The total numberpermarket chains often adopt heterogeneous pricing of stores in a particular chain in a given MSA is Lm lstrategies, suggesting that demand related forces can Ncm = lm =1 Nc m , while the total number of storessometimes outweigh the advantages of chain level in that chain across all MSAs is Nc = M Ncm . In m=1specialization. Second, local market factors play a key each local market, chains select a pricing strategyrole in shaping demand characteristics. Finally, any (action) a from the three element set K = E H P ,empirical analysis of pricing strategy must address where E ≡ EDLP, H ≡ HYBRID, and P ≡ PROMO.the role of competition. While investigating the role If we observe a market lm containing N lm = C Nc m l c=1of market demographics and firm characteristics is players for example, the set of possible action pro-not conceptually difficult, quantifying the structural lm files is then Alm = E H P Nc with generic elementimpact of rival pricing strategies on firm behavior alm = a1 a2 aicm l aNclm . The vector of actions ofrequires a formal game theoretic model of pricing lbehavior that accounts for the simultaneity of choices. store icm ’s competitors is denoted a−icm = a1l aicm −1 lIn the following section, we embed pricing strategy aic +1 lm aNc lm .in a discrete game that accommodates both local In a given market, a particular chain’s state vec- m mdemographics and the strategies of rival firms. We tor is denoted sc ∈ Sc , while the state vector for the m m Ncm mthen estimate this model using the two-step approach market as a whole is s m = s1 sNc ∈ c=1 Sc . Thedeveloped by Bajari et al. (2005). state vector s m is known to all firms and observed by the econometrician. It describes features of the mar- ket and characteristics of the firms that we assume3. A Strategic Model of are determined exogenously. For each firm, there are Supermarket Pricing also three unobserved state variables (correspondingA supermarket’s choice of pricing strategy is natu- to the three pricing strategies) that are treated asrally framed as a discrete game between a finite set private information of the firm. These unobservedof players. Each firm’s optimal choice is determined state variables are denoted icm aicm , or more com- l lby the underlying market conditions, its own charac- pactly icm , and represent firm specific shocks to the lteristics and relative strengths, as well as its expecta- profitability of each strategy. The private informa-tions regarding the actions of its rivals. Ignoring strate- tion assumption makes this a game of incompletegic expectations, pricing strategy could be modeled as or asymmetric information (e.g. Harsanyi 1973) anda straightforward discrete choice problem. However, the appropriate equilibrium concept one of Bayesiansince firms condition their strategies on their beliefs Nash Equilibrium (BNE). For any given market, theregarding rivals’ actions, this discrete choice must be l ’s are assumed to be i.i.d. across firms and actions, icmmodeled as a system of simultaneous equations. In and drawn from a distribution f icm that is known lour framework, firms (i.e., supermarket chains8 ) make to everyone, including the econometrician.a discrete choice of pricing strategy, selecting among Firms maximize store-level profits, choose pricingthree alternatives: everyday low pricing, promotional strategies independently across stores. In market lm ,pricing, and a hybrid strategy. While there is clearly the profit earned by store ic is given bya role for dynamics in determining an optimal pric-ing policy, we assume that firms act simultaneously in l icm = l icm s m aicm a−icm + l l l icm (1)a static environment, taking entry decisions as given. where icm is a known and deterministic function of lThis static treatment of competition is not altogether states and actions (both own and rival’s). Since theunrealistic since these pricing strategies involve sub- ’s are private information, each firm’s decision rulestantial store level investments in communication and aicm = dicm s m icm is a function of the common state l l lpositioning related costs that are not easily reversed.9 vector and its own , but not the private information We assume that competition takes place in “local” of its rivals. From the perspective of both its rivalsmarkets, each contained in a global market (here, an and the econometrician, the probability that a given8 firm chooses action k conditional on the common state Henceforth, we will use chains and firms interchangeably.9 vector is then given by As discussed above, pricing decisions are relatively sunk, due tothe positioning costs associated with conveying a consistent store- Picm aicm = k = l l 1 dicm s m l l =k f l d l (2)level message to a group of repeat customers. Furthermore, since icm icm icmthis is not an entry game, we are not particularly concerned aboutthe possibility of ex post regret that can sometimes arise in static where 1 dicm s icm = k is an indicator function equal l l lmgames (Einav 2003). to 1 if store ic chooses action k and 0 otherwise.
  • Ellickson and Misra: Supermarket Pricing StrategiesMarketing Science 27(5), pp. 811–828, © 2008 INFORMS 817We let Plm denote the set of these probabilities for a of three assumptions required for identification (wegiven local market. Since the firm does not observe its discuss our identification strategy in detail in §5.7).competitors actions prior to choosing its own action, In addition, we have assumed that the private infor- lit makes decisions based on its expectations. The mation available to store icm (i.e. icm ) can be decom- l lexpected profit for firm icm from choosing action aicm l posed into three additive stochastic componentsis then m l icm k = c k + c k + l icm k (7) m l icm aicm s l i Plm where icm k represents local market level private l = l aicm s m + l l (3) m icm icm information, c k is the private information that = l s m aicm a−icm P−icm + l l l l (4) a chain possesses about a particular global market icm icm a l −icm (MSA), and c k is a nonspatial component of pri- vate information that is chain specific. Following ourwhere P−icm = j=icm Pj aj s m . Given these expected l l earlier discussion, we assume that icm k is an i.i.d. lprofits, the optimal action for a store is then Gumbel error. We further assume that the two remain- ing components are jointly distributed with distribu- l icm = Pr icm l aicm s m + l l icm aicm l m tion function F c k c k , where is a set of parameters associated with F . Denoting the parameter > l ailm s m + l ailm ∀ ailm = aicm l (5) icm c icm c c vector = and letting icm k be an indicator l function such thatwhich is the system of equations that define the (pure strategy) BNE of the game. Because a firm’s optimal 1 if aicm = k laction is unique by construction, there is no need to l k = (8) icm consider mixed strategies. 0 if aicm = k l If the ’s are drawn from a Type I Extreme Valuedistribution, this BNE must satisfy a system of logit the optimal choice probabilities (conditional onequations (i.e. best response probability functions). m c k c k ) for a given store can be written asThe general framework described above has beenapplied in several economic settings and its properties l icm aicm = k l Plm X m c k c kare well understood. Existence of equilibrium follows E P exp s m k + −ilm m k1 + −ilm k2 + c k + c kdirectly from Brouwer’s fixed point theorem. = c c To proceed further, we need to choose a particular k∈E H P exp s m k + E −icm k 1 l + P lm k 2 + c k + c −ic m kspecification for the expected profit functions. We will l (9)assume that the profit that accrues to store icm fromchoosing strategy k in location lm is given by while the likelihood can be constructed as m m E P l icm aicm = k s l i Plm = s k+ l −icm k1 + l −icm k2 icm l aicm = k l Plm s m + c k + c k + l icm k (6) c∈C c k m∈M m c k lm ∈Lm ilm ∈N lm c c kwhere s m is the common state vector of both market m k c k l icm dF m k c k c c(local and MSA) and firm characteristics (chain andstore level). The E lm and P lm terms represent the s.t. Plm = lm Plm s m k c k (10) −ic −ic cexpected proportion of a store’s competitors in mar-ket lm that choose EDLP and PROMO strategies, Note that the construction of the likelihood involvesrespectively a system of discrete choice equations that must sat- isfy a fixed point constraint Plm = lm . There are two k 1 main approaches for dealing with the recursive struc- l = Pj aj = k −icm N lm l ture of this system, both based on methods originally j=icm applied to dynamic discrete choice problems. The first, Note that we have assumed that payoffs are a lin- based on Rust’s (1987) Nested Fixed Point (NFXP)ear function of the share of stores that choose EDLP algorithm, involves solving for the fixed point of theand PROMO, which simplifies the estimation prob- system at every candidate parameter vector and thenlem and eliminates the need to consider the share using these fixed point probabilities to evaluate thewho choose HYBRID H . We further normalize the likelihood. However, the NFXP approach is both com-average profit from the PROMO strategy to zero, one putationally demanding and straightforward to apply
  • Ellickson and Misra: Supermarket Pricing Strategies818 Marketing Science 27(5), pp. 811–828, © 2008 INFORMSonly when the equilibrium of the system is unique.10 a general basis: either EDLP, PROMO or HYBRID.An alternate method, based on Hotz and Miller’s The HYBRID strategy is included to account for the(1993) Conditional Choice Probability (CCP) estimator, fact that many practitioners and marketing theoristsinvolves using a two-step approach that is both com- view the spectrum of pricing strategies as more aputationally light and more robust to multiplicity.11 continuum than a simple EDLP-PROMO dichotomyThe first step of this procedure involves obtaining con- (Shankar and Bolton 2004). The fact that just over asistent estimates of each firm’s beliefs regarding the third of the respondents chose the HYBRID option isstrategic actions of its rivals. These “expectations” are consistent with this perception.then used in a second stage optimization procedure toobtain the structural parameters of interest. Given the 5. Empirical Implementationcomplexity of our problem, we chose to adopt a two- The empirical implementation of our frameworkstep approach based on Bajari et al. (2005), who were requires three primary inputs. First, we need tothe first to apply these methods to static games. choose an appropriate set of state variables. These will be the market, store and chain characteristics4. Data Set that are most relevant to pricing strategy. To deter-The data for the supermarket industry are drawn mine which specific variables to include, we drawfrom Trade Dimension’s 1998 Supermarkets Plus heavily on the existing marketing literature. Second,Database, while corresponding consumer demograph- we will need to define what we mean by a “mar-ics are taken from the decennial Census of the United ket.” Finally, we need to estimate beliefs and con-States. Descriptive statistics are presented in Table 1. struct the empirical likelihood. We outline each ofTrade Dimensions collects store level data from every these steps in the following subsections, concludingsupermarket operating in the United States for use in with a discussion of unobserved heterogeneity andtheir Marketing Guidebook and Market Scope publica- our strategy for identification.tions, as well as selected issues of Progressive Grocer 5.1. Determinants of Pricing Strategymagazine. The data are also sold to marketing firms The focus of this paper is the impact of rival pricingand food manufacturers for marketing purposes. The policies on a firm’s own pricing strategy. However,(establishment level) definition of a supermarket used there are clearly many other factors that influenceby Trade Dimensions is the government and industry pricing behavior. Researchers in both marketing andstandard: a store selling a full line of food products economics have identified several, including con-and generating at least $2 million in yearly revenues. sumer demographics, rival pricing behavior, and mar-Foodstores with less than $2 million in revenues are ket, chain, and store characteristics (Shankar andclassified as convenience stores and are not included Bolton 2004). Since we have already discussed the rolein the data set.12 of rival firms, we now focus on the additional deter- Information on pricing strategy, average weekly minants of pricing strategy.volume, store size, number of checkouts, and addi- Several marketing papers highlight the impact oftional store and chain level characteristics was gath- demographics on pricing strategy (Ortmeyer et al.ered using a survey of each store manager, conducted 1991, Hoch et al. 1994, Lal and Rao 1997, Bell andby their principal food broker. With regard to pric- Lattin 1998). Of particular importance are consumering strategy, managers are asked to choose the strat- factors such as income, family size, age, and accessegy that is closest to what their store practices on to automobiles. In most strategic pricing models, the PROMO strategy is motivated by some form of spa-10 It is relatively simple to construct the likelihood function when tial or temporal price discrimination. In the spatialthere is a unique equilibrium, although solving for the fixed point models (e.g. Lal and Rao 1997, Varian 1980), PROMOat each iteration can be computationally taxing. However, con- pricing is aimed at consumers who are either will-structing a proper likelihood (for the NFXP) is generally intractablein the event of multiplicity, since it involves both solving for all ing or able to visit more than one store (i.e. thosethe equilibria and specifying an appropriate selection mechanism. with low travel costs) or, more generally, those whoSimply using the first equilibrium you find will result in mispec- are more informed about prices. The EDLP strategyification. A version of the NFXP that is robust to multiplicity has instead targets consumers who have higher travelyet to be developed. costs or are less informed (perhaps due to hetero-11 Instead of requiring a unique equilibrium to the whole game, geneity in the cost of acquiring price information). Intwo-step estimators simply require a unique equilibrium be played the case of temporal discrimination (Bell and Lattinin the data. Futhermore, if the data can be partioned into distinctmarkets with sufficient observations (as is the case in our applica- 1998, Bliss 1988), PROMO pricing targets customerstion), this requirement can be weakened further. who are willing to either delay purchase or stockpile12 Firms in this segment operate very small stores and compete only products, while EDLP targets customers that preferwith the smallest supermarkets (Ellickson 2006, Smith 2006). to purchase their entire basket in a single trip or at a
  • Ellickson and Misra: Supermarket Pricing StrategiesMarketing Science 27(5), pp. 811–828, © 2008 INFORMS 819single store. Clearly, the ability to substitute over time competing stores. Intuitively, markets are groups ofor across stores will depend on consumer characteris- stores that are located close to one another. To con-tics. To account for these factors, we include measures struct these markets, we used a statistical clusteringof family size, household income, median vehicle method (K-means) based on latitude, longitude, andownership, and racial composition in our empirical ZipCode information.14 Our clustering approach pro-analysis. duced a large set of distinct clusters that we believe Since alternative pricing strategies will require dif- to be a good approximation of the actual markets infering levels of fixed investment (Lattin and Ortmeyer which supermarkets compete. These store clusters are1991), it is important to control for both store and somewhat larger than a typical ZipCode, but signifi-chain level characteristics. For example, large and cantly smaller than the average county. As robustnesssmall chains may differ in their ability to effi- checks, we experimented with the number of clus-ciently implement particular pricing strategies (Dhar ters, broader and narrower definitions of the marketand Hoch 1997). Store level factors also play a (e.g. ZipCodes and MSAs), as well as nearest neigh-role (Messinger and Narasimhan 1997). For example, bor methods and found qualitatively similar resultsEDLP stores may need to carry a larger inventory (to (see Appendix B.1).satisfy large basket shoppers), while PROMO storesmight need to advertise more heavily. Therefore, we 5.3. Estimation Strategyinclude a measure of store size and an indicator As noted above, the system of discrete choice equa-variable for whether the store is part of a vertically tions presents a challenge for estimation. We adopt aintegrated chain. Finally, since the effectiveness of two stage approach based on Bajari et al. (2005). Thepricing strategies might vary by market size (e.g. first step is to obtain a consistent estimate of Plm , theurban versus rural), we include measures of geo- probabilities that appear (implicitly) on the right handgraphic size, population density, and average expen- side of Equation (9).15 These estimates Plm are usedditures on food. to construct the −icm ’s, which are then plugged into l the likelihood function. Maximization of this (pseudo)5.2. Market Definition likelihood constitutes the second stage of the proce-The supermarket industry is composed of a large dure. Consistency and asymptotic normality has beennumber of firms operating anywhere from 1 to established for a broad class of two-step estimators1,200 outlets. We focus on the choice of pricing strat- by Newey and McFadden (1994), while Bajari et al.egy at an individual store, abstracting away from the (2005) provide formal results for the model estimatedmore complex issue of how decisions are made at here. We note in passing that consistency of the esti-the level of the chain. This requires identifying the mator is maintained even with the inclusion of theprimary trading area from which each store draws two random effect terms and , since these vari-potential customers. Without disaggregate, consumer- ables are treated as private information of each store.level information, the task of defining local markets A final comment relates to the construction of stan-requires some simplifying assumptions. In particular, dard errors. Because the two-step approach precludeswe assume markets can be defined by spatial prox- using the inverse information matrix, we use a boot-imity alone, a strong assumption in some circum- strap approach instead.16stances (Bell et al. 1998). However, absent detailedconsumer level purchase information, we cannot relax 5.4. The Likelihoodthis assumption further. Therefore, we will try to be In our econometric implementation, we will assumeas flexible as possible in defining spatial markets. that and are independent, mean zero normal Although there are many ways to group firms errors, so thatusing existing geographic boundaries (e.g. ZipCodesor Counties), these pre-specified regions all share the m F k c ksame drawback: they increase dramatically in size c mfrom east to west, reflecting established patterns of =F c k k ×F c k k (11)population density.13 Rather than imposing this struc-ture exogenously, we allow the data to sort itself by 14 ZipCodes are required to ensure contiguity: without ZipCodeusing cluster analysis. In particular, we assume that information, stores in Manhattan would be included in the samea market is a contiguous geographic area, measur- market as stores in New by geodesic distance and containing a set of 15 The l ’s are functions of Plm . −icm 16 In particular, we bootstrapped across markets (not individ-13 One exception is Census block groups, which are about half the ual stores) and held the pseudorandom draws in the simulatedsize of a typical ZipCode. However, we feel that these areas are too likelihood fixed across bootstrap iterations. To save time we usedsmall to constitute reasonably distinct supermarket trading areas. the full data estimates as starting values in each bootstrap iteration.
  • Ellickson and Misra: Supermarket Pricing Strategies820 Marketing Science 27(5), pp. 811–828, © 2008 INFORMSwhere both F and F are mean zero normal dis- obtain ˆ 1 , the SML estimate of 1 . Given these esti-tribution functions with finite covariance matrices. mates, and applying Bayes’ rule, the posterior expec-For simplicity, we also assume that the covariance m tation of P aicm = k s c k c k can be obtained via lmatrices are diagonal with elements 2 k and 2 k . the following computationFor identification, consistent with our earlier inde-pendence and normalization assumptions, we assume l icm aicm = k ˆ l m c k c k mthat c P = c P = 0 ∀ c ∈ C m ∈ M. We can thenuse a simulated maximum likelihood procedure that ˆ m m · l icm c k c k dF c k c k 1 kreplaces (10) with its sample analog −1 R R · l ˆ m k c k dF m k c k 1 k icm c c = R−1 R−1 l icm aicm = k l c∈C r =1 m∈M r =1 lm ∈Lm ilm ∈N lm c c (13) m l k Plm s c k c k icm (12) While this expression is difficult to evaluate analyt- ically, the vector of beliefs defined by m In the simulation procedure, c k r and c k r Picm aicm = k = l l l aicm = k ˆ l m k k (14) icm c care drawn from mean zero normal densities with vari-ances 2 k and 2 k respectively. We use R = R = can be approximated by its simulation analog500 and maximize (12) to obtain estimates of the struc-tural parameters. Note that the fixed point restriction, Picm aicm = k l lPlm = lm , no longer appears since we have replaced R l aicm = k ˆ l m k k l ˆ m k k r=1 icm c c r icm c c rPlm with Plm in the formulae for E lm and P lm , which R ˆ −ic −ic r=1 l icm m c k c k rare used in constructing icm (see (9)). We now turn lto estimating beliefs. (15) m in which c k c k r are draws from a distribution5.5. Estimating Beliefs m F c k c k with similar properties to those de-In an ideal setting, we could recover estimates of scribed in §5.4. Again, we use R = 500 simulationeach store’s beliefs regarding the conditional choice draws. Recalling that k ∈ K = E H P , we can nowprobabilities of its competitors using fully flexible k define a consistent estimator of lm as −icnonparametric methods (e.g. kernel regressions orsieves). Unfortunately, our large state vector makes k −1 lthis infeasible. Instead, we employ a parametric ˆ l = Nvm Pj aicm = k l (16) −icm v=c lapproach for estimating ˆ −icm , using a mixed multi- l j=icmnomial logit (MNL) specification to recover these first 5.6. Common Unobservablesstage choice probabilities (Appendix B.4 provides a While our data set is rich enough to include a largesemi-parametric robustness analysis). This is essen- number of covariates upon which firms may condi-tially the same specification employed in the sec- tion their actions, the strong emphasis we have placedond stage procedure (outlined above), only the store’s on strategic interaction may raise concerns regardingbeliefs regarding rivals’ actions are excluded from this the role of unobserved heterogeneity. In particular,reduced form. Note that we do not require an explicit how can we be sure that firms are actually reacting toexclusion restriction, since our specification already the actions of their rivals, rather than simply optimiz-contains natural exclusion restrictions due to the pres- ing over some common features of the local marketence of state variables that vary across stores and that we do not observe? Manski (1993) frames this aschains. the problem of distinguishing between endogenous We implement an estimator similar to (12), but with and correlated effects. Although the presence of boththe coefficients on the −icm ’s (i.e. ’s) set to zero. l effects yields collinearity in the linear in means modelLet the parameters in the first stage be denoted by that Manski analyzes (i.e. the reflection problem), the 1 = 1 1 17 and the first stage likelihood for a nonlinearity of the discrete choice framework elimi-given store be denoted by icm m nates this stark nonidentification result in our setting. l c k c k . Usinga simulated maximum likelihood (SML) approach, we However, the presence of correlated unobservables remains a concern. In what follows, we outline two strategies for handling this problem. The first incorpo-17 The subscript 1 indicates that these are first stage estimates. rates a fixed effect at the MSA level, while the second
  • Ellickson and Misra: Supermarket Pricing StrategiesMarketing Science 27(5), pp. 811–828, © 2008 INFORMS 821incorporates a random effect at the level of the cluster. local market (i.e., cluster)18 and are drawn from a dis-Our main results are robust to either alternative. tribution of known parametric form. This is clearly The most direct solution is to add a common unob- satisfied by the assumptions imposed above. The sec-servable, denoted lm , to the strategy specific profit ond assumption normalizes the expected profit asso-function of each store. Using the notation defined ear- ciated with one strategy to zero. This is a standardlier, this can be written identification condition of any multinomial choice model. We normalize the mean profit of the PROMO l aicm = k s m l i Plm strategy to zero. The final assumption is an exclusion icm restriction. = sm k + E l −icm k1 + P l −icm k2 + lm + l icm k (17) The need for an exclusion restriction can be illus- trated using Equation (9). Our two-step approach Ideally, one would estimate each lm as a cluster involves estimating the shares ( −icm ’s) on the right lspecific fixed effect. However, this would require esti- hand side of (9) in a first stage. These shares, whichmating 8,000 additional parameters with less than are simple functions of each firm’s beliefs regard-18,000 observations, which is clearly infeasible. A fea- ing the conditional choice probabilities of its rival’s,sible alternative is to model the common unobserv- depend on the same state vector s m as the firstable at the level of the MSA (i.e. include m instead term of the profit function s m k , creating a potentialof lm . In practice, this simply requires running the collinearity problem. Of course, identification can befirst stage separately for each MSA and then adding trivially preserved by the inherent nonlinearity of thean MSA level fixed effect to the second stage proce- discrete choice problem, but this follows directly fromdure. This has the added benefit of relaxing the equi- functional form. An alternative strategy (suggested bylibrium restriction: we need now only assume that a Bajari et al. 2005) involves identifying one or moreunique equilibrium is played in every MSA, instead continuous covariates that enter firm i’s payoffs, butof across all MSAs. We implement this strategy below. not the payoffs of any of its rivals. Note that eachHowever, given the local nature of the strategic firm’s private shock icm has already been assumed linteraction documented here, an MSA level common to satisfy this restriction, creating at least one set ofunobservable may not be sufficient to account for the “natural” exclusion restrictions. The characteristics ofrelevant correlated effects. rival firms constitute an additional exclusion. How- A second alternative is to use a cluster level ran- ever, a more subtle identification issue concerns thedom effect (i.e. assume the unobservables come from source of exogenous variation in the data that cana pre-specified density g lm ) and simply integrate pin down the form of strategic interaction. For this,out over lm in the second stage estimation procedure, we exploit the specific structure of the private infor-maximizing the resulting marginalized sample like- mation term and the presence of large multi-marketlihood. However, there is an additional impediment chains. The two random effect terms in (7) captureto implementing this strategy: the fact that lm is each firm’s tendency to employ a consistent strategy ma common unobservable prevents the econometrician within an MSA c k and/or across all stores c kfrom obtaining a consistent first stage estimate of Plm , in the chain. These firm level tendencies vary acrossa requirement of the two-stage procedure employed chains and markets, providing a source of variation for the local interactions that take place in any givenabove. (Note that this is not a problem if the first cluster. The key assumption is that we sometimesstage can be estimated separately for each market, as see firms that follow a consistent strategy (EDLP, forwas the case with the MSA level unobservable.) To example) at the market level deviate in a local clus-accommodate cluster level random effects, we adopt ter by playing either PROMO or HYBRID when thean approach based on Aguirregabiria and Mira (2007) demographics of the local market or its beliefs regard-that is tailored to our particular setting (the details of ing rival strategies outweigh its desire to follow aour algorithm are provided in Appendix C). consistent (chain or MSA-wide) strategy. This has the flavor of an instrumental variable approach, where5.7. Identification the instruments are measures of the overall strat-Bajari et al. (2005) establish identification of the struc- egy a chain adopts outside the local market or MSA.tural parameters of a broad class of discrete games of In order to maintain the static, local, simultaneousincomplete information, of which ours is a subcase. move structure of the game, we have restricted theseTheir identification argument rests on three assump-tions. The first two have already been (implicitly) 18 Note that the i.i.d. requirement need only hold at the cluster level.stated, but will be repeated here more formally. The In particular, it’s fine to include random effects in the error term, sofirst assumption is that the error terms are dis- long as they are treated as private information. This is the approachtributed i.i.d. across players and actions in any given we adopt in our main specification.
  • Ellickson and Misra: Supermarket Pricing Strategies822 Marketing Science 27(5), pp. 811–828, © 2008 INFORMSfirm level tendencies to be privately observed ran- the demographics of the market they serve. More-dom effects. However, an alternative specification in over, the impact of demographics corresponds closelywhich we conditioned directly on the average strate- to existing empirical studies of consumer preferencesgies that firms follow outside a given MSA yielded and conventional wisdom regarding search behavior.similar results. Second, we find that EDLP is favored by firms that operate larger stores and are vertically integrated into distribution. Again, this accords with conventional6. Results and Discussion wisdom regarding the main operational advantagesAs noted earlier, choosing an optimal pricing strat- of EDLP. Finally, with regard to strategic interaction,egy is a complex task, forcing firms to balance the we find that firms coordinate their actions, choosingpreferences of their customers against the strategic pricing strategies that match their rivals. This iden-actions of their rivals. A major advantage of our tifies an aspect of firm behavior that has not beentwo-step estimation approach is that, by estimating addressed in the existing literature: exactly how firmsbest response probability functions rather than equi- react to rival strategies.librium correspondences, we can separately identify Our main empirical results are presented in Table 6.strategic interactions, reactions to local and market The coefficients, which represent the parameterslevel demographics, and operational advantages asso- of the profit function represented in Equation (6),ciated with larger stores and proprietary distribution have the same interpretation as those of a stan-systems. The Bayesian structure of the game allows dard MNL model: positive values indicate a positiveus to account for different equilibria with the same impact on profitability, increasing the probability thatcovariates, due to the presence of unobserved types. the strategy is selected relative to the outside optionMore importantly, it allows us to model all 8,000 mar- (in this case, PROMO).kets as variations in the play of a game with thesame structure, but different conditioning variables. 6.1. The Role of DemographicsAs the conditioning variables vary, we are able to The coefficients on consumer demographics are pre-trace out the equilibrium correspondence and iden- sented in the second and third sections of Table 6.tify the impact of several distinct factors. First, we With the exception of two MSA-level covariates, everyfind that firms choose strategies that are tailored to demographic factor plays a significant role in the Table 6 Estimation Results EDLP HYBRID Estimate Std. err T -stat Estimate Std. err T -stat Effect Intercept −1 5483 0 2426 −6 3821 2 1344 0 2192 9 7372 Strategy variables ˆEDLP lm 4 4279 0 1646 26 9010 −2 0924 0 1595 −13 1185 −ic ˆPROMO lm −3 7733 0 1501 −25 1386 −6 3518 0 1351 −47 0155 −ic MSA characteristics Size (’000 sq. miles) 0 0394 0 0848 0 4645 −0 0566 0 0804 −0 7039 Density (pop 10,000 per sq. mile) −0 0001 0 0002 −0 4587 0 0006 0 0002 2 9552 Avg. food expenditure ($ ’000) −0 0375 0 0155 −2 4225 −0 0013 0 0141 −0 0904 Market variables Median household size 0 5566 0 1989 2 7983 0 2150 0 0900 2 3889 Median HH income −0 0067 0 0019 −3 5385 0 0056 0 0017 3 2309 Proportion Black 0 6833 0 1528 4 4719 0 0139 0 1443 0 0963 Proportion Hispanic 0 5666 0 2184 2 5943 −0 0754 0 2033 −0 3708 Median vehicles in HH −0 1610 0 0840 −1 9167 0 2263 0 1173 1 9292 Store characteristics Store size (sqft ’000) 0 0109 0 0015 7 2485 0 0123 0 0014 8 8512 Vertically integrated 0 1528 0 0614 2 4898 0 0239 0 0550 0 4343 Chain characteristics Number of stores in chain −0 0002 0 0001 −2 7692 0 0002 0 0001 3 5000 Chain effect 1 7278 0 0998 17 3176 2 8169 0 0820 34 3531 Chain/MSA effect 0 7992 0 0363 22 0408 0 9968 0 0278 35 8046
  • Ellickson and Misra: Supermarket Pricing StrategiesMarketing Science 27(5), pp. 811–828, © 2008 INFORMS 823choice of EDLP as a pricing strategy. This is important this is difficult to interpret since almost all the largefrom an econometric standpoint, since we use these chains are vertically integrated into distribution (i.e.very same factors to construct expectations in the first there are almost no large, nonvertically integratedstage. In particular, the significance of the estimates firms). Finally, both the chain specific and chain/MSAmeans that we do not have to worry about collinear- random effects are highly significant, which is not sur-ity. The statistical significance of the parameters is prising given the geographic patterns shown earlier.19also substantively important. It suggests that the evenafter accounting for competitive and supply side (e.g. 6.3. The Role of Competition:store/chain) characteristics, consumer demand plays Differentiation or Coordinationa strong role in determining pricing strategy. By constructing a formal model of strategic interac- Focusing more closely on the demand related tion, we are able to address the central question posedparameters, we find that (relative to PROMO), EDLP in this paper—what is impact of competitive expec-is the preferred strategy for geographic markets that tations on the choice of pricing strategy? Our conclu- HHhave larger households = 0 5566 , more racial sions are quite surprising. The first section of Table 6diversity in terms of African-American BL = 0 6833 reveals that firms facing competition from a high HIand Hispanic = 0 5666 populations, lower (expected) share of EDLP stores are far more likely to INCincome = −0 0067 , and fewer vehicles per choose EDLP than either HYBRID or PROMO k1 =household VH = −0 1610 . These results suggest that 4 4279 k2 = −3 7733 . The HYBRID case behavesEDLP is mostly aimed at lower income consumers analogously; when facing a high proportion of eitherwith larger families (i.e. more urbanized areas). Our EDLP or PROMO rivals, a store is least likely tofindings are clearly consistent with the consumer seg- choose HYBRID k1 = −2 0924 k2 = −6 3518 . Inments that firms like Wal-Mart and Food Lion are other words, we find no evidence that firms differenti-widely perceived to target. It also accords quite well ate themselves with regard to pricing strategy. To thewith the “fixed basket” model of shopping behav- contrary, we find that rather than isolating themselvesior (Bliss 1988, Bell and Lattin 1998), in which con- in strategy space, firms prefer to coordinate on asumers who are more sensitive to the price of an over- single pricing policy. Pricing strategies are strategicall basket of goods prefer EDLP. In particular, our complements.results suggest that the consumers who are unable This coordination result stands in sharp contrastto substitute inter-temporally are disproportionately to most formal models of pricing behavior, whichpoor, nonwhite, and from larger families. On the other (at least implicitly) assume that these strategies arehand, we find that consumers who are most able to vehicles for differentiation. Pricing strategy is typi-defer or stockpile purchases (wealthy suburbanites cally framed as a method for segmenting a hetero-with greater access to transportation) tend to prefer geneous market—firms soften price competition byPROMO or HYBRID pricing. moving further away from their rivals in strategy space. This is not the case for supermarkets. Instead6.2. Firm and Store Level Characteristics of finding the anti-correlation predicted by these spa-Turning next to chain and store level characteris- tial models, we find evidence of associative matching,tics, we again find that most parameter estimates which usually occurs in settings with network effectsare statistically significant. These effects, which are in or complementarities. This suggests that firms areline with both theory and broad intuition, providean additional empirical validation of our structural able to increase the overall level of demand by match-framework. ing their rivals’ strategies, a possibility we discuss in The last two sections of Table 6 reveal that stores more detail below. However, before discussing ourchoosing EDLP are both significantly larger SS = coordination result in greater detail, we must address0 0109 and far more likely to be vertically integrated the issue of correlated unobservables. VIinto distribution = 0 1528 . This is consistentwith the view that EDLP requires substantial firm 19 An earlier version of this paper also included the share oflevel investment, careful inventory management, and each firm’s stores outside the local MSA that employ EDLP and PROMO pricing as additional regressors. Not surprisingly, a firm’sa deeper selection of products in each store in order propensity to follow a particular strategy at the level of the chainto satisfy the demands of one-stop shoppers. It is had a large and significant impact on its strategy in a particularalso consistent with the logic of Lal and Rao (1997), store (and soaked up a lot of variance). While this suggests thewhereby pricing strategy involves developing an presence of significant scale economies in implementing pricingoverall positioning strategy, requiring complementary strategies, as suggested by both Lattin and Ortmeyer (1991) and Hoch et al. (1994), we omitted it from the current draft to maintaininvestments in store quality and product selection. the internal coherency of the underlying model (i.e. the simultane-Surprisingly, the total number of stores in the chain is ity of actions). However, these results are available from the authorsnegatively related to EDLP ST = −0 0002 , although upon request.
  • Ellickson and Misra: Supermarket Pricing Strategies824 Marketing Science 27(5), pp. 811–828, © 2008 INFORMSTable 7 Robustness by nonstrategic factors, including market characteris- tics, store and firm-level covariates, and the randomSpecification Strategy Strategy ˆ EDLP Variables ˆPROMO lm −ic −ic lm effects that we have treated as private information. In addition to this decomposition of profits, we alsoBaseline EDLP 4 4279 (0.1646) −3 7733 (0.1501) HYBRID −2 0924 (0.1595) −6 3518 (0.1351) conducted a policy experiment aimed at highlight-MSA by MSA EDLP 3 1867 (0.2522) −3 2823 (0.1771) ing the mechanism by which strategic effects influ- HYBRID −3 4418 (0.2603) −6 2746 (0.1701) ence pricing strategy. To do so, we simply shut off theNPL EDLP 1 7464 (0.1743) −2 5699 (0.1723) strategic effects and compared the odds of choosing HYBRID −0 7365 (0.1770) −4 9899 (0.1739) PROMO relative to EDLP under this counterfactual scenario to what we see in the data.20 The results are striking. At the aggregate level, the true odds ratio The surprising nature of our coordination result was around 1.31, implying that PROMO is roughlydemands careful consideration. Again, how can we be 31% more likely to be chosen than EDLP. However, insure that firms are actually reacting to the actions of the counterfactual scenario (without strategic effects)their rivals, rather than simply optimizing over some this drops to 4.1% (odds ratio = 1 041). This finding iscommon but unobserved feature of the local market? notable since it offers an explanation for why EDLPSection 5.6 described two alternative strategies for did not become the dominant paradigm in supermar-dealing with the potential presence of common unob- ket pricing. To see why, notice first that even with-servables. The first method involved adding an MSA out strategic effects, the odds ratio was greater thanlevel fixed effect to the baseline specification. In prac- one. This implies that there are market factors that,tice, this requires estimating the first stage separately on average, lead a market to lean towards one strat-for each MSA (to ensure a consistent first stage) and egy. With the same set of characteristics, the strate-then expanding the second stage likelihood to include gic effects induce a feedback effect that can causean MSA fixed effect. The main coordination results the market to tip more significantly in that direction.are presented in the section of Table 7 titled MSA While these are clearly aggregate trends, we observedby MSA (the demographic and chain level covariates similar phenomena in individual markets as well.have been suppressed for brevity, but are available Broadly speaking, strategic effects strengthen coordi-from the authors upon request). While the coefficients nation in markets where one strategy is weakly dom-have changed slightly in magnitude, the main coor- inant (under the counterfactual).dination result remains strong. The second methodinvolved adding a cluster level random effect, and re- 6.4. Discussion of Resultsestimating the model using Aguirregabiria and Mira’s The Bayesian structure of our game allows us to(2006) NPL algorithm. These results are presented in represent a quite complex game using a relativelythe section titled NPL. Here we find that the magni- simple structure. By tracing out the equilibrium corre-tudes of the coefficients fall relative to both the base- spondence, we have found that firms favor particularline and MSA by MSA specifications, as one might strategies in certain markets, in ways that are con-expect if firms are indeed reacting to a common unob- sistent with existing theory. We have also found thatservable. However, the coordination effects are still certain types of firms favor particular strategies, alsolarge and significant: pricing strategies are indeed consistent with existing theory. Finally, we have foundstrategic complements. that firms are more likely to choose a particular strat- But how important are these strategic effects? The egy if they expect their rivals to do the same. Thisparameter estimates from our baseline model can is a sharp departure from existing theory. It is worthbe used to gauge the relative influence that strate- emphasizing that reactions to market demographicsgic interactions have on profits. Because individual and firm characteristics help explain how firms arecovariates can influence profits either negatively or able to coordinate on consistent strategies. However,positively, a simple additive decomposition of prof- they do not explain why they choose to do so. Coordi-its by strategic and nonstrategic factors is inappropri- nation implies that firm’s conditional choice probabil-ate. To adjust for this, we adopted the method pro- ities act as strategic complements, meaning that theirposed in Silber et al. (1995), using the average squared best response probability functions (9) are upwardcontributions of each factor to construct a measure sloping. To support such complementarity, coordina-of the share of variance explained. This decomposi- tion must somehow increase the overall size of thetion reveals that, on average, strategic factors explain perceived market. In most cases, this means drawingabout 20.3% of the variation in EDLP profits and expenditures away from the outside good.13.2% of the variation in HYBRID profits, quite sub-stantial fractions. The remaining variance is explained 20 Note that PPROMO /PEDLP is now our object of interest.
  • Ellickson and Misra: Supermarket Pricing StrategiesMarketing Science 27(5), pp. 811–828, © 2008 INFORMS 825 In the context of supermarket pricing, this suggests drives consumer demand. The fact that firms coor-that coordination may actually increase the amount dinate with their rivals suggests that consumers pre-consumers are willing to spend on groceries, perhaps fer to receive a consistent message. While our resultsby drawing them away from substitutes like restau- pertain most directly to supermarkets, it seems likelyrants, convenience stores, and discount clubs. One that other industries could behave similarly. Futureway this might occur in practice is if consumers are research could examine the robustness of our findingsmore likely to trust retailers that provide a message by analyzing other retail industries, such as depart-that is consistent with those of their rivals. In other ment stores or consumer electronics outlets.words, if one firm tells you that providing the high- In this paper, our primary focus was the construc-est value involves high price variation while another tion and econometric implementation of a frameworktouts stable prices, you may be unwilling to trust for analyzing best responses to rival pricing strate-either, shifting your business to a discount club or gies. Our analysis describes the nature of strategicanother retail substitute. While this intuition has yet interactions, but does not delve into the details ofto be formalized, it is consistent with the empha- why these strategies are dominant. Decomposing thesis that Ortmeyer et al. (1991) place on maintaining why element of strategic coordination seems a fruitfulpricing credibility. Another possibility, consistent with area for research. We hasten to add that such researchLal and Rao (1997), is that price positioning is is needed not only on the empirical side but alsomulti-dimensional and by coordinating their strate- on the theoretical front. Building theoretical modelsgies stores can mitigate the costs of competing along that allow for the possibility of both differentiationseveral dimensions at once. Without a formal model and coordination is a challenging but undoubtedlyof consumer behavior and detailed purchase data, rewarding path for future research.we are unable to pin down the exact source of the The tendency to coordinate raises the possibilitycomplementarities we have documented here. How- that games such as this might support multiple equi-ever, we have provided strong empirical evidence libria. While this is not a concern in our current study,regarding how firms actually behave. Understand- it could play a central role when conducting pol-ing why firms find it profitable to coordinate their icy experiments or when analyzing settings in whichactions remains a promising area for future theoretical demographics (or other covariates) cannot effectivelyresearch. facilitate coordination. Developing methods that are The results presented above provide definitive robust to such possibilities remains an important areaanswers to the three questions posed in the intro- for future research.duction of this paper. We have found that demand On the methodological front, our research also cen-related factors (i.e. demographics) are important for ters its attention on the discrete choice aspect of strat-determining the choice of pricing strategy in a market; egy. There are a number of issues that emerge oncestore and firm level characteristics also play a central such strategic choices have been made such as therole. Both of these results are in line with the extant reaction of consumers (see e.g. Singh et al. 2006) andliterature. However, our results concerning competi- the overall demand faced by stores. Research thattive expectations are in sharp contrast to prevailing aims at incorporating such postgame outcome datatheory in both economics and marketing and warrant into the analysis promises to offer newer and crisperfurther attention. The final section outlines a research insights into the nature of competition in the market.agenda for extending the results found in this paper. Finally, in building our model of strategic interac- tion, we have assumed that firms interact in a static setting, making independent decisions in each store.7. Conclusions and Directions for A more involved model would allow chains to make Future Research joint decisions across all of their outlets and accountThis paper analyzes supermarket pricing strategies for richer (dynamic) aspects of investment. Develop-as discrete game. Using a system of simultaneous ing such a model is the focus of our current research.discrete choice models, we estimate a firm’s optimalchoice conditional on the underlying features of the Acknowledgments The authors thank participants at the Supermarket Retail-market, as well as each firm’s beliefs regarding its ing Conference at the University of Buffalo, the 2006competitor’s actions. We find evidence that firms BCRST Conference at the University of Toronto, thecluster by strategy, rather than isolating themselves 2005 QME Conference at the University of Chicago, andin product space. We also find that demographics the Supermarket Conference held at IFS London, as welland firm characteristics are strong determinants of as seminar participants at Duke, UCLA, Stanford, UTpricing strategy. From a theoretical perspective, it Dallas, and Yale. The authors also thank Victor Aguirrega-is clear that we have yet to fully understand what biria, Pat Bajari, J. P. Dubé, Han Hong, Paul Nelson, and
  • Ellickson and Misra: Supermarket Pricing Strategies826 Marketing Science 27(5), pp. 811–828, © 2008 INFORMSChris Timmins for their comments. All remaining errors are product of systematic reporting error, as this would requirethe authors’ responsibility. coordination between tens of thousands of managers and hundreds of brokers to willfully and consistently mis-reportAppendix A. Survey Validity their practices (for no obvious personal gain). However, toAll of the variables in the Trade Dimensions data, includ- further allay such fears, we cross-verified the data ourselvesing the information on pricing strategy, are self-reported. using publicly available data from the Dominick’s FinerThis may raise some concerns regarding accuracy, especially Foods (DFF) supermarket chain in Chicago. In particular,given the high degree of local variation we observe in the we extracted store level prices from four major product cate-data. Two questions naturally arise. First, are firms actually gories for every store in the DFF data set and matched themwilling and able to set prices at such local levels? Second, up to the pricing classifications reported by Trade Dimen-do these self-reported strategies reflect actual differences in sions. The vast majority of the Dominick’s stores are iden-pricing behavior? We will address both issues in turn. tified as PROMO (93%), while the remainder are HYBRID, First, with regard to local pricing, we should note that which is itself encouraging since Dominick’s is known to besupermarket firms clearly have the technological resources a PROMO chain. We then checked whether the incidenceto set prices (and therefore pricing strategy) at a very of promotions (i.e. whether a UPC was “on sale”) variedlocal level. Indeed, Montgomery (1997) provides a novel across PROMO and HYBRID stores. In all four categoriesmethod for profitably customizing prices at the store level, that we examined (Soft Drinks, Oatmeal, Paper Towels, andusing widely available scanner data.21 We contacted pricing Frozen Juice), we found a significantly lower incidence ofmanagers at several major chains and other industry pro- promotions at the HYBRID stores. The differences rangedfessionals regarding their ability to engage in such micro- from 8.1% in Soft Drinks (a very heavily promoted cate-marketing. Even on the condition of anonymity, they were gory) to 23.4% in Oatmeal. All differences were significantextremely reluctant to discuss the details of their actual at the 1% level.pricing strategies, but did acknowledge that they “certainly In addition, we also compared the HYBRID and PROMOhave the data and resources to do it.” Furthermore, a con- stores for equality in the variance of the prices using stan-sultant who was involved in several recent supermarket dard folded—F tests. One would expect PROMO storesmergers confirmed that the extent of local pricing was a key to have higher variances. For three of the four categoriesfactor in the approval process.22 (Oatmeal, Paper Towels, and Frozen Juice) the variance in A related issue is whether firms face significant pressure prices was indeed higher in the PROMO stores, validatingto maintain a consistent (pricing) image across stores. We the survey data. The difference was not significant for Softsuspect not. Unlike many other types of retail food services Drinks category. We also repeated each analysis for only(e.g., fast food establishments), supermarket customers do the highest selling UPC in each category and found qual-the majority of their shopping in a single store.23 There- itatively similar results. While these tests use only a fewfore, while consumers undoubtedly have strong preferences product categories from a single chain in a single market,over the pricing strategy of their chosen store, they have lit- the sharpness of the results should provide additional con-tle reason to care directly about the overall strategy of the fidence in the integrity of our data.chain. Of course, chains may have strong operational incen-tives (e.g. scale economies in distribution and advertising) Appendix B. Robustness Checksto maintain a consistent strategy across several (not neces-sarily proximate) stores, which might lead them to adopt a B.1. Market Delineation and Definitioncommon strategy in multiple outlets. Indeed, we are rely- As noted earlier, our empirical analysis uses specific marketing on just such incentives to provide the variation needed definitions based on spatial cluster analysis. We verified theto identify the effect of strategic interactions (see §5.7). The robustness of our results to alternative market definitionspoint is that firms may indeed have both strong incentives by repeating the analysis using ZipCodes, Counties, andand the ability to tailor pricing to the local environment. MSAs. In all cases, the results were qualitatively similar. We The second question concerns the validity of the survey also varied the number of clusters and did not find signifi-instrument itself. We note first that the survey was of store cant differences from the results reported above. Finally, wemanagers but administered by brokers (who explained the experimented with n-nearest neighbor methods (we tried 3questions), providing an additional level of cross-validation. and 5 nearest neighbors of a focal store) and again foundIt is unlikely that the results reported below could be the similar results.21 B.2. Multiplicity While the emphasis there is on maintaining a consistent image, As we noted in the main text, consistent estimation of aMontgomery argues that the potential gains to micro-marketing static (or dynamic) game requires some form of uniquenessare quite significant. Setting different sales frequencies in differentstores is simply an alternative method of micro-marketing. of equilibrium, either in the model or in the data.24 Con-22 sistency of our baseline model requires that only one equi- While detailed information on the degree of micro-marketing in librium be played in the data which, in our context, meansthe supermarket industry is not publicly available, explicit evidenceof local pricing was a major issue in the proposed merger betweenStaples and Office Depot (Ashenfelter et al. 2006). 24 Uniqueness may fail to hold in many settings. Brock and Durlauf23 According to the Food Marketing Institute, consumers allocate (2001) and Sweeting (2004) provide two such examples. Non-78% of their overall budget to their primary store. Moreover, their uniqueness can complicate policy experiments, which typicallysecondary store is almost always part of a different chain. involve solving for a new equilibrium.
  • Ellickson and Misra: Supermarket Pricing StrategiesMarketing Science 27(5), pp. 811–828, © 2008 INFORMS 827every location in every MSA. It is possible to relax this by In the rth iteration implement the following steps:estimating the first stage separately for each MSA, so the Step 1. Given Pr−1 updaterequirement becomes that a unique equilibrium be playedin each MSA (we do not have enough data to estimate the r = arg max ln l ailm = k icm cfirst stage separately for each cluster, which would elim- m∈M c∈C lm ∈Lm ilm ∈N lm k∈K c cinate the problem entirely). The results of this procedure kwere very close to the baseline model. For brevity, we report Plm s lm k l icm dF lmonly the coefficients on the strategy variables (see Table 7). r Step 2. Update Plm using , settingB.3. Format CharacterizationIn our baseline model, we assumed that firms care onlyabout the share of their rivals that choose each strategy. P rlm ailm = k = l icm ailm = k r Plr−1 X m lm k ic c cAn alternative, similar to what is done in the entry litera- r · Plr−1 s k ailm dF rture, is to assume that firms care instead about the number m lm c lmof rivals. We reestimated the baseline model using counts whereinstead of shares and found qualitatively similar results. ···B.4. Nonparametric Estimation of −i l k rAs noted above, the ideal approach for estimating beliefs = l icm ailm = k Plr−1 s m lm k icm c lm ∈Lm ilm ∈N lm k∈Kis nonparametric. However, the number of covariates we c cuse precludes us from adopting such a strategy. To assess rthe robustness of our results, we used a bivariate thin-plate · l icm ailm = k cspline to model pricing strategies as nonparametric func- lm ∈Lm ilm ∈N lm k∈K c c −1 ktions of the strategies chosen outside the MSA. Again, the Plr−1 s k l icm dF r lm lmmain results were qualitatively similar to those presented mabove. Step 3. If Plrm − Plr−1 is smaller than a predetermined mB.5. Nonlinearity of f −i value, stop and choose NPL = r . If not, increment r andTo examine the potentially nonlinear relationship between return to Step 1.payoffs ( ) and strategies ( −i ), we adopted a smoothing Note that there are a few key differences between oursplines approach to modeling f −il . In particular, we re- approach and the Nested Pseudo Likelihood (NPL) algo-estimated our model using a bivariate thin-plate spline, rithm proposed by Aguirregabiria and Mira (2007) (hence-treating the functional relationship as forth AM). First, unlike AM, our game is static. This does E P not alter the main econometric properties of the NPL esti- fj a−ilm = f (18) c l −icm l −icm mator, since a static game is simply a one-period subcase of a dynamic one. However, a natural consequence of the The qualitative results obtained using the linear specifica- static setting is that the state variables do not transition overtion continue to hold. For example, the probability of firms time, allowing us to extend the NPL approach to includechoosing EDLP increases with the proportion of competi- continuous states. A more significant point of departuretors that also choose EDLP. between our algorithm and the NPL is the inclusion of con- tinuous heterogeneity. Since the evolution of the observedB.6. Error Structure state variables naturally depends on the unobserved stateIn our analysis we assumed that firm types (the i ’s) were variables, AM restricted their estimator to a finite support.distributed Gumbel (Type I Extreme Value), allowing us to In our case, the static nature of the problem, coupled withspecify set of simultaneous multinomial logit choice prob- an independence assumption ( lm is orthogonal to s), allowsabilities for determining pricing policies. As an alternative us to simply integrate out over a continuous heterogeneityspecification, similar to the empirical application in Bajari distribution. An attractive feature of the NPL algorithm iset al. (2005), we also tested ordered logit/probit models in that it works even in the presence of inconsistent or poorlywhich the strategies were ranked ordered EDLP to PROMO. estimated initial probabilities. As long as the algorithm con-While qualitative findings were similar, these ordered spec- verges, it will do so to the root of the likelihood equations.ifications force a particular ranking of strategies that may In our experience, the procedure converged very quickly tonot be warranted. the same fixed point for several different starting values.Appendix C. Nested Pseudo Likelihood (NPL)Algorithm ReferencesWe assume that the common unobservables are jointly dis- Aguirregabiria, V., P. Mira. 2007. Sequential estimation of dynamictributed with distribution function F lm , where is a discrete games. Econometrica 75(1) 1–53.set of parameters associated with F . To start the algorithm, Ashenfelter, O., D. Ashmore, J. B. Baker, S. Gleason, D. S. Hosken.let Plm be the set of strategy choice probabilities across play- 2006. Econometric methods in merger analysis: Econometricers in a given local market lm . Finally, let Pl0m be some (not analysis of pricing in FTC v. staples. Internat. J. Econom. Bus.necessarily consistent) estimator for Plm . 13(2) 265–279.
  • Ellickson and Misra: Supermarket Pricing Strategies828 Marketing Science 27(5), pp. 811–828, © 2008 INFORMSAugereau, A., S. Greenstein, M. Rysman. 2006. Coordination ver- Lattin, J., G. Ortmeyer. 1991. A theoretical rationale for everyday sus differentiation in a standards war: 56K modems. RAND J. low pricing by grocery retailers. Working paper. Econom. 37(4) 887–909. Manski, C. 1993. Identification of endogenous social effects: TheBajari, P., C. L. Benkard, J. Levin. 2007. Estimating dynamic models reflection problem. Rev. Econom. Stud. 60(3) 531–42. of imperfect competition. Econometrica 75(5) 1331–1370. Messinger, P., C. Narasimhan. 1997. A model of retail formats basedBajari, P., H. Hong, J. Krainer, D. Nekipelov. 2005. Estimating static on consumers’ economizing on shopping time. Marketing Sci. models of strategic interactions. Working paper, University of 16(1) 1–23. Michigan. Montgomery, A. L. 1997. Creating micro-marketing pricing strate-Bayer, P., C. Timmins. 2007. Estimating equilibrium models of sort- gies using supermarket scanner data. Marketing Sci. 16(4) ing across locations. Econom. J. 117(518) 353–374. 315–337.Bell, D., J. Lattin. 1998. Shopping behavior and consumer prefer- Newey, W. K., D. McFadden. 1999. Large sample estimation and ence for store price format: Why “large basket” shoppers prefer hypothesis testing. D. McFadden, R. Engle, eds. Handbook EDLP. Marketing Sci. 17(1) 66–88. of Econometrics, Vol. 4, Chap. 36. Elsevier, North-Holland,Bell, D., T. Ho, C. S. Tang. 1998. Determining where to shop: Amsterdam, The Netherlands, 2113–2245. Fixed and variable costs of shopping. J. Marketing Res. 35(3) Orhun, A. Y. 2005. Spatial differentiation in the supermarket indus- 352–369. try. Working paper, University of California.Bliss, C. 1988. A theory of retail pricing. J. Indust. Econom. 36(4) Ortmeyer, G., J. Quelch, W. Salmon. 1991. Restoring credibility to 375–391. retail pricing. Sloan Management Rev. 33(1) 55–66.Brock, W., S. Durlauf. 2001. Discrete choice with social interactions. Rust, J. 1987. Optimal replacement of GMC bus engines: An empir- Rev. Econom. Stud. 62(2) 235–260. ical model of Harold Zurcher. Econometrica 55(5) 999–1033.Dhar, S. K., S. J. Hoch. 1997. Why store brand penetration varies by Seim, K. 2006. An empirical model of firm entry with endogenous retailer. Marketing Sci. 16(3) 208–227. product-type choices. RAND J. Econom. Forthcoming.Einav, L. 2003. Not all rivals look alike: Estimating an equilib- Shankar, V., R. Bolton. 2004. An empirical analysis of determinants rium model of the release date timing game. Working paper, of retailer pricing strategy. Marketing Sci. 23(1) 28–49. Stanford University. Silber J., P. Rosenbaum, R. Ross. 1995. Comparing the contributionsEllickson, P. 2006. Quality competition in retailing: A structural of groups of predictors: Which outcomes vary with hospital analysis. Internat. J. Indust. Organ. 24(3) 521–540. rather than patient characteristics. J. Amer. Statist. Assoc. 90(429)Gowrisankaran, G., J. Krainer. 2004. The welfare consequences 7–18. of ATM surcharges: Evidence from a structural entry model. Singh V., K. Hansen, R. Blattberg. 2006. Market entry and consumer Working paper, Washington University. behavior: An investigation of a wal-mart supercenter. Market-Harsanyi, J. 1973. Games with randomly disturbed payoffs: A new ing Sci. 25(5) 457–476. rationale for mixed-strategy equilibrium points. Internat. J. Smith, H. 2006. Supermarket choice and supermarket competition Game Theory 2(1) 1–23. in market equilibrium. Rev. Econom. Stud. Forthcoming.Hoch, S., X. Dreze, M. Purk. 1994. EDLP, Hi-Lo, and margin arith- Sweeting, A. 2004. Coordination games, multiple equilibria, and metic. J. Marketing 58(4) 16–27. the timing of radio commercials. Working paper, NorthwesternHotz, J., R. Miller. 1993. Conditional choice probabilities and University. the estimation of dynamic models. Rev. Econom. Stud. 60(3) Varian, H. 1980. A model of sales. Amer. Econom. Rev. 70(4) 651–659. 497–531. Watson, R. 2005. Product variety and competition in the retail mar-Kumar, N., R. Rao. 2006. Using basket composition data for intelli- ket for eyeglasses. Working paper, University of Texas. gent supermarket pricing. Marketing Sci. 25(2) 188–199. Zhu, T., V. Singh, M. Manuszak. 2005. Market structure and compe-Lal, R., R. Rao. 1997. Supermarket competition: The case of every- tition in the retail discount industry. Working paper, Carnegie day low pricing. Marketing Sci. 16(1) 60–81. Mellon University, Pittsburgh.