Product variety is an important strategic tool that firms can use to attract customers and respond to competition.

Size: px
Start display at page:

Download "Product variety is an important strategic tool that firms can use to attract customers and respond to competition."

Transcription

1 MANAGEMENT SCIENCE Vol. 57, No. 6, June 0, pp ssn essn do 0.87/mnsc INFORMS Managng Product Varety and Collocaton n a Compettve Envronment: An Emprcal Investgaton of Consumer Electroncs Retalng Charlotte R. Ren Krannert School of Management, Purdue Unversty, West Lafayette, Indana 47907, rren@purdue.edu Ye Hu Bauer College of Busness, Unversty of Houston, Houston, Texas 7704, yehu@uh.edu Yu Jeffrey) Hu Krannert School of Management, Purdue Unversty, West Lafayette, Indana 47907, yuhu@purdue.edu Jerry Hausman Department of Economcs, Massachusetts Insttute of Technology, Cambrdge, Massachusetts 04, jhausman@mt.edu Product varety s an mportant strategc tool that frms can use to attract customers and respond to competton. Ths study focuses on the retal ndustry and nvestgates how stores manage ther product varety, contngent on the presence of competton and ther actual dstance from rvals. Usng a unque data set that contans all Best Buy and Crcut Cty stores n the Unted States, the authors fnd that a store s product varety.e., number of stock-keepng unts) ncreases f a rval store exsts n ts market but, n the presence of such competton, decreases when the rval store s collocated wthn one mle of the focal store). Moreover, collocated rval stores tend to dfferentate themselves by overlappng less n product range than do noncollocated rvals. Ths smaller and more dfferentated product varety may be because of coordnated nteractons between collocated stores. In summary, ths paper presents evdence of both coordnaton and competton n retalers use of product varety. Key words: product varety; competton; collocaton; dfferentaton Hstory: Receved December 3, 008; accepted January 4, 0, by Bruno Cassman, busness strategy. Publshed onlne n Artcles n Advance Aprl 9, 0.. Introducton Managers n many ndustres use product varety as a strategc lever Bayus and Putss 999, Sorenson 000). Ths lever appears ncreasngly appealng n the retal ndustry, where more and more retalers have adopted prce match guarantees e.g., Best Buy, Sears, Staples, Vons, and sometmes Walmart) and cannot resort to temporary prce reductons as effectvely as they mght have n the past to attract consumers. Consderable research has studed why varety mght theoretcally beneft the frm e.g., Kahn 998a, b; Lancaster 990, 998) and examned emprcally the benefts e.g., Kekre and Srnvasan 990) and costs of expandng varety e.g., Randall and Ulrch 00). However, lttle research has focused on competton as a key determnant of product varety. Workng n the context of retal competton for consumer electroncs, we buld on pror efforts e.g., Olvares and Cachon 009, Watson 009) to study how competton nfluences product varety decsons. We extend these efforts by explorng two prevously gnored varables: collocaton and product range overlap. We address the followng questons: wll a store change ts level of product varety f a key compettor s present versus not present) n ts market area? Gven such competton, does t matter f the competng store s collocated? Apart from the level of product varety, wll collocaton nfluence the extent to whch the focal store s product assortment overlaps wth the competng store? We consder explctly whether collocaton, whch strengthens competton and also may expand the market, nfluences product varety over and above the general presence of competton. Conventonal studes on spatal competton presume that geographc proxmty ncreases competton, and heghtened competton reduces rents for all Chung and Kalnns 00, p. 969). But when collocated, frms also can enjoy agglomeraton gans, because geographc concentraton helps consumers reduce ther search costs Stahl 98), and thus heghtens demand for collocated stores Cachon et al. 008, McCann and Folta 008). 009

2 00 Management Scence 576), pp , 0 INFORMS In part because of the agglomeraton benefts, many drect rvals such as CVS and Walgreens, Home Depot and Lowe s, and Best Buy and Crcut Cty locate n the same shoppng plaza or open stores across a street. Jackson Lan, the founder and former owner of the PC Club chan, was referred to by Stavro 999) n a Los Angeles Tmes artcle as the strategst next door, because from 99 to 999 he strategcally located hs 9 PC Club stores wthn walkng dstance of CompUSA, Best Buy, or Fry s Electroncs. Hs ratonale was that ths copycat locaton strategy not only helped PC Club attract foot traffc but also avoded overspendng on advertsng campagns and market research about where to locate. Our focus on collocated competton allows us to see how the nterplay of these contrastng forces affects rval stores product varety decsons. Moreover, no prevous studes have ever examned whether collocated compettors dfferentate more by reducng the extent of product range overlap than do noncollocated compettors. We probe both the level and composton of product varety, that s, the number of varants and the proporton of overlappng products wth compettors. Ths ntal exploraton of product range overlap and our unque focus on collocaton consttute the prmary contrbutons of our research. Our research s further dstnct from pror work n that our data set covers every retal outlet of the two largest specalty consumer electroncs retal chans n the Unted States n 006, Best Buy and Crcut Cty. Wth ths extensve data set, we provde a more expansve and pertnent research settng than most prevous work e.g., Watson 009). Snce the tme of our data collecton, Crcut Cty has declared bankruptcy and exted the market. However, ths ndustry settng n 006 offered two key benefts for our study. Frst, the compettve dynamcs between Best Buy and Crcut Cty showcased the mportance of product varety and were potentally representatve of many pars of U.S. and nternatonal retal chans e.g., CVS versus Walgreens, Walmart versus Carrefour). Second, the locaton and n-store product varety nformaton for each Best Buy and Crcut Cty store was avalable on ther webstes and updated frequently. Advanced computer search and data collecton technques enabled us, through great effort, to retreve and comple nformaton wth better relablty than would be possble through manual data collecton. On November 0, 008, Crcut Cty fled for bankruptcy protecton under Chapter, followed by store closures and layoffs. When we frst collected the data n November 005, Crcut Cty was n no sgn of trouble. Accordng to Crcut Cty s quarterly earnngs flng wth the U.S. Securty and Exchange Commsson, durng the last quarter of 005, t was stll proftable net earnngs $0 mllon) and ssung dvdends. Our nvestgaton reveals that a store s product varety ncreases sgnfcantly when a rval store appears n the local geographc market wthn a 0-mle radus of the focal store). However, gven the presence of such competton, the focal store s product varety decreases relatve to noncollocated competton when the rval store s collocated wthn one mle). Moreover, collocated stores are more nclned to dfferentate themselves by overlappng less n product assortments than do pars of dstant rvals. Ths smaller and more dfferentated product varety mples the possblty of coordnated nteractons between collocated stores. Ther coordnated behavor can blunt ther competton and allow both stores to reap agglomeraton benefts by attractng more consumers wth lower stockng costs. In summary, we document emprcally that both coordnaton and competton wth rvals occur n retalng and that product varety, n terms of both level and overlap, can be managed strategcally to adapt to the compettve envronment. We organze the remander of ths paper as follows: We revew lterature on the motves for and consequences of changng product varety n. In 3, we descrbe a smple model to llustrate our ntutons and propose several hypotheses about how a store s product varety level and overlap) mght change accordng to the competton and collocated competton t faces. We outlne the emprcal methodology n 4 and present the emprcal results n 5. Fnally, we dscuss mplcatons of the key results, our contrbutons and lmtatons, and some drectons for further research.. Pror Lterature Research from many dscplnes, ncludng economcs, marketng, operatons management, and strategy, ndcates that product varety can help frms attract customers and respond to competton. Prevous emprcal studes report the benefts of hghvarety strateges, such as sales ncreases Kekre and Srnvasan 990), hgher prces Pgou 90), and enhanced survval rates Sorenson 000). Yet launchng and mantanng a large product varety ncurs consderable costs, because of hgher nventory levels Kekre 987), the loss of scale economes, and the mposton of supply-chan market medaton Randall and Ulrch 00). The benefts of product varety may derve from several sources. Marketng scholars have dentfed We avod the term mplct colluson, whch mples anttrust volatons. The terms coordnaton and coordnated nteracton come from the Horzontal Merger Gudelnes of the U.S. Department of Justce and the Federal Trade Commsson.

3 Management Scence 576), pp , 0 INFORMS 0 consumer-based motvatons for an ncrease n product varety for revews, see Kahn 998b, McAlster and Pessemer 98). If each consumer knows hs or her preference precsely n a product category and chooses the same opton repeatedly, more varety n [the] product lne wll make t more lkely that each consumer fnds exactly the opton he or she desres Kahn 998a, p. 46). Alternatvely, when consumers make dfferent choces over tme, especally n lownvolvement, low-rsk product categores, they may seek varety to meet ther ntrnsc drve for stmulaton Raju 980) or satsfy ther curosty about novel thngs Hrschman 980). In both scenaros, greater product varety can ncrease customer satsfacton and loyalty. Lancaster 998) also suggests some nformaton- and producer-based motvatons for ncreasng varety. An ncumbent frm may offer product varety that exceeds some long-run market equlbrum, p. 4) f t can attan economes of scope by dong so or f t lacks nformaton about consumer preferences and therefore needs to offer as many products as possble to fnd out whch ones consumers prefer. Moreover, greater product varety can establsh barrers to entry, n that the ncumbent frm preemptvely flls all potental market gaps n whch an entrant could have entered, as emprcally documented by Putss and Bayus 00) n the personal computer ndustry. Despte ths good understandng of the consequences of and theoretcal motves for ncreasng product varety, lttle emprcal research has examned product varety as a strategc tool for respondng to competton. Most studes focus on prcng, not product varety, as a compettve nstrument e.g., Mazzeo 00, McGahan and Ghemawat 994, Thomadsen 007). The few extant studes on product varety competton focus on the mpact of compettors actons and market structure on frm-level varety. For example, Bayus and Putss s 999) nvestgaton of the personal computer ndustry durng shows that f compettors broaden ther product lne, frms also should ncrease ther product varety. Berry and Waldfogel 00) examne commercal rado statons n 43 U.S. markets n 993 and 997 and fnd that the greater concentraton of ownershp n a market, wrought by the 996 Telecommuncatons Act, was assocated wth an ncrease n per-frm product varety.e., number of dfferent programmng formats relatve to the number of compettors n a market). Somewhat more related to our study, n ther nvestgaton of more than 00 U.S. General Motors dealershps, Olvares and Cachon 009) fnd that dealers carry more nventory when they face more competton measured as the number of dealershps n a local geographc market). Watson s 009) analyss of the dsplay nventores of eyewear retalers n the mdwestern Unted States presents contrastng results at two levels: At the market level, average perretaler varety decreases wth the number of rvals; at the retaler level, when more rvals are located nearby, retalers frst stock more product varety before eventually reducng the level. Buldng on these two recent studes, we am to pant a rcher pcture of product varety competton by explorng two new drectons: collocated competton and the extent of assortment overlap. 3. Model and Hypotheses We model how stores change ther product varety n a market that ncludes two drect, mutually acknowledged competng retal stores Chen 996). 3 Because many retal stores, such as Best Buy and Crcut Cty, offer prce match guarantees, we do not focus on prce competton 4 and nstead place our emphass on how rval stores nteract wth one another by choosng approprate product varety levels. 3.. The Model Consder a local geographc market consstng of two competng stores that carry product varetes V j and V to attract consumers j =, j). We begn by specfyng the demand functon of each store based on a set of assumptons. Frst, a store s absolute product varety drves up ts own demand by ncreasng the lkelhood that store vstors can fnd products to match ther tastes Kahn 998a). We use V to descrbe how store s product varety nfluences ts demand. Second, carryng more product varety than a compettor store ncreases the lkelhood that consumers wll choose to vst the focal store, whch n turn ncreases the focal store s market share and demand Shugan 989). We use V /V j to capture how the relatve varety between store and j nfluences store s demand, such that a larger represents a more 3 Chen 996) posts that a compettor analyss s based on two factors: whether frms have smlar types and amounts of resources resource smlarty) and whether they compete n many markets that are mportant to them market commonalty). If frms have both hgh resource smlarty and market commonalty, they are drect and mutually acknowledged compettors e.g., Sony and Toshba, Coke and Peps, Best Buy and Crcut Cty). 4 To valdate the pont, we collected prce data n March 006 for all Best Buy and Crcut Cty stores n two product categores: dgtal cameras and televsons. We conducted pared t-tests on the prces of dgtal cameras and televsons and found no sgnfcant dfference n prce p = 053 for dgtal cameras, 0.66 for televsons). Specfcally, for dgtal cameras, the average pared prce dfference between the two chans s 0.3% of the prce, and 47% of the matched products have same prces at both chans. For televsons, the average pared prce dfference s 0.45% of the prce, and 36% of products have equal prces. Note that even though we observe very small dfferences n regular prces, larger dfferences n promotonal prces mght arse.

4 0 Management Scence 576), pp , 0 INFORMS compettve market. Thrd, f the competng store s located close enough to the focal store, an agglomeraton effect may occur. Ths agglomeraton effect suggests that addtonal product varety by the proxmate rval helps ncrease the focal store s demand. to capture ths agglomeraton effect, wth a larger ndcatng a stronger agglomeraton effect. Puttng these components together, we follow Shugan 989) and adopt the commonly used multplcatve functonal form to set up store s demand functon: We use V j Q = KV V V j ) V j j = j ) where Q s store s demand; K s a constant that represents store s absolute market potental; denotes the margnal effect of own varety on demand; represents the level of compettve ntensty; and s the margnal agglomeraton effect on demand. In a compettve market, all these parameters are strctly postve, whereas n a monopoly market that ncludes only one store, both competton and the agglomeraton effect dsappear = 0 and = 0). The proft of store n turn s V V j = KV V V j ) V j V j = j ) where V denotes store s costs of carryng product varety V e.g., nventory, transportaton, opportunty cost of shelf space). 5 Based on the demand and proft functons, next we analyze three questons: ) How does a store change ts product varety f a compettor exsts n ts local geographc market? ) In a compettve market, do collocated rvals dfferentate more n ther product range than do dstant rvals? 3) How do rval stores that compete n the market change the level of product varety f they are collocated? We provde the proofs n the appendx. 3.. Product Varety and the Presence of Competton In ths subsecton, we compare store-level product varety n a compettve market to that n a monopoly market. In a compettve market, each compettor chooses ts own product varety to maxmze ts own proft. In other words, the problem for each store s max V V V j j = j. Solvng the proft maxmzaton problems of both stores allows us to propose the followng: Proposton. In a compettve market there exsts a symmetrc equlbrum: Each store chooses the optmal 5 The coeffcent s added for mathematcal convenence and does not change the results substantvely. We use a quadratc cost functon for the unqueness of the soluton. Such a cost functon s commonly employed to descrbe the ncreasng costs of operaton as nput ncreases Mas-Colell et al. 995). varety V d = K+ / =. In a monopoly market, the optmal product varety of the monopolstc store s V m addton, V d = K /. It holds that V d / > 0. > V m. In Proposton suggests that the optmal varety of a compettve store s always greater than that of a monopolstc store. More generally, a store s optmal product varety ncreases wth the level of compettve ntensty, and a monopoly market s a specal case when competton dsappears. The ntuton behnd ths proposton holds that more product varety contrbutes to store demand and proft because t s more lkely that each consumer fnds hs or her most preferred style Kahn 998a), and more varety allows each consumer to enjoy a dversty of optons McAlster and Pessemer 98). In a compettve market, any store that lags behnd ts compettor n provdng varety faces the consequences of lower demand and lower profts. Because of such an arms race, both stores must provde hgher levels of varety than ether would have provded n a monopoly market of the same sze. Accordngly, we emprcally test the followng hypothess: Hypothess H). A store s product varety ncreases f a rval store competes n the same market Dfferentaton and Collocated Competton We now focus on the case of a compettve market. We explore the possblty that rval stores may dfferentate themselves usng ther product ranges, and we examne how the extent of dfferentaton vares when rval stores are collocated. A store can dfferentate by carryng fewer products that also are avalable at the compettor store overlappng products) and/or ncreasng the number of products that the compettor does not carry nonoverlappng products). Dfferentaton softens competton Mazzeo 00, Porter 99). It can also strengthen the agglomeraton effect: If a store adds an overlappng product, the total product varety of both stores remans unchanged. In contrast, f a store dfferentates by ntroducng a nonoverlappng product, the total product varety ncreases, whch further reduces consumer search costs and enhances the attractveness of the overall marketplace. We ntroduce a parameter to capture the mpact of dfferentaton or the reverse of product range overlap) on competton and the agglomeraton effect. 6 We assume that, wth = representng no dfferentaton and > denotng the exstence of 6 We treat dfferentaton denoted by the parameter ) as a choce varable that needs to be coordnated by both stores. Because t can be adjusted only slowly through coordnaton, n the short run t can be treated as fxed, as we do n dervng our Proposton 3. Moreover, note that for Proposton, we only model a one-perod

5 Management Scence 576), pp , 0 INFORMS 03 dfferentaton. The proft functon n Equaton ) can be rewrtten as follows: ) / V V j = KV V V j V V j j = j 3) where we dvde by to ndcate that dfferentaton reduces competton and multply by to show that dfferentaton ncreases the agglomeraton gan. 7 Agan we solve the proft maxmzaton problems of the two rval stores and obtan each store s optmal varety and correspondng proft. Proposton. A symmetrc equlbrum exsts such that each store chooses the optmal varety to earn proft [ V = K + )] / [K = + )] / [ + / ] = As a result, we have a) / > 0. That s, a store s equlbrum proft ncreases wth the level of dfferentaton. b) / > 0 / > 0 That s, the postve mpact of dfferentaton on proft s greater when there s more competton and an agglomeraton effect. Proposton a) suggests that because more dfferentaton leads to hgher store proft, stores have an ncentve to dfferentate. Proposton b) ndcates that ths ncentve s stronger when stores face hgher levels of competton and agglomeraton gans, two features assocated wth collocaton. On the one hand, when rval stores are collocated e.g., n the same shoppng plaza), transport cost across collocated stores declnes to nearly nothng, whch should reduce consumers swtchng costs and thereby ncrease the ntensty of the rvalry between the sellers ). On the other hand, the demandheghtenng agglomeraton effect emerges wth collocaton ). Dfferentaton helps dampen the more game n whch rval stores may or may not coordnate to such extent that makes the level of dfferentaton consstent wth the equlbrum. Ths type of coordnaton, however, s most lkely to occur when rval stores nteract repeatedly, as s true n our emprcal study. 7 To model the dfferental mpacts of dfferentaton on and, we also could wrte / and ). In ths case, and ndcate the dfferental magntude of mpact. For parsmony though, we assume = and = wthout loss of generalty. ferce competton between collocated stores and further enhances the agglomeraton gans by ncreasng the collocated regon s total product varety, thereby contrbutng to store profts to a greater extent. As a result, collocated stores are more nclned than dstant rvals to dfferentate. In lne wth Proposton, we submt the followng hypothess: Hypothess H). Gven competton n a market, collocated rvals are more lkely than dstant rvals to dfferentate by reducng ther product range overlap Product Varety and Collocated Competton Many rval retalers have coexsted for a long tme and are lkely to nteract repeatedly. Some rval retalers even become ncreasngly smlar n ther resources and markets Chen 996), as exemplfed by Best Buy and Crcut Cty n our emprcal settng. Repeated nteracton among a few smlar players encourages coordnaton Trole 988, p. 40). In choosng ts compettve tactcs then, a store must take nto account not only the possble ncrease n shortterm profts but also the possblty of long-run losses from the rval s retalaton n the future. Moreover, when rval retalers encounter each other n multple geographc markets, they need to wegh gans n one market aganst dangers of retalaton n other markets. Such multmarket contact can blunt the ncentves for rvalry and further facltate coordnaton Trole 988, p. 43). We now use a smple case to demonstrate why coordnaton can result n lower product varety at each store. If two rval stores coordnate fully, they no longer behave noncooperatvely to maxmze ther ndvdual proft. Instead, ther combned problem becomes choosng product varety levels V c ) to maxmze jont profts. For a gven level of dfferentaton ), the common objectve shared by two rval stores s max V V V V + V V. By solvng ths problem we are able to state the followng: Proposton 3. For a gven level of, we have V c < V. That s, the optmal product varety when both stores fully coordnate s smaller than that when each store acts noncooperatvely wth the sole objectve of maxmzng ts own proft. Proposton 3 shows that the respectve product varety of two fully coordnated rvals s smaller than the optmal level chosen by two rvals that do not coordnate.e., compettve level of varety). Because two fully coordnated stores choose varety levels to maxmze ther jont profts, they are not engaged n a harmful arms race to add extra product varety and attract customers away from the competton, whch also would ncur hgher stockng costs. Instead, two rval stores behave as f they were one bg store, each

6 04 Management Scence 576), pp , 0 INFORMS offerng a lower product varety to reduce stockng costs. Such full coordnaton s more lkely n collocaton condtons. When collocated, rval stores forego the possblty of dfferentatng through locatons and confront more ferce competton, whch can reduce each other s proftablty Chung and Kalnns 00). They thus have greater ncentves to develop an mplctly coordnated market Shugan 985) to mtgate such head-to-head competton. If one of the stores defects from the coordnated nteracton by ncreasng product varety, t may enjoy an mmedate gan from the defecton, but t wll suffer a punshment loss n the future because the collocated rval can detect ts defecton easly and retalate by escalatng ts own product varety. Compared wth dstant rvals, collocated rvals can access each other s updated status, carryng costs, and planned product varety moves more easly through n-store vsts, nformal daly communcatons, and personal networks among employees, thereby reducng the nformaton lag and encouragng rapd retalaton. The smaller nformaton lag and qucker retalaton assocated wth collocaton make the mplct coordnaton more lkely to be sustanable Trole 988, p. 4). Consderng the collocated stores have both the motvaton and the capablty to coordnate effectvely, we conclude that collocaton may lead to lower level of product varety. Hypothess 3A H3A). Gven the exstence of a rval store n a market, a focal store s product varety decreases f the rval store s collocated. Yet fully coordnated stores also mght choose a level of dfferentaton that dffers from the level chosen by nonfully coordnated stores. In other words, may not be the same n the two cases. When we factor n ths potental varaton n dfferentaton, t s unclear whether V c < V stll holds.8 Ths adds ambguty to predctons about product varety level under collocaton. Intutvely, such ambguty can be explaned as a result of two ntertwned tendences of collocated rvals: on the one hand, compared wth dstant competng stores, collocated rvals tend to dfferentate more as predcted by Proposton ) by carryng fewer overlappng and/or more nonoverlappng products; on the other hand, they are also nclned to offer a smaller product varety to save stockng costs as dscussed n H3A). When rval stores are collocated, consumers can search both stores wth lttle transport cost and purchase from 8 Our proof n 3. of the appendx shows that V c < V may not hold when the level of dfferentaton ) n the fully coordnated case dffers from that n the nonfully coordnated case. ether, therefore what may be more mportant to consumers s the total product varety of both stores.e., the sum of two stores varetes less the number of overlappng products) rather than each store s varety. Two possble scenaros then emerge: If collocated rval stores dfferentate manly by carryng fewer overlappng products, even though each store offers a lower product varety, the total product varety of two collocated rvals can stll stay at the same level as that of dstant rvals. However, f collocated stores dfferentate prmarly by carryng more nonoverlappng products, then each store s respectve varety may not declne. Instead, wth the addton of nonoverlappng products, each store lkely offers a larger varety, as a result of whch the total product varety of the collocated regon becomes even hgher than that of the dstant rvals. Because ths second scenaro suggests the possblty that collocated rval stores may ncrease both ndvdual varety and dfferentaton, we submt a new hypothess that competes wth H3A: Hypothess 3B H3B). Gven the exstence of a rval store n a market, a focal store s product varety ncreases f the rval store s collocated. 4. Research Desgn 4.. Data We collected data of Best Buy and Crcut Cty n 006 from ther webstes, and Usng a Web crawlng program, we retreved the address of each Best Buy and Crcut Cty store n the Unted States ncludng Alaska and Hawa) as of March 006. Another Web crawlng program collected n-store product varety nformaton for each Best Buy and Crcut Cty store n a sngle product category: dgtal cameras. To ensure that the data crawlng processes wrapped up wthn the same day to mnmze any possble changes n the product nformaton, we used 0 computers that extracted webpages smultaneously. To verfy data accuracy, we ntervewed several store managers, who confrmed the consstency between ther actual n-store product varety and the nformaton lsted onlne. We also manually collected product varety nformaton n the dgtal camera category by vstng several stores; we found no dfference between the hand-collected nformaton and the data collected usng our Web crawlng program. In 006, Best Buy mantaned 70 stores and Crcut Cty had 69 stores, so we have,39 observatons for the dgtal camera product category. We delneate local markets accordng to the method proposed by Zhu and Sngh 009). The U.S. Census parttons each county nto subregons called census

7 Management Scence 576), pp , 0 INFORMS 05 Fgure Detals of a Local Market Tract Tract 3 Tract Tract 5 Tract 4 Tract 6 Store locaton Geometrc center of each census tract Harde 996), SKU choce s a more fttng descrpton than brand for consumers purchase decson processes, because consumers typcally choose among SKUs on the bass of varous product attrbutes, one of whch s the brand. We use a log transformaton because the number of SKUs exhbts approxmately log-normal dstrbutons n our data. To capture the extent of product range overlap, we compare and match the detaled product descrptons e.g., model numbers, major attrbutes) of both chans. We next count the number of overlappng SKUs for each store and ts nearest compettor COMPV). We then take nto account the possblty that f the unverse of SKUs n a partcular market s hgher, the lkelhood of overlap s larger by random probablty. Followng the sprt of the dartboard approach proposed by Ellson and Glaeser 997), we measure product range overlap for each par of stores as the raw number of overlappng SKUs, less the randomly drawn overlap: tracts. 9 Because the address of each store s avalable, we can map each store onto the correspondng census tract usng the Census Bureaus TIGER/Lne fles. We then draw a crcle of 0-mle radus around each store s locaton, wth the presumpton that ths dstance represents a relevant area n whch each store potentally competes wth other stores. In Fgure, we flesh out the detals of a local market, whch we defne as the 0-mle crcle around the focal store, across a number of tracts. In our data set, each market covers 35 tracts on average. Fgure depcts the locatons of all the stores n our data. 4.. Measures Before gong nto detal about varable measurements, we explan how we defne a par of compettors. Frst, usng each store s address, we fnd the lattude and longtude of each store. 0 Second, we calculate the sphercal dstance between a focal store and each store of the rval chan. Thrd, we select the rval store that s the shortest dstance from the focal store and defne these two stores as a par of compettors for our study. Our research consders two dependent varables. We measure the total product varety PV) of each store as the logged number of stock-keepng unts SKUs) n the dgtal camera product category. SKUs provde a good measure of product varety from the perspectve of consumers. Accordng to Fader and 9 More detaled nformaton about the defnton and basc characterstcs of census tracts s also avalable at 0 The prmary onlne source of our nformaton was terraserver-usa.com. Ths webste s now renamed as msrmaps.com. OVERLAP = logcompv lograndomcompv ) n n j = logcompv log 4) N where n denotes the number of SKUs avalable at the focal store, n j s the SKUs avalable at the rval store j, and N ndcates the total unverse of SKUs n the partcular market of store. All the elements of the calculaton are shown n Table. Whereas the actual number of overlappng SKUs s x, the expected number of overlappng SKUs based on probablty.e., RandomCOMPV ) s n /N n j /N N = n n j /N. Our two ndependent varables are ndcator varables that ndcate whether a compettor exsts COMP) and s collocated COLLOCATE) n a gven market. We code COMP as f the dstance between a focal store and ts nearest rval store s not greater than 0 mles that s, f a compettor of the focal store appears wthn the 0-mle crcle and 0 otherwse. Followng Rosenthal and Strange 003), we code COLLOCATE as f In some consumer electroncs categores, manufacturers put dfferent model numbers on dentcal products sold to dfferent retalers to soften downstream prce competton. Dgtal camera manufacturers sometmes do so across global markets. We do not observe such practces n the dgtal camera category across the two retal chans n the U.S. market though. In our data set, based on the product nformaton extracted, we fnd that no dentcal products are sold under dfferent names at these retalers. The SKUs dffer n tangble features, ncludng brand name, desgn type, megapxels, optcal versus dgtal zoom, and so on. We thank an anonymous revewer for pontng out ths possblty. Imagne the total unverse of SKUs s 00. One par of stores stocks 30 SKUs, and another par of stores stocks 50. For a par of stores, f we randomly assgn ther SKUs to the 00 possble cells throw darts at 00 cells), the lkelhood of overlap s smaller for the frst par than for the second par of stores, because we would throw two sets of 30 compared wth two sets of 50 darts.

8 06 Management Scence 576), pp , 0 INFORMS Fgure Locatons of Best Buy and Crcut Cty Retal Stores n Our Data Legend Best Buy Crcut Cty State boundares the nearest rval store s wthn one mle of the focal store, and 0 otherwse. By defnng these two varables, we can categorze our,39 observatons nto the followng types of market structures: 9 observatons appear n the no local competton category COMP = 0), and,38 observatons reveal local competton COMP = ); wthn ths compettve category, 609 observatons ndcate collocated competton COLLOCATE =, COMP = ), and 59 reveal noncollocated competton COLLOCATE = 0, COMP = ). As a control varable, we use an ndcator varable for Best Buy stores BESTBUY), whch captures the store-specfc characterstcs of the Best Buy chan, such as ts organzatonal culture, reputaton, standard customer servce, store cleanlness, employee tranng, and frendlness. We also add two marketlevel demographc varables calculated from census data: INCOME log of medum household ncome n a market) and POPDEN populaton densty n a market, or the sze of the populaton per square klometers/,000). Both INCOME and POPDEN relate to a Table Overlappng and Nonoverlappng SKUs n a Par of Rval Stores SKUs avalable n store j SKUs avalable n store Yes No Row sum Yes x n x n No n j x N n n j + x N n Column sum n j N n j N a a The total unverse of SKUs N) vares by local market. We also use the total unverse of SKUs across all geographc markets to measure N. That unverse s the same for both Best Buy and Crcut Cty, that s, 80 product varants n total. Our data analyss results hold wth ths alternatve defnton of the total unverse of SKUs. market s product varety; the former ndcates potental ncome constrants on purchasng behavor Hoch et al. 995), and the latter affects the market s demand level and relatve proftablty Watson 009). Fnally, we control for several demographc varables that may descrbe the market condtons, namely, average household sze HHSIZE), the fracton of the populaton wth a college educaton COLLEGE), the fracton of the populaton over 8 years of age ADULT), the fracton of male resdents MALE), and the fracton of consumers who are nonwhte NONWHITE). The demographc data reflect census nformaton for tracts wthn the local market, whch we aggregate to the market level by takng weghted averages usng tract-level populaton as the weght. If store s local market defned as ts 0-mle radus) overlaps wth the markets of many other stores, we dentfy common tracts shared by these n + stores. We then dvde the demographc data of those common tracts by n +, such that the shared tracts are equally allocated to nearby stores. Ths adjustment ensures each store s local market s mutually exclusve from other stores markets. In Table, we provde summary and descrptve statstcs for all the varables n ths study. The product varety and overlap profle for each type of market structure s presented n Table Statstcal Method To estmate how competton affects a store s product varety, we would normally estmate the model: PV = 0 + COMP + X + 5) where COMP s the key ndependent varable that ndcates whether a compettor for a focal store

9 Management Scence 576), pp , 0 INFORMS 07 Table Summary and Descrptve Statstcs of Varables Varable Descrpton Mean SD Mn Max Dependent varables PV Logged number of a store s product varety SKUs) OVERLAP Logged number of common products.e., products carred by a store and ts nearest compettor) mnus logged number of randomly drawn common products Independent varables COMP Indcator = f a store s dstance from ts nearest compettor 0 mles COLLOCATE Indcator = f a store s dstance from ts nearest compettor mle Control varables BESTBUY Indctor varable for Best Buy stores INCOME Medan household ncome n a market US$, logged unts) POPDEN Populaton densty n a market populaton sze per square klometers/,000) COLLEGE Decmal fracton of the populaton wth a college educaton n a market ADULT Decmal fracton of the populaton older than 8 years n a market MALE Decmal fracton of male consumers n a market NONWHITE Decmal fracton of consumers who are nonwhte n a market HHSIZE Average household sze n a market SUPERURBAN Indcator varable for superurban market Note. N = 39; March 006 dgtal camera full sample. appears n a gven market, s a coeffcent vector, and X s a vector of control varables that mght nfluence product varety e.g., demographc varables that descrbe market-level consumer taste heterogenety). However, we must allow for the possblty that the presence of competton n the market COMP = ), lke product varety, depends on market-level control varables. Unobservables that are not ncluded n the vector of control varables n Equaton 5) can yeld based and nconsstent estmates of the coeffcent for COMP usng an ordnary least squares OLS) estmaton Greene 990). To address the potental endogenety of COMP, we use the nstrumental varables IV) method. We frst apply a probt model to estmate the probablty that there exsts a compettor for a focal store n a gven market as a functon of the exogenous pre- determned) demographc varables. The functonal form we choose s PrCOMP = Z = Z 6a) where s the cumulatve dstrbuton functon of the standard normal dstrbuton and Z ncludes BESTBUY, INCOME, POPDEN, and the demographc varables MALE, ADULT, COLLEGE, NONWHITE, and HHSIZE) that may affect market demand and thus are related to the level of competton n a gven market. We then use the predcted value from Equaton 6a) COMP) as an nstrumental varable for COMP n Equaton 6b), the equaton of prmary nterest: PV = IV 0 + IV COMP + X IV + 6b) where X ncludes BESTBUY, INCOME, POPDEN, and four demographc varables MALE, ADULT, Table 3 Product Varety and Overlap by Market Structure Market Average no. of Average no. of Average no. of Average no. of structure No. of obs. a store-level SKUs overlappng SKUs b nonoverlappng SKUs b total SKUs b Monopoly N/A N/A N/A Noncollocated ) competton Collocated ) competton a In column ), the number n each parentheses refers to the number of observatons n whch the overlappng, nonoverlappng, and total SKUs data are avalable. b Columns 4) 6) apply only to the compettve markets where for each par of rvals, we compare the product range overlap and calculate the total varety of both stores. Note that the total varety s the sum of both stores SKUs less the number of overlappng SKUs.

10 08 Management Scence 576), pp , 0 INFORMS Table 4 Models of the Impact of Competton and Collocated Competton on Product Varety All areas ncluded Superurban effect controlled ) ) 3) 4) 5) 6) Dependent varable PV OVERLAP PV PV OVERLAP PV CONSTANT COMP a COLLOCATE a BESTBUY INCOME POPDEN COLLEGE ADULT MALE NONWHITE SUPERURBAN No. of obs.,39,0 b,38,39,0 b,38 Durbn-Wu- = 787 = 99 = 3 = 79 = 690 = 68 Hausman test p = 0005) p = 000) p = 0000) p = 0005) p = 0009) p = 0009) Note. Standard errors are n parentheses. a Predcted values of COMP and COLLOCATE from the frst-stage probt analyses are the nstrumental varables for COMP and COLLOCATE, respectvely. b The compettve subsample for models 3) and 6) contans,38 observatons, but only,0 contan product range overlap nformaton, so n = 0 for models ) and 5). Sgnfcant at the % level; sgnfcant at the 5% level; sgnfcant at the 0% level. COLLEGE, and NONWHITE) that reflect a market s dversty n gender, age, educaton, and ethncty and thus relate to the market s aggregate taste for varety. Such IV estmatons requre at least one extra explanatory varable that nfluences the frst stage but not the second stage, whch s often referred to as the excluson restrcton requrement Wooldrdge 00). We nclude the demographc varable HHSIZE n the frst-stage probt model but exclude t from the second stage. Household sze should relate to market demand and affect the number of stores n ths market, whch suggests that we should nclude ths varable n the probt model. Product varety, the dependent varable n the second stage, prmarly depends on consumer heterogenety n a market. But HHSIZE does not capture the heterogenety across households and therefore should not drectly nfluence product varety. 3 3 We note two addtonal ponts regardng the excluson restrcton requrement. Frst, ths restrcton s not absolutely requred n our To examne how a collocated compettor, compared wth a noncollocated compettor, affects product varety and product range overlap, we conduct smlar IV estmatons that account for the endogenety of COLLOCATE on a subsample of only compettve cases wth the followng second-stage functonal forms: OVERLAP = IV 0 + IV COLLOCATE +X IV + 7) PV = IV 0 + IV COLLOCATE + X IV + 8) case, because we use probt a nonlnear, bnary response model n the frst stage to generate our nstrumental varable Amemya 985). Although the nonlnearty of the frst-stage probt model allows our second-stage equaton to be techncally dentfed, we go further and apply the restrcton requrement to make the source of the dentfcaton clearer. Second, our estmaton results reman ntact regardless of whether we nclude ths extra varable n the frst stage.

11 Management Scence 576), pp , 0 INFORMS Results and Robustness Checks We provde the results of the IV estmatons n whch we examne the mpact of COMP or COLLOCATE on product varety n Table 4. 4 We report the results of Durbn-Wu-Hausman tests, whch confrm the endogenety of the COMP and COLLOCATE varables and ndcate that ther coeffcents, f estmated by OLS, should dffer sgnfcantly from those of the IV estmatons. Model ) contans the IV estmaton results for the mpact of competton on product varety. As expected, the coeffcent of COMP s postve and hghly sgnfcant IV = 0045 wth a t-statstc of.65), n support of H; that s, the exstence of a competng store s assocated wth greater total product varety of a focal store. If we hold the other regressors equal, the average number of product varants n a compettve market COMP = ) s 5% =e 0045 = 005 hgher than n a market wthout competton COMP = 0). Model ) reports the IV estmatons of the mpact of collocaton on product range overlap for only compettve cases. The IV coeffcent of COLLO- CATE s negatve and statstcally sgnfcant IV = 0346, wth a t-statstc of.0), whch mples that a store s less lkely to carry overlappng products when the compettor s collocated. All else beng equal, the average extent of product range overlap by collocated compettors COLLOCATE = ) s approxmately 9% = e 0346 = 09) lower than that carred by noncollocated compettors COLLOCATE = 0). H s thus supported. Model 3) presents the IV estmaton results of the mpact of collocated competton on product varety for compettve cases. The coeffcent of the COLLOCATE varable n model 3) s negatve and sgnfcant IV = 04, wth a t-statstc of.8); that s, collocated compettors carry a smaller product varety than do noncollocated compettors, n support of H3A. When the other regressors reman constant, collocated compettors dsplay a level of product varety that s approxmately 9% = e 04 = 09 lower than noncollocated compettors. The control varables have some effects of nterest. For example, Best Buy stores dsplay a lower level of product varety than Crcut Cty stores, and when they collocate wth ther nearest compettors, they dfferentate slghtly more than do Crcut Cty stores by dsplayng a lower level of product range overlap. A hgher fracton of male resdents n a market s assocated wth lower product varety, whch mples that men s tastes tend to be more homogenous than are women s. In the full sample analyss model )), we also fnd that markets wth hgher ncome levels, 4 The results of the frst-stage probt models that we used to generate the nstrumental varables can be obtaned from the authors upon request. larger populaton szes, and larger fractons of adults dsplay hgher levels of product varety. We use several alternatve specfcatons to test the robustness of our results. Frst, there are some hghly urbanzed areas such as New York Cty, Chcago, and Boston, whch we label as superurban areas, where compettors may be more lkely to collocate wthn one mle. In other areas, competng stores may not collocate as much and nstead coexst only wthn 0 mles. Thus, the effects we have observed for the ndependent varables COMP and COLLOCATE mght reflect not the mpact of compettve proxmty but rather whether the store s located n a superurban area. To control for the effects of these superurban areas, we generate an ndcator varable SUPERUR- BAN that equals f the focal store s zp code s categorzed by Census 000 to be n a core based statstcal area CBSA) that can be subdvded nto two or more metropoltan dvsons and 0 otherwse). 5 The estmaton results wth ths varable ncluded, as shown n models 4) 6) n Table 4, stll support H, H, and H3A. We also run the analyss on the subsample that ncludes only the nonsuperurban areas. The results n models 7) 9) n Table 5 ndcate that all our major fndngs stll hold n ths subsample. Second, we change the collocaton cut-off pont from mle to 0.5 mles. The results usng ths alternatve measure, as reported n models 0) and ) n Table 5, support both H and H3A. We also use the dstance from the nearest rval as a contnuous measure for the degree of collocaton and fnd that geographcally closer rvals tend to overlap less and each carres a smaller product varety. Ths fndng s consstent wth the results when collocaton s measured as noncontnuous varables. Thrd, we defne a local geographc market as the 5-mle radus of a focal store. Wth a larger market, we can explore compettve dynamcs n nonsuperurban areas, where stores may locate farther away from one another. After recalculatng all the market-level demographc varables, we repeat the analyses from models ) ); all the results stll hold. Fnally, we note the possble mpact of the presence of Walmart, whch sells consumer electroncs too. Natonwde data ndcate that approxmately 5 Wth the zp code nformaton of each store, we collected the correspondng mcro, metro, and CBSA dvson data. A CBSA s the offcal term for a functonal regon based around an urban center of at least 0,000 people, based on standards publshed by the Offce of Management and Budget n 000. A CBSA s broken nto mcro and metro subcategores, dependng on populaton denstes. To be consdered a mcro area, the populaton must be between 0,000 and 50,000. A metro area s populaton must exceed 50,000. If a metro area contans an urbanzed area of at least.5 mllon people, t can be subdvded nto two or more metropoltan dvsons. Of our,39 stores, 385 are n areas wth metropoltan dvsons, whch we call superurban areas.

12 00 Management Scence 576), pp , 0 INFORMS Table 5 Selected Models for Robustness Check Only nonsuperurban areas ncluded Superurban effect controlled, collocate as 05 mles 7) 8) 9) 0) ) Dependent varable PV OVERLAP PV OVERLAP PV CONSTANT COMP a COLLOCATE a BESTBUY INCOME POPDEN COLLEGE ADULT MALE NONWHITE SUPERURBAN No. of obs b 77,0 c,38 Durbn-Wu- = 06 = 443 = 899 = 067 = 563 Hausman test p = 000) p = 0035) p = 0003) p = 000) p = 0000 Note. Standard errors are n parentheses. a Predcted values of COMP and COLLOCATE from the frst-stage probt analyses are the nstrumental varables for COMP and COLLOCATE, respectvely. b The compettve subsample that ncludes only nonsuperurban areas for model 9) contans 77 observatons, but only 754 contan product range overlap nformaton, so n = 754 for model 8). c The compettve-only subsample for model ) contans,38 observatons, but only,0 contan product range overlap nformaton, so n = 0 for model 0). Sgnfcant at the % level; sgnfcant at the 5% level; sgnfcant at the 0% level. 3,000 Walmart stores exsted n the Unted States n March 006 recall that there were 70 Best Buy and 69 Crcut Cty stores). Therefore, at least one Walmart store appears n every defned local market we study, and the presence of Walmart s a common factor for all our data observatons. Ths form of competton therefore cannot explan our fndngs. Moreover, our fndngs contnue to receve support after we control for the dstance as a measure of the extent of collocaton) from Walmart. 6. Dscusson and Conclusons 6.. Implcatons of Major Fndngs A store s product varety expands sgnfcantly f a rval store exsts n ts market. In a compettve market, each rval store has an ncentve to add product varety, beyond the level that t would have stocked had t been a monopolst. Ths larger varety can enhance customer satsfacton and loyalty, as well as encourage customers to swtch away from competng stores. Ths frst key fndng generally parallels exstng studes. For example, Lancaster 990) shows theoretcally that a monopoly, facng no competton, should produce the least product varety of any market structures. In a study of the vdeo rental market n a dstrct of Edmonton, Alberta, de Palma et al. 994) fnd that centrally located stores, whch face the most competton, tend to offer more product varety than do stores on the market boundares. In the presence of competton, collocated rval stores tend to dfferentate more than dstant rvals do by overlappng less n product range. Frms wthn the same ndustry often collocate e.g., hotels, Baum and Haveman 997; Chung and Kalnns 00; footwear producers, Sorenson and Auda 000; new