Estimating Network Economies in Retail Chains: A Revealed Preference Approach. Duke University working paper

Many industries are characterized by a small number of firms making many entry decisions over a large choice set. Nowhere is this more true than in the case of the dis-count retail sector, where Target, Wal-Mart, and Kmart compete in a national market. Traditional discrete choice models of firm entry are ill-suited to this high dimensional choice problem. We instead draw upon recent innovations in the application of max-imum score estimators to models of revealed preference (Bajari and Fox (2006), Fox (2007)). Unlike previous work on this sector, our approach allows us to consider any number of potential rivals, any number of stores per location, the endogeneity of the distribution network, and unobserved (to the econometrician) location attributes that might cause firms to cluster their stores. Moreover, we show how recent innovations in set identification and inference can be used to separate the role of these unobservables from observed location attributes like population. We find that all firms (especially Target and Kmart) find it advantageous to cluster stores around distribution centers. Conditional upon that clustering, however, they find it costly to locate their stores in close proximity to one another (i.e., an “own business stealing ” effect). All firms (es-pecially Kmart) find it even more costly to locate in close proximity to a rival. Both of these strategic effects are understated if unobservable market attributes are ignored. Using counterfactual simulations, we explore the role of the distribution network in de-termining the level of retail competition experienced by consumers in markets of varying size. Density economies arising from the distribution network lead consumers in small, isolated markets to be underserved by Target, giving Wal-Mart more market power and the ability to raise prices there