Spatial Differentiation in Retail Markets for Gasoline

This paper studies an empirical model of spatial competition. The main feature of my approach is to formally specify commuting paths as the ``locations'' of consumers in a Hotelling-type model of spatial competition. This modeling choice is motivated by the fact that consumers are moving across the market when consuming the product. Although this feature is perhaps more relevant for gasoline markets, this also applies to most retail markets since consumers are not immobile. The consequence of this behavior is that competition is not fully localized as in the standard address-model. In particular, the substitution patterns between stations depend in an intuitive way on the structure of the road network and the direction of traffic flows. Another feature of the model is that consumers' available options are directly linked with their commuting behavior. Consumers who commute more encounter more stations, observe more prices, and therefore may pay less on average; a result observed empirically (Yatchew and No [2001]). The demand-side of the model is estimated by combining a model of traffic allocation with econometric techniques used to estimate models of demand for differentiated products (Berry, Levinsohn and Pakes [1995]). The empirical distribution of commuters is computed with a shortest-path (or Dijkstra) algorithm, combining detailed data on the Qu\'ebec City road network with aggregate Origin-Destination commuting probabilities. The model's parameters are then estimated using a unique panel data-set on the Qu\'ebec City gasoline market from $1991$ to $2001$. The empirical analysis of the model clearly shows that adding commuting behavior is important for explaining the distribution of sales across the city. In particular, gasoline sales are poorly correlated with the population distribution (address-model) but more closely related to the distribution of commuting. The results of the structural model also reveal important differences between the home-location and the commuting models. Contrary to the commuting model, the estimated transportation cost is unrealistically low under the home-location model. The evaluations of market power under the two models are also different. The home-location model leads to markup estimates that are close to $50\%$ higher than the commuting model. The estimates of the structural model are also used to assess the relative market power of major retail chains over independents. The results indicate that the role of independents in this market is marginal, since on average they are operating low-value and high-cost stations.