Network-based predictions of retail store commercial categories and optimal locations

I study the spatial organization of retail commercial activities. These are organized in a network comprising “antilinks,” i.e., links of negative weight. From pure location data, network analysis leads to a community structure that closely follows the commercial classification of the U.S. Department of Labor. The interaction network allows one to build a “quality” index of optimal location niches for stores, which has been empirically tested.

Figure 1

(Color online) Map of Lyon showing the location of all the retail stores, shoe stores, furniture dealers, and drugstores.

Figure 2

(Color online) The landscape defined by the quality index is closely correlated to the location decisions of bakeries. (a) The 19 bakeries that closed between 2003 and 2005 had an average quality of $-2.2\times {10}^{-3}$ to be compared to the average of all bakeries $(4.6\times {10}^{-3})$, the difference being significative with probability 0.997. Taking into account the small number of closed bakeries and the importance of many other factors in the closing decision (family problems, bad management, etc.), the sensitivity of the quality index is remarkable. (b) Concerning the 80 new bakeries in the 2005 database (20 truly new, the rest being an improvement of the database), their average quality is $-6.8\times {10}^{-4}$, to be compared to the average quality of all possible sites in Lyon $(-1.6\times {10}^{-2})$, a difference significant with a probability higher than 0.9999.