Location and co-location in retail: a probabilistic approach using geo-coded data for metropolitan retail markets

In this paper, we employ geo-coded data at a fine spatial resolution for Sweden's metropolitan areas to assess retail co-location. Retail clusters and their place in urban space are assessed from several angles. The probability of a specific type of retail unit to be established in a 250 by 250 m square is modelled as a function of (i) the presence of other similar retail establishments, (ii) the presence of stores that belong to other retail sectors and (iii) other characteristics of the square area, and its access to demand in the pertinent urban landscape. The analysis clarifies which types of retail clusters one can expect to find in a metropolitan region, as well as their relationship to the urban landscape. We analyse three distinct types of stores: clothing, household appliances, and specialized stores. Stores with high intensities of interaction are co-located, and predominantly located close to the urban cores, consistent with predictions from bid rent theory and central place theory. We further document negative location tendencies between shops that sell frequently purchased products and shops that sell durables. Moreover, our results highlight the importance of demand in the close surroundings, which is particularly strong for small-scale establishments.