Mapping Inter-City Trade Networks to Maximum Entropy Models using Electronic Invoice Data

Abstract: We analyze the network of transactions among cities based on the electronic invoice database for the municipalities in the Cear\'a state, Brazil. It consists of approximately 3.7 billion records, registered during the period between the years 2016 to 2019. All the transactions are grouped in a unique dataset and represented as an asymmetrical adjacency matrix corresponding to a directed graph with connections weighted by the number of transactions among cities. Due to the large size of Cear\'a state, its unequal distribution of wealth, and spatially heterogeneous population, we initially determine communities of cities based on their mutual intensity of trades and verify to which extent their economic interests reflect a community cohesiveness. For the first task, we use the Infomap algorithm to detect the partition which provides the shortest description length and captures the optimal community structure of the network in terms of its associated flow dynamics. Surprisingly, the partition identified has five modules, whose two-dimensional geographical projections are all simply-connected domains. We proceed with the analysis of traded products by building bipartite structures represented in terms of adjacency matrices between municipalities and products, considering selling and buying. Using the revealed comparative advantage (RCA) concept, we define a non-monetary and binary activity index that can distinguish the RCA of a city in a class of goods or services as evidenced by trade flows. Finally, through the pairwise Maximum Entropy Model, we can associate to the largest communities their corresponding binary Ising-like Hamiltonian models. In an analogy with critical phenomena, our results reveal that each community operates at a "temperature" that is close to the corresponding "critical point", suggesting a high degree of "economic cohesiveness" in its trade network of cities.