The Gestalt in Graphs: Prediction Using Economic Networks

Abstract

We define an economic network as a linked set of entities, where links are created by actual realizations of shared economic outcomes between entities. Such aggregations, which include a wide variety of Web-based product networks, are becoming increasingly available for business use. We analyze the predictive information contained in such networks using the copurchase network on Amazon where entities are books and links designate which pairs were purchased simultaneously. Our dataset covers a diverse set of books spanning over 400 categories over a period of three years with a total of over 70 million observations. We demonstrate that an entity’s future demand is more accurately predicted by combining its historical demand with that of its neighbors than by considering its demand alone. In other words, if you want to know what your state will be in the future, consider what is happening to your neighbors now. This result could apply to other economic networks where outcomes of sets of entities tend to be related and suggest substantial potential for real-time decision support systems that leverage such networks. To our knowledge, this is the first large scale study showing that an economic network contains useful distributed information for demand prediction.

Keywords: network effects, economic networks, copurchase networks, predictive models, data mining