[2H5-OS-2b] OS-2

4:45 PM - 5:00 PM JST(7:45 AM - 8:00 AM UTC)

[2H5-OS-2b-06] Shapley Values in Networks with Higher-Order Interactions

〇Tomoya Akamatsu¹, Kei Nakagawa² (1. MUFG Bank, Ltd., 2. Osaka Metropolitan University)

Keywords:

Shapley value,Edge-based Shapley value,Hypergraph,Game theory on networks

Network analysis is very important in the field of finance and economics for understanding complex interdependence, such as B2B trading ties and interlocking directorships. Analyzing higher order interactions, where more than two nodes belong to the same hyperedge at the same time, is harder than analyzing ordinary graphs that only describe pairwise relationships. Consequently, there is a growing need for established methods to evaluate the actual influence of each node. To address this challenge, we explore the edge-based Shapley value (ESh), an extension of the Shapley value from cooperative game theory to network settings. ESh evaluates the contribution of participating players (nodes) based on changes in the network structure, i.e., the organization (hyperedge) itself and its associated attribute values. This property enables a principled allocation of organizational value among nodes, thereby facilitating the identification of structurally important nodes in hypergraphs. In this study, we derive a representation of the ESh allocation within the Harsanyi dividend framework of cooperative game theory. Furthermore, we empirically compare ESh with the conventional Shapley value using a network built from director data of firms listed on the TOPIX 100.