Internalization of Externalities in Interdependent Security: Large Network Cases

TL;DR

This paper investigates how to incentivize selfish agents in large, interconnected systems to invest more in security by internalizing externalities, using a population game model on Chung-Lu random graphs, and identifies optimal cost modifications.

Contribution

It establishes a relation between social cost minimizers and Nash equilibria, and shows how to modify private costs to improve overall security in large networks.

Findings

Unique social cost minimizer coincides with Nash equilibrium under certain conditions.

Modifying agents' private costs can enhance security and reduce social costs.

Degree distribution affects security investments and overall network security.

Abstract

With increasing connectivity among comprising agents or (sub-)systems in large, complex systems, there is a growing interest in understanding interdependent security and dealing with inefficiency in security investments. Making use of a population game model and the well-known Chung-Lu random graph model, we study how one could encourage selfish agents to invest more in security by internalizing the externalities produced by their security investments. To this end, we first establish an interesting relation between the local minimizers of social cost and the Nash equilibria of a population game with slightly altered costs. Secondly, under a mild technical assumption, we demonstrate that there exists a unique minimizer of social cost and it coincides with the unique Nash equilibrium of the population game. This finding tells us how to modify the private cost functions of selfish agents in order to enhance the overall security and reduce social cost. In addition, it reveals how the sensitivity of overall security to security investments of agents influences their externalities and, consequently, penalties or taxes that should be imposed for internalization of externalities. Finally, we illustrate how the degree distribution of agents influences their security investments and overall security at both the NEs of population games and social optima.