The Shapley Axiomatization for Values in Partition Function Games

TL;DR

This paper proves that Shapley's original axioms uniquely determine a value for partition-function games with externalities and introduces a Monte Carlo algorithm to approximate these values.

Contribution

It establishes the uniqueness of Shapley's axiomatization for externality games and provides a practical approximation algorithm.

Findings

Shapley's axioms imply a unique value for externality games.

Existing proposed values can be derived from Shapley's axioms.

A Monte Carlo method is developed for approximating these values.

Abstract

One of the long-debated issues in coalitional game theory is how to extend the Shapley value to games with externalities (partition-function games). When externalities are present, not only can a player's marginal contribution - a central notion to the Shapley value - be defined in a variety of ways, but it is also not obvious which axiomatization should be used. Consequently, a number of authors extended the Shapley value using complex and often unintuitive axiomatizations. Furthermore, no algorithm to approximate any extension of the Shapley value to partition-function games has been proposed to date. Given this background, we prove in this paper that, for any well-defined measure of marginal contribution, Shapley's original four axioms imply a unique value for games with externalities. As an consequence of this general theorem, we show that values proposed by Macho-Stadler et al., McQuillin and Bolger can be derived from Shapley's axioms. Building upon our analysis of marginal contribution, we develop a general algorithm to approximate extensions of the Shapley value to games with externalities using a Monte Carlo simulation technique.