TL;DR
This paper introduces a new concept called "player nullification" in cooperative game theory, which is a softer version of player removal. By applying this nullification to existing axioms, the authors derive new characterizations for important solution concepts like the Shapley value and introduce novel values with interesting properties.
What Problem Does This Paper Address?
In cooperative game theory, a key aspect of analyzing games is understanding how the value of a game changes when players are removed. Axioms like "balanced contributions" and "balanced cycle contributions" formalize this by examining the impact of player removal on the payoffs of remaining players. However, removing a player entirely can be a drastic change. This paper explores a more nuanced approach: "player nullification." Instead of removing a player, nullification transforms the game such that the nullified player contributes nothing to any coalition they are part of. This paper investigates how existing axiomatic characterizations of game values change when player removal is replaced by this player nullification concept, and whether this leads to new insights or different solution concepts.
Methods
The core methodological contribution is the formalization and application of "player nullification." This involves defining a new game transformation where a player's contribution is effectively zeroed out for all coalitions. The authors then systematically revisit established axioms from cooperative game theory, such as Balanced Contributions (BC), Balanced Cycle Contributions (BCyC), and Balanced Collective Contributions (BCoC), and adapt them to this nullification framework, creating new axioms like Balanced Contributions under Nullification (BCN), Balanced Cycle Contributions under Nullification (BCyCN), and Balanced Collective Contributions under Nullification (BCoCN). They then use these new axioms, in conjunction with standard axioms like Efficiency (E) and Null Player (N), to derive new characterizations of known solution concepts (like the Shapley value) and to define and characterize novel values. The paper also revisits the "potential approach" to game theory, adapting it to the nullification concept.
Datasets
Not discussed in the paper. The paper deals with theoretical cooperative games, not empirical datasets.
Key Insights and Ideas
- Player Nullification as a Softer Alternative: Nullifying a player is a less drastic change than removing them, allowing for a more granular analysis of player influence and game dynamics.
- New Axioms, New Values: Adapting existing axioms to player nullification leads to new axiomatic systems that characterize both known and novel solution concepts.
- Shapley Value Robustness: The Shapley value often emerges as a key result, demonstrating its robustness across different axiomatic frameworks, including those involving nullification.
- Exogenous Solidarity: The characterization of an "affine Shapley value" (Shapley value plus a budget-balanced transfer scheme) highlights how player nullification can formalize exogenous forms of solidarity, independent of the game's specific characteristic function.
- Potential Approach Adaptation: The nullification concept can be applied to the potential function in game theory, leading to a "nullified potential" and a recursive formula for the Shapley value.
Key Results
- Balanced Contributions under Nullification (BCN): When combined with Efficiency (E), BCN characterizes a class of values that are the sum of the Shapley value and a budget-balanced transfer scheme. This class can be interpreted as incorporating exogenous solidarity.
- Balanced Collective Contributions under Nullification (BCoCN): This axiom, when combined with Efficiency (E) and Equal Treatment (ET), uniquely characterizes a new value with a closed-form expression. This value exhibits marginalistic properties, potentially favoring productive players.
- Balanced Cycle Contributions under Nullification (BCyCN): When combined with Efficiency (E) and the Null Player axiom (N), BCyCN characterizes the Shapley value on games with at least one null player. Further extensions with Linearity (L) or specific conditions on null players lead to other characterizations.
- Nullified Potential: A unique nullified potential function is defined, which is shown to be equal to the standard potential function. This leads to a recursive formula for the Shapley value.
Limitations and Future Work
- Fixed Player Sets: Many of the results are derived for TU-games with a fixed player set, which differs from some literature that considers variable player sets.
- Impossibility Results: The paper highlights situations where combining certain axioms (e.g., Efficiency, Null Player, and BCoCN) leads to impossibility results, suggesting careful selection of axioms is needed.
- Extension to Graph Games: The authors suggest that a natural future direction is to extend the nullification concept to TU-games augmented by a graph, where axioms might be based on link deletion rather than player removal. This could involve exploring axioms like fairness and component fairness in this new context.
Summary updated