Shapley Value
- Introduced by
- Lloyd Shapley
- Year
- 1953
- Type
- algorithm
- Family
- Game-theoretic
What does it do?
The Shapley Value is a solution concept for coalition games that distributes total payoff fairly among players based on their marginal contributions to coalitions.
Introduced by Lloyd Shapley in 1953, the Shapley Value is the unique payoff distribution that satisfies four intuitive axioms: efficiency (total payoff is distributed), symmetry (identical players receive equal payoff), null player (players contributing nothing receive nothing), and additivity across games.
The intuition behind the method
Imagine a group of workers forming a joint venture. The Shapley Value asks: how much should each worker earn? The answer is based on the average marginal contribution each worker makes when joining teams of different sizes. For each possible order in which workers join (from zero to all), compute how much each adds; then average these contributions across all possible orders. This gives each worker a payoff reflecting their indispensable role without bias toward timing or group size. It is the unique 'fair' division satisfying intuitive axioms.
History
Lloyd Shapley introduced the Shapley Value in 1953 as the unique solution to the fair division problem in cooperative games. His axiiomatic approach proved transformative, establishing cooperative game theory as a rigorous field. The Shapley Value has been applied to diverse domains from cost allocation to credit assignment and environmental regulation. Shapley's contributions to game theory earned him the Nobel Prize in Economics in 2012, recognizing the value's foundational importance.
When to use it
Use the Shapley Value when fairly distributing joint payoffs from coalition formation, such as profit sharing in partnerships, credit allocation in research teams, or valuing individual contributions to ensemble models. Use when players have heterogeneous contributions and no market prices reveal individual value. Appropriate when seeking an axiomatically justified fair division that satisfies basic equitable principles.
Limitations and considerations
- Computationally expensive: requires enumerating all 2^n coalitions for n players; infeasible for large n
- Requires complete information: coalition values must be known in advance for all subsets
- May not satisfy individual rationality: some players could earn more outside the grand coalition
- Non-unique in multiple coalition games: when many coalitions are feasible, determining which forms is ambiguous
Source conditions do not replace independent scientific review or validation by an analysis engine.
Strengths
- Axiomatically unique: only division satisfying efficiency, symmetry, null player, and additivity axioms
- Fair and theoretically grounded: reflects each player's average marginal contribution across coalitions
- Flexible: applies to any cooperative game where coalition values are well-defined
- Decomposable: can be used with any valuation function, including probabilistic and weighted versions
Applications
- Partnership accounting: fairly distributing profits among business partners with varying contributions
- Research team credit: allocating acknowledgments and credit among co-authors based on contribution magnitude
- Cost allocation: determining how to split infrastructure costs among users with different usage patterns
- Watershed resource management: distributing water from shared rivers among downstream users equitably
- Machine learning: attributing model predictions to feature contributions using SHAP (SHapley Additive exPlanations)
How it works
How it works
Identify all possible subsets (coalitions) of players that can form.
For each coalition S, calculate v(S), the total payoff that coalition can achieve without outside players.
For each player i and coalition S not containing i, compute the marginal contribution: v(S ∪ {i}) - v(S).
Average player i's marginal contribution across all orderings of players, weighting by the number of coalitions of each size.
Common pitfalls
- Assuming Shapley Value leads to stable coalitions; the value defines fair payoff division, not necessarily stable coalition structure
- Computing Shapley Value by simple averaging across all orderings; must weight by coalition size probability
- Confusing Shapley Value with Nash Bargaining or other solution concepts; each has different interpretations and properties
- Forgetting that Shapley Value assumes complete characteristic function; if coalition values are uncertain, extensions are needed
Frequently asked questions
Why is the Shapley Value unique?
Because it is the only payoff distribution satisfying four intuitive axioms: (1) Efficiency (all payoff distributed), (2) Symmetry (identical players get equal payoff), (3) Null player (zero-contribution players get nothing), and (4) Additivity (payoffs combine across games). These axioms uniquely determine the value.
How does Shapley Value handle games where the grand coalition does not form?
The Shapley Value assumes the grand coalition forms and distributes its payoff v(N). If coalitions other than the grand coalition might form, the Shapley Value does not directly address coalition structure stability; other game-theoretic concepts (core, bargaining) are needed.
Can the Shapley Value be negative?
Yes. If a player's average marginal contribution across coalitions is negative (they reduce total payoff), their Shapley Value is negative. This can occur in games with externalities or negative synergies.
Related methods
Sources
Shapley, L. S. (1953). A value for n-person games. In H. W. Kuhn & A. W. Tucker (Eds.), Contributions to the Theory of Games II (pp. 307-317). Princeton University Press.
Open DOI referenceCopied.Roth, A. E. (1988). The Shapley value as a von Neumann-Morgenstern utility. Econometrica, 56(4), 745-794.
Open publication referenceCopied.