Cooperative Game Theory: Value, Stability, and Fairness

Build a working mental model of cooperative game theory by learning how coalition value is modeled, what stability means, and how fairness rules split gains. You will see why the core can be empty, how Shapley value allocates credit, and what nucleolus does when stability fails.

Bargaining shows up whenever a group can create surplus together but cannot force a single outcome alone. The cooperative move is to talk in coalitions rather than play a step by step strategy, because the key question becomes which groups can credibly walk away with what. That shifts attention from tactics to enforceable promises and how the total value gets divided. Cooperative game theory is the toolkit for that shift. It asks two questions that stay linked throughout. How much is each coalition worth, and which payoff divisions keep every coalition from wanting to break off?

Why coalition value beats move-by-move strategy

A cooperative model starts by treating any subset of players as a potential deal-making unit. The object you specify is what each coalition can guarantee for itself if it forms and coordinates. Once you know those coalition values, stability and fairness become constraints on the division of the grand coalition’s gains.

Before getting formal, it helps to see how values across singletons, pairs, and the full group can pull in different directions.

A quick way to read a coalition diagram is to look for complementarity and redundancy.

• If two-player coalitions already capture most of the grand coalition value, the third player’s bargaining position is weak.
• If only the grand coalition creates big surplus, everyone is pivotal, and stability is easier.
• If a singleton is valuable alone, that player has a strong outside option that any agreement must beat.

Outside options
In cooperative settings, a player’s leverage often comes from what they can credibly earn without the full group, not from threats inside the group.

Characteristic function games and the TU vs NTU choice

A standard cooperative representation is a characteristic function $v(S)$ that assigns a value to each coalition $S$. The modeling fork is whether that value is a single transferable number or a set of feasible utility vectors.

To see the difference between transferable utility (TU) and non-transferable utility (NTU), compare what counts as an outcome and what can be moved across players.

What TU buys you

With TU, you assume the coalition produces a total surplus that can be redistributed freely using side payments. That makes solution concepts clean. You search for payoff vectors $x$ that sum to $v(N)$, where $N$ is the grand coalition.

When NTU is the honest model

NTU fits when outcomes are inherently multidimensional or constrained, like time, risk exposure, or indivisible goods. A coalition is not summarized by one number. It is summarized by a feasible set $V(S)$ of utility profiles. Stability and fairness then depend on which profiles are feasible, not just on totals.

Model first
If side payments are limited, forcing TU can manufacture fairness results that are impossible to implement.

The core as coalition-proof stability

The core is the most direct stability test in a TU game.

A payoff vector $x$ is in the core if it satisfies two conditions. Efficiency means $\sum_{i\in N}x_i=v(N)$. Coalitional rationality means for every coalition $S$, $\sum_{i\in S}x_i\ge v(S)$. If some coalition can do better on its own, it blocks the proposed allocation.

To get intuition for which inequalities actually matter in a given game, it helps to view the feasible payoff space and watch constraints bite.

Core emptiness is not a technical annoyance. It is the model telling you that the coalition values are mutually incompatible with any fully stable split. Common causes include strong pairwise synergies that cannot all be satisfied at once, or a grand coalition that is not valuable enough to beat the best deviations.

What binding constraints mean

If one coalition constraint is tight, that coalition is just indifferent between staying and leaving. Tight constraints identify the real bargaining threats. Loose constraints are irrelevant in that environment, even if they look important on paper.

Emptiness signal
An empty core means you cannot promise everyone enough to prevent some breakaway, given the coalition values you assumed.

Shapley value and fairness from marginal contributions

If the core is about stability, the Shapley value is about a principled fairness baseline. It assigns each player their average marginal contribution across all possible orders in which the grand coalition could form.

Formally, for player $i$ in an $n$ player TU game, the Shapley value is

$$\phi_i(v)=\sum_{S\subseteq N\setminus\{i\}}\frac{|S|!(n-|S|-1)!}{n!}\Big(v(S\cup\{i\})-v(S)\Big).$$

The weights are the probabilities that coalition $S$ precedes $i$ in a random ordering.

Seeing marginal contributions vary across orderings makes the logic click.

The appeal comes from axioms that match common fairness instincts.

• Symmetry gives equal payoffs to players who contribute the same to every coalition.
• Dummy player assigns zero to a player who never increases any coalition’s value.
• Additivity says if two value sources stack, allocations should stack too.

Where it can feel wrong is when bargaining power comes from outside options or coalition threats rather than contribution averaging. A player may have low average marginal contribution yet be essential to prevent a blocking coalition, so stability and Shapley fairness can point different ways.

When the core is empty, minimize complaints

When no allocation is unblocked, you can still look for a compromise that is as stable as possible. The nucleolus does this by ranking coalitions by their dissatisfaction, then choosing the allocation that lexicographically minimizes the worst complaints. In a TU game, a coalition’s complaint at $x$ is often captured by its excess $e(S,x)=v(S)-\sum_{i\in S}x_i$.

The bargaining set relaxes core stability more directly. It allows objections by a coalition only if they cannot be countered by another coalition, so some deviations are treated as non-credible.

A worked example makes the idea concrete, especially in a game where the core vanishes.

A useful way to interpret nucleolus outcomes is that they spend value to buy down the most dangerous grievances first. Even when you do not intend to compute it, this mindset helps you diagnose negotiations. Identify which coalition feels most shortchanged, then see what compensation would neutralize that complaint without triggering a bigger one elsewhere.

Complaint ordering
Nucleolus is less about splitting the pie evenly and more about eliminating the sharpest instability pressures in the coalition system.

Applications and common failure modes

Many applied problems reduce to specifying $v(S)$ and choosing a solution concept that matches your governance and information.

• Cost sharing in networks or projects, where $v(S)$ reflects savings from pooling resources.
• Joint ventures, where complementary assets create superadditive gains.
• Voting and influence, where coalition value may be winning probability or policy utility.

Failure modes usually come from violating cooperative assumptions. Externalities across coalitions break the idea that $v(S)$ depends only on members of $S$. Communication limits make certain coalitions infeasible. Weak enforceability makes stability results aspirational.

Try mapping one real scenario into the model and stress-testing it against multiple solution concepts.

If you find yourself adjusting $v(S)$ to make a favored solution look good, pause. That is often a sign that the real environment is NTU, has externalities, or needs an explicit coalition formation layer rather than a fixed grand coalition.

Choosing the right solution concept

Start with what the allocation must accomplish operationally, then pick the weakest concept that still fits.

If you need no subgroup to have a profitable deviation and enforcement is strong, the core is the natural target. If the core is empty, decide whether you want to relax stability or reinterpret it. Nucleolus keeps the stability frame but accepts that some coalitions will be unhappy and asks you to minimize the most explosive unhappiness. Bargaining set is closer to negotiation practice, because it filters deviations by whether they can be countered.

If the goal is credit assignment, benchmarking, or a fairness norm in repeated collaboration, Shapley value is often the right anchor, especially when players accept contribution-based narratives. When bargaining power is driven by outside options, you may need a stability-oriented concept, or you may need to model those options directly by adjusting coalition values or moving to an explicit bargaining model.

A practical selection checklist is to ask:

• Is the environment credibly TU, meaning side payments can really move value?
• Are coalitions enforceable, and is deviation a realistic threat?
• Do stakeholders care more about stability, perceived fairness, or minimizing conflict?
• Do externalities or coalition formation dynamics undermine $v(S)$ as a standalone object?

Pick one concept as your default, then compute or reason about one alternative as a robustness check. If they disagree sharply, the disagreement is the insight. It tells you which assumption is doing the real work and which political constraint you cannot ignore.