Game Theory

Game theory is the formal mathematical framework in business and economics theory, applied mathematics, computer science, and evolutionary biology that models strategic interaction among multiple decision-makers (players), where each agent's payoff or utility depends not only on their own chosen action but also on the actions chosen by other agents. Whereas classical price-taking consumer and producer models in supply and demand treat individual agents as optimizing against exogenous parametric prices, game theory resolves circular strategic expectations ("what I should do depends on what I expect you to do, which depends on what you expect me to do"). Following early oligopoly precursors by Antoine Augustin Cournot (1838) and Joseph Bertrand (1883) and Ernst Zermelo's backward-induction theorem for chess (1913), the modern discipline was established by John von Neumann's minimax theorem (1928), von Neumann and Oskar Morgenstern's Theory of Games and Economic Behavior (1944), and John F. Nash Jr.'s non-cooperative equilibrium theory (1950).

Strategic (Normal) Form Games

A finite N-player normal-form (or strategic-form) game is defined formally as an ordered triple G = ⟨N, (S_i)_i∈N, (u_i)_i∈N⟩:

1. Player Set (N): A finite set of strategic decision-makers N = {1, 2, …, n}.
2. Pure Strategy Spaces (S_i): For each player i ∈ N, a finite set S_i = {s_i1, s_i2, …, s_im} of feasible complete action plans. The Cartesian product S = S₁ × S₂ × … × S_n is the space of strategy profiles s = (s_i, s_−i), where s_−i denotes the vector of strategies selected by all players other than i.
3. Payoff (Utility) Functions (u_i): A von Neumann–Morgenstern cardinal utility mapping u_i: S → ℝ assigning a real numerical payoff u_i(s_i, s_−i) to each player i for every realized strategy profile s ∈ S.

When players randomize over their pure strategies, a mixed strategy σ_i ∈ Δ(S_i) is a probability distribution assigning probability σ_i(s_i) ≥ 0 to each pure strategy s_i ∈ S_i with ∑_si∈Si σ_i(s_i) = 1. Under independent randomization σ = (σ₁, …, σ_n), player i's expected payoff is the multilinear expectation u_i(σ) = ∑_s∈S [∏_j=1ⁿ σ_j(s_j)] · u_i(s).

Extensive Form Games and Information Sets

Whereas normal-form matrices represent simultaneous-move interactions, extensive-form games (formalized by Harold Kuhn in 1953) represent sequential interactions as a directed rooted tree ⟨H, Z, P, (I_i)_i∈N, (u_i)_i∈N⟩, specifying the exact chronological order of moves, chance nature nodes, terminal histories Z, and information sets I_i that partition player i's decision nodes according to what player i has observed. A game has perfect information if every information set is a singleton node (as in chess), and imperfect information if at least one information set contains multiple nodes (meaning a player cannot observe a prior move).

Zero-Sum vs. Non-Zero-Sum and the Minimax Theorem

In a two-player zero-sum (strictly competitive) game, one player's gain is identically the other player's loss for every strategy profile: u₁(s₁, s₂) + u₂(s₁, s₂) = 0. John von Neumann's (1928) Minimax Theorem proved using convex duality that for every finite two-player zero-sum game, the maximum expected payoff Player 1 can guarantee by choosing σ₁ equals the minimum expected loss Player 2 can enforce by choosing σ₂, defining the unique value of the game v^*:

In variable-sum (non-zero-sum) games—which characterize virtually all economic trade, oligopoly pricing, and collective action problems—mutual gains and mutual losses are possible, so minimax strategies no longer coincide with rational best responses. John Nash (1950) generalized von Neumann's solution to arbitrary N-player games via the Nash equilibrium, a strategy profile σ^* in which every player's strategy is a best response to the equilibrium strategies of all other players: u_i(σ_i^*, σ_−i^*) ≥ u_i(s_i, σ_−i^*) for all s_i ∈ S_i.

Subgame Perfection, Backward Induction, and Repeated Games

In sequential extensive-form games, an ordinary Nash equilibrium can rely on non-credible threats off the equilibrium path (such as an incumbent monopoly threatening a destructive price war that would be unprofitable to execute once a rival actually enters). Reinhard Selten (1965) eliminated non-credible threats by defining Subgame Perfect Equilibrium (SPE): a strategy profile that induces a Nash equilibrium in every proper subgame of the game tree, solvable in finite games of perfect information via dynamic programming backward induction.

When a stage game (such as the Prisoner's Dilemma) is repeated infinitely with discount factor δ ∈ (0, 1) (or repeated with continuation probability δ in a Markov chain), the Folk Theorems (Aumann & Shapley; Fudenberg & Maskin, 1986) prove that any feasible, individually rational payoff vector—including full Pareto efficiency cooperation—can be sustained as a subgame perfect equilibrium for sufficiently patient players (δ → 1) using history-dependent trigger or tit-for-tat punishment strategies.

Harsanyi Transformation and Signaling Games

For decades, economists could not apply game theory when players lacked complete knowledge of rivals' cost functions or valuations. John Harsanyi (1967–1968) solved this problem via the Harsanyi transformation, converting a game of incomplete information into a game of imperfect information in which "Nature" first draws each player's private type vector θ = (θ₁, …, θ_n) from a common-prior probability distribution P(θ). Players then choose type-contingent strategies s_i(θ_i) to maximize expected utility conditional on their own type, defining a Bayesian Nash Equilibrium (BNE). In sequential settings with adverse selection and moral hazard, informed agents can transmit credible information through costly actions (Michael Spence's 1973 education signaling model), whereas reverse game theory—known as mechanism design (Leonid Hurwicz, Eric Maskin, and Roger Myerson)—designs game rules and auction theory formats to achieve desired social outcomes under incentive-compatibility constraints.

Cooperative Games (The Shapley Value) and Evolutionary Stability

In cooperative game theory, where coalitions C ⊆ N can sign binding agreements to generate characteristic value v(C), Lloyd Shapley (1953) derived the unique Shapley value φ_i(v) that distributes total surplus fairly according to each player's average marginal contribution across all n! possible coalition arrival permutations:

In biology and behavioral economics (where agents exhibit bounded rationality rather than infinite deductive foresight), John Maynard Smith and George R. Price (1973) formulated Evolutionary Game Theory, defining an Evolutionarily Stable Strategy (ESS) and showing how natural selection under replicator dynamics (ẋ_k = x_k · [u(e_k, x) − u(x, x)]) converges toward symmetric Nash equilibria.

What is the difference between a pure strategy and a mixed strategy?

A pure strategy is a deterministic plan of action that specifies with 100% certainty which action a player will take at every decision node they might reach. A mixed strategy is a probability distribution over two or more pure strategies, where the player deliberately randomizes their choice (such as a soccer penalty kicker or an tax auditing agency) to prevent an opponent from exploiting a predictable action.

Why does backward induction in a finitely repeated Prisoner's Dilemma predict defection in every round?

If a Prisoner's Dilemma is repeated for a known, fixed finite number of rounds T (for example, T = 100), both players know that in the final round T = 100 there is no future round T + 1 in which defection can be punished. Therefore, mutual defection is the unique dominant outcome in round 100. Given that round 100's outcome is fixed regardless of round 99's play, round 99 effectively becomes the last strategic round, inducing defection in round 99, and this backward induction unravels cooperation all the way back to round 1. Cooperation is sustained in infinite or indefinite horizons where no known terminal round exists.

How does cooperative game theory differ from non-cooperative game theory?

In non-cooperative game theory, the fundamental unit of analysis is the individual player, and agreements (such as cartel output quotas) are only sustainable if they are self-enforcing equilibria for each participant. In cooperative game theory, the fundamental unit of analysis is the coalition of players, assuming an external legal or institutional framework can enforce binding contracts and surplus-sharing agreements among coalition members.

1. Von Neumann, J., & Morgenstern, O. (1944). Theory of Games and Economic Behavior. Princeton University Press. https://doi.org/10.1515/9781400829460
2. Nash, J. F. (1950). "Equilibrium Points in n-Person Games." Proceedings of the National Academy of Sciences, 36(1), 48–49. https://doi.org/10.1073/pnas.36.1.48
3. Selten, R. (1965). "Spieltheoretische Behandlung eines Oligopolmodells mit Nachfrageträgheit." Zeitschrift für die gesamte Staatswissenschaft, 121(2), 301–324. https://www.jstor.org/stable/40748884
4. Harsanyi, J. C. (1967). "Games with Incomplete Information Played by 'Bayesian' Players, I–III." Management Science, 14(3), 159–182. https://doi.org/10.1287/mnsc.14.3.159
5. Shapley, L. S. (1953). "A Value for n-Person Games." In Contributions to the Theory of Games (Vol. II, pp. 307–317). Princeton University Press. https://doi.org/10.1515/9781400881970-018