SCIENCE
Game Theory
Abstract. Game theory studies strategic interdependence: situations in which the consequences of an actor's decisions depend upon other actors' decisions, information, beliefs, commitments, and institutional rules. This monograph develops the discipline from its axiomatic and historical foundations through noncooperative equilibrium, dynamic credibility, incomplete information, coalition formation, bargaining, auctions, mechanism design, evolutionary adaptation, behavioral experiments, computation, and contemporary information design. It specifies the formal objects and assumptions behind influential solution concepts, works through reproducible strategic examples, compares the explanatory reach of competing approaches, and evaluates empirical interpretation. The central analytical result is a comparative understanding of game-theoretic explanation as an ordered process of specifying feasible actions, beliefs and information, institutional enforcement, equilibrium restrictions, behavioral evidence, and welfare criteria. The same observed pattern may be sustained by different mechanisms, so identification, robustness, and institutional specificity remain indispensable. This is an original scholarly synthesis of established research, with no claim to a new named theory, theorem, model, or doctrine.
Keywords: Game theory; strategic interaction; noncooperative games; Nash equilibrium; mixed strategies; extensive-form games; subgame perfection; sequential equilibrium; Bayesian games; correlated equilibrium; bargaining; cooperative games; Shapley value; repeated games; signaling; auctions; mechanism design; information design; evolutionary stability; behavioral game theory; computational complexity; market design; algorithmic pricing
1. Scientific Object, Domain, and Method
Game theory is the systematic study of interactions in which each participant evaluates available actions in light of possible actions of others. The defining condition is payoff-relevant interdependence. A monopoly producer facing a fixed technological constraint solves an optimization problem; the same producer facing an entrant whose decisions respond to its price and investment chooses within a game. A public authority deciding how to allocate a spectrum license confronts a game when bids reveal selected information and participants adapt to the allocation and payment rules. The relevant players may be people, firms, coalitions, states, institutions, software agents, or evolutionary types. Modeling a participant as a player establishes a role in a specified interaction; it does not presume psychological similarity between all participants.
Four ingredients are fundamental. There are feasible actors and actions; an outcome function translates action profiles and possibly chance events into consequences; preferences rank consequences; and information structures determine which contingencies and actions an actor can distinguish. The timing of moves, observation of earlier moves, enforceable commitments, available messages, and possibility of future interactions are additional parts of the economic institution being modeled. A game is therefore a representation of incentives within rules, and the same underlying resources can generate radically different strategic outcomes when the rules change.
The field's classifications are analytical tools. Noncooperative game theory models strategies chosen by separate decision makers; cooperative game theory analyzes what groups can secure and how a surplus may be allocated. Complete information means payoff-relevant characteristics are common knowledge within the modeled game, whereas perfect information means every decision maker observes the full history at each decision node. A game may have complete information and imperfect information because actions occur simultaneously. Uncertainty about another player's valuation or cost gives incomplete information even when earlier moves are observable. Ordinary language frequently conflates these distinctions, and doing so can invalidate an equilibrium argument.
Game theory uses several forms of scientific explanation: a logical characterization of behavior under assumptions; a comparative-static prediction about changed institutions or incentives; a mechanism connecting private information to outcomes; a welfare evaluation relative to a specified social criterion; and an empirically tested hypothesis about how real subjects act. Each demands distinct evidence. A theorem establishes an implication within a model. Empirical evidence tests whether the model's restrictions describe a population and setting. Institutional implementation requires engineering and ongoing monitoring. Roth's economist-as-engineer analysis explains why market design must join strategic theory with computation and experiments.
The disciplinary center of Game Theory is strategic interaction. Microeconomics supplies a broader theory of choice and exchange; Mathematical Economics covers formal economic analysis across strategic and nonstrategic environments; Information Economics investigates knowledge, incentives, and market consequences; Experimental Economics provides testing methods. Game Theory contributes a common language across these domains through explicit interdependence and strategic anticipation. This monograph treats their overlaps as fruitful research interfaces while keeping each Science's principal object conceptually identifiable.
2. Historical Formation of Game Theory
Formal game theory emerged from attempts to handle conflicts in which optimal choice depends recursively on other optimizing participants. Early mathematical treatment of mixed strategies and two-player adversarial problems culminated in John von Neumann's 1928 minimax paper. The joint 1944 work of von Neumann and Morgenstern made strategic choice, characteristic-function games, and expected-utility reasoning central to a research program in economics. The latter source is a Princeton University Press historical account of the book; the original work is separately identified in References.
John Nash shifted attention from two-player zero-sum conflict to general noncooperative interaction. His 1950 existence argument formalized a profile of mutual best replies, while Non-Cooperative Games (1951) developed the framework further. The essential advance was a logically consistent behavioral restriction applying to arbitrary finite games rather than a requirement that all parties have opposed interests. Nash equilibrium is a mathematical condition on strategies. Its existence does not by itself determine which equilibrium will be selected or how agents learn to play it.
The next generations studied what equilibrium predictions omit. Shapley's value for coalition games addressed cooperative allocation. Schelling's Strategy of Conflict developed strategic commitment, threats, bargaining, coordination, and focal behavior. Selten's dynamic oligopoly analysis contributed to the study of consistency after each history. Harsanyi's Bayesian games framework addressed asymmetric information by representing hidden payoff parameters as types drawn by nature. Aumann's correlated-strategy analysis made recommendations and shared signals legitimate objects of strategic explanation.
The discipline then differentiated into interlocking schools of research. Repeated-game analysis investigated how future consequences discipline current conduct; bargaining theory formalized proposals and delay; mechanism design treated the institutional rules themselves as design variables; evolutionary theory studied population adjustment and invasion resistance; experiments challenged assumptions of perfect rationality and purely selfish preferences; and computational game theory examined which equilibrium predictions can actually be found or implemented. The evolution of the field thus reflects deepening attention to timing, information, institutions, and behavioral feasibility.
This intellectual history also shows changing standards of application. Earlier equilibrium formulations asked whether consistent strategies exist; later research asks which outcomes survive refinements, what observables discriminate among mechanisms, which institutions can implement them, and whether computation or learning can find the relevant strategies. Historically foundational contributions remain relevant because these later questions often reuse the same formal objects while adding explicitly testable restrictions.
3. Formal Architecture of Strategic Games
A finite normal-form game is represented by G = (N, (A_i){i in N}, (u_i){i in N}). N is a finite player set, A_i is the set of pure actions available to player i, and u_i: A_1 × ... × A_n → R is the utility or payoff function. A profile a = (a_1, ..., a_n) combines one action per player. In a pure-strategy Nash equilibrium a*, every player satisfies u_i(a*i, a*-i) ≥ u_i(a_i, a*_-i) for every feasible a_i. The right-hand notation fixes others' equilibrium actions while varying only player i's move. This is the mathematical meaning of unilateral-deviation resistance.
A mixed strategy σ_i assigns probabilities to pure actions, with nonnegative entries summing to one. For independent mixed strategies, expected utility is the probability-weighted average of payoffs over pure profiles. A pure strategy is weakly dominated if another strategy gives at least as high a payoff against every opponent profile and a strictly higher payoff somewhere; strict domination requires a strict improvement everywhere. Rationalizability and iterative deletion of dominated strategies constrain choices without always selecting a unique equilibrium. A weakly dominated move can remain compatible with Nash equilibrium in some games; consequently one must name the dominance criterion before removing strategies.
Consider the following two-player prisoner's dilemma, expressed as ordered payoffs to Row and Column:
| Row / Column | Cooperate | Defect |
|---|---|---|
| Cooperate | (3, 3) | (0, 5) |
| Defect | (5, 0) | (1, 1) |
Defect strictly dominates Cooperate for both actors: against a cooperating counterpart, 5 exceeds 3; against a defecting counterpart, 1 exceeds 0. The unique Nash equilibrium is therefore (Defect, Defect), yielding (1, 1), although (Cooperate, Cooperate) gives both players (3, 3). This is an analytical demonstration that equilibrium and Pareto efficiency are distinct properties. The experiment assumes these exact utility numbers; they are illustrative, not observed social data. If the actors have inequality aversion, repeated encounters, outside enforcement, or reputational consequences, the strategic object itself changes.
A coordination game demonstrates another limitation:
| Row / Column | Left | Right |
|---|---|---|
| Left | (4, 4) | (0, 0) |
| Right | (0, 0) | (2, 2) |
There are two pure equilibria, (Left, Left) and (Right, Right), along with a mixed equilibrium. Coordinating on Left Pareto-dominates coordinating on Right, but unilateral best-response consistency alone selects neither. Beliefs, focal cues, conventions, history, risk exposure, and communication can govern selection. This is why the economic interpretation of a formal game requires identification of the institution and the process generating expectations.
Game representations are deliberately economical. A normal form abstracts from the path of play; an extensive form specifies the sequence of decision nodes, each player's information sets, chance moves, and outcomes. Moving from a decision tree to a normal form requires strategies that specify a complete contingent plan, including actions at decision nodes never reached on the predicted path. The two forms can support different discussions because the extensive form makes credible continuation behavior and observation structure visible.
4. Rationality, Knowledge, and Equilibrium
A best response maximizes a player's expected utility given a conjecture about opponents. Nash equilibrium imposes mutual consistency between the strategy profile and the beliefs used to evaluate unilateral deviations. In a simultaneous game, it does not assert that agents explicitly know the equilibrium or consciously calculate it. Its epistemic foundations require careful specification of rationality, beliefs, and what players know about others' rationality. Common knowledge means that everyone knows a proposition, everyone knows that everyone knows it, and so on through all finite orders. The distinction between shared information and common knowledge becomes central when expectations are strategic.
Aumann's agreement theorem illustrates the force and narrowness of these assumptions: with a common prior and common knowledge of two agents' posterior probabilities for the same event, the posteriors must agree. The theorem does not say that ordinary observers always agree; its assumptions may fail when priors differ, posterior beliefs are not common knowledge, or the event and information partitions are represented differently. The methodological lesson is to read information theorems as conditional propositions with explicit epistemic primitives.
Consider a two-action game with a player uncertain about whether the other will choose Left. If choosing Left yields 4 upon coordination and 0 otherwise, while choosing Right yields 2 upon coordination and 0 otherwise, a player with belief p that the opponent chooses Left earns 4p from Left and 2(1−p) from Right. Left is optimal if p ≥ 1/3, Right if p ≤ 1/3; indifference holds at p = 1/3. Belief thresholds are a compact bridge between qualitative coordination stories and falsifiable quantitative predictions. Yet an observed Left choice alone cannot identify the belief p without knowing preferences and unobserved payoffs.
Nash's existence theorem ensures at least one mixed-strategy equilibrium in each finite game, using fixed-point reasoning. It leaves open uniqueness, stability, efficiency, computational tractability, and the social process by which beliefs are coordinated. A theorem proving existence of a price or strategy profile is valuable even when no predictive process has been provided; the scientific task is to identify the additional assumptions needed for a stronger claim.
Strategic refinements respond to this need by eliminating equilibrium profiles inconsistent with specified forms of robustness. Their credibility depends on the informational and decision-theoretic conditions that justify each refinement. Using a stronger label does not automatically make its empirical predictions superior. A refinement can improve a logically unrealistic threat while leaving substantial multiplicity or producing predictions that are sensitive to minute model perturbations.
5. Zero-Sum Games, Minimax, and Strategic Randomization
In a two-player zero-sum game u_1(a_1, a_2) = −u_2(a_1, a_2), so any payoff gain to one participant is a loss to the other. The minimax theorem equates the maximum guaranteed expected payoff of one player with the minimum expected payoff the other can force under mixed strategies in finite zero-sum games. Von Neumann's mathematical treatment of strategic games provides the historical foundation. The result establishes a value and optimal randomized strategies under defined conditions; it does not make conflict a universal model of economic exchange.
In matching pennies, Row wins one unit when the two symbols match and loses one when they differ; Column's payoffs are the negatives. Each pure action by either player is exploitable. If Row plays Heads with probability p and Column plays Heads with probability q, the expected Row payoff is (2p−1)(2q−1). Row's maximin strategy is p = 1/2, Column's minimax strategy is q = 1/2, and the game value is zero. This is an exactly solvable example of mixing: randomization makes an opponent indifferent and removes a profitable predictable deviation. The derivation uses independent private randomization, not a public coin with correlated recommendations.
Under a general zero-sum payoff matrix M, a Row strategy x maximizes a guaranteed lower bound v subject to x_j ≥ 0, ∑_j x_j = 1, and ∑j x_j M{jk} ≥ v for every Column action k. The dual formulation lets Column minimize an upper bound w across all Row actions. Linear-programming duality connects these optimization problems and explains why their optimal values agree. Changing the matrix by adding a constant to all Row payoffs shifts the value but leaves best-response patterns unchanged; positive affine transformation of an individual player's von Neumann–Morgenstern utility also preserves its strategic preferences.
Many real conflicts combine zero-sum and common-interest dimensions. Trade negotiations, competition over market share, political bargaining, and disputes over public resources may contain a distributive conflict over shares and a shared interest in avoiding waste. Treating a nonzero-sum situation as strictly zero-sum can erase investments, surplus creation, joint risk, and mutually beneficial institutional reform. Thus minimax reasoning provides a robust local technique, while full economic explanation often requires the general-sum structure pioneered by Nash and subsequent authors.
6. Dynamic Games, Credibility, and Sequential Reasoning
An extensive-form game specifies who moves when, what each player observes, and how complete contingent strategies determine terminal outcomes. A subgame begins at a single decision node whose continuation includes all succeeding nodes without cutting any information set. A subgame-perfect equilibrium requires Nash equilibrium in every subgame, including those not reached on the equilibrium path. This guards against threats that would not be chosen if their decision node were actually reached. Selten's dynamic oligopoly contribution is part of the historical development of such sequential consistency.
Suppose an entrant first chooses Enter or Stay Out. Entry yields a subsequent incumbent decision to Accommodate or Fight. If entry occurs and accommodation gives (2, 2), fighting gives (−1, 0), and staying out gives (0, 3), where the first payoff is the entrant's, then backward induction predicts Accommodate after entry because the incumbent prefers 2 over 0. Anticipating this, the entrant enters because 2 exceeds 0. An incumbent's announcement that it will fight can sustain a Nash equilibrium representation with an implausible off-path threat, but it cannot survive subgame perfection with these payoffs. Credibility must be found in actual continuation incentives, enforceable investment, or changed commitment possibilities.
The prediction changes if the incumbent can make a prior investment altering the payoff from fighting, or if the entrant observes a costly action revealing incumbent costs. Commitment is therefore a transformation of the game tree or payoffs. Schelling's strategic commitment analysis emphasizes how limiting one's future choices can change an opponent's current decision. A legally enforceable contract, irreversible capacity, or delegated automated rule can perform this function if its enforcement assumptions are credible.
When actions are imperfectly observed or types are private, ordinary backward induction is insufficient. Kreps and Wilson's sequential equilibrium framework combines sequential rationality with a system of beliefs and consistency conditions that link those beliefs to limits of fully mixed behavior. A player's response at a previously unreached information set cannot be assessed without stating what the player would believe upon arrival. Perfect Bayesian equilibrium is commonly used as an organizing concept, but precise conditions vary by formulation and problem. One must state the exact refinement rather than treating all belief-consistency requirements as interchangeable.
Sequential models support analyses of entry deterrence, labor contracting, investment, bargaining, crisis escalation, and platform regulation. Their common insight is institutional: observed commitments and information at intermediate decisions can transform an outcome more strongly than a static change in terminal payoff numbers. The scientific test is whether the commitment technology and the informational path actually exist in the market under examination.
7. Repeated Games, Reputation, and Cooperation
A repeated game contains multiple plays of a stage game, with current strategies conditioned on histories and payoffs aggregated over time. With discount factor δ in [0,1), an infinite horizon may be represented by the normalized payoff (1−δ)∑_{t=0}^∞ δ^t u_i(a_t). Future losses can then support current cooperation, but cooperation requires sufficiently patient agents, suitable monitoring, and credible continuation strategies. Repetition alone establishes neither mutual trust nor an efficient outcome.
Take the prisoner's dilemma above with cooperation payoff 3, deviation reward 5, and punishment payoff 1. Under a grim-trigger strategy, players cooperate until a defection occurs and thereafter defect forever. The cooperative continuation value is 3/(1−δ) without normalization. A one-time deviation gives 5 + δ/(1−δ). Cooperation is sustainable against this deviation exactly when 3/(1−δ) ≥ 5 + δ/(1−δ), equivalently δ ≥ 1/2. The inequality is a worked illustrative calculation: it relies on perfect observation, the specified payoffs, deterministic punishment and infinite continuation. Under imperfect monitoring, a noisy signal may punish innocent actions or conceal true defection, changing the feasibility and welfare of discipline.
The folk theorem of Fudenberg and Maskin characterizes broad sets of feasible, individually rational outcomes that can arise in repeated-game equilibria when players are sufficiently patient under specified dimensionality or player-number conditions. The result shows that dynamic incentives enlarge sustainable conduct and also increase equilibrium multiplicity. It is therefore a possibility theorem, not a universal prediction that cooperation will emerge. An explanation of a particular cooperative arrangement still needs history, information, enforcement rules and a credible selection process.
Repeated games support models of long-term supply relationships, cartel incentives, credit repayment, arms-control monitoring, shared environmental resources and online-platform reputation. The same apparatus can rationalize punishment, exclusion, retaliation and sustained collusion. The social desirability of cooperation is contingent on whose payoffs and rights are represented: cooperation by competing firms to raise prices may harm customers, while cooperation to sustain a commons may enhance welfare. Game-theoretic description and welfare evaluation must remain distinct analytical operations.
For a finite horizon with a commonly known terminal date and a stage game having a unique equilibrium, backward induction can predict repeated play of the stage equilibrium throughout. Relaxing common knowledge of types, permitting uncertain duration, or introducing multiple stage equilibria changes the conclusion. The analytically relevant difference between an indefinite repeated relationship and a fixed final round is the credibility of future contingent incentives, not merely the number of times two agents meet.
8. Incomplete Information and Bayesian Strategic Choice
Incomplete information arises when some payoff-relevant characteristics, such as costs, valuations, risk preferences or capacities, are privately known. Harsanyi's 1967 Bayesian formulation replaces this uncertainty with a type space and a hypothetical initial move by nature. For each player i, a type t_i is drawn from a commonly modeled probability structure; an action strategy s_i maps information about that type into an action or mixture. A Bayesian Nash equilibrium requires each type to select a best response given its conditional beliefs about others' types and their strategies.
A formal Bayesian game consists of a player set, type spaces T_i, action spaces A_i, a joint distribution over types when a common prior is assumed, type-dependent utility functions, and information partitions or observed signals. A strategy is a map s_i: T_i → Δ(A_i), not merely a single action. The Bayesian equilibrium condition integrates payoffs against the conditional distribution of unobserved types and any opponent randomization. Harsanyi's 1968 treatment of basic distributions examines both consistent representations derived from a common probability model and circumstances where subjective distributions are inconsistent. The common-prior assumption is an explicit modeling restriction rather than a logical definition of uncertainty itself.
In a simple entry game, an incumbent may privately know whether its marginal cost is low or high. An entrant observes market prices but not cost. Low prices could reflect aggressive pricing by a genuinely efficient firm or costly mimicry by a higher-cost firm. The entrant's posterior belief after observing price affects its entry threshold, while the incumbent chooses price partly to influence that belief. This interaction produces a signaling game, rather than a static response to an exogenous price. Spence's job-market signaling analysis established a related logic: a costly observable signal may differentiate private types when its costs differ across types. Riley's screening and signaling survey documents how such mechanisms entered labor, industrial organization and finance.
A separating equilibrium assigns distinguishable observable actions to different types; a pooling equilibrium makes at least some types choose the same observable action. Whether either exists depends on incentive constraints, signal costs, prior beliefs and the allowed off-path inferences. Observed differentiation by education or price does not establish pure signaling because human capital, quality, preferences and other causal mechanisms may also affect the outcome. To identify a signal's strategic role, empirical work must distinguish its productivity effect, signaling cost and information revealed to relevant counterparties.
The scope of Bayesian games extends to procurement, political competition, monetary policy announcements, bargaining over unknown valuations, labor markets and insurance selection. The advantage is transparent representation of what each agent knows and what they infer from others. The cost is model sensitivity to priors, feasible type spaces, equilibrium refinements and observed communication. Where those inputs cannot be measured, rigorous analysis should describe the set of outcomes consistent with plausible information structures rather than report a falsely precise point prediction.
9. Communication, Correlation, and Information Design
A correlated equilibrium allows players to condition their moves on privately received recommendations from a mediator whose messages may be statistically dependent. In Aumann's correlated-strategy framework, an outcome distribution μ over action profiles is obedient when no player can gain by disobeying the recommended action, conditional on that recommendation. In a finite game the obedience condition for player i, recommended a_i, and alternative a'i is ∑{a_-i} μ(a_i,a_-i)[u_i(a_i,a_-i)−u_i(a'i,a-i)] ≥ 0. The constraint compares conditional consequences after receipt of information; it is stronger than a blanket claim that all recommendations are always followed.
Correlation can support outcomes unavailable through independent mixed strategies and can enlarge the attainable payoff set in many games. A mediator might recommend one of two congestion-reducing routes according to a schedule, or advise traders about compatible counterparties. Recommendation design works only if messages are private or suitably restricted and compliance is incentive compatible. If everyone hears the same recommendation, the corresponding obedience constraints and strategic incentives may differ. Communication should thus be modeled as an information technology with a precise disclosure rule, not as an unspecified social lubricant.
Bergemann and Morris's Bayes correlated equilibrium analysis develops a robust perspective on incomplete-information games. When an analyst knows a basic game and a baseline information structure but allows participants to possess additional signals, equilibrium outcome predictions can be characterized through obedience constraints expressed by Bayes correlated equilibrium. This is especially relevant where the observer sees market actions but cannot verify every private signal traders possess. Their information-design synthesis distinguishes a designer actively choosing information rules from an analyst using possible information structures to map the range of potential outcomes.
Persuasion is a particular case where one actor commits to a rule that generates and communicates information before a recipient chooses an action. Kamenica and Gentzkow's Bayesian persuasion result characterizes when an informed communication policy benefits the sender and how sender-optimal signaling can be constructed under the maintained assumptions. Credible commitment to an information policy is central. A sender free to misreport an arbitrary message after learning the state faces a different incentive problem. Moreover, a receiver's preferences and available independent information constrain how far disclosure can move decisions.
Information design connects auction disclosures, platform recommendation systems, medical-risk messaging, disclosure regulation and policy communication. The normative analysis must specify whose welfare the designer maximizes. Better information for one group can harm another through strategic responses, and a message system may improve coordination while facilitating exclusion or price discrimination. Information effects are strategic externalities: each receiver's response alters incentives and outcomes for others. The distinction between passive information and designed signals therefore expands traditional equilibrium analysis toward rules governing the production and allocation of knowledge.
10. Bargaining, Coalitions, and Cooperative Solutions
Bargaining theory studies feasible agreements under conflict over surplus. In a noncooperative bargaining game, offers, counteroffers, acceptance, rejection, and discounting are strategic actions within an explicit sequence. Rubinstein's alternating-offers model establishes a disciplined framework in which delay costs and the timing of proposals shape equilibrium divisions of a pie. Bargaining power is thereby partly endogenous to institutions: an actor moving first, enjoying a higher discount factor, or controlling an outside option may obtain a different allocation, subject to the precise game and equilibrium assumptions.
For a unit pie and infinite alternating offers with complete information, no breakdown risk, and discount factors δ_1 and δ_2, the standard stationary subgame-perfect division gives the first proposer share (1−δ_2)/(1−δ_1δ_2). The second player receives δ_2(1−δ_1)/(1−δ_1δ_2). With δ_1 = δ_2 = 0.9, the first proposer receives 1/1.9, approximately 0.5263, and the second receives approximately 0.4737. With discount factors tending to one together, first-mover advantage diminishes. These expressions describe a particular bargaining institution; they are not an empirical formula for all negotiated wages, treaties or acquisitions.
Cooperative game theory focuses instead on what coalitions can achieve and on desirable divisions of joint value. A transferable-utility characteristic-function game specifies a worth v(S) for each coalition S ⊆ N. An allocation vector x is efficient when ∑i x_i = v(N). The core comprises efficient allocations satisfying ∑{i∈S} x_i ≥ v(S) for each coalition, so that no coalition could secure more for its members in aggregate. The core may be empty. In contrast, the Shapley value selects a single allocation based on each participant's average marginal contribution across all possible orders of coalition formation. Shapley's original value construction gives an axiomatic allocation rule with efficiency, symmetry, dummy-player and additivity properties.
For n players the Shapley value is φ_i(v) = ∑_{S⊆N{i}} [|S|!(n−|S|−1)!/n!][v(S∪{i})−v(S)]. The weights equal the probability that coalition S precedes i in a uniformly random ordering. For two symmetric players with v({1}) = v({2}) = 0 and v({1,2}) = 100, each receives 50. If v({1}) = 20 and v({2}) = 0 while v({1,2}) = 100, then φ_1 = 60 and φ_2 = 40. This reflects both unilateral worth and the marginal contribution to joint production. Such an allocation is interpretable only if the coalition worths and transferability assumptions are defensible.
The core and Shapley value answer different questions. Core membership concerns coalitional blocking; a Shapley-value allocation can lie outside the core in some games. An empirically measured transfer or profit share is not evidence of Shapley bargaining unless the relevant outside coalition worths and institutional decision rule are established. Coalition theory provides a disciplined vocabulary for joint ventures, shared infrastructure, cooperative production and multilateral negotiations, but it requires measurement of credible coalition options rather than hypothetical values chosen to force a preferred conclusion.
Bargaining and coalition analysis join noncooperative theory through implementation: a cooperative allocation may emerge from negotiations with specific credible threats, information asymmetries and enforcement institutions. The discipline's research program is strongest when it connects abstract allocation principles to the strategic process that can produce them.
11. Mechanism Design, Auctions, and Incentives
Mechanism design reverses the ordinary predictive question. Instead of treating game rules as given and asking which behavior they induce, the designer specifies a rule for messages, allocations and payments and asks which incentives, outcomes and welfare properties it implements. A direct mechanism maps reported types into outcomes, often including monetary transfers. Incentive compatibility requires a participant to prefer the intended report or action given the stipulated beliefs and strategic conduct of others. Individual rationality requires participation to be preferable to an available outside option. Budget balance and feasibility impose further constraints. These requirements can conflict, so evaluating a design requires an explicit ranking of objectives.
A central mechanism-design tool is the revelation principle: under appropriate conditions, outcomes implementable by an equilibrium of an indirect mechanism can be represented by a direct mechanism in which truth-telling is an equilibrium. This simplifies the search over institutional rules, but the relevant equilibrium notion, commitment capabilities and information assumptions must be preserved. It is not a universal guarantee that a simple truthful mechanism is administratively feasible or computationally tractable. Myerson's optimal-auction analysis studies how a seller chooses allocation and payment rules when bidders' valuations are private and expected revenue is an objective.
In a standard private-value sealed-bid second-price auction with independent valuations, truthful bidding is weakly dominant for each participant: bidding one's own value secures the object when the highest competing bid is below that value and avoids a winning price above value when competing bids exceed it. The winner pays the second-highest bid, and a participant's own bid does not determine the price conditional on winning. This proof depends on private values, the payment rule and the absence of relevant collusion or identity manipulation. Vickrey's auction analysis is the foundational research source.
The strategic environment changes under common or affiliated values. In a common-value auction the object has a value that is ultimately common to participants, yet they receive different private estimates. Winning reveals that one's estimate was high relative to competitors, making the conditional value of winning lower than an unconditional estimate. Milgrom and Weber's competitive-bidding theory derives auction revenue comparisons under particular information and dependence assumptions. A blanket assertion that one auction format always earns more is therefore unsound. The mechanism's rules, bidder risk preferences and signal structure determine the ordering.
Some desired outcomes are impossible under combined constraints. Myerson and Satterthwaite's bilateral-trade impossibility result establishes, for the studied overlapping private-value environments and incentive/participation conditions, that full ex post efficiency cannot generally be combined with the required incentive, individual-rationality and no-outside-subsidy properties. This finding gives institutional design a sharp economic meaning: an apparently obvious target such as always transferring an object whenever the buyer values it more may be unattainable when each side privately knows its valuation and participation is voluntary.
Market design applies this logic to matching systems, spectrum auctions, procurement, digital advertising, health-care allocation and platform rules. Roth's practical design program insists on complementing equilibrium analysis with institutional detail, experiments and computation. A theoretical incentive guarantee is a building block; operating systems also need enforceability, comprehensible rules, entry conditions, fraud prevention, access, error handling and evidence from actual participants.
12. Matching, Networks, and Institutional Organization
Many economic allocations require matching counterparties rather than simply setting a single market-clearing price. A matching problem contains agents with preferences over partners, assignment constraints and a notion of stability. Gale and Shapley's college-admissions analysis developed the deferred-acceptance approach and foundational stability results. In a basic two-sided one-to-one setting, a matching is stable when no unmatched pair prefers one another to their assigned partners and the relevant individual-rationality requirements hold. The order and side making proposals can influence which stable allocation is produced.
A deferred-acceptance procedure repeatedly allows proposing agents to apply to their most preferred acceptable untried recipient, while recipients tentatively hold their most preferred proposals subject to capacity and reject others. With finite preferences and suitable responsive priority rules, the process terminates at a stable matching in the relevant domain. The resulting incentives can differ across sides. The institutional designation of applicants, priorities and capacities therefore forms part of the modeled mechanism, with consequences for fairness, strategic reporting and access.
Strategic network games introduce links as choices. Firms may form alliances; scientists collaborate; insurers share risk; users join online communities; financial institutions establish counterparties. Jackson and Wolinsky's network-formation analysis studies whether self-interested link formation and severance produce networks that are stable and socially efficient. Their findings demonstrate that stability need not coincide with network efficiency. The social benefits of a link may accrue to third parties who do not participate in the bilateral decision, so externalities arise even when each pair calculates its private gain carefully.
Networks add an endogenous structure to the strategic environment. A participant's payoff depends not only on direct neighbors but also, in many models, on paths, information spillovers, congestion and group formation. Policy can consequently act on link costs, data portability, interoperability, institutional membership rules, and the distribution of network surplus. Claims that a network is efficient require definition of the welfare function and counterfactual feasible networks; observed persistence by itself establishes only some form of institutional durability, not global optimality.
Matching and networks help connect game theory to labor allocation, public administration, health-care systems, financial intermediation and technological infrastructure. They also illustrate a general analytical constraint: the unit of strategic choice can be a partner, a coalition, or a link, and the equilibrium concept must fit the feasible alternatives and consent requirements of that unit.
13. Evolutionary, Learning, and Behavioral Approaches
Evolutionary game theory studies the distribution of strategies in populations that reproduce, imitate, adapt or persist according to payoffs. It is especially useful when explicit calculation of sophisticated beliefs is an inappropriate representation of the causal process. Maynard Smith's evolutionary-game framework developed evolutionarily stable strategies in environments where an individual's reproductive success depends on the prevalence of alternative behaviors. The underlying utility/payoff often represents reproductive fitness in biology; in economics it can represent survival, adoption or reinforcement, provided the interpretation is made explicit.
A strategy x is evolutionarily stable in a symmetric two-player game when, roughly, a small proportion of mutants using another strategy cannot obtain a fitness advantage over residents. For a pure strategy x against a mutant y, one sufficient classical criterion is that u(x,x) > u(y,x), or, in the event of equality, u(x,y) > u(y,y). An evolutionarily stable strategy is a symmetric Nash equilibrium under the usual setup, while symmetric Nash equilibrium alone need not guarantee evolutionary stability. Population dynamics such as replicator dynamics can show how relative strategy shares change, but attractor properties depend on the exact dynamic and payoff structure.
Learning models likewise replace the instant equilibrium leap with an updating process. Players can choose best responses to observed frequencies, reinforce rewarded actions, use regret minimization, or estimate beliefs from past play. The relation between the limit points of learning and Nash equilibrium varies with game class and information. Hart and Mas-Colell's uncoupled-learning limitation establishes an important impossibility in a specified setting: agents using only their own payoff information cannot be expected to converge to Nash equilibrium in all games under the relevant universal-convergence requirements. This is a warning against treating rational adaptation as a generic equilibrium-finding machine.
Behavioral game theory tests how actual players decide under strategic incentives, including bounded reasoning, noisy choice, social preferences and heterogeneous beliefs. McKelvey and Palfrey's quantal-response equilibrium replaces deterministic best response with probabilistic choice weighted toward higher expected utility and imposes mutual consistency on those choice probabilities. The model generates statistical predictions for experimental choice frequencies and connects equilibrium analysis to estimation. An observed mistake need not be interpreted as random noise alone: the structure of reasoning, stakes, cognitive load and social preferences may all matter.
Fehr and Schmidt's inequity-aversion model helps explain cooperation and punishment across experimental environments when some actors care about payoff distributions. Charness and Rabin's social-preference tests compare concerns for aggregate welfare, inequality and reciprocity. These studies establish that canonical self-interest assumptions are empirically contestable in some settings; they do not imply that one alternative preference specification is universally correct. In practice the scientific decision is whether the assumed motives and information restrictions fit the study's observed population and the institution under analysis.
A useful distinction is between strategic rationality and the measurable cognitive process of reaching a strategy. Equilibrium analysis specifies consistency; behavioral and learning approaches study the process that might produce or depart from it. Combining them can improve explanation when predictions are evaluated on held-out environments or field settings rather than fitted retrospectively to a single experiment.
14. Experiments, Empirical Identification, and Causal Claims
A well-designed game-theoretic experiment separates the rules of play from the mechanisms an investigator wants to test. Laboratory studies can vary information, matching protocols, repetition, communication and payment incentives while observing choices. Field studies add real stakes and institutional complexity but often make beliefs, private information and alternate strategies difficult to observe. Each design offers advantages and limitations. Strong causal identification of a treatment's effect can coexist with weak identification of why agents responded as they did.
Coordination experiments provide a direct example. Cooper and colleagues' communication-in-coordination research studies the effect of communication on selection and performance in coordination games. Changing communication rules can alter beliefs and reduce uncertainty over intended play. Yet to infer a general theory of social conventions from a laboratory treatment, the researcher must explain how real communication channels, participant identities and enforcement differ from experimental conditions. Strategic outcomes are sensitive to common knowledge of the experimental rules and to subjects' beliefs about each other's competence.
In common-pool resources, Ostrom, Walker and Gardner's self-governance experiments compare communication opportunities and sanctioning arrangements. Their findings show that institution-building and credible mutual commitments are relevant to cooperation. The studies support the possibility of self-governance within examined conditions, not a claim that all commons always self-organize successfully. Endogenous sanctions can improve incentive alignment and still involve costs, unequal power or external consequences that a complete policy assessment must record.
Social-preference experiments often use ultimatum, dictator, trust, public-good and punishment games. These games distinguish models only when their implied prediction sets differ within the actual experimental design. For example, rejection of a low offer in an ultimatum game may be consistent with fairness preferences, reputation under repeated or misunderstood interactions, strategic signaling, negative reciprocity or errors in belief. Changing anonymity, repetition and observable offers can help separate those explanations. A credible empirical conclusion should identify the population, stake scale, instructions, randomization and measured outcome, and distinguish a reduced-form treatment effect from a structural parameter estimate.
A systematic empirical workflow proceeds through preregistered or transparent hypotheses where feasible, clear payoff and information specification, random assignment or justified quasi-experimental design, uncertainty assessment, replication or robustness checks, and evaluation of transfer to new institutions. For quantitative prediction one should report uncertainty and alternative equilibrium selection rules. When multiple mechanisms fit the same choices, more observations of the same variable may add little information; changing an identifying institutional feature can be more revealing.
Strategic field interventions may produce general-equilibrium and spillover effects: raising incentives for one group changes the opportunities and beliefs of others. This is a distinctive challenge of causal inference in games because the treatment of one participant affects the exposure and payoffs of another. The stable-unit-treatment assumption can fail under interference, and the estimand must then describe the assignment mechanism, network structure or market-level policy response. Game theory and econometrics meet where the outcome process explicitly includes such strategic interdependence.
15. Oligopoly, Public Goods, Politics, and Security
Industrial organization makes game theory operational through pricing, capacity, advertising, product positioning, entry and research competition. Cournot quantity competition, Bertrand price competition and dynamic capacity investment produce different outcomes because available actions and timing differ. In the homogeneous-product Bertrand benchmark with identical constant marginal costs, sufficient capacities and simultaneous prices, undercutting pressures can yield marginal-cost pricing. With product differentiation, capacity constraints, search frictions, repeated interaction or private costs, the equilibrium implications change. The correct question is always what industry institution the strategic representation captures.
Price signaling can affect entry. Milgrom and Roberts' limit-pricing equilibrium study showed how an incumbent's observable price can convey information about privately known costs. A high-cost firm may have an incentive to imitate low-cost pricing to deter entry, making equilibrium signals costly and potentially distorted. The model's insight is an interaction between strategic information transmission and market structure. Evidence of a low incumbent price does not by itself establish deterrence: costs, demand, quality and temporary promotions are competing explanations.
Public-good and environmental games capture incentives to benefit from collective provision while privately avoiding contributions. Incentive-compatible enforcement, monitoring, recurring interaction, bounded group size and institutional trust can change the available equilibria. Ostrom and colleagues' commons experiments illustrate the role of rules formed or accepted by participants. Designing a collective-action mechanism also requires examining distributional fairness and rights, especially when sanctioning authority is unequal or affected parties cannot participate in the rule-setting process.
Political economy and international security use bargaining, incomplete information and commitment problems. Actors may risk conflict to signal resolve, yet public statements may be cheap talk when actions are not costly or verifiable. Schelling's analysis of deterrence and bargaining helped establish the role of credible threats and shared focal points. Interpretation requires factual grounding of the actors' information, objectives, capabilities and constraints. A model cannot turn an unobserved intention into a historical fact; it can organize counterfactual possibilities conditional on documented decisions and institutional context.
Financial-market games study liquidity, information acquisition, bank coordination, market microstructure, bilateral dealing and correlated portfolio choices. A market price can signal private information while simultaneously shaping incentives to collect and reveal it. Auctions are mechanisms of price discovery and allocation, and the 2020 Nobel committee's auction-theory assessment documents the practical importance of the work of Milgrom and Wilson. These applications require care with risk, price formation and enforcement constraints rather than immediate transfer from abstract bidding models.
Across these domains, an equilibrium may characterize inefficient competition, productive cooperation, discriminatory exclusion or harmful coordination. Game theory explains why conduct is strategically sustainable; regulation and public policy additionally ask which feasible rules alter incentives in a socially acceptable direction. Analytical neutrality about preferences is compatible with explicit welfare accounting and transparent institutional consequences.
16. Computation, Algorithms, and Artificial Agents
An equilibrium's mathematical existence is distinct from the complexity of computing it. Nash's fixed-point argument proves existence in finite games; it does not provide a universal efficient algorithm for finding an equilibrium of arbitrary precision. Daskalakis, Goldberg and Papadimitriou's complexity results establish PPAD-completeness for central finite-game equilibrium-computation problems. The class describes total search problems whose solutions are guaranteed to exist under certain parity/fixed-point structures. Such results do not say that every practically encountered game is hard; they show why a polynomial-time solver for arbitrary instances cannot be assumed from existence alone.
Algorithmic game theory therefore studies restricted game classes, approximation, decentralized learning, no-regret dynamics, mechanism implementation and welfare losses from decentralized choice. Roughgarden's algorithmic game theory lectures connect Nash-equilibrium computation with routing inefficiency, auction design and learning. For a network-routing problem, the price of anarchy compares a worst equilibrium outcome against a socially optimized feasible outcome, using a specified cost or welfare measure. Ratios are meaningful only with the same feasible set, demand and objective in numerator and denominator.
The practical problem of autonomous economic agents extends this agenda. Pricing algorithms can respond to one another's prices, trading programs can adapt to market microstructure, recommender systems can influence what others learn, and automated negotiators can make offers whose credibility depends on delegated authority. Such programs are strategic players when their actions affect counterparties and their objective functions govern updates. Their observable behavior does not establish human-like intentions; the measurable scientific object is the interaction among objectives, information, rules and adaptive policies.
A recent empirical contribution, Werner's algorithmic-and-human collusion study, published in January 2026, compares self-learning pricing algorithms with human sellers in controlled market environments. The abstract reports more collusive outcomes in specific simulated and experimental oligopoly treatments, with effects depending on market size, algorithm prevalence and seller experience. This evidence raises regulatory and economic questions about algorithmic adaptation. It does not establish that every pricing algorithm colludes or that observed high prices prove a tacit agreement: design, common shocks, demand changes, market concentration and algorithm heterogeneity remain alternative explanations requiring empirical assessment.
The strategic implications of artificial agents depend on observability of code and outputs, access to shared data, ability to make commitments, monitoring, replacement and shutoff rules, and legal accountability. Algorithmic interactions can involve reward specification errors and unexpected externalities; their analysis calls for interdisciplinary work with computer security, statistics, competition law and institutional design. Game theory supplies the incentive constraints, while audit logs, experiments and production telemetry supply evidence about behavior in the actual system.
17. Mathematical Limits, Welfare, and External Validity
Game theory supports precise deductions within an explicitly modeled interaction. The most important constraints on interpretation are internal consistency, multiplicity, unobserved beliefs, equilibrium selection, computational feasibility, social preferences, endogenous institutional change and empirical transfer. Each concern addresses a different gap between formal possibility and scientifically supported explanation. A model may be logically correct while a particular market fails its information assumptions; a model can also fit observed data without identifying which of several equilibrium mechanisms produced them.
Existence and uniqueness must be separated. Many finite games have mixed Nash equilibria; some have several pure equilibria; dynamic and repeated games can contain especially rich equilibrium sets. Existence is a structural property; selection asks which of several outcomes actual participants coordinate upon. Refinements use stronger reasoning conditions but can introduce sensitivity to off-path beliefs or small payoff changes. Stability adds a dynamic or learning process and tests whether nearby play returns to the candidate equilibrium. Robustness tests whether an implication remains true under credible alternative information or payoff specifications. Each further claim requires an additional argument beyond Nash's finite-game theorem.
An equilibrium is also distinct from an efficient arrangement. Pareto efficiency means there is no feasible alternative that makes some participant better off without making anyone worse off. This criterion depends on the set of participants and feasible alternatives. It does not determine a unique socially desirable allocation or resolve distributive justice. Cooperative surplus division, utilitarian welfare, cost-benefit analysis, equality of access and rights-based constraints can prescribe different evaluations. Comparing an observed equilibrium with a socially optimized benchmark therefore requires reporting the objective, the distribution of gains and losses, enforcement costs, participation conditions, and consequences for excluded parties.
The Myerson–Satterthwaite theorem demonstrates how efficiency may conflict with truthful information revelation and voluntary participation. The importance of such impossibility results is epistemic as well as practical: they identify where better institutional design cannot satisfy every desired property simultaneously under stated assumptions. Policymakers may then choose among efficiency loss, subsidies, information acquisition, redistribution and modified trading rights, with each alternative carrying its own institutional cost.
External validity depends on whether the underlying strategic mechanism transfers. A controlled auction with two bidders differs from a procurement system with coalitions, asymmetric bidder capacity, legal appeals and complex complementarities; a repeated public-goods experiment differs from a multigenerational commons with historical rights and unequal power. Strong inference requires an explicit correspondence between players, incentives, timing, available information, action constraints and credible enforcement in the model and in the target institution. If a central element is absent, transferring an equilibrium result is a new hypothesis requiring evidence.
Theoretical pluralism is useful when models are treated as complementary restrictions on the same problem. A signaling equilibrium describes how private information may appear in actions; a behavioral model describes how cognitive limits affect choice; a mechanism design model evaluates how institutional rules alter incentives; an econometric design investigates which causal restrictions actual outcomes support. A coherent study often needs all four, but importing the strongest conclusion from each without aligning assumptions risks internal contradiction.
18. Comparative Analytical Synthesis and Research Directions
The cumulative lesson of the field is that strategic explanation has several logically distinct levels. The first level specifies the participants, feasible actions, payoffs and interaction structure. The second determines what participants can observe and infer. The third identifies what they can commit to and what institutions can enforce. The fourth applies an appropriate consistency or stability criterion. The fifth examines whether actual behavior, learning and computational capabilities support the predicted conduct. The sixth evaluates consequences for welfare and distribution. These levels form an analytical sequence for studying existing theories, rather than a newly named model or a claim of original mathematical discovery.
A concrete comparison shows why the sequence matters. Suppose two firms charge high prices. A one-shot price game with homogeneous products may predict undercutting. A differentiated-products game may support positive markups; a repeated game may sustain mutual restraint; a Bayesian model may produce high prices reflecting private costs; a capacity-constrained game may generate scarcity markups; an algorithmic learning model may support elevated prices in a particular computational environment. The observation is the same, yet the causal accounts differ. To discriminate among them one must examine cost data, capacity, demand, entry, information, previous interactions and the algorithmic rules, possibly through policy variation or experiments. Game theory then supplies distinct falsifiable mechanisms rather than an all-purpose narrative label.
The distinction between mechanism and solution concept is equally important. Nash equilibrium, correlated equilibrium, subgame perfection, sequential equilibrium and an evolutionary stable strategy each impose a different consistency requirement on specified primitives. None uniquely determines the payoff functions or the institution. A researcher who infers a player's preferences from observed equilibrium behavior must consider observational equivalence: different unobserved preferences, information structures or equilibrium selections can generate the same observed actions. This creates an inverse problem in strategic empirical research.
A further synthesis concerns the role of information. The classical Bayesian approach treats type distributions and conditional beliefs as modeled inputs. Correlated equilibrium shows how information in recommendations alters the feasible distribution of strategic choices. The modern information-design literature investigates how the structure of information itself shapes the set of incentive-compatible outcomes. This progression explains why information governance is an economic institution: who may observe, aggregate, commit to, withhold or transmit information influences behavior as surely as allocation rules or prices do.
Another synthesis concerns time. Static equilibrium examines simultaneous mutual responses; dynamic refinements constrain behavior at intermediate decisions; repeated games make future incentives part of current opportunity costs; learning and evolutionary approaches replace an assumed equilibrium endpoint with an adjustment mechanism. These approaches can be combined if their assumptions are aligned. A repeated-game explanation of cooperation requires a credible monitoring technology and meaningful continuation. A learning account requires an update rule and enough time for its predictions to become relevant. Researchers should not invoke patience, reputation or convergence without measuring or justifying the corresponding conditions.
The computational perspective adds another necessary condition for implementation: agents or institutions must be able to find, approximate, or follow the recommended strategy. Algorithmic game theory studies this feasibility directly. For complex auctions, large networks and automated markets, tractable approximation and robustness to adversarial manipulation may matter more to public welfare than exact optimization in an idealized environment. Yet approximate implementation creates a further need for empirical audit, since computationally expedient rules alter incentives and may change the predicted equilibrium set.
These connections motivate a research agenda organized around falsifiable questions. How can belief distributions be identified when strategies and information co-evolve? Which equilibrium refinements predict out-of-sample behavior after institutional change? When do communication technologies improve coordination without encouraging exclusion or collusion? Which allocation mechanisms remain incentive compatible under realistic computational limits and strategic data manipulation? How should welfare be evaluated when information is distributed unevenly and participation constraints differ across groups? Under which market conditions does algorithmic pricing reduce competition, and which interventions improve outcomes without damaging beneficial automated coordination? Evidence from the 2026 algorithmic-pricing comparison supplies a concrete research setting for the last question rather than a universal regulatory answer.
The field's interdisciplinary reach is therefore structural. Logic and mathematics formalize equilibria; psychology and experimental science identify behavioral processes; statistics and econometrics test causal claims; computer science handles computation and learning; law specifies rights, institutions and enforcement; political science studies collective choice and coercion; biology investigates evolutionary stability. Game theory is most powerful when each discipline contributes a measurable part of the same interaction and the resulting explanation is checked for common assumptions.
Conclusion
Game Theory establishes a rigorous science of strategic interdependence. Its central achievement is a language for explaining how one participant's feasible choices and expectations depend on the anticipated choices of others. The discipline's foundations in minimax reasoning and Nash equilibrium expanded through dynamic credibility, Bayesian information, coalition allocation, bargaining, auctions, repeated cooperation, evolutionary selection and experimentally observed behavior. Contemporary work connects those foundations to designed information structures, computational complexity and interacting artificial agents.
The scientific unity of Game Theory lies in the relation between interaction rules and strategically coherent behavior. Formal equilibrium concepts reveal what behavior can be sustained under explicit assumptions. Empirical analysis determines which mechanisms operate in identifiable settings. Mechanism design and public policy use those findings to compare feasible institutions and their distributional consequences. The correct object of scholarly explanation is therefore the interaction, including its information, enforcement and temporal structure.
Three conclusions follow. First, equilibrium predictions inherit the scope and limits of their assumed preferences, beliefs, strategies and institutions. Second, socially desirable outcomes require explicit welfare criteria and feasible implementation conditions in addition to strategic consistency. Third, advancing the field depends on linking exact theory, identified behavior, computational practicality and institutional evidence in a reproducible way. These conclusions synthesize established results while defining a coherent research orientation for Game Theory within Economics by Bogdanova.
The manuscript provides a scientific foundation for subsequent independently commissioned Research Dossiers and Publications. Its reserved canonical URL will be activated only through a separate authorized publishing release and verified production readback.
References
-
Aumann, Robert J. (1974). “Subjectivity and Correlation in Randomized Strategies” Journal of Mathematical Economics 1 (1): 67–96.
-
Aumann, Robert J. (1976). “Agreeing to Disagree” Annals of Statistics 4 (6): 1236–1239.
-
Bergemann, Dirk, and Stephen Morris (2016). “Bayes Correlated Equilibrium and the Comparison of Information Structures in Games” Theoretical Economics 11 (2): 487–522.
-
Bergemann, Dirk, and Stephen Morris (2019). “Information Design: A Unified Perspective” Journal of Economic Literature 57 (1): 44–95.
-
Charness, Gary, and Matthew Rabin (2002). “Understanding Social Preferences with Simple Tests” Quarterly Journal of Economics 117 (3): 817–869.
-
Cooper, Russell, Douglas V. DeJong, Robert Forsythe, and Thomas W. Ross (1992). “Communication in Coordination Games” Quarterly Journal of Economics 107 (2): 739–771.
-
Daskalakis, Constantinos, Paul W. Goldberg, and Christos H. Papadimitriou (2009). “The Complexity of Computing a Nash Equilibrium” SIAM Journal on Computing 39 (1): 195–259.
-
Fehr, Ernst, and Klaus M. Schmidt (1999). “A Theory of Fairness, Competition, and Cooperation” Quarterly Journal of Economics 114 (3): 817–868.
-
Fudenberg, Drew, and Eric Maskin (1986). “The Folk Theorem in Repeated Games with Discounting or with Incomplete Information” Econometrica 54 (3): 533–554.
-
Gale, David, and Lloyd S. Shapley (1962). “College Admissions and the Stability of Marriage” American Mathematical Monthly 69 (1): 9–15.
-
Harsanyi, John C. (1967). “Games with Incomplete Information Played by ‘Bayesian’ Players, I–III. Part I: The Basic Model” Management Science 14 (3): 159–182.
-
Harsanyi, John C. (1968). “Games with Incomplete Information Played by ‘Bayesian’ Players, Part III: The Basic Probability Distribution of the Game” Management Science 14 (7): 486–502.
-
Hart, Sergiu, and Andreu Mas-Colell (2003). “Uncoupled Dynamics Do Not Lead to Nash Equilibrium” American Economic Review 93 (5): 1830–1836.
-
Jackson, Matthew O., and Asher Wolinsky (1996). “A Strategic Model of Social and Economic Networks” Journal of Economic Theory 71 (1): 44–74.
-
Kamenica, Emir, and Matthew Gentzkow (2011). “Bayesian Persuasion” American Economic Review 101 (6): 2590–2615.
-
Kreps, David M., and Robert Wilson (1982). “Sequential Equilibria” Econometrica 50 (4): 863–894.
-
Maynard Smith, John (1982). Evolution and the Theory of Games. Cambridge: Cambridge University Press.
-
McKelvey, Richard D., and Thomas R. Palfrey (1995). “Quantal Response Equilibria for Normal Form Games” Games and Economic Behavior 10 (1): 6–38.
-
Milgrom, Paul R., and John Roberts (1982). “Limit Pricing and Entry under Incomplete Information: An Equilibrium Analysis” Econometrica 50 (2): 443–459.
-
Milgrom, Paul R., and Robert J. Weber (1982). “A Theory of Auctions and Competitive Bidding” Econometrica 50 (5): 1089–1122.
-
Myerson, Roger B. (1981). “Optimal Auction Design” Mathematics of Operations Research 6 (1): 58–73.
-
Myerson, Roger B., and Mark A. Satterthwaite (1983). “Efficient Mechanisms for Bilateral Trading” Journal of Economic Theory 29 (2): 265–281.
-
Nash, John (1951). “Non-Cooperative Games” Annals of Mathematics 54 (2): 286–295.
-
Nash, John F. (1950). “Equilibrium Points in N-Person Games” Proceedings of the National Academy of Sciences 36 (1): 48–49.
-
Ostrom, Elinor, James Walker, and Roy Gardner (1992). “Covenants with and without a Sword: Self-Governance Is Possible” American Political Science Review 86 (2): 404–417.
-
Riley, John G. (2001). “Silver Signals: Twenty-Five Years of Screening and Signaling” Journal of Economic Literature 39 (2): 432–478.
-
Roth, Alvin E. (2002). “The Economist as Engineer: Game Theory, Experimentation, and Computation as Tools for Design Economics” Econometrica 70 (4): 1341–1378.
-
Roughgarden, Tim (2016). Twenty Lectures on Algorithmic Game Theory. Cambridge: Cambridge University Press.
-
Royal Swedish Academy of Sciences (2020). “The Prize in Economic Sciences 2020: Press Release” Stockholm, October 12.
-
Rubinstein, Ariel (1982). “Perfect Equilibrium in a Bargaining Model” Econometrica 50 (1): 97–109.
-
Schelling, Thomas C. (1960). The Strategy of Conflict. Cambridge, MA: Harvard University Press.
-
Selten, Reinhard (1965). “Spieltheoretische Behandlung eines Oligopolmodells mit Nachfrageträgheit: Teil I” Zeitschrift für die gesamte Staatswissenschaft 121 (2): 301–324.
-
Shapley, Lloyd S. (1953). “A Value for N-Person Games” In Contributions to the Theory of Games II, edited by Harold W. Kuhn and Albert W. Tucker, 307–318. Princeton: Princeton University Press.
-
Spence, Michael (1973). “Job Market Signaling” Quarterly Journal of Economics 87 (3): 355–374.
-
Vickrey, William (1961). “Counterspeculation, Auctions, and Competitive Sealed Tenders” Journal of Finance 16 (1): 8–37.
-
von Neumann, John (1928). “Zur Theorie der Gesellschaftsspiele” Mathematische Annalen 100: 295–320.
-
von Neumann, John, and Oskar Morgenstern (1944). Theory of Games and Economic Behavior. Princeton: Princeton University Press.
-
Werner, Tobias (2026). “Algorithmic and Human Collusion” Economic Journal, article ueag013, published January 27, 2026.