Global Games¶

Pick a single prediction out of a coordination game's many self-fulfilling equilibria by giving each player a private noisy signal of a common fundamental and deleting dominated strategies inward until one threshold cutoff survives.

Core Idea¶

Global games are a class of incomplete-information coordination games, introduced by Carlsson and van Damme (1993) and developed into a widely applied tool for crisis modeling by Morris and Shin (1998, 2003), in which a specific informational structure — each player receives a private noisy signal about a common but unobserved fundamental — is used to select a unique equilibrium in a game that, under complete information, would have multiple self-fulfilling equilibria. The equilibrium-selection mechanism operates through iterated deletion of strictly dominated strategies: at extreme signal values, one action dominates regardless of what others do (attacking a central bank whose reserves are certainly zero is dominant; not attacking when reserves are certainly unlimited is dominant); these dominance regions anchor beliefs at the boundaries; inward iteration then pins down each player's threshold strategy across the entire signal space, ultimately yielding a unique threshold equilibrium in which every player plays one action if and only if their private signal falls on the appropriate side of a calculable cutoff. The cutoff itself is determined by the noise structure of the signal distribution and the payoff parameters of the underlying game, not by exogenous coordination devices or focal points. The structural payoff is the dissolution of the multiple-equilibrium indeterminacy that had frustrated applied coordination-game analysis since Schelling: where complete information leaves the outcome anywhere in a wide range depending on arbitrary beliefs, the global-games perturbation yields a single equilibrium with sharp comparative statics. The canonical application is currency attacks (Morris and Shin 1998): each speculator receives a private estimate of the defending central bank's reserve position; under complete information there are two equilibria — universal attack and no attack — both self-fulfilling across a wide range of fundamentals; under the global-games informational structure a unique threshold strategy emerges, with the attack threshold determined by the noise variance, the cost of attacking, and the reserve requirement for a successful defense. The framework extends to bank runs (Goldstein and Pauzner 2005), sovereign-debt rollover crises, technology adoption under network externalities, and political participation in regime change, in each case converting an intractable multiplicity into a unique, empirically testable prediction.

Structural Signature¶

Sig role-phrases:

the base coordination game — a game with strategic complementarities that, under complete information, admits multiple self-fulfilling equilibria across a band of fundamentals
the common fundamental — an unobservable parameter that determines the game's payoffs
the private-signal informational structure — each player receives a private noisy signal of the fundamental, possibly with a small public component
the dominance regions — the signal extremes where one action is dominant regardless of others (attack when reserves are certainly zero; refrain when certainly unlimited), anchoring beliefs at the boundaries
the iterated-deletion engine — removal of strictly dominated strategies inward from the dominance regions, the mechanism that collapses the equilibrium set
the unique threshold strategy — the surviving equilibrium: each player acts iff their signal falls on the calculable side of a cutoff
the cutoff as a function of noise and payoffs — the threshold determined by the signal noise structure and the base game's payoff parameters, not by focal points or coordination devices
the comparative statics — the signed shifts of the threshold with reserves, cost of attacking, and noise variance, the source of testable predictions
the public-vs-private asymmetry — a public signal enters all beliefs at once and so moves the threshold non-linearly and disproportionately to its private-signal equivalent
the robustness scope (its boundary) — the selection applies only where a multiplicity is destroyed by a small informational perturbation; genuinely robust multiplicities admit no unique prediction

What It Is Not¶

Not a causal account of why crises happen. Global games is an equilibrium-selection method — an informational perturbation plus iterated dominance that picks one prediction out of many — not a theory of what triggers a currency attack or a bank run. It tells the analyst which coordination outcome obtains and where the threshold sits, not what shock set events in motion.
Not a discovery that crises have a unique outcome. Uniqueness is a property of the modeling assumption (private noisy signals about a common fundamental), not an empirical finding about the world. It dissolves the complete-information multiplicity by construction; change the informational structure and the multiplicity can return, so the single threshold is a derived artifact, not a fact that markets coordinate on one equilibrium.
Not the claim that crises are fundamentals-driven rather than sentiment-driven. The framework dissolves that dichotomy rather than taking a side: because the noise structure ties the unique equilibrium to the fundamental, fundamentals govern the outcome even where intuition sees pure self-fulfilling belief. It does not assert that sentiment is irrelevant — it shows the question "fundamental or sentiment?" is the wrong one.
Not stochastic resonance or noise-induced order. Physical systems where noise resolves multistability (double-well potentials, phase transitions) look similar but have different formal content — stochastic differential equations and Kramers' escape rates, not iterated dominance and threshold strategies. The resemblance is a loose family analogy, not the same mechanism; "a global game" applied to such systems is metaphor.
Not applicable to any multiple-equilibrium system. The selection result bites only where a multiplicity is destroyed by a small informational perturbation among payoff-interdependent strategic agents (the Carlsson–van Damme robustness scope). Where a multiplicity is genuinely robust, or the system is non-strategic, the framework claims no unique prediction — the powerful uniqueness theorem is conditional, not universal.

Scope of Application¶

As an equilibrium-selection method, global games live across the applied coordination-game fields of game theory and economics, wherever the three ingredients the uniqueness theorem requires recur: a base coordination game with strategic complementarities, a common unobserved fundamental, and a private-signal informational structure. Physical noise-induced-order phenomena (stochastic resonance, double-well escape) share only a family resemblance and a different formal content, so they sit outside the map; the portable selection insight belongs to the parent coordination_problem_and_equilibrium_selection prime. - Currency attacks — the originating application (Morris–Shin), resolving the Obstfeld-style multiple-equilibrium problem in self-fulfilling speculative attacks, with the attack threshold set by reserves, attack cost, and noise variance. - Bank runs — Goldstein–Pauzner derive a unique run threshold as a function of bank fundamentals, replacing the indeterminate run/no-run multiplicity of the Diamond–Dybvig coordination game. - Sovereign-debt rollover crises — each creditor's rollover decision is keyed to beliefs about the others, and the global-games threshold predicts the crisis boundary. - Technology adoption under network externalities — adoption decisions with strategic complementarities yield unique adoption thresholds rather than the usual chicken-and-egg multiplicity. - Political regime change and revolutions — each citizen's decision to participate in protest depends on beliefs about others' participation, and the framework predicts participation thresholds in regime-strength terms. - Market microstructure — liquidity provision and trader entry/exit under strategic complementarities, where the same threshold logic selects a unique outcome.

Clarity¶

Global games dissolve a problem that haunted applied coordination-game analysis for decades: which of several self-fulfilling equilibria do agents actually play? Under complete information a currency attack, a bank run, or a debt rollover admits both the crisis outcome and its absence across a wide band of fundamentals, and the model is silent on the choice — Schelling's focal points offered only an unsystematic, culturally contingent guess. By perturbing the game with private noisy signals about the fundamental and running iterated deletion of dominated strategies inward from the dominance regions, global games supply the first mechanism-grounded equilibrium selection: a unique threshold strategy whose cutoff is a calculable function of the noise variance and the payoffs. The clarifying payoff is the replacement of an indeterminacy — the outcome could be anywhere in a range — with a single prediction that carries sharp comparative statics, which is what made the framework reorganize the applied crisis literature.

It also sharpens the long-running quarrel between fundamentals and sentiment in crises. The complete-information multiplicity invited a stark dichotomy — either crises are driven by fundamentals or they are arbitrary sunspot coordination — and global games show the dichotomy is false: because the noise structure ties the unique equilibrium to the fundamental, fundamentals govern the outcome even where naive intuition sees pure self-fulfilling belief. That lets a practitioner stop asking the unanswerable "is this crisis fundamental or sentiment-driven?" and ask instead the tractable question the framework makes precise: given the signal noise and payoff structure, where is the threshold, and how does it move with reserves, with the cost of attacking, or with the precision of public information. The last of these surfaces a distinction the older analysis could not even state — that a small piece of public information shifts the threshold non-linearly and disproportionately to its private-signal equivalent, because it is common to all players' beliefs at once.

Manages Complexity¶

A coordination game with strategic complementarities under complete information is, analytically, a swamp: it admits a continuum of self-fulfilling equilibria over a wide band of fundamentals, the model is silent on which one obtains, and an applied theorist confronting a currency attack, a bank run, a debt rollover, and a protest movement faces four separate indeterminacies with no shared way to pin any of them down — the best the older apparatus offered was Schelling's focal points, an unsystematic and culturally contingent guess. Global games compress that swamp to a single procedure with a single output. By imposing one informational structure — each player gets a private noisy signal of the common fundamental — and running iterated deletion of dominated strategies inward from the dominance regions, the framework collapses the entire equilibrium set to a unique threshold strategy: every player acts if and only if their signal falls on the calculable side of a cutoff. The whole multiplicity problem, the thing that frustrated the field since Schelling, contracts to solving for one number. And the analyst need not re-derive that number's machinery for each application; the currency attack, the bank run, the sovereign rollover, the technology-adoption cascade, and the regime-change decision are recognized as one object — a base coordination game plus a common fundamental plus private signals — to which the same uniqueness theorem applies, so a family of intractable models becomes one tool deployed repeatedly.

The deeper compression is in what governs the surviving cutoff. The threshold is a calculable function of just two ingredients — the noise structure of the signal distribution and the payoff parameters of the base game — so the entire outcome space reduces to those few inputs, and the analyst tracks them and reads the crisis boundary off their relationship rather than reasoning through the belief hierarchy case by case. From that one cutoff the qualitative behavior follows by comparative statics: how the crisis threshold moves with reserves, with the cost of attacking, with the noise variance. The compression even tames the field's old fundamentals-versus-sentiment dichotomy, which had threatened to double the analytical burden by leaving every crisis to be classified as either fundamental or arbitrary sunspot: because the noise structure ties the unique equilibrium to the fundamental, the dichotomy dissolves, and "is this crisis fundamental or sentiment?" — unanswerable — is replaced by "where is the threshold and how does it move?", which the parameters answer directly. One subtlety the compression surfaces rather than hides is that public information must be tracked separately from private, because a public signal enters every player's beliefs at once and so shifts the threshold non-linearly and disproportionately to its private-signal equivalent — a distinction the pre-global-games analysis could not even state, now read straight off the same low-dimensional apparatus.

Abstract Reasoning¶

Global games license a tight sequence of reasoning moves, all flowing from the single result that the informational perturbation collapses the equilibrium set to a calculable threshold, so that the analyst reasons about a crisis by reasoning about a cutoff.

The foundational move is equilibrium selection by construction — turning an indeterminacy into a derivation. Faced with a coordination game that under complete information admits both the crisis outcome and its absence across a wide band of fundamentals, the analyst does not guess which equilibrium obtains (the Schelling focal-point move, unsystematic and culturally contingent) but deduces it: impose private noisy signals about the common fundamental, identify the dominance regions at the signal extremes where one action dominates regardless of others (attack when reserves are certainly zero; refrain when they are certainly unlimited), and iterate deletion of dominated strategies inward until a unique threshold strategy is pinned down. The reasoning runs from an informational assumption to a determinate prediction, replacing "the outcome could be anywhere in this range" with "the outcome is this, and here is why."

The second move is solve-for-the-cutoff, which is where the framework's predictive content lives. Once uniqueness is established, the analyst reasons to a single number — the threshold in signal space such that each player acts if and only if their signal falls on the appropriate side — and that number is a calculable function of just two ingredients: the noise structure of the signal distribution and the payoff parameters of the base game. The move is to read the crisis boundary off those inputs rather than reasoning through the infinite belief hierarchy ("what do I think others think I think…") case by case; the cutoff is the crisis boundary, and computing it is the analysis.

The third move is comparative statics on the threshold, the framework's main source of empirically testable predictions. Having located the cutoff, the analyst asks how it moves: raising reserves shifts the attack threshold one way, raising the cost of attacking shifts it another, increasing the noise variance another still. The reasoning is interventionist and directional — change a payoff or informational parameter, predict the signed shift in the crisis boundary — so the framework yields not just whether a crisis can occur but how its likelihood responds to policy and conditions, which is what made it reorganize the applied crisis literature.

A fourth, more counterintuitive move is public-versus-private information asymmetry. The analyst reasons that a public signal must be tracked separately from a private one of equal precision, because a public signal enters every player's beliefs simultaneously and so coordinates beliefs in a way private noise does not — shifting the threshold non-linearly and disproportionately. This licenses the surprising prediction that a small piece of public information (a central bank's statement) can move the crisis boundary far more than its private-signal equivalent, and it lets the analyst reason about the destabilizing potential of transparency itself — a distinction the pre-global-games complete-information analysis could not even state.

A fifth move is dissolving the fundamentals-versus-sentiment dichotomy by argument. Confronting the old question "is this crisis driven by fundamentals or by arbitrary self-fulfilling belief?", the analyst reasons that the dichotomy is false: because the noise structure ties the unique equilibrium to the fundamental, fundamentals govern the outcome even in cases naive intuition reads as pure sunspot coordination. The move replaces an unanswerable classification with the tractable "where is the threshold and how does it move?", and it is itself a piece of reasoning the framework makes available rather than a separate empirical finding.

Finally, the framework supports a boundary-drawing move on its own applicability via robustness reasoning. The analyst asks which features of a complete-information multiple-equilibrium structure survive a small informational perturbation and which do not — the Carlsson–van Damme robustness criterion — and so reasons about when the global-games selection legitimately applies (a base coordination game with strategic complementarities plus a common fundamental observed through private noise) versus when a multiplicity is genuinely robust and no unique prediction should be claimed. This keeps the powerful selection result scoped to the structures where its dominance-anchored argument actually bites.

Knowledge Transfer¶

Global games is a technical framework — an equilibrium-selection method built from a specific informational perturbation, iterated deletion of dominated strategies, and a threshold-strategy uniqueness theorem — so its transfer is the transfer of a piece of mathematical machinery, and that machinery carries as method within the strategic-interaction substrate while not carrying past it. Within game theory and applied economics the framework transfers cleanly and productively, because every target shares the same three ingredients the uniqueness theorem requires: a base coordination game with strategic complementarities, a common unobserved fundamental, and a private-signal informational structure. The whole apparatus — identify the dominance regions, iterate inward, solve for the cutoff as a calculable function of the noise structure and payoffs, run comparative statics on the threshold, and track public versus private information separately — carries without translation across currency attacks (Morris–Shin, the originating application, resolving the Obstfeld multiple-equilibrium problem), bank runs (Goldstein–Pauzner: a unique run threshold as a function of bank fundamentals), sovereign-debt rollover crises (each creditor's rollover decision keyed to beliefs about others), technology adoption under network externalities (unique adoption thresholds), political regime change (participation thresholds in protest), and market microstructure (liquidity provision under complementarities). These applications are not analogies but co-instances of one object — a base coordination game plus a common fundamental observed through private noise — to which the same theorem applies, which is exactly why a family of formerly intractable models became one repeatedly deployed tool. This is genuine within-domain mechanistic reach.

Beyond strategic-interaction substrates the framework does not transfer, and honesty requires saying so precisely. The mathematical content — Laplacian beliefs, iterated dominance, threshold-strategy uniqueness — is bound to games of payoff-interdependent agents under uncertainty; engineering, biological, and physical systems with multiple equilibria are not "global games" in the technical sense, and the framework carries neither its interventions nor its sharp predictions to them. There are surface-similar phenomena elsewhere — stochastic resonance and noise-induced order in physics, developmental bifurcations and ecosystem regime shifts in biology, where noise also seems to resolve multistability — but these are different mechanisms with different formal content (stochastic differential equations and Kramers' escape rates, not iterated dominance and threshold strategies), sharing a vague family resemblance rather than a structural identity. Invoking "a global game" for any noise-resolves-multiplicity system is therefore loose analogy, not mechanism transfer, and should be marked as such. What genuinely does carry — the substrate-portable insight that an informational perturbation can collapse a coordination multiplicity to a unique selection — is already held at the prime level by coordination_problem_and_equilibrium_selection, of which global games is one specific technical solution (available precisely when such a perturbation can be imposed). So the honest split is between method-reach (the full apparatus transfers across strategic-interaction economics, where the three ingredients recur, and there genuinely) and its boundary (it does not reach non-strategic substrates; the portable selection insight belongs to the parent coordination/equilibrium-selection prime, and physical noise-induced-order phenomena are analogy, not the same mechanism). The full boundary is drawn in Structural Core vs. Domain Accent.

Examples¶

Canonical¶

The founding demonstration is Carlsson and van Damme's 1993 Econometrica paper, which built the technique on a 2×2 coordination game. Two players choose "invest" or "not invest"; the payoff to investing depends on a fundamental θ and on whether the other invests. Under complete information, for a middle range of θ both (invest, invest) and (not, not) are Nash equilibria — the classic multiplicity. Carlsson and van Damme perturbed the game: each player observes θ only through a private signal with small noise. At extreme signals one action is strictly dominant (invest when θ is surely high, refrain when surely low). Iterating deletion of dominated strategies inward from these regions collapses the multiplicity to a single surviving threshold strategy, and as the noise vanishes it selects the risk-dominant equilibrium — a unique, non-arbitrary prediction.

Mapped back: The 2×2 investment game is the base coordination game; θ is the common fundamental seen through the private-signal informational structure. The surely-high and surely-low ends are the dominance regions that anchor the iterated-deletion engine, whose output is the unique threshold strategy, its cutoff fixed by the noise and payoffs.

Applied / In Practice¶

Morris and Shin (1998) turned the technique on speculative currency attacks, resolving the indeterminacy of Obstfeld's self-fulfilling-attack models. A continuum of speculators each decides whether to attack a pegged currency; the central bank defends if attacked by too few. Under complete information there are two equilibria — everyone attacks, or no one does — across a wide band of reserve levels. Morris and Shin give each speculator a private noisy estimate of the bank's reserves. Iterated dominance then yields a unique threshold: speculators attack if and only if their signal falls below a cutoff, and the peg breaks below a critical fundamental. The cutoff moves with reserves, the cost of attacking, and the signal noise.

Mapped back: Speculators form the base coordination game with complementarities; reserves are the common fundamental, each read through private signals. Certain-insolvency and certain-solvency are the dominance regions; the unique threshold strategy fixes who attacks, and the comparative statics — how the attack cutoff shifts with reserves and attack cost — deliver the framework's testable predictions.

Structural Tensions¶

T1: Manufactured uniqueness versus empirical fact (a property of the assumption, not of the world). The framework's headline achievement is dissolving the multiple-equilibrium indeterminacy into a single threshold — but that uniqueness is a property of the private-signal modeling assumption, not a discovery that markets actually coordinate on one outcome. The multiplicity is destroyed by construction; adopt a different informational structure and it can return. The tension is that the framework's greatest selling point (a determinate, testable prediction where older analysis was silent) rests on an informational device the analyst imposes, so the confidence the sharp threshold inspires can be spurious if the assumed signal structure does not describe how agents actually observe the fundamental. A unique equilibrium read as a fact about the economy, rather than as a consequence of a Laplacian-beliefs assumption, over-claims. Diagnostic: Is the single threshold being treated as a derived artifact of the assumed information structure, or as an empirical fact that the market has one equilibrium?

T2: Sharp prediction versus sensitivity to the information structure (the fine print of the noise). The cutoff is a calculable function of the noise structure and payoffs — beautifully sharp — but that sharpness makes the prediction hostage to the precise specification of how agents observe the fundamental: the common prior, the signal distribution, the correlation of noise across agents. The selection is cleanest in the vanishing-noise limit, yet the higher-order-beliefs literature (Weinstein–Yildiz) shows that small changes in the assumed information structure can move or unravel the selected equilibrium. The tension is that global games trade the honest indeterminacy of the complete-information game for a determinacy that is only as robust as the informational assumptions it smuggles in — the multiplicity is not eliminated so much as relocated into the specification of beliefs. Diagnostic: Is the threshold robust to plausible variation in the assumed noise and belief structure, or does the sharp prediction depend on a knife-edge specification of how agents observe the fundamental?

T3: Transparency as stabilizer versus destabilizer (the public-information asymmetry). A public signal enters every player's beliefs at once, so it coordinates beliefs in a way private noise cannot — and therefore moves the threshold non-linearly and disproportionately to its private-signal equivalent. This yields one of the framework's most striking predictions: a small piece of public information (a central-bank statement) can shift the crisis boundary far more than the same precision delivered privately, so more transparency can be destabilizing. The tension is genuinely two-edged: public information improves the accuracy of beliefs about the fundamental yet simultaneously amplifies the coordination that produces crises, so the welfare sign of disclosure is ambiguous and depends on parameters. A policy of "provide better information" is not automatically stabilizing under this logic. Diagnostic: Does the proposed disclosure sharpen agents' estimate of the fundamental, or does its very publicness coordinate beliefs enough to move the crisis threshold against stability?

T4: Selection method versus causal theory (what determines which outcome, not what triggers it). Global games answers "which of the coordination equilibria obtains, and where is the threshold" — it is an equilibrium-selection method, not an account of what shock set a crisis in motion. That abstraction from triggers is exactly what gives it reach: the same apparatus fits currency attacks, bank runs, and protests because it does not care what perturbed the fundamental. But the same abstraction is a limit — the framework can tell you the peg breaks below a critical fundamental without explaining why the fundamental deteriorated, so it locates the boundary of a crisis while remaining silent on its onset. The tension is that its cross-application power and its causal silence are the same feature: it selects among outcomes precisely by not theorizing their origin. Diagnostic: Is the question "which coordination outcome obtains and where is its threshold" (the framework's territory), or "what caused the fundamental to move" (which it does not address)?

T5: Robustness scope versus universal multiplicity-killer (where the selection legitimately bites). The uniqueness theorem is conditional, not universal: it selects an equilibrium only where a multiplicity is destroyed by a small informational perturbation among payoff-interdependent strategic agents (the Carlsson–van Damme scope). Where a multiplicity is genuinely robust to such perturbation, or the system is non-strategic, the framework claims no unique prediction. The tension is that the selection result is so powerful and so clean that it invites application to any multiple-equilibrium system, but doing so past its scope manufactures a false determinacy — reading a "global game" into a robustly multiple game, or into a physical noise-resolves-multistability system, claims a uniqueness the argument does not deliver. The method must be scoped by a robustness check, not applied wherever multiplicity is inconvenient. Diagnostic: Is the multiplicity here actually destroyed by a small informational perturbation among strategic agents, or is it robust — in which case no unique threshold is licensed?

T6: Autonomy versus reduction (a specific technical method or the instance of an equilibrium-selection parent). Global games is a named, heavily formalized apparatus — Laplacian beliefs, iterated dominance from dominance regions, threshold-strategy uniqueness — bound to strategic-interaction substrates; strip that machinery and it is not "global games" any more. What genuinely travels beyond game theory is the substrate-portable insight that an informational perturbation can collapse a coordination multiplicity to a unique selection — and that insight is already held at the prime level by coordination_problem_and_equilibrium_selection, of which global games is one specific technical solution, available precisely when such a perturbation can be imposed. Physical noise-induced-order phenomena (stochastic resonance, Kramers escape) share only a family resemblance with different formal content. The tension is between a canonical, discipline-reorganizing method and the recognition that its portable lesson belongs to the parent selection prime, not to its iterated-dominance machinery. Diagnostic: Resolve toward coordination_problem_and_equilibrium_selection when asking what carries beyond strategic games; toward "global games" specifically when the three ingredients (base coordination game, common fundamental, private signals) are present and the threshold is to be solved in situ.

Structural–Framed Character¶

Global games sits at mixed on the structural–framed spectrum — an evaluatively neutral piece of formal machinery, but one bound to a strategic-interaction substrate and stated in irreducibly game-theoretic vocabulary, and (unlike a naturally-occurring pattern) itself an analytical method an economist imposes rather than a mechanism nature runs. On evaluative_weight it reads structural: the framework convicts and praises nothing — it selects which coordination equilibrium obtains and where the threshold sits, a positive derivation, not a normative verdict; even its most policy-charged output (more transparency can destabilize) is a signed comparative static, not a value judgment. On human_practice_bound it reads mixed: the coordination dynamics it models — speculators attacking a peg, depositors running a bank — occur among real payoff-interdependent agents whether or not a theorist writes them down, so the subject is not observer-constituted; but the apparatus itself is a modeling practice (impose private signals, iterate dominance, solve for the cutoff) that exists only when someone performs it, and it requires strategic agents with beliefs, unlike isostasy's agent-free lithosphere. Institutional_origin reads framed: the framework is a designed artifact of a discipline — Carlsson–van Damme (1993), Morris–Shin — its uniqueness result an engineered consequence of the Laplacian-beliefs assumption, not a balance found in nature (the entry is explicit that uniqueness is a property of the assumption, not of the world). On vocab_travels it reads framed: dominance regions, iterated deletion of dominated strategies, threshold strategy, common fundamental, private-signal structure do not float free of games of incomplete information, and off that substrate there is no literal referent. And on import_vs_recognize the transfer is bimodal exactly as the entry documents — within strategic-interaction economics the method is recognized intact across currency attacks, bank runs, debt rollovers, adoption cascades, and protest (co-instances of one object), while physical noise-resolves-multiplicity phenomena (stochastic resonance, Kramers escape) are only import-by-analogy with different formal content.

The portable structural skeleton is coordination_problem_and_equilibrium_selection — the substrate-spanning insight that an informational perturbation can collapse a coordination multiplicity to a unique selection. That skeleton genuinely travels, but it is precisely what global games instantiates from its parent: the cross-domain reach belongs to the general selection prime, of which global games is one specific technical solution (available only when such a perturbation can be imposed), while the iterated-dominance machinery, the threshold-strategy theorem, and the noise-and-payoff cutoff stay home. Its character: an evaluatively neutral, discipline-engineered equilibrium-selection method, real in the strategic dynamics it models yet constituted by the theorist's apparatus and pinned to game-theoretic vocabulary, structural only in the coordination/equilibrium-selection skeleton it borrows from its parent.

Structural Core vs. Domain Accent¶

This section decides why global games is a domain-specific abstraction and not a prime, and carries the case for its domain-specificity in one place.

What is skeletal (could lift toward a cross-domain prime). Strip the game-theoretic machinery and one thin relational lesson survives: a small perturbation that ties each unit's local information to a common underlying state can collapse a self-reinforcing multiplicity to a single selected outcome. The portable pieces are abstract — a system that admits many mutually-consistent configurations, a shared but imperfectly-observed condition that ought to discipline which one obtains, and a perturbation that breaks the symmetry among the configurations to leave one. Nothing there requires speculators, reserves, or beliefs. That skeleton is genuinely substrate-portable, which is exactly why the entry files it under coordination_problem_and_equilibrium_selection, of which global games is one specific technical solution — available precisely when such an informational perturbation can be imposed. But that selection insight is the core global games shares, not what makes it global games.

What is domain-bound. Almost all the content is game-theoretic furniture, and none of it survives extraction. The perturbation is not generic — it is the private-signal informational structure, each strategic agent receiving a noisy read of the common fundamental. The selection is not generic — it is iterated deletion of strictly dominated strategies run inward from dominance regions at the signal extremes, terminating in a unique threshold strategy whose cutoff is a calculable function of the noise structure and the base game's payoffs. Its instruments (Laplacian beliefs, the comparative statics of the threshold, the public-versus-private asymmetry), its uniqueness theorem, and its worked cases (Morris–Shin currency attacks, Goldstein–Pauzner bank runs) are all bound to games of payoff-interdependent agents under uncertainty. The decisive test: remove the strategic agents with beliefs about one another and there is no iterated dominance and no threshold at all — only the looser, formally different fact that noise can resolve multistability, which in physics runs on stochastic differential equations and Kramers' escape rates, a family resemblance rather than the same mechanism.

Why this does not clear the prime bar. A prime's vocabulary travels and its transfer is recognition of the same mechanism, not analogy. Global games' transfer is bimodal. Within strategic-interaction economics the whole apparatus moves intact — identify the dominance regions, iterate inward, solve for the cutoff, run comparative statics, track public versus private information — across currency attacks, bank runs, sovereign-debt rollovers, adoption cascades, protest participation, and market microstructure, because these are co-instances of one object (a base coordination game plus a common fundamental observed through private noise), not analogies. That is genuine within-domain method-reach. Beyond strategic substrates it travels only by analogy: engineering, biological, and physical multiple-equilibrium systems are not "global games" in the technical sense, and calling a noise-resolves-multiplicity phenomenon a global game borrows the shape while discarding the iterated-dominance mechanism that names it. And when the bare structural lesson is wanted cross-domain — that an informational perturbation can collapse a coordination multiplicity to a unique selection — it is already carried, in more general form, by coordination_problem_and_equilibrium_selection. The cross-domain reach belongs to that parent; "global games," as named, carries the strategic-interaction machinery that should stay home.

Relationships to Other Abstractions¶

Current abstraction Global Games Domain-specific

Parents (4) — more general patterns this builds on

Global Games is part of Bayesian Nash Equilibrium Domain-specific

A Global Game contains a Bayesian equilibrium over signal-conditioned strategies; its uniqueness result selects one threshold member of that solution class.
Global Games is part of Dominated Strategy Domain-specific

Global Games contains iterated deletion of dominated strategies as the engine that moves inward from extreme-signal dominance regions.
Global Games is part of Threshold Prime

The selected equilibrium contains a cutoff that separates the signal region in which each of the two actions is optimal.
Global Games is a decomposition of Coordination Problem and Equilibrium Selection Prime

Removing private-signal game machinery leaves a method for selecting one outcome from several self-sustaining coordination equilibria.

Hierarchy paths (24) — routes to 15 parentless roots

Global Games → Bayesian Nash Equilibrium → Nash Equilibrium → Equilibrium → Fixed Point

Show alternative paths (23)

Not to Be Confused With¶

Sunspot equilibria. Rational-expectations equilibria in which agents coordinate on an extrinsic random signal — one uncorrelated with any fundamental (weather, an announcement, a "sunspot") — so that the outcome is arbitrary self-fulfilling belief with no tie to the underlying state. This is precisely the multiplicity global games dissolves: the private-signal perturbation ties the surviving equilibrium to the common fundamental, deleting the freedom to coordinate on payoff-irrelevant noise. Tell: does the signal agents key on carry information about the fundamental (global game) or is it deliberately fundamental-irrelevant, with the equilibrium arbitrary by construction (sunspot)?
Herding and information cascades. Models where agents move sequentially and each infers the fundamental by observing predecessors' actions, so private information gets swamped and a cascade locks in. Global games has agents move simultaneously on private signals of the fundamental, and the coupling is strategic complementarity in payoffs, not Bayesian inference from an observed action history. Tell: are agents learning the state by watching what earlier agents did (cascade), or acting at once on their own noisy read of the state with no action-observation channel (global game)?
Schelling focal points. The pre-global-games device for picking among multiple equilibria: agents converge on a culturally or contextually salient option (the "obvious" meeting spot) with no mechanism guaranteeing it. Global games replaces this unsystematic, non-derivable salience with a derived threshold — iterated dominance from the signal extremes yields the cutoff as a calculable function of noise and payoffs. Tell: is the selected equilibrium justified by out-of-model salience/convention (focal point) or deduced inside the model from dominance regions and the noise structure (global game)?
General Bayesian (Harsanyi) games of incomplete information. The super-type: any game where players hold private types drawn from a common prior and play Bayesian Nash equilibria — a framework that generically retains multiple equilibria. Global games is the special sub-class where the type is a private noisy signal of a common fundamental and the noise-and-complementarity structure lets iterated dominance collapse the multiplicity to one threshold. Part-vs-whole: every global game is a Bayesian game, but almost no Bayesian game delivers the uniqueness theorem. Tell: does the informational structure specifically pin a unique threshold via inward dominance-deletion (global game), or does it just describe private types under a prior with the equilibrium set left multiple (general Bayesian game)?
The base coordination game itself (e.g. Diamond–Dybvig, Obstfeld's currency model). The complete-information game to which the global-games perturbation is applied — the object that admits both the crisis and no-crisis equilibria across a band of fundamentals. Confusing the two collapses the perturbation into the thing perturbed: the base game is the multiplicity-bearing input, global games is the informational operation performed on it that selects one outcome. Tell: are you naming the strategic-complementarity game with its multiple self-fulfilling equilibria (base game), or the private-signal-plus-iterated-dominance procedure that reduces it to a single threshold (global game)?
coordination_problem_and_equilibrium_selection (the parent prime). The substrate-portable insight that an informational perturbation can collapse a coordination multiplicity to a unique selection — the cross-domain skeleton global games instantiates, treated more fully as its own prime. Global games is one specific, formalized technical solution under this umbrella, available only when a private-signal perturbation can actually be imposed among strategic agents. Tell: strip away the iterated dominance, threshold strategy, and noise-and-payoff cutoff and ask what remains portable — bare multiplicity-plus-selection is the parent prime, while the dominance machinery that names the tool is global games.

Neighborhood in Abstraction Space¶

Global Games sits in a crowded region of the domain-specific corpus (1^st percentile for distinctiveness): several abstractions share nearly its structure, so a description that fits it tends to fit its neighbors too.

Family — Strategic Interaction & Game Theory (23 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-07-12