Overview
Subgame perfect equilibrium (SPE) is a refinement of Nash equilibrium for dynamic or sequential interactions represented as game trees (extensive-form games). It requires that players’ strategies constitute a Nash equilibrium not just in the overall game, but in every subgame that could arise after any history of play. In plain terms, strategies must be credible at every stage: if the game were to reach any point, the planned continuation must still be optimal for all players. SPE matters because it rules out non-credible threats and promises, yielding more realistic predictions in settings like entry deterrence, bargaining, contract renegotiation, and repeated interactions.
Origins and Credit
Reinhard Selten introduced subgame perfect equilibrium in 1965 as part of his foundational work on equilibrium refinements, for which he shared the 1994 Nobel Prize in Economics with John Nash and John Harsanyi. Backward induction—the step-by-step solution method for finite perfect-information games—traces to early results like Zermelo’s theorem, and was formalized for extensive forms by Harold Kuhn. SPE provides the rigorous criterion that justifies backward induction and extends it to games with imperfect information when proper subgames exist.
Core Idea and Mechanics
An extensive-form game specifies the order of moves, available actions, information sets (what a player knows when moving), and payoffs. A strategy for a player prescribes an action at every information set where that player could move—on or off the path of play.
A subgame is a part of the game tree that starts at a decision node that begins an information set containing only that node (a singleton information set) and contains all subsequent nodes without cutting across any information set. The whole game is itself a subgame; “proper” subgames start somewhere later in the tree.
A strategy profile is a subgame perfect equilibrium if it induces a Nash equilibrium in every subgame. Two practical tools follow:
- Backward induction: In finite games with perfect information (players move sequentially and always know where they are in the tree), solve from the end: at the last moves, pick best responses; then fold back to earlier moves, anticipating later optimal play. The resulting strategy profile is the unique SPE.
- One-shot deviation principle: In many dynamic games (including discounted infinite-horizon settings), a strategy profile is SPE if and only if no player can gain by deviating in a single period at any information set, holding future play fixed. This simplifies verification.
Relation to other refinements:
- Nash equilibrium: Every SPE is a Nash equilibrium, but the converse need not hold when non-credible threats support a Nash equilibrium.
- Sequential equilibrium: In games with imperfect information, SPE may not fully discipline beliefs off the path of play; sequential equilibrium refines further by pairing strategies with Bayes-consistent beliefs at every information set.
Key Assumptions and Conditions
- Extensive-form representation: The game’s timing, actions, and information structure are explicitly modeled.
- Perfect recall: Players remember their past actions and information; standard existence and characterization results assume this.
- Best responses at every subgame: Optimality must hold after every feasible history, not just along the expected path.
- Common knowledge of the game and rationality: Players understand the structure and anticipate others’ optimal reactions.
- Finite horizon for backward induction: For strict backward induction, the game must end after finitely many moves; with discounting, the one-shot deviation principle aids characterization in infinite settings.
Implications
- Eliminates non-credible threats: Promises to take self-harming actions later (to deter rivals today) cannot support an SPE unless those actions are indeed optimal when the time comes.
- Dynamic consistency: Plans chosen at the outset are optimal at all later decision points, ensuring time-consistent strategic behavior.
- Clear comparative statics via backward reasoning: Anticipating optimal continuation play clarifies how earlier choices (like capacity, contracts, or prices) will be disciplined by later responses.
- Design insight: To make a threat or promise credible, firms or policymakers must change payoffs in the relevant subgame (e.g., commit capacity, adopt policies, or write contracts that alter incentives ex post).
Example in Practice
Entry deterrence with a non-credible threat. An incumbent faces a potential entrant. Stage 1: the entrant chooses Enter or Stay Out. Stage 2 (if Enter): the incumbent chooses Fight (price war) or Accommodate (peaceful coexistence). Suppose fighting is costly for the incumbent and hurts both firms compared with accommodation, while staying out gives the entrant zero and the incumbent a monopoly profit.
Backward induction solves the subgame that begins if entry occurs. Given entry, the incumbent strictly prefers Accommodate to Fight (accommodation yields higher profit than a price war). Knowing this, the entrant anticipates accommodation and chooses Enter. The action profile (Enter, then Accommodate) is the subgame perfect equilibrium. Any Nash equilibrium sustained by the incumbent’s threat to Fight is not subgame perfect because that threat is not optimal in the entry subgame.
Managerial lesson: deterrence requires commitment that changes payoffs in the post-entry subgame—e.g., install excess capacity or sign contracts that make aggressive pricing privately optimal if entry occurs. Otherwise, sophisticated entrants will discount cheap talk and enter.
Limitations and Common Misunderstandings
- Imperfect information limits: Some games have no proper subgames due to information sets; then SPE coincides with Nash and may not cull non-credible threats. More refined concepts (sequential equilibrium, perfect Bayesian equilibrium) add belief discipline.
- Multiplicity: Many dynamic games (especially repeated ones) have multiple SPE; selection may depend on norms, history, or additional refinements (e.g., renegotiation-proofness).
- Behavioral realism: Backward induction assumes common knowledge of rationality and sophisticated reasoning; real-world players may deviate due to bounded rationality or reputation concerns.
- Commitment vs. credibility: SPE assesses credibility given the modeled payoffs. If unmodeled commitment devices (laws, contracts, technology) are available, predicted actions can differ.
- Off-path beliefs: SPE does not specify how players interpret unexpected moves; without belief restrictions, some implausible contingencies may remain. Sequential equilibrium addresses this.
Related Concepts (names only)
Nash Equilibrium; Sequential Equilibrium; Perfect Bayesian Equilibrium; Backward Induction; One-Shot Deviation Principle; Folk Theorem.
