Subgame perfect equilibrium (source code)

= Subgame perfect equilibrium
{wiki}

= Subgame perfect equilibria
{synonym}

A <subgame perfect equilibrium> is a <strategy profile> inducing a <Nash equilibrium> in every subgame, including subgames not reached during equilibrium play. A proper subgame starts at a node that forms a singleton <information set>, includes all its descendants, and does not cut any other <information set>; one cannot treat indistinguishable nodes as separate games.

The requirement excludes noncredible threats. Suppose an entrant chooses Out, giving both players $1$, or Enter. After Enter, an incumbent chooses Fight with <payoffs> $(-1,-1)$ or Accommodate with <payoffs> $(2,0)$, with entrant payoff first. Out together with a prescribed Fight is a <Nash equilibrium>: the threat deters entry and the incumbent's off-path action does not change its realized payoff. It is not a <subgame perfect equilibrium>, because after entry the incumbent prefers $0$ to $-1$. The subgame-perfect outcome is Enter followed by Accommodate.

In finite <perfect-information games>, <backward induction in games> constructs such profiles. More generally, continuation subgames may require solving simultaneous-move games in <mixed strategies>, as in an <all-pay contest>. Node-by-node optimization cannot ignore <information sets> in games of imperfect information.