Lingo

Backward Induction

Solving sequential games by reasoning from the last move backward to the first.

What is it?

Backward induction is a method for solving sequential games—games where players move in a known order. You start at the last decision node, determine the best choice there, then work backward, eliminating earlier options that lead to suboptimal play. It produces a subgame-perfect equilibrium: a strategy that is optimal at every point of the game, not just at the start.

How it works

Start at the end: Identify the final move and write down what the last mover would rationally choose.

Roll back: Move to the previous decision and choose the option that produces the best payoff given the known future play—repeating until you reach the first move.

Credibility matters: Backward induction only works if future threats and promises are credible —i.e., if the player would actually follow through when the time comes. Non-credible threats are discarded.

The centipede game

The setup: Two players alternately take a growing pot or “pass” it onward; the pot grows each round. Passing eventually pays more for both, but stopping early pays the current player more immediately.

The prediction: Backward induction predicts the first player stops immediately—because the second player would rationally stop on their very first turn for a slightly larger immediate payoff.

The paradox: Real people usually pass for several rounds, cooperating far longer than theory predicts. This gap between backward induction and actual play exposes the limits of perfect-rationality assumptions.

Why it matters

Strategic credibility: Backward induction filters out empty threats, telling rational players which commitments are believable and which will be abandoned.

Chess and games: Chess engines evaluate positions by rolling forward many moves, but humans use backward reasoning to identify the consequences of a proposed move.

Business and negotiation: When deciding whether to enter a market or make an offer, anticipating how rivals and partners will respond down the line—and working backward from there—yields sound strategy.

Limits: With many players and long horizons, backward induction is computationally heavy, and real behavior often deviates, as the centipede game demonstrates.

References

  1. Wikipedia: Backward Induction— Overview of backward induction as a solution concept for sequential games.
  2. Wikipedia: Centipede Game— Description of the centipede game and its role in testing backward induction.
  3. Wikipedia: Subgame Perfect Equilibrium— Formal definition of subgame-perfect equilibrium, the concept backward induction produces.

Examples