Set out by John von Neumann and Oskar Morgenstern (Theory of Games and Economic Behavior, 1944), proved for every finite game of this kind by Harold W. Kuhn (1953), and refined by Reinhard Selten (1965).
Solves a sequence of moves by starting at the final choices and working back, picking each player's best move, to find the path rational players will take.
A three-move supply negotiation solved from the last move back
Each end shows the buyer's and the supplier's yearly profit. Start at the right: at each last move the mover takes the better payoff for itself, and that result rolls back to the move before, until the buyer's first offer is decided. Then raise the buyer's payoff for walking away after the $11 counter to $45K, as a second qualified supplier would, and watch the supplier's best reply change. Illustrative.
Payoffs read Buyer / Supplier.
Offer $10 a unit, then Accept $10
Offer $10 a unit, then Counter at $11, then Accept $11
Offer $10 a unit, then Counter at $11, then Walk away
Offer $12 a unit, then Accept $12
Offer $12 a unit, then Counter at $13, then Accept $13
Offer $12 a unit, then Counter at $13, then Walk away
An example
A packaging buyer at a food company was preparing to renegotiate a supply contract for molded trays. Her plan was to open at $12 a unit, a price the supplier would surely accept, and avoid a fight.
Her manager asked her to draw the negotiation as a sequence: she offers, the supplier accepts or counters, and she accepts the counter or walks away. They put rough yearly profits for both sides at every end and solved it from the last move. At the end, she would accept an $11 counter, because her only fallback supplier was slow and expensive. Knowing that, the supplier would counter any $10 offer at $11. A $12 opening would simply be accepted, and it left her $20,000 less than opening at $10. The drawing also showed where her real leverage lay. If she qualified a second supplier first, walking away would beat $11, and the supplier's best reply to $10 would be to accept. She spent six weeks qualifying a second source, opened at $10, and signed at $10.
Why it works
In a sequence of moves, what someone should do now depends on what everyone will do later, and later choices are easier to predict because fewer options remain. Starting at the end, where the mover simply takes the better payoff, gives a firm answer; replacing that decision with its result makes the move before it an end in turn. Repeating the step back to the first move produces the path that players who look ahead will actually follow.
The method also exposes empty threats. A promise to walk away counts only if walking away is the better choice at the moment it has to be carried out, and backward induction checks exactly that. It shows where changing a payoff, by finding an alternative or signing a commitment, changes the whole game.
The thinking behind it
Oskar Morgenstern Oskar Morgenstern, a German-born economist, founded game theory and its use in economics with John von Neumann in Theory of Games and Economic Behavior (Princeton University Press, 1944). Their book set out expected utility theory, the analysis of games of strategy, and the solution of games of perfect information by working back from the final move, the procedure now called backward induction.
Harold W. Kuhn Harold W. Kuhn, a mathematician who shaped game theory and optimization, gave the general form of games played in sequence, the extensive form with information sets, in Extensive Games and the Problem of Information (1953, in Contributions to the Theory of Games II). In the same paper he proved that every finite game of perfect information has an equilibrium in pure strategies, which backward induction finds.
Reinhard Selten Reinhard Selten, a German game theorist, proposed subgame perfect equilibrium in a 1965 paper on an oligopoly model, in the Zeitschrift für die gesamte Staatswissenschaft. The idea rules out threats that would not be credible at the moment they had to be carried out, and in games played in sequence it is found by backward induction. He shared the 1994 Nobel Memorial Prize in Economic Sciences with John Harsanyi and John Nash.
How to run it
- Draw the moves as a tree. Show who moves at each point and every option open to them, in order.
- Put payoffs at every end. Estimate each side's result at each end point, in the same units for everyone.
- Solve the last moves first. At each final decision, pick the option that is best for the player making it.
- Replace each solved move with its result. Carry the chosen payoffs back, so the move before becomes an end in its turn.
- Repeat back to the first move. The chosen branches from the start form the path the tool predicts.
- Test which payoffs change the answer. Vary the uncertain numbers, especially each side's fallback, and see where the predicted path flips.
- Act on the payoffs you can change. Look for moves before the game starts, such as a second supplier, that make a threat credible.
Where it goes wrong
The payoffs are guesses dressed as facts. A precise tree built on invented numbers gives a confident wrong answer. Find which payoffs the answer depends on, and spend the effort estimating those.
The other side is assumed to think as you do. Real counterparts misjudge, act on pride, or care about the relationship, which the payoffs may leave out. Put what they care about into their payoffs, or test how far off they would have to be to change the path.
The game is longer than the tree. A one-time model of a relationship that repeats every year misses reputation and future deals. When the parties will meet again, model the repetition or treat the answer as a lower bound.
When to use something else
- Nash Equilibrium. When the moves are simultaneous and nobody sees the other's choice before making theirs.
- Repeated Games. When the same parties play the same game again and again, so reputation and retaliation matter.
- BATNA and ZOPA. When the negotiation turns on one price, and the question is the range each side can accept.
Where it sits in SOLVE eX
Go to the source
- Theory of Games and Economic Behavior, section 15, 1944
- Extensive Games and the Problem of Information, Contributions to the Theory of Games II, 1953
- Spieltheoretische Behandlung eines Oligopolmodells mit Nachfrageträgheit, Zeitschrift für die gesamte Staatswissenschaft, 1965