Game Theory and Strategic Behavior

The Basic Setup

Game theory formalizes strategic interaction: situations where your optimal choice depends on what others choose, and they know that you know this, and you know that they know that you know this, and so on. The foundational concepts were developed by John von Neumann and Oskar Morgenstern in Theory of Games and Economic Behavior (1944), and the central equilibrium concept by John Nash in 1950.

A game requires: players (who are involved), strategies (what each player can choose), and payoffs (what each player gets for each combination of choices). The analysis asks: what will rational players do? And what outcomes will result from those choices?

The deceptively simple definition of a Nash equilibrium: a set of strategies, one for each player, such that no player can improve their payoff by unilaterally changing their strategy, given the strategies of all other players. It is a stable resting point — no one has a reason to deviate. Finding Nash equilibria is the core task of game theoretic analysis.

The Prisoner’s Dilemma

The prisoner’s dilemma is the single most discussed example in social science, and for good reason: it is a clear, simple model of why individually rational behavior produces collectively suboptimal outcomes.

Two suspects are held separately and interrogated. Each can either cooperate (stay silent) or defect (betray the other). If both cooperate, they each get a light sentence (say, 1 year). If both defect, they each get a moderate sentence (3 years). If one defects and the other cooperates, the defector goes free and the cooperator gets the maximum sentence (10 years).

For each prisoner, defecting is the dominant strategy — it produces a better outcome regardless of what the other does. If the other cooperates, defecting goes free versus 1 year. If the other defects, defecting gets 3 years versus 10 years. Dominance means: no matter what the other person does, you’re better off defecting.

Both prisoners reason this way. Both defect. Both get 3 years. Mutual cooperation, which would give each only 1 year, is not a Nash equilibrium because either prisoner would gain by defecting if the other cooperated. The individually rational outcome is collectively worse than the available alternative.

The prisoner’s dilemma structure appears everywhere: arms races (both sides building weapons that cancel each other out, at cost to both), price competition (both firms cutting prices until neither profits), climate change (each country emitting because abatement costs fall on them while the climate benefits are shared), commons overuse (each farmer grazing as many sheep as possible until the common is exhausted). The dilemma is pervasive because the underlying structure — individual incentive to free-ride on others’ cooperation — is pervasive.

Repeated Games and the Shadow of the Future

The one-shot prisoner’s dilemma has a clear solution: defect. The repeated prisoner’s dilemma — the same game played many times between the same players — has a radically different solution space.

Robert Axelrod’s tournaments in the early 1980s invited game theorists to submit strategies for a repeated prisoner’s dilemma. The winning strategy, submitted by Anatol Rapoport, was Tit for Tat: cooperate on the first move, then do whatever the other player did on the previous move. Tit for Tat is simple, cooperative (it starts by cooperating and reciprocates cooperation), retaliatory (it immediately punishes defection), and forgiving (it returns to cooperation as soon as the other player cooperates).

The formal result: in a repeated game where the probability of another interaction is sufficiently high (the “shadow of the future” is long enough), cooperation can be a Nash equilibrium. The logic: the long-run gain from sustained mutual cooperation outweighs the short-run gain from defecting. If the other player is playing Tit for Tat, defecting once produces a gain this round but triggers retaliation in all future rounds. The net present value of defecting can be negative if future rounds matter enough.

This is why stable cooperation tends to emerge between parties who interact repeatedly and can observe each other’s behavior. It’s also why anonymous, one-shot interactions (like anonymous online transactions, or interactions between strangers who will never meet again) tend toward defection — the shadow of the future is absent.

Coordination Games and Multiple Equilibria

Not all games have the prisoner’s dilemma structure. Coordination games have multiple equilibria, and the problem is selecting among them.

The most famous: the driving side coordination game. Everyone should drive on the same side of the road. Whether that’s left or right doesn’t matter; what matters is coordination. Both “everyone drives left” and “everyone drives right” are Nash equilibria — no one wants to unilaterally change. The selection between equilibria depends on history, convention, and sometimes explicit coordination mechanisms (laws).

Network effects in technology create coordination games. The value of a communications platform increases with the number of users. Multiple incompatible platforms might exist as Nash equilibria — users on each platform have no reason to unilaterally switch to the other. Historical accidents, early market share, and coordination by large adopters determine which equilibrium is selected.

Language is a coordination equilibrium. English is useful to learn because many people speak it; many people speak it because it’s useful to learn. This is a self-fulfilling equilibrium that persists not because English is intrinsically superior to alternatives but because it was the language of coordination in global commerce and science.

Mechanism Design — Engineering Incentives

Mechanism design is game theory applied to institutional engineering: given a desired outcome, what rules should govern the game to make that outcome an equilibrium? It’s sometimes called “reverse game theory” — instead of analyzing a given game, you design the game to achieve a target.

The auction literature is the canonical application. If you’re selling a good (spectrum licenses, government contracts, oil leases) and want to maximize revenue or allocate to the highest-value user, what auction format achieves this? The Vickrey auction (second-price sealed bid: submit your value, highest bidder wins but pays the second-highest bid) has the remarkable property that truthful bidding is a dominant strategy — you maximize your payoff by bidding exactly your true value, regardless of what others do. No strategic distortion. The mechanism makes truthful behavior the equilibrium.

FCC spectrum auctions, Google’s ad auction, and market designs for school assignment (matching students to schools using variants of the Gale-Shapley algorithm) are examples of applied mechanism design producing real institutions.

The limits of mechanism design: it assumes that the designer knows the players’ utility functions well enough to specify the mechanism, and that the players will play the game as designed and not find ways to game the mechanism in unanticipated directions. Both assumptions fail in practice. Real institutions are also embedded in social contexts where players have reputations and relationships that affect behavior beyond what the game theoretic analysis captures.

When Rationality Fails as a Predictor

The descriptive power of game theory depends on the rationality assumption. Players who are irrational, miscommunicate, have mistaken beliefs, or are motivated by factors the payoff matrix doesn’t capture (fairness, spite, honor) will not play Nash equilibria.

Behavioral game theory — developed by Kahneman, Thaler, Camerer, and others — documents systematic deviations from game-theoretic predictions. In ultimatum games, one player proposes a split of a sum and the other accepts or rejects (rejection means both get nothing). Game theory predicts the proposer will offer the minimum positive amount and the responder will accept any positive offer (something is better than nothing). In experiments across many cultures, proposers typically offer 40-50% and responders reject offers below 30% — even though rejection is costly to the responder.

The explanation is fairness norms: responders punish unfair offers at personal cost, and proposers anticipate this and offer fair amounts. This is not irrational in any deep sense — it reflects evolved social preferences for equity and punishment of defectors that make sense in repeated social contexts. But it means game theory’s purely self-interested rationality systematically fails to predict behavior in social contexts where norms of fairness and reciprocity are operative.

The synthesis: game theory is most predictive in contexts that closely resemble its assumptions — anonymous, one-shot, high-stakes interactions with clear payoffs and sophisticated players. It is less predictive in social contexts with strong norms, repeated interaction, and payoffs that include social and emotional dimensions. Knowing when to apply the model and when to adjust it is the practitioner’s skill.