Risk dominance
Risk dominance is an equilibrium selection criterion introduced by John Harsanyi and Reinhard Selten, favoring the equilibrium that is safer to play when a player is uncertain about the opponent's strategy choice. Unlike Pareto dominance, which selects the equilibrium with the highest joint payoff, risk dominance selects the equilibrium that maximizes the product of the players' deviations losses — the equilibrium from which unilateral deviation is most costly. In coordination games, risk-dominant and Pareto-dominant equilibria often conflict, producing a strategic tension between efficiency and security.
Risk dominance has been criticized for lacking axiomatic foundation and for predicting poorly in some experimental settings, yet it remains influential as a formalization of cautious rationality. The criterion's deeper significance lies in what it reveals about bounded rationality: when agents cannot compute or trust others' choices, they default to strategies that minimize worst-case exposure — a logic that extends far beyond games into institutional design and social norm formation.
The Formal Criterion
In a two-player coordination game with two pure-strategy Nash equilibria, risk dominance is computed from the deviation losses — the payoff each player sacrifices by unilaterally deviating from an equilibrium. For equilibria A and B, let a and b be the products of the two players' deviation losses from A and B respectively. Then A is risk-dominant if a > b. This product criterion, introduced by Harsanyi and Selten in their equilibrium selection theory, formalizes the intuition that an equilibrium is "safer" when defecting from it hurts both players more.
The canonical illustration is the Battle of the Sexes. In one version, the players achieve payoffs of (2, 1) at one equilibrium and (1, 2) at the other. Here neither equilibrium is payoff-dominant — neither Pareto-dominates the other. But if the off-diagonal payoffs (the miscoordination outcomes) are symmetric, both equilibria are equally risk-dominant. When the off-diagonal payoffs are asymmetric, the equilibrium with the higher product of deviation losses wins. This reveals that risk dominance is not about the attractiveness of coordination but about the cost of failure.
Crucially, risk dominance and payoff dominance can conflict. In the Stag Hunt, the mutual stag-hunt equilibrium is both payoff-dominant and risk-dominant. But in many coordination games with asymmetric payoffs, the efficient equilibrium (payoff-dominant) is not the safe one (risk-dominant). This tension is not a puzzle to be solved — it is the central strategic fact of social life. Every standard, every norm, every platform choice is a coordination problem where efficiency and security pull in different directions.
Risk Dominance as a Systems Principle
The logic of risk dominance extends far beyond laboratory games. In markets with network effects, the equilibrium that is risk-dominant is often the one with the larger installed base — not because it is better, but because switching away from it is costlier. The standards wars of the 1980s and 1990s (VHS vs. Betamax, QWERTY vs. Dvorak) are textbook cases where risk-dominant outcomes prevailed over payoff-dominant alternatives. Users chose the standard that minimized their exposure to coordination failure, even when a technically superior option existed.
This pattern repeats in social norm formation. Why do inefficient norms persist? Because the cost of being the first deviator is prohibitive. A risk-dominant norm is a path-dependent attractor in the dynamics of social interaction: once established, it is self-enforcing not because anyone believes it is optimal, but because unilateral defection is individually irrational. The norm is the equilibrium from which deviation is most costly — which is precisely the definition of risk dominance.
The connection to bounded rationality is intimate. A fully rational agent with unlimited computational resources and common knowledge of rationality might navigate to the payoff-dominant equilibrium through forward induction or sophisticated belief revision. But real agents lack these resources. They use heuristics. And the heuristic "pick the option where miscoordination hurts least" is exactly risk-dominant reasoning, executed without awareness of game theory.
Limits and Open Questions
Risk dominance has empirical vulnerabilities. In experimental settings, subjects frequently select payoff-dominant equilibria even when risk dominance predicts otherwise — particularly when communication is allowed or when subjects have prior experience with the game. This suggests that focal points, shared history, and communication can override the purely structural logic of deviation losses.
A deeper theoretical problem is that risk dominance lacks a fully satisfactory axiomatic foundation. Unlike Nash equilibrium — which can be derived from rationality and common knowledge — risk dominance is a refinement criterion whose normative justification remains intuitive rather than deductive. Harsanyi and Selten's tracing procedure provides one rationale, but it depends on assumptions about trembles and Bayesian updating that are themselves contested.
The most profound open question is whether risk dominance is a feature of games or a feature of observers. If risk-dominant equilibria are merely the ones that anxious, computationally limited agents converge on, then risk dominance is not a solution concept — it is a psychological prediction. The emergence of risk-dominant outcomes in social systems would then be an instance of emergent order from bounded rationality, not a discovery of game-theoretic structure.
The persistence of inefficient, risk-dominant equilibria in social systems is not a market failure or a coordination failure — it is evidence that security trumps efficiency whenever the cost of miscalculation is borne individually but the gains from efficiency are distributed collectively. Any theory of institutional design that ignores this asymmetry is designing for agents who do not exist.