Nash Equilibrium: Stable Strategies In Games

Build a working sense of Nash equilibrium by spotting best responses, handling mixed strategies, and recognizing when equilibrium is stable but undesirable. You will learn to read payoff tables, predict coordination failures, and know when refinements like subgame perfection matter.

Stable outcomes in games are not always fair, efficient, or even peaceful. A Nash equilibrium can lock players into a pattern simply because neither side can improve by changing their own move alone, even if both would prefer a different world. That no-one-can-blink property is why Nash shows up everywhere from pricing to politics to machine learning. The surprising part is that stability does not require trust, communication, or good intentions. It only requires that every alternative unilateral move looks worse to the mover.

Nash equilibrium means no profitable unilateral deviation

Here is the core idea, stated as a check you can actually run.

A strategy profile is a Nash equilibrium if, holding everyone else’s strategies fixed, no single player can increase their payoff by switching to a different strategy. Equivalently, each player’s strategy is a best response to the other players’ strategies.

This definition creates a sharp mental model. The equilibrium is stable against one person changing their mind, not necessarily against two people coordinating a change.

Take a moment to ground that in a concrete payoff table.

Sometimes the equilibrium cell is not the one you would pick if you could commit together. That is not a paradox. Nash is a statement about incentives given the other player’s choice, not a promise about good outcomes.

Stability, not goodness
Equilibrium answers will anyone regret their move given what happened, not did we get the best joint result.

Best responses turn payoff tables into equilibria

In small normal form games, you rarely find Nash by guessing the equilibrium directly. You find best responses, then look for where they line up.

The mechanic you want

For each possible action of the other player, ask what action maximizes your payoff. That mapping is your best response. A Nash equilibrium is a mutual best response, a cell where both players are playing a best response at the same time.

That sounds abstract until you do it a few times.

A couple practical notes that prevent common mistakes:

• If a player has two best responses to the same opponent action, that creates multiple candidate equilibria.
• A best response depends on what the other player does, not on what the other player prefers.
• If you see a strategy that is always worse than another, it will never be a best response and you can often delete it safely.

Once you can read best responses, Nash stops feeling like a magic label and starts feeling like bookkeeping on incentives.

Mixed strategies when no pure move can sit still

Some games have no pure strategy equilibrium, meaning every pure action can be exploited by a unilateral switch. The escape hatch is to randomize.

A mixed strategy assigns probabilities to actions. The point is not to be unpredictable for its own sake. The point is to pick probabilities that make the opponent indifferent among the actions you want them to mix between.

You can see the logic in one line. If Player A mixes between actions $a_1$ and $a_2$, then Player B’s mix must make A get equal expected payoff from $a_1$ and $a_2$. If one action paid strictly more in expectation, A would stop randomizing and switch.

Explore how the indifference point pins down the equilibrium mixture.

Indifference test
In a mixed equilibrium, every action used with positive probability must yield the same expected payoff.

This also clarifies a misconception. Randomizing is not a sign the model is broken. It is the stable choice when any predictable pure choice invites a profitable deviation.

Multiple equilibria and the selection problem

Many interesting games have more than one Nash equilibrium. Then Nash tells you what could persist, not what will be chosen.

Two forces often pull in different directions:

Efficiency versus safety

One equilibrium may be payoff-dominant, giving everyone more, but risky because it requires coordinated confidence. Another may be risk-dominant, meaning it is safer when you are unsure what the other player will do.

Seeing both side by side makes the tension obvious.

Selection then becomes about context and beliefs, not just incentives inside the table. People use conventions, history, and salient focal points to coordinate on one equilibrium rather than another. If you are modeling a real scenario, you often need to justify why one equilibrium is likely, not just list all of them.

Larger settings need the right game form

Payoff tables are the normal form view. They list strategies and payoffs as if choices happen simultaneously. Many strategic situations are sequential, with moves observed along the way. That is the extensive form view, where strategies are complete contingent plans.

A normal form Nash equilibrium can rely on threats that would never be carried out if the moment arrived. Sequential settings therefore push you toward a refinement called subgame perfect equilibrium (SPE), which rules out equilibria sustained by non-credible future actions.

A quick way to keep your modeling honest:

• Use normal form when timing and observability do not matter or can be ignored safely.
• Use extensive form when someone moves after observing something and that observation changes incentives.
• Reach for SPE when backward reasoning through future decision points is essential.

The meta-lesson is that equilibrium is not just about solving. It is about choosing a representation that matches what players know and when they know it.

Common failure modes when applying Nash

The fastest way to misuse Nash is to treat any equilibrium as an endorsement. Another is to compute equilibria without simplifying the game first.

Work through a few high-frequency gotchas.

Dominated strategies are the low-hanging fruit. If an action is strictly dominated, a rational player never uses it in equilibrium, and deleting it can reveal the real strategic core. Non-credible threats are the next trap, especially in sequential settings, which is why SPE exists. Finally, equilibrium and welfare are different questions. A stable outcome can be socially bad, and fixing it often means changing payoffs through rules, incentives, or commitment devices.

Nash in statistics and probability

Nash equilibrium has a mathematical feel because it is, at heart, a fixed point idea. Each player’s best-response rule maps beliefs about others into an optimal strategy. An equilibrium is where everyone’s strategy is a fixed point of that mapping.

This connects to learning dynamics. In repeated interaction, players may update beliefs from data, adjust strategies, and sometimes converge toward equilibrium or cycle around it depending on the game and the learning rule.

If you have a scenario in mind, map it into the ingredients of a game and see what equilibrium would even mean there.

A useful modeling trick from probability is to be explicit about what is random. Are players mixing intentionally, or are you uncertain about types, payoffs, or information? Those are different models that can produce similar-looking behavior.

Next-step intuition for using Nash well

Trust Nash when incentives are stable, players are roughly rational, and the rules of interaction are common knowledge. Be cautious when payoffs are misspecified, when communication or commitment changes the game, or when selection among multiple equilibria is the main story.

When you need more precision, refinements are your tools. SPE handles sequential credibility. Other refinements handle beliefs off the equilibrium path or eliminate implausible equilibria in richer games.

The most practical next step is to practice writing payoffs. If you can state what each player values, what they can do, what they observe, and what happens after each move, you will usually know whether Nash is the right lens or whether the model needs timing, uncertainty, or a mechanism change.