Understanding Mixed Strategy
Some games have no stable deterministic solution. In matching pennies, any fixed choice can be anticipated and beaten, so no pair of pure strategies is stable. Mixed strategies extend the space of options by permitting deliberate randomization, which restores a stable solution.
The critical property is unpredictability. If a player’s choice can be forecast, an opponent can prepare specifically against it. Randomizing according to fixed probabilities makes the specific choice unforecastable while keeping the long-run distribution deliberate. Russell and Norvig note that a solution existing only in mixed strategies does not require a player to be literally randomizing to be rational; mixed strategies are theoretical constructs for analyzing the game.
Equilibrium mixing probabilities have a characteristic and initially surprising structure: each player mixes so as to make the other indifferent between their options. If an opponent strictly preferred one response, they would play it exclusively, and the first player could then exploit that predictability. Indifference is exactly the condition under which no exploitable pattern exists.
This is the extension that makes Nash’s existence theorem work. Restricted to pure strategies, many finite games have no equilibrium at all. Allowing randomization guarantees at least one always exists, which is what makes equilibrium analysis applicable across the whole class of finite games rather than a fortunate subset.
Example of Mixed Strategy
In matching pennies, suppose one player showed heads 70% of the time. The opponent would then choose the response that wins against heads every time, and would profit on balance. Any departure from even mixing is exploitable in this way.
Mixing exactly half and half removes the exploit. Every response the opponent can choose yields the same expected payoff against a fair coin, so no counter-strategy is better than any other, and neither player has an incentive to change.
The same reasoning explains why competitors in genuinely adversarial settings avoid predictable patterns. Any regularity in behaviour is information the opponent can act on, and randomization is precisely the removal of that information.
Frequently Asked Questions
Why would randomizing ever be rational?
Because predictability is exploitable in adversarial settings. If any fixed choice can be countered, the best available option is to be unforecastable, and randomizing with the right probabilities is what achieves that.
Why do equilibrium probabilities make the opponent indifferent?
Because if the opponent strictly preferred one response they would always play it, and that predictability could then be exploited. Only when they are indifferent is there no pattern for the first player to take advantage of, which is what makes the situation stable.
Do real players actually randomize?
Not usually in a literal, calculated way. Mixed equilibria are best read as a description of aggregate behaviour or of the beliefs opponents hold, rather than as an instruction to consult a random number generator. The analytical content lies in the probabilities, not the mechanism.
The Bottom Line
A mixed strategy randomizes over actions, and admitting such strategies is what guarantees every finite game has a Nash equilibrium. Equilibrium mixing leaves opponents indifferent, which is precisely the condition under which no exploitable pattern remains.