Game Theory · GT-20
When you don't know your opponent's true type, you play against your best probability-weighted guess of who they are.
A game where players have private information (their 'type' — true costs, true preferences, true strength) unknown to others, who instead hold probability distributions (beliefs) over what type they're facing. Players choose strategies that are best responses given their beliefs, and the resulting equilibrium (Bayesian Nash equilibrium) accounts for this uncertainty directly.
Formalized by John Harsanyi in a series of 1967-68 papers that resolved a long-standing technical problem in modeling incomplete information — work that contributed to his shared 1994 Nobel Memorial Prize alongside Nash and Selten.
The Mechanism
Bidding against an unknown-type opponent — beliefs shape the best response
Your optimal bid shifts depending on what you believe about the opponent's hidden type — the same private valuation calls for a different action depending on the probability distribution you hold over who you're actually playing against.
01 · HARSANYI'S TRICK: TURN HIDDEN INFORMATION INTO A MOVE BY 'NATURE'
The key modeling insight
Harsanyi's breakthrough was reframing incomplete information as a game where an initial random move by 'Nature' assigns each player a private type, known only to them, with the assignment probabilities common knowledge — this converts an intractable-seeming problem (I don't know what you know) into a standard game with a well-defined equilibrium concept.
02 · BAYESIAN NASH EQUILIBRIUM: BEST RESPONSE GIVEN BELIEFS, NOT CERTAINTY
A strategy is a full plan, one action per possible type
In a Bayesian game, a player's 'strategy' actually specifies an action for every type they might privately be — the equilibrium condition requires each type's chosen action to be optimal in expectation over the distribution of the other player's possible types, not against one assumed certainty.
03 · IT UNDERPINS MODERN AUCTION AND MECHANISM DESIGN
The mathematical backbone of a huge applied literature
Bayesian games are the formal foundation for auction theory, mechanism design, and most modern information economics — anywhere private valuations or private costs matter (spectrum auctions, procurement bidding, insurance markets), the analysis runs through this Harsanyi framework.
Where It Fails / Inversion
Where it fails / inversion
The framework assumes players hold a well-specified probability distribution over opponent types and that this distribution itself is common knowledge — in reality, people often have poorly-calibrated, badly biased, or simply wrong beliefs about their counterparts, and small errors in the assumed distribution can produce large errors in the predicted equilibrium. The math is only as good as the beliefs fed into it.
How To Use It
Worked example · pricing when you don't know if a customer is price-sensitive
A salesperson negotiating with a lead of unknown price-sensitivity is playing a Bayesian game — their optimal opening offer depends on their belief about the probability distribution of customer 'types' (price-sensitive vs. value-focused), updated by any observable signals (company size, urgency of the inquiry, competing-vendor mentions). A well-calibrated sales process explicitly tracks and updates these type-probabilities rather than assuming one fixed customer type for every negotiation.
How to use it
When facing a counterpart whose true preferences, costs, or constraints you can't observe directly, don't guess a single assumed type and optimize against it — explicitly hold a probability distribution over their plausible types, update it as new signals arrive, and choose the action that's best in expectation across that whole distribution, not just against your best single guess.
See Also