Bayesian Games: Playing With Incomplete Information

How game theory handles situations where players don't know important facts about each other, only probabilities about them.

Most of the games covered earlier in this module assume every player knows exactly what payoffs everyone else is working with. Real strategic situations rarely offer that certainty - you often don’t know whether a rival is bluffing, whether a buyer genuinely values what you’re selling, or whether a negotiating counterpart is under real time pressure. Bayesian games are the branch of game theory built specifically to handle this kind of uncertainty.

What “incomplete information” means here

Incomplete information describes a game where at least one player doesn’t know some relevant characteristic of another player - their preferences, their available strategies, or the payoffs they’d receive from different outcomes. This is different from the imperfect information covered in the sequential games lesson, where players simply haven’t yet observed a move that already happened; here, a player may never fully learn some underlying fact about their opponent, even after the game concludes.

Types: a way of representing what you don’t know

Bayesian games handle this uncertainty by assigning each player a type - a specific version of that player, capturing the private information only they know, such as how much they genuinely value a good or how aggressive their true strategy is. Other players don’t know for certain which type they’re facing, but they do have a probability distribution over the possible types, usually built from experience, market data, or reasonable assumptions. A player’s optimal strategy in a Bayesian game has to work well against the whole range of types the opponent might be, weighted by how likely each type is.

Bidding without knowing your rival's budget

Picture two companies bidding for the same contract. Each knows its own maximum willingness to pay, but not its rival's - it only knows that, based on past bidding patterns, the rival is probably a "high-value" type most of the time and a "low-value" type occasionally. Each company has to choose a bidding strategy that performs reasonably well against both possibilities, weighted by how likely each one is, rather than a strategy tailored to one assumed opponent.

Bayesian Nash equilibrium

Just as the earlier Nash equilibrium lesson describes a stable outcome where no player wants to unilaterally change their strategy, a Bayesian Nash equilibrium is the equivalent concept for games with incomplete information: each type of each player is choosing the strategy that maximizes their expected payoff, given their beliefs about the probabilities of other players’ types and given that every other player is doing the same. It’s a more demanding concept than an ordinary Nash equilibrium, since it has to hold across every possible type simultaneously, not just for one fixed set of players.

Assuming uncertainty makes strategic analysis impossible

It might seem like not knowing your opponent's true preferences or payoffs would make careful strategic reasoning pointless - how can you calculate a best response to something you can't observe? Bayesian games show this isn't true: as long as you can form a reasonable probability distribution over the possibilities, you can still calculate an optimal strategy that performs well on average across that distribution, even without ever learning the specific truth.

Why this framework matters

Bayesian games underpin much of modern auction theory, connecting directly to the earlier lesson on the economics of bidding, since bidders essentially never know competitors’ exact valuations. They’re also central to analyzing negotiations, insurance markets where insurers don’t know a customer’s true risk level, and any strategic situation where a crucial piece of information is genuinely private to one side.

Key takeaways

Bayesian games model situations where players lack complete information about each other's true payoffs or preferences.
Each player has a "type" representing their private information, and others hold probabilities over possible types.
Optimal strategies must perform well across the whole range of an opponent's possible types, not just one guess.
A Bayesian Nash equilibrium requires every type of every player to be playing an expected-payoff-maximizing strategy.
This framework underlies auction theory, negotiation analysis, and markets like insurance with hidden information.