📌 “Incomplete information games require us to model not just actions, but also beliefs about others.” Bayesian Nash Equilibrium (BNE) and Perfect Bayesian Equilibrium (PBE) are two fundamental solution concepts for such games. This article explains their differences, when to use each, and common pitfalls.
Game theory studies strategic interactions where outcomes depend on the actions of multiple players. When players have incomplete information (i.e., they don't know everything about others), we need special solution concepts. The two most important ones are Bayesian Nash Equilibrium (BNE) and Perfect Bayesian Equilibrium (PBE). BNE handles static games with incomplete information, while PBE extends this to dynamic games (with sequences of moves) and adds requirements about beliefs.
What is Bayesian Nash Equilibrium (BNE)?
A Bayesian Nash Equilibrium is a strategy profile where each player's strategy is optimal given their beliefs about other players' types and the strategies those types are playing. It's used for static games (all players move simultaneously or without observing others' moves).
Two bidders compete for an item. Each bidder privately knows their own valuation (their "type")—either High ($100) or Low ($50)—with equal probability. They submit sealed bids simultaneously.
- Player A's strategy: "If my valuation is High, bid $80; if Low, bid $40."
- Player B's strategy: "If my valuation is High, bid $75; if Low, bid $35."
A firm (Entrant) decides whether to enter a market dominated by an Incumbent. The Incumbent's cost type is private: it can be Low-cost (aggressive) or High-cost (accommodating). The Entrant only knows the probabilities (60% Low-cost, 40% High-cost).
- Entrant's strategy: "Enter if my expected profit > 0, given the Incumbent's likely response based on its cost type."
- Incumbent's strategy: "If I am Low-cost, fight entry; if High-cost, accommodate."
What is Perfect Bayesian Equilibrium (PBE)?
Perfect Bayesian Equilibrium is a refinement for dynamic games with incomplete information (sequential moves). It requires:
- Sequential Rationality: Players choose optimal actions at every decision point, given their beliefs.
- Belief Consistency: Beliefs are updated using Bayes' rule wherever possible (i.e., when reaching a node with positive probability).
PBE adds belief systems to strategies, ensuring rationality throughout the game tree.
A worker's ability is private: High or Low. The worker first chooses an education level (High or Low). The firm then observes the education (but not ability) and decides to offer a High or Low wage.
- Worker's strategy: "If I am High ability, get High education; if Low ability, get Low education."
- Firm's strategy: "If I see High education, believe the worker is High ability with probability 1 and offer High wage; if Low education, believe Low ability and offer Low wage."
- Beliefs: The firm's beliefs are consistent with Bayes' rule given the worker's strategy.
Two players: You (with a weak hand) and Opponent. You can Bet or Fold. If you Bet, the Opponent (with a strong or medium hand) can Call or Fold.
- Your strategy: "Sometimes bluff (Bet with weak hand) to keep opponent guessing."
- Opponent's strategy: "If you Bet, I believe you might be strong or bluffing; I Call with probability based on my hand strength."
- Beliefs: Opponent updates beliefs about your hand strength after seeing your Bet, using Bayes' rule and your bluffing frequency.
Key Differences Between BNE and PBE
| Aspect | Bayesian Nash Equilibrium (BNE) | Perfect Bayesian Equilibrium (PBE) |
|---|---|---|
| Game Type | Static games (simultaneous moves) | Dynamic games (sequential moves) |
| Beliefs | Initial beliefs only (no updating) | Beliefs updated using Bayes' rule |
| Sequential Rationality | Not required (only ex-ante optimality) | Required at every decision point |
| Off-Path Beliefs | Not defined (no off-path moves) | Must specify beliefs even for zero-probability events |
| Common Applications | Auctions, static oligopoly, insurance markets | Signaling, bargaining, repeated games with incomplete info |
⚠️ Common Pitfalls & Misconceptions
- Using BNE for sequential games: BNE only checks optimality at the start. In dynamic games, it may allow irrational moves later. Always use PBE for games with observed actions.
- Ignoring belief consistency in PBE: PBE requires beliefs to be updated via Bayes' rule whenever possible. Arbitrary beliefs can support nonsensical equilibria.
- Confusing "perfect" with "complete information": Perfect Bayesian Equilibrium deals with incomplete information. "Perfect" refers to sequential rationality (like subgame perfection), not knowledge.
- Forgetting off-path beliefs: In PBE, you must specify what players believe after unexpected moves. These beliefs can affect equilibrium refinement.
When to Use BNE vs. PBE
Choose the solution concept based on the game structure:
- Use Bayesian Nash Equilibrium (BNE) when:
- Players move simultaneously (or without observing others' actions).
- You only care about ex-ante optimal strategies given initial beliefs.
- Examples: Sealed-bid auctions, static price competition with unknown costs.
- Use Perfect Bayesian Equilibrium (PBE) when:
- The game has a sequence of moves where players observe earlier actions.
- You need to model how beliefs evolve during the game.
- Examples: Signaling games (education, product quality), bargaining with offers and counteroffers, poker.
In short: BNE for static incomplete information, PBE for dynamic incomplete information.
Conclusion
Bayesian Nash Equilibrium and Perfect Bayesian Equilibrium are both essential for analyzing games with incomplete information. BNE provides a foundation for static settings, ensuring players choose optimal strategies given their initial beliefs. PBE builds on this by adding sequential rationality and belief consistency for dynamic games, preventing irrational moves later in the game. Understanding the difference helps you apply the right tool: use BNE for auctions and simultaneous moves, and PBE for signaling, bargaining, and any game where actions are observed over time.