완벽한 베이시안 평형
Perfect Bayesian equilibrium퍼펙트 베이시안 평형 | |
---|---|
게임 이론의 솔루션 개념 | |
관계 | |
부분 집합 | 베이시안 나시 평형 |
의의 | |
제안자 | 조와 크렙스[citation needed] |
에 사용됨 | 다이나믹 베이시안 게임 |
예 | 신호 게임 |
게임 이론에서 PBE(Perfect Bayesian Balance, PBE)는 불완전한 정보를 가진 동적 게임(시퀀셜 베이시안 게임)과 관련된 평형 개념이다. 베이시안 나시 평형(BNE)의 정교함이다. 완벽한 베이시안 평형에는 전략과 신념이라는 두 가지 요소가 있다.
- 주어진 정보 집합에서 플레이어의 전략은 해당 정보 집합에서 자신의 행동 선택을 명시하며, 이는 (이전에 게임에서 취한 행동에 따라) 이력에 따라 달라질 수 있다. 이것은 순차 게임과 비슷하다.
- 주어진 정보 세트에서 플레이어의 믿음이 게임이 도달했다고 믿는 정보 세트에서 어떤 노드에 도달했는지를 결정한다. 그 믿음은 정보 집합의 노드에 대한 확률 분포일 수 있으며, 일반적으로 다른 플레이어의 가능한 유형에 대한 확률 분포일 수 있다. 형식적으로, 믿음 시스템은 어떤 정보 세트에서든 확률의 합이 1이 되도록 게임의 모든 노드에 확률을 할당하는 것이다.
전략과 신념은 다음과 같은 조건을 만족시켜야 한다.
- 순차적 합리성: 각 전략은 신념에 따라 예상에 최적이어야 한다.
- 일관성: 각 믿음은 평형 전략, 관찰된 행동 및 양의 확률로 평형에 도달한 모든 경로에 대한 베이지스의 규칙에 따라 업데이트되어야 한다. 비균형 경로로 알려진 제로 확률의 경로에서, 신념은 명시되어야 하지만 임의적일 수 있다.
완벽한 베이시안 평형은 항상 내쉬 평형이다.
Examples of perfect Bayesian equilibria
Gift game 1
Consider the following game:
- The sender has two possible types: either a "friend" (with probability ) or an "enemy" (with probability ). Each type has two strategies: either give a gift, or not give.
- The receiver has only one type, and two strategies: either accept the gift, or reject it.
- The sender's utility is 1 if his gift is accepted, -1 if his gift is rejected, and 0 if he does not give any gift.
- The receiver's utility depends on who gives the gift:
- If the sender is a friend, then the receiver's utility is 1 (if he accepts) or 0 (if he rejects).
- If the sender is an enemy, then the receiver's utility is -1 (if he accepts) or 0 (if he rejects).
For any value of Equilibrium 1 exists, a pooling equilibrium in which both types of sender choose the same action:
- Equilibrium 1. Sender: Not give, whether of the friend type or the enemy type. Receiver: Do not accept, with the beliefs that Prob(Friend Not Give) = p and Prob(Friend Give) = x, choosing a value
The sender prefers the payoff of 0 from not giving to the payoff of -1 from sending and not being accepted. Thus, Give has zero probability in equilibrium and Bayes's Rule does not restrict the belief Prob(Friend Give) at all. That belief must be pessimistic enough that the receiver prefers the payoff of 0 from rejecting a gift to the expected payoff of from accepting, so the requirement that the receiver's strategy maximize his expected payoff given his beliefs necessitates that Prob(Friend Give) On the other hand, Prob(Friend Not give) = p is required by Bayes's Rule, since both types take that action and it is uninformative about the sender's type.
If , a second pooling equilibrium exists as well as Equilibrium 1, based on different beliefs:
- Equilibrium 2. Sender: Do not give, whether of the friend type or the enemy type. Receiver: Accept, with the beliefs that Prob(Friend Give) = p and Prob(Friend Not give) = x, choosing any value for
The sender prefers the payoff of 1 from giving to the payoff of 0 from not giving, expecting that his gift will be accepted. In equilibrium, Bayes's Rule requires the receiver to have the belief Prob(Friend Give) = p, since both types take that action and it is uninformative about the sender's type in this equilibrium. The out-of-equilibrium belief does not matter, since the sender would not want to deviate to Not give no matter what response the receiver would have.
Equilibrium 1 is perverse if The game could have so the sender is very likely a friend, but the receiver still would refuse any gift because he thinks enemies are much more likely than friends to give gifts. This shows how pessimistic beliefs can result in an equilibrium bad for both players, one that is not Pareto efficient. These beliefs seem unrealistic, though, and game theorists are often willing to reject some perfect Bayesian equilibria as implausible.
Equilibria 1 and 2 are the only equilibria that might exist, but we can also check for the two potential separating equilibria, in which the two types of sender choose different actions, and see why they do not exist as perfect Bayesian equilibria:
- Suppose the sender's strategy is: Give if a friend, Do not give if an enemy. The receiver's beliefs are updated accordingly: if he receives a gift, he believes the sender is a friend; otherwise, he believes the sender is an enemy. Thus, the receiver will respond with Accept. If the receiver chooses Accept, though, the enemy sender will deviate to Give, to increase his payoff from 0 to 1, so this cannot be an equilibrium.
- Suppose the sender's strategy is: Do not give if a friend, Give if an enemy. The receiver's beliefs are updated accordingly: if he receives a gift, he believes the sender is an enemy; otherwise, he believes the sender is a friend. The receiver's best-response strategy is Reject. If the receiver chooses Reject, though, the enemy sender will deviate to Do not give, to increase his payoff from -1 to 0, so this cannot be an equilibrium.
We conclude that in this game, there is no separating equilibrium.
Gift game 2
In the following example,[1] the set of PBEs is strictly smaller than the set of SPEs and BNEs. It is a variant of the above gift-game, with the following change to the receiver's utility:
- If the sender is a friend, then the receiver's utility is 1 (if they accept) or 0 (if they reject).
- If the sender is an enemy, then the receiver's utility is 0 (if they accept) or -1 (if they reject).
Note that in this variant, accepting is a weakly dominant strategy for the receiver.
Similarly to example 1, there is no separating equilibrium. Let's look at the following potential pooling equilibria:
- The sender's strategy is: always give. The receiver's beliefs are not updated: they still believe in the a-priori probability, that the sender is a friend with probability and an enemy with probability . Their payoff from accepting is always higher than from rejecting, so they accept (regardless of the value of ). This is a PBE - it is a best-response for both sender and receiver.
- The sender's strategy is: never give. Suppose the receiver's beliefs when receiving a gift is that the sender is a friend with probability , where is any number in . Regardless of , the receiver's optimal strategy is: accept. This is NOT a PBE, since the sender can improve their payoff from 0 to 1 by giving a gift.
- The sender's strategy is: never give, and the receiver's strategy is: reject. This is NOT a PBE, since for any belief of the receiver, rejecting is not a best-response.
Note that option 3 is a Nash equilibrium! If we ignore beliefs, then rejecting can be considered a best-response for the receiver, since it does not affect their payoff (since there is no gift anyway). Moreover, option 3 is even a SPE, since the only subgame here is the entire game! Such implausible equilibria might arise also in games with complete information, but they may be eliminated by applying subgame perfect Nash equilibrium. However, Bayesian games often contain non-singleton information sets and since subgames must contain complete information sets, sometimes there is only one subgame—the entire game—and so every Nash equilibrium is trivially subgame perfect. Even if a game does have more than one subgame, the inability of subgame perfection to cut through information sets can result in implausible equilibria not being eliminated.
To summarize: in this variant of the gift game, there are two SPEs: either the sender always gives and the receiver always accepts, or the sender always does not give and the receiver always rejects. From these, only the first one is a PBE; the other is not a PBE since it cannot be supported by any belief-system.
More examples
For further examples, see signaling game#Examples. See also [2] for more examples.
PBE in multi-stage games
A multi-stage game is a sequence of simultaneous games played one after the other. These games may be identical (as in repeated games) or different.
Repeated public-good game
Build | Don't | |
Build | 1-C1, 1-C2 | 1-C1, 1 |
Don't | 1, 1-C2 | 0,0 |
Public good game |
The following game[3]: section 6.2 is a simple representation of the free-rider problem. There are two players, each of whom can either build a public good or not build. Each player gains 1 if the public good is built and 0 if not; in addition, if player builds the public good, they have to pay a cost of . The costs are private information - each player knows their own cost but not the other's cost. It is only known that each cost is drawn independently at random from some probability distribution. This makes this game a Bayesian game.
In the one-stage game, each player builds if-and-only-if their cost is smaller than their expected gain from building. The expected gain from building is exactly 1 times the probability that the other player does NOT build. In equilibrium, for every player , there is a threshold cost , such that the player contributes if-and-only-if their cost is less than . This threshold cost can be calculated based on the probability distribution of the players' costs. For example, if the costs are distributed uniformly on , then there is a symmetric equilibrium in which the threshold cost of both players is 2/3. This means that a player whose cost is between 2/3 and 1 will not contribute, even though their cost is below the benefit, because of the possibility that the other player will contribute.
Now, suppose that this game is repeated two times.[3]: section 8.2.3 The two plays are independent, i.e., each day the players decide simultaneously whether to build a public good in that day, get a payoff of 1 if the good is built in that day, and pay their cost if they built in that day. The only connection between the games is that, by playing in the first day, the players may reveal some information about their costs, and this information might affect the play in the second day.
We are looking for a symmetric PBE. Denote by the threshold cost of both players in day 1 (so in day 1, each player builds if-and-only-if their cost is at most ). To calculate , we work backwards and analyze the players' actions in day 2. Their actions depend on the history (= the two actions in day 1), and there are three options:
- 첫날에는 어떤 선수도 만들지 않았다. 그래서 이제 두 선수 모두 상대의 비용이 이상이라는 것을 알고 있다 그들은 그에 따라 그들의 믿음을 갱신하고, 그들의 상대가 2일 안에 건설할 가능성이 더 적다고 결론짓는다. 따라서 임계비용을 증가시키고, 2일차 기준비용은 c > c이다
- 첫날, 두 선수 모두 만들었다. 그래서 이제 두 선수 모두 상대의 비용이 보다 낮다는 것을 안다 그들은 그에 따라 그들의 믿음을 갱신하고, 그들의 상대가 2일 안에 건설할 가능성이 더 크다고 결론짓는다. 따라서 임계값 비용을 절감하고, 2일차의 임계값 은 11 < ^ 이다
- 1일차에는 정확히 한 명의 플레이어가 구축되었다. 1번 플레이어를 가정해 보십시오. 그래서 지금은 플레이어 1의 비용이 이하, 2의비용이c ^ {\ 이상인 것으로 알려져 있다 2일의 동작이 플레이어 1의 빌드 작업과 동일하고 플레이어 2가 빌드하지 않는 평형이 있다.
이러한 각 상황에서 "임계 플레이어" (정확히 이c^ 의 기대 수익을 계산할 수 있다. 임계값 플레이어는 기여와 기여 사이에 무관심해야 하므로 1일 임계값 비용 을 계산할 수 있으며 이 임계값은 1단계 게임의 임계값인 c보다 낮은 것으로 나타났다. 2단계 경기에서는 선수들이 1단계 경기보다 구축 의지가 떨어진다는 의미다. 직감적으로 이유는 첫날에 기여하지 않을 때는 상대 선수가 자신의 비용이 높다고 믿게 하고, 이는 다른 선수가 둘째 날 기여를 더 적극적으로 하게 하기 때문이다.
점프입찰
공개적인 영국 경매에서 입찰자들은 작은 단계(예: 매회 1달러)로 현재의 가격을 올릴 수 있다. 그러나 종종 점프 입찰이 있다 - 일부 입찰자들은 최소한의 인상보다 훨씬 더 현재 가격을 인상한다. 이에 대한 한 가지 설명은 그것이 다른 입찰자들에게 신호 역할을 한다는 것이다. 각 입찰자가 특정 임계값을 초과하는 경우에만 점프하는 PBE가 있다. 점프 입찰#서명을 참조하십시오.
참고 항목
- 순차적 평형 - 불균형 정보 세트에 할당할 수 있는 믿음을 "합리적" 정보 세트로 제한하는 PBE의 정교함.
- 직관적 기준과 신성한 균형 - 신호 게임에 특화된 PBE의 기타 개선.
참조
- ^ James Peck. "Perfect Bayesian Equilibrium" (PDF). Ohio State University. Retrieved 6 December 2021.
- ^ Zack Grossman. "Perfect Bayesian Equilibrium" (PDF). University of California. Retrieved 2 September 2016.
- ^ a b Fudenberg, Drew; Tirole, Jean (1991). Game theory. Cambridge, Massachusetts: MIT Press. ISBN 9780262061414. 책 미리보기.