game theory 博弈论 01 rationality 理性

1. Game 博弈：

a process consisting in

-A set of plays (at least two) 一组玩家，至少两名；

-An initial situation 初始情况；

-Rules that the plays must follow 玩家必须遵守的规则；

-All possible final situations 所有可能的最终情况；

-The preferences of all players on the set of the final situation 所有玩家对最终情况的偏好

Game theory is the study of taking optimal decisions in presence of multiple players (agents).博弈论是研究在多个参与者（代理）存在的情况下做出最优决策的理论。

2. Assumptions 假设

Players are supposed to be

-Selfish: the players only care about their own preference with respect to the outcomes of the game (mathematical assumption) 玩家只关心他们自己对游戏结果的偏好

-Rational: for each type of these games we shall consider possible definitions of optimal behavior对于这些博弈的每一种类型，我们都将考虑最优行为的可能定义

  2.1 Preferences

   Definition: let X be a set. A preference relation on X is a binary relation

  Properties:

Thus x≥y means that alternative x is at leat as good as alternative y, while, naturally x>y means that x is strictly preferred or, differently said, that y≥x cannot hold.

因此，x≥y意味着备选方案x至少与备选方案y一样好，而自然地，x>y意味着x是严格优选的，或者换句话说，y≥x不能成立。

-The first rationality assumption is:

the players are able to provide a preference relation over the outcomes of the game, and the order must be consistent. 第一个理性假设是：玩家能够提供相对于游戏结果的偏好关系，并且顺序必须一致。

 2.2 Utility functions 效用函数

-Definition：

Let ⪰ be a preference relation over X. A utility function representing ⪰ is a function u : X → R such that

设⪰是相对于X的偏好关系。表示⪰的效用函数是函数u:X→ R使得

-Properties :

1. A utility function may not exist in particular cases, however it exists in the general setting, specifically when X is a finite set. 效用函数在特定情况下可能不存在，但它存在于一般设置中，特别是当X是有限集时
2. If a utility function exists, then there exist infinitely many utility functions, given by any strictly increasing transformation of the former. 如果一个效用函数存在，那么存在无限多个效用函数，由前者的任何严格递增变换给出。
3. To player i there is assigned a set Xi, representing all the choice available to her. Hence, the set X = xXi over which u is defined comprises the possible choices of all players. 对于玩家i，分配了一组Xi，代表她可以选择的所有选项。因此，定义u的集合X=xXi包括所有参与者的可能选择。

-The second rationality assumption reads:

the agents are able to provide a utility function representing their preferences relations, whenever it is necessary. 代理能够在必要时提供表示其偏好关系的效用函数

2.3 Allais experiment  Allais实验

a)In a sample of 72 people exposed to this experiment, 82% of them decided to play the Lottery B (preference for certainty of risk-averse people)在72名接受该实验的人的样本中，82%的人决定参加B彩票（偏好风险厌恶者的确定性）

Based on expected utilities, this is rational if means根据预期的效用，如果…，是理性的

b)In the same sample of people interviewed, 83% of them selected lottery C (preference for winning in the case of low probability) 在接受采访的同一样本中，83%的人选择了彩票C（在低概率的情况下更倾向于中奖）

Yet this is rational if , in contrast with experiment 1, thereby violating the independence axiom in expected utility theory. 然而，与实验1相比，如果…那么这是合理的，从而违反了预期效用理论中的独立性公理。

Allais paradox thus shows that people are not always rational players! 阿兰悖论由此表明，人们并不总是理性的玩家！

The third rationality assumption reads:

The players use consistently the laws of probability: in particular, they are consistent with the computation of the expected utilities, they are able to update probabilities according to Bayes rule...第三个合理性假设是：参与者一致地使用概率定律：特别是，它们与预期效用的计算一致，它们能够根据贝叶斯规则更新概率…

2.4 The beauty contest

A group of players is asked to write an integer between 1 and 100. The mean M is then calculated. The goal of the game is to write the number at the minimum distance from qM , with 0 < q < 1. 一组玩家被要求写一个介于1和100之间的整数。然后计算平均值M。游戏的目标是在距离qM的最小距离处写入数字，0＜q＜1。

A rational player will answer 1, independently of q. And he will probably lose!一个理性的玩家会独立于q回答1，他很可能会输！

For, let q = 1/2 . Given that M ≤ 100, at the first step it is irrational to write a number greater than 1/2 × 100, which is the actual number to guess. But then at the second step, since each player is rational and knows that also the other players are rational, one should repeat the previous reasoning so that it would be irrational to write a number greater than (1/2)^ 2 × 100. And so on and so forth, until at step n it would be irrational to write a number greater than ( 1/2 ) ^n × 100, from which one concludes that the only rational choice is to write the smallest possible number, namely 1.对于，设q=1/ 2。

假设M≤100，在第一步，写一个大于12×100的数字是不合理的，这是要猜测的实际数字。

但在第二步，由于每个参与者都是理性的，并且知道其他参与者也是理性的，因此应该重复前面的推理，这样写一个大于（1/2）^2×100的数字就是非理性的。

依此类推，直到在步骤n，写一个大于（1/2）^n×100的数字是不合理的，由此得出结论，唯一合理的选择是写尽可能小的数字，即1。

However, experiments show that the winning result is far higher than 1. In fact, it tends to grow with the value of the parameter q. 然而，实验表明，获胜的结果远高于1。事实上，它倾向于随着参数q的值而增长。

-The fourth rationality assumption reads:

The players are able to understand the consequences of all their actions, the consequences of this information on any other player, the consequences of the consequences and so on. 第四个理性假设是：玩家能够理解他们所有行为的后果，这些信息对任何其他玩家的后果，后果的后果等等。

2.5 Extending decision theory 扩展决策理论

-Finally, the fifth rationality assumption reads:

The players are able to use decision theory, whenever it is possible 最后，第五个合理性假设是：只要可能，参与者都可以使用决策理论。

That is, given a set of alternatives X, and a utility function u on X, each player seeks a  such that:  .也就是说，给定一组备选方案X和X上的效用函数u，每个参与者都寻求一个…，使得…。

3. Summary of the rationality assumptions合理性假设摘要

a)The players are able to rank the outcomes of the game. 玩家可以对游戏结果进行排名

b)The players are able to provide a utility function for their ranking. 玩家可以为他们的排名提供一个实用函数

c)The players use the expected value to build their utility function in presence of random event. 玩家使用期望值来建立他们在随机事件中的效用函数

d)The players are able to analyze all the consequences of their actions, and the consequences of the consequences and so on. 玩家能够分析他们行为的所有后果，以及后果的后果等等

e)The players use the tools of decision theory whenever possible. 玩家尽可能使用决策理论

4. An immediate and important consequence of the axioms公理的直接而重要的结果

-Principle of elimination of strictly dominated strategies.消除严格控制策略的原则。

-A player does not take an action ‘a’ it she has available an action ‘b’ providing her a strictly better result, no matter what the other players do. 无论其他玩家做什么，一个玩家都不会采取行动A，她有行动b，这会给她带来更好的结果。

- Example:

Player 1 action set is {18, … , 30}, whereas Player 2 action set is {accept, refuse}.

If the preference of Player 2 is passing the exam with any grade, rather than repeating it, the action refuse is strictly dominated.

Observe that asking for a better grade is not an available action in this game

示例：玩家1的动作集是{18，…，30}，而玩家2的动作集则是{accept，reject}。

如果玩家2的偏好是以任何分数通过考试，而不是重复考试，则严格控制行动拒绝。

注意，在这个游戏中，要求更好的分数是不可行的

5. Bidiagonal matrix 双对角矩阵

Conventionally, Player 1 chooses a row, while Player 2 chooses a colum.按照惯例，玩家1选择一行，而玩家2选择一列

This results in a pair of numbers, corresponding to the utilities of Player 1 and 2, respectively. The options can be summarized in a bimatrix, like the following:

这会产生一对数字，分别对应于玩家1和2的效用。这些选项可以总结为双矩阵，如下所示：

Utilities of Player 1:玩家1的效用

Since the second row is strictly dominated by the first, Player 1 selects the latter.

Likewise, Player 2 selects the first column, which strictly dominates the second one

由于第二行严格由第一行控制，玩家1选择后者第一行。同样，玩家2选择第一列，严格控制第二列。

Even if the Principle of elimination of strictly dominated actions may not be very informative, it has rather surprising consequences.

即使消除严格支配行动的原则可能信息量不大，但它也会产生相当令人惊讶的后果。

6. Comparisons of games博弈比较

Observe: relative to any single outcome both players are individually better off (i.e they always have greater utilities) in the first game rather than in the second

观察：相对于任何一个结果，两个玩家在第一场比赛中的表现都比在第二场比赛中要好（即他们总是有更大的效用）

Nevertheless, according to the principle, it is more convenient for them in the interactive situation to play the second game rather than the first! (for, the outcome pair (8, 8) is greater than (5, 5) )

尽管如此，根据原则，在互动的情况下，他们玩第二个游戏比玩第一个更方便！（例如，结果对（8，8）大于（5，5））

7. Less is better than more少胜于多

The second game contains all outcomes of the first, plus some further outcomes:

第二个游戏包含第一个游戏的所有结果，以及一些进一步的结果：

Here, the rationality axioms imply that in the first game the players should choose the outcome pair (10, 10), whereas in the second game they should choose the outcome pair (1, 1) in the first row and first column. 这里，合理性公理意味着在第一场比赛中，玩家应该选择结果对（10，10），而在第二场比赛中他们应该选择第一行和第一列的结果对（1，1）。

Therefore, just having less available actions can make the players better off!

因此，减少可用的动作可以让玩家过得更好！

8. Uniqueness issue唯一性问题

What are the rational outcomes of the following game?

以下游戏的合理结果是什么？

We formally do not know but it is obvious that the rational outcomes will be (1, 1)

However, the actions prescribed by (first row, second column) and (second row, first column) yield the same preferred outcomes but they cannot be distinguished, thereby creating a coordination problem between the players!

我们正式不知道，但很明显，合理的结果将是（1，1）。

然而，（第一行，第二列）和（第二行，第一列）规定的动作产生了相同的首选结果，但无法区分，从而在玩家之间产生了协调问题！

9. Elimination of dominated strategies 消除占主导地位的战略

Voting game with three players having distinct alternatives A, B, C available.

The players’ preferences are the following:

投票游戏，三个玩家有不同的选择A，B，C可用。玩家的偏好如下：

 

The rule is that the alternative that receives most votes will win. Yet, in case of three different votes, the alternative selected by Player 1 will win.

规则是，获得最多选票的候选人将获胜。然而，如果有三张不同的选票，由玩家1选择的替代者将获胜。

What can we expect to be the rational outcome of the game?

我们可以期待游戏的理性结果是什么？

Let us try with elimination of dominated actions让我们尝试消除被支配的行为：

1. Alternative A is a weakly dominant strategy for Player 1 备选方案A是玩家1的弱支配策略
2. Players 2 and 3 have their worst choice as weakly dominated strategy玩家2和3的最差选择是弱支配策略

In order to avoid their worst outcome, Pl2 keeps B and C (ordered in rows) and Pl3 keeps C and A (ordered in columns), while Pl1 always plays A.为了避免最坏的结果，Pl2保持B和C（按行排序），Pl3保持C和A（按列排序），而Pl1总是播放A。

Thus, the game reduces to a 2 × 2 table with the following outcomes因此，游戏简化为2×2的表格，结果如下：

Since C ⪶ A for both Pl2 and Pl3, they will choose the outcome in the second row and first column. Therefore the final result is C

由于Pl2和Pl3都是C⪶A，他们将选择第二行和第一列的结果。因此，最终结果为C。