文章目录
博弈论笔记
教材:(不确定,都可以)
Reference:
Osborm an introductory to game theory
Dixit and skeath,games of strategy
Gibbons,A Primer in Game Theory
Binmore Playing for Real:AText on game theory
评分:
participation(10%) + homework(10%) +project +presentation(20%) +final (50%)
考试题目用英文
presentation
博弈论案例或者问题,有趣的博弈问题
project
1000字(俚语俗语语言中的博弈论
别忘记细节
- 求纳什均衡的三种方法,剔除严格劣策略,画博弈矩阵,最优反应函数
- 描述一个博弈需要包括: player ; game tree(timing) , 终端历史;每个时间段或者历史对应的玩家p(h)={1};pay off ;不确定性,确定的信念。 (不需要描述策略,如果是静态博弈就把timing改成action)
- 不完全信息是对谁来说一定要弄清楚,不要把二者弄混淆这样就会错的。
- 剔除劣策略时,也要注意是看哪个payoff,这个payoff找错了,会发生剔除错的!!!一定要看清楚是哪个比哪个大,行比较列,列比较行!!!
- 策略需要把不完全信息加上去!!!所以就有很多啊!!!
- 贝叶斯均衡怎么描述qwq
- 博弈的扩展式就是博弈树,博弈的战略式就是博弈矩阵
- 概率论两个随机变量的和差
- 有范围的记住说一下范围,别忘啦
入门
9.13
博弈论是多人决策的问题
要素
players, choices,payoffs
timing
information
静态博弈
动态博弈
完全信息博弈
不完全信息博弈
囚徒困境,信号博弈,勇士博弈,谈判,上下游,拍卖,寡头
静态完全信息博弈,不完全信息动态博弈,动态完全信息博弈
解概念
占优策略,纳什均衡,贝叶斯均衡等
Introduction
game theory: 博弈论提供了正式研究互动策略的一系列工具。
strategic interaction: 多人决策 每个人的报酬都取决于其他人的策略
一些博弈:讨价还价 寡头竞争 拍卖 商业 政策 外交 军事冲突
博弈语言
players choices payoffs timing information
完全信息动态博弈 完全信息静态博弈 不完全信息动态博弈 不完全信息静态博弈
完全信息静态博弈
The set of players: N = { 1 , 2 , … , n } N=\{1,2,\dots ,n\} N={1,2,…,n}
player i’s feasible choices : A i A_i Ai
Player i’s payoff function: u i : A 1 × A 2 ⋯ × A n → R u_i:A_1\times A_2\dots \times A_n\to R ui:A1×A2⋯×An→R
解概念
dominance 占优
纳什均衡
背景介绍
贝叶斯均衡 等等
Prisoner’s Dilemma
囚徒困境 : 个人理性与社会最优相矛盾
占优策略,选择这个严格比另一个策略好
comments:
payoff and modeling; rationality assumption
囚徒困境的例子:
团队工作 军事竞赛 密谋勾结 公共资源(教育医疗排污环境)
解决方式: 国家 长期合作 第三方
Dominance
players choices payoff
p
l
a
y
e
r
s
:
{
1
,
2
,
…
,
N
}
p
l
a
y
e
r
i
′
s
c
h
o
i
c
e
s
:
{
A
i
}
p
l
a
y
e
r
i
′
s
p
a
y
o
f
f
:
u
i
:
A
1
×
⋯
×
A
n
→
R
players:\{1,2,\dots ,N\}\\player ~i's~ choices:\{A_i\}\\player ~i's ~payoff:u_i:A_1\times \dots \times A_n\to R
players:{1,2,…,N}player i′s choices:{Ai}player i′s payoff:ui:A1×⋯×An→R
a = ( a 1 , a 2 , … , a n ) a − i = ( a 1 , a 2 , … , 无 a i , a i + 1 , … , a n ) A 1 × A 2 = : { ( a 1 , a 2 ) ∣ a 1 ∈ A 1 , a 2 ∈ A 2 } A 1 × ⋯ × A n = ; { ( a 1 , a 2 , … , a n ) ∣ a i ∈ A i , i = 1 , 2 , … } A A − i a ∈ A a − i ∈ A − i a=(a_1,a_2,\dots ,a_n)\\ a_{-i}=(a_1,a_2,\dots ,无a_i,a_{i+1},\dots,a_n)\\ A_1\times A_2=:\{(a_1,a_2)|a_1\in A_1,a_2\in A_2\}\\ A_1\times \dots \times A_n=;\{(a_1,a_2,\dots,a_n)|a_i\in A_i,i=1,2,\dots\}\\ A \quad A_{-i} \quad a\in A\quad a_{-i}\in A_{-i} a=(a1,a2,…,an)a−i=(a1,a2,…,无ai,ai+1,…,an)A1×A2=:{(a1,a2)∣a1∈A1,a2∈A2}A1×⋯×An=;{(a1,a2,…,an)∣ai∈Ai,i=1,2,…}AA−ia∈Aa−i∈A−i
Dominance
a
i
a_{i}
ai (strictly) dominates
b
i
b_{i}
bi if
u
i
(
a
i
,
a
−
i
)
>
u
i
(
b
i
,
a
−
i
)
u_{i}\left(a_{i}, a_{-i}\right)>u_{i}\left(b_{i}, a_{-i}\right) \quad
ui(ai,a−i)>ui(bi,a−i) for all
a
−
i
∈
A
−
i
a_{-i} \in A_{-i}
a−i∈A−i.
b被占优,a是占优策略如果占优其他所有的行动
理性参与者不会采用劣势策略
劣势策略可以从考虑中消除
囚徒困境:每个人的占有选择导致的结果是社会次优的
Iterated Dominance(迭代的主导地位
逐步剔除被占优策略
占优可求:如果剔除的极限得到的集合是一个单元素集合
common knowledge of rationality
common knowledge: 我知道,你知道我知道,我知道你知道我知道,……
common knowledge of rationality:理性的共同知识
共同知识有更新,人们对理性的理解有更深一步的了解
Nash Equilibrium
Coordinated attack
-
没有期望报酬
-
期望去or不去
-
第一轮决定了后面的收敛结果:若第一次成功袭击,后面也会袭击,否则就不会袭击了
纳什均衡
( a 1 , a 2 , … , a n ) \left(a_{1}, a_{2}, \ldots, a_{n}\right) (a1,a2,…,an) is a Nash equilibrium
if u i ( a i , a − i ) ≥ u i ( a i ′ , a − i ) u_{i}\left(a_{i}, a_{-i}\right) \geq u_{i}\left(a_{i}^{\prime}, a_{-i}\right) ui(ai,a−i)≥ui(ai′,a−i) for all a i ′ ∈ A i a_{i}^{\prime} \in A_{i} ai′∈Ai and all i ∈ N i \in N i∈N.
pre-play communication
只有商量的结果是纳什均衡才会被大家遵守
最优反应函数
a
i
a_{i}
ai is a best response to
a
−
i
a_{-i}
a−i if
u
i
(
a
i
,
a
−
i
)
≥
u
i
(
a
i
′
,
a
−
i
)
f
o
r
a
l
l
a
i
′
∈
A
i
.
u_{i}\left(a_{i}, a_{-i}\right) \geq u_{i}\left(a_{i}^{\prime}, a_{-i}\right) \quad for~ all ~a_{i}^{\prime} \in A_{i} .
ui(ai,a−i)≥ui(ai′,a−i)for all ai′∈Ai.
B i ( a − i ) B_{i}\left(a_{-i}\right) Bi(a−i) : the set of best responses to a − i a_{-i} a−i.
纳什均衡的最佳反应函数解法
( a 1 , a 2 , … , a n ) \left(a_{1}, a_{2}, \ldots, a_{n}\right) (a1,a2,…,an) is a Nash equilibrium iff
a i ∈ B i ( a − i ) , a_{i} \in B_{i}\left(a_{-i}\right), \quad ai∈Bi(a−i), for all i . i . i.
零和博弈:唯一的纳什均衡 唯一的报酬
给定一个博弈一定存在混合策略的纳什均衡,不一定存在纳什均衡
Oligopoly
卖方寡头垄断
条件: 厂商少
战略互动的突出例子
古诺博弈:决定产量
反垄断的方法:
- 谈判,坦白
- 判断价格更接近垄断还是竞争
囚徒困境:
- 垄断并购
- 违约
- 长期博弈(无穷维度)
伯川德博弈:决定价格
Unit demand
n
n
n products:
i
=
1
,
2
,
…
,
n
i=1,2, \ldots, n
i=1,2,…,n
Values:
(
v
1
,
…
,
v
n
)
∼
F
×
⋯
×
F
.
\left(v_{1}, \ldots, v_{n}\right) \sim F \times \cdots \times F .
(v1,…,vn)∼F×⋯×F.
Demand under prices
p
=
(
p
1
,
p
2
,
…
,
p
n
)
p=\left(p_{1}, p_{2}, \ldots, p_{n}\right)
p=(p1,p2,…,pn) :
Q
i
(
p
)
=
Pr
(
v
i
−
p
i
>
v
j
−
p
j
Q_{i}(p)=\operatorname{Pr}\left(v_{i}-p_{i}>v_{j}-p_{j}\right.
Qi(p)=Pr(vi−pi>vj−pj for all
j
≠
i
)
⋅
1
\left.j \neq i\right) \cdot 1
j=i)⋅1
Profit:
π
i
(
p
)
=
(
p
i
−
c
i
)
Q
i
(
p
)
\pi_{i}(p)=\left(p_{i}-c_{i}\right) Q_{i}(p)
πi(p)=(pi−ci)Qi(p)
求纳什均衡和串谋时的决策
信息完全博弈
打牌 公开叫价 信号+博弈
威胁 承诺(信誉) 谈判(讨价还价) 报仇
完全信息博弈:采取行动时知道所有之前的行动
A tree
T
T
T is a set of finite sequences such that
(
a
1
,
…
,
a
k
,
a
k
+
1
,
…
,
a
m
)
∈
T
⟹
(
a
1
,
…
,
a
k
)
∈
T
f
o
r
a
l
l
k
<
m
\left(a^{1}, \ldots, a^{k}, a^{k+1}, \ldots, a^{m}\right) \in T \Longrightarrow\left(a^{1}, \ldots, a^{k}\right) \in T~for~ all ~k<m
(a1,…,ak,ak+1,…,am)∈T⟹(a1,…,ak)∈T for all k<m
Capture all possible sequences of actions
T的序列叫做历史
$ \emptyset \in T$:初始历史
N:终端历史
完全信息动态博弈包括:
- a set of players I = { 1 , 2 , … , n } I=\{1,2, \ldots, n\} I={1,2,…,n},
- a tree T T T with terminal sequences N N N,
- a player function P : T \ N → I P: T \backslash N \rightarrow I P:T\N→I,
- a payoff function u i : N → R u_{i}: N \rightarrow \mathbf{R} ui:N→R for each i ∈ I i \in I i∈I.
Stackberg duopoly
行动的定义
Actions available to the player who moves after
h
h
h :
A
(
h
)
≡
{
a
∣
(
h
,
a
)
∈
T
}
.
A(h) \equiv\{a \mid(h, a) \in T\} .
A(h)≡{a∣(h,a)∈T}.
策略的定义
A strategy
s
i
s_{i}
si for player
i
i
i specifies an action
s
i
(
h
)
∈
A
(
h
)
s_{i}(h) \in A(h)
si(h)∈A(h)
for each
h
h
h such that
P
(
h
)
=
i
P(h)=i
P(h)=i.
完整的终端历史 O ( s ) ≡ ( a 1 , a 2 , … , a m ) O(s) \equiv\left(a^{1}, a^{2}, \ldots, a^{m}\right) O(s)≡(a1,a2,…,am)
A strategy profile
s
s
s is a Nash equilibrium if
u
i
(
O
(
s
i
,
s
−
i
)
)
≥
u
i
(
O
(
s
i
′
,
s
−
i
)
)
u_{i}\left(O\left(s_{i}, s_{-i}\right)\right) \geq u_{i}\left(O\left(s_{i}^{\prime}, s_{-i}\right)\right)
ui(O(si,s−i))≥ui(O(si′,s−i))
for each
i
i
i and each
s
i
′
s_{i}^{\prime}
si′.
无效威胁,不可置信威胁
动态博弈:子博弈精炼纳什均衡
max
q
1
(
1
−
q
1
−
1
−
q
1
2
)
{
s
1
∗
=
1
2
s
2
∗
=
1
−
q
1
2
\begin{aligned} \max & q_{1}\left(1-q_{1}-\frac{1-q_1}{2}\right) \\ &\left\{\begin{array}{l} s_{1}^{*}=\frac{1}{2} \\ s_{2}^{*}=\frac{1-q_{1}}{2} \end{array}\right. \end{aligned}
maxq1(1−q1−21−q1){s1∗=21s2∗=21−q1
注意这个
s
2
∗
s_2^*
s2∗
子博弈精炼纳什均衡
s
s
s is a subgame perfect Nash equilibrium if
u
i
(
O
h
(
s
)
)
≥
u
i
(
O
h
(
s
i
′
,
s
−
i
)
)
u_{i}\left(O_{h}(s)\right) \geq u_{i}\left(O_{h}\left(s_{i}^{\prime}, s_{-i}\right)\right)
ui(Oh(s))≥ui(Oh(si′,s−i))
for each
i
i
i and each
s
i
′
s_{i}^{\prime}
si′ and each
h
∈
T
\
N
h \in T \backslash N
h∈T\N.
完全信息动态博弈描述
- Players: I = { 1 , 2 , … , n } I=\{1,2, \ldots, n\} I={1,2,…,n}
- Tree T h : = { h ′ ∣ ( h , h ′ ) ∈ T } T_{h}:=\left\{h^{\prime} \mid\left(h, h^{\prime}\right) \in T\right\} Th:={h′∣(h,h′)∈T}
- Terminal histories N h : = { h ′ ∣ ( h , h ′ ) ∈ N } N_{h}:=\left\{h^{\prime} \mid\left(h, h^{\prime}\right) \in N\right\} Nh:={h′∣(h,h′)∈N},
- Player function P h ( h ′ ) : = P ( h , h ′ ) P_{h}\left(h^{\prime}\right):=P\left(h, h^{\prime}\right) Ph(h′):=P(h,h′),
- Payoff function u h , i ( h ′ ) : = u i ( h , h ′ ) u_{h, i}\left(h^{\prime}\right):=u_{i}\left(h, h^{\prime}\right) uh,i(h′):=ui(h,h′).
- Induced strategy: s h , i ( h ′ ) : = s i ( h , h ′ ) s_{h, i}\left(h^{\prime}\right):=s_{i}\left(h, h^{\prime}\right) sh,i(h′):=si(h,h′).
backward induction:后退归纳法
无穷期博弈表示:
Let A = { ( C , C ) , ( C , D ) , ( D , C ) , ( D , D ) } A=\{(C, C),(C, D),(D, C),(D, D)\} A={(C,C),(C,D),(D,C),(D,D)}.
T
=
{
∅
}
∪
A
∪
A
2
∪
⋯
∪
A
∞
.
T=\{\emptyset\} \cup A \cup A^{2} \cup \cdots \cup A^{\infty} .
T={∅}∪A∪A2∪⋯∪A∞.
N
=
A
∞
.
N=A^{\infty} .
N=A∞.
P
(
h
)
=
{
1
,
2
}
P(h)=\{1,2\}
P(h)={1,2} for all
h
∈
T
/
N
.
h \in T / N .
h∈T/N.
Let
r
1
(
C
,
C
)
=
10
,
r
1
(
C
,
D
)
=
0
r
1
(
D
,
C
)
=
15
,
r
1
(
D
,
D
)
=
5
\begin{aligned} r_{1}(C, C) &=10, r_{1}(C, D)=0 \\ r_{1}(D, C) &=15, r_{1}(D, D)=5 \end{aligned}
r1(C,C)r1(D,C)=10,r1(C,D)=0=15,r1(D,D)=5
Then
u
1
(
a
1
,
a
2
,
…
)
=
r
1
(
a
1
)
+
δ
r
1
(
a
2
)
+
δ
2
r
1
(
a
3
)
+
⋯
.
u_{1}\left(a^{1}, a^{2}, \ldots\right)=r_{1}\left(a^{1}\right)+\delta r_{1}\left(a^{2}\right)+\delta^{2} r_{1}\left(a^{3}\right)+\cdots .
u1(a1,a2,…)=r1(a1)+δr1(a2)+δ2r1(a3)+⋯.
长期关系
Tit or tat ?
以牙还牙
Grim Trigger
冷酷策略
严格报复策略是纳什均衡,子博弈精炼纳什均衡
当历史有D无D,都为纳什均衡
取决于贴现KaTeX parse error: Undefined control sequence: \var at position 1: \̲v̲a̲r̲和明天还会玩的概率
参与人变化比如家族或者惩罚一村的人 信息传递
不完全信息动态博弈
信息不对称
炸金花博弈描述
Two players: 1 and 2 .
Game tree:
T
=
{
∅
,
(
F
)
,
(
R
)
,
(
R
,
F
)
,
(
R
,
C
)
}
T=\{\emptyset,(F),(R),(R, F),(R, C)\}
T={∅,(F),(R),(R,F),(R,C)}.
Player function:
P
(
∅
)
=
1
,
P
(
(
R
)
)
=
2
P(\emptyset)=1, P((R))=2
P(∅)=1,P((R))=2.
Uncertainty:
Θ
=
Θ
1
=
{
\Theta=\Theta_{1}=\{
Θ=Θ1={ Ace, King
}
\}
};
Pr
(
θ
=
\operatorname{Pr}(\theta=
Pr(θ= Ace
)
=
0.5
)=0.5
)=0.5.
Payoff:
u
1
(
(
F
)
,
θ
)
=
−
1
;
u
1
(
(
R
,
F
)
,
θ
)
=
1
,
…
u_{1}((F), \theta)=-1 ; u_{1}((R, F), \theta)=1, \ldots
u1((F),θ)=−1;u1((R,F),θ)=1,…
不完全信息博弈策略: 在每个可能做决策的地方对每个可能的你做一个决策行动
不完全信息是对方,别把自己给绕晕了!!!!
Given
(
σ
1
,
σ
2
,
b
)
:
\left(\sigma_{1}, \sigma_{2}, b\right):
(σ1,σ2,b):
E
[
u
1
(
m
1
,
σ
2
,
θ
)
]
=
u
1
(
m
1
,
a
1
,
θ
)
σ
2
+
u
1
(
m
1
,
a
2
,
θ
)
(
1
−
σ
2
)
\begin{aligned} E\left[u_{1}\left(m_{1}, \sigma_{2}, \theta\right)\right] &= u_{1}\left(m_{1}, a_{1}, \theta\right) \sigma_{2}+u_{1}\left(m_{1}, a_{2}, \theta\right)\left(1-\sigma_{2}\right) \end{aligned}
E[u1(m1,σ2,θ)]=u1(m1,a1,θ)σ2+u1(m1,a2,θ)(1−σ2)
E
[
u
2
(
m
1
,
a
,
θ
)
∣
b
]
=
u
2
(
m
1
,
a
,
t
1
)
b
+
u
2
(
m
1
,
a
,
t
2
)
(
1
−
b
)
\begin{aligned} &E\left[u_{2}\left(m_{1}, a, \theta\right) \mid b\right]= u_{2}\left(m_{1}, a, t_{1}\right) b+u_{2}\left(m_{1}, a, t_{2}\right)(1-b) \end{aligned}
E[u2(m1,a,θ)∣b]=u2(m1,a,t1)b+u2(m1,a,t2)(1−b)
贝叶斯公式
b
=
σ
1
(
t
1
)
γ
σ
1
(
t
1
)
γ
+
σ
1
(
t
2
)
(
1
−
γ
)
b=\frac{\sigma_{1}\left(t_{1}\right) \gamma}{\sigma_{1}\left(t_{1}\right) \gamma+\sigma_{1}\left(t_{2}\right)(1-\gamma)}
b=σ1(t1)γ+σ1(t2)(1−γ)σ1(t1)γ
Expected payoff:
E
[
u
i
(
a
i
,
s
j
(
θ
j
)
,
θ
i
,
θ
j
)
]
=
∑
θ
j
∈
Θ
j
u
i
(
a
i
,
s
j
(
θ
j
)
,
θ
i
,
θ
j
)
f
j
(
θ
j
)
E\left[u_{i}\left(a_{i}, s_{j}\left(\theta_{j}\right), \theta_{i}, \theta_{j}\right)\right]=\sum_{\theta_{j} \in \Theta_{j}} u_{i}\left(a_{i}, s_{j}\left(\theta_{j}\right), \theta_{i}, \theta_{j}\right) f_{j}\left(\theta_{j}\right)
E[ui(ai,sj(θj),θi,θj)]=θj∈Θj∑ui(ai,sj(θj),θi,θj)fj(θj)
Expected payoff:
E
[
u
i
(
a
i
,
s
j
(
θ
j
)
,
θ
)
]
=
∫
θ
j
∈
Θ
j
u
i
(
a
i
,
s
j
(
θ
j
)
,
θ
)
f
j
(
θ
j
)
d
θ
j
E\left[u_{i}\left(a_{i}, s_{j}\left(\theta_{j}\right), \theta\right)\right]=\int_{\theta_{j} \in \Theta_{j}} u_{i}\left(a_{i}, s_{j}\left(\theta_{j}\right), \theta\right) f_{j}\left(\theta_{j}\right) \mathrm{d} \theta_{\mathrm{j}}
E[ui(ai,sj(θj),θ)]=∫θj∈Θjui(ai,sj(θj),θ)fj(θj)dθj
s
=
(
s
1
,
s
2
)
s=\left(s_{1}, s_{2}\right)
s=(s1,s2) is a Bayesian equilibrium if
E
[
u
i
(
s
i
(
θ
i
)
,
s
j
(
θ
j
)
,
θ
)
]
≥
E
[
u
i
(
a
i
,
s
j
(
θ
j
)
,
θ
)
]
E\left[u_{i}\left(s_{i}\left(\theta_{i}\right), s_{j}\left(\theta_{j}\right), \theta\right)\right] \geq E\left[u_{i}\left(a_{i}, s_{j}\left(\theta_{j}\right), \theta\right)\right]
E[ui(si(θi),sj(θj),θ)]≥E[ui(ai,sj(θj),θ)]
for each
a
i
a_{i}
ai, each
θ
i
\theta_{i}
θi, and each
i
i
i.
s
i
s_{i}
si is a dominant strategy if
E
[
u
i
(
s
i
(
θ
i
)
,
s
j
(
θ
j
)
,
θ
)
]
≥
E
[
u
i
(
a
i
,
s
j
(
θ
j
)
,
θ
)
]
,
E\left[u_{i}\left(s_{i}\left(\theta_{i}\right), s_{j}\left(\theta_{j}\right), \theta\right)\right] \geq E\left[u_{i}\left(a_{i}, s_{j}\left(\theta_{j}\right), \theta\right)\right],
E[ui(si(θi),sj(θj),θ)]≥E[ui(ai,sj(θj),θ)],
for each
a
i
a_{i}
ai, each
θ
i
\theta_{i}
θi, and each
s
j
s_{j}
sj.
贝叶斯均衡
Perfect Bayesian equilibrium Defifinition
( σ 1 , σ 2 , b ) \left(\sigma_{1}, \sigma_{2}, b\right) (σ1,σ2,b) is a perfect Bayesian equilibrium if:
-
if σ 1 ( θ ) > 0 \sigma_{1}(\theta)>0 σ1(θ)>0, then
E [ u 1 ( m 1 , σ 2 , θ ) ] ≥ u 1 ( m 2 , θ ) ; E\left[u_{1}\left(m_{1}, \sigma_{2}, \theta\right)\right] \geq u_{1}\left(m_{2}, \theta\right) ; E[u1(m1,σ2,θ)]≥u1(m2,θ);
if σ 1 ( θ ) < 1 \sigma_{1}(\theta)<1 σ1(θ)<1, then
E [ u 1 ( m 1 , σ 2 , θ ) ] ≤ u 1 ( m 2 , θ ) ; E\left[u_{1}\left(m_{1}, \sigma_{2}, \theta\right)\right] \leq u_{1}\left(m_{2}, \theta\right) ; E[u1(m1,σ2,θ)]≤u1(m2,θ); -
if σ 2 > 0 \sigma_{2}>0 σ2>0, then
E [ u 2 ( m 1 , a 1 , θ ) ∣ b ] ≥ E [ u 2 ( m 1 , a 2 , θ ) ∣ b ] ; E\left[u_{2}\left(m_{1}, a_{1}, \theta\right) \mid b\right] \geq E\left[u_{2}\left(m_{1}, a_{2}, \theta\right) \mid b\right] ; E[u2(m1,a1,θ)∣b]≥E[u2(m1,a2,θ)∣b];
if σ 2 < 1 \sigma_{2}<1 σ2<1, then
E [ u 2 ( m 1 , a 1 , θ ) ∣ b ] ≤ E [ u 2 ( m 1 , a 2 , θ ) ∣ b ] ; E\left[u_{2}\left(m_{1}, a_{1}, \theta\right) \mid b\right] \leq E\left[u_{2}\left(m_{1}, a_{2}, \theta\right) \mid b\right] ; E[u2(m1,a1,θ)∣b]≤E[u2(m1,a2,θ)∣b]; -
if σ 1 ( t 1 ) > 0 \sigma_{1}\left(t_{1}\right)>0 σ1(t1)>0 or σ 1 ( t 2 ) > 0 \sigma_{1}\left(t_{2}\right)>0 σ1(t2)>0, then
b = σ 1 ( t 1 ) γ σ 1 ( t 1 ) γ + σ 1 ( t 2 ) ( 1 − γ ) . b=\frac{\sigma_{1}\left(t_{1}\right) \gamma}{\sigma_{1}\left(t_{1}\right) \gamma+\sigma_{1}\left(t_{2}\right)(1-\gamma)} . b=σ1(t1)γ+σ1(t2)(1−γ)σ1(t1)γ.
拍卖理论
Auctions
第二价格拍卖
价高者付第二高的价格
第一价格拍卖
价高者付第一高的价格
Players: buyers 1 and
2.
2 .
2.
Action=bid:
b
i
∈
[
0
,
∞
)
b_{i} \in[0, \infty)
bi∈[0,∞).
Uncertainty:
Θ
1
=
Θ
2
=
[
0
,
1
]
;
v
i
∼
U
[
0
,
1
]
\Theta_{1}=\Theta_{2}=[0,1] ; v_{i} \sim U[0,1]
Θ1=Θ2=[0,1];vi∼U[0,1].
Payoff:
u
i
(
b
i
,
b
j
,
v
i
)
=
{
v
i
−
b
i
,
if
b
i
>
b
j
;
1
2
(
v
i
−
b
i
)
,
if
b
i
=
b
j
;
0
,
if
b
i
<
b
j
u_{i}\left(b_{i}, b_{j}, v_{i}\right)= \begin{cases}v_{i}-b_{i}, & \text { if } b_{i}>b_{j} ; \\ \frac{1}{2}\left(v_{i}-b_{i}\right), & \text { if } b_{i}=b_{j} ; \\ 0, & \text { if } b_{i}<b_{j}\end{cases}
ui(bi,bj,vi)=⎩⎪⎨⎪⎧vi−bi,21(vi−bi),0, if bi>bj; if bi=bj; if bi<bj
Second-price auction:
∫
0
1
∫
0
1
min
{
v
1
,
v
2
}
d
v
1
d
v
2
=
1
3
.
\int_{0}^{1} \int_{0}^{1} \min \left\{v_{1}, v_{2}\right\} \mathrm{d} v_{1} \mathrm{~d} v_{2}=\frac{1}{3} .
∫01∫01min{v1,v2}dv1 dv2=31.
First-price auction:
∫
0
1
∫
0
1
max
{
v
1
2
,
v
2
2
}
d
v
1
d
v
2
=
1
3
.
\int_{0}^{1} \int_{0}^{1} \max \left\{\frac{v_{1}}{2}, \frac{v_{2}}{2}\right\} \mathrm{d} v_{1} \mathrm{~d} v_{2}=\frac{1}{3} .
∫01∫01max{2v1,2v2}dv1 dv2=31.
signaling
说真话要付出代价才会让人相信
分离均衡(Separating equilibrium)
可以完全分清楚类型
混同均衡(Pooling equilibrium)
没有信息的更新
半分离均衡(Semieparating equilibrium)
有一定的信息更新,可以分清楚一部分
有无信息的更新
完美贝叶斯博弈均衡时一个解概念
混同、分离、办分离均衡都属于完美贝叶斯博弈均衡的一种
总结:
[外链图片转存失败,源站可能有防盗链机制,建议将图片保存下来直接上传(img-gfiKtY5F-1641462637615)(https://gitee.com/zhang-yixin/markdown-drawing-bed/raw/master/img//202201041219492.png)]