Graph playground

\(1\)

\(n\)

\(\vdots\)

\(v_1 \sim F_1\)

\(v_n \sim F_n\)

\(b_1(\rm v_1)\)

\(b_n(\rm v_n)\)

\(\vdots\)

agents

strategy profile 

value distribution 

\(\bm F\)

\(\rm b_1\) 

\(\rm b_{\it n}\) 

actions 

i.e. bids

allocation

\({\bf x}=(\rm x_1, ..., x_n)\)

payment

\({\bf p}=(\rm p_1, ..., p_n)\)

\(\vdots\)

\(\bf b\)

\(\bf b\)

or: \({\bm b(\bm v) } = \bf b\)

mechanism

allocation is a function of the bids 

\(\bf x(b) = x(\bm b(v))\)

payment is a function of the bids 

\(\bf p(b) = p(\bm b(v))\)

ex ante

ex post

interim

three stages of a Bayesian game

agents know distribution of types 

[common prior]

\(v_i \sim\)\( F_i\)

agents learn their types 

\({\rm v}_i\)

game not played

game played; 

types and actions all known

also outcome and payments

\(\rm v\) 

\(\rm x\) 

\(\rm v_i\) 

\(\rm v_j\) 

\(\rm x_i\) 

\(\rm x_j\) 

\(\rm v_j (x_i - x_j)\) 

\(0\) 

\(q\)

\(x\)

\(1\)

\(s_L\)

\(s_H\)

\(b\)

quality: skill times budget

allocation: probability that one is going to get the item  

\(x \in [0, 1]\)

\(q = s \cdot b\)

\(q\)

\(x\)

\(1\)

\(s_L\)

\(s_H\)

Two skills; budget \(b=1\)

\(q\)

\(x\)

\(1\)

\(s_L\)

\(s_H\)

Feasible region

constraints agents face:

- non-negative utility

- does not exceed budget 

WTH is this feasible?

\(q\)

\(x\)

\(1\)

\(s_L\)

\(s_H\)

Two types of budget \(b_L, b_H\) 

\(b_L < 1\) by def

\(s_Lb_L\)

\(q\)

\(x\)

\(1\)

\(s_L\)

\(s_H\)

Feasible region now

\(b_L < 1\) by def

\(s_Lb_L\)

\(q\)

\(x\)

\(1\)

\(s_L\)

\(s_H\)

\(s_Lb_L\)

\(s_Hb_L\)

optimal menu 1

\(\{(s_Lb_L + \epsilon, 1)\}\)

\(q\)

\(x\)

\(1\)

\(s_L\)

\(s_H\)

optimal menu 2

\(\{(s_Hb_L, s_Hb_L/s_L - \epsilon), (s_L\min\{b_H, 1\}+\epsilon, 1)\}\)

\(s_Lb_L\)

\(s_Hb_L\)

When \( b_H < 1\) 

\(q\)

\(x\)

\(1\)

\(s_L\)

\(s_H\)

optimal menu 2

\(\{(s_Hb_L, s_Hb_L/s_L - \epsilon), (s_L\min\{b_H, 1\}+\epsilon, 1)\}\)

\(s_Lb_L\)

\(s_Hb_L\)

When \( b_H \geq 1\) 

\(q\)

\(x\)

\(1\)

\(s_L\)

\(s_H\)

Generalizing...

\(s_Lb_L\)

\(s_Hb_L\)

i.e. let \(b_L\) take different values

\(q\)

\(x\)

\(1\)

\(s_L\)

\(s_H\)

optimal menu 2

\(\{(s_Hb_L, s_Hb_L/s_L - \epsilon), (s_L\min\{b_H, 1\}+\epsilon, 1)\}\)

\(s_Lb_L\)

\(s_Hb_L\)

depending on the value of \(b_H\) 

\(q\)

\(x\)

\(1\)

\(s_L\)

\(s_H\)

\(b\)

\(q\)

\(x\)

\(1\)

\(s_L\)

\(s_H\)

\(b\)

\(q\)

\(x\)

\(1\)

\(s_L\)

\(s_H\)

Generalizing...

\(s_Lb_L\)

\(s_Hb_L\)

how do we justify this jump???

jump

first type of jump

\(q\)

\(x\)

\(1\)

\(s_L\)

\(s_H\)

Jump 2

\(s_Lb_i\)

\(s_Hb_j\)

\(q\)

\(x\)

\(1\)

\(s_L\)

\(s_H\)

Jump 3

\(s_Lb_i\)

\(s_Hb_j\)

Made with Slides.com