Network Economics -- Lecture 2: Incen5ves in online - - PowerPoint PPT Presentation

network economics lecture 2 incen5ves in online systems i
SMART_READER_LITE
LIVE PREVIEW

Network Economics -- Lecture 2: Incen5ves in online - - PowerPoint PPT Presentation

Network Economics -- Lecture 2: Incen5ves in online systems I: free riding and effort elicita5on Patrick Loiseau EURECOM Fall 2014 1


slide-1
SLIDE 1

Network ¡Economics ¡

  • ­‑-­‑ ¡

Lecture ¡2: ¡Incen5ves ¡in ¡online ¡ systems ¡I: ¡free ¡riding ¡and ¡effort ¡ elicita5on ¡ ¡

Patrick ¡Loiseau ¡ EURECOM ¡ Fall ¡2014 ¡

1 ¡

slide-2
SLIDE 2

References ¡

  • Main: ¡ ¡

– N. ¡Nisam, ¡T. ¡Roughgarden, ¡E. ¡Tardos ¡and ¡V. ¡Vazirani ¡(Eds). ¡ “Algorithmic ¡Game ¡Theory”, ¡CUP ¡2007. ¡Chapters ¡23 ¡(see ¡also ¡27). ¡ ¡

  • Available ¡online: ¡

hXp://www.cambridge.org/journals/nisan/downloads/Nisan_Non-­‑ printable.pdf ¡ ¡

  • Addi5onal: ¡ ¡

– Yiling ¡Chen ¡and ¡Arpita ¡Gosh, ¡“Social ¡Compu5ng ¡and ¡User ¡Generated ¡ Content,” ¡EC’13 ¡tutorial ¡

  • Slides ¡at ¡hXp://www.arpitaghosh.com/papers/ec13_tutorialSCUGC.pdf ¡and ¡

hXp://yiling.seas.harvard.edu/wp-­‑content/uploads/ SCUGC_tutorial_2013_Chen.pdf ¡ ¡

– M. ¡Chiang. ¡“Networked ¡Life, ¡20 ¡Ques5ons ¡and ¡Answers”, ¡CUP ¡2012. ¡ Chapters ¡3-­‑5. ¡

  • See ¡the ¡videos ¡on ¡www.coursera.org ¡ ¡

2 ¡

slide-3
SLIDE 3

Outline ¡

  • 1. Introduc5on ¡
  • 2. The ¡P2P ¡file ¡sharing ¡game ¡
  • 3. Free-­‑riding ¡and ¡incen5ves ¡for ¡contribu5on ¡
  • 4. Hidden ¡ac5ons: ¡the ¡principal-­‑agent ¡model ¡

3 ¡

slide-4
SLIDE 4

Outline ¡

  • 1. Introduc5on ¡
  • 2. The ¡P2P ¡file ¡sharing ¡game ¡
  • 3. Free-­‑riding ¡and ¡incen5ves ¡for ¡contribu5on ¡
  • 4. Hidden ¡ac5ons: ¡the ¡principal-­‑agent ¡model ¡

4 ¡

slide-5
SLIDE 5

Online ¡systems ¡

  • Resources ¡

– P2P ¡systems ¡

  • Informa5on ¡

– Ra5ngs ¡ – Opinion ¡polls ¡

  • Content ¡(user-­‑generated ¡content) ¡

– P2P ¡systems ¡ – Reviews ¡ – Forums ¡ ¡ – Wikipedia ¡

  • Labor ¡(crowdsourcing) ¡

– AMT ¡

  • In ¡all ¡these ¡systems, ¡there ¡is ¡a ¡need ¡for ¡users ¡contribu5on ¡

5 ¡

slide-6
SLIDE 6

P2P ¡networks ¡

  • First ¡ones: ¡Napster ¡(1999), ¡Gnutella ¡(2000) ¡

– ¡Free-­‑riding ¡problem ¡

  • Many ¡users ¡across ¡the ¡globe ¡self-­‑organizing ¡to ¡

share ¡files ¡

– Anonymity ¡ ¡ – One-­‑shot ¡interac5ons ¡ à Difficult ¡to ¡sustain ¡collabora5on ¡

  • Exacerbated ¡by ¡

– Hidden ¡ac5ons ¡(nondetectable ¡defec5on) ¡ – Cheap ¡pseudonyms ¡(mul5ple ¡iden55es ¡easy) ¡

6 ¡

slide-7
SLIDE 7

Incen5ve ¡mechanisms ¡

  • Good ¡technology ¡is ¡not ¡enough ¡
  • P2P ¡networks ¡need ¡incen5ve ¡mechanisms ¡to ¡

incen5vize ¡users ¡to ¡contribute ¡

– Reputa5on ¡(KaZaA) ¡ – Currency ¡(called ¡scrip) ¡ – Barter ¡(BitTorrent) ¡– ¡direct ¡reciprocity ¡

7 ¡

slide-8
SLIDE 8

Extensions ¡

  • Other ¡free-­‑riding ¡situa5ons ¡

– E.g., ¡mobile ¡ad-­‑hoc ¡networks, ¡P2P ¡storage ¡

  • Rich ¡strategy ¡space ¡

– Share/not ¡share ¡ – Amount ¡of ¡resources ¡commiXed ¡ – Iden5ty ¡management ¡

  • Other ¡applica5ons ¡of ¡incen5ves ¡/ ¡reputa5on ¡

systems ¡

– Online ¡shopping, ¡forums, ¡etc. ¡

8 ¡

slide-9
SLIDE 9

Outline ¡

  • 1. Introduc5on ¡
  • 2. The ¡P2P ¡file ¡sharing ¡game ¡
  • 3. Free-­‑riding ¡and ¡incen5ves ¡for ¡contribu5on ¡
  • 4. Hidden ¡ac5ons: ¡the ¡principal-­‑agent ¡model ¡

9 ¡

slide-10
SLIDE 10

The ¡P2P ¡file-­‑sharing ¡game ¡

  • Peer ¡

– Some5mes ¡download ¡à ¡benefit ¡ – Some5mes ¡upload ¡à ¡cost ¡

  • One ¡interac5on ¡~ ¡prisoner’s ¡dilemma ¡ ¡

C ¡ D ¡ C ¡ D ¡ 2, ¡2 ¡

  • ­‑1, ¡3 ¡

0, ¡0 ¡ 3, ¡-­‑1 ¡

10 ¡

slide-11
SLIDE 11

Prisoner’s ¡dilemma ¡

  • Dominant ¡strategy: ¡D ¡
  • Socially ¡op5mal ¡(C, ¡C) ¡
  • Single ¡shot ¡leads ¡to ¡(D, ¡D) ¡

– Socially ¡undesirable ¡

  • Iterated ¡prisoner’s ¡dilemma ¡

– Tit-­‑for-­‑tat ¡yields ¡socially ¡op5mal ¡outcome ¡

C ¡ D ¡ C ¡ D ¡ 2, ¡2 ¡

  • ­‑1, ¡3 ¡

0, ¡0 ¡ 3, ¡-­‑1 ¡

11 ¡

slide-12
SLIDE 12

P2P ¡

  • Many ¡users, ¡random ¡interac5ons ¡
  • Direct ¡reciprocity ¡does ¡not ¡scale ¡

Feldman ¡et ¡al. ¡2004 ¡

12 ¡

slide-13
SLIDE 13

P2P ¡

  • Direct ¡reciprocity

¡ ¡

– Enforced ¡by ¡BiXorrent ¡at ¡the ¡scale ¡of ¡one ¡file ¡but ¡ not ¡over ¡several ¡files ¡

  • Indirect ¡reciprocity ¡

– Reputa5on ¡system ¡ – Currency ¡system ¡

13 ¡

slide-14
SLIDE 14

How ¡to ¡treat ¡new ¡comers ¡

  • P2P ¡has ¡high ¡turnover ¡
  • Ooen ¡interact ¡with ¡stranger ¡with ¡no ¡history ¡
  • TFT ¡strategy ¡with ¡C ¡with ¡new ¡comers ¡

– Encourage ¡new ¡comers ¡ – BUT ¡Facilitates ¡whitewashing ¡

14 ¡

slide-15
SLIDE 15

Outline ¡

  • 1. Introduc5on ¡
  • 2. The ¡P2P ¡file ¡sharing ¡game ¡
  • 3. Free-­‑riding ¡and ¡incen5ves ¡for ¡contribu5on ¡
  • 4. Hidden ¡ac5ons: ¡the ¡principal-­‑agent ¡model ¡

15 ¡

slide-16
SLIDE 16

Reputa5on ¡

  • Long ¡history ¡of ¡facilita5ng ¡coopera5on ¡(e.g. ¡

eBay) ¡

  • In ¡general ¡coupled ¡with ¡service ¡differen5a5on ¡

– Good ¡reputa5on ¡= ¡good ¡service ¡ – Bad ¡reputa5on ¡= ¡bad ¡service ¡

  • Ex: ¡KaZaA ¡

16 ¡

slide-17
SLIDE 17

Trust ¡

  • EigenTrust ¡(Sep ¡Kamvar, ¡Mario ¡Schlosser, ¡and ¡

Hector ¡Garcia-­‑Molina, ¡2003) ¡

– Computes ¡a ¡global ¡trust ¡value ¡of ¡each ¡peer ¡based ¡

  • n ¡the ¡local ¡trust ¡values ¡
  • Used ¡to ¡limit ¡malicious/inauthen5c ¡files ¡

– Defense ¡against ¡pollu5on ¡aXacks ¡

17 ¡

slide-18
SLIDE 18

AXacks ¡against ¡pollu5on ¡systems ¡

  • Whitewashing ¡
  • Sybil ¡aXacks ¡
  • Collusion ¡
  • Dishonest ¡feedback ¡
  • See ¡next ¡lecture… ¡
  • This ¡lecture: ¡how ¡reputa5on ¡helps ¡in ¡elici5ng ¡

effort ¡

18 ¡

slide-19
SLIDE 19

A ¡minimalist ¡P2P ¡model ¡

  • Large ¡number ¡of ¡peers ¡(players) ¡
  • Peer ¡i ¡has ¡type ¡θi ¡(~ ¡“generosity”) ¡
  • Ac5on ¡space: ¡contribute ¡or ¡free-­‑ride ¡
  • x: ¡frac5on ¡of ¡contribu5ng ¡peers ¡

à 1/x: ¡cost ¡of ¡contribu5ng ¡

  • Ra5onal ¡peer: ¡

– Contribute ¡if ¡θi ¡> ¡1/x ¡ ¡ – Free-­‑ride ¡otherwise ¡

19 ¡

slide-20
SLIDE 20

Contribu5ons ¡with ¡no ¡incen5ve ¡ mechanism ¡

  • Assume ¡uniform ¡distribu5on ¡of ¡types ¡

20 ¡

slide-21
SLIDE 21

Contribu5ons ¡with ¡no ¡incen5ve ¡ mechanism ¡(2) ¡

  • Equilibria ¡stability ¡

21 ¡

slide-22
SLIDE 22

Contribu5ons ¡with ¡no ¡incen5ve ¡ mechanism ¡(3) ¡

  • Equilibria ¡computa5on ¡

22 ¡

slide-23
SLIDE 23

Contribu5ons ¡with ¡no ¡incen5ve ¡ mechanism ¡(4) ¡

  • Result: ¡The ¡highest ¡stable ¡equilibrium ¡

contribu5on ¡level ¡x1 ¡increases ¡with ¡θm ¡and ¡ converges ¡to ¡one ¡as ¡goes ¡θm ¡to ¡infinity ¡but ¡ falls ¡to ¡zero ¡if ¡θm ¡< ¡4 ¡

  • Remark: ¡if ¡the ¡distribu5on ¡is ¡not ¡uniform: ¡the ¡

graphical ¡method ¡s5ll ¡applies ¡

23 ¡

slide-24
SLIDE 24

Overall ¡system ¡performance ¡

  • W ¡= ¡ax-­‑(1/x)x ¡= ¡ax-­‑1 ¡
  • Even ¡if ¡par5cipa5on ¡provides ¡high ¡benefits, ¡

the ¡system ¡may ¡collapse ¡ ¡

24 ¡

slide-25
SLIDE 25

Reputa5on ¡and ¡service ¡differen5a5on ¡ in ¡P2P ¡

  • Consider ¡a ¡reputa5on ¡system ¡that ¡can ¡catch ¡

free-­‑riders ¡with ¡probability ¡p ¡and ¡exclude ¡ them ¡

– Alterna5vely: ¡catch ¡all ¡free-­‑riders ¡and ¡give ¡them ¡ service ¡altered ¡by ¡(1-­‑p) ¡

  • Two ¡effects ¡

– Load ¡reduced, ¡hence ¡cost ¡reduced ¡ – Penalty ¡introduces ¡a ¡threat ¡

25 ¡

slide-26
SLIDE 26

Equilibrium ¡with ¡reputa5on ¡

  • Q: ¡individual ¡benefit ¡
  • R: ¡reduced ¡contribu5on ¡
  • T: ¡threat ¡

26 ¡

slide-27
SLIDE 27

Equilibrium ¡with ¡reputa5on ¡(2) ¡

27 ¡

slide-28
SLIDE 28

System ¡performance ¡with ¡reputa5on ¡

  • W ¡= ¡x(Q-­‑R)+(1-­‑x)(Q-­‑T) ¡= ¡(ax-­‑1)(x+(1-­‑x)(1-­‑p)) ¡
  • Trade-­‑off: ¡Penalty ¡on ¡free ¡riders ¡increases ¡x ¡but ¡

entails ¡social ¡cost ¡

  • If ¡p>1/a, ¡the ¡threat ¡is ¡larger ¡than ¡the ¡cost ¡

à ¡No ¡free ¡rider, ¡op5mal ¡system ¡performance ¡a-­‑1 ¡

28 ¡

slide-29
SLIDE 29

FOX ¡(Fair ¡Op5mal ¡eXchange) ¡

  • Theore5cal ¡approach ¡
  • Assumes ¡all ¡peer ¡are ¡homogeneous, ¡with ¡

capacity ¡to ¡serve ¡k ¡requests ¡in ¡parallel ¡and ¡ seek ¡to ¡minimize ¡comple5on ¡5me ¡

  • FOX: ¡distributed ¡synchronized ¡protocol ¡giving ¡

the ¡op5mum ¡

– i.e., ¡all ¡peers ¡can ¡achieve ¡op5mum ¡if ¡they ¡comply ¡

  • “grim ¡trigger” ¡strategy: ¡each ¡peer ¡can ¡collapse ¡

the ¡system ¡if ¡he ¡finds ¡a ¡devia5ng ¡neighbor ¡

29 ¡

slide-30
SLIDE 30

FOX ¡equilibrium ¡

30 ¡

slide-31
SLIDE 31

Outline ¡

  • 1. Introduc5on ¡
  • 2. The ¡P2P ¡file ¡sharing ¡game ¡
  • 3. Free-­‑riding ¡and ¡incen5ves ¡for ¡contribu5on ¡
  • 4. Hidden ¡ac5ons: ¡the ¡principal-­‑agent ¡model ¡

31 ¡

slide-32
SLIDE 32

Hidden ¡ac5ons ¡

  • In ¡P2P, ¡many ¡strategic ¡ac5ons ¡are ¡not ¡directly ¡
  • bservable ¡

– Arrival/departure ¡ – Message ¡forwarding ¡

  • Same ¡with ¡many ¡other ¡contexts ¡

– Packet ¡forwarding ¡in ¡ad-­‑hoc ¡networks ¡ – Worker’s ¡effort ¡

  • Moral ¡hazard: ¡situa5on ¡in ¡which ¡a ¡party ¡is ¡more ¡

willing ¡to ¡take ¡a ¡risk ¡knowing ¡that ¡the ¡cost ¡will ¡be ¡ supported ¡(at ¡least ¡in ¡part) ¡by ¡others ¡

– E.g., ¡insurance ¡

32 ¡

slide-33
SLIDE 33

Principal-­‑agent ¡model ¡

  • A ¡principal ¡employs ¡a ¡set ¡of ¡n ¡agents: ¡N ¡= ¡{1, ¡…, ¡n} ¡ ¡
  • Ac5on ¡set ¡Ai ¡= ¡{0, ¡1} ¡
  • Cost ¡c(0)=0, ¡c(1)=c>0 ¡
  • The ¡ac5ons ¡of ¡agents ¡determine ¡(probabilis5cally) ¡an ¡
  • utcome ¡o ¡in ¡{0, ¡1} ¡
  • Principal ¡valua5on ¡of ¡success: ¡v>0 ¡(no ¡gain ¡in ¡case ¡of ¡

failure) ¡

  • Technology ¡(or ¡success ¡func5on) ¡t(a1, ¡…, ¡an): ¡probability ¡of ¡

success ¡

  • Remark: ¡many ¡different ¡models ¡exist ¡

– One ¡agent, ¡different ¡ac5on ¡sets ¡ – Etc. ¡ ¡

33 ¡

slide-34
SLIDE 34

Read-­‑once ¡networks ¡

  • One ¡graph ¡with ¡2 ¡special ¡nodes: ¡source ¡and ¡sink ¡
  • Each ¡agent ¡controls ¡1 ¡link ¡
  • Agents ¡ac5on: ¡ ¡

– low ¡effort ¡à ¡succeed ¡with ¡probability ¡γ ¡in ¡(0, ¡1/2) ¡ – High ¡effort ¡à ¡succeed ¡with ¡probability ¡1-­‑γ ¡in ¡(1/2, ¡1) ¡

  • The ¡project ¡succeeds ¡if ¡there ¡is ¡a ¡successful ¡

source-­‑sink ¡path ¡

34 ¡

slide-35
SLIDE 35

Example ¡

  • AND ¡technology ¡
  • OR ¡technology ¡

35 ¡

slide-36
SLIDE 36

Contract ¡

  • The ¡principal ¡agent ¡can ¡design ¡a ¡“contract” ¡

– Payment ¡of ¡pi≥0 ¡upon ¡success ¡ – Nothing ¡upon ¡failure ¡

  • The ¡agents ¡are ¡in ¡a ¡game: ¡ ¡
  • The ¡principal ¡wants ¡to ¡design ¡a ¡contract ¡such ¡

that ¡his ¡expected ¡profit ¡is ¡maximized ¡

ui(a) = pit(a)−c(ai) u(a,v) = t(a)⋅ v − pi

i∈N

% & ' ( ) *

36 ¡

slide-37
SLIDE 37

Defini5ons ¡and ¡assump5ons ¡

  • Assump5ons: ¡ ¡

– t(1, ¡a-­‑i)>t(0, ¡a-­‑i) ¡for ¡all ¡a-­‑i ¡ – t(a)>0 ¡for ¡all ¡a ¡

  • Defini5on: ¡the ¡marginal ¡contribu5on ¡of ¡agent ¡i ¡

given ¡a-­‑i ¡is ¡ ¡

  • Increase ¡in ¡success ¡probability ¡due ¡to ¡i’s ¡effort ¡

Δi(a−i) = t(1,a−i)−t(0,a−i)

37 ¡

slide-38
SLIDE 38

Individual ¡best ¡response ¡

  • Given ¡a-­‑i, ¡agent’s ¡i ¡best ¡strategy ¡is ¡

ai =1 if pi ≥ c Δi(a−i) ai = 0 if pi ≤ c Δi(a−i)

38 ¡

slide-39
SLIDE 39

Best ¡contract ¡inducing ¡a ¡

  • The ¡best ¡contract ¡for ¡the ¡principal ¡that ¡

induces ¡a ¡as ¡an ¡equilibrium ¡consists ¡in ¡ ¡

– ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡for ¡the ¡agents ¡choosing ¡ai=0 ¡ – ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡for ¡the ¡agents ¡choosing ¡ai=1 ¡

pi = c Δi(a−i)

pi = 0

39 ¡

slide-40
SLIDE 40

Best ¡contract ¡inducing ¡a ¡(2) ¡

  • With ¡this ¡best ¡contract, ¡expected ¡u5li5es ¡are ¡

– ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡for ¡the ¡agents ¡choosing ¡ai=0 ¡ – ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡for ¡the ¡agents ¡choosing ¡ai=1 ¡ – ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡for ¡the ¡principal ¡

ui = c⋅ t(1,a−i) Δi(a−i) −1 $ % & ' ( )

ui = 0

u(a,v) = t(a)⋅ v − c Δi(a−i)

i:ai=1

% & ' ' ( ) * *

40 ¡

slide-41
SLIDE 41

Principal’s ¡objec5ve ¡

  • Choosing ¡the ¡ac5ons ¡profile ¡a* ¡that ¡maximizes ¡

his ¡u5lity ¡u(a,v) ¡

  • Equivalent ¡to ¡choosing ¡the ¡set ¡S* ¡of ¡agents ¡

with ¡ai=1 ¡

  • Depends ¡on ¡v ¡à ¡S*(v) ¡
  • We ¡say ¡that ¡the ¡principal ¡contracts ¡with ¡i ¡if ¡

ai=1 ¡

41 ¡

slide-42
SLIDE 42

Hidden ¡vs ¡observable ¡ac5ons ¡

  • Hidden ¡ac5ons: ¡ ¡

¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡if ¡ai=1 ¡and ¡0 ¡otherwise ¡

  • If ¡ac5ons ¡were ¡observable ¡

– Give ¡pi=c ¡to ¡high-­‑effort ¡agents ¡regardless ¡of ¡success ¡ – Yields ¡for ¡the ¡principal ¡a ¡u5lity ¡equal ¡to ¡social ¡welfare ¡ à ¡Choose ¡a ¡to ¡maximize ¡social ¡welfare ¡

u(a,v) = t(a)⋅ v − c Δi(a−i)

i:ai=1

% & ' ' ( ) * * ui = c⋅ t(1,a−i) Δi(a−i) −1 $ % & ' ( ) u(a,v) = t(a)⋅v − c

i:ai=1

42 ¡

slide-43
SLIDE 43

(POU) ¡Price ¡of ¡Unaccountability ¡

  • S*(v): ¡op5mal ¡contract ¡in ¡hidden ¡case ¡
  • S0

*(v): ¡op5mal ¡contract ¡in ¡observable ¡case ¡

  • Defini5on: ¡the ¡POU(t) ¡of ¡a ¡technology ¡t ¡is ¡

defined ¡as ¡the ¡worst-­‑case ¡ra5o ¡over ¡v ¡of ¡the ¡ principal’s ¡u5lity ¡in ¡the ¡observable ¡and ¡hidden ¡ ac5ons ¡cases ¡

POU(t) = sup

v>0

t(S0

*(v))⋅v −

c

i∈S0

*(v)

t(S*(v))⋅ v − c t(S*(v))−t(S*(v) \ {i})

i∈S*(v)

% & ' ( ) *

43 ¡

slide-44
SLIDE 44

Remark ¡

  • POU(t)>1 ¡

44 ¡

slide-45
SLIDE 45

Op5mal ¡contract ¡

  • We ¡want ¡to ¡answer ¡the ¡ques5ons: ¡
  • How ¡to ¡select ¡the ¡op5mal ¡contract ¡(i.e., ¡the ¡
  • p5mal ¡set ¡of ¡contrac5ng ¡agents)? ¡
  • How ¡does ¡it ¡change ¡with ¡the ¡principal’s ¡

valua5on ¡v? ¡

45 ¡

slide-46
SLIDE 46

Monotonicity ¡

  • The ¡op5mal ¡contracts ¡weakly ¡improves ¡when ¡

v ¡increases: ¡ ¡

– For ¡any ¡technology, ¡in ¡both ¡the ¡hidden-­‑ ¡and ¡

  • bservable-­‑ac5ons ¡ ¡cases, ¡the ¡expected ¡u5lity ¡of ¡

the ¡principal, ¡the ¡success ¡probability ¡and ¡the ¡ expected ¡payment ¡of ¡the ¡op5mal ¡contract ¡are ¡all ¡ non-­‑decreasing ¡when ¡v ¡increases ¡

46 ¡

slide-47
SLIDE 47

Proof ¡

47 ¡

slide-48
SLIDE 48

Proof ¡(2) ¡

48 ¡

slide-49
SLIDE 49

Consequences ¡

  • Anonymous ¡technology: ¡the ¡success ¡

probability ¡is ¡symmetric ¡in ¡the ¡players ¡

  • For ¡technologies ¡for ¡which ¡the ¡success ¡

probability ¡depends ¡only ¡on ¡the ¡number ¡of ¡ contracted ¡agents ¡(e.g. ¡AND, ¡OR), ¡the ¡number ¡

  • f ¡contracted ¡agents ¡is ¡non-­‑decreasing ¡when ¡v ¡

increases ¡

49 ¡

slide-50
SLIDE 50

Op5mal ¡contract ¡for ¡the ¡AND ¡ technology ¡

  • Theorem: ¡For ¡any ¡anonymous ¡AND ¡

technology ¡with ¡γ ¡= ¡γi ¡= ¡1-­‑δi ¡for ¡all ¡i ¡

– There ¡exists ¡a ¡valua5on ¡finite ¡v* ¡such ¡that ¡for ¡any ¡ v<v*, ¡it ¡is ¡op5mal ¡to ¡contract ¡with ¡no ¡agent ¡and ¡ for ¡any ¡v>v*, ¡it ¡is ¡op5mal ¡to ¡contract ¡with ¡all ¡ agents ¡(for ¡v=v*, ¡both ¡contracts ¡are ¡op5mal) ¡ – The ¡price ¡of ¡unaccountability ¡is ¡ ¡

POU = 1 γ −1 " # $ % & '

n−1

+ 1− γ 1−γ " # $ % & '

50 ¡

slide-51
SLIDE 51

Remarks ¡

  • Proof ¡in ¡M. ¡Babaioff, ¡M. ¡Feldman ¡and ¡N. ¡

Nisan, ¡“Combinatorial ¡Agency”, ¡in ¡Proceedings ¡

  • f ¡EC ¡2006. ¡
  • POU ¡is ¡not ¡bounded! ¡

– Monitoring ¡can ¡be ¡beneficial, ¡even ¡if ¡costly ¡

51 ¡

slide-52
SLIDE 52

Example ¡

  • n=2, ¡c=1, ¡γ=1/4 ¡
  • Compute ¡for ¡all ¡number ¡of ¡agents ¡

– t ¡ – Δ ¡ – U5lity ¡of ¡principal ¡

52 ¡

slide-53
SLIDE 53

Op5mal ¡contract ¡for ¡the ¡OR ¡ technology ¡

  • Theorem: ¡For ¡any ¡anonymous ¡OR ¡technology ¡

with ¡γ ¡= ¡γi ¡= ¡1-­‑δi ¡for ¡all ¡i ¡

– There ¡exists ¡a ¡finite ¡posi5ve ¡values ¡v1, ¡…, ¡vn ¡such ¡ that ¡for ¡any ¡v ¡in ¡(vk, ¡vk+1), ¡it ¡is ¡op5mal ¡to ¡contract ¡k ¡

  • agent. ¡(For ¡v<v0, ¡it ¡is ¡op5mal ¡to ¡contract ¡0 ¡agent, ¡

for ¡v>vn, ¡it ¡is ¡op5mal ¡to ¡contract ¡n ¡agent ¡and ¡for ¡ v=vk, ¡the ¡principal ¡is ¡indifferent ¡between ¡ contrac5ng ¡k-­‑1 ¡or ¡k ¡agents.) ¡ – The ¡price ¡of ¡unaccountability ¡is ¡upper ¡bounded ¡by ¡ 5/2 ¡

53 ¡

slide-54
SLIDE 54

Example ¡

  • n=2, ¡c=1, ¡γ=1/4 ¡
  • Compute ¡for ¡all ¡number ¡of ¡agents ¡

– t ¡ – Δ ¡ – U5lity ¡of ¡principal ¡

54 ¡

slide-55
SLIDE 55

Illustra5on ¡

  • Number ¡of ¡contracted ¡agents ¡

v

3

gamma 200 150 0.4 100 50 0.3 0.2 0.1

2 1 3

12000 6000 8000 4000 2000 gamma 0.4 0.45 10000 0.3 0.35 0.25 0.2

55 ¡