csc304 lecture 9
play

CSC304 Lecture 9 Mechanism Design w/ Money: More examples of VCG, - PowerPoint PPT Presentation

CSC304 Lecture 9 Mechanism Design w/ Money: More examples of VCG, winner determination and truthful approximation CSC304 - Nisarg Shah 1 VCG Recap = = argmax ()


  1. CSC304 Lecture 9 Mechanism Design w/ Money: More examples of VCG, winner determination and truthful approximation CSC304 - Nisarg Shah 1

  2. VCG Recap β€’ 𝑔 𝑀 = 𝑏 βˆ— = argmax π‘βˆˆπ΅ Οƒ 𝑗 𝑀 𝑗 (𝑏) βˆ’ Οƒ π‘˜β‰ π‘— 𝑀 π‘˜ 𝑏 βˆ— β€’ π‘ž 𝑗 𝑀 = max Οƒ π‘˜β‰ π‘— 𝑀 π‘˜ 𝑏 𝑏 β€’ Procedure ➒ Step 1: Choose the allocation to maximize social welfare ➒ Step 2: Payment charged to each agent 𝑗 is the externality that 𝑗 imposes on others o [Max welfare of others | 𝑗 absent] – [welfare of others | 𝑗 present] Under 𝑏 βˆ— CSC304 - Nisarg Shah 2

  3. VCG Recap β€’ Four properties ➒ Maximize social welfare ➒ Dominant strategy incentive compatibility (DSIC) ➒ No payments to agents ➒ Individual rationality (IR) β€’ Vickrey auction satisfies the first two β€’ VCG adds Clarke’s pivot rule to satisfy all four CSC304 - Nisarg Shah 3

  4. VCG Example β€’ In the last lecture, we saw… ➒ Additive valuations: agent has value 𝑀 𝑗 𝑏 for each 𝑏 , 𝑀 𝑗 𝑇 = Οƒ π‘βˆˆπ‘‡ 𝑀 𝑗 𝑏 ➒ Unit-demand valuations: Still have 𝑀 𝑗 𝑏 for each 𝑏 , 𝑀 𝑗 𝑇 = max π‘βˆˆπ‘‡ 𝑀 𝑗 𝑏 o Goods are β€œsubstitutes” β€’ Another example… ➒ Complementary goods: value of the whole exceeds the sum of values of its parts CSC304 - Nisarg Shah 4

  5. VCG Example β€’ A chair ( 𝑑 ) and a table ( 𝑒 ) 𝑀 1 𝑑 = 3 𝑀 2 𝑒 = 4 β€’ Allocation? β€’ Payment? 𝑀 3 {𝑑, 𝑒} = 6 CSC304 - Nisarg Shah 5

  6. VCG Example β€’ A chair ( 𝑑 ) and a table ( 𝑒 ) 𝑀 1 𝑑 = 3 𝑀 2 𝑒 = 4 β€’ Allocation? β€’ Payment? 𝑀 3 {𝑑, 𝑒} = 8 CSC304 - Nisarg Shah 6

  7. VCG Example: Seller as Agent β€’ Seller ( 𝑇 ) wants to sell his car ( 𝑑 ) to buyer ( 𝐢 ) β€’ Seller has a value for his own car: 𝑀 𝑇 𝑑 ➒ Individual rationality for the seller mandates that seller must get revenue at least 𝑀 𝑇 𝑑 β€’ Idea: Add seller as another agent, and make his values part of the welfare calculations! CSC304 - Nisarg Shah 7

  8. VCG Example: Seller as Agent 𝑀 𝑇 𝑑 = 3 𝑀 𝐢 𝑑 = 5 β€’ What if… ➒ We give the car to buyer when 𝑀 𝐢 𝑑 > 𝑀 𝑇 (𝑑) and ➒ Buyer pays seller 𝑀 𝐢 𝑑 : Not DSIC for buyer! ➒ Buyer pays seller 𝑀 𝑇 (𝑑) : Not DSIC for seller! CSC304 - Nisarg Shah 8

  9. VCG Example: Seller as Agent 𝑀 𝑇 𝑑 = 3 𝑀 𝐢 𝑑 = 5 β€’ Allocation? Mechanism takes $3 from buyer, and gives ➒ Buyer gets the car (welfare = 5 ) $5 to the seller! β€’ Payment? β€’ Need external subsidy ➒ Buyer pays: 3 βˆ’ 0 = 3 ➒ Seller pays: 0 βˆ’ 5 = βˆ’5 CSC304 - Nisarg Shah 9

  10. Problems with VCG β€’ Difficult to understand in complex settings ➒ Need to reason about what allocation would maximize welfare if agent 𝑗 was absent β€’ Only cares about welfare, not revenue ➒ Though, as we will see in a few lectures, gets pretty good revenue β€’ With sellers and buyers, need external subsidy ➒ Actually, cannot get individual rationality, DSIC, no subsidy, and constant approximation of welfare β€’ Might be computationally difficult to implement ➒ Computing welfare maximizing allocation may be hard CSC304 - Nisarg Shah 10

  11. Single-Minded Bidders β€’ Combinatorial auction for a set of 𝑛 items 𝑇 β€’ Each agent 𝑗 has ➒ Value 𝑀 𝑗 if receives a subset 𝑇 𝑗 βŠ† 𝑇 ➒ Value 0 if doesn’t get a superset of 𝑇 𝑗 ➒ β€œSingle - minded” β€’ Welfare-maximizing allocation: ➒ Find a subset of players 𝑗 with the highest total value such that their sets 𝑇 𝑗 are disjoint CSC304 - Nisarg Shah 11

  12. Single-Minded Bidders β€’ Reduction to the Weighted Independent Set (WIS) problem in a graph ➒ NP-hard to find the welfare-maximizing allocation ➒ Note: not even thinking about computing payments yet ➒ In fact, hard to approximately optimize welfare 1 2 βˆ’πœ— ) approximation (unless 𝑂𝑄 βŠ† π‘Žπ‘„π‘„ ) o No O(𝑛 β€’ Luckily, a simple greedy algorithm gives 𝑛 -approximation (i.e., OPT/GREEDY ≀ 𝑛 ) CSC304 - Nisarg Shah 12

  13. Greedy Algorithm β€’ Input: (𝑀 𝑗 , 𝑇 𝑗 ) for each agent 𝑗 β€’ Output: Agents with mutually independent 𝑇 𝑗 β€’ Greedy Algorithm: ➒ Sort the agents. Go over them one-by-one. Accept each bid if no requested item is previously allocated. β€’ Sort by what? ➒ 𝑀 1 β‰₯ 𝑀 2 β‰₯ β‹― β‰₯ 𝑀 π‘œ ? 𝑛 -approximation 𝑀 1 𝑀 2 𝑀 π‘œ 𝑇 1 β‰₯ 𝑇 2 β‰₯ β‹― 𝑇 π‘œ ? 𝑛 -approximation ➒ 𝑀 1 𝑀 2 𝑀 π‘œ 𝑇 1 β‰₯ 𝑇 2 β‰₯ β‹― 𝑇 π‘œ ? 𝑛 -approximation [Lehmann et al. 2011] ➒ CSC304 - Nisarg Shah 13

  14. Greedy Algorithm β€’ (allocation rule, payments) truthful if and only if ➒ Allocation is monotonic: If agent 𝑗 wins with (𝑀 𝑗 , 𝑇 𝑗 ) , it β€² β‰₯ 𝑀 𝑗 and 𝑇 𝑗 β€² βŠ† 𝑇 𝑗 β€² , 𝑇 𝑗 β€² ) where 𝑀 𝑗 must win with (𝑀 𝑗 ➒ Payments are critical prices: Agent 𝑗 pays the least value (s)he could have reported and still won. |𝑇 𝑗 | β€’ π‘ž 𝑗 = 𝑀 π‘˜ βˆ— β‹… 𝑇 π‘˜βˆ— ➒ π‘˜ βˆ— is the smallest index π‘˜ such that 𝑇 π‘˜ ∩ 𝑇 𝑗 β‰  βˆ… and 𝑇 π‘˜ ∩ 𝑇 𝑙 = βˆ… for all 𝑙 < π‘˜, 𝑙 β‰  𝑗 ➒ If agent 𝑗 reports less than this value, agent π‘˜ gets 𝑇 π‘˜ first, and 𝑗 loses. CSC304 - Nisarg Shah 14

  15. Moral β€’ VCG can sometimes be too difficult to implement ➒ May look into approximately maximizing welfare ➒ Can set the payments right if the allocation rule is monotone β€’ Need for approximation is due to computational considerations β€’ Later in mechanism design without money… ➒ Can’t use payments to ensure truthfulness ➒ Will need to approximate welfare just to get truthfulness, even without computational considerations CSC304 - Nisarg Shah 15

  16. Sponsored Search Auctions CSC304 - Nisarg Shah 16

  17. Sponsored Search Auctions β€’ Suppose the search engine receives a search query β€’ 𝑙 advertisement slots ➒ β€œ Clickthrough rates” : 𝑑 1 β‰₯ 𝑑 2 β‰₯ β‹― β‰₯ 𝑑 𝑙 β‰₯ 𝑑 𝑙+1 = 0 β€’ π‘œ advertisers (bidders) For convenience ➒ Bidder 𝑗 derives value 𝑀 𝑗 *per click* ➒ Final value to bidder 𝑗 for receiving slot π‘˜ = 𝑀 𝑗 β‹… 𝑑 π‘˜ ➒ Without loss of generality, 𝑀 1 β‰₯ 𝑀 2 β‰₯ β‹― β‰₯ 𝑀 π‘œ β€’ Age-old question: ➒ Who gets which slot, and how much should they pay? CSC304 - Nisarg Shah 17

  18. Sponsored Search : VCG β€’ VCG ➒ Maximize welfare: π‘˜ th bidder gets π‘˜ th slot ( 1 ≀ π‘˜ ≀ 𝑙 ) ➒ Payment of π‘˜ th bidder? β€’ Increase in social welfare to others if π‘˜ abstains ➒ Bidders π‘˜ + 1 through 𝑙 + 1 get β€œupgraded” by one slot 𝑙+1 ➒ Payment of bidder π‘˜ = Οƒ 𝑗=π‘˜+1 𝑀 𝑗 β‹… (𝑑 π‘—βˆ’1 βˆ’ 𝑑 𝑗 ) 𝑑 π‘—βˆ’1 βˆ’π‘‘ 𝑗 𝑙+1 ➒ Payment to bidder π‘˜ β€œper click” = Οƒ 𝑗=π‘˜+1 𝑀 𝑗 β‹… 𝑑 π‘˜ ➒ Not very intuitive… CSC304 - Nisarg Shah 18

  19. Sponsored Search : VCG β€’ What happens if all clickthrough rates are same? ➒ 𝑑 1 = 𝑑 2 = β‹― = 𝑑 𝑙 > 𝑑 𝑙+1 = 0 β€’ Payment of bidder π‘˜ per click 𝑑 π‘—βˆ’1 βˆ’π‘‘ 𝑗 𝑙+1 ➒ Οƒ 𝑗=π‘˜+1 𝑀 𝑗 β‹… = 𝑀 𝑙+1 𝑑 π‘˜ β€’ Bidders 1 through 𝑙 pay the value of bidder 𝑙 + 1 ➒ Familiar? VCG for 𝑙 identical items CSC304 - Nisarg Shah 19

  20. Sponsored Search : GSP β€’ Generalized Second Price Auction (GSP) ➒ For 1 ≀ π‘˜ ≀ 𝑙 ➒ Bidder π‘˜ gets slot π‘˜ ➒ Bidder π‘˜ pays the bid of bidder π‘˜ + 1 β€’ A natural extension of the second price auction ➒ We already saw that this is not truthful even with two identical slots ➒ Highest bidder paying 2 nd highest bid β†’ wants to lower bid to become 2 nd highest bidder and pay 3 rd highest bid CSC304 - Nisarg Shah 20

  21. Sponsored Search : GSP β€’ Truth-telling is not a Nash equilibrium  β€’ But there is a good Nash equilibrium that realizes the VCG outcome, i.e., maximizes welfare and generates as much revenue as VCG ☺ [Edelman et al. 2007] β€’ Even the worst Nash equilibrium gives 1.282 - approximation to welfare ( 𝑄𝑝𝐡 ≀ 1.282 ) and generates at least half the revenue of VCG [Caragiannis et al. 2011, Dutting et al. 2011, Lucier et al. 2012] CSC304 - Nisarg Shah 21

  22. VCG vs GSP β€’ VCG ➒ Truthful in dominant strategy β†’ more confidence that players will bid truthfully ➒ Theoretical welfare/revenue guarantees will hold ➒ Though players might still misreport… ➒ Difficult to understand β€’ GSP ➒ Need to rely on players reaching a Nash equilibrium ➒ Good welfare and revenue ➒ Easy to understand CSC304 - Nisarg Shah 22

  23. VCG vs GSP β€’ Google uses GSP β€’ Facebook used GSP, but switched to VCG ➒ They argue that maximizing welfare has two benefits ➒ Advertisers are happy β†’ attract more advertisers β†’ more long-term revenue ➒ Users are happy (?!) β†’ users use FB more β†’ more slots to sell β†’ more long-term revenue β€’ No consensus CSC304 - Nisarg Shah 23

  24. Sponsored Search Reality β€’ Value is proportional to clickthrough rate ➒ Could it be that users clicking on the 2 nd slot are more likely buyers than those clicking on the 1 st slot? β€’ Ad engines also want to produce quality results ➒ An advertiser having a high value for a slot does not necessarily mean his ad is appropriate for the slot β€’ Theoretical analysis does not take into account market competition ➒ Advertiser divide their budget among ad engines CSC304 - Nisarg Shah 24

Recommend


More recommend