csc304 lecture 11
play

CSC304 Lecture 11 Mechanism Design w/ Money: Revelation principle; - PowerPoint PPT Presentation

CSC304 Lecture 11 Mechanism Design w/ Money: Revelation principle; First price, second price, and ascending auctions; Revenue equivalence CSC304 - Nisarg Shah 1 Recap: Bayesian Framework


  1. CSC304 Lecture 11 Mechanism Design w/ Money: Revelation principle; First price, second price, and ascending auctions; Revenue equivalence CSC304 - Nisarg Shah 1

  2. Recap: Bayesian Framework โ‹ฏ โ‹ฏ ๐‘ฌ ๐Ÿ ๐‘ฌ ๐’‹ ๐‘ฌ ๐’ โ‹ฏ โ‹ฏ ๐’˜ ๐Ÿ ๐’˜ ๐’‹ ๐’˜ ๐’ All distributions Private value of ๐‘— ๐’• ๐Ÿ ๐’• ๐’‹ ๐’• ๐’ known to all only known to ๐‘— โ‹ฏ โ‹ฏ agents ๐’„ ๐Ÿ ๐’„ ๐’‹ ๐’„ ๐’ All strategies known to all agents CSC304 - Nisarg Shah 2

  3. Recap: Bayesian Framework โ€ข Strategy profile ิฆ ๐‘ก = (๐‘ก 1 , โ€ฆ , ๐‘ก ๐‘œ ) โžข Interim utility of agent ๐‘— is ๐น ๐‘ค ๐‘˜ โˆผ๐ธ ๐‘˜ ๐‘˜โ‰ ๐‘— ๐‘ฃ ๐‘— ๐‘ก 1 ๐‘ค 1 , โ€ฆ , ๐‘ก ๐‘œ ๐‘ค ๐‘œ where utility ๐‘ฃ ๐‘— is โ€œvalue derived โ€“ payment chargedโ€ โžข ิฆ ๐‘ก is a Bayes-Nash equilibrium (BNE) if ๐‘ก ๐‘— is the best strategy for agent ๐‘— given ิฆ ๐‘ก โˆ’๐‘— (strategies of others) o NOTE: I donโ€™t know what othersโ€™ values are. But I know they are rational players, so I can reason about what strategies they might use. CSC304 - Nisarg Shah 3

  4. Recap: 1 st Price Auction โ€ข Sealed-bid first price auction for a single item โžข Each agent ๐‘— privately submits a bid ๐‘ ๐‘— โžข Agent ๐‘— โˆ— with the highest bid wins the item, pays ๐‘ ๐‘— โˆ— โ€ข Suppose there are two agents โžข Common prior: each has valuation drawn from ๐‘‰[0,1] โ€ข Claim: Both players using ๐‘ก ๐‘— ๐‘ค ๐‘— = ๐‘ค ๐‘— /2 is a BNE. โžข Proof on the board. CSC304 - Nisarg Shah 4

  5. Direct Revelation Mechanisms & The Revelation Principle CSC304 - Nisarg Shah 5

  6. Direct Revelation โ€ข Direct-revelation: mechanisms that ask you to report your private values โžข Doesnโ€™t mean agents will report their true values. โžข Makes sense to ask โ€œWould they, in equilibrium?โ€ โ€ข Non-direct-revelation: different action space than type space โžข Suppose your value for an item is in [0,1], but the mechanism asks you to either dive left or dive right. โžข Strategy ๐‘ก ๐‘— : 0,1 โ†’ {๐‘š๐‘“๐‘”๐‘ข, ๐‘ ๐‘—๐‘•โ„Ž๐‘ข} โžข Truthfulness doesnโ€™t make much sense. โžข But we can still ask: What is the outcome in equilibrium? CSC304 - Nisarg Shah 6

  7. BNIC Mechanisms โ€ข A direct revelation mechanism is Bayes-Nash incentive compatible (BNIC) if all players playing ๐‘ก ๐‘— ๐‘ค ๐‘— = ๐‘ค ๐‘— is a BNE. โžข I donโ€™t know what otherโ€™s valuations are, only the distributions theyโ€™re drawn from. โžข But as long as they report their true values, in expectation I would like to report my true value. โ€ข Compare to strategyproofness โžข I know what othersโ€™ values are, and for every possible values they can have, I want to report my true values. CSC304 - Nisarg Shah 7

  8. Revelation Principle โ€ข Outcome = (allocation, payments) โ€ข Strategyproof version [Gibbard , โ€˜73] โžข If a mechanism implements an outcome in dominant strategies, thereโ€™s a direct revelation strategyproof mechanism implementing the same outcome. โ€ข BNIC version [Dasgupta et al. โ€˜79, Holmstrom โ€˜77, Myerson โ€™79] โžข If a mechanism implements an outcome as BNE, thereโ€™s a direct revelation BNIC mechanism implementing the same outcome. CSC304 - Nisarg Shah 8

  9. Revelation Principle โ€ข Informal proof: Player 1 : ๐‘ค 1 Strategy s 1 โ‹ฎ โ‹ฎ Original Outcome Mechanism Strategy s ๐‘œ Player ๐‘œ : ๐‘ค ๐‘œ New direct revelation truthful mechanism CSC304 - Nisarg Shah 9

  10. Applying Revelation Principle โ€ข We already sawโ€ฆ โžข Sealed-bid 1 st price auction โžข 2 agents with valuations drawn from ๐‘‰[0,1] โžข Each player halving his value was a BNE โžข Not naturally BNIC (players donโ€™t report value) โ€ข Q: What is the BNIC variant of sealed-bid 1 st price auction that we get using the revelation principle? โ€ข Can also be used on non-direct-revelation mechs CSC304 - Nisarg Shah 10

  11. Revenue of Auction Mechanisms & Revenue Equivalence CSC304 - Nisarg Shah 11

  12. 1 st Price Auction โ€ข For ๐‘œ players with iid valuations from U[0,1], โ€œshadowingโ€ the bid by a factor of (๐‘œ โˆ’ 1)/๐‘œ is a BNE โ€ข ๐น[Revenue] to the auctioneer? ๐‘œโˆ’1 ๐‘œโˆ’1 โžข ๐น ๐‘ค ๐‘— โˆผ๐‘‰ 0,1 โˆ— max ๐‘ค ๐‘— = (Exercise!) ๐‘œ ๐‘œ ๐‘œ+1 ๐‘—=1 ๐‘— โ€ข Interestingly, this is equal to E[Revenue] from 2 nd price auction ๐‘œโˆ’1 ๐‘œ [2 nd highest ๐‘ค ๐‘— ] = โžข ๐น ๐‘ค ๐‘— โˆผ๐‘‰ 0,1 (Exercise!) ๐‘œ+1 ๐‘—=1 CSC304 - Nisarg Shah 12

  13. Revenue Equivalence โ€ข If two BNIC mechanisms A and B: 1. Always produce the same allocation; 2. Have the same expected payment to agent ๐‘— for some 0 (e.g., โ€œzero value for allโ€ โ†’ zero payment); type ๐‘ค ๐‘— 3. Have agent valuations drawn from distributions with โ€œpath - connected support setsโ€; โ€ข Then they: โžข Charge the same expected payment to all agent types; โžข Have the same expected total revenue. CSC304 - Nisarg Shah 13

  14. Revenue Equivalence โ€ข Informallyโ€ฆ โžข If two BNIC mechanisms always have the same allocation, then they have the same E[payments] and E[revenue]. โžข Very powerful as it applies to any pair of BNIC mechanism โ€ข 1 st price (BNIC variant) and 2 nd price auctions โžข Have the same allocation: Item always goes to the agent with the highest valuation โžข Thus, also have the same revenue CSC304 - Nisarg Shah 14

  15. Non-Direct-Revelation Auctions โ€ข Ascending auction (a.k.a. English auction) โžข All agents + auctioneer meet in a room. โžข Auctioneer starts the price at 0 . โžข All agents want the item, and have their hands raised. โžข Auctioneer raise the price continuously. โžข Agents drop out when price > value for them โ€ข Descending auction (a.k.a. Dutch auction) โžข Start price at a very high value. โžข Keep decreasing the price until some agent agrees to buy. CSC304 - Nisarg Shah 15

Recommend


More recommend