csc304 lecture 12
play

CSC304 Lecture 12 Ending Mechanism Design w/ Money: Recap revenue - PowerPoint PPT Presentation

CSC304 Lecture 12 Ending Mechanism Design w/ Money: Recap revenue maximization & Myersons auction Begin Mechanism Design w/o Money: Facility Location CSC304 - Nisarg Shah 1 Recap Single-item auction with 1 seller, buyers


  1. CSC304 Lecture 12 Ending Mechanism Design w/ Money: Recap revenue maximization & Myerson’s auction Begin Mechanism Design w/o Money: Facility Location CSC304 - Nisarg Shah 1

  2. Recap β€’ Single-item auction with 1 seller, π‘œ buyers β€’ Buyer 𝑗 has value 𝑀 𝑗 drawn from cdf 𝐺 𝑗 (pdf 𝑔 𝑗 ) 1βˆ’πΊ 𝑗 (𝑀 𝑗 ) β€’ Virtual value function: πœ’ 𝑗 𝑀 𝑗 = 𝑀 𝑗 βˆ’ 𝑔 𝑗 (𝑀 𝑗 ) β€’ Myerson’s theorem: E[Revenue] = E Οƒ 𝑗 πœ’ 𝑗 𝑀 𝑗 βˆ— 𝑦 𝑗 ➒ Maximize revenue = maximize virtual welfare subject to monotonic allocation rule CSC304 - Nisarg Shah 2

  3. Recap β€’ When all 𝐺 𝑗 ’s are regular ➒ Monotonicity is automatic β€’ Allocation: Give to agent 𝑗 with maximum πœ’ 𝑗 (𝑀 𝑗 ) if πœ’ 𝑗 𝑀 𝑗 β‰₯ 0 ➒ When the maximum πœ’ 𝑗 𝑀 𝑗 is negative, not selling the item is better (zero virtual welfare > negative virtual welfare) β€’ Payment: Charge βˆ— = min 𝑀 𝑗 β€² ∢ πœ’ 𝑗 𝑀 𝑗 β€² β‰₯ max 0, πœ’ π‘˜ 𝑀 π‘˜ 𝑀 𝑗 βˆ€π‘˜ β‰  𝑗 ➒ Least possible value for which the agent still gets the item ➒ If virtual value drops below any other virtual value or below 0 , the agent loses the item CSC304 - Nisarg Shah 3

  4. Recap β€’ Special case: All 𝐺 𝑗 = 𝐺 = Regular ➒ VCG with reserve price πœ’ βˆ’1 (0) β€’ Allocation: Give the item to agent 𝑗 with the maximum value 𝑀 𝑗 but only if 𝑀 𝑗 β‰₯ πœ’ βˆ’1 (0) ➒ Equivalent to πœ’ 𝑀 𝑗 β‰₯ 0 β€’ Payment: max πœ’ βˆ’1 0 , max π‘˜β‰ π‘— 𝑀 π‘˜ ➒ Least possible value for which the agent still gets the item ➒ The agent loses the item as soon as his value goes below either the 2 nd highest bid or the reserve price CSC304 - Nisarg Shah 4

  5. Approx. Optimal Auctions β€’ When 𝐺 𝑗 ’s are complex, the virtual valuation function is complex too ➒ The optimal auction is unintuitive ➒ Two simple auctions that achieve good revenue β€’ Theorem [Hartline & Roughgarden, 2009]: For independent regular distributions, VCG with bidder-specific reserve prices can guarantee 50% of the optimal revenue. CSC304 - Nisarg Shah 5

  6. Approx. Optimal Auctions β€’ Still relies on knowing bidders’ distributions ➒ Can break down if the true distribution is different than the assumed distribution β€’ Theorem [Bulow and Klemperer, 1996]: For i.i.d. bidder valuations, 𝐹[ Revenue of VCG with π‘œ + 1 bidders ] β‰₯ 𝐹[ Optimal revenue with π‘œ bidders ] β€’ β€œSpend effort in getting one more bidder than in figuring out the optimal auction” CSC304 - Nisarg Shah 6

  7. Simple Proof β€’ (n+1)-bidder VCG has the maximum expected revenue among all (n+1)-bidder DSIC auctions that always allocate the item ➒ Revenue Equivalence Theorem β€’ Consider the following (n+1)-bidder DSIC auction ➒ Run π‘œ -bidder Myerson on first π‘œ bidders. If the item is unallocated, give it to agent π‘œ + 1 for free. ➒ As much expected revenue as π‘œ -bidder Myerson auction ➒ No more expected revenue than (n+1)-bidder VCG β€’ QED! CSC304 - Nisarg Shah 7

  8. Optimizing Revenue is Hard β€’ Beyond single-parameter settings, the optimal auctions become even trickier β€’ Example: Two items, a single bidder with i.i.d. values for both items ➒ Q: Shouldn’t the optimal auction just sell each item individually using Myerson’s auction? ➒ A: No! Putting a take-it-or-leave-it offer on the two items bundled together can increase revenue! β€’ Slow progress on optimal auctions, but fast progress on simple and approximately optimal auctions CSC304 - Nisarg Shah 8

  9. Mechanism Design Without Money CSC304 - Nisarg Shah 9

  10. Lack of Money β€’ Mechanism design with money: ➒ VCG can implement the welfare maximizing outcome because it can charge payments β€’ Mechanism design without money: ➒ Suppose you want to give away a single item, but cannot charge any payments ➒ Impossible to get meaningful information about valuations from strategic agents ➒ How would you maximize welfare as much as possible? CSC304 - Nisarg Shah 10

  11. Lack of Money β€’ One possibility: Give the item to each of π‘œ bidders with probability 1/π‘œ . β€’ Does not maximize welfare ➒ It’s impossible to maximize welfare without money β€’ Achieves an π‘œ -approximation of maximum welfare max 𝑗 𝑀 𝑗 ➒ max (1/π‘œ) Οƒ 𝑗 𝑀 𝑗 ≀ π‘œ (What is this?) 𝑀 β€’ Can’t do better than π‘œ -approximation CSC304 - Nisarg Shah 11

  12. MD w/o Money Theme 1. Define the problem: agents, outcomes, valuations 2. Define the goal (e.g., maximizing social welfare) 3. Check if the goal can be achieved using a strategyproof mechanism ➒ β€œ strategyproof ” = DSIC 4. If not, find the strategyproof mechanism that provides the best approximation ratio ➒ Approximation ratio is similar to price of anarchy (PoA) CSC304 - Nisarg Shah 12

  13. Facility Location β€’ Set of agents 𝑂 β€’ Each agent 𝑗 has a true location 𝑦 𝑗 ∈ ℝ β€’ Mechanism 𝑔 ➒ Takes as input reports ΰ·€ 𝑦 = (ΰ·€ 𝑦 1 , ΰ·€ 𝑦 2 , … , ΰ·€ 𝑦 π‘œ ) ➒ Returns a location 𝑧 ∈ ℝ for the new facility β€’ Cost to agent 𝑗 : 𝑑 𝑗 𝑧 = 𝑧 βˆ’ 𝑦 𝑗 β€’ Social cost 𝐷 𝑧 = Οƒ 𝑗 𝑑 𝑗 𝑧 = Οƒ 𝑗 𝑧 βˆ’ 𝑦 𝑗 CSC304 - Nisarg Shah 13

  14. Facility Location β€’ Social cost 𝐷 𝑧 = Οƒ 𝑗 𝑑 𝑗 𝑧 = Οƒ 𝑗 𝑧 βˆ’ 𝑦 𝑗 β€’ Q: Ignoring incentives, what choice of 𝑧 would minimize the social cost? β€’ A: The median location med(𝑦 1 , … , 𝑦 π‘œ ) ➒ π‘œ is odd β†’ the unique β€œ(n+1)/2” th smallest value ➒ π‘œ is even β†’ β€œn/2” th or β€œ(n/2)+1” st smallest value ➒ Why? CSC304 - Nisarg Shah 14

  15. Facility Location β€’ Social cost 𝐷 𝑧 = Οƒ 𝑗 𝑑 𝑗 𝑧 = Οƒ 𝑗 𝑧 βˆ’ 𝑦 𝑗 β€’ Median is optimal (i.e., 1 -approximation) β€’ What about incentives? ➒ Median is also strategyproof (SP)! ➒ Irrespective of the reports of other agents, agent 𝑗 is best off reporting 𝑦 𝑗 CSC304 - Nisarg Shah 15

  16. Median is SP No manipulation can help CSC304 - Nisarg Shah 16

  17. Max Cost β€’ A different objective function 𝐷 𝑧 = max 𝑗 𝑧 βˆ’ 𝑦 𝑗 β€’ Q: Again ignoring incentives, what value of 𝑧 minimizes the maximum cost? β€’ A: The midpoint of the leftmost ( min 𝑦 𝑗 ) and the 𝑗 rightmost ( max 𝑦 𝑗 ) locations (WHY?) 𝑗 β€’ Q: Is this optimal rule strategyproof? β€’ A: No! (WHY?) CSC304 - Nisarg Shah 17

  18. Max Cost β€’ 𝐷 𝑧 = max 𝑗 𝑧 βˆ’ 𝑦 𝑗 β€’ We want to use a strategyproof mechanism. β€’ Question: What is the approximation ratio of median for maximum cost? 1. ∈ 1,2 2. ∈ 2,3 3. ∈ 3,4 4. ∈ 4, ∞ CSC304 - Nisarg Shah 18

  19. Max Cost β€’ Answer: 2 -approximation β€’ Other SP mechanisms that are 2 -approximation ➒ Leftmost: Choose the leftmost reported location ➒ Rightmost: Choose the rightmost reported location ➒ Dictatorship: Choose the location reported by agent 1 ➒ … CSC304 - Nisarg Shah 19

  20. Max Cost β€’ Theorem [Procaccia & Tennenholtz , β€˜09] No deterministic SP mechanism has approximation ratio < 2 for maximum cost. β€’ Proof: CSC304 - Nisarg Shah 20

  21. Max Cost + Randomized β€’ The Left-Right-Middle (LRM) Mechanism ➒ Choose min 𝑦 𝑗 with probability ΒΌ 𝑗 ➒ Choose max 𝑦 𝑗 with probability ΒΌ 𝑗 ➒ Choose (min 𝑦 𝑗 + max 𝑦 𝑗 )/2 with probability Β½ 𝑗 𝑗 β€’ Question: What is the approximation ratio of LRM for maximum cost? (1/4)βˆ—2𝐷+(1/4)βˆ—2𝐷+(1/2)βˆ—π· 3 = β€’ At most 𝐷 2 CSC304 - Nisarg Shah 21

  22. Max Cost + Randomized β€’ Theorem [Procaccia & Tennenholtz , β€˜09]: The LRM mechanism is strategyproof. β€’ Proof: 1/4 1/2 1/4 2πœ€ πœ€ 1/4 1/2 1/4 CSC304 - Nisarg Shah 22

  23. Max Cost + Randomized β€’ Exercise for you! Try showing that no randomized SP mechanism can achieve approximation ratio < 3/2 β€’ Suggested outline ➒ Consider two agents with 𝑦 1 = 0 and 𝑦 2 = 1 ➒ Show that one of them has expected cost at least Β½ ➒ What happens if that agent moves 1 unit farther from the other agent? CSC304 - Nisarg Shah 23

Recommend


More recommend