CSC304 Lecture 12 Ending Mechanism Design w/ Money: Recap revenue maximization & Myersonβs auction Begin Mechanism Design w/o Money: Facility Location CSC304 - Nisarg Shah 1
Recap β’ Single-item auction with 1 seller, π buyers β’ Buyer π has value π€ π drawn from cdf πΊ π (pdf π π ) 1βπΊ π (π€ π ) β’ Virtual value function: π π π€ π = π€ π β π π (π€ π ) β’ Myersonβs theorem: E[Revenue] = E Ο π π π π€ π β π¦ π β’ Maximize revenue = maximize virtual welfare subject to monotonic allocation rule CSC304 - Nisarg Shah 2
Recap β’ When all πΊ π βs are regular β’ Monotonicity is automatic β’ Allocation: Give to agent π with maximum π π (π€ π ) if π π π€ π β₯ 0 β’ When the maximum π π π€ π is negative, not selling the item is better (zero virtual welfare > negative virtual welfare) β’ Payment: Charge β = min π€ π β² βΆ π π π€ π β² β₯ max 0, π π π€ π π€ π βπ β π β’ Least possible value for which the agent still gets the item β’ If virtual value drops below any other virtual value or below 0 , the agent loses the item CSC304 - Nisarg Shah 3
Recap β’ Special case: All πΊ π = πΊ = Regular β’ VCG with reserve price π β1 (0) β’ Allocation: Give the item to agent π with the maximum value π€ π but only if π€ π β₯ π β1 (0) β’ Equivalent to π π€ π β₯ 0 β’ Payment: max π β1 0 , max πβ π π€ π β’ Least possible value for which the agent still gets the item β’ 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
Approx. Optimal Auctions β’ When πΊ π βs are complex, the virtual valuation function is complex too β’ The optimal auction is unintuitive β’ 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
Approx. Optimal Auctions β’ Still relies on knowing biddersβ distributions β’ 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, πΉ[ Revenue of VCG with π + 1 bidders ] β₯ πΉ[ Optimal revenue with π bidders ] β’ βSpend effort in getting one more bidder than in figuring out the optimal auctionβ CSC304 - Nisarg Shah 6
Simple Proof β’ (n+1)-bidder VCG has the maximum expected revenue among all (n+1)-bidder DSIC auctions that always allocate the item β’ Revenue Equivalence Theorem β’ Consider the following (n+1)-bidder DSIC auction β’ Run π -bidder Myerson on first π bidders. If the item is unallocated, give it to agent π + 1 for free. β’ As much expected revenue as π -bidder Myerson auction β’ No more expected revenue than (n+1)-bidder VCG β’ QED! CSC304 - Nisarg Shah 7
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 β’ Q: Shouldnβt the optimal auction just sell each item individually using Myersonβs auction? β’ 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
Mechanism Design Without Money CSC304 - Nisarg Shah 9
Lack of Money β’ Mechanism design with money: β’ VCG can implement the welfare maximizing outcome because it can charge payments β’ Mechanism design without money: β’ Suppose you want to give away a single item, but cannot charge any payments β’ Impossible to get meaningful information about valuations from strategic agents β’ How would you maximize welfare as much as possible? CSC304 - Nisarg Shah 10
Lack of Money β’ One possibility: Give the item to each of π bidders with probability 1/π . β’ Does not maximize welfare β’ Itβs impossible to maximize welfare without money β’ Achieves an π -approximation of maximum welfare max π π€ π β’ max (1/π) Ο π π€ π β€ π (What is this?) π€ β’ Canβt do better than π -approximation CSC304 - Nisarg Shah 11
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 β’ β strategyproof β = DSIC 4. If not, find the strategyproof mechanism that provides the best approximation ratio β’ Approximation ratio is similar to price of anarchy (PoA) CSC304 - Nisarg Shah 12
Facility Location β’ Set of agents π β’ Each agent π has a true location π¦ π β β β’ Mechanism π β’ Takes as input reports ΰ·€ π¦ = (ΰ·€ π¦ 1 , ΰ·€ π¦ 2 , β¦ , ΰ·€ π¦ π ) β’ Returns a location π§ β β for the new facility β’ Cost to agent π : π π π§ = π§ β π¦ π β’ Social cost π· π§ = Ο π π π π§ = Ο π π§ β π¦ π CSC304 - Nisarg Shah 13
Facility Location β’ Social cost π· π§ = Ο π π π π§ = Ο π π§ β π¦ π β’ Q: Ignoring incentives, what choice of π§ would minimize the social cost? β’ A: The median location med(π¦ 1 , β¦ , π¦ π ) β’ π is odd β the unique β(n+1)/2β th smallest value β’ π is even β βn/2β th or β(n/2)+1β st smallest value β’ Why? CSC304 - Nisarg Shah 14
Facility Location β’ Social cost π· π§ = Ο π π π π§ = Ο π π§ β π¦ π β’ Median is optimal (i.e., 1 -approximation) β’ What about incentives? β’ Median is also strategyproof (SP)! β’ Irrespective of the reports of other agents, agent π is best off reporting π¦ π CSC304 - Nisarg Shah 15
Median is SP No manipulation can help CSC304 - Nisarg Shah 16
Max Cost β’ A different objective function π· π§ = max π π§ β π¦ π β’ Q: Again ignoring incentives, what value of π§ minimizes the maximum cost? β’ A: The midpoint of the leftmost ( min π¦ π ) and the π rightmost ( max π¦ π ) locations (WHY?) π β’ Q: Is this optimal rule strategyproof? β’ A: No! (WHY?) CSC304 - Nisarg Shah 17
Max Cost β’ π· π§ = max π π§ β π¦ π β’ We want to use a strategyproof mechanism. β’ Question: What is the approximation ratio of median for maximum cost? 1. β 1,2 2. β 2,3 3. β 3,4 4. β 4, β CSC304 - Nisarg Shah 18
Max Cost β’ Answer: 2 -approximation β’ Other SP mechanisms that are 2 -approximation β’ Leftmost: Choose the leftmost reported location β’ Rightmost: Choose the rightmost reported location β’ Dictatorship: Choose the location reported by agent 1 β’ β¦ CSC304 - Nisarg Shah 19
Max Cost β’ Theorem [Procaccia & Tennenholtz , β09] No deterministic SP mechanism has approximation ratio < 2 for maximum cost. β’ Proof: CSC304 - Nisarg Shah 20
Max Cost + Randomized β’ The Left-Right-Middle (LRM) Mechanism β’ Choose min π¦ π with probability ΒΌ π β’ Choose max π¦ π with probability ΒΌ π β’ Choose (min π¦ π + max π¦ π )/2 with probability Β½ π π β’ Question: What is the approximation ratio of LRM for maximum cost? (1/4)β2π·+(1/4)β2π·+(1/2)βπ· 3 = β’ At most π· 2 CSC304 - Nisarg Shah 21
Max Cost + Randomized β’ Theorem [Procaccia & Tennenholtz , β09]: The LRM mechanism is strategyproof. β’ Proof: 1/4 1/2 1/4 2π π 1/4 1/2 1/4 CSC304 - Nisarg Shah 22
Max Cost + Randomized β’ Exercise for you! Try showing that no randomized SP mechanism can achieve approximation ratio < 3/2 β’ Suggested outline β’ Consider two agents with π¦ 1 = 0 and π¦ 2 = 1 β’ Show that one of them has expected cost at least Β½ β’ What happens if that agent moves 1 unit farther from the other agent? CSC304 - Nisarg Shah 23
Recommend
More recommend