USING LOTTERIES TO APPROXIMATE THE OPTIMAL REVENUE Paul W. GoldbergUniversity of Liverpool Carmine VentreTeesside University.

Slides:



Advertisements
Similar presentations
Combinatorial Auction
Advertisements

Combinatorial Auctions with Complement-Free Bidders – An Overview Speaker: Michael Schapira Based on joint works with Shahar Dobzinski & Noam Nisan.
6.896: Topics in Algorithmic Game Theory Lecture 21 Yang Cai.
(Single-item) auctions Vincent Conitzer v() = $5 v() = $3.
Collusion-Resistant Mechanisms with Verification Yielding Optimal Solutions Carmine Ventre (University of Liverpool) Joint work with: Paolo Penna (University.
6.896: Topics in Algorithmic Game Theory Lecture 20 Yang Cai.
Auction Theory Class 5 – single-parameter implementation and risk aversion 1.
Class 4 – Some applications of revenue equivalence
Yang Cai Oct 01, An overview of today’s class Myerson’s Auction Recap Challenge of Multi-Dimensional Settings Unit-Demand Pricing.
Prior-free auctions of digital goods Elias Koutsoupias University of Oxford.
Approximating optimal combinatorial auctions for complements using restricted welfare maximization Pingzhong Tang and Tuomas Sandholm Computer Science.
CPS Bayesian games and their use in auctions Vincent Conitzer
Mechanism Design, Machine Learning, and Pricing Problems Maria-Florina Balcan.
Seminar in Auctions and Mechanism Design Based on J. Hartline’s book: Approximation in Economic Design Presented by: Miki Dimenshtein & Noga Levy.
Prompt Mechanisms for Online Auctions Speaker: Shahar Dobzinski Joint work with Richard Cole and Lisa Fleischer.
An Approximate Truthful Mechanism for Combinatorial Auctions An Internet Mathematics paper by Aaron Archer, Christos Papadimitriou, Kunal Talwar and Éva.
Optimal auction design Roger Myerson Mathematics of Operations research 1981.
A Prior-Free Revenue Maximizing Auction for Secondary Spectrum Access Ajay Gopinathan and Zongpeng Li IEEE INFOCOM 2011, Shanghai, China.
Part 1: Optimal Multi-Item Auctions Constantinos Daskalakis EECS, MIT Reference: Yang Cai, Constantinos Daskalakis and Matt Weinberg: An Algorithmic Characterization.
A Sufficient Condition for Truthfulness with Single Parameter Agents Michael Zuckerman, Hebrew University 2006 Based on paper by Nir Andelman and Yishay.
Seminar In Game Theory Algorithms, TAU, Agenda  Introduction  Computational Complexity  Incentive Compatible Mechanism  LP Relaxation & Walrasian.
Yang Cai Oct 15, Interim Allocation rule aka. “REDUCED FORM” : Variables: Interim Allocation rule aka. “REDUCED FORM” : New Decision Variables j.
6.853: Topics in Algorithmic Game Theory Fall 2011 Matt Weinberg Lecture 24.
Maria-Florina Balcan Approximation Algorithms and Online Mechanisms for Item Pricing Maria-Florina Balcan & Avrim Blum CMU, CSD.
Item Pricing for Revenue Maximization in Combinatorial Auctions Maria-Florina Balcan, Carnegie Mellon University Joint with Avrim Blum and Yishay Mansour.
Combinatorial Auction. Conbinatorial auction t 1 =20 t 2 =15 t 3 =6 f(t): the set X  F with the highest total value the mechanism decides the set of.
Competitive Generalized Auctions Paper by Amos Fiat, Andrew Goldberg, Jason Hartine, Anna Karlin Presented by Chad R. Meiners.
Competitive Auctions and Digital Goods Andrew Goldberg, Jason Hartline, and Andrew Wright presenting: Keren Horowitz, Ziv Yirmeyahu.
Approximation Algorithms for Combinatorial Auctions with Complement-Free Bidders Speaker: Michael Schapira Joint work with Shahar Dobzinski & Noam Nisan.
Competitive Analysis of Incentive Compatible On-Line Auctions Ron Lavi and Noam Nisan SISL/IST, Cal-Tech Hebrew University.
Maria-Florina Balcan Mechanism Design, Machine Learning, and Pricing Problems Maria-Florina Balcan 11/13/2007.
Auction Theory Class 2 – Revenue equivalence 1. This class: revenue Revenue in auctions – Connection to order statistics The revelation principle The.
Mechanisms for Making Crowds Truthful Andrew Mao, Sergiy Nesterko.
CPS 173 Mechanism design Vincent Conitzer
Multi-Unit Auctions with Budget Limits Shahar Dobzinski, Ron Lavi, and Noam Nisan.
VCG Computational game theory Fall 2010 by Inna Kalp and Yosef Heskia.
Auction Seminar Optimal Mechanism Presentation by: Alon Resler Supervised by: Amos Fiat.
More on Social choice and implementations 1 TexPoint fonts used in EMF. Read the TexPoint manual before you delete this box.: AAA A Using slides by Uri.
Mechanism Design CS 886 Electronic Market Design University of Waterloo.
1 Deterministic Auctions and (In)Competitiveness Proof sketch: Show that for any 1  m  n there exists a bid vector b such that Theorem: Let A f be any.
By: Amir Ronen, Department of CS Stanford University Presented By: Oren Mizrahi Matan Protter Issues on border of economics & computation, 2002.
Yang Cai Oct 08, An overview of today’s class Basic LP Formulation for Multiple Bidders Succinct LP: Reduced Form of an Auction The Structure of.
Chapter 4 Bayesian Approximation By: Yotam Eliraz & Gilad Shohat Based on Chapter 4 on Jason Hartline’s book Seminar in Auctions and Mechanism.
Market Design and Analysis Lecture 5 Lecturer: Ning Chen ( 陈宁 )
USING LOTTERIES TO APPROXIMATE THE OPTIMAL REVENUE Paul W. GoldbergUniversity of Liverpool Carmine VentreTeesside University.
Unlimited Supply Infinitely many identical items. Each bidder wants one item. –Corresponds to a situation were we have no marginal production cost. –Very.
Algorithmic Mechanism Design: an Introduction Approximate (one-parameter) mechanisms, with an application to combinatorial auctions Guido Proietti Dipartimento.
Optimal mechanisms (part 2) seminar in auctions & mechanism design Presentor : orel levy.
AAMAS 2013 best-paper: “Mechanisms for Multi-Unit Combinatorial Auctions with a Few Distinct Goods” Piotr KrystaUniversity of Liverpool, UK Orestis TelelisAUEB,
Yang Cai Oct 06, An overview of today’s class Unit-Demand Pricing (cont’d) Multi-bidder Multi-item Setting Basic LP formulation.
Maria-Florina Balcan Mechanism Design, Machine Learning and Pricing Problems Maria-Florina Balcan Joint work with Avrim Blum, Jason Hartline, and Yishay.
Combinatorial Auctions without Money Dimitris Fotakis, NTUA Piotr Krysta, University of Liverpool Carmine Ventre, Teesside University.
6.853: Topics in Algorithmic Game Theory Fall 2011 Constantinos Daskalakis Lecture 22.
Item Pricing for Revenue Maximization in Combinatorial Auctions Maria-Florina Balcan.
Automated mechanism design Vincent Conitzer
Combinatorial Auction. A single item auction t 1 =10 t 2 =12 t 3 =7 r 1 =11 r 2 =10 Social-choice function: the winner should be the guy having in mind.
Approximation Algorithms for Combinatorial Auctions with Complement-Free Bidders Speaker: Shahar Dobzinski Joint work with Noam Nisan & Michael Schapira.
Lecture 4 on Auctions Multiunit Auctions We begin this lecture by comparing auctions with monopolies. We then discuss different pricing schemes for selling.
Comp/Math 553: Algorithmic Game Theory Lecture 10
Automated mechanism design
False-name Bids “The effect of false-name bids in combinatorial
Combinatorial Public Projects
Comp/Math 553: Algorithmic Game Theory Lecture 09
Game Theory in Wireless and Communication Networks: Theory, Models, and Applications Lecture 6 Auction Theory Zhu Han, Dusit Niyato, Walid Saad, Tamer.
Chapter 4 Bayesian Approximation
Competitive Auctions and Digital Goods
Automated mechanism design
Chapter 5. The Duality Theorem
Information, Incentives, and Mechanism Design
Class 2 – Revenue equivalence
Presentation transcript:

USING LOTTERIES TO APPROXIMATE THE OPTIMAL REVENUE Paul W. GoldbergUniversity of Liverpool Carmine VentreTeesside University

iTunes Store

Maximizing the revenue we_are_the_champions.mp3 £ 2.50 £ 3.00 iTunes Revenue = £ 2.97 Optimal Revenue = £ 8.00 More revenue!!!

Maximizing the revenue: eliciting “bids” we_are_the_champions.mp3 £2.50 £ 3.00 iTunes Revenue = £ 8.00 Optimal Revenue = £ 8.00 £ 2.50 £ 3.00 £ 2.50 £ 3.00 Promoted!?

Pay-what-you-say (aka 1 st price auction) weakness we_are_the_champions.mp3 £2.50 £ 3.00 iTunes Revenue = £ 0.03 Optimal Revenue = £ 8.00 £ 0.01 Fired! 1 st price

Incentive-compatibility (IC): truthfulness we_are_the_champions.mp3 v1v1 v1v1 b1b1 b1b1 v2v2 v2v2 b2b2 b2b2 v3v3 v3v3 b3b3 b3b3 is truthful Utility (v 1, b 2, b 3 ) ≥ Utility (b 1, b 2, b 3 ) for all b 1, b 2, b 3 def Utility (b 1, b 2, b 3 ) = v 1 – if song bought, 0 otherwise pricing(b 1, b 2, b 3 ) Def: Pricing truthful if all bidders are truthful pricing rule

IC: collusion-resistance we_are_the_champions.mp3 v1v1 v1v1 b1b1 b1b1 pricing rule v2v2 v2v2 b2b2 b2b2 v3v3 v3v3 b3b3 b3b3 Utility (b 1,b 2,b 3 ) + Utility (b 1,b 2,b 3 ) + Utility (b 1,b 2,b 3 ) maximized when bidders bid (v 1, v 2, v 3 ) def Pricing collusion- resistant

Designing “good” IC pricing rules We want to design IC pricing rules that approximate the optimal revenue as much as possible Not hard to see that “individually rational” deterministic pricing rules can only guarantee bad approximations Example: v 1, v 2, v 3 in {L,H}, L < H – aka, binary domain If bid vector is (L,L,L) then a bidder has to be charged at most L Bid vector (H,L,L): opt=H+2L, revenue=3L, apx ratio ≈ H/L v1v1 v1v1 v2v2 v2v2 v3v3 v3v3

Pricing “lotteries” We propose to price lotteries akin to [Briest et al, SODA10] Pay something for a chance to win the song A lottery has two components: Price p Win probability λ Risk-neutral bidders: Utility ( ) = λ * v 1 - p we_are_the_champions.mp3 v1v1 v1v1 v2v2 v2v2 v3v3 v3v3 b1b1 b1b1 b2b2 b2b2 b3b3 b3b3 Fact: Lotteries truthful iff λ i (b i, b -i ) ≥ λ i (b i ’, b -i ) iff b i ≥ b i ’ and collusion-resistant iff truthful and singular, ie, λ i (b i, b -i ) = λ i (b i, b’ -i ) for all b -i, b’ -i Fact: Lotteries truthful iff λ i (b i, b -i ) ≥ λ i (b i ’, b -i ) iff b i ≥ b i ’ and collusion-resistant iff truthful and singular, ie, λ i (b i, b -i ) = λ i (b i, b’ -i ) for all b -i, b’ -i

Lotteries for binary domains {L,H} Let us consider the following lottery: λ(L) = ½, priced at L/2 λ(H) = 1, priced at H/2 Properties collusion-resistant truthful since monotone non-decreasing singular (offer depends only on the bidder’s bid) anonymous (no bidder id used) approximation guarantee: ½ Tweaking the probabilities we can achieve an approximation guarantee of (2H-L)/H Can a truthful lottery do any better?

Summary of results

Lower bound technique, step 1: Upper bounding the payments Take any truthful lottery (λ j, p j ) for bidder j By individual rationality, the lottery must satisfy L * λ j (L, b -j ) – p j (L, b -j ) ≥ 0 in case j has type L By truthfulness, the lottery must satisfy H * λ j (H, b -j ) – p j (H, b -j ) ≥ H * λ j (L, b -j ) – p j (L, b -j ) in case j has type H We then have the following upper bounds on the payments p j (L, b -j ) ≤ L * λ j (L, b -j ) p j (H, b -j ) ≤ H * λ j (H, b -j ) – H * λ j (L, b -j ) + p j (L, b -j ) ≤ H – (H–L) * λ j (L, b -j )

Lower bound technique, step 2: setting up a linear system Requesting an approximation guarantee better than α implies α * Σ j p j (b) > OPT(b) = H * n H (b) + L * n L (b) for all bid vectors b In step 1, we obtained the following upper bounds on the payments: p j (L, b -j ) ≤ L * λ j (L, b -j ) p j (H, b -j ) ≤ H – (H–L) * λ j (L, b -j ) Then, to get a better than α approximation of OPT the following system of linear inequalities must be satisfied – (H–L) Σ j bidding H in b λ j (L, b -j ) + L Σ j bidding L in b λ j (L, b -j ) > H * n H (b) * (α-1)/α – L * n L (b) * 1/α for any bid vector b x j (b -j )

Lower bound technique, step 3: Carver’s theorem [Carver, 1922] – (H–L) Σ j bidding H in b x j (b -j ) + L Σ j bidding L in b x j (b -j ) > H * n H (b) * (α-1)/α – L * n L (b) * 1/α for any bid vector b n = 2 #bidders - 1 m = 2 #bidders - β i Σ j α ij x j

Lower bound technique, step 4: finding Carver’s constants (2 bidders) – (H–L) Σ j bidding H in b x j (b -j ) + L Σ j bidding L in b x j (b -j ) > H * n H (b) * (α-1)/α – L * n L (b) * 1/α for any bid vector b (LL) L x 1 (L) + L x 2 (L) > – L * 2 * 1/α (LH) L x 1 (H) – (H–L) x 2 (L) > H * (α-1)/α – L * 1/α (HL) – (H–L) x 1 (L) + L x 2 (H) > H * (α-1)/α – L * 1/α (HH)– (H–L) x 1 (H) – (H–L) x 2 (H) > H * 2 * (α-1)/α HH HL LH LL

weighted sum is function of α only weighted sum is 0 Lower bound: concluding the proof L x 1 (L) + L x 2 (L) + L * 2 * 1/α L x 1 (H) – (H–L) x 2 (L) – H * (α-1)/α + L * 1/α – (H–L) x 1 (L) + L x 2 (H) – H * (α-1)/α + L * 1/α – (H–L) x 1 (H) – (H–L) x 2 (H) – H * 2 * (α-1)/α weighted sum is non-positive Lottery cannot apx better than αSystem does not have solutionsk m+1 ≥ 0 α ≤ (2H-L)/H

Conclusions & future research Take home points Collusion-resistance = truthfulness, when approximating OPT with lotteries for digital goods Lotteries much more expressive than universally truthful auctions New lower bounding technique based on Carver’s result about inconsistent systems of linear inequalities What next? Further applications/implications of Carver’s theorem? Lotteries for settings different than digital goods? E.g., goods with limited supply

Extension to any finite domain For a domain of size d, we can easily get a d- approximation of OPT with a collusion-resistant lottery Straightforward generalization of the upper bound for {L,H} domain For any d and ε>0, there exist d values such that no truthful lottery can approximate OPT better than d-ε over the domain given by those values (Non-trivial) generalization of the lower bound for binary domains

Lotteries for domains [1,H] Easy to come up with a collusion-resistant lottery guaranteeing a ln(H)+1 approximation of OPT Bidder bidding b i wins with probability ln(e*b i )/ln(e*H) Matching Lower bound can be proved for any truthful lottery over [1,H] Proof uses a technique designed for universally truthful auctions by [Goldberg et al, GEB 2006]… … And a bijection between truthful lotteries and universally truthful auctions E.g., the ½-approximating lottery for {L,H} can be viewed as a uniform distribution over two simple auctions charging H and L respectively

Related literature Focus for “competitive auctions” [Goldberg et al, GEB 2006] is on F (2) rather than OPT, as OPT is “impossible” to approximate This research can then also be seen as the study of the implications of the knowledge of the domain on the approximation of OPT Lotteries are a more natural interpretation of universally truthful auctions In certain cases (ie, when Cumulative Distribution Functions do not have Probability density functions) lotteries are far more expressive than randomized auctions