Competitive Analysis of Incentive Compatible On-line Auctions

Similar documents
1 Mechanism Design (incentive-aware algorithms, inverse game theory)

1 Mechanism Design (incentive-aware algorithms, inverse game theory)

Lecture 7 - Auctions and Mechanism Design

1 Mechanism Design (incentive-aware algorithms, inverse game theory)

Price of anarchy in auctions & the smoothness framework. Faidra Monachou Algorithmic Game Theory 2016 CoReLab, NTUA

Game Theory: Spring 2017

Harvard University Department of Economics

Searching for the Possibility Impossibility Border of Truthful Mechanism Design

Mechanism Design in Social Networks

Reserve Price Auctions for Heterogeneous Spectrum Sharing

Non-decreasing Payment Rules in Combinatorial Auctions

Sponsored Search Markets

HIERARCHICAL decision making problems arise naturally

CS364B: Frontiers in Mechanism Design Lecture #17: Part I: Demand Reduction in Multi-Unit Auctions Revisited

Note on webpage about sequential ascending auctions

Robust Multi-unit Auction Protocol against False-name Bids

Approximation in Algorithmic Game Theory

Games, Auctions, Learning, and the Price of Anarchy. Éva Tardos Cornell University

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL., NO., 1. Dynamic Spectrum Sharing Auction with Time-Evolving Channel Qualities

Multi-player and Multi-round Auctions with Severely Bounded Communication

Title: A Mechanism for Fair Distribution of Resources with Application to Sponsored Search

An Auction Mechanism for Resource Allocation in Mobile Cloud Computing Systems

Concurrent Auctions Across The Supply Chain

Activity Rules and Equilibria in the Combinatorial Clock Auction

Methods for boosting revenue in combinatorial auctions

Three New Connections Between Complexity Theory and Algorithmic Game Theory. Tim Roughgarden (Stanford)

A Cooperative Approach to Collusion in Auctions

CSC304: Algorithmic Game Theory and Mechanism Design Fall 2016

VALUE OF SHARING DATA

The Relevance of a Choice of Auction Format in a Competitive Environment

Characterization of Strategy/False-name Proof Combinatorial Auction Protocols: Price-oriented, Rationing-free Protocol

A Framework for Truthful Online Auctions in Cloud Computing with Heterogeneous User Demands

Computationally Feasible VCG Mechanisms. by Alpha Chau (feat. MC Bryce Wiedenbeck)

Efficiency Guarantees in Market Design

Supply-Limiting Mechanisms

The Need for Information

arxiv: v1 [cs.gt] 11 Feb 2015

Optimal Shill Bidding in the VCG Mechanism

The Need for Information

Online Ad Auctions. By Hal R. Varian. Draft: December 25, I describe how search engines sell ad space using an auction.

Multiple Sourcing and Procurement Process Selection with Bidding Events

REDESIGNING BITCOIN S FEE MARKET arxiv:

Modified Truthful Greedy Mechanisms for Dynamic Virtual Machine Provisioning and Allocation in Clouds

Notes on Introduction to Contract Theory

The Ascending Bid Auction Experiment:

Incentive-Compatible, Budget-Balanced, yet Highly Efficient Auctions for Supply Chain Formation

CS269I: Incentives in Computer Science Lecture #16: Revenue-Maximizing Auctions

Public Economics by Luca Spataro. Market failures: Externalities (Myles ch. 10. sections 4.4, 5, 7.2 & 7.3 excluded)

Miscomputing Ratio: The Social Cost of Selfish Computing

A Study of Limited-Precision, Incremental Elicitation in Auctions

Chapter 17. Auction-based spectrum markets in cognitive radio networks

An Ascending Price Auction for Producer-Consumer Economy

Efficient Ex Post Implementable Auctions and English Auctions for Bidders with Non-Quasilinear Preferences

Topics in ALGORITHMIC GAME THEORY *

Intro to Algorithmic Economics, Fall 2013 Lecture 1

Simultaneous Single-Item Auctions

When Are Equilibria of Simple Auctions Near-Optimal?

Auctions for online ad space among advertisers sensitive to both views and clicks

Approximation and Mechanism Design

Auction Theory An Intrroduction into Mechanism Design. Dirk Bergemann

CS364B: Frontiers in Mechanism Design Lecture #11: Undominated Implementations and the Shrinking Auction

Review from last. Econ 2230: Public Economics. Outline. 1. Dominant strategy implementation. Lecture 9: Mechanism Design lab evidence

Journal of Computer and System Sciences

Modeling of competition in revenue management Petr Fiala 1

Quality-Driven Auction based Incentive Mechanism for Mobile Crowd Sensing

Valuation Uncertainty and Imperfect Introspection in Second-Price Auctions

Alberta Capacity Market

Copyright (C) 2001 David K. Levine This document is an open textbook; you can redistribute it and/or modify it under the terms of version 1 of the

Bidding for Sponsored Link Advertisements at Internet

Recap Beyond IPV Multiunit auctions Combinatorial Auctions Bidding Languages. Multi-Good Auctions. CPSC 532A Lecture 23.

University of California, Davis Date: July 3, PRELIMINARY EXAMINATION FOR THE Ph.D. DEGREE

Modeling Heterogeneous User. Churn and Local Resilience of Unstructured P2P Networks

The Sealed Bid Auction Experiment:

Problem Set 2. Externalities

Introduction to Auctions

FIRST FUNDAMENTAL THEOREM OF WELFARE ECONOMICS

Competitive Markets. Jeffrey Ely. January 13, This work is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 3.0 License.

Ad Auctions with Data

1 Introduction 1. 2 Forecasting and Demand Modeling 5. 3 Deterministic Inventory Models Stochastic Inventory Models 63

On Optimal Multidimensional Mechanism Design

Information and the Coase Theorem. Joseph Farrell. "The Journal of Economic Perspectives," Vol. 1, No. 2 (Autumn, 1987), pp.

Truth Revelation in Approximately Efficient Combinatorial Auctions Daniel Lehmann School of Computer Science and Engineering, Hebrew University, Jerus

WITH limited resources and high population density,

Public Advertisement Broker Markets.

COMP/MATH 553 Algorithmic Game Theory Lecture 8: Combinatorial Auctions & Spectrum Auctions. Sep 29, Yang Cai

Crowdsourcing Contests

How Does Computer Science Inform Modern Auction Design?

Supplimentary material for Research at the Auction Block: Problems for the Fair Benefits Approach to International Research

Pricing in Dynamic Advance Reservation Games

Econ 121b: Intermediate Microeconomics

Prompt Mechanisms for Online Auctions

Strategic Alliances, Joint Investments, and Market Structure

First-Price Path Auctions

Mechanisms. A thesis presented. Swara S. Kopparty. Applied Mathematics. for the degree of. Bachelor of Arts. Harvard College. Cambridge, Massachusetts

Bidding Clubs in First-Price Auctions Extended Abstract

We survey different models, techniques, and some recent results to tackle machine scheduling

Incentive Compatible Mechanisms for Resource Procurement in Computational Grids with Rational Resource Providers

Competing Mechanisms with Multi-Unit Consumer Demand

Electricity Pooling Markets with Elastic Demand: A Mechanism Design Approach

FlexAuc: Serving Dynamic Demands in a Spectrum Trading Market with Flexible Auction

Transcription:

Competitive Analysis of Incentive Compatible On-line Auctions Ron Lavi and Noam Nisan Theoretical Computer Science, 310 (2004) 159-180 Presented by Xi LI Apr 2006 COMP670O HKUST

Outline The On-line Auction Model Incentive & Supply Curves Terminologies Global Supply Curve Revenue & Social Efficiency Off-line Vickrey Auction Competitiveness One Divisible Good k Indivisible Goods A Randomized Auction A Deterministic Auction Revenue Analysis on Uniform Distribution

The Model The goods k identical indivisible goods when k is very large one divisible good Players valuations and utilities Player i has marginal valuation v i (j) for good j, 1 j k Assume that i, j: v i (j+1) v i (j) When player i receives q goods and pays P i his utility is Each player aims to maximize his utility v i

The Model The on-line game and players strategies At time t i, player i declares function b i ( ) as his marginal function b i : [1 k] R, non-increasing (of coz he could lie, i.e. b i ( ) v i ( ) ). The auctioneer must answer bidder i immediately with q i and P i Player i b i ( ) q i & P i t i t 1... t i+1 Auctioneer Applications: CPU time, cache space, bandwidth

Outline The On-line Auction Model Incentive & Supply Curves Terminologies Global Supply Curve Revenue & Social Efficiency Off-line Vickrey Auction Competitiveness One Divisible Good k Indivisible Goods A Randomized Auction A Deterministic Auction Revenue Analysis on Uniform Distribution

Incentive A strategy (bid) of player i is called dominant if for any other bid and for any sequence of the past and future bids of the other players,. An auction is called incentive compatible if for any valuation, declaring the true valuation is a dominant strategy. Comments: the motivation of incentive to free the bidders from strategic considerations (Vickrey et al. 1961); when all bidders are telling the truth, it is easy to maximize the social efficiency.

Supply Curves for On-line Auctions Definition 1 (Supply curves). An on-line auction is called based on supply curves if before receiving the i th bid it fixes a function (supply curve) p i (q) based on previous bids, and, 1. The quantity q i sold to bidder i is the quantity q that maximizes the sum j=1q (b i (j)-p i (j)), i.e. the bidder s utility. 2. The price paid by bidder i is j=1 qi p i (j). Assume each supply curve is non-decreasing.

Incentive & Supply Curves Theorem 1. A deterministic on-line auction is incentive compatible if and only if it is based on supply curves. Proof. This is proved in two directions by Lemma 1 and Lemma 2. Lemma 1. An on-line auction that is based on supply curves is incentive compatible. Proof. According to Definition 1, is maximized if based on supply curve. So that is always maximized iff. Lemma 2. Any deterministic incentive compatible on-line auction is based on supply curves. Proof. Next slides.

Proof of Lemma 2 Lemma 2. Any deterministic incentive compatible on-line auction is based on supply curves. Proof. For each player i, P i is uniquely determined by q i. Otherwise there exists bids v and v, where P<P, so that a player which has valuation v will lie by declaring v to increase his utility, which contradicts incentive compatibility. Denote P i (q): [1,k] R, the total payment of player i for q items. The allocation to player i must maximize otherwise player i will lie to increase his utility. Denote. Since, the allocation maximizes, and the payment is so that is the supply curve according to Definition 1.

Special Case: Fixed Marginal Valuation Lemma 3. Assume that for any player i, the marginal valuation is fixed to v i. Then any incentive compatible on-line auction is based on non-decreasing supply curves. marginal valuation Proof. 1. q i (v) is non-decreasing. v i quantity 2. Define p i (q) = inf { v q i (v) q }. Since q i (v) is non-dreasing, p i (q) is non-decreasing as well. 3. Any incentive compatible on-line auction A is based on p i (q).

Outline The On-line Auction Model Incentive & Supply Curves Terminologies Global Supply Curve Revenue & Social Efficiency Off-line Vickrey Auction Competitiveness One Divisible Good k Indivisible Goods A Randomized Auction A Deterministic Auction Revenue Analysis on Uniform Distribution

Global Supply Curve Definition 2 (A global supply curve). An online auction is called based on a global supply curve p(q) if it is based on supply curves and if where q j is the quantity sold to the jth bidder. q i-1 q i

Revenue and Social Efficiency Definition 3 (Revenue and social efficiency). In auction A, for a valuation sequence σ, the revenue is The social efficiency is Assumptions: 1. All marginal valuations are taken from some known interval, without assuming any distribution on them. 2. p is the salvage price of the auctioneer.

Off-line Vickrey Auction Definition 4 (The Vickrey auction). In the Vickrey auction, each player declares his marginal valuation function. The allocation chosen is the one that maximizes the social efficiency (according to the players declarations). The price charged from player i for the quantity q i he receives is the worth of this additional quantity to the other players. Formally, denote by E -i the optimal social efficiency when player i is missing, and by E the actual optimal social efficiency. Then the price that i pays is E -i -(E-v i (q i )). V 1 Social efficiency Price paid by player 1 V 2 q total q total q total q 1 q 2 q 1

Competitiveness Definition 5 (Competitiveness). An on-line auction A is c-competitive with respect to the revenue if for every valuation sequence σ, R A (σ) R vic (σ)/c. Similarly, A is c-competitive with respect to the social efficiency if for every valuation sequence σ, E A (σ) E vic (σ)/c. Comments: E vic is always optimal; while R vic is not necessarily optimal, i.e. sometimes can be far from the optimal revenue.

Outline The On-line Auction Model Incentive & Supply Curves Terminologies Global Supply Curve Revenue & Social Efficiency Off-line Vickrey Auction Competitiveness One Divisible Good k Indivisible Goods A Randomized Auction A Deterministic Auction Revenue Analysis on Uniform Distribution

One Divisible Good Let c be the unique solution to the equation: Comments: it can be shown that. For example, if =2 then c=1.28, and if =8 then c=1.97. Definition 6 (The competitive on-line auction). Define the competitive supply curve by The competitive on-line auction has the competitive supply curve as its global supply curve.

One Divisible Good Lemma 4. (El-Yaniv et al) The functions q(x), r(x) preserve the following conditions: 1. 2. 3. Where c is as defined in Eq. (1).

One Divisible Good Theorem 2. The competitive on-line auction is c- competitive with respect to the revenue and the social efficiency. Lemma 6. For any sequence of valuations σ, R cola (σ) R vic (σ)/c, where cola is the competitive online auction and vic is the Vickrey auction. Lemma 7. For any sequence of valuations σ, E cola (σ) E opt (σ)/c, where E opt (σ) is the optimal social efficiency for σ.

One Divisible Good Theorem 3. Any incentive compatible on-line auction must have a competitive ratio of at least c with respect to both the revenue and the social efficiency, where c is the solution to Eq. (1). Lemma 5. For any constant, there is no function such that Where and

Outline The On-line Auction Model Incentive & Supply Curves Terminologies Global Supply Curve Revenue & Social Efficiency Off-line Vickrey Auction Competitiveness One Divisible Good k Indivisible Goods A Randomized Auction A Deterministic Auction Revenue Analysis on Uniform Distribution

A Randomized Auction for k Indivisible Goods Definition 7. The randomized on-line auction: the supply curve is fixed with p(x)=p on, where p on is chosen by using the cumulative distribution q( ). p on Theorem 4. For any sequence of valuations σ, the expected revenue of the randomized auction is at least 1/c times the optimal efficiency, i.e. E(R on (σ)) E opt (σ)/c.

Define the cdf: v Proof of Theorem 4, Pr(x v)=q(v),

A Deterministic Auction for k Indivisible Goods Definition 8 (The discrete on-line auction). The discrete online auction is based on the following global supply curve: Theorem 5. The discrete on-line auction is k Φ 1/(k+1) - competitive with respect to the revenue and to the social efficiency. When k 2 lnφ then the discrete on-line auction is also 2 e (ln(φ)+1)-competitive with respect to the revenue and to the social efficiency. Theorem 6. Any incentive compatible on-line auction of k goods has a competitive ratio of at least m=max{φ 1/(k+1),c} with respect to the revenue and to the efficiency.

Revenue Analysis on Uniform Distribution We compare the expected revenue of the competitive on-line auction to the expected revenue of the Vickrey off-line auction for a divisible good in the special case of fixed marginal valuations uniformly distributed in.