Appendix for An Efficient Ascending-Bid Auction for Multiple Objects: Comment For Online Publication

Similar documents
we have E Y x t ( ( xl)) 1 ( xl), e a in I( Λ ) are as follows:

SELECTED SOLUTIONS, SECTION (Weak duality) Prove that the primal and dual values p and d defined by equations (4.3.2) and (4.3.3) satisfy p d.

Xin Li Department of Information Systems, College of Business, City University of Hong Kong, Hong Kong, CHINA

Appendix B. Criterion of Riemann-Stieltjes Integrability

Price Competition under Linear Demand and Finite Inventories: Contraction and Approximate Equilibria

ON AUTOMATIC CONTINUITY OF DERIVATIONS FOR BANACH ALGEBRAS WITH INVOLUTION

2.3 Nilpotent endomorphisms

Solutions to exam in SF1811 Optimization, Jan 14, 2015

Genericity of Critical Types

Module 17: Mechanism Design & Optimal Auctions

Appendix for Causal Interaction in Factorial Experiments: Application to Conjoint Analysis

Vickrey Auction VCG Combinatorial Auctions. Mechanism Design. Algorithms and Data Structures. Winter 2016

Pricing and Resource Allocation Game Theoretic Models

Lectures - Week 4 Matrix norms, Conditioning, Vector Spaces, Linear Independence, Spanning sets and Basis, Null space and Range of a Matrix

Perfect Competition and the Nash Bargaining Solution

Lower bounds for the Crossing Number of the Cartesian Product of a Vertex-transitive Graph with a Cycle

L-Edge Chromatic Number Of A Graph

Single-Source/Sink Network Error Correction Is as Hard as Multiple-Unicast

Computing Correlated Equilibria in Multi-Player Games

find (x): given element x, return the canonical element of the set containing x;

ISOTONE EQUILIBRIUM IN GAMES OF INCOMPLETE INFORMATION. By David McAdams 1

Vickrey Auctions with Reserve Pricing

ON THE EQUIVALENCE OF ORDINAL BAYESIAN INCENTIVE COMPATIBILITY AND DOMINANT STRATEGY INCENTIVE COMPATIBILITY FOR RANDOM RULES

Economics 101. Lecture 4 - Equilibrium and Efficiency

(1 ) (1 ) 0 (1 ) (1 ) 0

FIRST AND SECOND ORDER NECESSARY OPTIMALITY CONDITIONS FOR DISCRETE OPTIMAL CONTROL PROBLEMS

NP-Completeness : Proofs

Development Pattern and Prediction Error for the Stochastic Bornhuetter-Ferguson Claims Reserving Method

Price competition with capacity constraints. Consumers are rationed at the low-price firm. But who are the rationed ones?

FINITELY-GENERATED MODULES OVER A PRINCIPAL IDEAL DOMAIN

Linear, affine, and convex sets and hulls In the sequel, unless otherwise specified, X will denote a real vector space.

Modelli Clamfim Equazioni differenziali 7 ottobre 2013

Assortment Optimization under MNL

MARKOV CHAIN AND HIDDEN MARKOV MODEL

An Analysis of Troubled Assets Reverse Auction

Associative Memories

Randić Energy and Randić Estrada Index of a Graph

Tit-For-Tat Equilibria in Discounted Repeated Games with. Private Monitoring

Statistical Mechanics and Combinatorics : Lecture III

MATH 5707 HOMEWORK 4 SOLUTIONS 2. 2 i 2p i E(X i ) + E(Xi 2 ) ä i=1. i=1

DISCRIMINATION VIA SYMMETRIC AUCTIONS: ONLINE APPENDIX

Inner Product. Euclidean Space. Orthonormal Basis. Orthogonal

e - c o m p a n i o n

Ex post implementation in environments with private goods

Supplement: Proofs and Technical Details for The Solution Path of the Generalized Lasso

College of Computer & Information Science Fall 2009 Northeastern University 20 October 2009

Problem Set 4: Sketch of Solutions

n-step cycle inequalities: facets for continuous n-mixing set and strong cuts for multi-module capacitated lot-sizing problem

A Generalized Vickrey Auction

Note 2. Ling fong Li. 1 Klein Gordon Equation Probablity interpretation Solutions to Klein-Gordon Equation... 2

Welfare Properties of General Equilibrium. What can be said about optimality properties of resource allocation implied by general equilibrium?

763622S ADVANCED QUANTUM MECHANICS Solution Set 1 Spring c n a n. c n 2 = 1.

Characterization of Pure-Strategy Equilibria in Bayesian Games

Affine transformations and convexity

Boning Yang. March 8, 2018

Volume 18 Figure 1. Notation 1. Notation 2. Observation 1. Remark 1. Remark 2. Remark 3. Remark 4. Remark 5. Remark 6. Theorem A [2]. Theorem B [2].

Revenue Maximization in a Spectrum Auction for Dynamic Spectrum Access

Hila Etzion. Min-Seok Pang

Online Appendix. t=1 (p t w)q t. Then the first order condition shows that

Collusion in repeated auctions: a simple dynamic mechanism

6. Stochastic processes (2)

PAijpam.eu SOME NEW SUM PERFECT SQUARE GRAPHS S.G. Sonchhatra 1, G.V. Ghodasara 2

NODAL PRICES IN THE DAY-AHEAD MARKET

6. Stochastic processes (2)

MAT 578 Functional Analysis

On Tacit Collusion among Asymmetric Firms in Bertrand Competition

COXREG. Estimation (1)

Approximately-Strategyproof and Tractable Multi-Unit Auctions

Journal of Mathematical Economics

a b a In case b 0, a being divisible by b is the same as to say that

Bernoulli Numbers and Polynomials

k t+1 + c t A t k t, t=0

REAL ANALYSIS I HOMEWORK 1

Journal of Multivariate Analysis

Anti-van der Waerden numbers of 3-term arithmetic progressions.

MA 323 Geometric Modelling Course Notes: Day 13 Bezier Curves & Bernstein Polynomials

Random Walks on Digraphs

Estimation: Part 2. Chapter GREG estimation

MMA and GCMMA two methods for nonlinear optimization

THE ARIMOTO-BLAHUT ALGORITHM FOR COMPUTATION OF CHANNEL CAPACITY. William A. Pearlman. References: S. Arimoto - IEEE Trans. Inform. Thy., Jan.

FACTORIZATION IN KRULL MONOIDS WITH INFINITE CLASS GROUP

n α j x j = 0 j=1 has a nontrivial solution. Here A is the n k matrix whose jth column is the vector for all t j=0

Simultaneous Optimization of Berth Allocation, Quay Crane Assignment and Quay Crane Scheduling Problems in Container Terminals

Ballot Paths Avoiding Depth Zero Patterns

First day August 1, Problems and Solutions

Suggested solutions for the exam in SF2863 Systems Engineering. June 12,

ON THE JACOBIAN CONJECTURE

Optimal Dynamic Mechanism Design and the Virtual Pivot Mechanism

coordinates. Then, the position vectors are described by

where a is any ideal of R. Lemma 5.4. Let R be a ring. Then X = Spec R is a topological space Moreover the open sets

and decompose in cycles of length two

Games of Threats. Elon Kohlberg Abraham Neyman. Working Paper

A A Non-Constructible Equilibrium 1

(c) (cos θ + i sin θ) 5 = cos 5 θ + 5 cos 4 θ (i sin θ) + 10 cos 3 θ(i sin θ) cos 2 θ(i sin θ) 3 + 5cos θ (i sin θ) 4 + (i sin θ) 5 (A1)

ALGEBRA MID-TERM. 1 Suppose I is a principal ideal of the integral domain R. Prove that the R-module I R I has no non-zero torsion elements.

Interference Alignment and Degrees of Freedom Region of Cellular Sigma Channel

Mixed Taxation and Production Efficiency

Additional Codes using Finite Difference Method. 1 HJB Equation for Consumption-Saving Problem Without Uncertainty

Non-negative Matrices and Distributed Control

MATH 241B FUNCTIONAL ANALYSIS - NOTES EXAMPLES OF C ALGEBRAS

Transcription:

Appendx for An Effcent Ascendng-Bd Aucton for Mutpe Objects: Comment For Onne Pubcaton Norak Okamoto The foowng counterexampe shows that sncere bddng by a bdders s not aways an ex post perfect equbrum under a ratonng rues that satsfy the monotoncty property. Exampe 3 Consder a case where there are two bdders, A and B, and three quanttes of an object. Let u A, u B be the margna vaue functons of the two bdders such that 5 f q [0, 1 u A (q = u B (q = 1 f q [1, 3]. Consder the hstory h 4 = (x t A, xt B t=0,1,2,3 = ( (3, 3, (3, 3, (3, 3, (3, 3. After h 4, sncere bddng of each bdder s 1. The resut under sncere bddng x 4 A = 1 If the bdders report x 4 A = x4 B = 1 after h4, then the aucton ends at z 5 = (h 4, (1, 1, yedng an assgnment (x A, x B such that 1 x A 3, 1 x B 3, x A + x B = 3. Wthout oss of generaty, suppose that x A 3. Snce bdder A dd not cnch 2 at t 3, A s payment s ya = 4x A. Therefore, A s utty s U A(x A 4x A at z5. 1

The resut under msreportng ˆx 4 A = 0 If bdder A reports ˆx 4 A = 0, and bdder B reports x4 B = 1 after h4, then the aucton ends at ẑ 5 = (h 4, (0, 1, yedng an assgnment (ˆx A, ˆx B. Snce 0 = ˆx 4 A < x4 A = 1 and any other condton of ẑ5 s the same as z 5, by the monotoncty property, ˆx A must be strcty ess than x A. Smary to case wth sncere bddng, A s utty at ẑ 5 s U A (ˆx A 4ˆx A. We cacuate the dfference between A s uttes at z 5 and ẑ 5, ( U A (x A 4x A (U A (ˆx A 4ˆx A ( x ( A ˆxA = u A (qdq 4x A u A (qdq 4ˆx A 0 0 ( x A ˆxA = u A (qdq u A (qdq 4 (x A ˆx A 0 0 x A = u A (qdq 4 (x A ˆx A. (1 ˆx A Case 1: ˆx A 1. We cacuate (1 such that x A ˆx A 4 (x A ˆx A = 3 (x A ˆx A < 0. Case 2: ˆx A < 1. We cacuate (1 such that ( x A 1 + 5 (1 ˆx A 4 (x A ˆx A = 3x A ˆx A + 4 < 0 ( x A 3 2. Thus, A s utty at ẑ 5 s strcty greater than that at z 5, and bdder A has an ncentve to msreport after h 4. Therefore, sncere bddng by a bdder s not an ex post perfect equbrum. 2

Proof of Lemma 1 Snce u s a weaky decreasng nteger-vaued functon, there s a partton {a 0,..., a m } X wth 0 = a 0 < < a m = λ and vaues {b 1,..., b m } {0, 1,..., u} wth b 1 > b 2 > > b m such that for each k wth 1 k m, u (x = b k f a k 1 < x < a k. Note that m T. Consder any x X. Let By the defnton of Remann Integra, U (x = x 0 k = arg mn{a : x a }. k 1 u (qdq = b (a a 1 + b k (x a k 1. (2 =1 Take any p {1,..., T }. Defne b 0 = T + 1. Let By equaton (2, we can verfy that r = arg mn{b : p 1 < b }, r = arg mn{b : p b }. Because b Z for each, a r = mn{arg max x X U (x (p 1x }, a r = max{arg max x X U (x px }. {b : p 1 < b } = {b : p b }. Therefore a r = a r, that s, mn{arg max U (x (p 1x } = max{arg max U (x px }. x X x X 3

To prove Lemma 2 and Proposton 1, we expan some notaton and a property of the Ausube aucton. Notaton Wth fu bd nformaton, a strategy σ s a functon that maps each nontermna hstory h H \ Z to a quantty x X, that s, σ : H \ Z X. For each non-termna hstory h H \Z, the set of hstores n the subgame that foows h s gven by H h = {h H : h = (h, h for some sequence h }, and the set of termna hstores n the subgame s gven by Z h = Z H h. For each non-termna hstory h H \ Z and each strategy σ, we denote σ h : H h \ Z h X the nduced strategy n the subgame that foows h. For each h H h \ Z h, σ (h = σ h (h. Let π ( be the utty of bdder at an n-tupe of strateges. Property 1 For each t 1, f there exsts a bdder N such that x t = C t 1 and C t 1 > 0, then the aucton ends at t,.e., t = L. Therefore, f the aucton does not end at t, then for each bdder N, x t C t 1 Proof. Suppose that x t = C t 1 and C t 1 or C t 1 = 0. > 0. Then, x t = M j xt 1 j. By bddng constrant for each j N, x t j x t 1 j. Therefore j N xt j M. Note that ths property hods under a ratonng rues. We use the property n proofs of Lemma 2 and Proposton 1. 4

Proof of Lemma 2 Consder any t {0, 1,..., T }, h t = (x s 1, x s 2,..., x s n s t 1 H t \ Z t, and (u j j N. For each j N, et σ j be sncere bddng whch s correspondng to u j, and σj h t be nduced sncere bddng n the subgame that foows h t. Take any N and σ Σ h t. Suppose that x t 1 Q (p t 1. We sha show that Let π ((σ j h t j N π (σ, (σ j h t j. z L+1 = (x s 1, x s 2,..., x s n s L be the termna hstory whch s reached by (σ j h t j N, and w L +1 = (ˆx s 1, ˆx s 2,..., ˆx s n s L be the termna hstory whch s reached by (σ, (σj h t j. Denote {(Cj t j N } L t=0 the cumuatve cnches of z L+1, and {(Ĉt j j N } L t=0 the cumuatve cnches of w L +1. Step 1. x L 1 Q (p L 1. If L 1 = t 1, x L 1 = x t 1 Q (p t 1 = Q (p L 1. Then, et L 1 t. By the defnton of sncere bddng, x L 1 = σ h t By Property 1, x L 1 ( (x 1,..., x n L 2 = mn{x L 2, max{c L 2, Q (p L 1 }}. C L 2 Therefore, x L 1 Q (p L 1. or C L 2 = 0. Then, x L 1 = mn{x L 2, Q (p L 1 }. Step 2. For each j and s mn{l 1, L 1}, x s j = ˆx s j. Ths mpes that for each s mn{l 1, L 1}, C s = M j x s j = M j ˆx s j = Ĉs. 5

For each s t 1, obvousy x s j = ˆx s j. For the cases wth t s mn{l 1, L 1}, we sha show by nducton. Let s = t. Because x t j = σ j h t(h t and ˆx t j = σ j h t(h t, x t j = ˆx t j. Let s = k wth t + 1 k mn{l 1, L 1}. Suppose that x j = ˆx j for a wth t + 1 k 1. By the defnton of sncere bddng, ( x k j = σj h t (x 1,..., x n k 1 = mn{x k 1 j, max{c k 1 j, Q j (p k }}, ( ˆx k j = σj h t (ˆx 1,..., ˆx n k 1 = mn{ˆx k 1 j, max{ĉk 1 j, Q j (p k }}. Snce k mn{l 1, L 1}, by Property 1, x k j C k 1 j or C k 1 j = 0. Thus, x k j = mn{x k 1 j, Q j (k}. Smary, we have ˆx k j = mn{ˆx k 1 j, Q j (k}. Snce x k 1 j = ˆx k 1 j, x k j = ˆx k j. Step 3. π ((σ j h t j N π (σ, (σ j h t j. We consder three cases; L = L, L > L and L < L. Case 1. L = L. By step 2, for a s L 1 = L 1, C s = Ĉs. We cacuate C L and ĈL for two cases wth x L Q (p L and x L < Q (p L. Case 1-1. x L Q (p L. By step 1, x L 1 Therefore, by emma 1, Q L 1. Thus, Q (p L x L C L x L 1 Q (p L 1. mn{arg max x X (U (x p L x } C L max{arg max x X (U (x p L x }. Hence, π ((σ j h t j N π (σ, (σ j h t j. Case 1-2. Let x L < Q (p L. We sha show that x L = x t 1. By the defnton of sncere bddng, x L = σ h t((x s 1,..., x s n s L 1 = mn{x L 1, max{c L 1, Q (p L }}. 6

Snce x L < Q (p L, x L = x L 1. If L 1 = t 1, x L 1 t 1 L 1. By the defnton of sncere bddng, = x t 1. Then, we assume x L 1 = σ h t((x s 1,..., x s n s L 2 = mn{x L 2, max{c L 2, Q (p L 1 }}. Snce Q (p L Q (p L 1, x L 1 = x L < Q (p L 1. Hence, we have x L 1 = x L 2. By repeatng ths procedure, x L = x L 1 = = x t 1. Thus, C L = x t 1. Snce bdder cannot bd more quantty than x t 1 after h t, ĈL x t 1. Then, Ĉ L C L < Q (p L. Hence, π ((σ j h t j N π (σ, (σ j h t j. Case 2. L > L. By step 2, for each s L 1, C s = Ĉs. Then, we cacuate {C s } L s=l and Ĉ L. Snce the aucton does not end at L n the hstory z L+1, by Property 1 for each j, x L j C L 1 j or C L 1 j = 0. Then, by the defnton of sncere bddng, for each j, On the other hand, x L j = mn{x L 1 j, Q j (p L }. ˆx L j = mn{ˆx L 1 1 j, max{ĉl j, Q j (p L }}. Snce x L 1 j = ˆx L 1 j, x L j ˆx L j. Thus, Ĉ L M j ˆx L j M j x L j = C L. By the defnton of cumuatve cnches, for each s {L,..., L 1}, C s x s and x L C L x L 1. For each s {L,..., L 1}, because s t, x s s sncere bddng. That s, x s = mn{x s 1, max{c s 1, Q (p s }}. 7

Snce the aucton does not end at s L 1 n the hstory z L+1, by Property 1, x s = mn{x s 1, Q (p s }. Therefore, for each s {L,..., L 1}, x s Q (p s. Thus, C s Q (p s s {L,..., L 1}, Ĉ L C L Q (p L, C L x L 1 Q (p L 1 = max{arg max x X (U (x p L x }. Hence, π ((σ j h t j N π (σ, (σ j h t j. Case 3. L < L. We frst show that Q (p L x L. Suppose that Q (p L > x L. Smary to case 1-2, we have x L = x t 1. By bddng constrant, ˆx L x t 1. Then, ˆx L x L. Snce L < L, the aucton does not end at L n the hstory w L +1. Therefore, by Property 1, for each j, ˆx L ĈL or ĈL = 0. By the defnton of sncere bddng x L j ˆx L j = mn{x L 1 j, max{c L 1 j, Q j (p L }}, = mn{ˆx L 1 j, max{ĉl 1 j, Q j (p L }} = mn{ˆx L 1 j, Q j (p L }. For each j, snce by step 2, x L 1 j = ˆx L 1 j, we have x L j ˆx L j. Hence for each j N, x L j ˆx L j. Snce the aucton ends at L n the hstory z L+1, j N xl j M. Therefore, j N ˆxL j M. Ths mpes the aucton ends at L n the hstory w L +1. Ths contradcts to L < L. Thus, Q (p L x L. By step 2, for each s L 1, C s = Ĉs. Smary to case 2, we have C L ĈL. Because x L C L, Q (p L C L ĈL. Moreover, for each s L, Ĉ L Ĉs and Q (p s Q (p L. Thus, for each s L, Q (p s Ĉs. Hence, π ((σ j h t j N π (σ, (σ j h t j. 8

Proof of Proposton 1 Consder any t {0, 1,..., T }, h t = (x s 1, x s 2,..., x s n s t 1 H t \ Z t, and (u j j N. For each j N, et σ j be sncere bddng whch s correspondng to u j, and σ j h t be nduced sncere bddng n the subgame that foows h t. Take any N and σ Σ h t. We sha show that π ((σ j h t j N π (σ, (σ j h t j. If x t 1 Q (p t 1, we can show by Lemma 2. Suppose that x t 1 > Q (p t 1. Let z L+1 = (x s 1, x s 2,..., x s n s L be the termna hstory whch s reached by (σ j h t j N, and w L +1 = (ˆx s 1, ˆx s 2,..., ˆx s n s L be the termna hstory whch s reached by (σ, (σj h t j. Denote {(Cj t j N } L t=0 the cumuatve cnches of z L+1, and {(Ĉt j j N } L t=0 the cumuatve cnches of w L +1. We consder three cases; L > t, L > L = t and L = L = t. Case 1. L > t. Snce L 1 t, by the defnton of sncere bddng, x L 1 = σ h t By Property 1, x L 1 ( (x 1,..., x n L 2 = mn{x L 2, max{c L 2, Q (p L 1 }}. C L 2 or C L 2 = 0. Then, x L 1 = mn{x L 2, Q (p L 1 }. Therefore, x L 1 Q (p L 1, whch s the same argument as step 1 of Lemma 2. Note that we ony use the assumpton x t 1 Q (p t 1 n step 1 of Lemma 2. Thus, we can prove ths case smary to Lemma 2. Case 2. L > L = t. For each j N and each s t 1, obvousy x s j = ˆx s j. Therefore, for each 9

s t 1 = L 1, C s = Ĉs. We w cacuate C L and {Ĉs } L s=l. We frst show that Q (p L C L. By the defnton of sncere bddng, x L = mn{x L 1, max{c L 1, Q (p L }}. Snce x L 1 C L 1 and x L 1 > Q (p L 1 Q (p L, x L = max{c L 1, Q (p L }. Therefore, Q (p L x L. Because x L C L x L 1, Q (p L C L. Next we show that C L ĈL. For each j, because t = L, x L j = σ j h t(h t and ˆx L j = σ j h t(h t. Therefore, for each j, x L j = ˆx L j. Snce the aucton does not end at L n the hstory w L +1, Ĉ L = M j ˆx L j = M j x L j. On the other hand, snce the aucton ends at L n the hstory z L+1, C L M j x L j. Therefore, C L ĈL. Hence, Q (p L C L ĈL. Furthermore, for a s L + 1, Q (p s Ĉs, because Q (p s Q (p L and ĈL Ĉs. Thus, π ((σ j h t j N π (σ, (σ j h t j. Case 3. L = L = t. For each j N and each s t 1, obvousy x s j = ˆx s j. Furthermore, for each j, x L j = σj h t(h t = ˆx L j. Snce for each s L 1, C s = Ĉs, we cacuate C L and ĈL. Case 3-1. C L = x L. By the defnton of sncere bddng, x L = mn{x L 1, max{c L 1, Q (p L }}. 10

Snce x L 1 C L 1 and x L 1 > Q (p L 1 Q (p L, x L = max{c L 1, Q (p L }. If x L = Q (p L, then C L = Q (p L and we have Suppose that x L and C L 1 Ĉ L 1 π ((σ j h t j N π (σ, (σ j h t j. = C L 1. Then, C L 1 Q (p L and C L = C L 1. Snce ĈL = ĈL 1, Ĉ L C L. Therefore, Ĉ L C L Q (p L. Hence π ((σ j h t j N π (σ, (σ j h t j. Case 3-2. C L > x L. Frst, we show that for each j {1,..., 1}, Cj L = x L 1 j. Suppose that there exsts j {1,..., 1} such that Cj L x L 1 j. By the defnton of our ratonng rue, C L j = mn{x L 1 j, x L j + M = x L j + M j 1 x L k Ck L } k=j j 1 x L k Ck L. k=j k=1 k=1 Therefore, j M = x L k Ck L. k=j+1 k=1 Snce for each k N, x L k CL k, and k N CL k = M, j M = x L k Ck L = M. k=j+1 k=1 C L k k N Therefore, for each k j + 1, x L k = CL k. Because j + 1, ths contradcts to C L > x L. 11

Next, we show that C L ĈL. By the defnton of our ratonng rue, C L Ĉ L = mn{x L 1, x L + M 1 x L j Cj L } j= = mn{x L 1, ˆx L + M ˆx L j j=+1 j=1 1 x L j Ĉj L } j=1 = mn{x L 1, M = mn{x L 1, M j=+1 j=+1 1 x L j Cj L }, j=1 1 x L j Ĉj L }. j=1 For each j 1, snce x L 1 j = C L j and x L 1 j ĈL j, Therefore, mn{x L 1, M Hence, C L ĈL. C L j=+1 j=1 C L j ĈL j. 1 x L j Cj L } mn{x L 1, M j=+1 1 x L j Ĉj L }. Smary to case 2, we can show that Q (p L C L. Therefore, Q (p L ĈL. Thus, π ((σ j h t j N π (σ, (σ j h t j. j=1 12