Upload
devika
View
45
Download
0
Tags:
Embed Size (px)
DESCRIPTION
Competitive Auctions Review. Rattapon Limprasittiporn. Outlines. Bibliography Introduction Software seller problem Truthful Competitive Goal & Solution. Bibliography. Andrew V. Goldberg Microsoft Research - PowerPoint PPT Presentation
Citation preview
Competitive Auctions Review
Rattapon Limprasittiporn
Feb 9, 2004 Rattapon Limprasittiporn 2
Outlines Bibliography Introduction Software seller problem Truthful Competitive Goal & Solution
Feb 9, 2004 Rattapon Limprasittiporn 3
Bibliography Andrew V. Goldberg
Microsoft Research MASSACHUSETTS INSTITUTE OF
TECHNOLOGY, Doctor of Philosophy degree in Computer Science, January 1987.
Digital commerce models and languages. Auctions. Algorithm Design and Analysis. Implementation and Computational
Evaluation of Efficient Algorithms. Archival Intermemory.
Feb 9, 2004 Rattapon Limprasittiporn 4
Bibliography Jason D. Hartline
University of Washington Postdoctoral research fellow at
Carnegie Mellon University with the ALADDIN Center
Economic aspects of algorithms Optimization problems when input is
private information of selfish agents Game theoretic
Feb 9, 2004 Rattapon Limprasittiporn 5
Bibliography Anna R. Karlin
Stanford University Theoretical computer Design and analysis of algorithms Probabilistic algorithms Online algorithms
Feb 9, 2004 Rattapon Limprasittiporn 6
Outlines Bibliography Introduction Software seller problem Truthful Competitive Goal & Solution
Feb 9, 2004 Rattapon Limprasittiporn 7
Introduction Bidder: person who bid Utility Value
Max price that bidder willing to pay Not the price that bidder pays
Bidder is happy if he pay less than his utility value
Auctioneer: person who sell
Feb 9, 2004 Rattapon Limprasittiporn 8
Example Bidders: Alice, Bob, and Carrol
Bob wins, but ...
Alice
7
Bob
10
Carrol
4
37
Nah!
4Nah!
69
Feb 9, 2004 Rattapon Limprasittiporn 9
Example Bidder’s Goal
Pay minimum price which greater than all other people’s utility values
Problem Lots of bidding tactic
Single-round sealed-bid auction
Feb 9, 2004 Rattapon Limprasittiporn 10
Single-Round Sealed-Bid Single-round
Each bidder submits bid only once Sealed-bid
Bidder blinded from other bidder’s bid Who win? Vickrey auction
The highest bid wins Pay 2nd-highest-bid price
Feb 9, 2004 Rattapon Limprasittiporn 11
Example Bidders: Alice, Bob, and Carrol
Bob wins and pay $7
Alice
7
Bob
10
Carrol
4
7 410
Feb 9, 2004 Rattapon Limprasittiporn 12
Vickrey Auction At the end
The highest bid wins Pay 2nd-highest-bid price
k-item Vickrey auction Have k items to sell
Single price auction k highest bidders win All winners pay the k+1th highest bid
Feb 9, 2004 Rattapon Limprasittiporn 13
Have 2 items to sell Bidders: Alice, Bob, Carrol, Daniel, and Eve
Bob and Denial win Both winner and pay $5
Example
Alice
5
Bob
11
Carrol
4
5 411
Deniel
12
Eve
2
212
Feb 9, 2004 Rattapon Limprasittiporn 14
Truthfulness Should bidder bid their utility value?
Yes, at least in k-item Vickrey Auction An auction is “truthful” if it
encorages bidder to bid their utility K-item Vickrey Auction is a truthful
auction
Feb 9, 2004 Rattapon Limprasittiporn 15
Outlines Bibliography Introduction Software seller problem Truthful Competitive Goal & Solution
Feb 9, 2004 Rattapon Limprasittiporn 16
Software Seller’s Problem
Seller has unlimitted amount of products How can he put them in auction?
Choose k that maximizes his revenue
Selling 3 items to Alice, Bob, and Deniel is better than sell 2 items to Bob and Deniel
Is this a good auction?
Alice
5
Bob
11
Carrol
4
5 411
Deniel
12
Eve
2
212
Feb 9, 2004 Rattapon Limprasittiporn 17
Software Seller’s Problem
Bidders: Alice, Bob, and Carrol
To maximize revenue, seller sells software to Carrol only
Alice
10
Bob
30
Carrol
40
10 4030
Feb 9, 2004 Rattapon Limprasittiporn 18
Software Seller’s Problem
What if Bob changes his bid from 30 to 11
To maximize revenue, seller sells software to Bob and Carrol at the price of 10
Hey, this is not right! This auction is thus “not truthful”
Alice
10
Bob
30
Carrol
40
10 4011
Feb 9, 2004 Rattapon Limprasittiporn 19
Problem Find a good way to “auction” for
software seller We are on the seller side
What is a “good auction” Truthful Yield “good” revenue
Good compared to an “ideal” case “Competitive” (to the ideal case)
Feb 9, 2004 Rattapon Limprasittiporn 20
Outlines Bibliography Introduction Software seller problem Truthful Competitive Goal & Solution
Feb 9, 2004 Rattapon Limprasittiporn 21
Truthful (revisit) What is “truthful” auction?
Encorage bidders to bid their utility Prevent tactic and strategy
How to make an auction “truthful”? Process result of bidderi without
looking at his bid “Bid-Independent Auction”
Feb 9, 2004 Rattapon Limprasittiporn 22
Bid-Independent Auction b = set of all bids that bidders bid For each bidderi
Exclude bid from bidderi to get b-i , the set of all bids except the bid from bidderi
Compute “auction funtion”, f, on b-i to get threshold ti
If bidderi bids more than ti , he wins at price ti , otherwise, he loses
Feb 9, 2004 Rattapon Limprasittiporn 23
Example b = {5, 11, 4, 12, 2} Let “auction function”, f, = “maximum of” For bidder1
b-1 = {11, 4, 12, 2} f(b-1) = 12 Since 5 < 12, bidder1 loses the auction
For bidder2 b-2 = {5, 4, 12, 2} f(b-1) = 12 Since 11 < 12, bidder2 loses the auction
For bidder4 b-4 = {5, 11, 4, 2} f(b-4) = 11 Since 12 > 11 , bidder4 wins the auction at price 11
Feb 9, 2004 Rattapon Limprasittiporn 24
Auction Function Auction function, f, is a core of bid-
independent auction f is “maximum of” = 1-item Vickrey Auction f is “kth maximum of” = k-item Vickrey
Auction Must be “monotone”
If b-i > b-j then f(b-i) > f(b-j)
Every monotone bid-independent auction is truthful
Feb 9, 2004 Rattapon Limprasittiporn 25
Outlines Bibliography Introduction Software seller problem Truthful Competitive Goal & Solution
Feb 9, 2004 Rattapon Limprasittiporn 26
Competitive Good revenue compared to an ideal (in
seller’s sense) case Ideal case: Optimal single price
omniscient auction (F) k highest bids win at price kth highest bid Find k > 1 that yields highest revenue to be
the revenue of F Ex. b = {5, 11, 4, 12, 2, 8}
“k = 3” yields max revenue of 24 Revenue of F is F(b) = 24
Seller is happy if the revenue is close to F
Feb 9, 2004 Rattapon Limprasittiporn 27
Competitive Competitive = good revenue
Competitive to the ideal case Gives revenue within constant factor far
away form F Auction A is competitive if
A(b) F(b) / for some constant , and for all possible bid input b (worst case analysis)
Feb 9, 2004 Rattapon Limprasittiporn 28
Example Is 3-item Vickrey Auction competitive? Let b = {20, 20, 20, 1, 1, 1} 3-item Vickrey Auction gives revenue of 3 Optimal single price omniscient auction F
gives revenue of 60 3-item Vickrey Auction is not competitive In fact, all deterministic auctions are not
competitive!
Feb 9, 2004 Rattapon Limprasittiporn 29
Outlines Bibliography Introduction Software seller problem Truthful Competitive Goal & Solution
Feb 9, 2004 Rattapon Limprasittiporn 30
Goal Find a way of auction to make
software seller happy Truthful Competitive
No deterministic auction is competitive
Sol: randomized auction Compute twice might not be the same
Feb 9, 2004 Rattapon Limprasittiporn 31
Solutions Dual-Price Sampling Optimal
Threshold Auction (DSOT) Sampling Cost-Sharing Auction
(SCS)
Feb 9, 2004 Rattapon Limprasittiporn 32
DSOT Partition bids b randomly into two sets b’
and b’’ Use omniscient auction F to compute ideal
revenue of b’ and b’’ and get p’ and p’’ p’: price that each winner in b’ pay to get F(b’) p’’: price that each winner in b’’ pay to get F(b’’)
Use p’ as a threshold for all bids in b’’ All bids in b’’ less than p’ are rejected All bids in b’’ greater than p’ win at price p’
Use p’’ as a threshold for all bids in b’
Feb 9, 2004 Rattapon Limprasittiporn 33
Example b = {14, 21, 13, 4, 23, 15, 6, 12, 7} Random partition:
b’ = {14, 15, 21, 6, 12, 7} b’’ = {13, 4, 23}
Compute threshold F(b’) = 48 which sell 4 items at price 12 = p’ F(b’’) = 26 which sell 2 items at price 13 = p’’
Use p’’ = 13 as a threshold in b’ Bidders who bid 14, 15, 21 win at price 13
Use p’ = 12 as a threshold in b’’ Bidders who bid 13, 23 win at price 12
Feb 9, 2004 Rattapon Limprasittiporn 34
DSOT Analysis Truthful
Bid-Independent Auction Competitive
Get some factor of F Multiple price?
Feb 9, 2004 Rattapon Limprasittiporn 35
Solutions Dual-Price Sampling Optimal
Threshold Auction (DSOT) Sampling Cost-Sharing Auction
(SCS)
Feb 9, 2004 Rattapon Limprasittiporn 36
SCS Partition bids b randomly into two sets b’
and b’’ Use omniscient auction F to compute ideal
revenue of b’ and b’’ and get F(b’) and F(b’’)
The highest k’ bids in b’ that each bid higher than F(b’’) / k’ win at price F(b’’) / k’
The highest k’’ bids in b’’ that each bid higher than F(b’) / k’’ win at price F(b’) / k’’
Feb 9, 2004 Rattapon Limprasittiporn 37
Example b = {14, 21, 13, 4, 23, 15, 6, 12, 7} Random partition:
b’ = {14, 15, 21, 6, 12, 7} b’’ = {13, 4, 23}
Compute threshold F(b’) = 48 which sell 4 items F(b’’) = 26 which sell 2 items
Two bids in b’, 21 and 15, can share the cost of 26 by paying 13 each
Bidders who bid 21 and 15 win at price 13 No bid in b’’ can share the cost of 48
No bid in b’’ wins
Feb 9, 2004 Rattapon Limprasittiporn 38
SCS Analysis Truthful
Bid-Independent Auction Competitive
Get some factor of F
Feb 9, 2004 Rattapon Limprasittiporn 39
Conclusion Two models for software seller
DSOT SCS
Truthful and competitive Worst case analysis
Feb 9, 2004 Rattapon Limprasittiporn 40
Thank you
Question?