Auctions for Digital Goods Ali Echihabi University of
Auctions for Digital Goods Ali Echihabi University of Waterloo – Nov 2004 Auction for Digital Goods - Ali Echihabi
Talk Overview o o o Review of four papers. New context, new approach. Focus on: “Competitive Auctions”, Goldberg, Hartline, Karlin, Wright, Saks. o o Extra: experiments and extensions. Discussion. Nov 2004 Auction for Digital Goods - Ali Echihabi 2
Our Context o o o Unlimited supply. Identical goods (digital). Utilities may vary a lot. Allocation no longer a problem. Pricing is. Holds for bounded supply. Why auctions then? Nov 2004 Auction for Digital Goods - Ali Echihabi 3
New Approach o o What CS people do best! Algorithms are like auctions: n n n o Correctness : Truthfulness Efficiency: Time Performance: Revenue New concepts for auctions: n n Nov 2004 Provable performance guarantees. Over all possible inputs. Auction for Digital Goods - Ali Echihabi 4
Competitive Auctions o o o Bayesian can go wrong. Run detail-free auctions. Quantify the value of information: n n Nov 2004 How much revenue did we miss? What metric do we use? Auction for Digital Goods - Ali Echihabi 5
Benchmark o Two metrics: n o o o Maximum possible ever. You sell at bidder’s valuation. Multi-price. n o o o Nov 2004 Optimal price-fixing. Single-price for all bidders. We use this one. Auction for Digital Goods - Ali Echihabi 6
Definitions Nov 2004 Auction for Digital Goods - Ali Echihabi 7
Definitions (2) Nov 2004 Auction for Digital Goods - Ali Echihabi 8
Bid-independent Auction Nov 2004 Auction for Digital Goods - Ali Echihabi 9
Important Result o Deterministic Auctions are not competitive. n o Symmetric: outcome is independent of the order of the bids. Need to consider Randomized. Nov 2004 Auction for Digital Goods - Ali Echihabi 10
Randomized Auctions o Lower Bound: o Random Sampling: 1. 2. Nov 2004 Dual-Price Sampling Optimal Threshold (DOST) Sampling Cost Sharing (SCS). Auction for Digital Goods - Ali Echihabi 11
Definition of DOST Nov 2004 Auction for Digital Goods - Ali Echihabi 12
Properties of DOST o Truthful. More is better: o Best DOST can do is 4 -competitive. o Nov 2004 Auction for Digital Goods - Ali Echihabi 13
Definition of SCS o o Special case of Moulin-Shenker mechanism A. k. a Consensus Revenue Estimate (CORE) Nov 2004 Auction for Digital Goods - Ali Echihabi 14
Properties of SCS o o o Truthful. 4 -competitive. May lose half of potential profit: n n o Pick a ratio r < 1. Competitive ratio grows: 4/r. Only sells if revenue reachable. Nov 2004 Auction for Digital Goods - Ali Echihabi 15
Important Result: o is good Competitive auctions do not outperform: n Nov 2004 Considering all possible inputs. Auction for Digital Goods - Ali Echihabi 16
Results Summary o o o No truthful deterministic auction is competitive. Several randomized auctions are truthful and competitive. is a good benchmark. Nov 2004 Auction for Digital Goods - Ali Echihabi 17
Experiments: Number of Bids Nov 2004 Auction for Digital Goods - Ali Echihabi 18
Experiments: Sample Size Nov 2004 Auction for Digital Goods - Ali Echihabi 19
Experiments: Sample Size (2) Nov 2004 Auction for Digital Goods - Ali Echihabi 20
Extensions: Online vs Offline o o Paper borrows concepts from online algorithms That doesn’t make the auction online: n Nov 2004 Decide the price of current bid before next bid arrives [3] Auction for Digital Goods - Ali Echihabi 21
Extensions: Envy o o Bidder X rejected. Bidder Y wins and pays a price lower than X’s bid. Envy is bad for seller. No constant-competitive truthful auction is envy-free. Relax constraints. n n Nov 2004 CORE: Consensus Revenue Estimate (or SCS) Truthful, envy-free, competitive, group-strategy proof Auction for Digital Goods - Ali Echihabi 22
Extensions: Online vs. Offline o o Using theory for online algorithms doesn’t make your auction online. Online auction: determine price for bid i, before next bid arrives. Can’t make all pricing decisions at once [3] gives a randomized online competitive auction within Nov 2004 Auction for Digital Goods - Ali Echihabi 23
Excellent References o [1]: “Competitive Auctions”, Goldberg, Hartline, Karlin, Wright, Saks. o o o [2]: “Competitive Auctions and Digital Goods”, Goldberg, Hartline, Wrigth. [3]: “Incentive-Compatible Online Auctions for Digital Goods”, Bar-Yossef, Hildrum, Wu. [4]: “Envy-Free Auctions for Digital Goods”, Goldberg, Hartline. Nov 2004 Auction for Digital Goods - Ali Echihabi 24
Thank you! Discussion Nov 2004 Auction for Digital Goods - Ali Echihabi 25
- Slides: 25