Design of Combinatorial Auctions for Allocation and Procurement Processes Michael Schwind JWG-University Frankfurt CEC-2005 21.7.2005 Technical University of Munich Dipl. Wirtsch. Ing. Michael Schwind, Projekt PREMIUM Internetökonomie Basics of the Combinatorial Auction Design of an Auction Framework Economic Validation of Auction Design Summary and Outlook Literature Dipl. Wirtsch. Ing. Michael Schwind, Projekt PREMIUM Internetökonomie Combinatorial Auction Basics • Bidders` Valuations for Bundles of Goods: – Substitutionalities Subadditivity – Complementarities Superadditivity • Winner Determination Problem (WDP): – Allocation Auction Weighted Set Packing Problem – Procurement Auction Weighted Set Covering Problem • Procurement Auction: min x (S ) p (S ) i xi ( S ) S Bi i i s.t.c. x (S ) 1 j SBi ,S j i xi ( s) i 0;1 i, S Dipl. Wirtsch. Ing. Michael Schwind, Projekt PREMIUM Internetökonomie Combinatorial Auction Variants • Multidimensional Auction: – Exchange of complex preference information – Various dimensions: e.g. quality, delivery time • Multi-attributive Auction: – Impact of attributes on W2P is determined by valuation functions • Multi-item Auction: – Single items of different goods are bundled in bids • Multi-unit Auction: – Multiple items of a good type are bundled in bids Dipl. Wirtsch. Ing. Michael Schwind, Projekt PREMIUM Internetökonomie Combinatorial Auction Advantages / Problems • Advantages: – Higher efficiency in final allocation – Lower transaction costs – Higher transparency • Problems: – NP-hardness of WDP: • Exact solutions: Integer programming, branch-and-bound • Heuristics: Simulated annealing, genetic algorithms – Pricing Problem: • Linear prices / Non-linear prices (anonymous / personalized) – Preference Elicitation Problem: • 2j-1 combinations of bids in worst case – Incentive Compatibility / Stability of Mechanism: • Vickrey-Clarke-Groves (n+1 * NP-hard) Dipl. Wirtsch. Ing. Michael Schwind, Projekt PREMIUM Internetökonomie Combinatorial Auction Process Design • Modeling of the pre and post auction phase: – Organization of the auction preparation and post processing phase – E.g. publication of auction rules, transaction management • Design of the main auction phase: – Major impact on the auction outcome – Design of the allocation mechanism • Modeling of the auction process flow control: – Timing of bidding sequence, closing, clearing time • Legal, security and system stability issues: – Transaction management protocol, etc. Dipl. Wirtsch. Ing. Michael Schwind, Projekt PREMIUM Internetökonomie Basics of the Combinatorial Auction Design of an Auction Framework Economic Validation of Auction Design Summary and Outlook Literature Dipl. Wirtsch. Ing. Michael Schwind, Projekt PREMIUM Internetökonomie Combinatorial Auction Decision Support Ascending / Descending Auction • Fundamental Decisions: Price feedback – – One-shot: sealed-bid VCG usable, only acceptance Iterative: price feedback, anonymous pricing, usage of sealed bid proxy agents, clock auction Bid formation – Bid valuation: multiattributive, manual / automated bid construction (logistics), preference elicitation by questions, bid withdrawal (leveled-commitment) allowed in connection with proxy agents Open-Outcry Clock-Auction Quality Quantity anonymous non-linear pricing allowed only bid acceptance notification required Iterative Auction anonymous linear pricing required Time other individual non-linear pricing allowed Proxy-Agent Sealed-Bidding CA Price Feedback Bid-Valuation Module SealedBidding leveled commitment alllowed CA Bid Formation One-Shot Auction bidwithdrawal allowed manual valuation allowed automated bidgeneration required multiattributive valuation required OR OR-of-XOR AND-OR Bidding Language Constraints AND Vickrey-ClarkeGroves fast result calculation required Integer-Programming Solver Dipl. Wirtsch. Ing. Michael Schwind, Projekt PREMIUM Internetökonomie min. Provider exact result calculation required CA Winner Determination Quantity Turnover approximate result calculation allowed allocation result constraints required GA / SA / Greedy Solver other Winner-Determination Constraints Combinatorial Auction Decision Support • Fundamental Decisions: Bid formation (contd.) – Bidding language constraints: Logic (AND / OR, XOR, OR-of XOR), expressiveness vs. simplicity Ascending / Descending Auction – – Integer programming: small problem size, exact, slow, VCG GA / SA / Greedy: big problem size, approximate, fast computational speed vs. economic efficiency Winner determination constraints: quantity / turnover share, no. provider Quality Quantity anonymous non-linear pricing allowed only bid acceptance notification required Iterative Auction anonymous linear pricing required Time other individual non-linear pricing allowed Proxy-Agent Sealed-Bidding CA Price Feedback Bid-Valuation Module SealedBidding Winner determination: – Open-Outcry Clock-Auction leveled commitment alllowed CA Bid Formation One-Shot Auction bidwithdrawal allowed manual valuation allowed automated bidgeneration required multiattributive valuation required OR OR-of-XOR AND-OR Bidding Language Constraints AND Vickrey-ClarkeGroves fast result calculation required Integer-Programming Solver Dipl. Wirtsch. Ing. Michael Schwind, Projekt PREMIUM Internetökonomie min. Provider exact result calculation required CA Winner Determination Quantity Turnover approximate result calculation allowed allocation result constraints required GA / SA / Greedy Solver other Winner-Determination Constraints Basics of the Combinatorial Auction Design of an Auction Framework Economic Validation of Auction Design Summary and Outlook Literature Dipl. Wirtsch. Ing. Michael Schwind, Projekt PREMIUM Internetökonomie Combinatorial Auction Economic Validation • Analysis and Prototype Design: – Properties of procurement / allocation process • Experimental Game Theory: – – – Field implementation of prototype Small scale experimental field evaluation Iterative redesign • Automated Mechanism Design: – – – Simulation implementation Evaluation using benchmark Iterative parameter optimization • Evaluation: – Mechanism evaluation using benchmark • Meta language description: – Analysis of procurement and allocation process properties and design of auction prototype according to process properties Field implementation of auction prototype Small scale experimental field evaluation Sufficient allocation quality reached ? no yes Implementation of auction prototype in mechanism design optimizer Simulative evaluation of auction using benchmark Optimal allocation quality reached ? no yes Evaluation of mechanism using benchmark Auction description using XML-based CAMeL Description in auction meta language Dipl. Wirtsch. Ing. Michael Schwind, Projekt PREMIUM Internetökonomie auction redesign auction parameter optimization Basics of the Combinatorial Auction Design of an Auction Framework Economic Validation of Auction Design Summary and Outlook Literature Dipl. Wirtsch. Ing. Michael Schwind, Projekt PREMIUM Internetökonomie Combinatorial Auction Summary & Outlook • Advantages of the approach: – Enables trade off in practical environments – Two-step validation of economic properties • Development of a Combinatorial Auction Meta Language (CAMeL): – Enables description of auction in all phases of design process – CAMeL integrates: • Bidding Language description • Auction constraints and admission rules • Auction process control Dipl. Wirtsch. Ing. Michael Schwind, Projekt PREMIUM Internetökonomie Basics of the Combinatorial Auction Design of an Auction Framework Economic Validation of Auction Design Summary and Outlook Literature Dipl. Wirtsch. Ing. Michael Schwind, Projekt PREMIUM Internetökonomie Literatur – Ausubel, L. M., Cramton, P. and Milgrom, P. (2005) The Clock-Proxy Auction: A Practical Combinatorial Auction Design. In Combinatorial Auctions.(Eds, Cramton, P., Shoham, Y. and Steinberg, R.) MIT Press. – Bichler, M., Pikovsky, A., Setzer T. (2005) Kombinatorische Auktionen in der betrieblichen Beschaffung - Eine Analyse grundlegender Entwurfsprobleme. Wirtschaftsinformatik. – Hohner, G., Rich, J., Ng, E., Reid, G., Davenport, A. J., Kalagnanam, J., Lee, H. S. and Chae, A. (2003) Combinatorial and Quantity-Discount Procurement Auctions Benefit Mars, Incorporated and its Suppliers. Interfaces, 33, 23-35. – Kalagnanam, J. and Parkes, D. C. (2003) Auctions, Bidding and Exchange Design. In Supply Chain Analysis in the eBusiness Area.(Eds, Simchi-Levi, D., Wu, S. D. and Shen, M. Z.) Kluwer Academic Publishers. – Kameshwaran, S. and Narahari, Y. (2001) Auction Algorithms for Achieving Efficiencies in Logistics Marketplaces. Proceedings of the International Conference on Energy, Automation and Information Technology. – McAfee, P. and McMillan, J. (1987) Auctions and Bidding. Journal of Economic Literature, 25, 699-738. Dipl. Wirtsch. Ing. Michael Schwind, Projekt PREMIUM Internetökonomie Literatur – McMillan, J. (1995) Why Auction the Spectrum? Telecommunications Policy, 19, 191-199. – Nisan, N. (2005) Bidding Languages. In Combinatorial Auctions.(Eds, Cramton, P., Shoham, Y. and Steinberg, R.) MIT Press. – Porter, D., Rassenti, S. J., Smith, V. L. and Roopnarine, A. (2003) Combinatorial Auction Design. Interdisciplinary Center for Economic Science, George Mason University. – Sandholm, T. (2002a) Algorithm for optimal winner determination in combinatorial auctions. Artificial Intelligence, 135, 1-54. – Schwind, M., Stockheim, T. and Rothlauf, F. (2003) Optimization Heuristics for the Combinatorial Auction Problem. Proceedings of the Congress on Evolutionary Computation CEC 2003, Canberra, Australia, pp. 1588-1595. – Schwind, M., Weiss, K. and Stockheim, T. (2004) CAMeL - Eine MetaSprache für Kombinatorische Auktionen. 2004-111, Institut für Wirtschaftsinformatik, Johann Wolfgang Goethe-Universität. – Smith, V. L. (1994) Economics in Laboratory. The Journal of Economic Perspectives, 8, 113-131. – Vickrey, W. (1963) Counterspeculation, Auctions, and Competitive Sealed Tenders. Journal of Finance, 16, 8-37. Dipl. Wirtsch. Ing. Michael Schwind, Projekt PREMIUM Internetökonomie