Learning to Bid in Multi-Unit Auctions: Applications in Emissions Trading Systems and Beyond
Abstract: Motivated by Carbon Emissions Trading Schemes (ETS), Treasury Auctions, Procurement Auctions, and Wholesale Electricity Markets—where homogeneous multiple units are auctioned—we address the challenge of learning how to bid effectively in repeated multi-unit pay-as-bid (PAB) and uniform price auctions. In these auctions, a large number of identical items are allocated to the highest submitted bids. In PAB auctions, each winning bid pays its own bid price, whereas in uniform price auctions, the price is set by the smallest winning bid or the highest losing bid. This talk focuses on learning to optimize strategies in both PAB and uniform price auctions from the perspective of a single bidder.
Effective bidding in these auctions is complex due to the combinatorial nature of the action space. To tackle this, we concentrate on the offline setting, where bidders optimize their bids based on historical participant data. We demonstrate that the optimal solution to the offline problem can be achieved using a polynomial-time dynamic programming (DP) scheme, which decouples the bidder's utility across units. Leveraging the structure of this DP scheme, we design online learning algorithms with polynomial time and space complexity, applicable in both full information and bandit feedback settings. Additionally, we establish regret lower bounds to accompany our results.
Numerical experiments suggest that our no-regret learning algorithms drive market dynamics toward a welfare-maximizing equilibrium. Increased competition reduces the impact of strategization, accelerating convergence to higher revenue and welfare levels. Our experiments consistently show that PAB auctions outperform uniform price auctions in generating higher revenue, making them a compelling choice in contexts like ETS, where revenue generation holds substantial social value.
This talk is based on the results in the following two papers:
1- Learning in Repeated Multi-Unit Pay-As-Bid Auctions, Manufacturing & Service Operations Management, 2024.
2- Learning and Collusion in Multi-unit Auctions, The Proceedings of the Conference on Neural Information Processing Systems (NeurIPS), 2023.
Biography: Negin Golrezaei is the W. Maurice Young (1961) Career Development Associate Professor of Management and an Associate Professor of Operations Management at the MIT Sloan School of Management. Her research focuses on advancing online marketplaces—such as e-commerce, online advertising, and emissions trading systems—by designing and implementing data-driven strategies and algorithmic innovations. She aims to create more resilient, equitable, and sustainable digital ecosystems.
Before joining MIT, Negin was a postdoctoral fellow at Google Research in New York, where she collaborated with the Market Algorithm team to develop and test new mechanisms for online marketplaces. She holds a BSc (2007) and MSc (2009) in electrical engineering from Sharif University of Technology, Iran, and a PhD (2017) in operations research from the University of Southern California. Negin serves as an associate editor for Operations Research, Production and Operations Management, Operations Research Letters, and Naval Research Logistics. Her recognitions include the 2021 ONR Young Investigator Award, the 2018 Google Faculty Research Award, the 2017 George B. Dantzig Dissertation Award, the INFORMS Revenue Management and Pricing Section Dissertation Prize, and the USC Outstanding Teaching Award (2017).