创造管理新知,造就业界精英,践行持续发展,应对全球挑战。
Our Mission
【2026年11月25日】【管理高等研究院公开学术报告-第166期】平台双向拍卖的智能机制设计:基于可扩展错报搜索实现盈利目标
Speaker: Jiayin Liu, CUHK (Shenzhen)
Date & Time: Wednesday, 25 November 2026, from 10:00 to 11:30 AM (Beijing Time)
Location: Tongji Building A2101
ABSTRACT
In platform-mediated markets, digital platforms increasingly act as market governors by determining the allocation and payment rules that shape trade and intermediary profit. Designing such rules is challenging because platforms must jointly account for multi-level two-sided information interdependencies, strategic reporting by buyers and sellers, and variations in market composition. These challenges make optimal double-auction mechanisms difficult to derive and apply in complex digital markets, motivating a smart-market approach that learns implementable trading rules from market data. We develop DAT-EMS, a learning-based design artifact for profit-oriented platform-mediated double auctions. Its Dual-sided Auction Transformer (DAT) integrates pair-level, within-side, and cross-side information to learn allocation and payment rules. Shared, permutation-equivariant representations enable DAT to accommodate variable market structures while guaranteeing allocation feasibility and individual rationality by construction. To limit gains from misreporting, DAT-EMS penalizes empirical incentive violations during training. Because conventional regret estimation requires a separate misreport search for each trader, its computational cost increases substantially with market size. We therefore introduce Entropy-Regularized Misreport Search (EMS), which replaces these agent-wise searches with a relaxed side-level optimization over candidate misreports and trader-selection masks, thereby enabling scalable identification of severe incentive violations. Experiments show that DAT-EMS achieves near-optimal profit with low empirical incentive violations, generalizes across market conditions, and substantially reduces misreport-search costs. This study provides design knowledge for adaptive, profitable, and approximately incentive-compatible platform trading rules.
Keywords: Double Auction, Deep Learning
GUEST BIO
Jiayin Liu is a Ph.D. Candidate in Economics at The Chinese University of Hong Kong, Shenzhen. Supported by the CSC, she is a visiting joint‑Ph.D. student in Information Systems at the University of Texas at Dallas. Her research covers AI economics and governance, AI‑enabled market design, and algorithmic fairness. Several papers are conditionally accepted or under review at Information Systems Research. She has presented at INFORMS and AISTATS and worked as a teaching assistant for multiple economics‑related courses.